INTER-SECTORAL LABOUR MOBILITY IN KOREA: ITS ORIGINS AND RELATIONSHIP WITH UNEMPLOYMENT by Fiona Ai Lin Tan Bachelor of Economics (Hons) (UWA) This thesis is presented for the degree of Doctor of Philosophy of The University of Western Australia Business School University of Western Australia September 2008
403
Embed
INTER-SECTORAL LABOUR MOBILITY IN KOREA: ITS ORIGINS …€¦ · the micro policies to control or reduce mobility rates using the relevant variables (to alleviate unemployment) should
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
INTER-SECTORAL LABOUR MOBILITY IN KOREA:
ITS ORIGINS AND RELATIONSHIP WITH UNEMPLOYMENT
by
Fiona Ai Lin Tan
Bachelor of Economics (Hons) (UWA)
This thesis is presented for the degree of Doctor of Philosophy of
The University of Western Australia
Business School
University of Western Australia
September 2008
i
ABSTRACT
The Asian Financial Crisis was a wake-up call to the South Korean economy that a change
to its economic structure was needed. Prior to the Crisis, South Korea enjoyed healthy
economic growth and low unemployment. With the onset of the Crisis, Korea experienced
severe recession. Unemployment levels soared and turnover in the labour market became
commonplace. The Korean government enacted a series of policies and succeeded in
combating unemployment in the short-term. To the present time, unemployment levels
have been lowered, albeit with job instability and insecurity. A more effective longer-term
solution is needed to increase the resilience of this NIE.
The role of inter-sector labour mobility as a policy tool to combat unemployment using the
relevant determinants of mobility has not been explored in Korea (Asia), although it has
been debated at length in the West since the 1980s. Part of the reason for this lies in the
lack of longitudinal data to facilitate appropriate research. Recently, such data have been
made available by the Korean Labour Institute (KLI). This thesis extends research into the
labour mobility-unemployment relationship to South Korea. The priority is to establish
whether a mobility-unemployment relationship exists in Korea, and to obtain a thorough
understanding of the factors affecting sectoral mobility in this country in order to facilitate
the crafting of potential tools for addressing the unemployment problem.
The thesis is organised into two parts. Prior to the main study, however, the economic
history of Korea is outlined and sectoral labour reallocation patterns are associated with
economic growth. This preliminary work establishes the potential for the detailed research
for the Korean labour market that follows to make contribution to policy solutions to the
unemployment problem along the lines of the earlier research undertaken in Western
countries. Part I, entitled ‘Sectoral Mobility and Unemployment’, details the theoretical
hypotheses and empirical evidence concerning sectoral mobility and unemployment, and
extends the empirical application to Korea. The general finding is that whilst the
hypotheses [Sectoral Shift Hypothesis (SSH), Aggregate Demand Hypothesis (ADH) and
stage-of-the-business-cycle effect] are not relevant for Korea in the pre-Crisis era (1970-
1997), they have some support in the post-Crisis period (1998-2001). However, data
limitations, in the form of the short time period available for analysis, prevent strong
conclusions from being formed. The tentative conclusion is that a new mobility-
ii
unemployment relationship may exist for Korea in the post-Crisis period, thereby giving
rise to the potential for mobility as a policy tool for controlling unemployment. The in-
depth understanding of the factors of sectoral mobility required to implement such policy
provides the basis for the remainder of the thesis.
Part II, titled ‘The Factors Affecting Mobility’, develops a theoretical model for sectoral
mobility, and provides a literature review on other forms of labour mobility (union/non-
union, public-private and rural-urban mobility) as well as an empirical review of sectoral
mobility. These chapters set the stage for the empirical analysis of the determinants of
sectoral mobility for Korea. For the overall workforce, the main conclusion is that sectoral
mobility is a multi-facetted phenomenon involving a spread of factors. Of significance are
the expected and lifetime incomes, which are the pull factors of mobility, the deterrent
effect of the new sector’s unemployment rate and the direct effects of unanticipated sectoral
shocks. The multi-dimensional nature of these factors is replicated in the separate analyses
undertaken for males and females. The main finding is that whilst the monetary variables
and worker/industry characteristics impact male and female mobility differently, sectoral
unemployment and sectoral shock affect male and female mobility similarly.
The thesis is summarised and some policy measures provided in the sypnosis. It is argued
that the ‘new’ mobility-unemployment phenomenon appears to have emerged in Korea
after the Crisis, whereas it had been a feature of Western economies in much earlier time
periods. Traditional monetary and fiscal policies are inadequate when it comes to
combating unemployment in the presence of this mobility-unemployment phenomenon. A
combination of macro-policies, given the relevance of the ADH, and micro-policies, given
the validity of the SSH, is required. The multi-dimensional nature of mobility implies that
the micro policies to control or reduce mobility rates using the relevant variables (to
alleviate unemployment) should cover measures related to monetary wages, labour market
groups and sector performance. The sypnosis notes a dearth of Asian studies on sectoral
mobility, possibly due to the lack of longitudinal data. The collection of quality
longitudinal data for other Asian countries, so that research along the lines conducted in the
thesis could be undertaken for other NIEs, was seen as being of vital importance. With
such data, the standard of research on Asian economies can be at par with that of the
Western countries, and the apparently considerable potential benefits of microeconomic
policies via sectoral mobility for Asia could be realised.
iii
TABLE OF CONTENTS
Page
Abstract
Table of Contents
List of Tables
List of Figures
List of Common Acronyms
Acknowledgements
i
iii
xi
xii
xiii
xv
Chapter
Description
1
1.1
1.1.1
1.1.2
1.1.3
1.1.4
1.2
1.3
INTRODUCTION
Aims of Thesis
Definition of Sectoral Mobility
The Empirical Studies
Sectoral Mobility vis-à-vis Unemployment
The Factors Motivating Mobility
Organisation of the Thesis
Contributions to Labour Economics
1
1
1
1
2
4
5
7
2
2.1
2.2
2.2.1
2.2.2
2.2.3
2.2.4
2.2.5
2.2.6
2.3
2.3.1
2.3.2
2.3.3
2.4
2.4.1
2.4.2
2.4.3
2.5
ECONOMIC HISTORY OF SOUTH KOREA
Introduction
Korea’s Economic History
The Three Kingdoms
Koryo Dynasty
Choson Dynasty (1392-1910)
Japanese Colonial Rule (1910-1945)
Korean War (1950-1953)
Post-war South Korea
Korea’s Economic History in the Post-War Era
The 1970s
The 1980s
The 1990s
Economic Growth, Sectoral Changes and Labour Mobility
The Three Decades: 1970-2000
The 1998-2001 Period
Possible Structural Break during Asian Financial Crisis
Concluding Remarks
10
10
11
11
12
13
13
14
14
15
15
18
19
20
20
25
28
28
iv
Chapter Description Page
PART I:
SECTORAL MOBILITY AND UNEMPLOYMENT
PREAMBLE
29
29
3
3.1
3.2
3.2.1
3.2.2
3.2.3
3.3
3.3.1
3.3.2
3.4
3.5
3.5.1
3.5.2
3.6
3.6.1
3.6.2
3.6.3
3.7
3.7.1
3.7.2
3.7.3
3.8
3.9
THEORETICAL HYPOTHESES CONCERNING
SECTORAL MOBILITY AND UNEMPLOYMENT
Introduction
The Sectoral Shift Hypothesis
Impact of Sectoral Mobility on Aggregate Unemployment
SSH and Supply Shocks
SSH and the Natural Unemployment Rate
Aggregate Demand Hypothesis
U-V Relationship
The σ-U Co-movement Approach
Predicted and Unpredicted Mobility Indices
The Reallocation Timing Hypothesis and Stage-of-the-
Business-Cycle Effect
The Reallocation Timing Hypothesis
The Stage-of-the-Business-Cycle Effect
Conceptual Differences Between the SSH, ADH and RTH
Source of Sectoral Mobility
Chain of Causation
Nature of Unemployment
Methodological Differences
Methods to Test the SSH
Methods to Test the ADH
Methods to Test the RTH and Stage-of-the-Business-Cycle
Effect
Critique of the Mobility Indices
Summary
31
31
32
32
34
36
38
38
40
42
46
46
48
49
49
49
50
50
50
51
54
54
58
v
Chapter
4
4.1
4.2
4.2.1
4.2.2
4.2.3
4.2.4
4.3
4.3.1
4.3.2
4.4
4.4.1
4.4.2
4.4.3
4.5
4.6
4.6.1
4.6.2
4.6.3
4.6.3.1
4.6.3.2
4.6.3.3
4.6.4
4.6.4.1
4.6.4.2
4.6.5
4.6.6
4.6.7
4.7
4.8
Description
THE IMPACT OF SECTORAL MOBILITY ON
UNEMPLOYMENT: A REVIEW OF THE EMPIRICAL
LITERATURE
Introduction
Empirical Review on the SSH
The Raw Lilien Index
The Index Generated by Supply-side Disturbances
Pure Sectoral Shift Measures
The Natural Unemployment Rate Approach
Empirical Findings on the ADH
The Predicted Mobility Indices
The U-V Relationship
Findings on the RTH and Stage-of-the-Business-Cycle
The Horizon Covariance Index
Interaction Variables
Labour Reallocations and Foregone Production
Summary of Empirical Findings
Empirical Application
Type and Frequency of Data
Time Period
Model Estimation
Single-Equation Models
2-Stage Least Squares (2SLS)
Dual-Equation Models
Model Specification
Dependent Variable
Explanatory Variables
Number of σ’s in the Regression Equation
Natural Unemployment Rate Approach
Sectoral Mobility and Gender Unemployment
Summary of Empirical Application
Links with Research on Determinants of Mobility
Page
60
60
60
61
62
62
65
67
67
68
69
69
70
70
71
72
72
73
73
73
80
80
84
84
84
88
89
92
93
94
vi
Chapter Description
Page
5
5.1
5.2
5.2.1
5.2.2
5.3
5.3.1
5.3.2
5.3.3
5.3.4
5.4
5.4.1
5.4.1.1
5.4.1.2
5.4.2
5.4.2.1
5.4.2.2
5.4.3
5.4.3.1
5.4.3.2
5.4.3.3
5.4.3.4
5.4.3.5
5.4.4
5.5
5.5.1
5.5.2
5.6
5.6.1
5.6.2
5.6.3
5.6.4
5.7
SECTORAL MOBILITY AND UNEMPLOYMENT: AN
EMPIRICAL EXAMINATION FOR KOREA
Introduction
Trends in Aggregate and Sectoral Unemployment
Aggregate and Sectoral Unemployment
Sector-specific Employment and Unemployment
Model Framework
Baseline Model
Methodology
Descriptive Statistics
Stationarity
Dual-Equation Modelling
Estimation of Money Growth Equation
Review of Empirical Studies Estimating DMRt
Application to the Korean Case
Specification of Unemployment Equation
Unrestricted to Restricted Models
Preliminary Model Estimation
Structural Change
Prior Knowledge on Korean Unemployment
Tests for Model Stability
Phase I and Phase II
Phase II and Phase III
Accommodation of Structural Change
Re-specification of Unemployment Models
Final Model Estimation
Treatment for Serial Correlation
Sectoral Mobility during the Pre-Crisis Period (1971-1997)
Validity of the Hypotheses
Validity of the SSH
Relevance of the ADH
Applicability of the RTH
Sectoral Movements and Stage-of-the-Business-Cycle Effect
Concluding Remarks
97
97
98
98
98
100
100
100
101
105
106
106
106
108
111
111
114
116
116
116
120
120
123
125
127
127
129
130
131
134
135
136
138
vii
Chapter Description Page
PART II:
THE FACTORS AFFECTING SECTORAL MOBILITY
143
PREAMBLE
143
6
6.1
6.2
6.3
6.3.1
6.3.2
6.3.3
6.4
6.5
6.5.1
6.5.2
6.5.3
6.5.4
6.5.5
6.6
THE THEORETICAL AND CONCEPTUAL ISSUES IN
LABOUR/SECTORAL MOBILITY
Introduction
What is Labour Mobility
Theories of Sectoral/Industrial Mobility
Worker-Employer Mismatch Theory
Sectoral Shock Theory
Bridging Theory
Model of Labour Mobility
Empirical Models of Sectoral Mobility
Probability Choice Models
Simultaneous Equation Models
Vector Auto-regression Models
Sectoral Shock Measures
Time Periods
Summary: Model Application for Current Research
145
145
145
149
149
150
151
151
157
158
162
163
163
164
164
7
7.1
7.2
7.3
7.4
7.5
REVIEW OF THE EMPIRICAL LITERATURE ON
OTHER FORMS OF LABOUR MOBILITY
Introduction
Union versus Non-Union Mobility
Public versus Private Sector Mobility
Rural-Urban Mobility
Summary: Salient Points for Empirical Model
166
166
166
173
179
184
viii
Chapter
Description Page
8
8.1
8.2
8.3
8.3.1
8.3.2
8.3.3
8.3.4
8.4
8.5
8.6
8.7
8.8
EMPIRICAL EVIDENCE: FACTORS MOTIVATING
SECTORAL/INDUSTRIAL MOBILITY
Introduction
Sectoral/Industrial Mobility
Determinants under the Mismatch Theory
Monetary Wages
Macroeconomic Factors
Worker Characteristics
Job/Industry Characteristics
Determinants under Sectoral Shock Theory
Determinants under Bridging Theory
Assessment of Empirical Studies of Sectoral Mobility for
Modelling
Summary of Empirical Studies of Sectoral Mobility
Summary of Lessons Drawn from the Literature
186
186
186
188
194
198
203
220
226
230
231
233
237
9
9.1
9.2
9.2.1
9.2.2
9.2.3
9.2.3.1
9.2.3.2
9.2.3.3
9.2.3.4
9.3
9.4
9.4.1
9.4.1.1
9.4.1.2
9.4.1.3
9.4.2
EMPIRICAL STUDY ON THE DETERMINANTS OF
SECTORAL/INDUSTRIAL MOBILITY IN KOREA
Introduction
Data Sources, Concepts and Coverage
KLIPS Data
Korea NSO Data
The Role of Interim State of Unemployment
Sectoral Labour Flows
Missing Industry Information
Missing Survey Information
Interim States of Unemployment
Generic Model of Sectoral/Industrial Mobility
Descriptive Statistics
Survey Weights
Wave 1 Weights and the Population
Weights for Sample Attrition
Weights for New Entrants
Descriptive Statistics: Complex Statistics
240
240
241
241
245
245
245
247
249
250
251
252
253
254
255
255
256
ix
Chapter
9.5
9.5.1
9.5.2
9.5.3
9.6
9.6.1
9.6.2
9.6.3
9.6.4
9.6.5
9.7
9.7.1
9.7.2
9.8
Description
Derivation of Predicted/Recomputed Variables
Predicted Sectoral Wages
Sector-level Variables
Descriptive Statistics of Predicted/Recomputed Variables
Empirical Analysis: Determinants of Sectoral Mobility
GENDER DIFFERENCES IN SECTORAL/MOBILITY IN KOREA Introduction Model and Sample Dataset Validity of Pooling the Dataset Descriptive Statistics for Males and Females Gender Differences in the Determinants of Sectoral Mobility Monetary Variables Macroeconomic Variables Worker Characteristics Industry Characteristics Sectoral Shock A Gender Perspective on Theories of Sectoral Mobility Worker-Employer Mismatch Theory Sectoral Shock Theory Bridging Theory Decomposition Analysis An Overview of the Standard Decomposition Technique Application to Logit Models Decomposition Results Explanatory Power of Observed Variables Concluding Remarks
THE SYPNOSIS Introduction Part I: Sectoral Mobility and Unemployment Part II: The Factors Affecting Sectoral Mobility The Policy Implications Policy Measures in Post-Crisis Period Assessment of Policy Measures and Current Situation Policy Recommendations Direction for Future Research
337 337 337 342 354 354 355 356 362
REFERENCES
364
LIST OF APPENDICES* 388
* Available on enclosed CD.
xi
LIST OF TABLES
Table Description Page
Table 2.1 Annual % Change in GDP, CPI and Employment (EMP) and 16
Unemployment Rate (UR)
Table 2.2 GDP by Sector, 1970-2000 21
Table 2.3 Employed Persons by Sector, 1970-2000 23
Table 2.4 GDP at Current Prices by Sector, 1998-2001 26
Table 2.5 Employed Persons by Sector, 1998-2001 27
Table 4.1 Studies on the Impact of Sectoral Mobility on Aggregate 63
Unemployment in the U.S.
Table 4.2 R2 between Actual Unemployment Rate and Natural Unemployment 66
Rate
Table 4.3 Contemporaneous Correlations between Labour Reallocation 71
and Average Value Proxies of Foregone Production
Table 4.4 Unemployment and Money Growth Equations used in Selected 74
Studies of Sectoral Mobility
Table 5.1 Employment and Unemployment By Sector 99
Table 5.2 Symbols of Sectoral Mobility 102
Table 5.3 Descriptive Statistics of Ut, DMRt and σ 103
Table 5.4 Initial Parameter Estimates of σ 115
Table 5.5 Phases in the Korean Labour Market from the CUSUMSQ Test 118
Table 5.6 F- and Harvey-Collier Statistics from Tests of Structural Change 122
Table 5.7 Final Model: Parameter Estimates of σ, D and σD and LM statistic 128
Table 5.8 1971-1997: Parameter Estimates of σ 130
Table 5.9 Parameter Estimates of σ, σSt and/or σStD 137
Table 7.1 Selected Studies of Union/Non-Union Mobility 168
Table 7.2 Selected Studies of Public-Private Sector Mobility 175
Table 7.3 Selected Studies of Rural-Urban Sector Mobility 181
Table 8.1 Probability Choice Studies of Sectoral/Industrial Mobility under 190
Worker-Employer Mismatch Theory
Table 8.2 Wages and Sectoral/Industrial Mobility 197
Table 8.3 Unemployment, Employment, GNP and Sectoral/Industrial Mobility 202
Table 8.4 Age and Sectoral/Industrial Mobility 204
Table 8.5 Gender and Sectoral/Industrial Mobility 206
Table 8.6 Marital Status/Head of Household and Sectoral/Industrial Mobility 207
Table 8.7 Education and Sectoral/Industrial Mobility 208
Table 8.8 On-the-job Training and Sectoral/Industrial Mobility 213
Table 8.9 Occupation and Industrial Mobility 217
Table 8.10 Initial Industry and Industrial Mobility 218
Table 8.11 Employment Status and Industrial Mobility 220
Table 8.12 Working Hours, Product Similarity, Work Similarity and Industrial 223
Mobility
Table 8.13 Sectoral Performance Indicators and Sectoral/Industrial Mobility 226
Table 8.14 Sectoral Shocks and Sectoral/Industrial Mobility under Sectoral Shock 229
Theory
Table 8.15 Assessment of the Explanatory Variables 235
xii
Table Description Page
Table 9.1 Gross and Net Labour Flows based on Sample of 29,474 Observations 246
Table 9.2 Gross and Net Labour Flows based on Sample of 29,474 Observations 248
and the Interim State of Unemployment
Table 9.3 Industry Breakdown of 29,474 Sample with/without Survey Information 249
Table 9.4 Gross and Net Labour Flows based on Sample of 10,691 Observations 251
Table 9.5 Wave 1 Weights 254
Table 9.6 Means and Standard Deviations for Korean workers, Aged 20-64 years 258
Table 9.7 Actual versus Predicted Monetary Variables 269
Table 9.8 Means and Standard Deviations for Predicted and Recomputed
Variables 275
Table 9.9 Unrestricted Model: Logit Regression on Probability of
Sectoral/Industrial Mobility 278
Table 9.10 Main Model: Logit Regression on Probability of Sectoral/Industrial
Mobility 281
Table 9.11 Logit Regression on Probability of Sectoral/Industrial Mobility:
A Focus on the Initial Industry, Selected Coefficients 292
Table 9.12 Logistic Regression of Sectoral/Industrial Mobility on Wages and
Alternative Measures of Sectoral Shock, Selected Coefficients 296
Table 10.1 Logistic Regression of ‘Full’ Model 306
Table 10.2 Means and Standard Deviations for Male and Female workers, Aged
20-64 years 308
Table 10.3 Logistic Regression of Sectoral/Industrial Mobility by Gender 314
Table 10.4 Logistic Regression of Sectoral/Industrial Mobility on the Standard
Error of Wage Distribution and Sectoral Shock for Males and Females 324
Table 10.5 Decomposition Results 330
Table 10.6 Explanatory Power of Observed Characteristics in Decomposition 332
Table 11.1 Micro-policy Targets for Korea 360
LIST OF FIGURES
Figure Description Page
Figure 1.1 Lilien Index (σt) and Annual Unemployment Rate (Ut) 3
Figure 2.1 Korea’s Historical Timeline 11
Figure 2.2 Annual % Change in GDP, 1970-2001 17
Figure 2.3 Annual % Change in Employment and Unemployment Rate, 17
1970-2001
Figure 2.4 Annual % Change in CPI, 1970-2001 18
Figure 5.1 DMRt series 110
Figure 9.1 Probability of Sectoral Mobility and Age 284
Figure 9.2 Probability of Sectoral Mobility and Tenure 285
xiii
LIST OF COMMON ACRONYMS
ABS Australian Bureau of Statistics
ACGR Average Annual Compound Growth Rate
AD Aggregate Demand
ADH Aggregate Demand Hypothesis
APEC Asia-Pacific Economic Cooperation
AR Auto-regression
BLS Bureau of Labor Statistics
CILSS Cöte d’Ivoire Living Standards Survey
CO Cochrane-Orcutt
CPI Consumer Price Index
CPS Current Population Survey
CSO Central Statistical Organisation
CSV Cross-Section Volatility
CUSUM Cumulated Sum of Residuals
CUSUMSQ Cumulated Sum of Squared Residuals
DMR Unanticipated Money Growth
DME Anticipated Money Growth
DOLS Dynamic Ordinary Least Squares
DWS Displaced Workers Survey
ECM Error Correction Models
EP Energy Price Index
ESS Error Sum of Squares
GDP Gross Domestic Product
GIC Government Investment Corporation
GNP Gross National Product
HC Harvey-Collier
HILDA Household, Income and Labour Dynamics in Australia
ILO International Labor Organisation
IMF International Monetary Fund
IQ Intelligence Quotient
IT Information Technology
IV Instrumental Variables
KLI Korean Labor Institute
KLIPS Korea Labor Income Panel Study
LM Lagrange Multiplier
LMAS Labour Market Activity Survey
MLE Maximum Likelihood Estimation
NHWI National Help-Wanted Index
NIE Newly Industrialised Economy
NILF Not In Labour Force
NLS National Labor Survey
NSO National Statistical Office
NSW New South Wales
OECD Organisation for Economic Cooperation and Development
OLS Ordinary Least Squares
PID Personal Identification
PPI Producer Price Index
PSC Post-School Certificate
PSID Panel Study of Income Dynamics
RTH Reallocation Timing Hypothesis
SA South Australia
xiv
SSH Sectoral Shift Hypothesis
SME Small and Medium-sized Enterprises
TSM Time-Series Models
TQ Trade Qualification
UI Unemployment Insurance
U-V Unemployment-Vacancies
VAR Vector-autoregression
WA Western Australia
2SE 2-Step Estimation
2SLS 2-Stage Least Squares
σ-U Mobility-Unemployment
σ-V Mobility-Vacancies
Note: Excludes annotations for variables and mathematical symbols.
xv
ACKNOWLEDGEMENTS
This thesis has moved with me through three continents, seven houses and varied states of
employment. It’s a wonder it is finished. I have several people to be grateful for.
My principal supervisor, Professor Paul Miller; who was instrumental in the evolution of
this thesis. His expertise on the area of labour economics and excellent supervision
throughout the thesis will be treasured. His suggestions on modelling and empirical issues,
conscientious attention in reviewing the hundreds of drafts and empirical results, and clear
suggestions in the written drafts are valued, considering that the study was done long-
distance with minimal face-to-face contact. He is the ideal supervisor one could have.
My coordinating supervisor in my initial country of residence, Dr Chai Tai Tee, from the
Government of Singapore Investment Corporation, who gave direction on the choice of
topic and data collection. He sketched a realistic picture on the effort involved and was
willing to give the moral backup outside of the university.
The Korea Labor Institute for their assistance in the execution of the KLIPS software and in
explaining the survey questionnaires and data items which initially appeared in the Korean
language onscreen. The UWA Economics Programme for providing me with the necessary
resources during my residency at the university. A note of appreciation to Ms Derby Voon
for going through my list of references.
Special thanks to my family. To my mother who has been there for me these years; her
perfect blend of kindness and wisdom never ceases to amaze me. To my husband whose
job stint in the Middle East made this study possible and who became the IT helpdesk at
home. To my brother who assisted in the merging of several datasets. To ‘Moses’, my
little Maltese, my source of fun and amusement.
I thank God, too, for bringing these people to me, for without them, this thesis would not be
complete.
1
CHAPTER 1
INTRODUCTION
1.1 AIMS OF THESIS
1.1.1 Definition of Sectoral Mobility
Labour mobility is an area of labour economics that has generated considerable attention in
studies across the world. It involves labour movements across sectors, and can take various
forms, including between union and non-union sectors, public and private sectors, and rural
and urban sectors. This study examines labour mobility across industries or sectors of the
economy. Such labour movements are termed as „industrial‟, „inter-industrial‟, or more
simply „intersectoral‟ or „sectoral‟ mobility.
1.1.2 The Empirical Studies
There has been widespread interest in the study of sectoral mobility, from the perspective
of its causes and consequences. The latter has been particularly popular as a research topic,
with many studies looking at the links between sectoral mobility and employment and
unemployment. These links have been examined for the U.S. [Lilien (1982), Abraham and
Katz (1986), Blanchard and Diamond (1989), Parker (1992), Palley (1992), Brainard and
Cutler (1993), Davis (1987), Mills, Pelloni and Zervoyianni (1995), Loungani (1986),
Murphy and Topel (1987a) and Lu (1996)], Canada [Neelin (1987) and Samson (1985)],
Europe [Saint-Paul (1997) for France, Garonna and Sica (2000) for Italy], and Asia [Prasad
(1997) for Japan]. The seminal paper was Lilien (1982) on the relationship between
sectoral mobility and unemployment1. The subsequent development of this led to several
hypotheses, namely, the Sectoral Shift Hypothesis (SSH), Aggregate Demand Hypothesis
(ADH), Reallocation Timing Hypothesis (RTH) and stage-of-the-business-cycle effect.
This research has policy significance, as if the underlying relationship holds, inter-sector
mobility could be an instrument in combating unemployment via adjustments of its relevant
determinants2. Putting it figuratively, this is analogous to sculpting a new tool to solve an
old problem.
2
Given the links between mobility and unemployment, and the potential of mobility as a
policy tool, the need to understand the factors that motivate it emerges. Interest in these
only followed nearly a decade later, in the late 1980s and 1990s. The studies covering this
topic are for the U.S. [Loungani and Rogerson (1989), Jovanovic and Moffitt (1990),
McLaughlin and Bils (2001), Brainard and Cutler (1993), Fallick (1993), Thomas (1996b),
Neal (1995), Clark (1998) and Kim (1998)], Canada [Osberg (1991), Osberg, Gordon and
Lin (1994), Vanderkamp (1977) and Altonji and Ham (1990)], Europe [Ottersen (1993) for
Sweden and Gulde and Wolf (1998) for the European Union (France, Italy, Germany and
Spain)] and Asia [Prasad (1997) for Japan and Jayadevan (1997) for India]. Though the
number of studies is by no means sparse, given its belated entry into the field of labour
economics, sectoral mobility can be considered to be an infant topic of research.
1.1.3 Sectoral Mobility vis-à-vis Unemployment
The empirical basis for the research into the links between sectoral mobility and
unemployment can be seen for several continents/countries, namely, Oceania (Australia),
North America (U.S. and Canada), Europe (U.K., Sweden and Finland), and Asia (Japan,
South Korea and Singapore). The indicator of unemployment is its rate (Ut). Inter-sector
labour movements can be represented by the raw Lilien index (ζt), the derivation of which
will be outlined in chapter 33. Inter-sector labour movements and the unemployment rate
both fluctuate for each country over the 1970-2001 period4 (see country charts under Figure
1.1). Of relevance is South Korea, where sectoral mobility moved in tandem with
unemployment, especially during the 1998-20015 post-Crisis period where unemployment
reached 7% in 1998, the highest since 1970.
The various empirical investigations into data like that presented in Figure 1.1 suggest that
a mobility-unemployment relationship exists in most Western countries. The first review
of the aggregate-level data in Figure 1.1 suggests that a mobility-unemployment
relationship may also exist for Korea. Given the potential for this relationship to be
exploited in unemployment policy in Korea, it is important to establish more formally its
strength, and to determine how it arises. Part I of this thesis addresses these issues.
3
Figure 1.1 Lilien Index (ζt) and Annual Unemployment Rate (Ut)
Australia U.S.
0.00
2.00
4.00
6.00
8.00
10.00
12.00
1970
1972
1974
1976
1978
1980
1982
1984
1986
1988
1990
1992
1994
1996
1998
2000
0.00
2.00
4.00
6.00
8.00
10.00
12.00
1970
1972
1974
1976
1978
1980
1982
1984
1986
1988
1990
1992
1994
1996
1998
2000
Canada U.K.
0.00
2.00
4.00
6.00
8.00
10.00
12.00
14.00
1970
1972
1974
1976
1978
1980
1982
1984
1986
1988
1990
1992
1994
1996
1998
2000
0.00
2.00
4.00
6.00
8.00
10.00
12.00
14.00
1971
1973
1975
1977
1979
1981
1983
1985
1987
1989
1991
1993
1995
1997
1999
2001
Note: Methodology for 1999 revised; data not strictly comparable.
Finland Sweden
0.00
2.00
4.00
6.00
8.00
10.00
12.00
14.00
16.00
18.00
1971
1973
1975
1977
1979
1981
1983
1985
1987
1989
1991
1993
1995
1997
1999
2001
0.00
1.00
2.00
3.00
4.00
5.00
6.00
7.00
8.00
9.00
1970
1972
1974
1976
1978
1980
1982
1984
1986
1988
1990
1992
1994
1996
1998
2000
Note: Methodology for 1989 revised; data not strictly comparable. Note: Methodology for 1993 revised; data not strictly comparable.
[log (Et-1) - log (Et-2)]. This index provides a workable method of conditioning on past
patterns of labour reallocation in time-series data. Relatively large (small) values for ζH
t-1,
47
ζH
t-2,. indicate that the time t direction of labour reallocation reinforces (reverses) past
patterns of labour reallocation. Reinforcement (reversal) of recent past patterns of labour
reallocation exacerbates (mitigates) skill, location, and informational mismatches between
workers. Davis (1987) estimated the impact of the horizon covariance index using the
following specification of the unemployment equation:
J J J
Ut = βo + β1 DUM74 + ∑β2 ζt + ∑ β3jζH
t-j + ∑ β4j DMEt-j + j=0 j=0 j=0 J
∑ β5j DMRt-j + β6 µt-1 + β7 µt-2 + εt (3.34) j=0
where DUM74 is a dummy variable that equals zero prior to 1974 and one thereafter and
DME is the anticipated money supply growth rate. The error terms, β6µt-1 + β7µt-2 + εt,
follow an AR(2) process. A positive impact for the horizon covariance index on
unemployment was to be interpreted as evidence in favour of the RTH.
The alternative method by Davis (1987) is based on estimating the contemporaneous
correlations between labour reallocation measures and proxies for monetary compensation
and finished goods. According to the RTH, labour mobility and turnover are substitutable
over time. Labour mobility involves unemployment and other forms of foregone
production, which implies that movements in the value of foregone production across
sectors are negatively correlated with unemployment and the pace of labour reallocation.
The two monetary compensation proxies for the cross-sectoral average value of foregone
production are log [compensation index/producer price index (PPI)] and log [compensation
index/consumer price index (CPI)]11
. As these are broad-based measures to proxy for the
average value of foregone production across sectors, they are termed by Davis (1987) as
„cross-sectoral‟.
The deficiencies associated with the monetary compensation measures include the
difficulty in estimating the number of effective hours worked for workers, and the fact that
real wages (under long-term contracts) need not follow short-term movements in the
marginal product of labour. Two alternative proxies based on finished goods were also
adopted, namely, log (manufacturing finished goods inventory at constant prices) and log
(constant prices inventory/manufacturing sales). These are based on the argument that high
48
finished goods inventory levels indicate a lower level of foregone production which triggers
layoffs. Hence, a positive correlation between the finished goods proxies and labour
reallocation should result.
Davis (1987) tested this by letting the raw Lilien index series and the simulated
unemployment series represent the labour reallocation measures. The simulated series is
estimated as per equation (3.34) but without the ζH
t-j component, and with the raw Lilien
index taking its sample values and fixing other regressors in the equation at their sample
means. The proxies of the cross-sectoral average values of monetary compensation and
foregone production discussed above were adopted. A negative correlation between labour
reallocation and the monetary compensation measures and a positive one for the finished
goods were argued to be indicative of confirmation of the fundamental prediction of the
RTH.
3.5.2 The Stage-of-the-Business-Cycle Effect
Whilst the SSH is independent of aggregate economic conditions, the stage-of-the-
business-cycle effect stresses their role in explaining unemployment conditions. During a
recession, there is a tendency for unemployment spells to lengthen and so shocks inducing
sectoral mobility will therefore lead to higher unemployment. Conversely, during an
upturn, unemployment spells tend to be shorter. Consequently, a given rise in sectoral
mobility should result in a smaller increase in unemployment during upturns compared to
downturns. The stage-of-the-business-cycle effect asserts that although the direction of the
ζ-U relation is the same under the SSH and RTH, the magnitude of the increase in
unemployment will be higher during recessions than during boom periods.
The method of testing this type of asymmetric effect involves constructing a multiplicative
interaction variable and testing its effect on unemployment. In Mills, Pelloni and
Zervoyianni (1995), it was the product of the dispersion index, ζt, and S (where S takes
value one when real GNP is below its trend value and 0 otherwise). In Davis (1987), the
interaction variable was the product of RECESS (number of months to recession during the
quarter divided by three) and βζ (where β is the estimated coefficient of ζ). A positive
49
effect of ζt and these interaction variables was to be interpreted as evidence of the stage-of-
the-business-cycle effect.
3.6 CONCEPTUAL DIFFERENCES BETWEEN THE SSH, ADH AND RTH
3.6.1 Source of Sectoral Mobility
Some conceptual differences relating to the role/impact of sectoral mobility on
unemployment are established in this section. Although all hypotheses postulate a positive
ζ-U relationship, there is a difference in the acknowledgement of sectoral mobility as the
cause of unemployment. Whilst the SSH and RTH play up its significance, the ADH
excludes the possibility of this. This difference stems from the source of a sectoral shift. In
the SSH, the shifts are independent of aggregate demand disturbances and can be either
“pure” shifts arising from changes in worker characteristics and sectoral earnings
differentials, or sectoral mobility shifts arising from a supply-side disturbance. In the
ADH, the shifts are derived from pure shocks to aggregate demand, meaning that mobility
is a by-product of an aggregate demand disturbance. Whilst the ADH is based on factors
affecting the macro economy, the SSH has microeconomic foundations. The RTH marries
the two approaches. It is dependent on aggregate economic fluctuations, but it is these
macroeconomic conditions that influence the microeconomic behaviour of individuals.
3.6.2 Chain of Causation
The chain of causation between ζ and U differs between the hypotheses. Whilst it is
sectoral mobility that induces a recession (unemployment) in the SSH and RTH, it is a
recession (generated by an AD disturbance) that causes sectoral movements (which then
lead to unemployment) in the ADH. Although the RTH depends on the stage-of-the-
business-cycle, since past mobility behaviour governs current behaviour, sectoral mobility
is acknowledged as an integral factor generating unemployment. In other words, whilst the
primary cause of unemployment in the SSH and RTH is sectoral mobility, in the ADH it is
the AD disturbance itself.
50
3.6.3 Nature of Unemployment
There is a difference as to how the significance of the components of aggregate
unemployment is perceived amongst the hypotheses. The SSH argues that it is natural
unemployment arising from sectoral movements that causes aggregate unemployment
levels to rise. According to Lilien (1982), “much of the cyclical unemployment is better
described as fluctuations of the natural rate itself”. In contrast, the ADH asserts the role of
cyclical unemployment, since it can be viewed as a form of demand-deficient
unemployment. A negative demand shock would result in cyclical unemployment. The
RTH, however, does not explicitly indicate the nature of the unemployment generated.
3.7 METHODOLOGICAL DIFFERENCES
3.7.1 Methods to Test the SSH
There are methodological differences in the ways the hypotheses have been tested in the
empirical literature. For the SSH, the validation rests upon the statistical significance of the
mobility indices in unemployment models. In relation to the differences in the source of
the sectoral shifts, several measures of the mobility indices have been constructed.
The base measure is the raw Lilien index of equation (3.1). Numerous studies have used
this raw index to test the SSH, namely Lilien (1982), Loungani (1986), Davis (1987), Mills,
Pelloni and Zervoyianni (1995), Loungani and Rogerson (1989), Parker (1992), Palley
(1992), Lu (1996), Neelin (1987), Samson (1985), Saint-Paul (1993), Prasad (1997) and
Brainard and Cutler (1993). Two limitations of the raw index are noted: (a) it captures net
labour flows rather than gross labour flows [Prasad (1997)]; and (b) it assumes that only
pure sectoral shifts affect the dispersion in employment growth rates [Mills, Pelloni and
Zervoyianni (1995)] when aggregate demand and supply shocks could also influence the
dispersion in employment. Owing to these limitations, supply-side and unpredicted
mobility indices have been developed in other studies.
51
The mobility index attributed to supply shocks, introduced by Loungani (1986), is as
expressed in equation (3.9), but it should be noted this index captures only oil price shocks
and not other forms of aggregate supply disturbances. There are variants of the unpredicted
mobility indices intended to capture the sectoral movements from pure sectoral shifts
and/or sector-specific shocks. These indices are purged of aggregate demand disturbances
proxies, namely DMR [equation (3.25)] by Garonna and Sica (2000) and Neelin (1987),
aggregate employment [equations (3.26) and (3.29)] by Lu (1996) and Palley (1992), DMR
and DME [equation (3.31)] and government deficit [equation (32)] by Mills, Pelloni and
Zervoyianni (1995). Others include the index purged of aggregate supply shocks, i.e. an oil
price shock [equations (3.9), (3.11a) and (3.11b)] by Loungani (1986) and Mills, Pelloni
and Zervoyianni (1995).
3.7.2 Methods to Test the ADH
Several methods have been adopted to substantiate the relevance of the ADH, including the
use of predicted mobility indices, ζ-U co-movement approach and the U-V method.
Predicted Mobility Indices
The predicted indices capture the anticipated component of sectoral mobility attributable to
aggregate demand shocks, and their statistical significance in unemployment models is used
to inform on the relevance of the ADH. Two proxies of AD disturbances have been applied
when predicted indices have been constructed, i.e. DMR [equation (3.24)] by Garonna and
Sica (2000) and Neelin (1987) and aggregate employment [equation (3.30)] by Palley
(1992).
The σ-U Co-movement Approach
In the ζ-U co-movement approach, Abraham and Katz (1986) regressed sectoral mobility
on a variable for AD shocks (denoted by deviations of GNP from its trend growth), as per
equation (3.19). First, it was shown that ζt and ∆Ut were positively correlated, given the
negative correlation between the industrial trend growth rates [i.e. d(ln eit)/dt] and their
responsiveness in employment to cyclical fluctuations [i.e. the sum of the δ coefficients
52
from the regression of equation (3.19)]. Second, ΔUt and Ut bore a positive correlation for
the U.S. over 1949-1980 for 11 major sectors. From this, it was concluded that an
aggregate demand-driven ζt-∆Ut correlation could, through a positive ΔUt-Ut relationship,
lead to positive ζ-U co-movements.
The main critique of this method is that there is no direct assessment of the impact of ζ on
U to see if sectoral mobility really does or does not affect aggregate unemployment. As
such, there are limited studies using this method to validate the ADH. The majority of the
studies have used the predicted index approach, as it tackles the issue of whether sectoral
movements (arising from AD shocks) impact U directly.
U-V Method
The U-V correlation was used by Abraham and Katz (1986) to reveal which factors
(aggregate demand disturbances or sectoral shifts) were important in explaining aggregate
unemployment. As mentioned, Abraham and Katz (1986) stated that if the pure SSH
captured why the relationship between ζ and U was positive, that between ζ and V should
be positive, which implies a positive U-V relation. In contrast, the pure ADH suggests
there is a negative U-V relationship. There are, however, doubts over the U-V method
from the theoretical and empirical points of view.
Theoretically, the U-V relation need not be positive under the SSH. A negative U-V
relation is also consistent with the SSH in the presence of asymmetric hiring and firing
costs12
. Thus, by extending Weiss‟ (1984) model to include vacancies, Palley (1992)
illustrated that whilst a negative demand shock to a sector tends to reduce vacancies in that
sector, a positive shock to other sectors, which increases their job vacancies, may not
necessarily offset the decline in vacancies in the sector with the negative shock if it is costly
to hire additional workers in the other sectors. The net effect is an increase in aggregate U
and decline in aggregate V, i.e. a negative U-V relationship. Furthermore, it has been
shown that the correlation between U and V could be negative under the pure SSH and
positive under the pure ADH. Using an equilibrium job matching model, Hosios (1994)
identified circumstances where these particular correlations were possible13
. Under the
SSH, an increase in the price dispersion of output of firms in different sectors could lead to
53
a rise in the number of job searchers, a decline in the actual number of jobs and an increase
in the probability of finding a job, which result in higher unemployment and lower job
vacancies. Under the ADH, an increase in the separation rate could result in a higher
number of workers searching for jobs, leading to a rise in the layoff rate, which causes
higher unemployment and job vacancies.
The criticisms of this empirical approach focus on modelling and measurement issues.
The first lies in the approach to modelling with respect to the posited theoretical
relationship. Abraham and Katz (1986) computed 2 regressions for the U.S. and U.K. as
per equations (3.13) and (3.14). From the resultant positive ζ-U and negative ζ-U
relations, a negative U-V relationship was inferred, suggesting support for the ADH. As
both estimating equations are independent regressions, estimated separately for U on ζ and
V on ζ, respectively, it is not really appropriate to make inference about the U-V
correlation. Since the essence of the positive ζ-U relation is derived from the foundations
of the SSH, it appears that the conclusion arising from the 2 regressions tends to “mix” the
arguments of the ADH and SSH.
The other criticism is that the NHWI used in many studies may be an inadequate proxy for
the vacancy rate. Although Abraham and Katz (1986) concluded that the NHWI tracked
actual vacancies relatively well (by showing that the index and the actual vacancy rate for
Minneapolis/St. Paul were positively correlated with an R2 = 0.8), the evidence is only
based on 1 state. Also, as the NHWI is derived from the number of job advertisements, it is
possible that its increase is attributed to changes in employers‟ advertising practices and
declining newspaper competition [Wachter (1987)] and that the index had not been adjusted
for structural change in the labour market. The 35 per cent reduction of the NHWI by
Abraham and Katz (1986) to accommodate these concerns has been criticized as being
large [Wachter (1987)]. Moreover, since it is based on help-wanted advertising, shifts in
the demand and supply of help-wanted advertising that are unrelated to any change in
vacancies are not considered [Zagorsky (1990)]. Furthermore, even if registered vacancy
data were available for other countries, there may be other vacant jobs not registered with
the authorities14
. For example, adversely affected sectors could reduce job vacancies whilst
favourably affected sectors could recall former workers without relying on registered
advertising [Shin (1997a)].
54
Although the U-V relationship was the central argument of Abraham and Katz (1986), the
criticisms on its directional relationship, modelling techniques and statistical inference, and
adequacy of the NHWI as a proxy for the vacancy rate, appear to make it an inappropriate
approach. Not surprisingly, only a few studies [Davis (1987), Brainard and Cutler (1993),
Palley (1992) and Edin and Holmlund (1997)] have examined the U-V argument. The
majority have instead tested the ζ-U correlation with predicted indices instead.
3.7.3 Methods to Test the RTH and Stage-of-the-Business-Cycle Effect
The statistical significance of the horizon covariance index, as per equation (3.33), in
unemployment models and the contemporaneous correlations between labour reallocation
and proxies for the value of foregone production (as mentioned above) have been used by
Davis (1987) to examine the RTH. To ascertain if the stage-of-the-business-cycle effect
exists, the interaction variable, ζ.S, devised by Mills, Pelloni and Zervoyianni (1995), has
been used.
Therefore, taking into account the criticisms associated with some methods, the preferred
methodology for testing the hypotheses in this thesis is the use of the mobility indices in
unemployment regression models. The section below presents a critique of the mobility
indices in order to assess their suitability for the current study. It does not look at an
exhaustive list of the indices but rather highlights those where deficiencies in terms of
concept and methodology could occur and potentially pose problems for model estimation
and interpretation.
3.8 CRITIQUE OF THE MOBILITY INDICES
Raw Lilien Index
The most widely-used raw Lilien index has led to conflicting claims about the SSH for the
various economies for which research has been undertaken. From its construction outlined
in equation (3.1), the raw Lilien index accounts for the change in inter-sector movements in
excess of aggregate-level labour movements, and as such appears to be a suitable measure
55
of sectoral mobility. The main criticism of this index lies in its inability to capture pure
sectoral mobility purged of aggregate disturbances. This is critical because the SSH
postulates a positive relationship between unemployment and pure mobility. Nonetheless, it
will be applied in this thesis as it forms the baseline index from which other indices are
developed. This is in alignment with the other empirical studies which have used the raw
Lilien index as a benchmark, albeit with awareness of its major limitations.
The σt(p) and σt(up) Indices
The predicted ζt(p) and unpredicted ζt(up) indices were introduced by Garonna and Sica
(2000) for the Italian labour market, as added measures of inter-sector labour movements.
The construction of ζt(p) involves two sets of industry regressions, where:
log eit – log eit-1 = β0 + β1DMRt + β2DMRt-1 + β3T + β4AvgRest + uit
log Et – log Et-1 = β0 + β1DMRt + β2DMRt-1 + β3T + β4AvgRest + ut
with AvgRest as the weighted averages of the residuals from the estimated regressions for
each sector, where the shares of each sector in the total employment are utilized as weights.
The regressions for each sector consisted of regressing (log eit – log eit-1) on DMRt,
DMRt-1 and a time trend.
The introduction of the AvgRest measure rides on the argument that the unanticipated
money growth may not take into consideration all aggregate effects on both the sectoral and
aggregate employment growth rates which it would presumably take care of. However, the
inclusion of AvgRest need not necessarily pick up these additional aggregate effects, and,
more importantly, may introduce measurement errors since it can be considered as a
generated regressor which could potentially lead to inefficient estimates15
. In this case, the
construction may not have a solid basis, as the index is meant to pick up predicted shifts.
Consequently, its unpredicted counterpart, ζt(up), may not be the best indicator of
unpredicted mobility shifts. For these reasons, these two indices will not be used in this
thesis.
56
The σa1
t(up) Index
The ζa1
t(up) of equation (3.26) was used by Lu (1996) in an assessment of the SSH for the
U.S. labour market. The limitation of this index lies in the purging indicator, aggregate
employment, which may not be a strong indicator of aggregate demand per se as it is
negatively related with some other aggregate demand indicators. For example, in the case
of South Korea, it was negatively correlated (correlation coefficient in brackets) with the
ratio of public debt to GDP (-0.299) and ratio of exports to GDP (-0.388). For this reason,
the ζa1
t(up) index will not used in the current work.
The σt(r) Index
The unpredicted ζt(r) index of equation (3.10) was recommended by Loungani (1986) in a
study of the SSH in the U.S. It is a mobility index purged of both aggregate demand (i.e.
unanticipated money supply) and supply (i.e. PPI) factors. Because of this, there is a
concern that the index has been over-purged, and this probably accounts for the
inconsistency of the empirical findings associated with it. Depending on the number of
quarterly lags used, the Loungani (1986) study gave conflicting results, namely, the
estimated impact of the index was positive for ζt-1(r) and ζt-7(r), negative for ζt-3(r) and
ζt-4(r), and insignificant for the other lagged indices. Thus, this measure does seem to be
suitable for empirical work.
The σt(s) Index
Another index introduced by Loungani (1986) is the ζt(s) index of equation (3.9), and it is
the only one reflecting labour movements arising from supply side shocks in the form of oil
prices (PPI). The main concern with this index is its lack of currency. Since there have
only been minute changes in oil prices from 1980 to 2001, the supply shifts are negligible
and there is not much use for the index in the current thesis, particularly if the interest is on
the latter years.
57
The σp1
t(up) and σp2
t(up) Indices
Two SSH indices purged of aggregate supply influences in the form of the energy price
index (EP) are the ζp1
t(up) and ζp2
t(up) applied in the Mills, Pelloni and Zervoyianni (1995)
study for the U.S. economy. Unlike the ζt(s) index, these indices do not reflect movements
from supply influences, but rather movements that have been purged of such. These two
indices will be examined as their plotted series (see Appendix 5C in chapter 5) show
sufficient variability to warrant an examination, and „supply-side‟ indices are needed to
complement the array of the demand-related ones.
Horizon Covariance Index
The horizon covariance index captures inter-sectoral movements from past periods which
are deducted from movements in the present period. The index may not therefore be
suitable for the present study as the present study will be based on annual data and sectoral
movements from up to two years ago may be too distant to exert any influence on current
unemployment. Nonetheless, the use of the index should not be ruled out at this early
stage, as it has won favour in some of the influential literature [see Davis (1987)].
Interaction variable, σtSt
The interaction variable was introduced by Mills, Pelloni and Zervoyianni (1995) to assess
the stage-of-the-business-cycle effect. Given that the Asian Financial Crisis marks a major
turning point in Korea, the interaction variable may not be suitable for the current study.
This is because the post-Crisis era has only 4 annual data points, and because the cycle was
incomplete at the end of the data period examined. However, the interaction variable
should not be omitted at this stage, given the importance of the Asian Financial Crisis.
58
3.9 SUMMARY
This chapter has described the SSH, ADH, RTH and stage-of-the-business-cycle effect and
provided insights into the main conceptual and methodological issues. In summary:
a) The hypotheses not only differ in terms of the presumed impact of sectoral
mobility on unemployment but they also differ conceptually (i.e. source of
sectoral shifts, chain of causation and nature of unemployment) and
methodologically.
b) The method of testing the SSH involves several versions of the mobility index
(raw, supply-side and unpredicted indices purged of aggregate demand and/or
supply disturbances) as regressors in models of unemployment.
c) The methods to test the ADH comprise the ζ-U co-movement approach, U-V
argument and regressing the unemployment rate on a predicted mobility index.
d) The RTH and stage-of-the-business-cycle effect were subject to tests that took
the form of regressing unemployment on the horizon covariance index and
interaction variable, and computing the contemporaneous correlations between
labour reallocation measures and proxies for monetary compensation and
finished goods.
e) The current study will examine the hypotheses using mobility indices (i.e. raw
Lilien index, unpredicted/predicted indices, horizon covariance index and
interaction variable) in unemployment models.
f) Taking into consideration their suitability in terms of concept and methodology,
a total of eleven mobility indices will be utilised in this thesis to test the
hypotheses. They are the SSH indices [ζt, ζa2
t(up), ζm
t(up), ζgt(up), ζ
p1t(up), ζ
p2t(up)],
the ADH indices [ζa2
t(p), ζm
t(p), ζgt(p)], the horizon covariance index (ζH) and the
interaction variable (ζtSt).
The SSH, ADH, RTH and stage-of-the-business-cycle effect have been tested in many
studies in the empirical literature. The findings of these studies, covering various
economies, will be presented and discussed in the following chapter.
59
Endnotes:
1. Other studies examine the role of sectoral shocks on output growth rate [Long, Plosser and Charles (1987), Horvath
(2000) and Norbin and Schlagenhauf (1991)], or sectoral labour mobility on labour productivity [McCombie (1991)] or
sectoral mobility on the recall/retention of jobs [Idson and Valletta (1996)].
2. The other factors influencing the natural rate are labour market distortions (e.g. influence of unions, minimum wage
laws and unemployment insurance) and a change in the profile of the labour force.
3. The Conference Board is an independent global business membership and research organization, and equips businesses
with practical knowledge through issues-oriented research and senior executive meetings.
4. The connection between unemployment and output growth is often formally summarized by the statistical relationship
known as Okun‟s Law. The law relates decreasing unemployment rates with increasing output growth.
5. Abraham and Katz (1986) started the analysis from 1949, rather than 1948 as in Lilien (1982), since 1949 was the
earliest year for which they could obtain data on the help-wanted index.
6. Since Abraham and Katz (1986) stated that „ln Yt is log (GNPt)‟, it can be assumed that it is ln Yt = loge (GNPt), and
hence Yt is the GNPt series.
7. In the process of deriving the predicted and unpredicted indices, some studies have expressed the regression(s) in a
logarithmic series, i.e. as „log‟ [Garonna and Sica (2000), Neelin (1987) and Lu (1996)], whilst others have stated the
series was in natural logarithmic terms [Palley (1992)]. For the studies that do not give the base for the logs, it can be
assumed that they have used natural logs. However, to maintain comparability, this study presents the methodologies
according to the way each empirical study has stated their respective indices.
8. Neelin (1987) mentioned that these predicted and unpredicted indices are analogous to the one used in Lilien (1983).
9. The ζa2t(up) index is a hybrid of predicted and unpredicted indices. The predictive component is captured from the
change in sectoral employment and the unpredicted element arises from the residual component (εit) which presumably
captures pure sectoral influences. Whilst the index has elements of predictability, it is also considered as an index purged
of aggregate demand influences and will be used to test for the SSH in this thesis.
10. Using the average rate of return on capital for industry i in period t, Shin (1997a) computed a cross-sectoral variance
index, ζyit and purged ζy
it by regressing it on the current and lagged growth rates of GNP. However, since the rate of
return on capital is used instead of the employment growth rate, Shin‟s (1997a) study is not relevant to this thesis.
11. These proxies were estimated in logarithmic terms in Davis (1987).
12. Garonna and Sica (2000) also highlighted the role of hiring and firing costs for sectoral shifts in the Italian labour
market.
13. There was no empirical work undertaken for the equilibrium model outlined by Hosios (1994).
14. See Bureau of Labour Market Research, „Structural Change and the Labour Market‟, Research Report no. 11,
Australian Government Publishing Service, Canberra, 1987.
15. Refer to the next chapter for the problem of generated regressors.
60
CHAPTER 4
THE IMPACT OF SECTORAL MOBILITY ON UNEMPLOYMENT:
A REVIEW OF THE EMPIRICAL LITERATURE
4.1 INTRODUCTION
The previous chapter introduced the hypotheses concerning the impact of sectoral mobility
on unemployment: the SSH, ADH, RTH and the stage-of-the-business-cycle effect, and
discussed their conceptual and methodological differences. This chapter aims to review the
related empirical findings on the hypotheses. It then uses this review to indicate how the
empirical application for Korea (chapter 5) might proceed. The organization of the chapter
is as follows. The empirical review is conducted under sections 4.2 to 4.5. Sections 4.6 and
4.7 draw out practical implications for the study of the Korean labour market to be
undertaken in chapter 5, in terms of model specification and estimation. The link to the
microeconomic research on the determinants of sectoral mobility is identified in the
concluding section.
4.2 EMPIRICAL REVIEW ON THE SSH
Sections 4.2 to 4.4 discuss the empirical findings for the various hypotheses with reference
to Table 4.1, which presents the results for the impact of sectoral mobility on
unemployment for the U.S., with those for Canada and Italy reported in the footnote (owing
to the different variables used). Details on these studies are presented here to highlight
differences in the empirical approach. The findings of studies for several other countries –
Japan, Europe and Canada – will also be introduced below. The methodologies in these
studies are generally variants of those employed in the studies for the U.S. listed in Table
4.1. Given the differing hypotheses, and the spread of mobility indicators adopted
within/across the hypotheses, studies with similar indices will be compared under each
hypothesis. This approach establishes conceptual similarity and minimizes the need for
undue explanations where differing results across studies are merely due to conceptual or
methodological dissimilarity. A summary of the findings is found in section 4.5.
61
This section presents the empirical findings for the SSH. Owing to the array of mobility
indices used to test the SSH, the organization of each section is by index-type where the
findings of studies adopting the same type of mobility index are compared. The section
commences with findings from the raw Lilien index – the most commonly-used index, and
ends with the natural unemployment rate approach to testing the hypothesis.
4.2.1 The Raw Lilien Index
Among studies using the raw Lilien index, all for North America and Japan confirm a
positive and significant impact of sectoral mobility on aggregate unemployment. These
include Lilien (1982), Loungani (1986), Parker (1992), Loungani and Rogerson (1989),
Brainard and Cutler (1993), Davis (1987), Mills, Pelloni and Zervoyianni (1995) and Lu
(1996) for the U.S. and Samson (1985) and Neelin (1987) for Canada. Prasad (1997)
computed the raw Lilien index for Japan over 1970-1994 and found a negative relationship
with aggregate employment via graphical analysis, meaning that its impact on
unemployment was probably positive. Similar results were produced when lagged values
of the raw Lilien indices were introduced to cater for long run responses of sectoral
mobility on the macro economy in Lilien (1982), Loungani (1986), Brainard and Cutler
(1993), Davis (1987) and Lu (1996).
For the European economies, the studies acknowledged the role of sectoral mobility on
unemployment but the effect worked in the opposite direction. A negative correlation of the
raw Lilien index with aggregate unemployment was plotted by Saint-Paul (1997) for France
over 1964-1991 and by Garonna and Sica (2000) for Italy for 1952-1994. Rising public
employment in response to higher unemployment and labour market rigidities via
temporary contracts that hampered workers‟ ability to relocate to growth sectors or those
requiring more of specific products/services were the reasons cited for France for this
perverse finding. In Italy, the greater cyclical sensitivity of manufacturing employment vis-
à-vis services employment as compared to the U.S, and firing costs that exceeded hiring
costs (given the high job security) such that unemployment was kept low through sectoral
reallocations (via new hires and pull of new sectors) and interregional mobility were both
argued to have contributed to the negative ζ-U relation.
62
4.2.2 The Index Generated by Supply-side Disturbances
Sectoral movements generated by supply disturbances (oil price shocks) were captured by
an index developed in a single study by Loungani (1986). The lack of similar analyses
reflect the datedness of the 1970s global oil crisis, which has not warranted much research
given the stability of oil prices in latter years. Loungani‟s (1986) index generated by
supply-side disturbances had a positive impact on the unemployment rate for up to 4 lagged
periods.
4.2.3 Pure Sectoral Shift Measures
The pure sectoral shift measures are those that have been purged of the influences of
various aggregate demand and supply variables. With regards to the mobility index purged
of oil price shocks, Mills, Pelloni and Zervoyianni (1995) reported a positive ζ-U
correlation for the U.S. when a one-period lagged version of the mobility index was used.
Again, this is an isolated study which has not generated much interest in the literature,
possibly owing to the datedness of the 1970s oil shock.
Pertaining to the index purged of money growth, the effect on unemployment was positive
for the current period variable and negative for the four-period lagged variable for the U.S.
[Mills, Pelloni and Zervoyianni (1995)] but insignificant for Canada [Neelin (1987)].
When purged of government debt, the index‟s lagged effect was positive for a one-period
lagged variable but negative for a four-period lagged variable [Mills, Pelloni and
Zervoyianni (1995)]1. Conflicting results were produced when the index was purged of
aggregate employment. Palley (1992) reported positive influences for the current and one-
period lagged variables, but Lu (1996) reported an insignificant relationship for all three
lagged indices included in the estimating equation (see Table 4.1). This conflict may
reflect the inappropriateness of using overall employment as a purging tool rather than
specific AD variables.
Loungani (1986) purged the index of aggregate demand (DMR) and supply (oil price
shocks) variables. The one-period and seven-period lagged indices were found to have
significant effects on unemployment. There does not appear to be a convincing argument
for this pattern of significant effects. However, the index may be over-purged, and the
degree of support for the hypothesis under test provided by such results is therefore open to
debate.
Table 4.1 Studies on the Impact of Sectoral Mobility on Aggregate Unemployment in the U.S. Study Lilien (1982) Loungani
(1986)
Parker (1992) Brainard and
Cutler (1993)
Palley (1992) Loungani and
Rogerson
(1989)
Davis
(1987)
Mills, Pelloni and
Zervoyianni
(1995)
Lu
(1996)
Raw Lilien index Over 13
sectors
Over 65
industries
MLE 2SE
ζt 55.9** 0.29** 0.051** 0.0034 0.71+ 0.361**
ζt-1 18.9* 0.40** 0.708** 0.559** 0.37**
ζt-2 0.16 0.467** 0.384** -0.21*
ζt-3 -0.03 -0.136 0.335** 0.04
ζt-4 0.15 -0.123 0.138
ζt-5 0.39** 0.113
ζt-6 0.36** 0.224**
ζt-7 0.46** 0.288**
ζt-8 0.20 0.218**
ζt-9 0.203**
ζt-10 0.138
ζt-11 0.151
ζt-12 0.252**
Δζt 3.856**
Δζt-1 4.690**
Δζt-2 3.333**
Index purged of AD variables
ζmt(up) 2.268**
ζmt-4(up) -1.993**
ζgt-1(up) 4.364**
ζgt-4(up) -2.529*
ζa1t-1(up) 0.09
ζa1t-2(up) 0.32
ζa1t-3(up) -0.07
ζa2t(up) 18.00** 24.15**
ζa2t-1(up) 15.24* 14.72*
ζa2t-2(up) -5.99 -7.61
ζa2t-3(up) 7.91 6.19
Index purged of AS variables
ζp1t-1(up) 3.820**
ζp1t-3(up) -3.519**
ζp2t-3(up) -3.190**
Index purged of AD and AS
variables
ζt(r) 0.24
ζt-1(r) 0.38**
ζt-2(r) 0.10
ζt-3(r) -0.23
ζt-4(r) -0.20
ζt-5(r) 0.06
ζt-6(r) 0.08
ζt-7(r) 0.28**
ζt-8(r) 0.17
64
Table 4.1 Studies on the Impact of Sectoral Mobility on Aggregate Unemployment in the U.S. (continued) Study Lilien (1982) Loungani
(1986)
Parker (1992) Brainard
and Cutler
(1993)
Palley (1992) Loungani and
Rogerson
(1989)
Davis
(1987)
Mills, Pelloni
and
Zervoyianni
(1995)
Lu (1996)
MLE 2SE
Index attributed to AS shocks
ζt(s) 0.60**
ζt-1(s) 0.83**
ζt-2(s) 0.49**
ζt-3(s) 0.47**
ζt-4(s) 0.34**
ζt-5(s) -0.19
ζt-6(s) -0.12
ζt-7(s) 0.07
ζt-8(s) 0.09
Index attributed to Aggregate
shocks
ζat(p) -22.07** -47.42**
ζat-1(p) -21.19** -25.44**
ζat-2(p) -19.00** -14.40
ζat-3(p) -15.89* -13.63
Horizon Covariance Index
Quarterly data series
ζHt-1 0.598**
ζHt-6 0.279**
ζHt-12 0.166**
Annual data series
ζHt-1 -7.9
ζHt-4 28.9*
ζHt-1 6.13
ζHt-3 44.18**
Interaction variable
Quarterly data series
RECESS(βtζt + βt-1ζt-1) 0.172* 5
RECESS ∑ (βiζi ) i=0
0.046
10
RECESS ∑ (βiζi ) i=0
0.077
12
RECESS ∑ (βtζt + βt-1ζt-1) i=0
0.038
∆ (Stζt + St-1ζt-1) 1.210**
** significant at 5% level. * significant at 10% level.
+ : Based on bivariate correlation coefficient.
Note: 1. Although the Lilien indices have been categorized by type (raw, purged of AD/AS/Aggregate variables, pure indices, interaction variable) and by the number of lags, including that for the current period, the indices are not directly comparable across studies. This arises as the impacts of the Lilien indices on unemployment for these studies were based
on different methods of estimation, model specification and number of industries.
2. Findings on coefficients (in brackets) for other countries
Canada: Neelin (1987): Raw Lilien index (31.71) and Index attributed to AD shocks (101.52) were significant at the 5% level. Index purged of AD variables (-7.63) was insignificant.
Samson (1985): Raw Lilien index (81.7) was reported to be significant at the 5% level.
Italy: Garonna and Sica (2000): Index purged of AD variables (0.30) was significant at the 10% level and Index attributed to AD shocks (-0.70) was significant at the 5% level.
65
4.2.4 The Natural Unemployment Rate Approach
The empirical findings on the significance of the natural unemployment rate concur with
the foundations of the SSH (Table 4.2). Thus, the natural unemployment rate explained a
significant portion of the variations in the aggregate unemployment rate in Lilien (1982)
and Mills, Pelloni and Zervoyianni (1995). Furthermore, when alternative indices (index
purged of aggregate demand influences and interaction variables for the stage-of-the-
business-cycle effect) were included in the unemployment equation used to estimate the
natural rate in the latter study, the natural rate accounted for a slightly larger proportion of
the actual rate. Parker (1992) estimated the natural rate as the fitted value of the
unemployment equation with the unanticipated money growth and residual obtained from
the unemployment regression set equal to zero, and plotted this series against that of U. It
was observed that U exceeded U* during 1956-1964. Towards the late 1960s and early
1970s, U declined due to microeconomic factors (labour supply shortage associated with
the Vietnam war) and the U-U* disparity lessened. Given that the natural rate is that which
is attributable solely to microeconomic factors, the narrowing of the U*-U gap implies that
much of the unemployment in the 1970s was accounted for by natural (microeconomic)
factors, i.e. the natural rate. Samson (1985) plotted the U* series against the actual U series
over the 1957-1983 period and found a small deviation of 0.45 between the two series
(based on their average absolute values), implying that the natural rate could explain the
actual series relatively well. Loungani (1986) constructed two measures of the natural rate,
U*t(s) and U
*t(r), where the former was estimated from the regression associated with the
index attributed to the oil shock, and the latter with the unpredicted index. Using quarterly
data for the U.S. over j = 8 lags, these were calculated as:
8
U*t(s) = β0 + ∑ βj ζ t(s)-j and
j=0
8
U*t(r) = β0 + ∑ γj ζ t(r)-j
j=0
where β0 is the intercept estimate from the regressions and βj and γj were each the
estimated coefficients attached to the ζt(s) and ζt(r) variables. It was reported that U*t(s)
accounted for 20% of U, higher than the 5% reported for U*t(r).
66
Table 4.2 R2 between Actual Unemployment Rate and Natural Unemployment Rate
Study Lilien (1982) Mills, Pelloni and
Zervoyianni (1995)
Loungani (1986)
Between Ut and U*t Between ∆Ut and ∆U*t Between Ut and U*t
Index used
Actual
series
Detrended
series
Detrended series
ζt 0.74 0.60 0.52
ζt(s) 0.20
ζt(r) 0.05
ζm
t(up) 0.55
∆(Stζt + St-1ζt-1) 0.57
∆(Stζm
t(up)+St-1ζm
t(up)-1) 0.56
Whilst several studies have supported the general influence U* has on U, namely, Parker
(1992), Loungani (1986) and Samson (1985), a counter argument questioning the influence
of U* was presented by Murphy and Topel (1987a). They made use of a constant natural
rate argument to conclude a lack of support for the SSH. In theory, sectoral movements
under the SSH generate frictional unemployment which should lead to changes in the
natural rate. Using unit-record cross-sectional data for male employees in the U.S., it was
shown that only 2.4-4.0 per cent of the total unemployed were industry movers during
1968-1985, and that this proportion remained virtually constant throughout this period,
implying a constant natural rate of unemployment. Since changes in the natural rate are
implied under the SSH in that U* varies with frictional inter-sector labour movements,
Murphy and Topel (1987a) concluded that the constant natural rate (implied from the non-
varying 2.4-4 per cent) did not concur with the hypothesis. This contrasts with the other
evidence for the U.S., such as Parker (1992) and Lilien (1982), which illustrated a
fluctuating U* series over the period, but it should be noted that two different sets of data
are used to support the evidence for the SSH or lack thereof. Murphy and Topel‟s (1987a)
constant rate is implied from descriptive data covering males only, whilst Parker‟s (1992)
and Lilien‟s (1982) U* series has been estimated formally and covers both males and
females. Furthermore, in response, Lilien (1987) argued that Murphy and Topel (1987a)
misinterpreted the underlying implications of the SSH, where sectoral mobility is generated
by frictional movements as well as economic shocks.2 Thus, whilst Murphy and Topel
(1987a) interpreted inter-industry movements to arise from frictional labour movements,
such sectoral movements can also originate from economic shocks.
67
4.3 EMPIRICAL FINDINGS ON THE ADH
The methodologies to test the ADH comprise the use of predicted mobility indices, the U-V
relationship and ζ-U co-movement approach. The organization of this section is to
evaluate the empirical works according to these methodologies. For the latter approach, the
findings are not assessed since Abraham and Katz (1986) appears to be the single study
using the ζ-U co-movement method, and this has been described in the previous chapter.
4.3.1 The Predicted Mobility Indices
Predicted indices from aggregate demand disturbances have been used to test the ADH.
The two studies adopting this approach have reported opposite results. The ζ-U relation
was positive in Neelin (1987) for Canada but negative in Garonna and Sica‟s (2000)
analysis for Italy. For the latter, the inverse relation was held to reflect the differing
cyclical responsiveness of economic sectors which triggers unemployment, thereby leading
to the deduction that sectoral movements in Italy were generated from an AD disturbance
and that unemployment is cyclical. Thus, whilst the Canadian experience appears to
provide clear support for the ADH via its positive ζ-U relation, the Italian outcome can be
argued to be aligned to the ADH only in terms of the source of sectoral shifts and nature of
unemployment, and certainly not from the evidence of the directional influence of ζ on U.
A predicted index generated by aggregate employment was examined by Palley (1992) for
the U.S. over 1951-1988 for 11 sectors. A negative influence on unemployment from
mobility for the current-period index and indices lagged by three periods was reported.
This does not support the ADH, since the study for the U.S. by Abraham and Katz (1986)
asserted that there should be positive ζ-U co-movements. Since Palley (1992) and
Abraham and Katz (1986) cover the same economy and almost similar time periods but
reveal differing results, the method of filtering involved when constructing this form of
predicted index (i.e. the use of aggregate employment) remains questionable.
68
4.3.2 The U-V Relationship
As explained in the previous chapter, under the SSH the U-V relationship should be
positive, and ζ-U and ζ-V should both be positively related as well. Under the ADH, the
positive ζ-U association generates a negative ζ-V relationship following an aggregate
demand shock and the resulting U-V relation is inverse. In this section the empirical
support for the ADH from the correlation results of U-V and/or ζ-U and/or ζ-V is
examined. Generally, it is shown that the conclusions on the ADH using the U-V
relationship are contradictory.
Davis (1987) plotted the ζ, NHWI and unemployment inflows and outflow series for 1948-
1986. It was shown that: (a) periods of high (low) unemployment inflow and outflow rates
coincided with declining (rising) NHWI levels; and (b) periods of rapid rates of labour
reallocations accompanied high unemployment rates. The negative U and V correlation
and the positive ζ and U relationship appeared to be consistent with the ADH rather than
the SSH.
Brainard and Cutler (1993) estimated the Beveridge Curve by regressing the logarithm of
the vacancy rate against the current and lagged values of a cross-section volatility (CSV)
measure, the raw Lilien index and the unemployment rate3:
Note: As the author did not specify the money growth equation, it is
assumed the equation follows that of Barro (1977).
75
Table 4.4 Unemployment and Money Growth Equations used
in Selected Studies of Sectoral Mobility (continued) Study Country/ Time
Period/Data-
type/Method of
Estimation for
Aggregate
Unemployment
Model Specification
Davis (1987) U.S., 1953-1986, Aggregate-level time- series quarterly data. Method of Estimation Joint-estimation of the unemployment and money growth equation using non-linear least squares.
+ β5 DUM74 + β6 μt-1 + β7 μt-2 + εt Note: a: J varies as 6 regressions were run separately with ζHt-1, ζHt-6,
ζHt-12, (ζHt-1 and ζHt-12), (ζHt-1, ζHt-6 and ζHt-10) and (ζHt-1, ζHt-4, ζHt-8 and ζHt-10). Ut is seasonally-adjusted for quarterly series. The error terms, βj μt-1 + βj μt-2 + εt, follow an AR(2) process. Regression 3: 12 9 12
Note: Ut is seasonally-adjusted for the quarterly series. 4 regressions were estimated separately for RECESS(∑βjζt-j) lagged by 1, 5, 10 and 12 periods. The error terms, βjμt-1 + βjμt-2 + εt, follow an AR(2) process. Money Growth Rate 12 4 4
Davis (1987) U.S., 1924-1985, Aggregate-level time- series, annual data. Method of Estimation Joint-estimation of the unemployment and money growth equation using non-linear least squares.
where ECMt = DMt - β1yt - β2 pt + β3 T + β4Rt + β5Ut,, y is income, p is
prices, T is the time trend, i is the inflation rate, g is government deficit and
R is the interest rate. Palley (1992)
U.S., 1951-1988, Aggregate-level time-series quarterly data (seasonally-adjusted). Method of Estimation Maximum likelihood estimation (MLE) for single-equation regression.
3 3
Ut = βo + ∑ β1jζa2
t(p)-j + ∑ β2 jζa2
t(up)-j + β3 Tt + β4 Ut-1 + εt
j=0 j=0
Note: Ut is seasonally-adjusted.
77
Table 4.4 Unemployment and Money Growth Equations used
in Selected Studies of Sectoral Mobility (continued) Study Country/ Time
Note: U.S. GNP is the logarithm of real GNP, BILL is the 3-month treasury
bill rate and EX is the logarithm of U.S. exports. The U.S. variables were
included to avoid simultaneity bias.
Samson
(1985)
Canada, 1957-1983, Aggregate-level time-series annual data. Method of Estimation: 2SE.
Aggregate Unemployment Regression 1: Ut = βo + β1 ζt + β2DMRt-1 + β3 Ut-1 + β4T + εt Regression 2: Ut = βo + β1 ζt + β2DMRt-1 + β3 Ut-1 + β4 LFPRWt + εt Note: LFPRW is the ratio of women in labour force to total labour force Regression 3: Ut = βo + β1 ζt + β2DMRt-1 + β3 Ut-1 + β4 U.S. Ut + β5T + εt Regression 4: Ut = βo + β1 ζt + β2DMRt-1 + β3 Ut-1 + β4 U.S. U*t + β5 U.S. MSt
+ β6 U.S. MSt-1 + εt Note: MS is money supply. The sample period for regression 4 is 1957-1980. Natural Unemployment Rate (from Regression 3) U*t = βo + β1 ζt + β3 U*t-1 + β4 U.S. Ut + β5T As initial U*t-1 was not observable, actual U was used for the first observation. The value obtained was substituted back into the equation to generate the next U* until the last observations of ζt, U.S. Ut and T were utilized. Money Growth Rate DMt = α0 + α1 DMt-1 + α2DMt-2 + α3DMt-3 + α4FEDVt + α5Ut-1 + α6 DMtU.S. + DMRt Note: DMtU.S. is the U.S. M1 growth rate.
79
Table 4.4 Unemployment and Money Growth Equations used
in Selected Studies of Sectoral Mobility (continued) Study Country/ Time
and applying Barro‟s (1977) proposition to equation (1) that DMR is obtained solely from the history of DM
gives:
96
DMt = α0 + α1 DMt-1 + α2DMt-2 + DMRt (3)
The estimated unemployment equation with the estimated DMR from equation (3) should have a poorer fit (as
proven by Barro (1977) for the U.S. for 1946-1973). Substituting into the estimated equation (2), from the ^ ^
condition that DMR ≡ DM - DM, where DM is from the estimated equation (1), the „reduced form‟ unemployment becomes a function of (DMt-1 …..DMt-j), DMEt-j , DUM74, BILLt and UNt-1. The restrictions could be that the parameters associated with DMEt-j, DUM74, BILLt and UNt-1 are zero, such
that Ut = f (DMt-1 …..DMt-j). Davis (1987) did mention that BILLt and UNt-1 did not enter the unemployment
equation except through the money growth equation. Hence, the cross-equation restrictions could be
tantamount to testing the joint hypothesis: H0: α3 = α4 = β3 = β4 = 0.
In the same manner, applying Barro‟s (1977) proposition to Parker‟s (1992) equations below:
Table 5.7 presents the OLS estimates of the mobility indices from the final specifications.
The LM statistic was used to test for the presence of serially-correlated errors15
. Given the
use of annual data, only first-order serial correlation was considered. First-order serial
correlation was prevalent in equation (5.1*) with ζH and equation (4.9a*) with ζp1
t(up),
ζH*100 and ζtSt. This is of particular concern as the t-values will typically be inflated with
positively correlated error terms. Accordingly, a Cochrane-Orcutt (CO) correction for serial
correlation was applied to the affected equations16
.
Where serial correlation has been detected, the CO estimates are displayed alongside the
OLS estimates. As using CO estimation results in a loss of the first observation, which is
substantial in the current application considering that the dataset is based on only 31 data
points17
, a Prais-Winsten transformation is applied to preserve the first observation,
whereby it is written as:
U1* = U1 (1- ρ2)1/2
; and
X1* = X1 (1- ρ2)1/2
where X denotes the set of explanatory variables in the estimating equation. Compared to
the OLS estimates, there were differences in the size of the parameter estimates and t-
values when the estimates corrected for serial correlation are considered. In particular, the
t-values of several of the OLS estimates were larger in the presence of serial correlation.
Having specified the models to reflect structural change, reduced multicollinearity and
serial correlation, and omitted irrelevant variables, reliable, consistent and efficient CO
estimates can be obtained to examine the validity of the SSH, ADH and RTH hypotheses.
In general, compared to the OLS estimates of Table 5.4, the fit of the model has improved
substantially, with the adjusted R-squared ranging from 0.7 to 0.9. This improvement
arises due to the modelling of structural change.
128
Table 5.7 Final Model: Parameter Estimates of ζ, D and ζD and LM Statistic Regression with: Equation (4.9a*) Equation (5.1*) OLS CO LM statistic OLS CO LM statistic ζt 48.951
(2.923) n.u. 2.13 8.992
(0.470) n.u. 1.62
ζtD 61.209 (3.245)
102.679 (4.942)
ζm
t(up) -2.632 (-1.752)
n.u. 0.93 -2.189 (-1.532)
n.u. 0.15
ζm
t(up) D 33.046 (6.873)
27.576 (5.276)
D -5.957 (-5.574)
ζgt(up) 17.057
(1.226) n.u. 0.48 4.775
(0.388) n.u. 0.33
ζgt(up) D 127.108
(4.639) 120.865
(5.152)
D -2.734 (-3.453)
ζa2
t(up) 0.534 (1.209)
n.u. 1.41 -0.017 (-0.036)
n.u. 0.36
ζa2
t(up) D 3.583 (4.794)
4.453 (5.488)
ζp1
t(up) 41.745 (2.862)
37.936 (3.145)
10.14 9.647 (0.791)
n.u. 1.62
ζp1
t(up) D 40.369 (2.779)
38.484 (3.184)
70.992 (5.488)
ζp2
(up) 4.908 (0.432)
n.u. 0.84 -11.504 (-0.971)
n.u. 0.30
ζp2
(up) D 130.095 (4.954)
131.703 (5.011)
D -3.281 (-3.641)
-2.502 (-2.050)
ζa2
t(p) -9.096 (-0.557)
n.u. 2.19 -34.363 (-2.301)
n.u. 2.10
ζa2
t(p)D 303.477 (3.149)
103.779 (5.272)
D -7.982 (-2.743)
ζm
t(p) -1.698 (-0.933)
n.u. 0.18 -0.317 (-0.181)
n.u. 1.74
ζm
t(p)D 21.052 (4.295)
11.857 (4.461)
D -2.685
(-2.826)
ζgt(p) 19.137
(1.210) n.u. 0.15 -10.173
(-0.645) n.u. 1.59
ζgt(p)D -340.299
(-7.254) -272.188
(-5.095)
D 7.772 (8.010)
7.552 (7.915)
ζt 67.704 (4.115)
48.106 (2.820)
6.30 46.071 (1.920)
19.849 (0.962)
5.97
ζH*100 0.825 (0.165)
0.256 (0.060)
2.746 (0.515)
1.057 (0.244)
ζH*100D 132.806 (3.666)
191.558 (2.229)
165.745 (2.651)
105.074 (1.316)
ζt 20.535 (1.325)
39.579 (2.691)
3.09 11.868 (0.601)
n.u. 1.14
ζtSt 158.621 (3.723)
9.646 (0.954)
-6.115 (-0.712)
ζtStD 158.621 (3.723)
61.011 (3.810)
108.882 (4.793)
Note: t-values in parentheses. n.u.: not undertaken.
129
The regressions with three indices (ζp1
t(up), ζH*100 and ζtSt) had to be corrected for serial
correlation. Comparing the OLS and CO estimates in Table 5.7, the standard errors were
higher for ζp1
t(up) and ζtSt under equation (4.9a*), those for ζH*100 under equations (5.1*)
and (4.9a*) under OLS estimation. A likely reason is that their post-Crisis interaction
dummy variables constitute 4 data points, and perhaps this period is not sufficiently long to
give the higher standard errors under CO estimation. Nonetheless, the standard errors of the
regressions under CO estimation are fairly close to those under OLS estimation. The CO
estimates can be considered to be valid for statistical inference.
It is noted that serial correlation is not prevalent in regression equations with the other
indices, and their OLS estimates are unbiased, consistent and efficient, and valid for
statistical inference. Compared to the OLS estimates of Table 5.4, the fit of the newly
specified models improved, with the adjusted R-squared of at least 0.8. For reference
purposes, the estimating equations of these final unemployment models (as per Table 5.7)
can be found in Appendix 5G.
5.5.2 Sectoral Mobility during the Pre-Crisis Period (1971-1997)
The unemployment equations (5.1*) and (4.9a*) were also estimated for 1971-1997 to
ascertain if the mobility-unemployment relationship found for this truncated data period is
consistent with that established with the larger sample (1971-2001) after accommodating
the structural break. This is a test of the adequacy of the way the structural break has been
modelled. Since the 1998-2001 observations are removed, the interaction variables with D,
and D itself, are excluded. The same equations cannot be estimated for 1998-2001 as the
number of parameters exceeds the number of observations.
The SSH/ADH indices and horizon covariance index, which were insignificant for 1971-
2001, were also insignificant for 1971-1997 in both models, confirming that predicted and
unpredicted mobility as well as past labour reallocations did not cause unemployment
during the pre-Crisis period. Pertaining to the stage-of-the-business-cycle effect, ζtSt
remained insignificant for equations (5.1*) and (4.9a*) for 1971-1997.
130
Thus, where the mobility indices were insignificant variables for 1971-2001, they were also
insignificant for 1971-1997. Likewise, whilst the index [i.e. ζa2
t(p)] was significant for the
full data period under equation (5.1*), it was also significant for the pre-Crisis period. This
implies that modelling of the 1997 structural break using the dummy variable does not
introduce any distortions to the fundamental relationships between mobility and
unemployment over 1971-1997.
Table 5.8 1971-1997: Parameter Estimates of ζ Regression with: Equation
(4.9a*)
Equation
(5.1*)
ζt 16.147
(1.020)
9.619
(0.627)
ζm
t(up) -2.296
(-1.848)
-2.140
(-1.776)
ζgt(up) 6.336
(0.459)
5.038
(0.389)
ζa2
t(up) 0.227
(0.777)
0.047
(0.128)
ζp1
t(up) 14.603
(1.267)
5.289
(0.425)
ζp2
t(up) 6.563
(0.658)
-10.989
(-0.980)
ζa2
t(p) -14.500
(-1.469)
-28.891
(-3.155)
ζm
t(p) -1.563
(-1.312)
-1.172
(-0.879)
ζgt(p) 10.870
(0.728)
-8.213
(-0.550)
ζH*100 0.261
(0.079)
0.854
(0.242)
ζtSt 8.945
(1.325)
-2.148
(-0.293)
1. All figures are OLS except for the CO
estimates (applied with Prais-Winsten
transformation) of ζt, ζm
t(up), ζgt(up), ζ
gt(p) and
ζp2
t(up) under equation (5.1*).
2. t-values in parentheses.
5.6 VALIDITY OF THE HYPOTHESES
The main focus of this analysis is to see if sectoral labour movements generate aggregate
unemployment and this is to be done through examination of the statistical significance of
the mobility indices. With the variety of indices, there arises a need to establish the set of
indices which are robust under the alternative restricted models. Since equations (4.9a*)
and (5.1*) are similar in that: (i) equation (5.1*) has its roots in the basic Lilien approach,
131
while equation (4.9a*) is a more encompassing version of the former equation; and (ii) the
structural change has been modelled into both equations, the result in terms of the
significance of the mobility indices must at least be broadly consistent across these two
models.
5.6.1 Validity of the SSH
Each SSH index has been purged of differing influences and it is important to examine the
validity of the SSH by comparing the findings with studies of similar index type.
Raw Lilien Index
The raw Lilien index was significant under equation (4.9a*) but insignificant under
equation (5.1*). However, its interaction with the dummy variable was significant and
positive for both models. The finding under equation (4.9a*) is consistent with the
numerous studies for North America and Japan reporting a significant positive impact on
unemployment: Lilien (1982), Loungani (1986), Abraham and Katz (1986), Parker (1992),
Loungani and Rogerson (1989), Brainard and Cutler (1993), Davis (1987), Mills, Pelloni
and Zervoyianni (1995) and Lu (1996) for the U.S., Neelin (1987) for Canada and Prasad
(1997) for Japan. It contradicts findings from Europe: France [Saint-Paul (1997)] and Italy
[Garonna and Sica (2000)], which showed mobility to affect unemployment in the opposite
direction. As mentioned in chapter 4, the negative effect has been attributed to labour
market rigidities in France, i.e. temporary contracts and rising public sector employment,
and the high firing costs and differences in cyclical sensitivities in Italy‟s manufacturing
and services sectors.
The sensitivity of the results under the alternative models may, in addition to the potential
omitted variables bias in the two non-nested models, reflect the deficiency of the raw index
in aligning with the concept of the SSH. Several studies have criticized the index and opted
and/or recommended alternative indices to examine the SSH, even though it was reported
to be significant. These include Loungani (1986), Palley (1992), Mills, Pelloni and
Zervoyianni (1995), Lu (1996), Neelin (1985) and Garonna and Sica (2000). It is the intent
of the following sub-sections to do likewise.
132
Pure Sectoral Shifts Purged of AD disturbances
Similar results were produced for ζm
t(up), ζgt(up) and ζ
a2t(up) under equations (5.1*) and
(4.9a*), in that they had an insignificant impact on unemployment. In the unpredicted
sense, a mobility-unemployment relationship did not exist before 1997 in Korea.
Consistency in findings was displayed in the post-1997 findings. All three interaction
variables, i.e. ζm
t(up)D, ζgt(up)D and ζ
a2t(up)D, had positive and significant coefficients under
equations (5.1*) and (4.9a*). The post-Crisis finding agreed with the reports of the
empirical studies in terms of the ζ-U impact, although the data periods differ. Whilst the
findings for the first two were in tandem with Mills, Pelloni and Zervoyianni (1995), that of
ζa2
t(up) replicated the findings of Palley (1992) for the U.S.
The structural change brought about by the Crisis caused a phenomenal change in the way
sectoral mobility affected unemployment. Taking equation (5.1*) with ζgt(up) as an
example, the coefficient estimate during the pre-Crisis period was 4.775 while the post-
Crisis impact, estimated as ∂Ut/∂ζgt(up) = 4.775 + 120.865ζ
gt(up), exceeds that of the pre-
Crisis magnitude. Evaluated at the mean of ζgt(up), it equals 7.500. The onset of the Crisis
led to a much greater influence of mobility movements on unemployment.
The robustness in the result and its concurrence with the empirical literature (in terms of
statistical significance only) seems to point towards the existence of the unpredicted ζ-U
relationship during 1998-2001. For this period, it seems possible to validate the claims of
the SSH for pure mobility purged of demand disturbances.
Pure Sectoral Shifts Purged of Supply Influences
Whilst ζp2
t(up) was insignificant under both equations, ζp1
t(up) was only positive and
significant under equation (4.9a*). Given the lack of robustness in results, not much can be
deduced from the impact of pure shifts purged of supply shocks on unemployment.
Moreover, the insignificant results are in conflict with the outcome for ζp1
t(up) in Mills,
Pelloni and Zervoyianni (1995) for the U.S., but it should be noted that that study covered
the earlier 1961-1991 period, where supply shocks may have been more important.
133
The post-Crisis mobility effect for ζp1
t(up) and ζp2
t(up) showed results similar to those for the
indices purged of AD disturbances, in that both ζp1
t(up)D and ζp2
t(up)D became positive and
significant under both models. In terms of the ζp1
t(up)-U impact, the result is consistent with
Mills, Pelloni and Zervoyianni (1995) for the U.S., although the data periods differ. The
mobility effect was magnified dramatically in the post-Crisis period. The coefficient for
ζp2
t(up) in equation (4.9a*) was 4.908 for the pre-Crisis period but was very much larger
after the Crisis, i.e. ∂Ut/∂ζp2
t(up) = 4.908 + 130.095ζp2
t(up). Evaluated at the mean of ζp2
t(up),
this equals 8.700.
In summary, the post-Crisis findings for Korea suggest that an unpredicted ζ-U relationship
existed. What appears to hold is the following:
a) The impact of unpredicted mobility on unemployment was not felt prior to
1998. This is a common finding across unemployment models with the pure
mobility indices. The finding coincides with results obtained from estimates
based only on 1971-1997 data.
b) Pure sectoral movements purged of demand and supply disturbances seem,
however, to have led to higher aggregate unemployment during the post-Crisis
period. During this period, workers changing sectors will have exacerbated any
unemployment problem. Job replacements may not have been as easy as in the
past as job seekers in this more recent turbulent period will need further time to
acquire skills in the emerging high-skilled jobs, compete with existing workers
with higher productivity, and compete with technology which has made much of
the unskilled labour redundant, i.e. the jobless growth phenomena to be
mentioned in chapter 9.
The claims of the SSH seem valid for the post-Crisis period for Korea, given the robustness
of the results across the various forms of unpredicted indices and the consistency with
related empirical work. However, as highlighted earlier, data limitations in terms of the
low number of observations for the post-Crisis period prevent strong conclusions from
being drawn. The unpredicted ζ-U relationship could be more effectively studied if the
dataset for the post-Crisis period covered a longer time frame.
134
As an added comment, preliminary estimations suggest that much of the unemployment
generated in the post-Crisis period could be non-frictional. As the SSH itself has not been
fully validated, statements about the nature of unemployment generated by pure inter-sector
movements can only be tentative at this stage. For reference purposes, however, a
discussion on the SSH and the natural unemployment rate is provided in Appendix 5H.
5.6.2 Relevance of the ADH
The predicted indices capturing aggregate demand shocks are ζm
t(p), ζgt(p) and ζ
a2t(p). The
former two ADH indices had an insignificant impact on unemployment for equations
(4.9a*) and (5.1*). Thus, from the perspective of mobility predicted from changes in
anticipated money supply and the government deficit to GDP ratio, the ADH is irrelevant to
Korea during the pre-Crisis period. However, whilst the interaction variable of ζm
t(p)D was
positive and significant under both models, the interaction variable of ζgt(p)D gave a
negative and significant result, thereby suggesting that the ADH could only be validated for
Korean mobility arising from changes in the money supply. It also suggests that in the
post-Crisis period, monetary policy would increase the mobility rate as compared to fiscal
policy (via a reduction in the public deficit), which works in the reverse direction.
Where predicted mobility is measured by removing unanticipated deviations in the sectoral
labour movements from changes in aggregate employment [i.e. ζa2
t(p)], the index was
insignificant under equation (4.9a*) but negative and significant in equation (5.1*). Despite
this inconsistency in statistical significance, the fundamental relationship between
unemployment and ζa2
t(p) appears to be inverse - the coefficient in the significant instance is
-34.363 and the point estimate is also negative in the equation with the insignificant
estimate. The inverse relationship continued to prevail during the Crisis period. For
example, from equation (5.1*), the mobility effect on unemployment, i.e. ∂Ut/∂ζa2
t(p) =
-34.363 + 103.779ζa2
t(p), was equal to -31.60 when evaluated at the mean of ζa2
t(p).
It is possible to conclude the lack of relevance of the ADH for Korea in the pre-Crisis
period for predicted mobility arising from changes in the money supply and government
deficit. Though the post-Crisis results indicate a predicted ζ-U correlation, caution must be
exercised in forming conclusions given the limited number of data observations.
135
5.6.3 Applicability of the RTH
Earlier, the issue of the horizon covariance index being a poor measure for the annual data
series was raised. The regression findings appear to confirm this. First, under equations
(5.1*) and (4.9a*), ζH*100 was an insignificant variable. Second, the standardized
coefficient18
for the index (0.033) was one of the smallest values in equation (5.1*) and was
the lowest value in equation (4.9a*) [0.020], signifying it had the least impact on
standardized unemployment. This is not surprising as the index captures the influence of
labour mobility over the preceding two years, which seems too wide an interval to affect
unemployment in the present year. Consequently, interpreting the results of its interaction
variable, ζH*100D, though positive and significant, would be meaningless as we are left
with two data points. The 1998 data point reflects movements of 1996 and 1999, 2000 and
2001 each point towards mobility in 1997, 1998 and 1999. Since the Crisis started
sometime around 1997-1998, only the 2000 and 2001 data points will reflect the Crisis‟
impact. Thus, estimation with the horizon covariance index with an annual data series
appears to be impeded by a major, insurmountable, measurement issue.
Furthermore, the results do not concur with the Davis (1987) study for the U.S. Using
annual data, Davis (1987) reported the coefficient for ζHt-1 to be insignificant. The point
estimate reported was negative, whereas those in this study were positive. It is noted that
the support for the RTH in Davis‟ (1987) study was rooted in regressions using quarterly
data, and for the annual data series for ζHt-3 and ζHt-4 only. Oi (1987) also questioned the
influence of the RTH, since Davis (1987) reported weak correlations.
In short, the RTH cannot be validated for Korea for two reasons. First, the measurement
problem associated with ζH*100 for annual data renders it an unsuitable index, and its
insignificance in the various models estimated appears to confirm this. Second, the
findings for Korea differ from those reported in the empirical literature [Davis (1987)] but
this is primarily due to the general lack of robustness in the results obtained using,
alternatively, annual and quarterly data series.
136
5.6.4 Sectoral Movements and Stage-of-the-Business-Cycle Effect
The regressions of equations (5.1*) and (4.9a*) showed that ζtSt was an insignificant
variable. This finding for the post-Crisis phase in Korea does not concur with that for a
more stable economic setting reported by Mills, Pelloni and Zervoyianni (1995) for the
U.S. Since there are limitations associated with the raw Lilien index, the variable St was
made to interact with other unpredicted and predicted indices.
Pre-Crisis Finding
To assess if the stage-of-the-business-cycle effect applies to unpredicted sectoral
movements, additional interaction variables were created by multiplying St by each
unpredicted index. Each of the five sets of new variables [(i) ζm
t(up), ζm
t(up)St and
ζm
t(up)St D, (ii) ζgt(up), ζ
gt(up)St and ζ
gt(up)St D, (iii) ζ
a2t(up), ζ
a2t(up)St and ζ
a2t(up)St D, (iv) ζ
p1t(up),
ζp1
t(up)St and ζp1
t(up)St D, and (v) ζp2
t(up), ζp2
t(up)St and ζp2
t(up)St D] was entered into the
regressions of equations (5.1*) and (4.9a*) in place of ζt, ζtSt and ζtStD. The regressions
revealed the unpredicted ζSt‟s to be insignificant (Table 5.9), implying that pure inter-
sector labour movements did not lead to higher aggregate unemployment during the period
1971-2001.
The stage-of-the-business-cycle hypothesis was also tested for mobility predicted by
demand disturbances. For the regressions with the ADH indices, each set of variables, i.e.
ζa2
t(p), ζa2
t(p)St and ζa2
t(p)St D, ζm
t(p), ζm
t(p)St and ζm
t(p)St D and ζgt(p), ζ
gt(p)St and ζ
gt(p)St D, was
entered into equations (5.1*) and (4.9a*) instead of ζt, ζtSt and ζtStD2. The interaction
variables ζa2
t(p)St, ζm
t(p)St and ζgt(p)St were insignificant, suggesting that the business cycle
effect is inapplicable during 1971-2001 for predicted labour movements.
Thus, the stage-of-the-business-cycle argument cannot be extended to predicted and
unpredicted sectoral labour movements. As the SSH and ADH did not exist for most forms
of mobility during 1971-2001, it is not surprising that the stage-of-the-business-cycle effect
does not apply to predicted and unpredicted mobility.
137
Table 5.9 Parameter Estimates of ζ, ζSt and/or ζStD Equation
(4.9a*)
Equation
(5.1*) ζt 39.579*
(2.691) 11.868 (0.601)
ζtSt 9.646 (0.955)
-6.115 (-0.712)
ζtStD 61.011* (3.810)
108.882* (4.793)
ζm
t(up) -1.735 (-1.114)
-1.582 (-1.050)
ζm
t(up)St 1.548 (0.626)
-0.208 (-0.095)
ζm
t(up)StD 10.808* (3.824)
13.553* (4.868)
ζg t(up) 21.176
(1.402) 12.803 (0.958)
ζg t(up)St 8.635
(0.565) -8.604
(-0.983) ζ
g t(up)StD 53.955*
(3.319) 88.535* (5.410)
ζa2
t(up) 0.450 (1.005)
-0.012 (-0.024)
ζa2
t(up)St 1.047 (1.237)
0.161 (0.197)
ζa2
t(up)StD 2.638* (2.448)
4.293* (3.692)
ζp1
t(up) 35.263* (3.604)
11.276 (0.914)
ζp1
t(up)St 9.603 (0.960)
-7.632 (-0.956)
ζp1
t(up)StD 29.660* (2.535)
77.691* (5.244)
ζp2
t(up) 26.182 (1.786)
+
-9.153 (-0.546)
ζp2
t(up)St 6.143 (0.646)
-0.890 (-0.105)
ζp2
t(up)StD 36.470* (3.243)
80.124* (5.243)
ζa2
t(p) 2.438 (0.140)
-37.531* (-2.531)
ζa2
t(p)St 14.901 (0.890)
-15.145 (-1.400)
ζa2
t(p)StD 51.966* (2.306)
125.556* (5.063)
ζm
t(p) -0.149 (-0.081)
-0.309 (-0.173)
ζm
t(p)St 4.173 (1.658)
-0.363 (-0.152)
ζm
t(p)StD 4.476 (1.426)
12.244* (3.299)
ζg t(p) -55.588
(-1.377) -40.205 (-0.943)
ζg t(p)St 43.784
+
(1.772) 10.465 (0.383)
ζg t(p)StD -0.732
(-0.024) 96.677
+
(1.859)
* significant at 5% level. + significant at 10% level.
Note: All are OLSE except for equation (4.9a*) with ζa2
t(p),
ζa2
t(up), ζm
t(up) and ζgt(p), and equation (5.1*) with ζ
gt(p) and
ζm
t(up) which are CO estimates applied with a Prais-Winsten
transformation.
138
Post-Crisis Finding
From the results of the predicted and unpredicted ζStD variables, the majority of the indices
point towards a positive and significant coefficient. Although this may suggest some
evidence in favour of the business-cycle effect during the post-Crisis period, two notable
limitations with regards to the dataset prevent the formation of any conclusion pertaining to
the post-Crisis period. The 4 annual data points for the 1998-2001 period are too few
observations to confidently ascertain if the stage-of-the-business-cycle effect really existed.
This is compounded by the fact that at the end of the data period (2001), the most
pronounced business cycle in the data series was incomplete. Any examination of this
hypothesis for Korea can only be conducted over a longer time frame. This is outside the
scope of this thesis.
5.7 CONCLUDING REMARKS
The impact of sectoral mobility on unemployment with regards to the Korean labour
market was examined in this chapter. This examination was conducted from the
perspective of the four main hypotheses: SSH, ADH, RTH, and the stage-of-the-business-
cycle effect. Owing to the contradictory evidence from the empirical studies, the extensive
list of indices and wealth of explanatory variables gathered from the literature, lengthy,
rigorous steps had to be followed to ensure reliable estimates were obtained prior to
statistical inference.
In terms of specification, the approach was to move from an unrestricted to a restricted
model. As an all-encompassing model would have led to over-parameterisation and
multicollinearity, tests were conducted to determine if any of the regressors were
correlated. It was also necessary to ensure the series of each variable was stationary and to
avoid instability in the regression.
With a time period of 31 years, the likelihood of a change in the deterministic relationship
in the variables within each model would be quite high. This is especially so following
from events like the 1998 Asian Financial Crisis, which led to a severe recession in Korea
139
[The World Bank (1999)]. For the majority of models and indices, the tests indicated a
structural break between 1971-1997 and 1998-2001. The identification of these structural
breaks gave rise to a more appropriate functional form with the creation of dummy and
interaction variables. Serial correlation was detected in several estimating equations and
was corrected with the Cochrane-Orcutt iterative method where applicable. The final
estimates gave rise to a better fit of the models and were considered to be valid for
statistical inference.
The non-uniqueness in model specification meant that robustness of the regression results
under alternative models had to be established a priori. Having established robustness in
the results between the two models [i.e. equations (4.9a*) and (5.1*)], it was possible for
conclusions pertaining to the validity of the SSH, ADH, RTH and stage-of-the-business-
cycle to be reached. However, it was not possible to conclude if the RTH applies to Korea,
owing to the problems in using the horizon covariance index for an annual data series, lack
of congruency with the empirical literature, and lack of robustness in the results of the
related empirical literature.
In terms of the pre-Crisis period, there is a general lack of relevance of the SSH, ADH and
stage-of-the-business-cycle effect for the Korean economy. For the post-Crisis period, the
results tend to support the first two hypotheses, but do not support the stage-of-the-
business-cycle effect.
For the post-1997 period, the limited data (4 observation points) seem to provide evidence
in favour of the post-Crisis effect for the SSH and ADH. However, this limitation (i.e.
short span) of the aggregate-level data prevents the full validation of these two hypotheses,
and it is only when more data become available for the post-Crisis period that appropriate
empirical testing will be possible. The implications of the stage-of-the-business-cycle
effect could not be examined effectively owing to the limited data available (i.e. only 4 data
points). Furthermore, it may not be meaningful to examine this hypothesis when the last
data point reflects a mid-point of the most pronounced cycle in the data series.
Nonetheless, since the aggregate-level data findings have indicated that the SSH/ADH
could apply to Korea, and that the nature of unemployment arising from pure sectoral
140
movements could be non-frictional after the Crisis, it could imply that the SSH/ADH are
new phenomena for Korea. What existed for the developed countries much earlier in the
last century appears to have only started for this NIE in recent years19
.
What we would like do in this current thesis, therefore, is extend the research from
aggregate-level data to longitudinal data using the same period of 1998-2001. Part II of the
thesis therefore uses unit-record data to study the factors that motivate inter-sector mobility.
Such knowledge may be useful when seeking solutions to future unemployment problems
through changes to labour mobility.
Endnotes:
1. South Korea will be referred to as Korea hereinafter.
2. The formal indices of sectoral mobility gathered from the literature review are to be analysed later.
3. A part of the reason is that a substantial portion of the ADH discussions has been centred around the U-V
correlation initiated by Abraham and Katz (1986).
4. Additional measures that were examined but not reported on are ζa1
t(up), ζt(s), ζt(r), ζt(p) and its corresponding
unpredicted series, ζt(up). See chapter 3 for a critique of these indices.
5. DMt and UNt were found to be stationary.
6. In the current study, these U.S. variables will not be included because, unlike Canada, the U.S. and Korean
economies are not integrated. To address the possibility of simultaneity bias, checks on the correlation
between DMRt and the other explanatory variables of the unemployment equation will be carried out. These
checks are undertaken as correlation between variables may indicate simultaneity bias, and if overlooked this
may lead to biased estimation [Barrows (2004)].
7. Garonna and Sica (2000) referred to the series as „money growth rate‟. Since it is an aggregate demand
indicator, it is most likely to be the money supply growth for Italy.
8. Barro (1977) indicated that the growth of the public sector at 5% per year (ρ = 0.2) would not seem to be
permanently sustainable. The Korean case of ρ = 0.05 appears reasonable.
9. The term, ζ, represents the generic sectoral mobility index covering all predicted and unpredicted indices.
It does not have a subscript for time t. The term with the subscript for time t, ζt, is the raw Lilien index. The
change variables denoted by the prefix „∆‟ are with respect to the immediate past period.
10. The similarity is based on the type of the variables as the number of lags and difference operators will
differ depending on tests of stationarity and data frequency in the case for Korea. Since annual data are used,
the number of lags for each variable was kept to a minimum. If not, the influence of a variable lagged by, say,
more than one year becomes dated. For the ζ‟s, the indices lagged by one time period were insignificant for
most regressions under equation (5.1). The DMR was lagged by one time period since DMRt was
insignificant in most regressions. It was not lagged by two periods since the annual data series for the
regressions would start from 1972, meaning a further loss of observations and degrees of freedom.
11. The CUSUM and CUSUMSQ techniques, and the associated significance lines, are often viewed as
„yardsticks‟ rather than formal statistical tests. The timing of any structural change is difficult to pinpoint
accurately using this procedure, and the plots need to be examined in association with prior knowledge. The
point at which the structural change commences is often when the plotted line starts to deviate upwards or
downwards.
12. It should be noted that since the absence of a structural break between phase 1 and phase 2 was
established, the Chow test could also have been conducted for (phase 1 plus phase 2) versus phase 3.
However, this is not necessary given the varied findings above. It would be better to proceed straight to the
structural dummy variable approach to ascertain the occurrence and source of structural change.
13. The only study was Parker (1992), where a military variable was incorporated into the unemployment
model to pick up the manpower influences of the Vietnam war.
141
14. A prior test that involved estimating a model with only the mobility dummy first, followed by the
inclusion of the intercept and mobility dummies, showed the statistical significance of the dummies to alter
according to the order of inclusion.
15. This LM statistic is computed as (N-1)R2, where R
2 comes from the auxiliary regression of the estimated
OLS error term on all explanatory variables together with the lagged value of the estimated error term for
(N-1) observations, with N being the total number of observations.
16. The tightening of the models could equally have been done following a correction for serial correlation
without any material change to the findings. The regressions with ζH and ζtSt under equation (4.9a*) and ζH
under equation (5.1*) were corrected for serial correlation. The findings showed that the intercept dummy
and/or mobility interaction dummies remained insignificant under both equations.
17. Greene (2003) conducted a Lagrange Multiplier test with a sample size of 19.
18. The standardized coefficient in the SPSS program expresses the impact of the independent variable in
terms of standard deviation units, i.e., whether the number of standard deviations the dependent variable
increases or decreases with a one standard deviation increase in the independent variable. The standardized
coefficient is calculated by multiplying the non-standardized coefficient by the ratio of the standard deviations
for the independent and dependent variables. A standardized coefficient of a single independent variable in a
multiple regression will assist in determining whether it has a greater or lesser effect on the dependent
variable as compared to the effects of other independent variables.
19. See „The World Bank (1999) „Republic of Korea: Establishing a New Foundation for Sustainable
Growth‟, Report No. 19595 KO, Nov 2, 1999‟.
142
(This page is left blank intentionally.)
143
PART II: THE FACTORS AFFECTING SECTORAL MOBILITY
PREAMBLE
Part I examined, from the perspective of the four hypotheses: SSH, ADH, RTH, and the
stage-of-the-business-cycle effect, the impact of sectoral mobility on unemployment.
Based on aggregate-level data, the key finding was the significant impact that unpredicted
and predicted sectoral mobility had on aggregate unemployment in Korea during the post-
Crisis 1998-2001 period. The four hypotheses, however, had little relevance for periods
prior to this. The mobility-unemployment relationship therefore appears to be a new
phenomenon in Korea, though it has been a characteristic of developed countries like the
U.S. and Canada for earlier periods. There is a need to understand the sectoral mobility
associated with this type of unemployment in Korea. This is the aim of the research
presented in Part II, where the causal factors of sectoral mobility are analysed using
longitudinal data for Korea for the 1998-2001 period.
The literature review in the next three chapters seeks to gather ideas for the current work
from the research undertaken on various forms of labour mobility. Chapter 6 presents
theoretical and conceptual issues in labour mobility and outlines the proposed empirical
framework. Chapter 7 reviews the literature on forms of labour mobility other than sectoral
mobility and extracts salient points for the current research. Chapter 8 conducts a review of
the empirical evidence on the factors affecting sectoral mobility. Finally, chapters 9 and 10
contain the empirical application for Korea, with emphasis on the overall labour force in
the former and separate analyses for males and females in the latter. The key conclusion is
that sectoral mobility is a multi-facetted phenomenon encompassing a range of factors,
including monetary and macroeconomic variables, worker and industry characteristics, and
the sectoral shock.
144
145
CHAPTER 6
THE THEORETICAL AND CONCEPTUAL ISSUES
IN LABOUR/SECTORAL MOBILITY
6.1 INTRODUCTION
This chapter introduces the main theoretical and conceptual issues in the microeconometric
study of labour mobility. The various forms of labour mobility are defined in section 6.2
and the three theories of sectoral/industrial mobility are presented in section 6.3. A generic
theoretical model describing the main motivations behind labour mobility is outlined in
section 6.4. This model provides a framework within which the empirical literature can be
studied (see chapter 8). It also forms the basis for the empirical analyses presented later in
the thesis (see chapters 9 and 10). While the focus of this thesis is on sectoral or industrial
mobility, the empirical literature reviewed covers other forms of labour mobility, including
union/non-union mobility, public-private sector mobility and rural-urban mobility. The
reason for this broad approach is that research into sectoral/industrial mobility appears less
advanced, and hence there may be much to be learned from careful study of the
econometric techniques, databanks and research questions from these other types of labour
mobility. Then, the empirical models used in study of the various forms of mobility are
presented in section 6.5. A summary of the chapter and the implications for the empirical
model are presented in the final section.
6.2 WHAT IS LABOUR MOBILITY?
Labour mobility is a very general term. It can be applied to movement of labour across
countries, across regions within a country, across occupations, industries or broad sectors of
an economy, such as the union and non-union sectors, government versus non-government
sectors and rural versus urban sectors.
There is a vast amount of literature dealing with the movement of labour across countries:
International migration has been a major research issue for most of the last century [Borjas
146
(1994), Bartel (1989), Chiswick (1991), Chiswick and Miller (1985), Chiswick, Le and
Miller (2008) and Dustmann (1993)]. A range of international migration issues have been
examined in Asia. Seok (1999), for example, examined Korea‟s foreign worker labour
immobility during the post-1997 Asian Financial Crisis, and attributed this to the fact that
small and medium-sized firms preferred to hire migrant labour at lower wages, while these
workers remained as their potential gain in earnings from re-migration did not exceed the
costs of returning to their countries of origin. Chew (1990) investigated issues related to
the brain drain in Singapore and highlighted the number of Singapore emigrations in the
1980s. Manning‟s (1999) study focused on implications of the influx of foreign labour into
Singapore from developing countries. Bartram‟s (2000) study highlighted that, in contrast
to other advanced industrial countries with positive migrant inflow, Japan experienced
negative labour migration in the post-World War II period.
Intra-regional migration is also of importance, with researchers attempting to account for
the rise and decline of parts of a country, the growth and demise of regional concentrations
of specific groups of people, and even patterns of settlement within cities [Tomes and
Robinson (1982a), Antolin and Bover (1997), and Fanni, Galli, Gennari and Rossi (1997)].
There have been a number of studies on intra-regional migration and these have
emphasized various patterns. Rogers and Henning (1999), for example, reported that
during the periods 1975-1980 and 1985-1990, foreign-born Americans showed a slightly
higher likelihood of crossing state boundaries than their native-born counterparts. Cutler,
Glaesar and Vigdor (1999) highlighted a trend for black migration during the period 1980-
1996 from the ghettos to cities/suburbs that previously had a predominantly all-white
population. Jeong (2003) showed that wages and large corporate employment raised the
likelihood of regional mobility in Korea over the period 1995-2002.
Occupational mobility is a popular field of study for economists interested in individual
economic well-being. A person‟s occupation offers a good guide to their economic
standing in society, and changes in the individual‟s occupation over time offer useful
insights into their economic progress. Occupational mobility can also be studied at the
aggregate (group) level where a change in the occupational mix over time can help explain
147
why particular groups fare better than others in the job market. For example, if males are
concentrated in trades occupations, and women in services, a shift in the jobs generated in
the economy away from trades towards services would, ceteris paribus, lead to more
favourable labour market outcomes for females than for males. Similarly, if the scope for
productivity gains differs across occupations, knowing how the occupational mix changes
over time will be fundamental to an appreciation of the origins of economic growth, for
example, whether it is so-called jobless growth or is associated with employment growth.
Examples of studies on occupational mobility include Flyer (1997), Kim (1998),
Greenhalgh and Stewart (1985), Miller (1984) and Chiswick, Lee and Miller (2005). Flyer
(1997) reported that the projected earnings was a positive and significant variable in the
initial occupation choice of college graduates. Greenhalgh and Stewart (1985) showed that
British men experienced greater upward mobility and achieved higher occupational status
than women. Miller (1984) presented a model of job matching and occupational choice,
demonstrating that it was optimal for young workers with lesser work experience to switch
occupations. Kim (1998) found that workers who change occupations experienced smaller
wage gains, were less skilled, lower educated and had lower market experience than
workers who do not change occupations. Chiswick, Lee and Miller (2005) found that
although there was a drop in occupational attainment from the last job in the origin to the
first job in the destination for male immigrants in Australia, upward occupational mobility
was possible with post-immigration investments.
As with occupational mobility, labour mobility across broad sectors of the economy
involves individual behaviour which could have implications for the economy. Sectoral
mobility takes various forms. One of the more common types is mobility between union
and non-union sectors [Heywood (1993) and Hahn (1996)]. Other forms of labour mobility
include that between government and non-government sectors [Borland, Hirschberg and
Lye (1998) and Blank (1985)] and rural and urban sectors [Todaro (1981), Zahn (1971) and
Tcha (1993)]. For the latter form of mobility, Zahn (1971) and Tcha (1993) examined the
determinants of worker movements for Japan and Korea, respectively.
Industrial or sectoral mobility is the main topic for the current study. As mentioned in Part
I of the thesis, one of the reasons for this study is that sectoral or industrial mobility is often
associated with structural changes and cyclical movements in the economy. This link has
148
been drawn in a number of studies. Studies associating sectoral mobility and cyclical
variations in unemployment include Abraham and Katz (1986), Blanchard and Diamond
(1989), Brainard and Cutler (1993), Lilien (1982) and Loungani and Rogerson (1989) for
the U.S. labour market, Garonna and Sica (2000) for the Italian labour market, and Prasad
(1997) on industrial mobility for the Japanese manufacturing sector. One point worth
noting is that these studies adopt aggregate-level time-series data. Interest in the individual
behaviour that leads to sectoral (industrial) mobility commenced around the late 1980s, and
this was facilitated by access to unit-record longitudinal data. Studies taking this approach
include Osberg (1991), Osberg, Gordon and Lin (1994), Vanderkamp (1977) for Canada,
Loungani and Rogerson (1989), McLaughlin and Bils (2001), Fallick (1993) and Neal
(1995) for the U.S.
These studies focus on the conventional definitions of economic sectors/industries.
Alternative definitions using micro-level datasets were developed in other studies. For
example, Thomas (1996b) constructed two sectors: (a) pre-displacement sector, which is
the original sector of employment of displaced workers; and (b) the remainder of the labour
market for Canada. Osberg, Mazany, Apostle and Clairmont (1986) categorized sectors as
central or marginal, where the former consisted of the goods-producing primary sector that
used capital intensive technology, including the resource and construction sectors, and the
latter comprised other manufacturing firms not in the central sector and personal services
industries.
The studies above have primarily been concerned with the movement of labour across
economic sectors of the economy. All the types of mobility considered can be examined
within a common framework. This framework is a standard neo-classical model that
depicts individuals as moving from one state (country, region, occupation or
industry/sector) to another if the gains from moving outweigh the costs. These gains and
costs can be either monetary or non-monetary. A model of mobility is outlined and used as
a basis for a more detailed review of the literature in section 6.4. Prior to that, the theories
pertaining to the origins of sectoral mobility will be presented in the section below.
149
6.3 THEORIES OF SECTORAL/INDUSTRIAL MOBILITY
Three theories on the origins of sectoral mobility emerge from the literature, namely, the
worker-employer mismatch theory, the sectoral shock theory and the bridging theory.
These theories are basically about model specification.
6.3.1 Worker-Employer Mismatch Theory
The worker-employer mismatch theory relies on the mismatch between workers and jobs to
generate sectoral mobility. Mobility is modelled as a function of wages and worker/job
characteristics. Workers change sectors if there is a change between their current and
expected circumstances; in the form of higher perceived wages in the new sector and/or
non-pecuniary benefits in new sector and/or or a better job-match between worker
characteristics and new job requirements. Hence, workers could change sectors if the
following matches occur:
a) workers‟ expected wages match with the prospective employers‟ wage offer;
b) workers‟ expectations of the non-wage benefits of the job, e.g. working hours
and benefits, match with the new job characteristics; and
c) workers‟ individual skill sets, e.g. demographic profile, qualification and
experience, match with employer demands and requirements for the job.
Whether one or all of the above matches occur following a sectoral switch really depends
on the individual worker and employer. For example, whilst one worker will switch sectors
if his skill set meets a firm‟s requirements even if the wages do not, another worker will
require that both his skill set and wage levels are in accordance with expectation before a
sectoral change takes place.
As both workers and employers are heterogeneous, the probability of moving to another
sector will differ across workers. It takes time and resources for workers to acquire
information about available job prospects and for employers to acquire information about
potential applicants. Moreover, there is uncertainty about this job information. The theory
150
suggests that workers will seek to maximize their expected wages based on the information
acquired and make a decision on a sectoral switch. Employers will optimally assign jobs to
workers based on the available information about the workers. Optimizing behaviour
within this framework can generate sectoral mobility for some workers and stability for
others. The theory relies on worker heterogeneity and imperfect information in job markets
to generate mobility and is applicable to all forms of sectoral mobility. Many studies of
union/non-union, public-private and rural-urban mobility and the majority of the studies of
sectoral/industrial mobility are based on this theory.
6.3.2 Sectoral Shock Theory
The sectoral shock theory subscribes to the view that sectoral shocks are responsible for
generating sectoral/industrial mobility. A sectoral shock can take the form of changing
tastes, technology, input price, product demand and productivity. Sector-specific shocks
are believed to affect the pattern of labour demand which leads to sectoral reallocations in
the labour market [Helwege (1992) and Clark (1998)]. For example, after a sectoral shock,
the demand for the product of that sector rises, the wages in that sector rise and this attracts
workers from other sectors, thereby generating labour mobility. There are a number of
studies measuring the impact of a sectoral shock on mobility, namely Gulde and Wolf
(1998), Jovanovic and Moffitt (1990), Brainard and Cutler (1993), Altonji and Ham (1990)
and Clark (1998).
Two distinctions must be made between the worker-employer mismatch and sectoral shock
theories. First, whilst the former relies on worker heterogeneity and imperfect markets to
generate mobility, it is implied in the latter that labour movements can occur even when
workers are homogeneous in a perfectly competitive labour market [Clive and Jovanovic
(1988)]. Specifically, in a perfect market, each homogenous worker is deemed to have an
equal probability of changing sectors [Mincer and Jovanovic (1981)] following a sectoral
shock. Second, there are implications pertaining to the empirical application. The sectoral
shock approach could be used if gross flows were equal to net flows. That is, if sectoral
shocks are the only reason for generating mobility, workers move from one specific sector
to another in response to a sectoral shock. Since the sectoral shock theory rules out other
causes of mobility, it implies that gross flows of labour should be equal to net flows. In
151
contrast, under the mismatch theory, mobility occurs owing to reasons other than a sectoral
shock. Workers move across sectors in both directions, and gross flows can be larger than
net flows.
6.3.3 Bridging Theory
Bull and Jovanovic (1986) argued that labour mobility may be caused by shifts in the
derived demand for labour on the part of firms/sectors and by mismatches between workers
and jobs. Furthermore, Jovanovic and Moffitt (1990) stated that a model that relies solely
on the impact of sectoral shocks on labour demand or concentrates on sector/worker
mismatch is likely to lead to misinterpretation of the empirical results. Hence, the
“bridging” theory subscribes to the view that labour mobility can be modelled in two ways:
via a shift in labour demand and also generated by employer-employee mismatch. This
bridging view was mooted by Clive and Jovanovic (1988) in theory only. The study that
tested this theory was Jovanovic and Moffitt (1990), where wages, worker characteristics
and a sectoral shock variable were incorporated into the mobility equation. Since mobility
is also generated by an employer-employee mismatch, it operates in the presence of
worker/employer heterogeneity and imperfect job markets. Each worker faces an unequal
probability of a sectoral switch.
6.4 MODEL OF LABOUR MOBILITY
The model outlined below is developed as a tool to explain worker movement from one
sector to another. The model has as its starting point the approach taken by Le and Miller
(1998). They developed a model of labour market choice incorporating an individual‟s
current and future earnings streams as well as the non-pecuniary aspects of alternative
employment states. This is a conceptual advance over other models of labour market
choice that are based only on the differential in current earnings associated with alternative
employment states.
Let yai(t) represent the annual earnings of an individual i in sector „a‟ in period t, and ybi(t)
be the annual earnings of the individual in sector „b‟ for period t. The lifetime earnings of
this individual in sectors „a‟ and „b‟ would each be:
152
T T
Yai = ∫ yai(t)e-rt
dt and Ybi = ∫ ybi(t)e-rt
dt 0 0
where r is a discount rate that is constant across individuals.
If the individual aims to maximize the net present value of their lifetime wealth, then they
will choose to move to sector „a‟ if Yai – Ybi – Ci > 0, where Ci reflects the difference in
non-pecuniary aspects and any non-recoverable costs of moving between sectors. This
decision rule may be approximated by ln Yai – ln Ybi – ci > 0, where ci is the cost of shifting
sector (and differential in non-pecuniary benefits) normalized by the earnings in sector „b‟.
This model may be rendered empirically tractable by using Willis and Rosen‟s (1979)
specification for the earnings generation process. This incorporates current earnings and
initial earnings in a simple geometric growth model, namely,
( ) ( ) aig t
ai aiy t y t e dt and ( ) ( ) big t
bi biy t y t e dt ,
_ _
where yai and ybi are the individual‟s initial earnings in sectors „a‟ and „b‟, and gai and gbi
are the growth rates of earnings in these two sectors, with r > gai, gbi. Over an infinite time
horizon,
∞ __
Yai = ∫ yai(t)e-rt
dt will be equal to Yai = yai / (r - gai) and
0
∞ __
Ybi = ∫ ybi(t)e-rt
dt equal to Ybi = ybi / (r - gbi).
0
When considering the choice of sector of employment, it is useful to work with this model
in the context of a discrete choice framework. Hence define an index function
Ii = ln Yai - ln Ybi - ci . (6.1)
The individual is assumed to choose sector „a‟ where Ii 0 and sector „b‟ where Ii < 0. By
_ _
expressing the index function as Ii = ln yai - ln ybi – ln(r-gai) + ln(r-gbi) - ci, and applying a
153
Taylors series expansion for ln(r-gai) and ln(r-gbi) around the mean values of the arguments,
the following expression may be derived:
1 2 3 4[ln ln ]i ai bi iai biI y y g g c . (6.2)
where the βs are the parameters that will be estimated to show how initial earnings in each
sector and growth rates in earnings in these sectors affect the underlying index that is used
to determine the choice of sector.
The index function could also be expressed in terms of current earnings and growth rates,
_ _
using the fact that ln yai(t) = ln yai + gait and ln ybi(t) = ln ybi + gbit. Thus,
public-private sector mobility while section 7.4 reviews rural-urban mobility. The final
section presents a summary of findings of relevance to the current empirical work.
7.2 UNION VERSUS NON-UNION MOBILITY
Union/non-union mobility refers to worker movements from the original non-union (union)
sector to the union (non-union) sector1. The rationale is that wealth maximizing individuals
will join the union sector if expected wages in that sector exceed current wages in the non-
union sector, ceteris paribus. As it is anticipated that the union wage will be above market-
clearing levels, there will be involuntary or wait unemployment, and it is the expected
rather than the actual wage in the union sector that will enter into the worker‟s calculations.
The studies on union choice generally adopt a model similar to that of equation (6.7),
except that the component of expected wages has not been factored into the worker‟s
calculations2.
There are numerous empirical studies focused on the determination of union choice, some
of which include Christie (1992), Borland and Ouliaris (1994), Sharpe (1971) and Kenyon
and Lewis (1997) for Australia, Booth (1983), Bain and Elsheikh (1976) and Carruth and
167
Disney (1988) for the U.K., and Neumann and Rissman (1984) and Farber and Saks (1992)
for the U.S. These studies generally relate union choice to monetary, macroeconomic and
non-pecuniary variables.
Table 7.1 outlines the main features of these studies, providing the data source, data-type,
coverage, model specification, method of estimation and relevant findings from the studies.
With regards to the data-type, whilst studies with micro cross-sectional data have generally
included a monetary variable (e.g. sectoral wage advantage) and a wide range of non-
pecuniary factors, the time series analyses with aggregate-level data have focused on the
macroeconomic factors and included lagged dependent variables. The approach to model
specification therefore depends to a certain extent on the type of data available.
The specification of variables is of particular interest to the current research. The effect of
monetary influences is generally captured by the earnings differential between the union
and non-union sectors. This component of the model corresponds to the current wage
advantage in the models of Todaro (1981) and Le and Miller (1998). An assessment of
whether cyclical fluctuations account for union/non-union mobility is generally made by
including the macroeconomic variables of unemployment, prices, wages and employment
rates in the union choice equations. These economic variables are generally included in
time-series analyses either as lagged independent variables or are differenced to the first-,
second- or third-order. This is in line with an earlier paper by Shister (1953), who argued
that both the rate and pattern of economic change were possible causes of unionization.
The influence of lagged dependent variables (i.e. union membership in previous time
periods) was also considered in several studies dealing with aggregate-level data [Sharpe
(1971), Booth (1983), Kenyon and Lewis (1990), Carruth and Disney (1988) and Borland
and Ouliaris (1994)]. There are several reasons for including the lagged dependent
variable. The “saturation effect” suggests that it is more difficult to increase trade union
membership in already highly unionized sectors owing to resistance from the remaining
non-unionised workers [Sharpe (1971) and Booth (1983)], and hence union density in
preceding periods might be expected to exert negative influences on current membership.
Another reason is that owing to reporting delays, union membership in preceding periods
might reflect some of the membership numbers for the current period [Carruth and Disney
(1988)].
168
Table 7.1 Selected Studies of Union/Non-Union Mobility
Study Source/Country/Time
Period, Data-Type and
Coverage
Dependent Variable,
No. of Regressors and
Explanatory Variables
Method of Estimation
and Relevant Findings
Christie (1992)
Source/Country/Time Period Australian National Social Science Survey, 1984. Data-type Unit-record cross-sectional data. Sample of population. Coverage 1,316 full-time and part-time wage earners aged 18 years and over from all Australian states.
Dependent Variable: Probability of union membership. No. of Regressors: 8. Explanatory Variables: Monetary: union/non-union
wage differential.
Socio-economic:
educational qualification,
experience, industry and
occupation.
Demographic: marital
status, sex and state.
Method of Estimation: Logit model. Relevant Findings: Workers are likely to join unions if the expected wages are higher. Males, diploma holders, experienced workers and those in Tasmania have a higher probability of joining unions. Marital status did not have an influential effect on the probability of union membership. Workers from agriculture, manufacturing, construction, wholesale trade, finance and public administration are less likely to join unions. Professionals, administrators, clerical, sales and service workers have a lower chance of union membership.
Farber and Saks (1980)
Source/Country/Time Period Individual votes from National Labor Relations Board elections, U.S., Jan 1972-Sep 1973. Data-type Unit-record cross-sectional data. Random sample of workers from 29 establishments in various industries. Coverage 817 union and non-union workers who were asked to participate in the vote, i.e. whether they preferred to join a union job or not.
Dependent Variable: Probability of an individual voting for a union job. No. of Regressors: 12. Explanatory Variables: Monetary: individual‟s position in intra-firm earnings distribution. Socio-economic: seniority, education, indicators for union causing relationship deterioration, union causing fairness improvement, chances for promotion, difficulty of finding job (DIFF), dissatisfied with job security (DS) and interaction variable (DIFF*DS). Demographic: race, sex, location and age.
Method of Estimation: Probit model. Relevant Findings: Workers who are at the lower end of the intra-firm earnings distribution, feel that they are unfairly treated, feel that chances for promotion in the non-union sector are not good, find difficulty in replacing jobs and are dissatisfied with job security are more likely to vote for unionization. Blacks are more likely to vote for unionization but older workers are not. Seniority, sex, education and location had little impact on the vote. The effects of demographic factors were controlled for in the study.
Borland and Ouliaris (1994)
Source/Country/Time Period Australian union membership data, 1913-1989. Data-type Aggregate-level time-series data. Coverage Total Australian workforce.
Dependent Variable: Change in union membership. No. of Regressors: 6. Explanatory Variables: Macroeconomic: employment in manufacturing and non-manufacturing sectors, UR in period t - UR in period t-2 and RW in period t-1 - RW in period t-3. Lagged dependent variable: union density in period t-1 and union density in period t-3.
Method of Estimation Engel and Granger (1987) method of co-integration using an error correction model. Relevant Findings: Employment in manufacturing and non-manufacturing have significant positive and negative impacts, respectively, on union membership. Unemployment and real wages showed a negative impact on union membership. An increase in union membership in previous periods increases union membership in the current period.
169
Table 7.1 Selected Studies of Union/Non-Union Mobility (continued)
Study Source/Country/Time
Period, Data-Type and
Coverage
Dependent Variable, No.
of Regressors and
Explanatory Variables
Method of Estimation
and Relevant Findings
Kenyon and Lewis (1997)
Source/Country/Time Period The data period covers 1948 to 1995. Data are from the Australian Bureau of Statistics‟ publications, including Trade Union Members and Trade Union Statistics. Data-type Aggregate-level time-series data. Coverage Total Australian workforce.
Dependent Variable: Change in union membership. No. of Regressors: 9. Explanatory Variables: Macroeconomic: RW in period t-1, UR in period t-1, employment in union sector in periods t and t-1, female employment in period t-1 and government employment in period t. Lagged dependent variable: union membership - total civilian employment in period t-1. Political: political dummy variable (1 = Labor Party in power, 0 = otherwise), Accord dummy variable (1 = during 1983-1990, 0 = otherwise) and dummy variable for post-1990 period.
Method of Estimation: OLS. Relevant Findings: Real wages had a positive effect on union membership. Any change in union employment in periods t and t-1, and government employment showed a positive influence. A change in female employment and a net increase in union membership over total employment in the previous period had a negative impact on union membership. Whilst the presence of the Labor party raised union membership, the Accord did not. The addition of a post-1990 dummy variable caused a negative shift in union membership. The unemployment rate had an insignificant effect on union membership.
Sharpe (1971) Source/Country/Time Period The data period covers 1907-1969. Data on union membership, unemployment and real wages obtained from the Labour Report of the Bureau of Census and Statistics. Employment data from the Australian Economic History Review and Yearbook of Commonwealth of Australia. Data-type Aggregate-level time-series data. Coverage Total Australian workforce.
Dependent Variable: Annual growth in trade union membership. No. of Regressors: 5. Explanatory Variables: Macroeconomic: growth in employment in the union sector, UR and RW in period t-1. Lagged dependent variable: Ratio of union membership to employment in period t-1. Political: dummy variable for institutional factors.
Method of Estimation: OLS. Relevant Findings: An increase in union sector employment leads to an increase in union membership. Real wages had an insignificant effect on unionization. The ratio of union membership to total employment in the previous period and the overall unemployment rate had negative effects on union membership. Institution factors exerted a positive impact.
Carruth and Disney (1988)
Source/Country/Time Period The data period covers 1896 to 1984 and are obtained from the following publications: U.K. Department of Employment (DE) Gazette (various issues), DE surveys of Trade Union membership and from the Census of Employment. Data-type Aggregate-level time-series data. Coverage Total British workforce.
Dependent Variable: Change in union membership. No. of Regressors: 10. Explanatory Variables: Macroeconomic: employment, employment, differential between wages and price in period t-1, UR, UR and union membership – employment in period t-1. Lagged dependent variable: union membership in previous periods t-1, t-2, and t-3. Political: dummy variable for political climate (0=Conservative government in power, 1=non-Conservative government).
Method of Estimation: OLS. Relevant Findings: Results are extracted from the model based on real wages after incorporating a dummy variable for political climate. Union membership has a positive effect on membership growth for up to period t-1. Real wages and unemployment have negative effects on the incentive to unionise. Any deceleration/acceleration to the change in unemployment has an offsetting effect. A positive change in employment and the presence of a non-Conservative government raises union density. Union membership net of employment in the previous period had an insignificant effect.
170
Table 7.1 Selected Studies of Union/Non-Union Mobility (continued)
Study Source/Country/Time Period,
Data-Type and Coverage
Dependent Variable, No.
of Regressors and
Explanatory Variables
Method of Estimation
and Relevant Findings
Booth (1983) Source/Country/Time Period
The data period covers 1895 to 1980. The data are obtained from the U.K. Census of Population and the following publications: Employment Gazette, The British Economy: Key Statistics and U.K. Annual Abstract of Statistics. Data-type Aggregate-level time-series data. Coverage Total British workforce.
Dependent Variable: Logistic transformation of union density1. No. of Regressors: 6. Explanatory Variables: Macroeconomic: price inflation, wage inflation and UR in periods t and t-1. Lagged dependent variable: percentage of union membership to the total workforce in periods t-1 and t-2.
Method of Estimation: OLS. Relevant Findings: Union membership for periods t-1 and t-2 exerted positive and negative impacts on union membership, respectively. The unemployment rate in the current period reduced union membership but the same variable in period t-1 tended to increase membership. Price inflation was an insignificant explanatory variable but wage inflation displayed a direct relation with union membership.
Bain and Elsheikh (1976)
Source/Country/Time Period As in Booth (1983). Data-type Aggregate-level time-series data. Coverage Total British workforce.
Dependent Variable: % change in union membership. No. of Regressors: 5. Explanatory Variables: Macroeconomic: prices, wages, and unemployment in periods t-1 and t-2. Lagged dependent variable: union density in period t-1.
Method of Estimation: OLS. Relevant Findings: Changes in prices and wages, and unemployment in period t-2 exerted positive effects on union membership. Union density and unemployment in period t-1 showed negative effects.
Neumann and Rissman (1984)
Source/Country/Time Period The data covers the period 1904-1980, obtained from the U.S. Bureau of Labor Statistics‟ Handbook of Labor Statistics, Wolman (1936) and Troy (1965). All sources are based on membership figures reported by unions. Data-type Aggregate-level time-series data. Coverage Total U.S. workforce.
Dependent Variable: % unionised. No. of Regressors: 11. Explanatory Variables:
Macroeconomic: inflation
rate, employment,
employment in periods t-1,
t-2 and t-3, UR and %
unemployed in periods t-1
and t-2.
Socio-economic: welfare in
period t (depicted by
government expenditure on
social welfare as a %GNP),
% representation elections
won by unions and %
demographic representation
in Congress.
Method of Estimation: OLS. Relevant Findings: Higher inflation increases union
membership. The unemployment rate
and change in employment in period
t-1 showed positive effects. Whilst
the change in employment in period t
had a negative impact, the change in
employment in periods t-2 and t-3
had insignificant effects. The %
unemployed in period t-1 exhibits a
positive impact but there is an
insignificant impact for the same
variable for period t-2. Social welfare
benefits reduce the attractiveness of
union membership. The higher the
percentage of representation
elections won by unions, the higher
the % unionised. The percentage of
demographic representation in
Congress had an non-influential
impact on union membership.
1. Derived as Z = ln [D/(1-D)]t where D is union density with a one time period lag. Annotation: UR denotes unemployment rate, RW denotes real wages.
denotes change in the variable between two time periods. denotes a second-order change in the variable between time periods. For example, yt = (yt – yt-1) = (yt – yt-1) – (yt-1 – yt-2).
171
A range of non-pecuniary influences has been examined in the union choice literature.
Included are the demographic and socio-economic composition of the labour force, in terms
of the age, race, sex, marital status, education, industry, occupation, seniority and
employment status of workers. Several studies have also added what Shister (1954) termed
as a “proximity influence”, which can be viewed as either physical proximity (e.g. rural-
urban-suburb, state), an employer- or employment-specific factor (e.g. relationship with
supervisor, fairness treatment of employee, promotional prospects, job security) or political
proximity, e.g. influence of political party on the union as in Kenyon and Lewis (1997) and
Carruth and Disney (1988).
The empirical findings from this body of research are also of interest, as they can show the
success or otherwise of this approach to modelling. The union/non-union wage differential
was found to have a positive and significant effect on the union choice decision by Christie
(1992). Farber and Saks (1980) went a step further to add a threshold point - workers at the
lower end of the intra-firm earnings distribution (earning less than $0.21/hour above the
infra-firm mean earnings of $0.49/hour) were more likely to join unions. This is consistent
with suggestions in the union literature that unions represent the political interests of lower-
income and disadvantaged persons. For example, see the discussion of the collective voice
“face” of unions in Freeman and Medoff (1984). These and the other studies demonstrate,
therefore, that monetary incentives can be modelled successfully when analyzing worker
mobility.
The studies that have examined the impact of economic variables on the rate of
unionization, however, have produced conflicting results in relation to the possible impact
of unemployment, level of employment and real wages. The relationship between
unemployment and union membership was negative in Borland and Ouliaris (1994), Sharpe
(1971) and Carruth and Disney (1988), but positive in Neumann and Rissman (1984),
Ashenfelter and Pencavel (1969) and Freeman (1989), and insignificant in Kenyon and
Lewis (1997). In the studies by Booth (1983) and Bain and Elsheikh (1976), the
unemployment rate lagged by different time periods also exhibited conflicting results.
Unemployment exerted a negative influence for the current period in Booth (1983) and for
period t-1 in Bain and Elsheikh (1976). The unemployment rate for period t-1 and that for
period t-2 tended to be associated with increased membership in Booth (1983) and Bain and
172
Elsheikh (1976), respectively. In part, the conflicting empirical evidence may reflect the
ambiguous nature of the theoretical predictions. On the one hand, it has been argued that
higher unemployment raises union density because unions are able to increase job security.
On the other hand, if unions are seen as a source of higher unemployment owing to their
wage-setting powers, the incentive to unionise will decline during periods of high
unemployment3.
Similarly, changes in the level of overall employment were found to have a positive effect
on union density in Carruth and Disney (1988), but a negative bearing in Neumann and
Rissman (1984). It should be noted that Sharpe (1971) considered a sectoral breakdown for
the employment variable, with the inclusion of employment in the union sector, which was
found to have a positive impact on union membership. However, as the number of studies
using sector-specific economic indicators is few, no firm conclusions can be formed.
Theoretically, the association of real wage with union choice is indeterminate. It has
generally been argued that this association will be negative, as decreases in wages that lead
to worker dissatisfaction should increase the desire to unionise. However, it is also possible
that workers might join unions to defend real wage gains so that real wages and
unionization will be positively correlated. Given these competing views, it should come as
little surprise that the empirical findings on the union density – real wage relationship are
mixed. Whilst Borland and Ouliaris (1994) and Carruth and Disney (1988)4 concluded that
real wages had a negative effect on union density, Kenyon and Lewis (1990) and Peetz
(1990)5 reported a positive impact. Real wages were found to have an insignificant effect
on unionization in the study by Sharpe (1971).
It was earlier argued that workers seek to maximize their economic wealth and so will join
unions if the perceived union wages are higher. This, however, is tied to a ceteris paribus
assumption that needs to be accommodated in empirical work. Worker characteristics differ
and each labour market exhibits different characteristics. The studies reviewed in Table 7.1
take account of such factors, although the empirical evidence is not always conclusive. For
example, sex and education were found to be significant explanatory variables in Christie‟s
(1992) model for Australia, but not in the study undertaken by Farber and Saks (1980) for
the U.S.
173
There are, however, limitations to these union choice models which the current study
should attempt to overcome. In particular, there appears to be a fundamental oversight
which could explain the conflicting results for business cycle variables. Earlier, it was
observed that sectoral labour movements are determined, in part, by unemployment in
sector „a‟ versus sector „b‟. In the union choice studies, however, an overall unemployment
rate variable is used. That is, there is no distinction as to whether the pool of the
unemployed is generated from the union sector or the non-union sector. This same line of
argument applies to the real wage and employment variables, where empirical studies fail
to differentiate between union/non-union employment and real wages. Kelly and
Richardson (1989) and Booth (1983) have also expressed doubts concerning the
explanatory power of estimating equations based on business cycle models. Sharpe (1971)
also indicated that a disaggregated sectoral unemployment rate might help to explain trade
union growth. The exclusion of these apparently appropriate explanatory variables could
lead to model misspecification, giving rise to misleading results. The empirical review in
chapter 8 addresses this issue6.
7.3 PUBLIC VERSUS PRIVATE SECTOR MOBILITY
Public-private sector mobility is another form of labour mobility7. Both of these sectors are
associated with different characteristics and influences. Whilst the private sector tends to
follow principles of profit-maximisation or cost-minimisation, the public sector is more
often subject to other social, political and non-economic influences. Workers with differing
personal characteristics have differing probabilities of choosing public versus private sector
employment, as they seek the job (or sector) where their specific set of characteristics will
receive the highest rewards. The public-private sector divide resembles more closely the
setting that will be used in the empirical work to be undertaken in this thesis, in that
movement between sectors for public and private sector workers generally involves a
greater set of changes than does movements between sectors for the union and non-union
workers examined above.
Many of the models of public-private sector mobility have a structure that is quite similar to
the empirical model of equation (6.7). Accordingly, this research has included both the
public-private sectoral wage differential and worker/job characteristics as explanatory
174
variables in the estimating equations used. Some examples include Borland, Hirschberg
and Lye (1996) for Australia, Blank (1985), Gyourko and Tracy (1988), Long (1975), Long
(1976) and Utgoff (1983) for the U.S., Hartog and Oosterbeek (1993) for the Netherlands
and van der Gaag and Vijverberg (1988) for Cöte d‟Ivoire.
Table 7.2 overviews the major approaches and findings from research into public-private
sector mobility. The public-private sector studies selected use unit-record cross-sectional
data and incorporate a wealth of monetary and non-pecuniary factors as regressors. The
exception, in this context, is Utgoff (1983) who uses aggregate-level cross-sectional data
and with a smaller number of regressors. However, unlike union/non-union studies with
aggregate-level time-series data, there do not appear to be any studies in the public-private
sector with macroeconomic factors and lagged dependent variables.
The explanatory variables in the private-public sector selection models consist of the
sectoral earnings differential and non-monetary factors. The latter comprises personal
(IQ), veteran status, years and levels of education, past school in non-English speaking
country, school attended, whether born in Asian/non-English speaking country, year of
arrival in Australia, field of study, reading, writing and arithmetic skills, geographic region,
age of youngest child and whether in capital city) and job characteristics (occupation, firm
size, experience and blue-collar versus white-collar status). In addition to these personal
and job characteristics, Hartog and Oosterbeek (1993) added social background indicators,
such as number of siblings, father‟s occupation and father and mother‟s education.
175
Table 7.2 Selected Studies of Public-Private Sector Mobility Study Source/Country/Time
Period, Data-Type and
Coverage
Dependent Variable,
No. of Regressors and
Explanatory Variables
Method of Estimation and Relevant
Findings
Borland, Hirschberg and Lye (1996)
Source/Country/Time Period Australian Bureau of Statistics (ABS) Training and Education Experience Survey 1993. Data-type Unit-record cross-sectional data. Sample of employees. Coverage 5,969 males and 3,376 females aged 15-64 years who were employed full-time as wage and salary earners.
Dependent Variable: Probability of selecting a public sector job. No. of Regressors: 13. Explanatory Variables: Demographic: age, marital status, age of youngest child and whether in capital city. Socio-economic: education level, age minus age left school, experience-squared, year of arrival in Australia, last school attended in non-English speaking country, field of study, whether born in Asian/non-English speaking country and state of residence.
Method of Estimation: Separate probit models for male and female employees. Relevant Findings: Males: The more experienced males residing in Victoria, South Australia (SA) and Western Australia (WA), who have attended a school in a non-English speaking country, arrived in Australia between 1964-1967, 1972-1975, 1986-1987 and 1990-1991, and who have studied Trade Qualification (TQ) in vehicle and food, Post-School Certificate (PSC) in science, computing and agriculture will have a lower likelihood of choosing the public sector. Those who have degrees in law, education, medicine, mathematics, IT, veterinary science, engineering, social sciences and TQ in electricals and electronics, arts, social sciences and crafts, are more likely to choose the public sector. Females: Women with longer job tenures, who are residing in NSW, Victoria, Queensland, SA and WA who arrived in Australia between 1984-1985 have a greater probability of choosing the private sector. Those with children between 0-2 years who have completed degrees in law, education and the social sciences, PSC in education, teacher training, nursing, other health and para-medical, and who arrived in Australia between 1968-1971 and 1972-1975 are more likely to choose the public sector instead.
Blank (1985) Source/Country/Time Period U.S. 1979 Current Population Survey (CPS). Data-type Unit-record cross-sectional data. Random sub-sample of the CPS, i.e. one-fourth of employed heads of households. Coverage 10,908 employed heads of households, of whom 8,344 are in the private sector and 2,564 are in the public sector.
Dependent Variable: Probability of individual being a private sector worker. No. of Regressors: 6. Explanatory Variables: Demographic: sex, race and geographic region. Socio-economic: veteran status, occupation, education level and experience.
Method of Estimation: Probit model. Relevant Findings: For the non-monetary variables, veterans, non-whites, higher-educated persons, workers in services and those in Washington D.C. and with more experience have a higher probability of choosing the public sector. Women showed no statistically distinguishable preference.
176
Table 7.2 Selected Studies of Public-Private Sector Mobility (continued) Study Source/Country/Time
Period, Data-Type and
Coverage
Dependent Variable,
No. of Regressors and
Explanatory Variables
Method of Estimation and Relevant
Findings
Hartog and Oosterbeek (1993)
Source/Country/Time Period Individuals from the Dutch province of Noord-Brabant obtained from addresses in the city population register of the Netherlands in 1983. Data-type Unit-record cross-sectional data. Sample of population. Coverage Males and females from 2,726 addresses in a single province.
Dependent Variable: Probability of selecting a public sector job. No. of Regressors: 8. Explanatory Variables: Monetary: public-private sector wage differential. Demographic: sex (female). Socio-economic: social background (no. of siblings, father‟s occupation, education of father and education of mother), personal characteristics (IQ, education level).
Method of Estimation: Endogenous switching regression model. Relevant Findings: Variables related to social background were unimportant in the determination of public sector employment, except for father‟s education which showed a positive effect. For personal characteristics, vocational and university graduates are more likely to work in the public sector. The higher the IQ, the lower the probability of the individual working in the public sector. Females were less likely to become public servants. The likelihood of public sector employment is higher the larger the predicted wage gain in the public sector.
van der Gaag and Vijverberg (1988)
Source/Country.Time Period Cöte d‟Ívoire Living Standards Survey (CILSS), 1985. Data-type Unit-record cross-sectional data. Sample of households. Coverage 513 wage earners from 1,600 households.
Dependent Variable: Probability of obtaining public sector job. No. of Regressors: 6. Explanatory Variables: Monetary: public-private sector wage differential. Demographic: sex, age and age-squared. Socio-economic: indicators for diploma at elementary, high school, higher and technical diplomas, reading, writing and arithmetic (RRR) skills and years of schooling.
Method of Estimation: Probit model. Relevant Findings: Women are more likely than men to be employed in the public sector. Age (up to 50 years) shows a positive effect on public sector employment. Elementary and high-school diplomas increase the likelihood of a public sector job. Higher and technical diplomas, years of schooling and RRR skills have insignificant effects. The sectoral wage differential is not significantly different from zero.
Gyourko and Tracy (1988)
Source/Country/Time Period U.S. 1977 CPS. Data-type Unit-record cross-sectional data. Sample of population. Coverage Full-time wage earners.
Dependent Variable: Probability of selecting private/union or private/non-union or public/union or public/non-union sector. No. of Regressors: 8. Explanatory Variables: Demographic: marital status, race, gender and region of residence (northeast, central, south and west). Socio-economic: veteran status, seniority status (junior and senior), level of college (1st, 2nd, 3rd and 4th year) and graduate status.
Method of Estimation: Multinomial logit model. The model had 4 distinct labour markets: private/union, private/non-union, public/union and public/non-union sectors. Relevant Findings1: Veterans, juniors,
graduates, those attending higher levels
of college education (3rd-4th year) and
lived in the western region had a higher
chance of selecting a public union/non-
union job. There was no distinct
preference for a public union/non-
union job versus private union/non-
union job for workers who are white,
married, male, seniors, have lower
level of college education (1st and 2nd
year) and lived in the northeast, central
and south regions.
177
Table 7.2 Selected Studies of Public-Private Sector Mobility (continued) Study Source/Country/Time
Period, Data-Type and
Coverage
Dependent Variable,
No. of Regressors and
Explanatory Variables
Method of Estimation and Relevant
Findings
Long (1975) Source/Country/Time Period
U.S. 1970 Census of Population. The reference year is 1969. Data-type Unit-record cross-sectional data. 1-in-1,000 public use sample of the 1970 census data. Coverage 40,578 males aged 14 years and over, of which 3,886 were black.
Dependent Variable: Probability of public sector employment. No. of Regressors: 6. Explanatory Variables: Demographic: indicators for males aged 14 years and over, males aged 18-34 years, workers in the southern region and non-southern region. Socio-economic: white-collar workers and blue-collar workers.
Method of Estimation: Linear probability model. Relevant Findings: The study concentrates on sectoral differences in employment for blacks relative to whites. The probability of public sector employment was higher for black white-collar and blue-collar workers, black males aged 14 years and over, and those aged 18-34 years. In both southern and non-southern regions, blacks were relatively more likely to be employed in the public sector rather than the private sector.
Long (1976) Source/Country/Time Period U.S. 1970 Census of Population. The reference year is 1970. Data-type Unit-record cross-sectional data. 1-in-1,000 public use sample of the 1970 census data. Coverage Male and female employees.
Dependent Variable: Probability of Federal Employment. No. of Regressors: 4. Explanatory Variables: Demographic: marital status (married and single) of workers. Socio-economic: indicators for white-collar workers, and occupation (professionals, managers and administrators).
Method of Estimation: Linear probability model. Relevant Findings: Females are less likely to be employed in the public service. Specifically, females who are white-collar workers, professionals, administrators and managers tend to be under-represented in the public service. Marriage has a negative impact on the probability of public employment among females, while being single had a positive but insignificant effect.
Utgoff (1983) Source/Country/Time Period U.S. 1972 Bureau of Labor Statistics‟ (BLS) data and 1972 Census of Manufactures. Data-type Aggregate-level cross-sectional data. Sample of the population (for BLS data on quit rates). Coverage Government employees.
Dependent Variable: Probability of quitting the public sector. No. of Regressors: 2. Explanatory Variables: Monetary: average hourly earnings. Socio-economic: firm size.
Method of Estimation: OLS. Relevant Findings: Larger firm size and higher average hourly earnings had a negative effect on the probability of quitting the public sector.
1. The Gyourko and Tracy (1988) study had four labour market choices: public union, public/non-union, private
union and private non-union. Since the focus is on the choice between two labour markets (public versus
private sector), the findings presented reflect the most significant result which will be independent of
union/non-union choice.
As argued previously, it would be expected that higher wages in the public sector would
induce wealth maximizing individuals to seek employment in that sector. Hartog and
Oosterbeek (1993) found that the larger the predicted wage gain in the public sector, the
higher the likelihood of public sector employment. In Borland, Hirschberg and Lye (1996),
public sector male and female employees had higher wages than their counterparts in the
private sector, implying that the higher-paid public sector would attract individuals to seek
employment in that sector. However, it has also been argued that there may be non-wage
178
benefits, e.g. job stability, working hours and fringe benefits, that are generally not
considered in the statistical analyses, and the presence of which mean that workers may
prefer the public sector even if monetary wages are higher in the private sector. In the case
of the Ivorian market, the sectoral earnings differential did not have a significant impact on
the choice of sectoral employment. Thus, with the exception of the Ivorian labour market,
the evidence for public-private sector mobility sits comfortably alongside that for
union/non-union choice models. Hence, emphasis can be placed on the estimated impact of
the sectoral wage advantage in both the union choice equation and the public sector/private
sector model.
The findings on education levels were consistent for all the studies represented in Table 7.2,
except for Borland, Hirschberg and Lye (1996). Specifically, public sector choice tended to
be associated with higher education levels, even though the specification of the education
variables differed across studies. In particular, Blank (1985) found that higher-educated
persons in the U.S. had a higher probability of choosing the public sector. Hartog and
Oosterbeek (1993) demonstrated that university graduates in the Netherlands were more
likely to work in the public sector than in the private sector. van der Gaag and Vijverberg
(1988) also found that public sector employees were, on average, better educated than
private sector employees for the Ivorian labour market. Gyourko and Tracy (1988) found
that graduates and persons with higher levels of diplomas had a greater likelihood of
choosing a public sector job. This is not surprising as the public sector has been perceived
as recruiting better educated persons to assist in the planning of policies and evaluation of
programmes.
It has been argued that certain groups, e.g. non-whites and women, may have a higher
probability of choosing the public sector. Non-whites may choose the public sector as the
work practices are less discriminatory. Women may prefer the public service given that the
work environment there is more family friendly, with practices facilitating intermittent
labour market attachment. The findings on the racial divide were consistent for the U.S.
Blacks and other non-whites were found to have a higher probability of choosing public
sector employment in the three studies conducted for the U.S. for the different periods of
analysis [Blank (1985), Long (1975) and Gyourko and Tracy (1988)]. For Australia,
179
being born in an Asian country did not have a significant impact on public-private sector
choice [Borland, Hirschberg and Lye (1996)].
The findings on the role of gender on public sector choice were inconclusive. Whilst van
der Gaag and Vijverberg (1988) and Gyourko and Tracy (1988) found that women were
more likely than men to choose public sector employment, Hartog and Oosterbeek (1993)
and Long (1976) reported that females were less likely to become public servants. Blank
(1985), however, reported that being female had a statistically insignificant impact on
public sector choice. Apart from the fact that the role of women at work varies in
importance across different countries, a possible reason for the mixed evidence could be a
failure to take adequate account of the composition of female employment by occupation,
industry etc. For instance, if the public sector in a country had a higher proportion of
clerical personnel compared to professionals, and if such workers are predominantly
female, then it could be expected that women, on the whole, would have a higher chance of
securing public sector employment, particularly if insufficient account is taken of
occupational structure in the estimations.
A final point to note in relation to the public-private sector selection models is that sectoral
unemployment does not appear to have been recognized in the analyses. This may be due
to data limitations. While datasets are available that contain information on the type of
work (e.g. occupation/industry) that individuals are seeking, and hence facilitate the
estimation of unemployment models for different sectors, this information may not be
available in the datasets used for the studies reviewed in Table 7.2.
7.4 RURAL-URBAN MOBILITY
Rural-urban mobility (or migration) is another type of labour mobility that has been
researched extensively8. Todaro (1969) and Harris and Todaro (1970) hypothesized that
rural-urban mobility is stimulated primarily by rational economic considerations of relative
benefits and costs, and it is mainly the expected urban-rural income differential that will
influence the individual‟s decision to move9. The studies on rural-urban mobility have
mainly been conducted for developing countries with a predominantly agrarian population,
180
i.e. Schultz (1971) for Columbia and Ghatak (1996) for India, as well as NIEs with a
substantial share of rural workers, namely Tcha (1993) for Korea and Zahn (1971) for
Japan. The relevance of these studies to the current analysis lies in the emphasis placed on
the monetary and non-pecuniary determinants.
Table 7.3 summarises the main findings of several studies conducted for the U.S., South
America and Asia. With the exception of Ghatak (1996), the authors have included the
sectoral wage differential as well as macroeconomic (e.g. overall unemployment and
economic growth rate), demographic (e.g. urban-rural population ratios, rural labour supply
and the age-sex distribution of the population) and a rich array of socio-economic (e.g.
education, violence indicators and travel time to city) elements in the rural-urban mobility
function.
These rural-urban studies use aggregate-level data. Whilst the aggregate-level time-series
analyses have incorporated macroeconomic variables, those with cross-sectional data, i.e.
Ghatak (1996) and Schultz (1971), have not. The absence of macroeconomic determinants
for cross-sectional analyses was also observed in the union/non-union and public-private
sector studies. This arises as cross-sectional studies are unable to track the consequences of
a structural change in the macroeconomy. In addition, as in the case of public-sector
studies, the rural-urban studies generally have a fewer number of regressors (6 or less)
when aggregate-level data are applied.
The primary explanatory variable in these studies is the rural-urban wage differential10
.
According to Todaro‟s model and the model outlined in chapter 6, there should be a clear
positive relationship between the wage differential and worker mobility, especially when
the probability of obtaining work in the urban sector is taken into account. However, this
clear theoretical prediction is not reflected in the empirical literature. The studies that are
consistent with the Todarian hypothesis, and report a positive relationship between the
urban-rural income ratio/differential and rural-urban movements, include Tcha (1993) for
the U.S. and Zahn (1971) for Japan. The sectoral wage differential was insignificant in the
study by Ghatak (1996). In contrast, Tcha‟s (1996) findings for the Korean labour force
did not support Todaro‟s hypothesis: he found that Korean villagers were willing to
181
sacrifice higher rural wages for lower incomes in return for better living conditions in urban
areas, provided the expected urban wages were at least 75 per cent of the original rural
income.
Table 7.3 Selected Studies of Rural-Urban Sector Mobility Study Source/Country/Time
Period, Data-Type and
Coverage
Dependent Variable, No.
of Regressors and
Explanatory Variables
Method of Estimation and
Relevant Findings
Tcha (1993)
Source/Country/Time Period Annual data of U.S., 1960-1987. Data-type Aggregate-level time-series data. Sample of population. Coverage U.S. migrants.
Dependent Variable: Net rural-urban migration rate1. No. of Regressors: 4. Explanatory Variables: Monetary: dynastic rural-urban income ratio. Macroeconomic: real growth rate of the economy and overall unemployment rate. Demographic: ratio of the rural population to urban population.
Method of Estimation: OLS (log-linear). Relevant Findings: Dynastic income ratio and economic growth rate had significant positive effects on rural-urban migration. Effects of the overall unemployment rate and ratio of the rural population to urban population were insignificant.
Schultz (1971)
Source/Country/Time Period Columbia, 1951 and 1964. Data-type Aggregate-level cross-sectional data. Sample of 131 Columbian municipalities drawn from 1951 and 1964 Population Census data. Coverage Males and females aged 7-51 years.
Dependent Variable: Net migration rate for rural population2. No. of Regressors: 5. Explanatory Variables: Monetary: rural wage. Socio-economic: school enrolment for children aged 5-9 years and 10-14 years, frequency of political violence and distance to travel to the next city. Demographic: growth rate of rural labour supply.
Method of Estimation: OLS. Relevant Findings: Rural wage has a negative effect on out-migration. An increase in the growth rate of the rural labour supply accelerates out-migration. School children aged 10-14 years are more likely to move to urban areas than those 5-9 years. The effect of greater distance to the next city spurs migration. An increase in rural violence encourages persons to move to urban areas.
Tcha (1993) Source/Country/Time Period Annual data of Korea, 1963-1988. Data-type Aggregate-level time-series data. Sample of population. Coverage Korean migrants.
Dependent Variable: Net rural-urban migration rate1. No. of Regressors: 4. Explanatory Variables: Monetary: dynastic rural-urban income ratio. Macroeconomic: real growth rate of the economy and overall unemployment rate. Demographic: ratio of the rural population to urban population.
Method of Estimation: OLS (log-linear). Relevant Findings: For the income variable, people are willing to move to urban areas until the expected dynastic income is 75% of the rural income. Economic growth rate had a positive effect on migration and the unemployment rate had a negative effect. The result for the ratio of the rural population to urban population was insignificant.
182
Table 7.3 Selected Studies of Rural-Urban Sector Mobility (continued)
Study Source/Country/Time
Period, Data-Type and
Coverage
Dependent Variable, No.
of Regressors and
Explanatory Variables
Method of Estimation and
Relevant Findings
Zahn (1971) Source/Country/Time Period
The data covers the period 1878 to 1937. Agricultural and industrial labour, real output and working age population are obtained from Ohkawa (1957). The capital stock series are estimated from Ohkawa‟s (1957) capital stock estimate and Rosovsky‟s (1961) savings data. Population data are from the Bank of Japan. Data-type Aggregate-level time-series data. Coverage Males and females aged 14 years and over.
Dependent Variable Industrial-agrarian labour force ratio. No. of Regressors Demand equation: 2. Supply equation: 2. Explanatory Variables:
Demand Equation
Industrial-agrarian required labour ratio is expressed as a function of the socio-economic (industrial-agrarian capital stock ratio) and macroeconomic (real output ratio and technical progress) factors.
Supply Equation
Industrial-agrarian labour force ratio is expressed as a function of the monetary (expected urban-rural income ratio) and demographic (an index of the age-sex distribution of the population) factors.
Method of Estimation: Simultaneous equation model using 2-stage least squares estimation. Relevant Findings: Demand Equation The industrial-agrarian real output ratio and technical progress have positive effects on the industrial-agrarian labour ratio. The industrial-agrarian capital stock ratio had a negative effect. Supply Equation An increase in the actual urban-rural wage ratio leads to out-migration. A higher number of working age persons and females both induce out-migration.
Ghatak (1996)
Source/Country/Time Period Census of Population 1971 and 1981 obtained from the Statistical Abstract of the Indian Union. Data-type Aggregate-level cross-sectional data. Coverage Rural and urban population for all Indian States.
Dependent Variable 3 variables: size/growth rate/density of urban population (UP). No. of Regressors: 1. Explanatory variables: Monetary: estimated rural-urban income differential.
Method of Estimation: OLS. Relevant Findings: For these 3 regressions, a higher urban-rural wage differential does not appear to induce out-migration.3
1. The rate is calculated using the actual and expected rural population (RP) data. The expected RP in period t is
calculated by multiplying RP in period t-1 (RPt-1) by the natural population growth rate allowing for births and
deaths (δt). Subtracting the actual RPt from RPt-1(1+ δt) gives net rural-urban migration.
2. Net migration rate is defined as the ratio of a net migration flow in the rural sector to the average size of the
local population. A negative migration rate means a net out-migration from the rural sector, and conversely for
a positive migration rate. 3. According to Ghatak (1996), several factors, i.e. moving costs, expected wages, skill levels, risk-taking
behaviour of individuals and borrowing and liquidity constraints, were not taken into consideration, and this could explain why the Todarian hypothesis was not supported.
The conflicting findings in relation to the estimated impact of the monetary variable could
be due to two factors. Firstly, as in the public-private sector studies, differences exist
between rural-urban work environments, individual preferences, culture of country etc., and
the effects of the monetary element are therefore not expected to be the same for studies
183
conducted in different contexts. Secondly, different measurements of the variable have
been used. Ghatak (1996), for example, computed the rural-urban income ratio using the
estimated agricultural and industrial incomes for India (in Rupees) at current prices, while
Zahn (1971) used the actual urban-rural income differential. In comparison, Tcha (1993)
calculated a dynastic income ratio of the weighted average of blue-collar and white-collar
incomes in the urban area to the income in the rural area, where the weights were chosen
iteratively and were related to time and altruistic discount rates between generations.
Schultz (1971) only considered the push factor of the rural wage as a determinant of
mobility.
The dynastic income ratio used by Tcha (1983) suggests that the wage differential should
account for altruism between generations which has multiplicative effects on a family‟s
decision to migrate. The average of blue-/white-collar incomes was also used in the
dynastic measure as migrants from rural areas typically have insufficient physical and
human capital, and are more likely to use the urban blue-collar income as their expected
income. This type of dynastic income measure is not relevant to analysis of
movements could cause multi-generational family migration, sectoral/industrial mobility is
usually independent of family migration.
The rural-urban mobility models do not have a sectoral breakdown of the unemployment
rate, unlike that suggested for the current empirical model. Such a breakdown is important
theoretically as sectoral movements could prevail in the presence of higher unemployment
in the new sector where migrants would be underemployed in the informal sector.
Underemployment is, however, difficult to measure, and this measurement problem may
explain the practice (of omitting unemployment rate variables) in applied work. In
comparison, the inclusion of the sectoral unemployment rates in the empirical work is
likely to be important in the current research, and it is practical to include relevant measures
in the estimating equations.
184
Several aspects of the rural-urban migration analyses are of value to the current study.
First, the use of the sectoral wage differential is notable and this variable can be considered
for inclusion in the current work. Second, the use of sectoral performance indicators in
rural-urban studies is an approach that can be followed. In the study of rural-urban
migration, individuals are argued to move from the lower-growth sector to the rapidly
growing sector. The variables indicative of sectoral growth comprise the industrial-
agrarian labour force, capital stock and real output [Zahn (1971)] and growth rate of rural
labour supply [Schultz (1971)]. Whilst the industrial-agrarian labour force ratio acts as a
demand-pull factor, where a higher ratio (e.g. from technology shock) causes urban wages
to rise and pulls people to migrate to urban areas, higher rural labour supply acts as a
supply-side factor causing rural unemployment and pushing people to migrate. For
sectoral/industrial mobility, it would be the growth (declining) sectors that induce workers
to move to (out of) their sectors.
7.5 SUMMARY: SALIENT POINTS FOR EMPIRICAL MODEL
The review of the literature in this chapter has highlighted the following points which
should inform the empirical work to be undertaken in chapters 9 and 10.
a) The determinants of sectoral mobility should be modelled within a framework
comprising a sectoral wage differential and the macroeconomic and non-
monetary factors associated with mobility.
b) A sectoral distinction in the macroeconomic variables, especially on
unemployment, is desirable.
c) The non-pecuniary determinants should also be measured on a sector-by-sector
basis where possible.
d) The type of data to be used affects model specification. The longitudinal data
that are to be used have the advantage that macroeconomic and lagged
dependent variables, which have been demonstrated to be significant
determinants of labour mobility, can be incorporated into the estimating
equation.
185
Endnotes:
1. It is recognized that the worker movement can be from the non-union sector to the union sector or from the
union sector to the non-union sector. For ease of exposition the discussion here is in terms of the former flow
of workers.
2. The sole exception to this appears to be Gyourko and Tracy (1988).
3. See Borland and Ouliaris (1994).
4. Carruth and Disney (1988) incorporated a dummy variable for political climate: 1 when the
Labour/Liberals were in power and 0 for the presence of a Conservative government. The initial regression
in the absence of the political dummy revealed that the real wage had a negligible effect on union
membership.
5. Peetz (1990) presented evidence on workers in the manufacturing sector. Those who experienced a decline
in real wage in the previous two years had a higher desire to unionise.
6. The exception applies to changing price levels, which were found to be directly related to union
membership in Neumann and Rissman (1984), Carruth and Disney (1988) and Bain and Elsheikh (1976). A
possible explanation for the non-conflicting result in this instance is that, unlike real wages and
unemployment, prices are not sector-specific. All individuals/workers, regardless of whether they are union
or non-union members, face similar price levels. It is noted that for Carruth and Disney (1988), the findings
reported are obtained from their nominal inflation model.
7. Private-public sector labour flows do occur, but for ease of exposition the review examines the public-
private sector mobility.
8. While there may be urban-rural labour flows, most literature addresses the more important rural-urban
flow, and this is the focus of this section.
9. Several authors, like Mincer (1978) and Borjas (1990), have questioned the hypothesis that migration
behaviour can be explained solely by the individual‟s income. Mincer (1978) examined decision making
within the family unit, particularly the effect of interactions between husband and wife on the probability of
migration. Borjas (1990) considered the welfare of children as a determinant in the migration function.
10. Earlier studies prior to Todaro [Jorgensen (1967) and Ranis and Fei (1964)] examined rural-urban
mobility using the wage differential.
186
CHAPTER 8
EMPIRICAL EVIDENCE:
FACTORS MOTIVATING SECTORAL/INDUSTRIAL MOBILITY
8.1 INTRODUCTION
This chapter reviews the empirical evidence on the factors motivating sectoral/industrial
mobility. These factors include monetary and macroeconomic characteristics, worker and
job characteristics, and sectoral shocks. Various labour markets are covered. Moreover,
where possible, results for both males and females, as well as for the overall labour force,
are reviewed.
The chapter categorises the studies according to labour mismatch, sectoral shock and
bridging theories. Section 8.2 gives a general introduction to the concept of
sectoral/industrial mobility and introduces the studies covered. Section 8.3 outlines the
impact of explanatory variables under the labour mismatch theory. These variables include
monetary and macroeconomic factors and worker and job characteristics. Section 8.4
focuses on the effects of a sectoral shock on mobility under the shock theory. The
implications from a single study based on the bridging theory are covered in section 8.5.
The impact of the explanatory variables on overall (i.e. both male and female workers)
mobility will first be reviewed. This will be followed by an examination of gender
differences in sectoral labour market outcomes. An assessment of the empirical studies of
sectoral mobility for the purpose of empirical modelling is given in section 8.6. A
summary of the findings is provided in the final section, together with suggestions on the
applicability of the explanatory variables to the current empirical study of Korea.
8.2 SECTORAL/INDUSTRIAL MOBILITY
Sectoral/industrial mobility is the main form of labour mobility of interest to the current
research. It is a complex matter involving a spectrum of factors in the individual‟s decision-
making process, and the costs and benefits involved are usually of far greater importance
than those that need to be considered in union/non-union or intra-sectoral mobility. In the
187
latter forms of mobility, wealth-maximising individuals select jobs similar to their former
jobs/sectors so that much of their human capital can be transferred to the new sector in
return for wage gains. Consequently, the costs of moving, and the wage gains required to
induce mobility, will usually be relatively minor. In comparison, the costs and barriers to
entry in inter-sectoral mobility are greater. As sector-specific skills may not be easily
transferable to other sectors, skills relevant to the new sector will need to be acquired, and
this means the investment costs necessary to facilitate the move may be considerable.
Moreover, limited market knowledge of the new sector may act as a barrier to entry
[Subrahmaniam, Veena and Parikh (1982) and Gallaway (1965)]. There may also be
psychic costs to moving that are of greater importance than in intra-sectoral mobility, e.g.
uncertainty about prospects in the new sector which pose as artificial barriers to entry
[Greenwood (1975), Gallaway (1965) and Vanderkamp (1977)]. These real and artificial
barriers to entry constitute a further cost that workers need to take into account in their
choice of sector/industry.
These issues are prominent in empirical studies of sectoral mobility. Under the labour
mismatch theory, the main studies of sectoral/industrial mobility are Osberg (1991),
Osberg, Gordon and Lin (1994) and Vanderkamp (1977) for Canada, Loungani and
Rogerson (1989), McLaughlin and Bils (2001) and Brainard and Cutler (1993) for the U.S.,
Prasad (1997) for Japan and Jayadevan (1997) for India. These studies covered employed
workers, where sectoral mobility rates of around 13% [U.S. males in Jovanovic and Moffitt
(1990) and Canadian employees in Osberg, Gordon and Lin (1994)] have been reported.
Inter-industrial movements are higher for the unemployed (about two-thirds of the
unemployed who gained employment changed their industry of employment in Thomas
(1996b)1 and Neal (1995)
2), but there are only a small number of studies of their behaviour.
The main contributions are Fallick (1993), Thomas (1996b), Neal (1995) and Kim (1998)
for the U.S.,3 and Ottersen (1993) for Sweden.
The studies of the sectoral shock hypothesis cover a range of countries, including Brainard
and Cutler (1993) and Clark (1998) for the U.S., Gulde and Wolf (1998) for the European
Union (France, Italy, Germany and Spain) and Altonji and Ham (1990) for Canada. The
sole study based on the bridging hypothesis was undertaken by Jovanovic and Moffitt
(1990) for the U.S.
188
8.3 DETERMINANTS UNDER THE MISMATCH THEORY
The aim of this section is to review the determinants covered in the studies of sectoral
mobility under the mismatch theory. Table 8.1 provides a selection of the relevant
empirical literature. The explanatory variables under the bridging theory in Jovanovic and
Moffitt (1990) are also presented so that the variables (excluding the sectoral shock) can be
compared with those under the mismatch theory. These studies are chosen as they cover a
wide range of the monetary, macroeconomic and non-pecuniary factors that appear to be
directly related to the current work. Additionally, the studies cover the main worker
groups, namely, the overall workforce as well as males and/or females. This is relevant to
the separate analyses undertaken later for males and females in the Korean labour market.
The determinants include a sectoral distinction for the monetary variable (e.g. wage
differential between the old and new industries or wages in the original and/or wages in the
new industries, industry per worker real wage growth rate and sectoral wages relative to
total wages) and macroeconomic factors (overall GNP and employment, employment in the
old and new industries, unemployment in the old and new industries and the ratio of
employment in the old industry to that of the new industry). Some of the macroeconomic
factors, e.g. average and real GNP growth, overall employment and unemployment and
unemployment duration, do not have a sectoral breakdown. Most of these variables were
analysed from the perspective of the ways they affected overall mobility. However, the
analysis of the impact of the overall unemployment rate and duration of unemployment
spell was extended to male and female mobility, and the analyses of the effects of wages in
the old/new sector and relative sectoral wages were undertaken for male mobility.
The non-pecuniary factors include worker and job characteristics. The worker
characteristics consist of demographic factors (sex, age, age of entry and marital status) and
socio-economic characteristics (education status, language ability, unionization, head of
household status, having children, skill levels, e.g. white-collar job, job tenure and
employment status). These characteristics are fixed for the worker and so there is no need
to consider sector-specific measures. The examination of the impacts of marital status,
tenure, employment status, occupational status, initial industry and working hours was
carried out for male/female mobility, and that of formal education for male mobility.
189
The job characteristics comprise working hours/weeks, training, industry, industry
performance, occupation, industry size, turnover, output, whether the industry provides
unemployment insurance/social benefits, product/work similarity4, type of job loss (e.g.
advance notification, due to slackness or shift in position) and male-female mix in the
industry. These variables tend to be sector- or industry-specific, and there are quite
considerable variations across industries in these factors. Consequently, it is expected that
most variables will exert greater influence on sectoral mobility than on other forms of
worker mobility. In addition, sector-specific shocks which are believed to affect certain
economic sectors causing sectoral mobility were included in many studies.
These studies either cover the employed or unemployed. As employed workers have greater
access to information markets, i.e. networks in other sectors and capital markets, and face
greater opportunity costs in changing sectors [Pissarides and Wadsworth (1989)], it is
expected that the personal characteristics and market conditions that affect their mobility
will differ from those for the unemployed. Where possible therefore, the empirical
determinants of mobility for these two groups need to be assessed separately.
It was earlier highlighted that several studies examine the determinants of sectoral mobility
separately for males and/or females. However, Osberg (1991) appears to be the only study
that compares the inter-industry mobility patterns of male and female employees. Gender
comparisons regarding the determinants of industrial mobility are therefore limited to the
explanatory variables covered in his study. Many studies focus on male mobility behaviour,
including Osberg (1991), Osberg, Gordon and Lin (1994), and Jovanovic and Moffit (1990)
for the employed, and Neal (1995), Fallick (1993) and Thomas (1996b) for the
unemployed. Other studies have covered either the overall workforce [Vanderkamp
(1977), Loungani and Rogerson (1989) and Ottersen (1993)], industry establishments
[McLaughlin and Bils (2001) and Jayadevan (1997)] or sectoral employment [Prasad
(1997), Gulde and Wolf (1998), Altonji and Ham (1990), Brainard and Cutler (1993) and
Clark (1998)].
190
Table 8.1 Probability Choice Studies of Sectoral/Industrial Mobility
under Worker-Employer Mismatch Theory Study Source/Country/Time-
period, Data-type and
Coverage
Dependent Variable, No. of
Regressors and Explanatory
Variables
Method of Estimation and Relevant
Findings
Studies of Employees
Osberg (1991) Source/Country/Time Period Labour Force Survey, Statistics Canada, 1980/1981, 1982/1983 and 1985/1986. Data-type Longitudinal data. Stratified random sample of households.
Dependent Variable Probability that a worker changed industry of employment between Sep in year t and Feb in year t+1. No. of Regressors: 11. Explanatory Variables: Macroeconomic: unemployment rate and weeks unemployed. Socio-economic: Worker characteristics: initial industry, initial occupation, education status (years of schooling) and part-time worker status. Work characteristics: change in usual weekly hours, job tenure and job tenure squared. Demographic: Age and marital status (single = 1, 0 otherwise).
Method of Estimation: Separate regressions are run for males and females using a logit model. Relevant Findings: Unemployment did not exert any influence for males in the three periods; its effect for females during 1980/1981 and 1985/1986 was negative. Education status and marital status had non-influential impacts on mobility. Single and higher-educated men had higher industrial mobility only in 1980/1981. Age was an insignificant variable except for its negative influence on women in 1985/1986. Job tenure showed negative effects for both men and women in all time periods. For job tenure squared, positive effects were exhibited in all 3 periods except in 1985/1986 for females. This means that as job tenure increases, its positive influence increases at a less than linear rate. Male workers in the construction industry were more likely to leave the industry in all 3 time periods, and those in manufacturing and resources had higher mobility rates in 1982/1983. Females in construction and government had higher mobility rates in 1985/1986, and those in manufacturing, trade and finance, utitlies and transport exhibited higher mobility rates in 1982/1983. The occupational status effect was insignificant for men except for managers/professionals/technicians in 1980/1981, who were less likely to change sectors. Among women, higher mobility rates were seen by those in personal services in 1980/1981 and 1982/1983, those in clerical services in 1980/1981 and managers/professionals/technicians in 1982/1983. Women in all occupational groups displayed lower incidences of mobility in 1985/1986. The longer the number of weeks unemployed, the higher the probability of mobility for males and females for all periods. The changes in working hours did not have a significant impact except for males in 1982/1983 and 1985/1986. Except for females in 1980/1981, part-time workers are more likely to change industry.
191
Table 8.1 Probability Choice Studies of Sectoral/Industrial Mobility
under Worker-Employer Mismatch Theory (continued) Study Source/Country/Time-
period, Data-type and
Coverage
Dependent Variable, No. of
Regressors and Explanatory
Variables
Method of Estimation and Relevant
Findings
Studies of Employees
Vanderkamp (1977)
Source/Country/Time Period Canadian Insured Population for years 1965/1966, 1966/1967 and 1967/1968. Data-type Unit-record, cross-sectional data. Sample of Insured Population. Coverage 4,692 employees.
Dependent Variable Proportion of moves from industry i to j. No. of Regressors (linear model): 25. Explanatory Variables: Monetary: wage in original industry (Yi) and wage in new industry (Yj). Macroeconomic: unemployment in original industry (Ui) and unemployment in new industry (Uj). Socio-economic: Worker characteristics: formal education in original industry (EDi), education in new industry (EDj), (EDiEDj)
1/2, unionization in original industry (CAi), unionization in new industry (CAj), (CAiCAj)
1/2, change in occupation, change in province of employment and change in occupation and province of employment. Work characteristics: industry size (Pj), industry turnover and dummy variables for product similarity, location similarity and work similarity. Demographic: age of entry in original industry (Ai), age of entry in new industry (Aj), (AiAj)
1/2, male-female specialization in the original industry (Fi), male-female specialization in the new industry (Fj) and (FiFj)
1/2.
Method of Estimation: OLS for linear mobility model. Relevant Findings from Linear Model: For 1965/1966 and 1966/1967, the lower the wage in the original industry, the higher the likelihood of mobility. Conversely, higher wages in the new industry induce industrial mobility for all three periods. Higher unemployment in the original industry encourages industrial mobility. Unemployment in the new industry did not have a significant effect for 1966/1967 and 1967/1968. The larger the industry size and turnover, the higher the probability of mobility. Education, age of entry and male-female specialisation in the new and original industries displayed negative effects on industrial mobility. However, the effects of the coefficients on (EDiEDj)
1/2, (AiAj)1/2 and
(FiFj)1/2 were positive. Unionization in the
old and new industries had negative effects on mobility except for unionization in the new industry in 1965/1966. The effect of the coefficient on (CAiCAj)
1/2 was positive for 1965/1966 and 1966/1967. The change in occupation and change in occupation and province indicators had negative effects on industrial mobility. However, the change in the province of employment had a positive impact on mobility. The 3 dummy variables for product, location and work similarity showed positive and significant effects on inter-industry mobility.
Osberg, Gordon and Lin (1994)
Source/Country/Time Period 1986-1987 Labour Market Activity Survey (LMAS) extracted from the Labour Force Survey, Statistics Canada. Original interview conducted in Jan/Feb 1987 on labour market activities with a re-interview concerning activities in 1987 being conducted in Jan/Feb 1988. Data-type Longitudinal data. Stratified sample of households. Coverage Prairie and Atlantic male employees aged 16-69 years with hourly wages in both surveys. Initial sample is 8,570 males, out of which 1,095 changed industries.
Dependent Variable Probability of inter-industry mobility. No. of Regressors: 14. Explanatory Variables: Monetary: wage differential. Macroeconomic: no. of weeks unemployed. Socio-economic: Work characteristics: index of job availability, desire for more working weeks per year and desire for more work hours. Worker characteristics: education qualification (elementary, post-secondary, diploma, university), language (French speaking indicator), job tenure, received unemployment insurance indicator, received training in 1986 indicator, received social assistance indicator and used CEC1 in 1986 indicator. Demographic: age (16-19 years, 20-24 years and 25-34 years) and marital status (married).
Method of Estimation: Bivariate probit model of simultaneous choice between 3 states: immobility, inter-regional and inter-industry mobility during 1987. Relevant Findings: The wage differential did not exert any influence on inter-industry mobility. The greater the availability of jobs and the shorter the job tenure, the higher the probability of inter-industry mobility. Persons aged 16-19, 20-24 and 25-34 years, desiring a higher number of working weeks per year and more working hours per week, with a post-secondary education, with longer duration of unemployment and who received unemployment insurance and those who received training in 1986 showed a higher incidence of inter-industry mobility. French speakers, married persons, those with elementary, diploma or university education, those who used CEC and received social assistance displayed insignificant effects on industrial mobility.
192
Table 8.1 Probability Choice Studies of Sectoral/Industrial Mobility
under Worker-Employer Mismatch Theory (continued) Study Source/Country/Time-
period, Data-type and
Coverage
Dependent Variable, No. of
Regressors and Explanatory
Variables
Method of Estimation and Relevant
Findings
Studies of Employees
Jovanovic and Moffitt (1990)
Source/Country/Time Period National Longitudinal Survey of Young Men, U.S., 1968-1981. Data-type Longitudinal data. Survey sample of males. Coverage Male employees aged 14-24 years were interviewed in 1966 and who were 29-39 years in 1981 at the last interview. There are a total of 9,963 observations: 492 (1965-1968), 754 (1968-1970), 628 (1967-1969), 887 (1969-1971), 1,357 (1971-1973), 1,846 (1973-1975), 2,032 (1976-1978) and 1,967 (1978-1980).
Dependent Variable Probability of a sectoral move. No. of Regressors: 2. Explanatory Variables: Monetary: standard deviation of wage distribution. Socio-economic: Worker characteristics: Job experience (5 years, 8 years and 11 years). Aggregate Disturbance term: sectoral shocks.
Method of Estimation 1. Probit mobility equation estimated separately for each year as a function of education, experience, experience-squared, and race. Only the fitted probabilities at 5, 8 and 11 years of job experience are shown for each regression estimated for the years 1968 to 1973, 1975, 1978 and 1980. 2. The probability of a sectoral move was regressed on the standard deviation of the log (wage distribution) and standard deviation of sectoral shocks at 5, 8 and 11 years of job experience. Relevant Findings: 1. The probability of a sectoral move decreased for all experience levels (5 years, 8 years and 11 years of experience). Mobility fell much faster from 1968 to 1973 than from 1973 to 1981. 2. The larger the standard deviation in wages, the higher the probability of a sectoral move for all experience groups. The sectoral shock had a positive impact on the probability of sectoral mobility for workers with 5 and 8 years of work experience.
Loungani and Rogerson (1989)
Source/Country/Time Period U.S. Michigan Panel Study of Income Dynamics (PSID) 1974-1984. Data-type Longitudinal data. Sample of population. Coverage Workers in the labour force covering 26 industries with 8 time periods (208 observations).
Dependent Variable: Proportion of permanent industry switchers. 3 regressions were estimated for sectoral mobility: all sectoral switchers, from goods to services sector, and from services to goods sector. No. of Regressors: 3 Explanatory Variables: Macroeconomic: average real GNP growth between periods t and t+1 and real GNP growth in period t+2. Socio-economic: Worker characteristics: skill-mix (proportion of individuals in skill-intensive industries).
Method of Estimation: OLS. Relevant Findings: The average real GNP growth exerted a negative effect on the proportion of industry switchers from the goods to services sector and a positive effect for workers switching from the services to goods sector. Its impact in the regression for all industry switchers was insignificant. The higher the real GNP growth, the lower the proportion of industry switchers from the goods to services sector and for all industry switchers. The effect on movements from services to the goods sectors was not significant. The skill mix did not have a significant influence on the proportion of industrial switchers.
McLaughlin and Bils (2001)
Source/Country/Time Period U.S. Bureau of Labor Statistics Survey of Establishments, 1964-1995. Time period for business & repair services, personal services and other professional services is 1972-1995. Data-type Panel data. Sample of establishments weighted to represent the aggregate. Coverage U.S. establishments in 22 industries.
Dependent Variable Natural logarithm of the proportion of industry‟s employment to aggregate employment. No. of Regressors: 2. Explanatory Variable: First difference of the natural logarithm of aggregate employment and the time trend variable.
Method of Estimation: OLS. Relevant Findings: Employment fluctuations in construction and all durable manufacturing industries are more than twice the size of aggregate employment fluctuations. Industries that exhibit cyclical movements in employment that are less than half the size of aggregate employment include agriculture, food and tobacco, communication and utilities, public administration, and several service industries.
193
Table 8.1 Probability Choice Studies of Sectoral/Industrial Mobility
under Worker-Employer Mismatch Theory (continued) Study Source/Country/Time-
period, Data-type and
Coverage
Dependent Variable, No. of
Regressors and Explanatory
Variables
Method of Estimation and Relevant
Findings
Studies of Employees
Jayadevan (1997)
Source/Country/Time Period Annual Survey of Industries published by Central Statistical Organisation (CSO) for the years 1973/1974 to 1979/1980, 1980/1981 to 1990/1991. Data-type Panel data. Sample of establishments weighted to represent the aggregate. Coverage Indian establishments in 18 manufacturing industries.
Dependent Variable Growth rate of employment in industry. No. of Regressors: 2. Explanatory Variables: Macroeconomic: output growth rate and real wages per worker growth rate.
Method of Estimation: OLS. Relevant Findings: Industries with higher output growth experienced higher employment growth for the two time periods and industries with higher growth in per worker real wages had lower employment growth for the 1980/1981 to 1990/1991 period.
Studies of the Unemployed
Neal (1995) Source/Country/Time Period U.S. Displaced Workers Survey (DWS), 1984/1986/1988/1990. The DWS was supplemented with the Current Population Survey. Data-type Unit-record, cross-sectional data. Sample of population. Coverage Unemployed males aged 20-61 years at survey dates.
Dependent Variable: Probability of switching industries. No. of Regressors: 12. Explanatory Variables: Macroeconomic: unemployment spell. Socio-economic: Worker characteristics: original industry employment/employment growth, experience, experience-squared, tenure, tenure-squared, years of schooling and occupation. Demographic: race (white), marital status (married) and indicator for persons with children.
Method of Estimation: Probit Model. Relevant Findings: The probability of switching industries was higher the longer the duration of unemployment. Married males, whites and those with a longer job tenure had a lower probability of changing industries. Professionals, craftsmen and operators showed lower probabilities of changing industries. The effects of having children, years of schooling and experience were insignificant. The probability of switching industries was higher the lower the original industry employment and employment growth.
Ottersen (1993)
Source/Country/Time Period Statistics Sweden. Monthly data on the number of layoffs for the years 1978-1987. Data-type Aggregated, time-series data. Coverage Unemployed workers.
Dependent Variable Probability of being hired in the new sector after being laid off from the original sector. No. of Regressors: 3. Explanatory Variables: Work characteristics: number of lay-offs in the original industry. Other variables: Monthly dummy variables and time trend variable.
Method of Estimation OLS. Relevant Findings: The higher the number of layoffs in the original industry, the lower the probability of being hired in the new sector.
1. Osberg, Gordon and Lin (1994) did not specify what CEC stands for. It could be some form of a social funding in Canada, e.g.
Council for Exceptional Children, which aims to assist children/youth with disabilities or those who are exceptionally gifted.
Note: Vanderkamp (1977) also estimated a non-linear mobility equation with multiplicative interaction variables using a composite costs of adjustment variable Vij and interacting it with Yi, Yj, Ui, Uj, Pj and industry turnover. Vij is constructed using variables in the linear
mobility equation weighted by coefficient estimates. Results are not shown.
The structure of this review of each determinant is as follows. First, where possible, a
priori knowledge of the variable‟s impact on mobility will be highlighted. Second, the
empirical findings, irrespective of whether the studies focus on the overall, male or female
mobility, are presented. Separate findings for the employed and unemployed will also be
given where they are available. The review for each explanatory variable will highlight
whether the studies have been extended to the disaggregated analysis by gender. Finally,
194
the variables‟ feasibility in terms of alignment with the theoretical model and applicability
to the current in-depth, unit-record research for the current thesis is considered.
8.3.1 MONETARY WAGES
Overall Wages
A number of studies use overall wages as an explanatory variable even though separate
wage measures for each sector would be preferred. Overall wages was expressed in terms
of the mean income ratio [computed as the ratio of (1963 income + 1964 income) to (1961
income + 1962 income)] in Cox (1971) and in logarithmic terms in Thomas (1996b).
Jovanovic and Moffitt (1990) used the standard deviation derived from a log wage
regression on race, education and experience to test the mismatch theory of sectoral
mobility. The use of the standard deviation follows from their theory, where the probability
of a worker changing jobs was inversely related to the ratio of the costs of moving to the
standard error of the wage distribution.
The empirical findings across studies based on employed workers are consistent (Table
8.2). A positive wage-mobility relationship was found in Cox (1971), where workers who
changed industries had higher incomes than those who did not. Jovanovic and Moffitt
(1990) found that the probability of a sectoral move was higher the larger the standard
deviation of log wages for all levels of work experience (5, 8 and 11 years). An overall
wage variable was included in the study of the unemployed by Thomas (1996b), but was
found to be insignificant.
The absence of a sectoral distinction in the wages variable in the studies noted above is a
major limitation. In the absence of this sectoral distinction, it needs to be assumed that the
overall wages influence mobility via wages in the new or old industry, though the actual
channel of influence cannot be ascertained. Where possible the wages variables should be
constructed on a sector-by-sector basis to enable the origin of its influence to be
established. This ideal practice will be followed in the empirical work for Korea reported
on later in this thesis.
195
Wages in New and Original Industry
Relatively high wages in the new sector are usually viewed as a pull factor in mobility
studies. However, they may not induce industrial mobility where the higher wages do not
offset any loss of industry-specific skills [Helwege (1992)] and the costs of moving. This is
in line with the model outlined in chapter 6, where both the monetary benefits and costs of
sectoral mobility are considered.
Mixed findings have been reported on the role of the wages as a pull factor. The studies of
mobility among employees by Vanderkamp (1977) showed that higher wages in the new
industry were a significant pull factor. Osberg, Gordon and Lin (1994) reported an
insignificant effect on inter-industry mobility for wages in the new sector. Jayadevan
(1997), however, found that rising per worker real wage growth lowered industrial
employment growth during 1980/1981 to 1990/1991. This meant that at the aggregate
level, higher industrial wages did not generate greater industrial mobility. The results were
also mixed among displaced workers. Fallick (1993) reported that rising wages in the new
industry induced higher industrial mobility. Kim (1998) reported that the industrial wage
premiums of switchers were about 50 per cent smaller than for stayers. This suggests the
unemployed are willing to accept wages at below market-clearing levels in the new sector;
possibly because they are faced with liquidity constraints [Mortensen (1986)] and their
reservation wage decreases with increasing lengths of unemployment [Kasper (1967)].
The original industry‟s wage level would generally be expected to impact on industrial
mobility. It works in the opposite direction to that outlined above for the „new‟ industry‟s
pull factor. The mitigating factors outlined above for wages in the new sector are also
relevant to wages in the old sector. The empirical findings are associated with mixed
results. For employees, a net negative effect was established by Vanderkamp (1977) for
two time periods and an insignificant effect was reported by Osberg, Gordon and Lin
(1994). For the displaced workers, however, higher wages appear to result in workers
becoming unemployed, and this in turn leads such workers to change industries in Fallick
(1993).
Hence, results pertaining to old and new sector wages are associated with mixed findings.
Expectations concerning the links between sectoral mobility and monetary variables cannot
therefore be formed on the basis of the empirical literature.
196
There is limited information on whether the impacts of the old and new sector wages differ
for males and females. Only Osberg, Gordon and Lin (1994) examine this issue, and they
focused only on male employees. They found that the old and new sector wages had a non-
influential impact on mobility among males.
It is worth pointing out that the element of expectations is absent in the empirical literature,
which means that the wage variables presented in Table 8.2 will not conform to the
theoretical model exposited in equation (6.7). In addition, most of the studies use the old
and/or new industry wages, and not the sectoral wage differential in their analyses. These
studies are therefore not fully comparable with the analyses to be conducted in chapters 9
and 10, which are based on the expected sectoral wage differential. One exception is
Osberg, Gordon and Lin (1994), who used the wage differential between movers and
stayers. The wage differential has a strong theoretical basis (see chapter 6), and will be
included in the empirical analyses to be conducted below5.
Table 8.2 Wages and Sectoral/Industrial Mobility Studies of Employees Osberg,
Standard deviation of Log Wage Distribution 0.91* 1.26* 0.66* Industrial Real Wages Per Worker Growth Rate -0.91 -0.79*** Wage Differential between New and Original Industry
0.0037
Wages in Original Industry -0.0652** -0.0585** -0.0119 Wages in New Industry 0.0765** 0.0971** 0.0321** Growth Rates of Relative Sectoral Wages in VAR
Agriculture -0.26** -0.30** Construction -0.38** 0.11 Finance -0.44** -0.27** Manufacturing -0.25** -0.06 Mining -0.62** -0.80** Public Administration -0.81** -0.66** Services 0.12 -0.05 Trade -0.02 -0.27** Transport and Communications -0.10 -0.63** Utilities -0.50** -0.66** Same State, Same Industry 1.146 Same State, Different Industry 1.279 Different State, Same Industry 1.228 Different State, Different Industry 1.538 All Categories
1.177
Studies of the Unemployed Thomas (1996b) Fallick (1993) Kim (1998)
Wage in Original Industry 0.0025* Wages in New Industry Standard deviation of Industry Wage Premiums
0.16***
0.166 0.110
*** significant at 1% level. ** significant at 5% level. * significant at 10% level. 1. The mean income ratio is computed as (1963 income + 1964 income) / (1961 income + 1962 income). Note: The regression estimates shown are not comparable as their methods of estimation differ. Where the regression estimates are significant, the effect of wages on sectoral/industrial mobility is greater the larger the absolute magnitude of the estimate.
198
8.3.2 MACROECONOMIC FACTORS
Overall Unemployment
No clear links between inter-industry mobility and the overall unemployment rate have
been established in the empirical literature (see Table 8.3). For employees, Osberg (1991)
reported that unemployment had a negative effect only for females during 1980/1981 and
1985/1986. For the unemployed, the overall unemployment rate does not appear to be a
significant influence [Fallick (1993)].
The overall unemployment rate was used by Osberg (1991) in separate analyses of male
and female mobility. The results from these analyses showed that male mobility was not
affected by the overall unemployment rate, but this aggregate-level variable had a negative
effect on female mobility during several of the time periods analysed. In large part the
limited statistical success from the inclusion of the overall unemployment rate in models of
industrial mobility may be because this is not the best measure to capture any of the
influences noted above. A better measure would be to examine the separate roles of
unemployment in the old and new industries.
Unemployment in Old and New Industry
The theoretical model of chapter 6 asserts that higher unemployment in the original
industry induces sectoral movements to the new sector. Amongst employed persons,
higher unemployment in the original industry acted as a push factor for out-mobility in
Vanderkamp (1977) for all three time periods analysed. This is thus consistent with the
implications of the theoretical model. The effect of unemployment in the original sector
was, however, insignificant in Fallick‟s (1993) study of unemployed persons.
Unemployment in the new industry would generally be expected to discourage potential
entrants from moving into the new sector as their chances of securing a job in that sector
are lowered. However, this effect may be small where the wage gap between the sectors is
considerable. The empirical studies reveal mixed results. Unemployment in the new
industry did not exert any influence on the extent of inter-industry mobility among
employed workers in Vanderkamp (1977) in 1966/1967 and 1967/1968. However, there
was a surprising positive coefficient in Vanderkamp‟s (1977) study for 1965/1966.
199
Displaced workers in Fallick‟s (1993) study were not affected by unemployment levels in
the new industry.
The general observation is that studies using the old/new sectors‟ unemployment as
explanatory variables generate conflicting results, and hence predetermined views on the
unemployment-mobility relationship cannot be formed on this basis. In addition, the
literature does not appear to have examined whether sectoral unemployment rate variables
have different impacts on mobility for males and females, or whether within the separate
studies for males and females, the old/new sectors unemployment rate variable should be
defined in a gender-specific way rather than cover both males and females. Nonetheless,
the sectoral distinction in these variables aligns with the theoretical model where the push
or pull factor of mobility can be determined. From this perspective there is merit to their
inclusion in the empirical work in chapters 9 and 10.
Unemployment Spell
The theories/studies of labour mobility other than those of industrial mobility have assumed
either no intervening period of unemployment [Jovanovic and Moffitt (1990), Tobin (1972)
and Mattila (1974)] or that every job change involves an intervening unemployment spell
[Lucas and Prescott (1974)]. In comparison, the role of a spell of unemployment has been
examined in a number of studies of industrial mobility. Theoretically, as the spell
lengthens, workers would be expected to shift their search efforts towards new sectors and
to lower their earnings expectations [Pissarides and Wadsworth (1989)]. A positive effect
of spells of unemployment on industrial mobility among employed workers was reported
by Osberg (1991) and among male employees by Osberg, Gordon and Lin (1984).
The results for the unemployment spell variable in analyses for displaced workers have
been ambiguous. This may be attributed to the different coverage groups (i.e. quitters
versus losers, UI recipients versus non-UI recipients). Unemployed workers in Neal (1995)
and job quitters/losers who did not receive UI and losers who received UI in Thomas
(1996b) were more likely to change industries with a longer duration of unemployment6.
Conversely, the probability of moving sectors decreased with a longer spell among job
quitters who were UI-recipients. A likely reason for this is that since job quitters receive
some monetary compensation from UI, the opportunity cost of unemployment is lower than
200
when UI is not available, which mitigates the expected tendency to shift sectors as an
unemployment spell lengthens.
The analysis of the impact of the duration of unemployment on sectoral mobility has been
extended to separate analysis for males and females in several studies. Longer
unemployment spells were associated with greater mobility for both men and women in
Osberg (1991) during each of the three time periods examined, and for men in Osberg,
Gordon and Lin (1984). In particular, in Osberg (1991), the marginal effect of an
unemployment spell was higher for males in 1982/1983 and higher for females in
1980/1981 and 1985/19867.
Given the diversity of these findings for the unemployment spell variable, particularly for
the group of unemployed individuals for whom the variable should be more relevant, there
is arguably little benefit from including an unemployment spell variable in the empirical
application of chapters 9 and 10.
Overall Economic Growth and Employment
There are alternative viewpoints on the cyclical patterns associated with sectoral mobility
when aggregate indicators are used. Economic growth is usually associated with increases
in the rate of sectoral mobility. This occurs where the greater job availabilities associated
with an upturn encourage workers to switch sectors voluntarily. Alternatively, an economic
downturn can be associated with greater sectoral mobility where job losses/retrenchments
cause workers to seek employment in a new sector. The major study on this issue is
Loungani and Rogerson‟s (1989) analysis over the 1974 to 1984 period using a micro-
dataset for the U.S. Industry switchers in this study were defined as those who were
employed at the time of the base-year interview, i.e. year t, changed industries in year t+1
and who did not return to the original industry by year t+3. It was reported that the average
real GNP growth between years t and t+1 and the real GNP growth in year t+2 (i.e. the
growth rate of real GNP in the year following the initial industry switch) had negative and
significant effects on the proportion of industry switchers from the goods sector to the
services sector. This implies that if the declining goods sector was cyclically more
sensitive, mobility accelerates during a downturn. However, only the average GNP growth
between year t and year t+1 had a positive effect on the proportion of switchers from the
201
services to the goods sector. This implies that mobility from the acyclical services sector to
the cyclical goods sector accelerated during an economic upturn.
McLaughlin and Bils (2001) estimated the cyclical sensitivity of industries by regressing
each industry‟s share of employment on aggregate employment. The cyclical sensitivity
can be regarded as a measure of the degree of sectoral mobility in each industry. Aggregate
employment fluctuations could be expected to influence a particular industry‟s share of
employment, which in turn reflects the labour inflow/outflow from that sector.
It was shown that some industries had cyclical movements in employment less than half the
size of that of aggregate employment (agriculture, food and tobacco, communication and
utilities, public administration and several service industries), while other industries
(construction and all durable manufacturing) had employment fluctuations that were more
than twice that of aggregate employment.
The studies above therefore show that the sign of the relationship between economic
growth and mobility depends on the cyclical sensitivity of industries. This prevents strong
general priors on the impact of economic growth on mobility from being drawn. A further
reason why strong general priors cannot be drawn is that the coverage of industries is also
not all-encompassing: McLaughlin and Bils‟ (2001) approach involved industry-specific
regressions and Loungani and Rogerson‟s (1989) analysis focused only on the broad
industry grouping of „goods‟ versus „services‟. Including individuals from various
industries in a single regression appears to offer a superior encompassing test of the
relationship between economic growth and mobility. Notwithstanding the limitations
associated with GDP growth itself, overall employment can be regarded as inferior to the
GDP variable since the employment variable can be viewed as largely duplicating the
information content of an unemployment variable (as employment plus unemployment
equals the labour force, which may not vary greatly from period to period). To avoid this
possible duplication, a GDP growth variable will be used as the measure of economic
performance in the applied work in this thesis.
202
Table 8.3 Unemployment, Employment, GNP and Sectoral/Industrial Mobility
Vanderkamp (1977) Loungani and Rogerson (1989) McLaughlin and Bils (2001)
OLS estimates
OLS estimates
OLS estimates
1965/66 1966/67 1967/68 All Industry switchers
Goods to services
Services to Goods
Unemployment in Old Industry
0.1006** 0.0708** 0.0706**
Unemployment in New Industry
0.0674** -0.0025 0.0217
Average Real GNP Growth between Periods t and t+1
-0.008 -0.037* 0.076**
Real GNP Growth in Period t+2
-0.040** -0.041** -0.014
Overall Employment in:
Agriculture -1.06***
Mining -0.56
Construction 1.67***
Metals 1.76***
Machinery 1.74***
Transportation Equipment 1.73***
Other Durables 1.21***
Food & Tobacco -0.65***
Textiles, Apparel & Leather 0.33
Paper, Printing & Publishing 0.03
Chemicals, Petroleum & Rubber
0.48***
Transportation 0.35***
Communications and Utilities -0.67***
Wholesale Trade -0.01
Retail Trade -0.11
Finance, Insurance & Real Estate
-0.39***
Business & Repair Services 0.55***
Personal Services -0.33*
Health Services -1.00***
Education -0.59***
Other Professional Services -0.63***
Public Administration -0.53***
Studies of the Unemployed Fallick (1993) Neal (1995)
Hazard rate estimates Probit estimates
Overall Unemployment Rate -0.021
Unemployment in Old Industry
-0.029
Unemployment in New Industry
-0.15
Years since Displacement 0.05**
*** significant at 1% level, ** significant at 5% level, * significant at 10% level.
Note: The regression estimates shown are not comparable as their methods of estimation differ. Where the regression estimates are significant, the effect
of unemployment/employment/GNP on sectoral/industrial mobility is greater the larger the absolute magnitude of the estimate.
203
8.3.3 WORKER CHARACTERISTICS
Age
In general, increasing age is expected to be linked to a lower incidence of sectoral/industrial
mobility. Younger workers are expected to have a higher degree of sectoral mobility as
they have a longer period over which they can gain any rewards associated with the change
of jobs [Creedy and Thomas (1982)]. In comparison, older workers who face greater costs
of moving [Jovanovic and Moffitt (1990)] and have a smaller amount of time to recoup the
costs [Creedy and Thomas (1982)] are expected to be less mobile. Moreover, the chances
of moving are likely to fall with age because of accumulation of sector-specific experience
and knowledge. The ability to learn new job skills required in the new sector also
diminishes with age.
It is observed that age has been viewed in empirical research as a personal characteristic
applicable to all individuals (i.e. age of individual) and with reference to specific industries
(i.e. age of entry into industry). The same expectations apply to both forms of the variable.
A decline in labour mobility with increasing age has been found in many studies [see for
example, Mincer and Jovanovic (1981) and Antolin and Bover (1997)], and has come to be
termed a socioeconomic by-law [Byrne (1975)]. This finding carries over to the literature
on industrial mobility among employed workers (see Table 8.4). Thus, older women had a
lower incidence of industrial mobility in Osberg (1991) in 1985/1986 whilst younger males
had a greater industrial mobility in Cox (1971) and Osberg, Gordon and Lin (1994). Both
the age of entry into the old industry and into the new industry had negative impacts on
industrial mobility in Vanderkamp (1977)8.
Declining mobility with age is also a characteristic of the unemployed. Thus, Thomas
(1996b) reported that younger job quitters (UI and non-UI recipients) and job losers who
did not receive UI aged 16-19 years had higher probabilities of switching industries.
Similarly, older displaced workers in Fallick (1993), and job quitters and losers aged 45-49
years who received UI in Thomas (1996b), were less likely to switch industries.
204
Where separate analyses have been undertaken for males and females, age has been
negatively related to industrial mobility among male employees in Cox (1971) and Osberg,
Gordon and Lin (1994) and among female employees in Osberg (1991) for 1985/1986.
However, age did not exert any significant influence on mobility over a number of time
periods in Osberg (1991): 1980/1981, 1982/1983 and 1985/1986 for males and 1980/1981
Age -0.1013*** Job Quitters Job Losers Job Quitters Job Losers
16-19 years
45-49 years
1.54**
-0.63*
0.43
-0.71**
1.02**
-0.40
1.35**
0.07
*** significant at 1% level, ** significant at 5% level, * significant at 10% level.
Note: The regression estimates shown are not comparable as their methods of estimation differ. Where the regression estimates are
significant, the effect of age on sectoral/industrial mobility is greater the larger the absolute magnitude of the estimate.
205
From the studies above, the negative age-mobility relationship is nearly always reported for
all labour groups, and this expectation is to be carried over to the empirical analysis later in
this thesis. Either the linear variable of Osberg (1991), the dummy variables of Osberg,
Gordon and Lin (1994) or a quadratic function in age could be used. The dummy variable
and quadratic function in age are more general and therefore appear to offer a sounder
starting point for empirical analysis.
Gender
A gender variable has been included in several studies to capture mobility differences
between men and women. Two forms have been used: as an industry characteristic (e.g.
male/female mix in the industry) and as a personal characteristic. Vanderkamp‟s (1977)
industry characteristic variable reflected male-female specialization. It was computed for
both the original and new industries, and each of these measures was associated with
negative effects on industrial mobility. This was interpreted to mean that a higher
proportion of males relative to females in the original industry acted as a barrier to outward
mobility whilst a higher proportion in the new industry was a barrier to entry. Fallick
(1993) incorporated a dummy variable for females in his study of the unemployed and
reported that they had a higher likelihood of industrial mobility (Table 8.5).
Based on the findings of these empirical studies, as well as patterns established in the
general labour economics literature, a gender difference in mobility behaviour is expected
for the empirical work presented later in this thesis. Moreover, this expectation is a basis
for conducting the separate analyses for males and females. The gender variable will be
used as a personal characteristic rather than as an industry characteristic (i.e. the gender mix
of the industry) as this is the usual practice in recent applied labour economics research.
206
Table 8.5 Gender and Sectoral/Industrial Mobility Study of Employees Study of the Unemployed
Vanderkamp (1977) Fallick (1993)
OLS estimates Hazard rate estimates
1965/67 1966/67 1967/68
As a Personal Characteristic
Female 0.27***
As an Industry Characteristic
Male-female Specialization in Old
Industry (Fi)
-0.0910** -0.0834** -0.0662**
Male-female Specialization in New
Industry (Fj)
-0.0248** -0.0364** -0.0459**
(FiFj)1/2 0.1057** 0.0957** 0.0999**
*** significant at 1% level, ** significant at 5% level.
Note: The regression estimates shown are not comparable as their methods of estimation differ. Where the regression estimates are significant, the effect of gender on sectoral/industrial mobility is greater the larger the
absolute magnitude of the estimate.
Family Indicators: Marital Status, Head of Household and Children
It has been suggested that workers with greater family commitments, e.g. married persons,
heads of households and persons with children, have lower propensities to switch industries
as any adverse consequences (e.g. temporary loss of income) will impact more intensely on
them than on other groups. However, marital status did not exert significant effects on the
propensity to switch sectors for employees in Osberg (1991), except for 1980/1981. Among
the unemployed, married persons displayed lower incidences of mobility in Neal (1995).
The results for unemployed heads of households in Fallick (1993) were as expected, with
lower probabilities of changing industries. However, having children was not associated
with any significant influence on industrial mobility in Neal (1995). The studies that have
examined the determinants of mobility behaviour separately for men and women have
concluded that marital status generally did not have any significant impact on industrial
mobility for either males or females, as seen from Osberg (1991) and Osberg, Gordon and
Lin (1994)9. There is no evidence on whether the impacts of the household head and
children indicators on mobility differ between men and women.
Owing to these conflicting findings and the limited number of studies, preconceived views
about marital status/head of household status vis-à-vis mobility are difficult to arrive at.
207
This contrasts with the situation with respect to marital status and head of household
variables in many other areas of labour market research (e.g. wage determination,
occupational attainment). This difference is likely due to the sparse nature of research on
sectoral mobility at the present time. Accordingly, the marital status and head of household
variables will be used in the current unit-record analysis.
Table 8.6 Marital Status/Head of Household and Sectoral/Industrial Mobility
Studies of Employees Osberg (1991) Osberg, Gordon and
Lin (1994)
Males Females
1980/81 1982/83 1985/86 1980/81 1982/83 1985/86
Marital Status
Single = 1
Otherwise = 0
0.345* 0.209 -0.126 0.862 0.236 0.1005
Married = 1
Otherwise = 0
-0.012
Studies of the Unemployed
Fallick (1993) Neal (1995)
Marital Status
Currently married -0.151**
Head of Household -0.40***
With Children 0.016
*** significant at 1% level, ** significant at 5% level, * significant at 10% level.
Formal Education
The influence of formal education on the probability of moving to a new sector/industry is
indeterminate a priori. Education level is both a stock of acquired skills and a signal of
one‟s ability to learn. Whether these attributes are rewarded more in the original sector
(which would retard mobility) or new sector (which would encourage mobility) is an
empirical matter. This is viewing education as a personal characteristic.
Education can also be viewed as an industry characteristic, measured as the mean education
level or the education composition of the workforce of the industry in question. A highly
educated workforce within an industry could act as a barrier to entry into that industry,
especially if potential entrants view it as lessening their chances for securing higher paid
jobs which are generally associated with higher education.
208
Table 8.7 Education and Sectoral/Industrial Mobility
Studies of Employees Osberg, Gordon
and Lin (1994)
Vanderkamp (1977)
Probit estimates OLS estimates
1965/67 1966/67 1967/68
As a Personal Characteristic
Elementary -0.016
Post-secondary 0.15**
Diploma 0.057
University 0.032
As an Industry Characteristic
Education in Old Industry (EDi) -0.3018** -0.3735** -0.4052**
Education in New Industry (EDj) -0.2635** -0.3553** -0.3807**
(EDiEDj)1/2 0.5318** 0.6765** 0.7673**
Studies of the Unemployed Fallick (1993) Neal (1995) Kim (1998)
Hazard rate estimates
Probit estimates
Descriptive data
Industry
Switcher
Industry
Stayer
No. of grades of school
completed
0.052***
Years of schooling -0.007
Standard deviation of education 0.46 0.83
*** significant at 1% level, ** significant at 5% level.
Note: The regression estimates shown are not comparable as their methods of estimation differ. Where the
regression estimates are significant, the effect of education on sectoral/industrial mobility is greater the larger the absolute magnitude of the estimate.
From Table 8.7, the effects of education on mobility appear to differ according to how it is
measured and according to the population studied. Higher education (diploma and
university compared to post-secondary levels) did not appear to affect the mobility of the
employed in the study by Osberg, Gordon and Lin (1994). Vanderkamp (1977) included
separate education variables (ED) for the original and new industries as industry
characteristics. The coefficient of both variables were negative and of similar magnitude.
For the unemployed, the effects of education were mixed. Fallick (1993) showed that
education had a positive effect on industrial mobility. However, Neal (1995) reported the
years of schooling to have an insignificant impact. Kim (1998) compared the standard
deviation of the education levels of industry stayers and switchers, and found that industry
209
switchers have smaller industry dispersions in education. Kim (1998) proposed that the
probability of an industry switch was greater among marginal workers, e.g. low- (high-)
educated workers in high- (low-) wage industries, and the results showed that switchers
were marginal in terms of the smaller standard deviation measure.
There is limited evidence on whether the links between educational attainment and mobility
differ between males and females, and so comment on this is not provided. The superiority
of the education variable as a personal or industry characteristic cannot be determined on
the basis of consistent findings, although the former is more relevant for unit-record
analysis, is consistent with the practice in most recent applied labour market studies, and
has the merit of capturing a personal characteristic that can readily be seen as a policy
variable. For these reasons the level of education of the individual will be included in the
estimating equation used for the study of the Korean labour market.
On-the-job Training
It has been suggested that on-the-job training may be a more appropriate measure of a
worker‟s knowledge of the job than formal education and hence have a greater influence on
labour mobility [Parent (1999)]. On-the-job training can take two forms: general and
firm/sector-specific [Creedy and Thomas (1982) and Becker (1964)]. In general training,
the marginal productivity of trainees is the same across sectors. If the post-training wage is
below the workers‟ improved marginal productivity, it would be economically irrational for
the trained person to remain in the same firm/sector and the likelihood of switching sectors
is greater. In firm/sector-specific training, the worker‟s improved productivity is not
transferable to other firms/sectors. In this situation, an organization would be more willing
to bear some of the costs of training, and Becker (1964) argues that both the costs of, and
returns to, firm/sector specific training will be shared by employer and employee, which
will tend to lock workers into their existing jobs and limit mobility.
210
The fundamental difficulty with on-the-job training compared to formal education is that it
is not easily measured. Most on-the-job skills are acquired through learning-by-doing
[Oatey (1970)] rather than from formal training programmes, and learning-by-doing is
generally not quantifiable. The usual proxy variables of tenure and labour market
experience capture the influence of a range of factors (e.g. life-cycle factors, cohort effects)
and attributing any statistical relationships between these variables and labour mobility to
on-the-job training is therefore difficult.
The proxy measures for on-the-job training considered in the literature are job tenure and
labour market experience [Neal (1995) and Burdett (1978)]. An individual with a longer
job tenure or labour market experience is more likely to switch sectors if the relevant work
experience/training acquired in the years worked represents general training. Offsetting
this are other factors like benefits received. Thus, workers with longer job tenures,
experience and greater training may be less willing to move and give up seniority rights
like job security, pension benefits, seniority-based pay, longer vacation periods and
promotional advantages [Mincer and Jovanovic (1981) and Neal (1995)].
There is an issue of the specification of the tenure/experience variables that also is of
relevance. Some studies use a linear specification for these variables and others a more
general quadratic function. The studies that use a linear specification are reviewed first
below followed by those that present tenure in its quadratic form. For the linear
specification where there are more studies, the findings for job tenure are covered prior to
the findings for experience. For the quadratic functional form, however, there are fewer
studies and the two proxies are dealt with together.
Studies using job tenure as a proxy for on-the-job training have produced consistent results
among employed workers but not among the unemployed. Osberg (1991) reported that job
tenure reduced the likelihood of industrial mobility for all three time periods examined for
males and females. Job tenure was also associated with a reduced likelihood of industrial
mobility for males in Osberg, Gordon and Lin (1994). Thus, the evidence on the links
between tenure and mobility is similar for both male and female workers. The findings,
211
however, differed among unemployed persons. A longer prior job tenure reduced the
probability of inter-industry mobility for displaced workers in Fallick (1993) and Neal
(1995), and for job quitters and job losers who received UI, and job quitters who did not
receive UI in Thomas (1996b). The tenure effect was insignificant for job losers who did
not receive UI in Thomas (1996b). Kim (1998) showed that industry switchers had smaller
standard deviations in tenure at the time of job displacement compared to industry stayers.
This can be interpreted to mean that switchers would have had shorter tenures during the
pre-displacement period. Hence, a shorter tenure tends to increase the chances of an
industry switch.
In terms of labour market experience, the probability of a sectoral move decreased for all
employed workers with longer work experience (5, 8 and 11 years) in Jovanovic and
Moffitt (1990). Contradictory findings were reported for the unemployed. Labour market
experience had an insignificant effect on the probability of inter-industry mobility for
displaced workers in Neal (1995). Industry switchers and stayers both had fairly similar
standard deviations in work experience during the period of job displacement in Kim
(1998). It can be inferred that as switchers and stayers would both have similar work
experience in the pre-displacement period, the effect of experience on the likelihood of an
industry switch is non-influential.
In cases where tenure/experience is entered in quadratic form, an examination of the partial
derivatives shows that the effects are in the same direction across all reasonable
tenure/experience levels, and the discussion that follows focuses on the most common
effects without digressing to deal with turning points that occur at high levels of tenure/
experience. From Table 8.8, where the partial derivative of tenure/experience is negative
across all reasonable tenure levels, the coefficient of the quadratic term is positive in
Osberg (1991) for males for all three periods and females for 1980/1981 and 1982/1983,
and in Neal (1995) for the unemployed. This indicates that the negative influence of
tenure/experience on sectoral mobility diminishes as tenure/experience increases.
The only study that captured training directly was Osberg, Gordon and Lin (1994), where it
was reported that training received in the previous sector exerted a positive and significant
212
impact on inter-industry mobility. However, this measure will not be dealt with in the
current work owing to the difficulty in determining a suitable across-the-board measure of
training for individuals in the KLIPS10
.
In summary, consistent results are revealed in the studies of job tenure/experience and
mobility for employed workers. Given the strength of these findings, a negative
relationship between tenure/experience and mobility would be expected in the current
work, both for the aggregate-level analyses and for the separate analyses to be undertaken
for males and females. Given the evidence in favour of non-linear relationships between
mobility and tenure/experience, tenure/experience should be examined using a quadratic
function.
Table 8.8 On-the-job Training and Sectoral/Industrial Mobility
Table 8.8 On-the-job Training and Sectoral/Industrial Mobility (continued)
Studies of the
Unemployed
Thomas (1996b) Fallick
(1993)
Neal (1995) Kim (1998)
Weibull-competing risk estimates Hazard rate
estimates
Probit estimates Descriptive data
UI recipient Non-UI recipient
Job
Quitters
Job
Losers
Job
Quitters
Job
Losers
Industry
Switcher
Industry
Stayer
Tenure/10 -0.02** -0.01** -0.01* -0.004
Job Tenure -0.023*** -0.024**
Job Tenure-squared 0.001***
Standard deviation
for Tenure
1.35 1.85
Experience -0.010
Experience-squared 0.0001
Standard deviation
for Experience
1.64 1.71
*** significant at 1% level, ** significant at 5% level, * significant at 10% level.
Note: The regression estimates shown are not comparable as their methods of estimation differ. Where the regression estimates are significant, the effect of on-the-job training on
sectoral/industrial mobility is greater the larger the absolute magnitude of the estimate.
215
Occupation
Occupation is reflective of skill levels, and has been measured in mobility studies by initial
occupation, change in occupation and as a proportion of individuals in skill-intensive
occupations (see Table 8.9). There are alternative viewpoints regarding the industrial
mobility behaviour of skilled and semi-skilled workers. On the one hand, skilled workers
could exhibit „mobility stickiness‟ [Subrahmanian, Veena and Parikh (1982)] if skills are
team-specific and jobs rely on the existing net of workers [Mailath and Postlewaith (1990)
and Chillemi and Gui (1997)] in the original sector. On the other hand, by virtue of their
skill being vital in certain industries [Neal (1995)], skilled workers may be scouted for their
talent, potential injection of new ideas or productivity [Murphy and Topel (1990)]. White-
collar jobs are also more likely to be advertised than blue-collar jobs [Abraham (1987)] and
the rate of mobility for such workers may be higher. The change in occupation was
examined by Vanderkamp (1977), who argued that such changes impose additional costs to
mobility in the form of acquiring new skills and retraining for a different occupation.
The initial occupation was examined by Osberg (1991) for the employed and Neal (1995)
and Fallick (1993) for the unemployed. Osberg (1991) reported that higher probabilities of
industrial mobility were reported by female personal service workers for all three periods
examined (1980/1981, 1982/1983 and 1985/1986), female managers, professionals and
technicians in 1982/1983 and 1985/1986, and female clerical and sales workers in
1980/1981 and 1985/1986. Male managers, professionals and technicians in 1980/1981
were also more likely to change industries. Fallick (1993) reported that displaced workers
who were in the technical, sales or administration, precision production, craft and repairs or
who were operators, fabricators and labourers in the old industry had higher propensities to
switch sectors. In contrast, Neal (1995) reported that unemployed workers who were
professionals, craftsmen and those who were operators were less likely to be industry
switchers. The empirical findings revealed mobility stickiness for job losers who received
UI [Thomas (1996b)]11
. The change in occupation was a deterrent to industrial mobility in
Vanderkamp‟s (1977) empirical work. In terms of the proportion of individuals in skill-
intensive occupations, higher skill levels did not exert any significant effect on sectoral
mobility in Loungani and Rogerson (1989).
216
This mixed evidence therefore does not provide a basis for establishing priors on whether
the mobility behaviour of skilled and unskilled workers will differ. Similarly, as Osberg
(1991) is the only study that addresses whether mobility patterns across occupations differ
for males and females, and the extent to which his findings generalize to other countries
and time periods is not clear, priors for the role that occupation may have in the separate
analyses to be conducted for men and women cannot be formed.
Part of the reason for the conflicting results on the role of occupation in the empirical
literature may be the different variables used (initial occupation, change in occupation,
dummy variables, industry averages). However, it seems that there are grounds for a
reasoned choice in this regard. Between the first two types of variables mentioned above,
the change in occupation is less preferred as it depicts another form of labour mobility,
namely, occupational mobility, and so may potentially be endogenous. In contrast, the
initial occupation variable is exogenous, and has an unambiguous interpretation, and for
this reason is preferred for the empirical work. For consistency with the representation of
other worker characteristics in the model, the initial occupation will be categorized as a
dummy variable (i.e skilled versus unskilled).
Table 8.9 Occupation and Industrial Mobility
Studies of Employees Osberg (1991) Vanderkamp (1977) Loungani and Rogerson (1989)
Logit estimates OLS estimates OLS estimates
Males Females
Occupational status 1980/81 1982/83 1985/86 1980/81 1982/83 1985/86 1965-66 1966-67 1967-68 All Industry
Switchers
Goods to Services
Services to Goods
Proportion of individuals in skill-intensive occupations (professionals/managers/craftsmen)
-0.70 0.94 -1.42
Initial Occupation
Personal service -0.216 0.159 0.151 1.313** 2.215*** -0.623**
*** significant at 1% level, ** significant at 5% level, * significant at 10% level. Note: The regression estimates shown are not comparable as their methods of estimation differ. Where the regression estimates are significant, the effect of occupation on sectoral/industrial mobility is greater the
larger the absolute magnitude of the estimate.
218
Industry
Two studies have analysed the role of the initial industry in determining sectoral mobility,
namely Osberg (1991) and Thomas (1996b) [see Table 8.10]. Osberg (1991) found that
workers were more likely to move out of certain industries. These included male workers
from the construction industry for each of the three time periods examined, males in
manufacturing and resources for 1982/1983, females in construction and government in
1985/1986 and females in manufacturing, trade and finance, and utilities and transport in
1982/1983. Thus, between males and females, it is evident that their probabilities of
changing sectors vary depending on their initial industry. The results pertaining to the
initial industry were not significant for the unemployed in the study by Thomas (1996b).
Table 8.10 Initial Industry and Industrial Mobility
Studies of Employees Osberg (1991)
Logit estimates
Males Females
1980/81 1982/83 1985/86 1980/81 1982/83 1985/86
Construction 0.849*** 1.279*** 0.665** 0.759 2.127***
Trade, finance 1.060***
Government 0.559 1.262***
Utilities, transport 1.700***
Manufacturing 0.776*** 1.820***
Resources 0.82*
Studies of the Unemployed Thomas (1996b)
Weibull-competing risk estimates
UI-recipients Non-UI recipients
Job Quitters
Job Losers
Job Quitters
Job Losers
Primary/manufacturing -0.06 0.02 -0.12 0.16
*** significant at 1% level, ** significant at 5% level, * significant at 10% level.
Note: The regression estimates shown are not comparable as their methods of estimation differ. Where the
regression estimates are significant, the effect of industry on sectoral/industrial mobility is greater the larger the
absolute magnitude of the estimate.
Thus, while the empirical basis is limited, it appears that the initial industry is likely to
impact inter-industry mobility and hence this variable should be considered in the current
study. This is particularly the case if policy relevance is an issue, as it is obviously
important to know if mobility varies across industries. Gender differences from this
perspective also seem likely [see Osberg (1991)] and should be established for the Korean
219
labour market if possible. Hence variables for the initial industry will be included in the
mobility equations used in chapters 9 and 10.
Employment Status, Unionisation, Alternative Sources of Income and Region
A number of other potential determinants of worker mobility have been considered, e.g.
employment status in Osberg (1991), unionization in Vanderkamp (1977), unemployment
insurance in Osberg, Gordon and Lin (1994) and Fallick (1993), social assistance in
Osberg, Gordon and Lin (1994) and region by Vanderkamp (1977) and Thomas (1996b).
The empirical evidence relating to the latter four factors is not reviewed as it is not relevant
to the empirical analyses to be conducted below.
The research on the impact of employment status on sectoral mobility is relevant to the
current work and the finding by Osberg (1991) will be reported here. Osberg (1991)
focused on full-time versus part-time workers, although it seems that other categorizations
could be used, such as employees versus employers, own account workers and workers in
family business. Part-timers and employees are expected to have a higher incidence of
industrial mobility as they are less emotionally attached to their current job/sector than full-
time employees. This was confirmed in Osberg (1991), where male and female part-time
workers were both associated with higher mobility rates (see Table 8.11). Since Osberg
(1991) is the sole study of sectoral mobility that covers employment status, the findings
should not be generalized to other samples. However, the apparent strength of the results,
their accord with intuition, and the policy and social relevance of knowledge of whether
sectoral mobility varies by employment status provide sound reasons for considering an
employment status variable in the study of mobility in the Korean labour market.
220
Table 8.11 Employment Status and Industrial Mobility
*** significant at 1% level, ** significant at 5% level. 1. Dummy variable indicating product similarity (Dummy x 10-2). The industries are classified into 11 product groups.
2. Dummy variable indicating work similarity (Dummy x 10-2). The industries are divided according to two types of work effort: light (including mental activity) and heavy (including
manual and physical activity). Note: The regression estimates shown are not comparable as their methods of estimation differ. Where the regression estimates are significant, the effect of the working hours/product
similarity/ work similarity on sectoral/industrial mobility is greater the larger the absolute magnitude of the estimate.
224
Size of Original and New Industries
Industry size is typically measured in empirical studies of worker mobility by the level of
employment or changes in employment and, to the extent that the contractions and
expansions in employment reveal declining or growing trends for specific industries, the
variable can reflect industry performance. Higher employment in the initial industry could
be viewed as a good indication of the industry‟s job market prospects/availability and lower
costs of finding a job in that industry [Neal (1995)]. Thus, a negative correlation between
mobility and employment growth in the old industry, and a positive association between
mobility and size of the new industry, are to be expected.
In terms of the size of the old industry, the expected negative relationship was reported by
both Fallick (1993) and Neal (1995) [see Table 8.13]. However, both of these studies cover
displaced workers only, and this approach does not appear to have been applied to other
groups or in separate analyses of the mobility of male and female employees.
With respect to the size of the new sector, higher employment was shown to be positively
correlated with industrial mobility in Vanderkamp (1977) for the three time periods
analysed. Similarly, lower job availabilities in the new industry (observed from the no-job-
available index) reduced the likelihood of industrial mobility for males in Osberg, Gordon
and Lin (1994). To the extent that lower job availabilities is an indication of lower
employment in a sector, the positive correlation between the new sector‟s size and mobility
is implied in the latter study.
While these findings are interesting, the limited number of studies of employed workers is a
shortcoming of research in this particular area. It prevents strong conclusions from being
drawn, whether for the aggregate labour force or for males and females separately.
Nevertheless, the ready availability of measures of employment size means that the impacts
of size of the original and new industries can be easily addressed in the current empirical
study. In this context, it is noted that the industry size variables (i.e. for original and new)
can be both examined in a single regression model as simultaneity does not appear to be an
225
issue. At any time t, the pool of workers in one industry differs from the pool of workers in
another.
Industry Turnover and Output
It is believed that higher turnover or output for an industry is generally indicative of its
ability to recruit, which raises the probability of finding a job and encourages inter-industry
mobility. However, higher turnover or output may also mean a greater chance of
retrenchment, which would be expected to be a deterrent to industrial mobility.
Vanderkamp (1997) reported a positive and significant relationship between turnover in the
new industry and industrial mobility. Jayadevan (1997) introduced the industrial output
growth rate as a performance indicator in an analysis of the manufacturing sector. Using
establishment data that covered all workers, it was found that a higher growth in output was
positively associated with higher employment growth or a higher net inflow of labour.
Among the unemployed, Ottersen (1993) showed that the number of layoffs in the original
sector was negatively associated with the probability of being hired in the new sector.
The different measures employed in these studies, as well as the vastly different
populations studied, prevents a consensus on the impact of industry performance on
sectoral mobility from emerging. Moreover, analysis of the industry performance –
sectoral mobility relationship has not been conducted for separate samples of males and
females. However, as with some other variables, there is considerable practical/policy
appeal in having knowledge of the links between industry performance and sectoral
mobility. For this reason, the applied work below will consider this relationship. In
assessing the variables, Vanderkamp‟s (1977) turnover indicator has the merit of
distinguishing the old versus new sectors so that the actual channel of influence on mobility
can be determined. However, the information is not about sectoral performance per se but
the sector‟s ability to recruit or retrench. Jayadevan‟s (1997) industry output (or GDP
growth) is a viable indicator as it reflects the economic performance of the sector, but it
lacks the sectoral distinction preferred for the current study and is confined to the
manufacturing sector. Hence, for the current work, it is recommended that a sectoral
performance indicator in the form of GDP growth be adopted with a sectoral distinction.
226
Table 8.13 Sectoral Performance Indicators and Sectoral/Industrial Mobility
Studies of Employees Jayadevan (1997) Osberg, Gordon
and Lin
(1994)
Vanderkamp (1977)
OLS estimates Probit
estimates
OLS estimates
1973/74 to 1979/80 1980/81 to 1990/91 1965/66 1966/67 1967/68 Industry Size
Employment in New Industry 0.0553** 0.0488** 0.0423**
No-Jobs Available Index in New Industry1
-1.98**
Industry Turnover Turnover in New Industry 0.0167** 0.0105** 0.0171**
Output Growth Rate 0.64*** 0.47***
Studies of the Unemployed Fallick (1993) Neal
(1995)
Ottersen
(1993)2
Hazard rate estimates Probit
estimates OLS
estimates
Industry Size Employment in Old Industry -4.7** -0.032***
Employment Growth in Old
Industry
-1.460***
Employment in New Industry -44.0
Ratio of Employment in the
New Industry to Employment in Old Industry
0.0012
Industry Turnover Layoffs in the Old Sector -2.01***
*** significant at 1% level, ** significant at 5% level.
1. The index is calculated as the difference between individuals from the same industry (who responded that a shortage of jobs created
difficulties in finding employment during their periods of non-employment) against the weighted national average of individuals in
another industry for the same occupation.
2. The dependent variable was the probability of being hired in the new sector conditional upon the fact that the workers were laid off
from the original sector. Note: The regression estimates shown are not comparable as their methods of estimation differ. Where the regression estimates are
significant, the effect of sectoral performance indicators on sectoral/industrial mobility is greater the larger the absolute magnitude of the estimate.
8.4 DETERMINANTS UNDER SECTORAL SHOCK THEORY
Table 8.14 outlines the main features of three studies that have examined the impact of
sectoral shocks on shifts in sectoral employment under the sectoral shock theory. Six
measures of sectoral shocks were considered: the annual sectoral employment growth rate
[Gulde and Wolf (1998)], the residual of an AR regression on the lagged growth rate of
employment [Gulde and Wolf (1998)], the standard error of a sectoral shock [Jovanovic
and Moffitt (1990)], the residual from a regression on the innovation (steady-state) variance
in aggregate employment [Altonji and Ham (1990)], the residual of a VAR regression of
the variance in industrial employment [Clark (1998)] and the industry-specific excess stock
returns [Brainard and Cutler (1993)].
227
The approach taken varies considerably across these studies. Gulde and Wolf (1998) used
the spatial correlation of the shock measures along the national (i.e. European Union) and
sectoral dimension for both the AR(1) measure of the shock and sectoral employment
growth rates. This correlation technique is not recommended as it merely measures the
association of sectoral shocks amongst countries and various sectors, and not the impact of
a shock as measured under a formal regression. Jovanovic and Moffitt (1990) regressed the
probability of a sectoral move for various groups of workers on a sectoral shock variable,
where the measure of a shock was provided by the standard deviation of residuals from
sector-specific AR(2) regressions of the log annual U.S. employment. Altonji and Ham‟s
(1990) shock variable was derived from the residual of a regression on the innovation
variance in aggregate employment in the Canadian labour market. This innovation
variance was expressed as a function of the variances of the national, provincial, sectoral
and U.S. shocks [derived from the residual of an AR(2) regression of U.S. GNP]. The
impact of the sectoral shock was determined via regression of the shock on each sector‟s
employment growth rate. Clark‟s (1998) model involved estimation in two forms, namely,
a VAR model and error models. The VAR model (for K lags) was estimated as:
K
Xt = ∑ ζk Xt-k + et k=1
where Xt is a vector of regional and industrial employment rates of growth, ζk is the vector
of regression coefficients and et is the error term. The error models were estimated as:
er,t = θrct + ∑ αrt εit + µrt i
ei,t = θict + εit + ∑ βrtµrt r
for industry i and region r, with ct, εit and µrt each representing unobserved national,
industrial and regional shocks. The θ coefficient represents the impact of a national shock.
Whilst the α coefficient measures the impact of the industry i shock in region r, the β
coefficient measures the impact of the region r shock in industry i. In particular, the
response of an industrial (sectoral) shock can be measured by the α coefficient in the error
model.
A concern with the measures adopted in Altonji and Ham (1990) and Clark (1998) is that
the various shock measures (national, provincial and sectoral) may be mutually
228
correlated12
, which could mean potential problems of multicollinearity. Brainard and
Cutler (1993) regressed the excess industrial employment change on the industry specific
excess stock returns. This dependent variable was the sum of residuals over several time
horizons arising from a regression of the change in the logarithm of each industry‟s
employment on a constant and the change in the logarithm of total employment. A concern
with the excess returns measure is that it is a capital measure rather than a labour market
measure. Furthermore, the excess returns variable does not appear to have great
explanatory power (R2 ranged from 0.001 to 0.007).
Despite the differences in methodology, the studies reviewed in Table 8.14 have a common
finding, namely that sector-specific shocks affect sectoral/industrial mobility. Gulde and
Wolf (1998) reported that sectoral shocks to the transport and food industries exhibited the
strongest spatial correlations, while the agricultural and textiles industries had smaller
correlations. Jovanovic and Moffitt (1990) found that sectoral shocks (as measured by the
residual of an AR regression on the lagged growth rate of employment) affected labour
mobility positively for workers with 5 and 8 years of experience. Alternative measures of
sectoral shocks were used (Lilien index and net flows of sectoral employment) but these
had insignificant effects on labour mobility. It was argued that the reason for the poor
performance of these alternative measures is that they include foreseen components of
changes arising from a sectoral shock, compared to the standard error of sectoral shocks
which accounts for the unanticipated effects following an exogenous shock. Altonji and
Ham (1990) established a positive impact of a sectoral shock on each sector‟s employment
growth for most industries up to 5 years, after which the effect dissipated. Clark‟s (1998)
study indicated that sector-specific shocks had a greater influence than the national shock
on the variance of industrial employment. Brainard and Cutler (1993) reported that the
excess stock returns to industry significantly predicted industrial employment growth,
although the effects were small.
229
Table 8.14 Sectoral Shocks and Sectoral/Industrial Mobility under Sectoral Shock Theory Gulde and Wolf (1998) Brainard and Cutler
(1993) Jovanovic and Moffitt (1990)
Annual
Employment
Growth Rate Measure
Residual of
AR(1)
Regression Measure
Industry‟s Excess
Returns2
Standard deviation
of Sectoral Shocks3
Lilien
Index3
Net Flows in
Sectoral
Employment3
Years of Experience
5 years 3.28* -4.51 -0.96
8 years 2.78*
-7.40 -1.54*
11 years 1.39
-8.23
-0.64
Quarters
1 0.0061* 4 0.0523***
8 0.0815***
12 0.0964***
16 0.1165***
20
0.1340***
Correlation patterns:
Shocks to
Employment Growth1
Agriculture 0.0532 0.0686
Construction 0.449 0.0211
Food 0.3007 0.2505 Chemicals 0.1623 0.0480
NM minerals 0.3099 0.2093
Metal products 0.1457 0.1108 Textiles 0.0136 0.0223
Paper, printing 0.1120 0.1139
Transport equipment 0.2557 0.1252 Other manufactures 0.2228 0.2066
Transportation 0.1713 0.2909
Fuel and power 0.1711 0.2067 Market services 0.1682 0.0815
*** significant at 1% level. * significant at 10% level. 1. The higher the correlation, the greater the association of a sectoral shock with that sector‟s employment.
2. The larger the magnitude of the industry‟s excess returns, the greater the impact of a sectoral shock. The effects occur up to 20
quarters. 3. Where the regression estimates are significant, the effect of a sectoral shock on sectoral/industrial mobility is greater the larger the
absolute magnitude of the estimate. The Lilien index was described in chapter 3. 4. The impact of a sectoral shock on the growth rate of each industry‟s employment is higher the larger the absolute magnitude of the
residual estimate. The effects disappear after 5 years. 5. The impact of an industry shock is higher the larger the share of the fitted variance due to industry. Note: The regression estimates shown are not comparable as their methods of estimation differ.
230
In summary, whilst a range of methods have been used in the literature, several of these
seem less suited to the current study than others. Gulde and Wolf‟s (1998) correlation
technique is unsuitable, as it offers only a measure of association. Moreover, the spatial
component of the shock measure computed across the European countries is not relevant to
the current study which focuses on one country (i.e. Korea). The shock measures in Altonji
and Ham (1989) and Clark (1998) could have multicollinearity problems and Brainard and
Cutler‟s (1993) capital measure may not be relevant where the focus is on labour market
pressures. Thus, it appears that the AR technique employed by Gulde and Wolf (1998) and
Jovanovic and Moffitt (1990) to measure an industrial shock is the preferred approach for
the current work.
8.5 DETERMINANTS UNDER BRIDGING THEORY
The study by Jovanovic and Moffitt (1990) attempted to model sectoral/industrial mobility
based on the bridging theory (refer to Table 8.1). This section will not review the findings
for specific variables from this study since this was done in the sections above. Instead, the
further implications for modelling will be highlighted.
Jovanovic and Moffitt‟s (1990) model included one monetary variable (the standard
deviation of the wage distribution) and the sectoral shock variable. The monetary variable
was constructed by first estimating a log wage regression by year as a function of
education, experience, experience-squared and race. The predicted monetary variable and
sectoral shock were then entered into the mobility regression, which was estimated using
samples of workers at 3 levels of experience, 5, 8 and 11 years.
The main feature of the Jovanovic and Moffitt (1990) study of relevance to the current
study is the inclusion of the variables typically included in tests of the mismatch theory as
well as the sectoral shock variable. This approach will be followed in the empirical
chapters of this thesis.
231
8.6 ASSESSMENT OF EMPIRICAL STUDIES OF
SECTORAL MOBILITY FOR MODELLING
The sections above have reviewed the findings on specific variables in the empirical
literature, and where possible have commented on whether parallel variables should be
included in the empirical application to Korea. The lessons from past research can be
extended to issues associated with data-type, coverage, model specification, variable-type
and method of estimation. Under each of these headings, general observations from
empirical studies are identified below, followed by a critical assessment. Where possible,
links with the theoretical model and studies of other forms of mobility (in chapters 6 and 7)
are made. For reference, Table 8.1 has outlined the features of the studies of
sectoral/industrial mobility pertaining to the source, data-type, coverage, model
specification, dependent variable and method of estimation.
Data-type
Studies of sectoral mobility have been based on three types of data: longitudinal or unit-
record time-series13
, unit-record cross-sectional and aggregate-level datasets. Studies with
longitudinal data include Osberg (1991), Osberg, Gordon and Lin (1994), Jovanovic and
Moffitt (1990) and Loungani and Rogerson (1989), whilst those with cross-sectional data
comprise Vanderkamp (1977) and Neal (1995). The studies using aggregate-level datasets
are McLaughlin and Bils (2001), Jayadevan (1997) and Ottersen (1993).
Chapters 6 and 7 (and Part I) recommended the use of micro-level data for the study of the
factors affecting mobility to facilitate an in-depth understanding. Thus, studies with
aggregate-level data are less favoured since only broad-level patterns can be identified.
Amongst studies with unit-record data, analyses with time-series or cross-sectional data
appear to be equally valuable, as the former is able to cater for a time dimension in the
analyses, and the latter can provide an in-depth profile of industry movers. The KLIPS is a
unit-record longitudinal dataset which marries the two data categories bringing together the
benefits of both into a single estimating equation.
232
Coverage
The studies reviewed in Table 8.1 cover three labour groups5: the overall workforce
[Vanderkamp (1977), Loungani and Rogerson (1989) and Ottersen (1993)], males [Osberg,
Gordon and Lin (1994), Jovanovic and Moffitt (1990) and Neal (1995)] and separate
analyses for males and females [Osberg (1991)]. It is seen that most studies have examined
either the overall or male workforce, save for Osberg (1991). As the mobility patterns of
males and females may differ, chapter 6 recommended that the analyses be disaggregated
by gender. Against this background, the Osberg (1991) study appears to have an advantage
over the others by extending its analysis to female mobility. The current thesis will examine
mobility behaviour for the pooled workforce, and also conduct separate analyses for the
male and female workforces in Korea.
Model Specification
The model specification recommended in the previous chapter involved a mix of monetary
and non-monetary variables. Several studies of sectoral mobility have used such a
specification, namely, Osberg (1991), Vanderkamp (1977), Osberg, Gordon and Lin
(1994), Jovanovic and Moffitt (1990), Loungani and Rogerson (1989) and McLaughlin and
Bils (2001). Jovanovic and Moffitt (1990) is distinguished by also including a stochastic
shock variable. Therefore, in terms of the specification, these studies are more relevant for
the current work than those focusing only on the sectoral shock variable, i.e. Gulde and
Wolfe (1998), Brainard and Cutler (1993), Clark (1998) and Altonji and Ham (1990) [refer
to Table 8.14].
Sectoral Distinction of Variables
One of the points gathered from chapters 6 and 7 was that a sectoral breakdown for both
monetary and non-pecuniary variables was desirable. However, only a few of the studies
listed in Table 8.1 have accommodated this, namely the sectoral wage differential in
Osberg, Gordon and Lin (1994), sectoral wages, size and unemployment in Vanderkamp
(1977) and sectoral performance in Jayadevan (1997). The current study will be based on
sectoral-specific variables where possible.
233
Dependent Variable and Method of Estimation
Chapter 6 indicated that the dependent variable should be binary, indicative of an
individual‟s change of sectors/industries, and this should be analysed with a probit or logit
model. Selected studies, i.e. Osberg (1991), Osberg, Gordon and Lin (1994), Jovanovic
and Moffitt (1990) and Neal (1995), have used the probit or logit model to examine the
probability of a sectoral move. These are more relevant to the empirical model of equation
(6.7) than those studies using the proportion/growth rate of industry movers/moves and
OLS for estimation.
In summary, there is no one single study that accommodates all the relevant features for the
current work. The empirical model is a combination of the features implied from its
theoretical origins (outlined in chapter 6) and extracted from other studies of various forms
of mobility (in chapter 7).
8.7 SUMMARY OF EMPIRICAL STUDIES OF SECTORAL MOBILITY
This chapter has described the various ways worker mobility has been modelled. The
empirical evidence from sectoral/industrial mobility can be succinctly stated. It should be
noted that only variables where at least two studies reported a similar finding are discussed
in this paragraph. The main determinants of sectoral/industrial mobility appear to be overall
wages, unemployment duration, age, tenure, working hours, size of the old/new industry
and sectoral shocks. Among the employed, the variables that were positively associated
with sectoral/industrial mobility were overall wages, working hours, size of the new
industry and sectoral shocks. Unemployment duration, age and tenure were shown to have
a negative impact on mobility. Among the unemployed, only age, tenure and size of old
industry, which had negative effects on industrial mobility, were statistically significant.
Only a few studies reported that separate analyses were conducted for males and females.
Unemployment spell, employment status, working hours/weeks and size of new industry
were positively associated with male mobility, whilst age and tenure had negative impacts.
234
Employment status and unemployment spell had positive influences on female mobility,
whilst age, tenure and the overall unemployment rate had negative effects on female
mobility. The remaining variables considered for the separate gender groups, namely
marital status for both groups and working hours/week for females, were statistically
insignificant.
Table 8.15 lists the explanatory variables used and findings in the studies of
sectoral/industrial mobility. The applicability of these variables with respect to the current
analysis for the Korean workforce is also summarized in this table. Whilst sections 8.3 and
8.4 have reviewed the explanatory variable in terms of conceptual alignment with the
theoretical model and research on sectoral mobility, an assessment of the variables on
issues of measurement and applicability for the Korean labour market, and availability of
data in the KLIPS dataset, is also provided in the table.
Given the varied evidence and few common findings for each explanatory variable, few
firm general conclusions can be drawn. Even fewer can be drawn concerning gender
differences in the determinants of mobility. However, this should not be viewed as overly
alarming, as there are conflicting hypotheses regarding the impact of most variables on
industrial mobility. Thus, the impact of many variables cannot be determined prior to the
empirical application. The analysis on the determinants of industrial mobility in Korea to
be undertaken in chapters 9 and 10 will adopt a comprehensive approach which may enable
comment on the array of findings in the literature to date.
235
Table 8.15 Assessment of the Explanatory Variables
Yes Individual‟s average monthly income in the new sector, adjusted for the chances of finding employment. This variable will be used to compute the expected sectoral wage differential.
Wages in the Original Sector/Industry
M (E) P(U)
Yes Individual‟s average monthly income in the original sector. This variable will be used to compute the expected sectoral wage differential.
Sectoral Wage Differential I (E) Yes Wage differential between individual‟s monthly income in the new sector and individual‟s monthly income in the original sector. This variable will be adjusted for the chances of finding employment.
Wage Growth in the New Sector/Industry
n.r. Yes The annual growth rate of wages (in percentage terms) in the individual‟s new sector/industry.
Wage Growth in the Original Sector/Industry
n.r. Yes The annual growth rate of wages (in percentage terms) in the individual‟s original sector/industry.
Macroeconomic Variables Overall Unemployment N (E)
I (U) No This variable is not recommended as
there is no sectoral distinction.
Unemployment in the New Sector
M (E) Yes The unemployment rate of the individual‟s new sector. A lagged variable should be used if simultaneity occurs with sectoral/industrial mobility.
Unemployment in the Original Sector
P (E) I (U)
Yes The unemployment rate of the individual‟s original sector. A lagged variable should be used if simultaneity occurs with sectoral/industrial mobility.
Unemployment Duration
P (E) M (U)
No The data pertain to the unemployed for aggregate-level datasets.
Overall Economic Growth M Yes Overall GDP growth rate.
Overall Employment M No This variable is not recommended as it should simply capture influences similar to the unemployment variable.
Inflation Rate n.r. No All workers face the same price levels and changes regardless of sectors.
236
Table 8.15 Assessment of the Explanatory Variables (continued)
Explanatory Variables Findings Applicability Measurement/Remarks Worker Characteristics Age N (E)
N (U) Yes Age of individual (in years).
Gender M (E-males)
P (U-females) Yes Gender of individual entered as a
dummy variable for males versus females.
Race n.r. No A racial distinction is not relevant for predominantly mono-cultural societies like Korea.
Language n.r. No The findings were insignificant. Furthermore, the variable is not needed as the common/commercial language in Korea is Korean.
Marital Status P (E) N (U)
Yes Marital status of an individual (married versus non-married) entered as a dummy variable.
Household Head N (U) Yes Household head status of an individual entered as a dummy variable to distinguish if the person was a household head.
Children I (U) No This variable is not recommended as its
effect was found to be insignificant.
Formal Education M (E) M (U)
Yes Education status entered as a dummy variable to distinguish tertiary versus non-tertiary educated workers.
On-the-job Training P (E) No The data on on-the-job training are difficult to quantify.
Tenure N (E) N (U)
Yes Job tenure of the individual measured in years is available.
Initial Industry M (E) I (U)
Yes The initial industry of an individual.
Occupation Status M (E) M (U)
Yes Occupational status of an individual (skilled versus semi-skilled) entered as a dummy variable.
Employment Status M (E) Yes Employment status of an individual [employee versus other workers (employer, own account workers, family workers)] entered as a dummy variable.
Unionisation n.r. No The data are not available for non-employees in Korea.
Region n.r. No The data are only available for region of birth, and not region of present residence.
Alternative Sources of Income
n.r. No The KLIPS had poor data quality as the majority of respondents did not know whether they had social assistance.
237
Table 8.15 Assessment of the Explanatory Variables (continued)
Explanatory Variables Findings Applicability Measurement/Remarks Job/Industry Characteristics Working Hours/Weeks P (E) No There was a relatively high number of
KLIPS respondents who did not indicate their working hours/weeks. Therefore, the number of observations for this variable compared to the other explanatory variables is relatively low.
Product/Work Similarity P (E) No The level of product/work similarity cannot be ascertained from the KLIPS.
Size of Original Industry N (U) Yes Level of industry employment in the individual‟s original industry.
Size of New Industry P (E) I (U)
Yes Sum of industries‟ employment except that of the individual‟s original industry.
Industry Turnover P (E) N (U)
No The separation and accession rates were not available for the agricultural sector.
Performance of Original Industry
n.r. Yes Value-added growth rate of individual‟s original industry.
Performance of New Industry n.r. Yes Value-added growth rate of individual‟s new industry (i.e. all other industries except individual‟s original industry).
Sectoral Shock
P
Yes
Residual of an AR regression on employment that is lagged by one or more time-periods.
Annotation: P : one or more studies reported a positive effect on mobility. N : one or more studies reported a negative effect on mobility. I : one or more studies reported an insignificant effect on mobility. M : Mixed findings among studies/groups/periods. It can refer to differing results among multiple studies and/or
across time periods for same work group in the same study, or for different groups (e.g. males and females) in the same study.
E : Employed persons. U : Unemployed persons. n.r.: The variable was not reviewed. Note: Where neither „E‟ nor „U‟ is indicated, the variable covers the macro-economy.
8.8 SUMMARY OF LESSONS DRAWN FROM THE LITERATURE
Numerous lessons have been drawn from the theoretical and empirical review in the first
three chapters of Part II, covering both sectoral mobility and other forms of mobility.
Chapter 6 provides the theoretical basis for model application and estimation. The
extended Le and Miller (1998) model (as per equation 6.7) is the recommended tool where
conceptual advancements are introduced in the form of the expected sectoral wage
differential and lifetime earnings. Probit- or logit-type regressions are deemed as most
suitable, catering for the use of dichotomous dependent variables which adequately reflect
238
dual mobility states, namely, to move or to stay. Gender analyses on mobility, which are
usually neglected in the literature, could be conducted depending on findings of gender
differences in Korea. The model can be applied to test the three theories of sectoral
mobility covering the worker-employer mismatch, sectoral shock and bridging hypotheses.
Chapter 7 reviews other forms of labour mobility (union/non-union, public-private, rural-
urban) and gives a general framework for specification of the current model. The
recommendations are to establish a model that includes a sectoral wage differential as well
as macroeconomic and non-monetary factors; a sectoral distinction for the unemployment
variable, and sectoral breakdown for non-pecuniary variables where possible. Longitudinal
datasets are superior to cross-sectional datasets as they are a rich data source and cater for
time-series analyses. The latter attribute is desirable for the current study as
macroeconomic and lagged dependent variables, which have been found to be significant
determinants of mobility, can be incorporated into the empirical model.
Chapter 8 is the critical literature review where empirical evidence is canvassed from
studies of sectoral/industrial mobility that could be a yardstick against which analyses of
the determinants of mobility in Korea are assessed. However, varied evidence and
conflicting hypotheses prevent the formation of firm conclusions for each variable. The
determinants reported to be significant covered an array of monetary factors (overall
tenure and working hours), job characteristics (size of the old/new industry) and sectoral
shocks. This spread of factors, coupled with the limitation in the number of common
findings, gives rise to the adoption of a comprehensive approach in model specification for
the current work.
The final section of chapter 8 (section 8.7) summarizes the applicability of the explanatory
variables with respect to the current analysis, taking into account issues related to
measurement, the Korean labour market and data availability. The determinants to enter
into worker‟s mobility function are the sectoral wage differential, sectoral wage
growth/unemployment/size/performance, GDP growth, sex, age, marital status, educational
attainment, head of household status, occupational status, employer status, job tenure and
239
the sectoral shock. The investigation into the determinants of sectoral mobility in Korea
begins in chapter 9, followed by the study disaggregated by gender in chapter 10.
Endnotes:
1. Refers to 751 job losers and quitters who changed industries out of a total of 1,089 who went through at
least 1 week of unemployment but gained employment either in the new or old industry.
2. Refers to 1,685 workers who changed industries out of a total of 2,641 employees. The 2,641 were
unemployed in the 5 years before the survey date but they had gained full-time employment at the point of the
survey.
3. Several studies [Podgursky and Swaim (1987), Madden (1987, 1988) and Addison and Portugal (1989)]
recognized the importance of industrial mobility but did not examine this form of worker mobility behaviour.
These studies focused on the wage losses of displaced workers instead.
4. The industry was classified into eleven product groups to distinguish product similarity and two work
groups (light – mental activity, and heavy – manual/physical activity) for classifying work similarity.
5. It is noted that Prasad (1997) examined the correlations between the growth rates in relative sectoral
employment and relative sectoral wages. The relative measures of these variables were the deviations from
the aggregate growth rates. Negative and significant correlations between relative wages and employment
were found in agriculture, construction, finance, manufacturing, mining, public administration and utilities
during the 1959-1993 period. This measure is not suitable for the current study as it is a bi-variate correlation
study, not regression analysis. 6. Results from Thomas (1996b) were inferred from Figures 1 and 2 for a standard worker who receives/does not receive UI according to the route of job separation (quit or loss). The standard worker is one who worked in a service industry in a blue-collar occupation for 2.2 years, did not belong to a union and had an hourly wage rate of $9.16. The results are the probability of the u-r transition from being unemployed in sector „a‟ to being employed in sector „b‟, Mjb (tu), conditional upon the length of the unemployment spell (tu) and job separation status (j). Mjb (tu) = hjb (tu) / [hjb (tu) + hja (tu)], where hj represents the hazard rates of transition. 7. The marginal effect is measured by β*ρ*(1-ρ)*100, where β is the regression coefficient and ρ is the
proportion of industry movers. In Osberg (1991), the marginal effects for males were 0.23, 0.18 and 0.21 for
1980/1981, 1982/1983 and 1985/1986, respectively. For the same corresponding periods, the marginal effects
for females were 0.26, 0.12 and 0.32.
8. It is noted that the coefficient of (AiAj)1/2
was positive.
9. Amongst the unemployed, Neal (1995) reported that married males had a lower likelihood of switching
industries.
10. The KLIPS questionnaire classifies training (excluding regular schooling) according to training in a
private institution, authorized vocational institute, public vocational institution, in-house training by firms etc.
Since there are a variety of training programmes, those who received training would have had different types
of training which may or may not be relevant to the new sector. Furthermore, not all individuals would have
received training, especially the non-employees.
11. This does not support the view in Neal (1995) that switchers forfeit compensation for industry skills.
12. Altonji and Ham (1990) and Clark (1998) made mention of possible correlations between shock measures
but assumed the errors were independently distributed in the models.
13. A number of studies use datasets that follow individuals or firms over time. These can be described as
unit-record time-series datasets. In this thesis, these will be referred to via the usual terminology of
longitudinal or panel datasets.
240
CHAPTER 9
EMPIRICAL STUDY ON THE DETERMINANTS
OF SECTORAL/INDUSTRIAL MOBILITY IN KOREA
9.1 INTRODUCTION
The literature review in chapter 8 indicated that there is a solid empirical foundation for the
understanding of worker mobility. It was suggested that the main determinants of sectoral
labour mobility in the U.S., Canada, Sweden and India are monetary, economic,
demographic and socio-economic factors. However, three gaps in the research were
identified. First, there is a dearth of mobility studies for Asia. This chapter attempts to fill
this void in the literature by modelling mobility behaviour in Korea. Second, there are a
number of inconsistencies in the results reported. For example, there are few common
findings with regards to the relationship between inter-industry mobility and worker
characteristics. The current study will attempt to account for these inconsistencies. Third,
the studies reviewed often focus on one set of variables (e.g. demographic) to the exclusion
of others (e.g. monetary). It is the intent of this chapter to conduct an all-encompassing
formal study covering monetary, economic, demographic and socio-economic factors,
which no other study has done.
The objective of this chapter will therefore be to comprehensively examine the
determinants of sectoral/industrial mobility for the Korean workforce. The chapter is
organized as follows. Section 9.2 introduces the data source, concepts, coverage and time
periods used in the empirical work. A generic model of sectoral/industrial mobility is
presented in Section 9.3. Sections 9.4 and 9.5 present descriptive statistics of the variables
used in the regression analysis of sectoral mobility, as well as selected
predicted/recomputed monetary and sector-level variables. The results of the empirical
analysis of the determinants of sectoral mobility are presented and discussed in Section 9.6.
A series of extensions of the empirical model are considered in Section 9.7. These include
assessing the impact of an individual‟s industry of origin on sectoral mobility as well as an
empirical test of three theories of mobility: worker-employer mismatch, sectoral shock and
bridging theories. A summary of the empirical findings and concluding comments are
241
given in the final section. The list of variables and the rules followed when deriving them
are provided in Appendices 9A and 9B.
9.2 DATA SOURCES, CONCEPTS AND COVERAGE
This chapter is based on both unit-record and aggregate-level data. The unit-record data
were obtained from the Korean Labor and Income Panel Study (KLIPS) conducted by the
Korea Labor Institute (KLI)1. The aggregate-level data were obtained from the Korea
National Statistical Office (NSO).
There are five ideal prerequisites for a micro-level dataset on inter-industry mobility: (i) the
sample should be representative of the working population; (ii) the dataset should be large;
(iii) individuals must be surveyed at least twice; (iv) the data should extend over a fairly
long period; and (v) individuals should report income and industry at the time of the
interview rather than over the past year [McLaughlin and Bils (2001)]. The unit-record
data available in the KLIPS sample satisfy these prerequisites.
9.2.1 KLIPS Data
Although there are several national labour surveys in Korea, i.e. Current Population Survey,
Special Survey of Employment, Survey of Labour Mobility and Basic Survey of Wages,
these are cross-sectional in nature. Cross-sectional data do not cater for the construction of
many of the variables that are prominent in studies of mobility behaviour, and the KLI was
set up to enable the collection of longitudinal data that might overcome these and to
facilitate in-depth study of the labour market and mobility issues. The KLIPS that it has
collected is a longitudinal study of a representative sample of households and individuals
living in urban areas in Korea. It is the first panel survey in Korea on labour-related issues.
The first wave was launched in 1998 in the midst of the Asian Financial Crisis.
The KLIPS sample is an equal probability sample of households from the seven
metropolitan cities and urban areas in eight provinces in Korea. The sampling frame was
from the 1995 Korean mid-term census. Out of 21,675 census unit areas, 951 sampling
unit areas were selected. For each sampling unit, five to six households were randomly
242
chosen. The KLIPS sample yielded 5,000 households in the first wave, with some 13,321
household members aged 15 years and over being successfully interviewed. These 5,000
households represent the original panel in the study.
The study comprises four waves of data that were collected from 1998 till 2001. The initial
sample of households were interviewed in 1998 (wave 1), with follow-up interviews in
1999 (wave 2), 2000 (wave 3) and 2001 (wave 4). New joiners, namely those who have
blood or economic ties to the original panel members, were added to the sample in waves
2-4. Where a panel member moved out and formed an independent household with his/her
new family (e.g. spouse), then the new family members were treated as new joiners to the
original panel. Additionally, if an outside party joins one of the existing households
surveyed (e.g. via marriage), he/she was also included in the interview. Each person is
identified by a unique personal identification number (PID).
The field work for the KLIPS started in May and ended around September each year, with
the majority of household interviews being completed by end-August. Consequently, the
survey reference month is treated as end-June (mid-point) each year.
Any analysis on mobility needs a time dimension to assess if mobility occurred. This paper
looks at the mobility over one year, i.e. between year t-1 and year t. Sectoral/industrial
mobility is defined as having occurred if a person switched industries between year t-1 and
year t. This establishes one of the selection criteria for the current study‟s dataset, namely
that a person must participate in at least two consecutive survey years and have reported
positive incomes and valid data on the industry of employment in the two years. It is
possible to identify such persons by using the PID to match respondents in the datasets of
adjacent years. One advantage in using a one-year time dimension to examine mobility is
that persons who switch industries with longer intervening unemployment spells can be
included in the study.2
As the KLIPS attempts to track members who moved out of the original household, the
KLIPS dataset does not depend on residential stability. Since at the aggregate level, inter-
industry mobility includes inter-industry movers who change their place of residence, this
means that the measure of industrial mobility available for use in this thesis will not
243
underestimate aggregate inter-industry labour mobility. This an advantage over Osberg‟s
(1991) study, for example, where movers were dropped from the original Canadian sample
of households and thus his measure of the dependent variable was the conditional
probability of inter-industry mobility, given residential stability. Since the residential
movers in Canada comprised 3% of the initial sample, if the probabilities of residential and
inter-industry mobility are positively correlated, Osberg‟s (1991) estimate of the probability
of sectoral mobility would be biased downwards.
Notwithstanding these advantages, several limitations of the dataset must be pointed out.
The past year‟s data can be captured fairly accurately for waves 2-4, as respondents, having
been interviewed in the first wave, are aware of the subsequent interviews and will
probably record and report the information at the time of the interview and/or provide an
update of the change of information from the previous year‟s survey. The responses for the
initial wave, however, could be subject to greater recall error as respondents were being
interviewed for the first time and did not know previously that they had to provide answers
to their income/industry over the past year. So the actual dollar income earned or specific
industry group for the initial wave may not be accurate. Nevertheless, there is consistency
in all the waves in the sense that the past year‟s information refers to the past 12 months.
This even applies to the initial wave. Some respondents may have reported a series of jobs
in the past but it is possible to ascertain the previous year‟s income/industry based on the
start dates and quit dates of the previous job. So the data for wave 1 are fixed to a specific
time period of one year, i.e. as at June 1997.
The other limitation lies in the length of the time period available for research. The four-
year time series, though adequate for research on some labour market characteristics (e.g.
unemployment duration), may not be sufficiently long to capture effects on mobility over
an individual‟s working life. Furthermore, as mobility patterns are detected at the same
month every year (i.e. June), it would not be possible to ascertain the seasonal responses in
mobility behaviour. Unless mobility patterns can be tracked quarterly/six-monthly instead
of annually, any seasonal effects on inter-industry mobility should be interpreted with
caution.
To obtain information about individuals during the pre-move period (year t-1) and post-
move period (year t), the data items from the previous wave‟s dataset were appended to the
244
current wave‟s dataset via matching with the PID. The list of data items and their
derivations are supplied in Appendix 9A. Moreover, for the initial 1998 wave, respondents
were asked to provide their previous income, industry, occupation, employment size,
employment status, start dates and quit dates, and so these records could be considered for
inclusion in the set of inter-industry movers for the study below.
The analysis focuses on a subset of the population in the dataset, namely persons aged 20-
64 years. The age group is chosen as the mobility patterns for younger workers (aged less
than 20 years) are affected by schooling behaviour, whilst those for older workers (beyond
64 years) are influenced by retirement behaviour. To model mobility that is affected by
schooling or retirement behaviour is beyond the scope of the thesis. Therefore, the focus is
on workers aged 20-64 years. This is in line with Oi‟s (1987) recommendation to include
adults aged 20-64 years as this offers a „cleaner statistic measuring variations in labour
market activity‟. Respondents with non-positive income, those who did not report either an
old/new industry, and those who did not provide valid data on any other question used in
the analysis are excluded from the sample. Consequently, the sample for the current study
amounts to 10,691 person-year observations covering the period 1998-2001 (about 4 years
per person). In addition, one variable, working hours in the individual‟s original sector, was
excluded owing to its significantly fewer number of observations (6,161).
The structure of the sample dataset varies from that used in past studies. Cross-sectional
analyses of mobility behaviour have been conducted for different periods with periodic
gaps. For instance, Osberg (1991) analysed 3 sets of years: 1980 to 1981, 1982 to 1983 and
1985 to 1986, with a periodic gap between the second and third set. This study‟s panel
dataset combines the mobility records of individuals from 4 consecutive waves: 1997 to
1998, 1998 to 1999, 1999 to 2000 and 2000 to 2001. By repeatedly interviewing the same
respondents over the years, there are no periodic gaps and mobility behaviour can be more
readily analysed in conjunction with the continuous time-series macroeconomic data which
can be embedded into this type of data structure. With the inclusion of relevant time-series
macroeconomic data in micro-level panel data, the explanatory power of the regression
analysis of sectoral mobility is likely to be enhanced.
245
9.2.2 Korea NSO Data
The macroeconomic variables and sectoral indicator variables considered from the
literature review to have influence on mobility decisions, namely: overall/sectoral GDP
growth rate, sectoral/industrial employment size and unemployment rate, and annual
growth rates in sectoral/industrial income, are obtained from the Korea NSO. In addition,
the sectoral shock measure is estimated using industrial employment data from the NSO.
9.2.3 The Role of Interim State of Unemployment
The sample of 10,691 covers employed persons who reported an industry of employment as
at year t-1 and year t. This does not preclude the possibility of such persons being out of
employment between periods t-1 and t. The purpose of this section is to demonstrate that
ignoring any interim states of unemployment does not affect the analysis of sectoral labour
flows. By doing so, the various states of employment/non-employment and the relationship
between gross and net labour flows are highlighted. The approach here is to examine the
inflows and outflows of labour from a larger sample with fewer restrictions, and to compare
these to the proposed final sample of 10,691 observations, where additional restrictions are
imposed to permit a more refined analysis. In addition, this preliminary analysis will
provide the reader with information on how the sample of 10,691 observations evolved.
9.2.3.1 Sectoral Labour Flows
Table 9.1 shows the labour flows based on a sample of persons aged 20-64 years, where
individuals can report either their new or old industry, or both. That is, respondents need
not report all of the information on wages, job tenure, employment status, occupational
status and educational attainment. This gives us a larger sample size of 29,474 person-year
observations. This sample constitutes industry stayers (Es), inter-industry movers,
employed workers in year t-1 who become unemployed, moved out of the labour force or
did not report any industry in year t (denoted by Uo, where the subscript refers to outflows)
as well as the unemployed, those not in the labour force (NILF) or who did not report any
industry in year t-1 who entered into employment and reported their industry in year t
246
(denoted by Ui, where the subscript refers to inflows). Among inter-industry movers, the
inflow of entrants into a particular industry is represented by Ei, and the outflow, by Eo.
The gross inflow of labour into an industry from year t-1 to year t will comprise workers
from other industries as well as the unemployed and those formerly NILF3. That is, gross
labour inflow = Ei + Ui. For example, the gross inflow of labour into the agricultural
sector (302) consists of workers from the non-agricultural sector (237) and those formerly
unemployed or persons NILF (65).
At the same time, the gross outflow of labour from an industry consists of workers who
changed to other industries as well as those who became unemployed or moved out of the
labour force4. The gross outflow from the agricultural sector, for example, is 390,
and this includes movers into the non-agricultural sector (328) and persons who
become unemployed or choose not to participate in the labour force (62). Thus,
gross outflow = Eo + Uo.
Table 9.1 Gross and Net Labour Flows based on Sample of 29,474 Observations New Industry
Communications), 8 (Financial, Real Estate & Business Services) and 9 (Community, Social & Personal Services). Uo : Employees in an industry in year t-1 who become unemployed or moved out of the labour force or did not report any
industry in year t.
Ui : Unemployed or those not in the labour force/did not report any industry in year t-1 who entered into an industry of employment in year t.
Note: As at year t-1, Uo and Ui are mutually exclusive. As at year t, Uo and Ui are mutually exclusive.
The net labour flow is taken as the difference between the gross outflows and inflows.
Mathematically, the net flow = (Eo + Uo) – (Ei + Ui.). The example of the agricultural
sector reveals a net outflow of 88 persons. From Table 9.1, the gross outflow exceeds the
247
gross inflow for all sectors/industries except for transport, storage and communications.
This pattern is not surprising since the data collection was during the post-Asian Financial
Crisis period which witnessed numerous business closures and job losses on an economy-
wide scale. An outflow of labour in nearly all sectors/industries is thus to be expected.
9.2.3.2 Missing Industry Information
The provision of a respondent‟s industry information in period t-1 and period t is a critical
key for the empirical exercise. In the dataset, some workers did not state their industry of
employment in either survey period t-1 or period t. Their numbers are represented by
Uiinterim
and Uointerim
in Table 9.2. From Tables 9.1 and 9.2, Ui = Ui*
+ Uiinterim
and
Uo = Uo* + Uo
interim. From the first equality, Ui
interim denotes persons who did not report any
industry in period t-1 but reported an industry of employment in period t. The Ui* category
comprises persons formerly in non-employment in period t-1 who entered into employment
in period t. For the second equation, Uointerim
are those who had a job/industry reported in
period t-1 but did not provide their industry of employment in period t. The Uo*
category
represents workers formerly in employment in period t-1 who became unemployed or left
the labour force in period t. Since such persons under Uiinterim
and Uointerim
categories did
not report any industry information in one of the time periods, they are excluded from the
final sample.
It is observed that missing industry information does not really affect the comparison of
gross flows and net flows in the KLIPS. Compared to Table 9.1, when Uiinterim
and Uointerim
are ignored, gross outflows still exceed the gross inflows for all sectors/industries in Table
9.2. Furthermore, the labour movements of inter-industry movers (i.e. the Eo‟s and Ei‟s) of
Table 9.2 are very similar to those of Table 9.1. One difference between Table 9.1 and
Table 9.2 is the net flow data for the transport, storage and communications industry, where
the net flow turned positive, from -1 to 5. The small disparity of 6 persons stems from the
difference between Uiinterim
(19) and Uointerim
(13). Thus, from this exercise, the non-
importance of the non-stated industry categories is illustrated.
Table 9.2 Gross and Net Labour Flows based on Sample of 29,474 Observations New Ignoring Uo
Annotation for Industry, Uo and Ui : See Table 9.1. * : This comprises respondents who reported both their original and new industries but did not provide any information for at least
one of the following variables: old wage, new wage, job tenure or occupation. These records were excluded to obtain the main sample of 10,691 observations.
9.2.3.4 Interim States of Unemployment
The sample of 10,691 observations will include those who may have experienced an
interrupted spell of unemployment between period t-1 and period t. Out of this sample,
there are some 826 workers who encountered an unemployment spell during the interim
period, as shown in Table 9.4. The difference between these workers and the Uiinterim
and
Uointerim
groups is that they reported their industry of employment as at the survey reference
dates. Hence, they can be effectively classified under their industry of employment, as
shown in Table 9.4.
The purpose in this section is to illustrate that even if the 826 persons were excluded, the
main features of the comparison of the gross outflows and gross inflows carry over from
the comparisons shown in Table 9.3. There is a net outflow of labour from the
agricultural, mining and manufacturing sectors, and a net inflow into the utilities,
construction, commerce and services industries. Therefore, ignoring the state of intervening
unemployment does not affect the labour flows in this study of inter-industry mobility.
251
Table 9.4 Gross and Net Labour Flows based on Sample of 10,691 Observations New
Industry
Old
Industry
1 2 3 4 5 6 7 8 9
Interim Unemployment between period t-1 and period t
1 6 0 3 0 1 3 1 0 1
2 0 0 0 0 0 0 0 0 0
3 0 0 170 0 9 14 4 15 4
4 0 0 1 2 1 0 1 0 0
5 0 0 7 0 85 3 5 5 2
6 0 0 16 1 7 147 17 15 6
7 0 0 2 0 0 4 31 4 1
8 0 0 11 0 3 12 2 82 7
9 0 0 5 0 4 9 2 10 85
Number of persons with uninterrupted employment
Gross
Outflows
Net
Flows
1 407 0 36 1 27 38 11 30 13 156 74
2 4 5 2 0 7 4 0 0 1 18 13
3 15 0 1889 4 67 251 49 111 95 592 179
4 1 0 1 25 1 3 2 4 0 12 -5
5 15 1 54 3 672 61 16 64 22 236 -8
6 26 0 164 0 71 1744 43 121 112 537 -25
7 5 1 31 0 22 36 588 28 22 145 -30
8 7 3 78 8 34 85 35 1167 70 320 -97
9 9 0 47 1 15 84 19 59 1118 234 -101
Gross Inflows 82 5 413 17 244 562 175 417 335
Annotation for Industry: See Table 9.1.
9.3 GENERIC MODEL OF SECTORAL/INDUSTRIAL MOBILITY
The generic model of sectoral labour mobility adopted for the study of individuals‟ choice
between two sectors given in the index function of equation (6.7) is restated here:
Ii = γ1 + γ2 [ ln pi + ln yai – ln ybi] + γ3 gai + γ4 gbi - Ziδ - Siφ.
The actual and expected incomes, measured over the individuals‟ lifetimes, represented in
the model are as described earlier. Zi is an all encompassing vector of economic,
demographic and socio-economic factors, Si represents the stochastic shock term, δ is a
vector of coefficients for Zi and θ is the coefficient for Si. This generic model enables us to
test three theories of sectoral mobility: the worker-employer mismatch, sectoral shock and
bridging theories of sectoral mobility that were outlined in chapter 6.
The index, Ii, is a latent variable for the propensity of workers to move across industries. It
is not observed. Rather, what is observed is a binary indicator of whether workers moved
252
(I*i). It takes the value 0 when individual i did not switch sectors/industries, and the value 1
if individual i did switch sectors/industries, between period t-1 and period t. It can be
linked to the latent index Ii as follows: I*i = 1 if Ii ≥ 0 ; I
*i = 0 otherwise.
The dataset classifies the sectors/industries according to the Korean Standard Classification
of Industries. The sectors/industries are categorized into nine major groups: agriculture;
Source: KLIPS dataset, KLI. Note: Since the annual GDP growth rate does not vary within each wave, the weighted standard deviation is zero. Hence, its design
effect cannot be computed. See Appendix 9C.
1. The lower standard deviation (weighted series) is due to fewer individuals in mining. The standard deviation is zero for waves 2 and 3 for new entrants and wave 4 for new entrants/survivors. Hence, the design effect is relatively low.
2. The lower design effect reflects the lower standard deviation (weighted series) for old/new industry growth for new entrants in waves
3 and 4. Since these variables vary by wave only, the standard deviation for each wave merely reflects distributional differences amongst new entrants and survivors.
3. The standard deviations on a by-wave basis are generally less than 100 except for wave 3 for survivors. This accounts for the lower
overall weighted standard deviation of 56. Hence, the design effect is lower. 4. Since the residual is derived from a regression for each wave, the residual on a by-wave basis is negligible. Any deviation within each
wave reflects distributional differences between survivors and new entrants and is negligible. Thus, the descriptive statistics between
the weighted and non-weighted sets differ and the design effect is small.
259
In terms of worker and job characteristics, there are proportionately more males (65%) than
females (35%) in the Korean sample. There are proportionately more married persons
(73%), employees (80%) and household heads (53%) in the sample than the respective
complementary categories (the non-married, non-employees and non-household heads).
There are relatively fewer graduates (15%) and professionals (8%). The typical worker is
40 years of age and has accumulated 7 years of work experience in his original/current job.
The initial industries for most individuals are concentrated in the manufacturing sector
(25%), commerce sector (23%) and financial, real estate and business services industries
(15%).
The monetary indicators favour the new sector/industry. Thus, the new industry‟s expected
wages (in natural logarithms) exceed the old industry‟s actual wages, and the annual
average growth rates in income of the workers‟ new industries are marginally higher than
the rates in their old industries.
On average, the GDP growth rates and employment sizes of workers‟ new industries were
lower than those of their original industries. The lower growth suggests that the influence
of the sector‟s past performance on a sectoral move may not be compelling. The average
lagged annual unemployment rate was slightly higher for new industries. That is, at the
aggregate level, movement to a new sector will involve a trade off of higher unemployment
for higher wages.
It is observed that the comparison of old versus new sector monetary and sector-level
variables remains unchanged under the weighted and non-weighted series. The average
GDP growth rate of the industries of employment reported by Korean workers in the
sample was 4%.
The typical worker experienced a sectoral shock whilst working in their original industry
during 1998-2001. This is revealed by the positive mean values from the alternative
measures of a sectoral shock. There are two approaches to estimating the AR(1) residual
designed to capture the unanticipated effects on sector-specific employment between two
time periods. The first, the residual of an AR(1) regression7 (micro-level), is computed by
regressing the individual industry‟s employment in period t on that of period t-1, and
260
taking the difference between the fitted and estimated values of industry employment for
each individual record. The second AR(1) residual (by wave) was estimated in a similar
fashion to that of Jovanovic and Moffitt (1990). The natural logarithm of the industry‟s
employment AR(1) regression was estimated for each year (i.e. four years) and the
corresponding four standard errors from these regressions were then inserted into the
dataset for each record8. All individual records within the same wave (or year) would have
the same value. It is not surprising that the standard deviation was significantly smaller
than for the first AR(1) measure.
The shock measures are estimated across all sectors/industries of the economy and
comparison data need to be constructed in a similar way9. The positive mean value of the
AR(1) residual (by wave) is comparable in size with Jovanovic and Moffitt‟s (1990) shock
measures, which ranged from 0.006 to 0.031 during 1968-1980 in the U.S.
The third measure of sectoral shock is given by the cross-sector standard error of the
residual of an AR(1) regression of the natural logarithm of industry employment. This
estimates the unobservable effects on sectoral employment independent of the effects on
aggregate employment. For each observation, it is computed as:
[eit/Et x (res(ln eit) – res(ln Et))]1/2
,
where eit is the industry‟s employment, Et is aggregate employment, res(ln eit) is the
residual of an AR(1) regression of industry employment and res(ln Et) is the residual of an
AR(1) regression of aggregate employment.
Table 9.6 shows that the mean value of the cross-sectoral standard error of this AR(1)
residual was 0.18, and this is much smaller than the AR(1) residual (micro-level) measure.
This is due to the fact that it removes the unanticipated effects of a shock on overall
employment, and as such it allows the effect on a specific sector to be examined in
isolation. Therefore, it might be reasonable to expect that this might be the most
appropriate measure for the empirical study.
These descriptive statistics give a preliminary indication of some worker/job and monetary
variables that might be influential in the mobility decision. The patterns between old and
261
new sectors in many of these variables are consistent with expectations. However, in the
case of sectoral growth rates, the new sector‟s rate is lower than the old sector‟s, and for
unemployment rates, the new sector‟s rate is higher. A further examination of these
variables is required. The extent to which patterns more consistent with economic theory
emerge from the study of individual-level data will be examined in the later part of this
chapter.
9.5 DERIVATION OF PREDICTED/RECOMPUTED VARIABLES
As the examination of sectoral mobility involves industry movers and stayers, and the study
focuses on the motivation behind a movement from the old sector to the new, there must be
some observable differences in the explanatory variables for movers and stayers. This
leads to problems for the researcher in the case of stayers, for whom there is no designated
new industry, especially for variables that make use of aggregate-level industry data, and
for the monetary variable which involves computation of the new sector‟s wage, i.e. the
sectoral wage differential. The purpose of this section is to compute suitable values for
these variables for industry stayers.
Sector-level Variables
The potential new sector-level values for industry stayers are not observed for the wage
growth, lagged unemployment rate, sectoral size and sectoral performance variables. If the
current sector values are assigned to the new sector variables, there is a high correlation
between the old-new variables: ga versus gb (0.889), Ua,t-1 versus Ub,t-1 (0.875), sizea versus
sizeb (0.728) and GDPa versus GDPb (0.793). These correlation coefficients are less than
one only because the old and new industries for movers differ.
To overcome this problem, the sector-level variables for movers are constructed using the
average across all industries other than the stayer‟s original industry value as the new
industry values. These re-computed new sector values should not exhibit high correlations
with the values for the old sectors. The remaining explanatory variables, namely the
262
individual characteristics and GDP growth, are not affected as they remain unchanged for
each individual in the sample datasets.
The list of the data items following these changes and their corresponding annotation are
supplied in Appendix 9B.
Sectoral Wage Differential
The issue of determining an appropriate „new sector‟ value also arises for the sectoral wage
differential for stayers, as their old sector wage, but not their potential new sector wage, is
observed. A related issue is that the actual wage data available at the micro-level contain
both an observed predictable element and an unobserved stochastic component. This could
lead to biased estimates if the stochastic component is related to the error term in the
mobility equation.
The possibility of biased estimates with the use of actual wage data can be accommodated
through the use of predicted wages. The variable in question is the expected sectoral wage
differential. For this, the individual‟s new and old sectors‟ wages, and the new sector‟s
unemployment rate, need to be derived to arrive at the wage differential term. For
reference purposes, these derived data items will be termed the „predicted new sector‟s
wage‟, „predicted old sector‟s wage‟, „predicted new sector‟s unemployment rate‟ and
„predicted expected sectoral wage differential‟.
9.5.1 Predicted Sectoral Wages
The computation of the predicted old/new sector‟s wages adopts the methodology of Tomes
and Robinson (1982a) in their estimation of the determination of wages for two different
regions in the context of Canadian interprovincial migration. The methodology of Tomes
and Robinson (1982a) is an example of what Borjas (1980) refers to as using a „clean‟
proxy for a wage variable that has econometric problems associated with it. Offered wages
are usually held to be dependent on personal attributes, and so the wage functions for sector
a (ln yai) and sector b (ln ybi) for an individual i can be constructed as:
263
ln yai = Xiβa + uai (9.1)
ln ybi = Xiβb + ubi (9.2)
where Xi is the set of observable personal characteristics (sex, age, marital status,
educational attainment, head of household status, occupational status, employer status and
job tenure)10
, βa and βb are the vectors of parameters associated with each sector to be
estimated for the corresponding explanatory variables, and uai and ubi are the unobservable
components representing the general ability and other factors applicable to sectors a and b,
respectively, but which are not captured under Xi. The dependent variables, ln yai, and
ln ybi, are the individual‟s actual wages (expressed in natural logarithms) in the new sector
and old sector, respectively.
Since industry movers and stayers each experience different levels of utility from changing
sectors or remaining immobile, the sample is self-selected into mover and stayer sub-
samples, and the estimation of the individual wage equations (9.1) and (9.2) is based on
these truncated samples. Predicted wages are then obtained from the industry-specific
wage regressions for movers and stayers. It is noted that these equations are not corrected
for sample selection bias, as undertaken by Tomes and Robinson (1982a), as there is a lack
of variables in the dataset that can be used as legitimate identifying variables in the
selection equation. Furthermore, there have been recent dissatisfaction with the selection
bias methodology where the correction might reduce the accuracy of coefficient estimates
[Puhani (2000) and Stolzenberg and Relles (1997)].
As several studies have revealed inter-industry wages to vary [Carrington and Zaman
(1994), Dickens and Katz (1987), Gibbons and Katz (1992), Helwege (1992), Keane (1993)
and Krueger and Summers (1987)], and some industries may value particular employee
attributes more highly than others, an industry-specific approach to the estimation of the
wage regressions of equations (9.1) and (9.2) is recommended11
.
Predicted Wage for Movers
As in the case of Tomes and Robinson (1982a), wage regressions are undertaken for each
mover and stayer sub-sample. For movers, the term ln ybi is the actual wage reported by
264
individual i in period t-1 in the old industry and ln yai is the corresponding wage reported in
period t in the new industry. Specifically, as mobility in the KLIPS sample is measured
annually, new (old) sector earnings are based on data reported in the first (previous) year of
each survey wave, i.e. 1998 (1997) for wave 1, 1999 (1998) for wave 2 and 2000 (1999)
and 2001 (2000) for waves 3 and 4, respectively.
To obtain the predicted wages for movers in the new (old) industry, ln yai (ln ybi) is
regressed on the Xi‟s for each industry. The fitted values, ln yap
(ln ybp), constitute the
predicted wages for movers in the new (old) industry. Each mover will have a different
predicted wage for the old/new sector computed for incorporation into the mobility
equation. In this regard, the KLIPS dataset is superior to the Tomes and Robinson (1982a)
data, as wages for movers are estimated before and after a sectoral switch. Therefore, the
predicted wages for movers would be:
ˆ ln ya = Xβa from the new industry; and ˆ ln yb= Xβb from the old industry.
It is noted that incomes in the four years of the survey need not be adjusted to the real
(1998) values using the CPI. Since the final model is about the sectoral wage differential,
the real and actual wage differentials are the same as both old/new sector wages are
adjusted by the same deflator12
. This is in line with empirical studies of sectoral mobility
adopting a sectoral wage differential [Osberg, Gordon and Lin (1994)] or old/new sector
wages [Vanderkamp (1977) and Fallick (1993)] which do not make use of a wage deflator.
In addition, recent studies using the KLIPS wage data (though not about sectoral mobility)
have not adopted real wages in their analyses, including Son (2007a), Son (2007b), Kim
(2003), Kang, Park and Lee (2007) and Kang (2004).
Predicted Wage for Stayers
For stayers, the predicted wages in the old industry are obtained from an industry-specific
regression of ln ybi on Xi. The reported old sector wages are based on the previous year,
namely 1997 for wave 1, 1998 for wave 2, 1999 for wave 3 and 2000 for wave 4. Each
stayer will have a different predicted wage for the old sector.
265
As mentioned earlier, the estimation of predicted wages in the new industry for stayers will
pose a problem since their old and new industries are the same. The dataset gives the actual
wage of period t, and not the potential wage that could be achieved in an alternative
industry. To overcome this limitation, the procedure adopted by Tomes and Robinson
(1982a) is used. They treated the new destination as a single alternative comprising all
destinations other than that in which the individual (stayer) was observed in period t. For
example, for an industry stayer working in the commerce sector, the new sector‟s wage is
the average wage across all sectors other than the commerce sector.
To obtain „ln yai‟ for a typical stayer, the aggregated earnings for stayers reported across all
industries other than the original industry is divided by the number of stayers in all
industries other than the original industry, which is then expressed in natural logarithmic
terms. These averages are argued to provide a general idea of what could be earned
following a sectoral move. Earnings (i.e. ya) are based on data observed in the first year
each survey wave was conducted, i.e. 1998 for wave 1, 1999 for wave 2 and 2000 and 2001
for waves 3 and 4, respectively. Stayers from the various „original‟ industries will have
differing ln yai‟s, but those from the same „original‟ industries (for the same survey wave)
will have similar ln yai‟s.
It should be noted that many observations would have the same value for ln yai for the
obvious reason that earnings in the new sector (ya) are not realized or observed by industry
stayers. The variability within each stayer sub-sample needed for the industry-specific
regression to be feasible arises because stayers from different survey waves will have
dissimilar ln yai‟s.
Having derived stayers‟ new sector wages, industry-specific regressions of ln yai on Xi were
then estimated and used to generate a predicted „new‟ sector wage. The corresponding
fitted values, ln yap, constitute the predicted potential wages for stayers in the new industry.
Since the regression is undertaken on Xi which contains a set of characteristics unique to
each individual, each ln yap value would be unique. Therefore, in the case of stayers, the
predicted wages would be:
266
n-1 ˆ ln ya = ∑ Xβj / n-1, which is the average of predictions for the n-1 industries; and j=1 ˆ ln yb= Xβb for the original industry.
The Tomes and Robinson (1982a) method is preferred over that used by Osberg, Gordon
and Lin (1994) for the current study. The former method reflects the sectoral wage
differential in that there is an old sector wage and a potential new sector wage for each
individual. This treatment is consistent with the fact that mobility decisions are made in the
ex ante period. The latter method, where predicted wages were obtained from regressions
as per equations (9.1) and (9.2) for each mover/stayer subsample using individuals‟ actual
reported incomes, merely reflects the prevailing wage differential between movers and
stayers. That is, the new sector wage for stayers is based on the regression estimates for all
movers rather than being the average wage for all industries other than the stayer‟s original
industry.
At this stage, it should be noted that the modelling of the sectoral wages as per equations
(9.1) and (9.2) for inclusion in the mobility equation requires some identifying
restriction(s). This issue of variable identification was also raised in Tomes and Robinson
(1982a). Sectoral mobility, as per the main model in Table 9.10, is found to be independent
of marital status and occupational status. Although these two variables were placed into the
mobility equations, marital status was insignificant in all three regressions of the
unrestricted model and occupational status was insignificant in regression 3 (refer to section
9.6 below) as well as in the main model in Table 9.10. Thus, marital status and
occupational status influence sectoral mobility only via sectoral wages.
In addition, another form of identification arises from aggregation. Rewrite the predicted
wages for movers as:
ˆ ln ya
p = Xβa from the new industry; and
ˆ ln yb
p = Xβb from the old industry;
and for stayers as:
267
n-1 ˆ ln ya
p = ∑ Xβj / n-1; and
j=1 ˆ ln yb
p = Xβb for the original industry.
It can be seen that it is this averaging process, in addition to variable identification, which
gives rise to the low correlations between the predicted variables and the other variables
included in the mobility equation.
Predicted New Sector’s Unemployment Rate
The new sector‟s unemployment rate (Uat) can be derived in the same manner as outlined
above for wages via the industry-specific regression of Uat on the individual characteristics
for inter-industry movers. This gives the predicted unemployment rate (Uatp) for movers.
For stayers, the new industry is once again treated as all industries outside the original
industry and the corresponding Uat is the average rate across all industries other than the
stayer‟s original industry. For example, for stayers from commerce, Uat for the new
industry (non-commerce) is computed as:
[UNEMPnon-commerce / (UNEMPnon-commerce + EMPnon-commerce)] x 100,
where UNEMPnon-commerce and EMPnon-commerce are the levels of unemployment and
employment, respectively in all sectors outside commerce. It is noted that the employment
and unemployment data are obtained from the Korean NSO. Published data are used since
these can be considered to be the information available in the marketplace that rational
income-maximizing individuals will make use of to assess their potential mobility
outcomes. The regression of Uat on individual characteristics is undertaken to derive the
predicted rate (Uatp) for stayers.
Each mover/stayer will have a different predicted rate. For movers, the predicted
unemployment rate would be:
ˆ Uat
p = Xβa for the new industry; and
ˆ Ubt
p = Xβb for the original industry.
268
Uatp and Ubt
p for movers are derived via industry-specific regressions using the mover sub-
sample covering four waves of data in the KLIPS.
For stayers, the predicted unemployment rates are:
n-1 ˆ Uat
p = ∑ Xβj / n-1; which is the average of predictions for the n-1 industries; and
j=1 ˆ Ubt
p = Xβb for the industry of origin.
It is noted that the predicted unemployment rates are estimated from industry-specific
regressions using the stayer sub-sample covering four waves of data.
This approach to modelling unemployment rates for inclusion in the mobility equation
embodies the same form of variable identification as was used for wages in that marital
status and occupational status affect inter-sectoral mobility only via the sectoral
unemployment rates. The averaging process of Uat for stayers, another form of
identification, implies that perfect collinearity between the predicted variables and the other
determinants of mobility would never arise.
Predicted Expected Sectoral Wage Differential
With the predicted variables, Uatp, ln yai
p and ln ybi
p, the predicted expected sectoral wage
differential can be computed for each individual as:
ln(pya)p
- lnybp = ln [(1- Uat
p) x yai
p] - ln (ybi
p).
Actual versus Predicted Wage Differential
A comparison between actual (as per the original dataset in Table 9.6) and predicted wages
is undertaken to ensure that the algorithms used to derive the predicted variables have not
altered the basic patterns in the data. The key variable for comparison is the expected
sectoral wage differential. The mean values for the actual and predicted variables are
presented in Table 9.7. It is clear that the use of predicted variables in the empirical work
should not introduce any major distortions, as the mean values of the actual and predicted
variables are fairly similar in magnitude. Thus, the mean values of the actual and predicted
269
expected sectoral wage differential are both higher for movers than for stayers, conforming
to the theory that income-maximising individuals will switch sectors for the monetary
benefit. Moreover, it is observed that the predicted variables have lower standard errors
than the actual variables, attributed in large part to the fact that the stochastic elements
associated with actual wages have been removed.
Table 9.7 Actual versus Predicted Monetary Variables Actual Predicted
Movers
Mean
Standard
Deviation
Mean
Standard
Deviation
Expected New Sector Wages 4.43 0.66 4.37 0.28
Old Sector Wages 3.90 1.33 3.91 0.64
Unemployment Rate 5.43 3.44 5.43 3.06
Expected Sectoral Wage
Differential
0.52 1.35 0.45 0.60
Stayers
Expected New Sector Wages 4.63 0.66 4.81 0.04
Old Sector Wages 4.47 0.88 4.47 0.33
Unemployment Rate 3.88 2.71 3.88 1.84
Expected Sectoral Wage
Differential
0.16 0.76 0.34 0.34
Source: KLIPS dataset
Note: For ease of comparison between actual and predicted variables, the non-
weighted series is presented.
9.5.2 Sector-level Variables
Sectoral Unemployment, Size and Performance
Given the high correlation between the old-new sector-level variables mentioned earlier,
that arises primarily because stayers‟ old and new industries are the same, a rework of the
„new‟ sector variables (the lagged unemployment rate, sectoral size and performance, and
sectoral wage growth) for stayers is in order. Following Tomes and Robinson‟s (1982a)
treatment of the new destination as a single aggregated alternative, the new industry for the
stayers‟ sector-level variables will be all industries other than the original industry. The
sector-level data are obtained from the Korean NSO. As mentioned above, these can be
treated as the data available in the marketplace that income-maximizing agents make use of
in their mobility decisions13
.
270
The new sector‟s size is the average size of all industries other than the stayer‟s original
industry. The new sector‟s GDP growth and lagged unemployment rate are computed
using aggregates for all industries other than the stayer‟s original industry. For instance,
the growth in GDP at current prices across all industries other than agriculture in
period t (GDPnon-agriculture,t) over GDP in period t-1, (GDPnon-agriculture,t-1), is computed as:
[(GDPnon-agriculture,t - GDPnon-agriculture,t-1)/GDPnon-agriculture,t-1] x 100.
The average unemployment rate across all industries other than agriculture is computed as:
[UNEMPnon-agriculture /(UNEMPnon-agriculture + EMPnon-agriculture)] x 100,
where UNEMPnon-agriculture and EMPnon-agriculture each represent the unemployment level and
employment level in all industries other than agriculture. The pair-wise correlations
between these recomputed variables (marked with an asterisk) are substantially lower than
those initially presented: U*a,t-1 versus Ub,t-1 (0.460), size*a versus sizeb (-0.227) and GDP*a
versus GDPb (0.380).
New Sector’s Wage Growth
The descriptive statistics for the annual sectoral wage growth are given in the first two
columns of Table 9.6. Given that the data cover the Asian Financial Crisis period, it is
observed that all industries experienced negative wage growth in 1998. As the wage
growth variable is supposed to represent lifetime earnings, an individual would probably
look at „the industry average‟ rather than a year-on-year measure so that his/her decision
would be more fully informed.
Based on this principle, the wage growth variables were computed using information on the
previous five years. A 5-year period is chosen as it appears to be long enough to average
out the fluctuations observed in the data. Thus, for wave 1, the old/new industry wage
growth rate is computed based on the average annual compound growth rate (ACGR) over
the 1994-1998 period; for wave 2 over the 1995-1999 period; for wave 3 covering 1996-
2000; and for wave 4 extending from 1997 to 2001. For example, for T number of periods
(i.e. 5 in this study), the ACGR for the 1998 wages (y1998
) over the 1994 wages (y1994
) is
computed as follows:
271
ACGR = [(y1998
/ y1994
)1/(T-1)
- 1 ] x 100.
With regards to the new sector, the decision would be whether to move using information
about the new sector‟s future earnings. For each mover, the wage growth is based on the 5-
year average annual compound growth rate in the mover‟s new industry. In line with the
method used for the wage data, fitted values (g*p
at) from a regression of the new sector‟s
wage growth on individual characteristics (sex, age, marital/occupational/employment
status, educational attainment and tenure) are used. This method is similar to Willis and
Rosen (1979), where the wage growth functions for college and non-college attendees were
regressed on individual characteristics. The wage growth regressions are not corrected for
selection bias for the reasons mentioned above. The wage growth effects in the modelling
of worker mobility are identified through marital status and occupational status influencing
mobility only via the wage growth rates. This type of identification is again coupled with
that achieved through the averaging of the predictions for the larger stayers component of
the sample, as discussed in the next paragraph.
In the case of stayers, as their old and new sectors are the same in periods t-1 and t, the
Tomes and Robinson‟s (1982a) treatment of the new sector as an aggregated alternative is
employed. The new sector income is computed as the earnings for all industries other than
the stayer‟s original industry. Stayers for each different year (1998 till 2001) from each
„original‟ industry will have differing earnings. Thereafter, the 5-year average annual
compound growth is applied. For wave 1, the new industry wage growth rate is computed
over 1994-1998, wave 2, over 1995-1999, wave 3 over 1996-2000, and wave 4 over 1997-
2001. Predicted values, obtained from industry-specific regressions of average wage growth
on stayer‟s individual characteristics, are used in the model. In addition to variable
identification, the averaging process of the new wage growth rates for stayers suggests that
the problem of perfect collinearity in the mobility equation that includes the predicted wage
growth variable would be avoided.
Old Sector’s Wage Growth
The lifetime earnings potential in a mover‟s original industry is proxied by using the
average annual wage growth of that old industry over the last five years. Again, fitted
272
values (g*p
bt), from the regression of the old sector‟s wage growth on individual
characteristics, are used in the estimations.
In the case of stayers, since their old and new industries are the same, their annual wage
growth, derived from their reported earnings in period t-1 and period t, is indicative of the
old sector‟s wage growth. It is noted that the use of individual data is in tandem with Willis
and Rosen‟s (1979) estimation of lifetime earnings (conditioned on actual school choices in
the U.S.), which made use of each individual‟s reported initial and latest earnings to
compute a wage growth variable. However, there are anomalies associated with the use of
individual data in the present application, and these will be addressed below.
The average wage growth was an astounding 127.66%, with a large standard deviation of
258.87. These values could perhaps be explained by considering the data period covered,
which is just after the onset of the Asian Financial Crisis when the Korean labour market
underwent tremendous adjustments. These adjustments can be seen in the pattern in the
average wage growth over time. As an example, whilst the average growth was 110.80% in
1998 and 146.96% in 1999, it later tapered to 84.01% in 2000 and 34.00% by 2001. On the
one hand, there are workers who experience phenomenal recovery in actual wages after
apparently suffering major income setbacks during the Crisis period. On the other hand,
there are workers with negative wage growth who failed to recover from the Crisis. Such
outliers should be discarded for the purpose of modelling lifetime wage growth, as this
wage growth would be expected to follow a reasonably steady pattern, and should not
reflect temporary surges or dips arising from external disturbances. The outliers should be
progressively excluded until a reliable growth pattern is achieved.
Given that the 5-year industry average annual wage growth over 1998-2001 using
aggregate-level data is about 6-7%, the KLIPS sample unit-record data spanning all
industries should have a similar average wage growth. To achieve a reliable growth pattern
across respondents in the KLIPS, the top 10% and bottom 5% of outliers had to be removed
in the calculation of lifetime wages. It is noted that more observations from the high-end
growth distribution are removed since the magnitude of wage increases exceeded that of the
decreases, and there are more respondents with positive growth. Additionally, workers who
changed job status between periods were removed in the computation of lifetime wages as
273
their inclusion could distort the annual wage growth rate. These cover a change from part-
time to full-time work and vice versa, or from regular (job contracts exceeding 1 month) to
irregular work (include job contracts of less than 1 month, including daily-rated work) and
vice versa14
.
Removing the top 10%, bottom 5% and individuals with a changed job status gives an
average wage growth of 6.20 with a standard deviation of 25.92 for industry stayers15
. The
average growth in the sample now reflects the industry average of 6-7% with aggregate-
level data. The annual wage growth for each industry was then regressed on the personal
attributes of this sub-set of stayers. The estimated regression equation was then used to
predict a wage for all industry stayers. This enables maximum usage of the dataset. The
average predicted wage growth rate from this exercise is 6.12%, with a much lower
standard deviation of 3.45. This approach to constructing a wage growth variable for
inclusion in the mobility equation identifies the effects of the old sector‟s wage growth via
exclusion restrictions (marital status and occupational status) and through having multiple
(for each industry) equations for generating the predicted variables.
A sensitivity test was conducted to ensure the regression results of the main model
presented later in Table 9.10 are not sensitive to changes in the sample of stayers used to
construct the wage growth variable. An extra 1% of observations at the upper end of the
distribution were removed. The average predicted wage growth rate was then computed
again, and the inclusion of this alternative measure in the main model resulted only in slight
changes to the regression coefficients (of around 1 decimal point). Thus, the results are
robust to this change in the sample used for the underlying regression for the wage growth
calculations for industry stayers.
With this combination of past industry data and the individual-level variables, the pair-wise
correlation coefficient between g*p
at and g*p
bt is 0.007, which is much lower than that for
the aggregate-level, of 0.889.
274
9.5.3 Descriptive Statistics of Predicted/Recomputed Variables
Having derived predicted monetary variables and recomputed the sector-level variables, a
re-look at the descriptive statistics that are to be used in the main model is in order. For
comparative purposes, the non-weighted and weighted series are presented. Similar to the
case for actual data, the design effect is less than 1 for most predicted/recomputed
variables. However, it is usually not much less than 1, implying that little is lost by
focusing on the unweighted data. The description in this section therefore applies to the
non-weighted series.
As in the case for actual variables, the average predicted expected new sector‟s wage was
higher than the average predicted old sector‟s wage. With regards to lifetime earnings, the
average predicted wage growth in the new sector (5.72%) is slightly less than that of the old
sector (6.12%), a pattern different from when the actual variables (as per the first two
columns of Table 9.6) are used. A breakdown revealed that this pattern applied to industry
stayers only. Since stayers constitute most of the sample, this pattern is to some extent the
result of averaging over five years for the new sector compared with the two years for the
old sector. It could also be due to the fact that an arbitrary decision was made by leaving
out the top 10% and bottom 5% to compute the sample 2-year average wage growth in the
old sector, and the descriptive statistics simply reflect the conservative approach taken in
this regard.
The average lagged annual unemployment rate was also slightly higher for the new sectors
when the recomputed rates were used. This is a similar scenario to when the actual
variables were used, suggesting again that a move to the high-wage new sector involves a
trade off for higher unemployment. Conforming to the pattern of Table 9.6, the
recomputed new sector size is smaller than the old sector size, and the recomputed new
sector‟s average growth rate is lower than the old sector‟s, once again suggesting that the
impact of the sector‟s past performance on mobility may not be as great.
275
Table 9.8 Means and Standard Deviations for Predicted and Recomputed Variables
Mean
Standard
Deviation Mean
Standard
Deviation
___
√deff
Monetary variables Non-weighted series Weighted series
ln(pya)p 4.71 0.23 4.74 0.17 0.739
lnybp 4.34 0.48 4.38 0.38 0.792
g*p
at (%) 5.72 0.90 5.74 0.77 0.856
g*p
bt (%) 6.12 3.46 6.15 3.61 1.043
Sector-level variables
U*a,t-1 (%) 4.64 2.12 5.49 1.04 0.4911
Ub,t-1 (%) 3.82 2.78 4.57 2.36 0.849
size*a (no.) 2,432 950 2,365 784 0.825
sizeb (no.) 3,556 1,573 3,555 1,578 1.003
GDP*a (%) 4.59 5.93 6.69 2.63 0.4441
GDPb (%) 5.36 8.33 7.76 5.93 0.712
Sample size 10,691 10,691
Source: KLIPS dataset. Annotations and description of variables are in Appendix 9B.
1. Compared to the weighted series, the high standard deviation for the non-weighted
U*a,t-1 and GDP*a series (and hence their lower design effects) reflects the higher
values in wave 2, the period following the Crisis when Korean workers are in the
initial stages of adjustment. It is noted that the design effects were closer to 1 for the
non-predicted series since aggregate-level data was used.
It can be seen that the predicted/recomputed statistics (non-weighted series) are generally
consistent with the actual variables. Thus, the derivations of predicted and recomputed
variables has not altered the way the mean monetary, macroeconomic and industry
characteristics differ in terms of the old-new sector comparison.
9.6 EMPIRICAL ANALYSIS: DETERMINANTS OF SECTORAL
MOBILITY
Having established the extended Le and Miller (1998) model, a reliable dataset and derived
variables to accommodate econometric issues associated with the original data, the
statistical analysis can be conducted. Prior to examining the determinants of sectoral
mobility, an attempt has to be made to arrive at the main model from the most general
model suggested by the review of past studies. There are three issues: (i) moving from an
unrestricted to a restricted model; (ii) determining if the weighted results should be used;
and (iii) deciding on the most suitable measure of sectoral shock.
Prior to the conduct of unrestricted-to-restricted modelling, a correlation matrix was
computed for all explanatory variables to detect multicollinearity. It was found that the
276
overall GDP growth rate was highly correlated with the new sector‟s performance and
lagged unemployment rate, with a correlation coefficient of at least 0.8. The other
variables were not highly correlated, having correlation coefficients of less than 0.6. To
minimize potential multicollinearity, and given the data shortcomings in that there is a
discrepancy in the time periods [i.e. the GDP growth rate is estimated at year-end (January
till December) whilst mobility (and annual employment) is estimated at mid-year (June
year t-1 till June year t)], and that GDP growth, being measured at yearly intervals, will not
capture any shorter-term cyclical effects, this variable will be excluded from subsequent
specifications.
Table 9.9 lists the estimated coefficients from an unrestricted model, using alternative
measures of a sectoral shock. This unrestricted model is based on a non-linear mobility
relationship for both age and job tenure. The first set of estimates in the left-hand panel is
based on the residual of the industry-specific AR(1) regression (micro level). Those in the
middle panel are based on the residual of the industry-specific AR(1) regression (by wave),
which was adopted by Jovanovic and Moffitt (1990) for the U.S. The third set in the right-
hand side panel is based on the cross-sectoral standard error of residuals from the industry-
specific AR(1) regression16
. It should be noted that the three models presented in Table 9.9
are non-nested and this may make comparison of the models difficult. While a nested
model can include all three shock regressors in the estimating equation, collinearity,
especially between the AR(1) (by wave) and AR(1) (micro level) residuals, prevents this17
.
The determination of the main model and the appropriate shock regressor will have to be
based on the number of significant regressors in the model as well as the overall fit of the
model.
Before the main model is determined, an attempt is made to weight the results and compare
the weighted and non-weighted results. Although the coefficient estimates will not be
seriously affected with complex sampling in large samples, the t-statistic, confidence
intervals and model selection will be biased if these complexities are not taken into
account. The t-statistic (tβ‟) under the complex KLIPS sample can be computed as:
___
tβ‟ = tβ /√deff
277
where tβ is the t-statistic for a regression estimate of β under the assumption of simple
random sampling18
. The weighted t-values are presented in Table 9.9. The comparison is
thus based on the statistical significance of the weighted and non-weighted t-values.
It can be seen that the statistical significance of the explanatory variables does not alter
whether the weighted or non-weighted t-statistic was used. This applies to all three
regressions. The design effect appears to have a modest effect at best. From here, the
logit estimates are therefore based on the non-weighted series. The use of non-weighted
KLIPS data concurs with previous studies, including Nam (2007), Kim (2004a), Cho
(2005), Sawangfa (2007), Kim (2003), Young (2005), Kang (2004), Kim (2004b), Young
(2006), Chang and Yang (2007), Seong (2007), Son (2007a), Son (2007b) and Jung, Moon
and Hahm (2007).
When the residual of an AR(1) regression (micro level) of the individual industry‟s
employment is used, the t-test indicates that sex, age, age-squared, marital status,
employment status, the new sector‟s wage growth and old sector performance did not
significantly influence the probability of a sectoral move at the 5% level. In addition to the
first five variables, the new sector‟s performance and tenure-squared were shown by the
corresponding t-statistic to be insignificant when the AR(1) residual (by wave) was used in
the regression. When the cross-sectoral standard error of residuals from the sector-specific
AR(1) regression of the natural logarithm of annual employment was used, only
occupational status and marital status were not significant influences (at the 5% level) on
the probability of a worker being classified as an inter-sector mover.
Table 9.9 Unrestricted Model: Logit Regression on Probability of Sectoral/Industrial Mobility Variable Regression 1 Regression 2 Regression 3
Iim and Iif are the latent (i.e. unobserved) variables for the probability of a sectoral move for
males and females, respectively. The indices, I*
im and I*if, are the observed dependent
variables for males and females, respectively. They indicate whether a sectoral move has
taken place. I*im and I
*if take the value 1 if male/female workers changed sectors, and the
value 0 if a change did not occur. These observed variables are linked to their
corresponding latent indices as shown below:
I*im = 0 if Iim < 0;
= 1 if Iim ≥ 0; and
I*if = 0 if Iif < 0;
= 1 if Iif ≥ 0.
The explanatory variables are as described earlier, with the m and f subscripts each
denoting male and female workers. The terms vim and vif denote the stochastic disturbances
for males and females. As in the estimation for the pooled dataset, the logit method of
estimation will be used in this disaggregated analysis.
The sample dataset for males and females are subsets of the pooled dataset. There are
6,906 person-year observations for the regression for males and 3,785 person-year
305
observations for the regression for females. The list of data items for males and females is
given in Appendix 10A. The general characteristics of the data follow that of the pooled
data in that the re-computed sector-level variables exhibit lower correlations than those
displayed for the original variables2.
10.3 VALIDITY OF POOLING THE DATASET
Prior to the analysis, it would be worthwhile to test if the male and female samples should
be pooled into a single sample. The approach is to make use of a gender dummy variable
(F = 1 for females, F = 0 for males) to test if the individual coefficients differ between the
gender groups. A chi-squared test of whether all coefficients for males are simultaneously
different from the respective coefficients for females is also presented. Thus, to test if the
mobility relationship is the same for both gender groups, the female dummy variable (F) is
inserted in the general equation together with a series of interaction terms (FX‟s):
Ii = γ1 + γ*1F + (β1X1 + β2X2 +……….βkXk) + (β
*1FX1 + β
*2FX2 +……….β
*kFXk) + ui
It can readily be seen that the estimated models for each gender group are:
Males: Ii = γ1 + (β1X1 + β2X2 +……….βkXk) + ui
Females: Ii = (γ1 + γ*1) + (β1+β
*1)X1+ (β2+β
*2)X2 + …….. (βk+β
*k)Xk + ui
Whilst the β‟s will be the estimated coefficients for males, the (β+β*)‟s will be the
coefficients for females. The t-tests of the null hypotheses H0: β*‟
s = 0 will indicate if the
individual slope coefficients for females differ from those for males. The t-test of the null
hypothesis H0: γ*1 =0 will show if the intercepts in the male and female mobility
regressions are similar.
306
Table 10.1 Logistic Regression of „Full‟ Model Variable Coefficient Standard
Error
t-value
Constant -16.844 2.177 -7.737
ln(pya)p-lnyb
p 1.181 0.259 4.560
g*p
at 0.821 0.047 17.468
g*p
bt -0.400 0.022 -18.182
U*a,t-1 -0.713 0.065 -10.969
Ub,t-1p 0.384 0.050 7.680
∆ GDP 0.484 0.067 7.224
AGE -0.055 0.010 -5.500
TENURE -0.067 0.011 -6.091
MS (Non-married) 1.435 0.233 6.159
HEAD (Non-heads) 1.175 0.210 5.595
EDA (Non-graduates) 0.330 0.243 1.358*
OCC (Non-professionals, non-associate
professionals) -2.030 0.453 -4.481
ES (Non-employees) 0.278 0.234 1.188*
SIZE b/1000 1.130 0.247 4.575
SIZE*a/1000 3.410 0.414 8.237
∆ GDPb 0.025 0.018 1.389*
∆ GDP*a -0.415 0.034 -12.206
SHOCK 17.924 2.653 6.756
Female Dummy (F) -10.949 2.313 -4.734
F x [ln(pya)p-lnyb
p] 1.579 0.491 3.216
F x g*pat 0.606 0.148 4.095
Fx g*p
bt 0.246 0.039 6.308
F x U*a,t-1 -1.852 0.244 -7.590
F x Ub,t-1 0.169 0.139 1.216*
F x ∆ GDP 0.715 0.151 4.735
F x AGE 0.006 0.020 0.300*
F x TENURE 0.059 0.024 2.458
F x MS (Non-married) -0.471 0.394 -1.195*
F x HEAD (Non-heads) -0.143 0.437 -0.327*
F x EDA (Non-graduates) 0.311 0.433 0.718*
F x OCC (Non-professionals, non-
associate professionals) 0.839 0.843 0.995*
F x ES (Non-employees) 0.772 0.435 1.775
F x SIZEb/1000 -3.551 0.474 -7.492
F x SIZE*a/1000 0.567 0.798 0.711*
F x ∆ GDPb -0.051 0.032 -1.594*
F x ∆ GDP*a -0.154 0.067 -2.299
F x SHOCK 57.356 6.973 8.225
Chi-square statistic 10,044.774
Nagelkerke R-squared 0.922
Sample (6,906 males and 3,785 females) 10,691 Source: Pooled KLIPS dataset. This dataset differs from the pooled dataset in the previous chapter as several variables were constructed separately for the male and female samples: monetary variables, lagged unemployment rates, sectoral sizes and sectoral shock.
* insignificant at 10% level.
307
Table 10.1 shows the t-statistic for each coefficient. The intercepts in the male and female
regressions are clearly different. There are eight gender interaction terms that are
insignificant at the 10% level, and ten statistically significant interaction variables. In
particular, the individual coefficients for females that are significantly different from the
corresponding coefficient for males are the three monetary variables, lagged new sector‟s
unemployment rate, GDP growth, employee status, tenure, old sector size, new sector
performance and sectoral shock. Furthermore, the chi-square statistic for the test that the
female dummy and interaction terms can be excluded from the model is 550. This exceeds
the critical value and so the hypothesis that there is no significant difference between the
models of worker mobility for males and females is clearly rejected. The industrial
mobility relationship should therefore be estimated separately for males and females.
10.4 DESCRIPTIVE STATISTICS FOR MALES AND FEMALES
This section outlines the similarities and differences between male and female workers in
Korea in terms of the monetary and macroeconomic variables, and worker and job
characteristics that are the basis of the model of industrial mobility. As noted in the
introduction, differences between males and females in these variables may contribute to an
understanding of the reasons for any gender differences in the sectoral mobility of the two
groups.
The means and standard deviations for the KLIPS male sample of 6,906 observations and
the female sample of 3,785 observations covering the four job waves (1998 till 2001) are
listed in Table 10.2. A „t‟ statistic for the test of whether the mean values for each male
and female variable are significantly different from each other is reported in the last column
of the table3.
As in the case of the pooled sample, the majority of male and female workers are industry
stayers, with the mean mobility rates being 23.5% for males and 22.3% for females.
Although the share of movers for men and women do not differ significantly, there could be
differences in the means of the explanatory variables between men and women. If this is
the case, it implies that there must be differences in the estimated coefficients of the
308
individual and job characteristic, since these differences in behavioural patterns are needed
to offset differences in mean values of characteristics to give the similar mean rates of
sectoral mobility for males and females.
Table 10.2 Means and Standard Deviations for Male and Female Workers, Aged 20-64 years Males Females tβ
Mean (or %)
Standard Deviation
Mean (or %)
Standard Deviation
statistic
Monetary variables
Ln (Expected New Industry Wage) 4.82 0.28 4.65 0.69 -17.20*
Ln (Original Industry Wage) 4.13 2.25 4.01 0.58 -3.10*
Growth Rate of New Industry Wage (%) 3.13 3.65 6.74 1.76 57.23*
Growth Rate of Original Industry Wage (%) 13.50 6.33 10.60 5.42 -23.81*
Macroeconomic variables
Unemployment Rate in New Industry in Period t-1 (%) 5.07 2.39 3.97 1.57 -25.37*
Unemployment Rate in Original Industry in
Period t-1(%) 4.20 2.98 3.23 1.79 -18.27*
GDP Growth Rate (%) 4.21 4.17 4.81 4.09 7.10*
Worker characteristics
Industry Mover (%) 23.5 42.40 22.3 41.63 -1.38
Age at Former Interview (yrs) 40.44 10.19 38.31 11.33 -9.95*
Original Job Tenure (yrs) 7.88 8.34 5.72 7.80 -13.09*
ES (Non-employees) 0.218 0.974* 3.92 1.112 3.669 19.27
SIZEb/1000 0.492 1.940+ 8.85 -2.855 -8.378 -49.46
SIZE*a/1000 2.937 7.702 52.79 3.150 6.729 54.58
∆ GDP*b 0.057 3.706 1.02 0.158 7.871 2.74
∆ GDP*a -0.240 -12.255 -4.32 -0.315 -8.583 -5.46
SHOCK 20.221 6.591 15.47 74.706 13.082 58.05
Nagelkerke R-squared 0.912 0.908
Chi-square statistic (16) 6,426.520 3,410.659
Sample size 6,906 3,785
* insignificant at 5% level.
+ significant at 10% level.
n.a. : not applicable
Note : For the SHOCK variable, the elasticity measure is used. The elasticity of mobility w.r.t. a change in
the sectoral shock is measured as: elasticity = marginal effect / p = β x p(1-p)/p. For males, elasticity =
20.221 x 0.235(0.765)/0.235 = 15.47. For females, elasticity = 74.706 x 0.223(0.777)/0.223 = 58.05.
Lifetime Earnings
For the pooled sample analyses of chapter 9, the effects of the individuals‟ lifetime earnings
streams, as captured by the wage growth terms, were found to differ between sectors. The
probability of moving was raised by higher wage growth in the new sector. For the original
sector‟s wage growth, a move out of the original sector was less likely if the original
sector‟s permanent earnings were higher. The separate analyses undertaken by gender
mirrored these results.
Male workers were more likely to move to the new sector for higher permanent earnings.
For a unit increase in income growth, male mobility increased by 13 percentage points.
Similarly, lifetime earnings in the new sector exerted a positive impact on female mobility
315
behaviour. Females were 19 percentage points more likely to enter into the new sector for
every unit increase in wage growth.
For the original sector, lower lifetime earnings induce male workers to change sectors. For
every unit decrease in income growth, men are 7.33 percentage points more likely to move.
The deterrent effect was also evident among women: women were 3.97 percentage points
more likely to change sectors for every unit decline in the growth rate.
From this examination of the results for the monetary variables, it can be concluded that
both men and women are motivated by the initial monetary gains and are equally
responsive to the prospect of earning higher lifetime incomes in the new sector. The extent
of the impact of lifetime income is also greater in the new sector than the old sector,
judging from the magnitude of the parameter estimates and the marginal effects. The
gender analyses therefore supports one of the predictions of the Le and Miller (1998)
model, that individuals move to alternative employment states in anticipation of higher
lifetime earnings. It also confirms the expectation that higher earnings in the new sector act
as a pull factor and lower earnings in the original industry act as a push factor for male and
female mobility. The consistency of the findings for males and females is reassuring, from
the perspective of informing on the robustness of the results.
10.5.2 Macroeconomic Variables
The macroeconomic variables included in this empirical analysis are the unemployment
rates in the original and new industries. The results obtained for the study of males and
females in the Korean labour market do not appear to have any direct counterparts in the
literature; hence only comparisons with general, aggregate-level, studies can be provided
below.
Sectoral Unemployment Rate
In the analysis of chapter 9, where the sample was pooled across males and females, the
original sector‟s unemployment rate had a positive effect on sectoral mobility. Similarly, in
316
the analyses disaggregated by gender, the old sector‟s unemployment rate had a positive
impact on the likelihood of both males and females moving across sectors. For males, the
marginal effect of a one percentage point increase in the unemployment rate was 8.57
percentage points. For females, the marginal effect was 17.97 percentage points. Thus, it
appears that the original sector‟s unemployment rate acts as a push factor of mobility.
For the new sector‟s lagged unemployment rate, the separate analyses for males and
females replicate the overall sample result that the odds of a move to the new sector are
lowered the higher the new sector‟s unemployment rate. For male workers the marginal
effect was 6.58 percentage points. For female workers, the marginal effect was 24.27
percentage points for every one percent increase in the unemployment rate8. The Table
10.3 results are consistent with the Todarian hypothesis, which asserts a negative
relationship between the unemployment rate and the probability of gaining employment.
These findings for males and females are internally consistent in that lower job
opportunities lead to out-mobility from the old sector and deter mobility into the new
sector. The results for Ub,t-1 (U*a,t-1) also held when U*a,t-1 (Ub,t-1) was omitted from the
estimating gender equations. These features, and the similarity with the findings reported
in chapter 9, point to the results with respect to the macroeconomic variables being quite
robust.
10.5.3 Worker Characteristics
Gender differences in the impact of a number of worker characteristics, i.e. age, job tenure,
family indicators, employment/occupational status and educational attainment variables, are
of interest. Several recent studies have focused on these issues, and they afford a basis,
albeit limited, for comparison in this section. Where applicable, these comparisons will be
highlighted.
Age
The analyses conducted for the pooled sample indicated a non-linear, negative age-mobility
relationship. As noted above, the preliminary examination of the data indicated that the
317
mobility-age relationship for the separate male and female samples were linear. Consistent
with the aggregate-level analysis, however, these relationships were also negative. Among
men, the chances of moving decreased by 0.84 percentage points per additional year of age.
Among women, the probability of moving sectors declined by 0.74 percentage points for
every extra year of age. These gender findings correspond with the negative age-mobility
correlation for males in Osberg, Gordon and Lin (1994) and for females in 1985/1986 in
Osberg (1991). As noted before, this relationship is generally held to arise as older workers
have a shorter working period over which they can derive benefits from a different job
[Creedy and Thomas (1982)] and, due to the experience and knowledge they have acquired
in the original job/sector, they face greater costs in moving [Jovanovic and Moffitt (1990)].
Job Tenure
The analysis of the data pooled across the male and female samples revealed a non-linear,
generally negative tenure-mobility relation. When the analyses were undertaken separately
for males and females, however, the relationship between mobility and job tenure for males
was linear and negative, while that for females was statistically insignificant. However, for
consistency with the analyses for males, a linear specification was used in the estimating
equation for females.
Male workers with lengthy tenures were less likely to switch sectors, with their propensity
to move reducing by 1.10 percentage points per extra year of tenure. This finding conforms
with other studies for males: Osberg (1991), Osberg, Gordon and Lin (1994) and Neal
(1995) all reported a negative tenure effect. However, the insignificant tenure effect for
females differs from Osberg‟s (1991) study, which showed females to have lower mobility
probabilities with rising tenure.
Various suggestions for the difference in the impact of tenure on the mobility behaviour of
males and females can be advanced. These include a greater importance of firm-specific
training for male workers than for female workers, and, in general, a higher opportunity
cost of moving sectors for male workers than for their female counterparts. This could
arise from a higher proportion of older male workers with lengthy tenures holding senior
318
positions, and the high wages of their senior positions, in addition to other non-pecuniary
benefits, may not be readily available in the new sector.
Family Indicators: Marital Status and Household Head
The preliminary analyses in chapter 9 of the data pooled across males and females showed
that marital status was not a significant determinant of the propensity to change sectors.
The marital status variable was subsequently omitted from the estimating equation. The
separate analyses conducted for men and women, however, revealed marital status to be a
significant determinant of mobility. Whilst married men were 24.67 percentage points less
likely than their non-married counterparts to switch industries, females were 16.62
percentage points less likely than their non-married counterparts to change sectors. This
result for males is in line with Neal (1995), who shows married men to have lower
propensities of moving sectors. However, it differs from results reported by Osberg (1991)
for 1982/1983 and 1985/1986, and Osberg, Gordon and Lin (1994), where marital status
did not have a significant effect on sectoral mobility. The finding for females in the current
study, to the extent that married women have greater household responsibilities, comes
across as intuitively reasonable. However, the finding for females does not concur with
Osberg‟s (1991) study, where marital status did not impact on female sectoral mobility in
any of the three phases, 1980/1981, 1982/1983 and 1985/1986, for which the statistical
analyses were undertaken.
In the analysis for the entire sample of workers, heads of households had a lower incidence
of industrial mobility. Given the considerably higher proportion of male heads in the
sample, it is not surprising that this impact is mirrored in male mobility behaviour. Male
heads were 20.15 percentage points less likely to change sectors than males who were not
the household head. This finding aligns with the study of Fallick (1993), which revealed
unemployed male heads to have a lower incidence of industrial mobility, and supports the
view of household heads facing greater risks from an industrial switch arising from their
family responsibilities. Household head status also showed up as a significant deterrent of
female mobility, with the marginal effect being 21.58 percentage points. This could be due
to the fact that females heading households are single parents who are not able to afford to
incur the risks of changing sectors. A breakdown of the data revealed that nearly three-
319
fifths of the 840 female household heads in the sample were single, divorced, widowed or
separated.
Educational Attainment
In the previous chapter where the sample was pooled across male and female workers,
educational attainment was shown to have a significant impact on sectoral mobility, with
graduates being less likely to switch sectors. The preliminary analyses conducted for males
and females revealed that little, if any, of the variation in male or female mobility could be
attributed to educational attainment. The insignificant education variable was therefore
omitted from the estimating equation used in this chapter. The finding for males reported
here is consistent with Osberg, Gordon and Lin (1994) and Neal (1995). In this earlier
study, male elementary/diploma and university degree holders had similar rates of mobility.
There is no readily apparent reason for the different findings for the aggregate-level
analysis and these estimations undertaken on the separate male and female datasets, other
than the size of the sample. In this regard it is noted that the „t‟ on the educational
attainment variable in the aggregate-level analysis was typically around 2, and a halving of
the sample would itself reduce this value by around one-third in the preliminary analysis.
Occupational Status
Occupational status was shown to be an insignificant determinant of mobility in the overall
sample in chapter 9, and the variable was omitted from main set of analyses in that chapter.
The disaggregated gender analysis revealed a similar finding for women. Thus, among
Korean females, occupational status did not exert any impact on gender mobility. This
finding is consistent with the some of the results reported by Osberg (1991), where the
likelihood of a sectoral change was not significantly different for women in clerical and
sales occupations in 1982/1983 and for female managers, professionals and technicians in
1980/1981.
In the analyses undertaken separately in this thesis for males, however, skilled workers had
a higher likelihood of moving sectors, with their chances being 38.25 percentage points
greater than that of their unskilled counterparts. This finding is consistent with the results
320
in Osberg (1991), where men in the higher-skilled managerial/professional/technical
occupations were found to have a higher probability of mobility for 1980/1981. This
finding for males in Korea supports the view of skill levels being vital to certain industries‟
operations [Neal (1995)] and workers with vital skills will be more likely to switch to
industries requiring such skills. Also, the finding supports the idea of skilled workers being
scouted for their talent and productivity [Murphy and Topel (1990)] and they will
consequently have a higher likelihood of switching sectors. The finding for females,
however, reflects mobility stickiness. Skilled females could be in more specialized
occupations which limit their range of alternatives and thus results in limited mobility.
Employment Status
In the analysis of the data pooled across males and females, non-employees had a greater
likelihood than employees of moving to new sectors/industries. This finding was replicated
in the analysis for females. Specifically, females who were non-employees were 19.27
percentage points more likely than employees to move across sectors. Among males,
however, employment status did not have a significant effect on mobility. It is not clear
why this variable should have different effects for men and women. The Asian Financial
Crisis adversely affected businesses and caused many employers/business owners to close
down. It could be that female employers/business owners were more likely to be in small
firms: such businesses have emerged as significant avenues for the economic empowerment
of women in the Asia-Pacific, as their flexibility in operations with minimum technology
and capital start-ups, and family-based nature favour women‟s decision to participate in the
labour force9. With limited access to resources and business networks, however, these
small businesses may have been more vulnerable and the hardest hit by the Crisis. The
mobility literature does not provide any evidence in this regard.
10.5.4 Industry Characteristics
The industry characteristics considered for inclusion in the model of sectoral mobility are
industry size and sectoral performance. Whilst the findings linking industry size to sectoral
mobility in the separate samples of males and females can be compared with findings from
the empirical literature, no comparison appears possible for sectoral performance.
321
Sectoral Size
In the previous chapter, the aggregate-level analyses revealed that a larger size of the
original sector reduced the probability of a sectoral move. This result carried over to the
separate study of female workers. The elasticity of female mobility with respect to an
increase in the size of the original sector was -2.22. In contrast, the original sector size
increased the likelihood of male mobility, with the corresponding elasticity of mobility at
0.38. The finding for males contradicts that in Neal (1995), where it was reported that the
industry size had a negative effect on mobility (although the study pertains to displaced
workers).
To account for the contrasting results between men and women, the female variable for the
original sectoral size was placed into the estimating equation for males. The coefficient of
this alternative variable was negative but insignificant, compared to the positive one when
the male variable was used. This suggests that a reason for the conflicting results is the
gender differences in the industrial composition, with males being concentrated in the
construction sector and females being concentrated in the commerce sector and in
community, social and personal services industries. In other words, it is not so much size
per se that is important, but it is the size of particular sectors of the economy.
The aggregate-level analyses also indicated that the probability of a sectoral move was
higher the larger the size of the new sector, and it was suggested that Korean workers
moved for greater employment opportunities. The separate analyses for males and females
reinforce this result. The elasticity of mobility in response to an increase in the new sector
size for females (2.45) was slightly larger than for males (2.25). This finding corresponds
with Osberg, Gordon and Lin (1994), where a positive association between male mobility
and industry size was reported, and it portrays the idea of workers moving in response to
employment availabilities in the new sector.
Sectoral Performance
The GDP by sector variable cannot be constructed separately for men and women. Hence
differences between men and women in this variable will only reflect gender differences in
322
the distribution of workers across industries. The earlier study of the overall sample showed
that stronger growth in the original sector raised the likelihood of a sectoral move.
Likewise, the gender studies replicated this finding, with higher growth in the original
sector increasing the probabilities of both male and female mobility, with the marginal
effects being 1.02 percentage points for males and 2.74 percentage points for females. As
in the case of the result from the study of the pooled sample, it appears that both male and
female mobility support the jobless growth hypothesis. This may be associated with the
technological advances in Korea, where the high-performing original sector has limited job
vacancies associated with the growth thereby leading to higher out-mobility rates.
For the new sector, better performance reduced the likelihood of a sectoral switch in the
aggregate-level analyses, and the disaggregated analyses conducted here revealed similar
results for both males and females. That is, a higher (lower) GDP growth rate reported in
the new industry decreased (increased) the probability of moving industries for males and
females. The marginal effects were -4.32 percentage points for males and -5.46 percentage
points for females. These findings are consistent with the jobless growth hypothesis, where
the high-performing new sectors with technological advancements have fewer job
opportunities, and hence the chances of a sectoral switch to the new sectors are lowered.
10.5.5 Sectoral Shock
The effect of the sectoral shock was large, positive, and significant for the overall
workforce. This result was mirrored in the analyses conducted for the separate samples of
males and females. The elasticity of mobility with respect to a change in the sectoral shock
variable was 15.47 for males and 58.05 for females. This finding for males is similar to the
result reported by Jovanovic and Moffitt (1990), who demonstrated the sectoral shock to
have a positive impact on the probability of male sectoral mobility10
. When this exogenous
sector shock variable was omitted from the regressions, the fit under the male model
dropped from 0.912 to 0.905, and that under the female model dropped from 0.908 to
0.761. It can thus be seen that the effect of the sectoral shock variable is greater among
female workers than among male workers, which is also reflected in the gender difference
in the coefficients of the sectoral shock variable given in Table 10.3.
323
10.6 A GENDER PERSPECTIVE ON THEORIES OF SECTORAL MOBILITY
The objective of this section is to assess the empirical relevance of the three theories of
sectoral mobility outlined earlier to the separate samples of male and female workers. The
tests of null hypothesis under each theory are the same as in the previous chapter.
10.6.1 Worker-Employer Mismatch Theory
With the regression for males, the t-statistic for the coefficient of the industrial shock, θ,
was 6.60. With the regression for females, the t-statistic for H0: θ = 0 was 13.08. Thus, the
null hypothesis of H0: θ = 0 is rejected for both gender groups. That is, the sectoral shock
is a significant factor in accounting for both male and female mobility patterns. For the
Korean workforce, gender models that are based on the worker-employer mismatch theory
and disregard the sectoral shock effect on mobility would be inadequate in accounting for
sectoral labour movements.
10.6.2 Sectoral Shock Theory
From the results displayed in Table 10.3, fourteen variables in the equation for males are
significantly different from zero at the 5% level. The chi-square statistics for the test of
whether all the non-sectoral shock variables add to the explanatory power of the model was
4,339. This exceeds the critical value and so the null hypothesis that these non-sectoral
shock variables are not important is rejected. The pure sectoral shock theory cannot be
applied to the study of male mobility. For the regression for females, there were fourteen
significant variables at the 5% level. The test of the null that all the non-sectoral shock
variables did not contribute to the explanatory power of the model yielded a test statistic of
1,599, which is far higher than the critical value. As in the case of the study of male
mobility, the pure sectoral shock theory is not applicable to the analysis of female mobility
behaviour.
324
10.6.3 Bridging Theory
Table 10.3 shows that the monetary, macroeconomic, demographic, socio-economic and
sectoral shock variables are significant in explaining male and female mobility. The
mismatch theory (i.e. testing H0 : θ = 0) and sectoral shock theory (i.e. testing H0 : β1 = β2
….. = βk = 0) are rejected for both groups and it appears that gender movements are best
described using the bridging theory.
The validity of these results was checked using the approach along the lines of the
Jovanovic and Moffitt‟s (1990) method that was discussed in Chapter 9. This involved
regressing the probability of a sectoral move on the standard error of the log wage
distribution and the sectoral shock. The wage and sectoral shock measures were significant
for both the male and female regressions, as shown in Table 10.4. It can be concluded,
therefore, that the bridging theory of sectoral mobility, that was previously held to apply to
the pooled data regression, applies also to the study of male and female mobility.
Table 10.4 Logistic Regression of Sectoral/Industrial Mobility on the Standard
Error of Wage Distribution and Sectoral Shock for Males and Females Variable Coefficient t-statistic
Male Regression
Constant -7.714 -42.810 Standard Error of ln(original industry wage) 3.167 25.256 Standard Error of Sectoral Shock 19.134 28.542 Nagelkerke‟s R-squared 0.498 Female Regression Constant -7.580 -28.128 Standard Error of ln(original industry wage) 2.778 12.518 Standard Error of Sectoral Shock 17.371 23.036 Nagelkerke‟s R-squared 0.622
Source: KLIPS dataset
Whilst the model specification of the above approach is similar to Jovanovic and Moffitt
(1990), the method is not a direct application in terms of data (4 years for the Korean case
compared with 13 years), difference in the sectoral shock variable (i.e. cross-sectoral
measure in this case), estimation of the probability function for males and females using
data for the entire sample period rather than using a series of sub-periods. With regards to
325
the latter, the current application was based on the male and female datasets of 6,906 and
3,785 observations respectively. Jovanovic and Moffitt (1990), however, estimated the
probability function for a typical worker for each period with the number of observations in
and sectoral shock variable. The robustness of these results across the pooled and
disaggregated analyses therefore gives a high degree of confidence in the analyses of
sectoral mobility.
The Gender Decomposition Result
The analyses conducted in chapter 10 on the separate samples of male and female workers
revealed that the factors affecting sectoral mobility differed between men and women. The
decomposition results showed that male workers in Korea have slightly higher average
probabilities of sectoral mobility than their female counterparts. The explained difference
shows that men have relatively less (more) of those characteristics associated with a higher
354
(lower) likelihood of switching sectors. The variable that contributed most to the explained
difference is the new sector wage growth. In terms of the unexplained difference, it was
found that for a similar set of male (or female) endowments, males had a higher chance of a
sectoral switch than females. That is, in terms of sectoral mobility behaviour, males are
more sensitive to changes in worker and/or industry characteristics than females.
11.4 THE POLICY IMPLICATIONS
Policy implications are presented in this section. The aim is to assess the current policy
measures in Korea in the post-Crisis era and see if further recommendations could be made
from this study of sectoral mobility.
11.4.1 Policy Measures in Post-Crisis Period
In chapter 2 it was reported that the unemployment levels, which had been low before
1997, soared during the Crisis. The Financial Crisis was an indication that Korea‟s
economic and labour market structure required a fundamental revision [Cheon and Jung
(2004)]. A Tripatite Commission, consisting of Government, Union and Employer‟s
Association, was formed to oversee and implement the revision.
The revisions comprised an IMF rescue package which involved restructuring industry and
various unemployment measures. Industrial restructuring applied to the chaebols, financial
sector and government investment corporations (GICs). There was a reduced reliance on
state-funding, a push for reforms to the ownership, supervision and accounting practices of
corporations, privatization of GICs and innovations in the public sector.
The comprehensive unemployment package comprised active measures to maintain and
create jobs [Jeong (2002)] and measures for the unemployed [Cheon and Kim (2004) and
Yoo (2005)]. Under the active measures, job maintenance included providing support for
employment adjustment and creation, introducing childcare centres at work and assisting in
job information and having mutual aid programmes for construction workers. Job creation
involved introducing jobs in SMEs, implementing training programmes (i.e. government-
355
supported internships, human resource development at SMEs, forest-cultivation
programmes) and establishing databases on public sector jobs and social welfare services to
supply information on temporary relief work for the unemployed.
The measures assisting the unemployed consisted of income support to the poor via
unemployment loans/benefits and wage guarantees, vocational training for re-employment
of the unemployed and female-householders, expansion of job security offices, and the
establishment of centres for working women. Each measure catered to different groups,
including those displaced as a result of dismissal, those who have difficulty with finding a
job, those who have become unemployed as a result of business closures or bankruptcies,
those from SMEs not covered by employment insurance, the middle-aged, elderly or non-
regular workers [Yoo (2005)].
11.4.2 Assessment of Policy Measures and Current Situation
The restructuring of the corporate/financial sector was prompt and the unemployment
measures were successful in that they appeared to help reduce unemployment for thousands
[Jeong (2002)]. The unemployment rate dropped from 7% in 1998 to 3% by 2001. These
measures were, however, short-term ones enacted by Korea to help overcome the Crisis.
Although Korea is in the aftermath of the Crisis, there are lingering effects. The prompt
restructuring of industry carried with it a social cost [Yoo (2005)]. Unemployment became
higher than the pre-Crisis levels, owing in part to the retraction of jobs in large companies
which have downsized and outsourced their business activities in response to industrial
restructuring. Job instability has increased as workers are re-employed into lower quality,
non-regular jobs, face recurrent unemployment and precarious earnings (as wages have
become more flexible) [Cheon and Jung (2004)]6. Job prospects are not glamorous for
youth owing to the higher demand for experienced personnel, and they are dim for elderly
women. There is a skill mismatch in the labour market: high-skilled jobs are shrinking as
a result of restructuring and there is a need to fill low-skilled work. The higher-educated
seem to have little desire to work in these lower-skilled jobs.
356
11.4.3 Policy Recommendations
KLI’s Long term Plan Required
The emergency measures were implemented to provide a short-term solution to the Crisis.
However, given the lingering effects of the Crisis noted above, a longer-term plan is
required. One such long-term plan has been recommended by the Korean Labor Institute
(KLI) [Jeong (2002)]. In summary, the basis of this plan is to:
a) Improve the quality of the labour force via investments in
vocational/professional training and labour market information and provision of
wage/promotion incentives.
b) Emphasise the importance of regional labour markets with a call for regional
unemployment data to be made available.
c) Assist younger unemployed workers via smooth transition from school to work,
creating job opportunities and increasing job market information.
d) Create more opportunities for public works programmes, e.g. IT jobs for
younger workers, forest cultivation/public road works for older workers.
e) Enhance work conditions for non-standard workers by catering for social
insurance and leave, improving administration and supervision, ensuring wage
equality for comparable work, identifying jobs which can be converted to
regular jobs, and providing vocational training to convert to regular jobs.
These recommendations are geared towards maintaining low levels of unemployment and
reducing job instability. These policies could be expanded to incorporate the implications
derived from the findings of Parts I and II of this thesis.
Combination of Macro- and Micro-policies
Part I of this thesis showed that the SSH and ADH applied to Korea over 1998-2001, but
given the short period, any policy inference is tentative. From this, a combination of macro-
and micro-level policies is implied. First, the relevance of the ADH to the Korean
economy points towards the adoption of macro-policy. The findings indicated that mobility
predicted from changes in money supply and government debt were significant
357
determinants of unemployment. Aggregate demand policies, via tight controls on the
money supply and a reduction in the public debt to reduce unemployment, would therefore
be relevant. The mechanism is that these lead to smaller predicted inter-sector labour
movements, which our empirical evidence has shown will alleviate unemployment.
Second, the applicability of the SSH suggests that macro-policies are insufficient.
Implementation of these in isolation would lead to a problem where the „government
cannot perfectly identity the characteristics of agents to implement the first-best re-
distributive policy‟ [Andersen (1997)]. Remedial action is required at the micro-level.
Possible Micro-policy Targets
The microeconomic analyses of Part II of this thesis form the basis for appropriate policy
responses. Prior to the identification of policy targets, the nature of unemployment in
relation to the type of mobility must be established. From the SSH, frictional
unemployment, occurring as a consequence of pure sectoral shifts, is not the problem as
mobility is regarded as part of reallocating resources to better sectors following a successful
job match.
The problem arises if there is an inefficient reallocation of labour resources. The SSH and
ADH suggest unemployment accompanies the reallocation of resources in response to
demand and supply shocks. Structural and cyclical unemployment generated from mobility
attributed to demand and supply shocks (SSH and ADH) would be the area of concern.
The situation is worsened if unemployment becomes prolonged and is coupled with job
instability. There may be a role for policy in encouraging better initial job matches and this
can be achieved through identifying characteristics of the labour force and sectors that are
associated with lower mobility so that the labour force can be made more resilient to
shocks. Sectoral mobility arising from these shocks should be minimized so that
unemployment is kept at low levels. This is where empirical findings on the determinants
of mobility become relevant. The SSH and ADH asserted a positive mobility-
unemployment relationship, and from this the main impetus is to reduce sectoral mobility
to lower non-frictional unemployment for Korea.
358
Table 11.1 lists the possible targets derived from the empirical findings on the determinants
of mobility. From the pooled and disaggregated analyses, the target groups are also
indicated. That is, in order to reduce mobility rates, the following targets and measures
could be introduced:
a) Narrow the expected sectoral wage differential gap.
This can best be done by raising the income levels of low-wage sectors via
increases in output and turnover. Ceteris paribus, this would mean that
prevailing rewards in high-wage sectors may no longer be sufficient to entice
worker movements across sectors.
b) Increase permanent incomes in low-wage sectors whilst maintaining7 permanent
incomes in high-wage sectors.
This can be achieved by emphasizing the concept of lifetime employment in all
sectors via skills upgrading for workers, funding businesses and encouraging
higher output and turnover in industries.
c) Enhance job stability for females.
By introducing more non-casual female employment, encouraging skills-
training and education for women, women would be better able to see a career
progression in their existing jobs which could encourage job stability.
d) Encourage young workers/inexperienced men to remain in their original sectors.
For younger persons, the recommendations by Jeong (2002), to create more job
opportunities and improve job information, clearly apply. In addition, official
recruitment policies in the public service for fresh graduates could be enacted so
that younger persons can look forward to a longer-term career path [Addison
(1997)]. For the more inexperienced male workers, a balance between
seniority- and performance-based wage systems should be established so they
can foresee a longer-term progression in their careers, thereby discouraging
them from considering a sectoral switch.
e) Encourage married men/women and household heads to return to the workforce
or to remain in their existing jobs.
This could be achieved by continuing to develop alternative arrangements in
family rearing (like childcare centres) for working married men/women and
household heads. This was first implemented as an emergency measure during
the Crisis period and should be ongoing for longer-term success.
359
f) Raise the standard of formal education.
This recommendation involves continuing with the training programmes and
vocational training for re-employment of the unemployed that were
implemented during the post-Crisis period as a short-term solution. These
should be ongoing to achieve longer-term success.
g) Provide career incentives for skilled men.
Under this initiative, it is envisaged that corporations could give incentives for
existing skilled male workers to remain in their current establishments, either
via monetary or non-pecuniary benefits. For newly recruited skilled male
workers, a progression in their career path could be made known in order to
increase job satisfaction and discourage job quits.
h) Promote entrepreneurship in existing sectors and assist employers in their
businesses to prevent business closures.
Under this proposal, funding could be provided for new business start-ups. In
other words, SMEs should be given more recognition [Garonna and Sica
(2000)]. This was enacted as an immediate measure in the Crisis period and
should be ongoing for longer-term success. In addition, there could be some
funding backup if new businesses are on the verge of failure so that business
owners need not resort to a sectoral switch. For female enterpreneurs, additional
maternity leave and childcare cover could be provided to encourage business
start-ups.
i) Raise GDP growth of all sectors by raising labour productivity8.
The main issue under this recommendation is to cater for multi-sector
production by increasing labour productivity in all sectors. As the phenomenon
of the jobless growth hypothesis appears to be at play in the Korean economy,
the focus should be to increase labour productivity in all sectors by increasing
the skill and technical competency of workers from all sectors in order to ease
mobility rates.
j) Make sectors more resilient so they can better respond to sectoral shocks.
The measures under item (i) can be applied to the sectors to achieve this further
objective.
k) Reduce out-mobility rates of workers in agriculture, and financial, business
services and real estate, and community, social and personal services industries.
360
The measures under items (a) to (i) can be applied with particular force for these
sectors.
In general, these policies are targeted at the overall, male and/or female labour force. The
exceptions pertain to marital status and occupational status, where the policies need to be
specific to males and/or females.
Some policy targets covered in the analysis in this thesis were not recommended even
though, from a simple application of the empirical findings, they can reduce sectoral
mobility. These include increasing (decreasing) the size of the old (new) sector and raising
(easing) the new (old) sector‟s unemployment rate. The reason for this is that the indirect
effect on unemployment via mobility may be more than offset by other more direct impacts
on the level of unemployment. It can also be noted that the potential policy target of
influencing sectoral GDP growth in order to achieve a differential impact on the sectoral
mobility, and hence the unemployment, of males and females is deemed inappropriate, as
the finding in chapter 10 reflects only gender differences in industrial distributions.
Table 11.1 Micro-policy Targets for Korea Target Target Group a) Narrow the expected sectoral wage differential gap. All b) Increase permanent incomes in low-wage sectors whilst
maintaining permanent incomes in high-wage sectors. All
c) Enhance job stability for females. Females d) Encourage young workers and the more inexperienced
males to remain in their original industries. Age Effect: All Tenure Effect: Men
e) Encourage married men/women and/or household heads to return to the workforce/remain in their existing jobs.
Married: Men and Women Household heads: All
f) Raise standard of formal education. All g) Provide career incentives for skilled men. Men h) Promote entrepreneurship in existing sectors and assist
employers in their businesses to prevent business closures.
All Priority group: Female entrepreneurs
i) Raise GDP growth in all sectors by raising labour productivity.
n.r.
j) Make sectors more resilient so they can better respond to sectoral shocks.
All
k) Reduce sectoral mobility rates of agriculture, and financial, business services and real estate, and community, social and personal services industries.
n.r.
All : The policy target is applicable to the overall, male and female labour forces. Men and Women: The policy target is applicable only to the separate analyses undertaken for males and females. n.r: not recommended from the separate analyses undertaken for men and women.
361
It is envisaged that these policies would aid in reducing the social costs of higher
unemployment as well as the job instability mentioned above. Whilst unemployment can be
alleviated via a mix of macro- and micro-policies, job stability could be achieved by
moderating mobility through the micro measures stated.
Integration of Policies with KLI’s Recommendation
Several of the policy measures stated above are inter-linked with the KLI‟s long-term plan.
The KLI‟s recommendation of investment in vocational training is related to the suggested
measures of training to increase lifetime incomes, moderating mobility amongst women
and younger workers, and raising the level of education of the workforce in general [items
(b), (c), (d) and (f)]. The measures to encourage younger workers to remain in the original
sectors are also applicable to the KLI‟s suggestion to tackle youth unemployment [item
(d)]. The measures to lower female mobility, encourage married women to work or remain
in their jobs can be related to the KLI‟s plan to improve job prospects for elderly women
[items (c) and (e)]. Lastly, whilst the suggestion to create IT jobs for younger persons is
related to the measures for moderating mobility rates of young workers [item (d)], the
forest cultivation public works programme for older workers will assist in reducing
mobility rates for workers in agriculture [item (k)].
Of interest to note also is that several suggestions from the KLI to tackle job instability
have the implied result of reducing sectoral mobility, which is the very goal the policies
from this study are geared at. The provision of wage incentives could prevent workers
from switching sectors (since higher wages in the existing jobs reduce mobility) and
assisting disadvantaged workers (young, elderly women, older workers and workers in non-
regular jobs) could discourage them from switching jobs/sectors, thereby lowering mobility
rates and subsequently alleviating unemployment problems.
Therefore, it can be seen that the policy targets derived from this study are in line with the
KLI‟s long-term plan. An integrated effort is thus required to combat the social costs of
unemployment in Korea.
362
11.5 DIRECTION FOR FUTURE RESEARCH
Part I of the thesis provided some preliminary insight into the positive mobility-
unemployment relationship for Korea from the perspectives of the SSH and ADH. Part II
augmented the research via an in-depth analysis of the factors that motivate sectoral
mobility. The thesis has reported an ample range of findings which can provide a basis for
further research.
The empirical support for the SSH/ADH and stage-of-the-business-cycle effect for Korea
was rather tentative owing to the limited amount of data available. The results suggest that
the „new‟ mobility-unemployment phenomenon appears to have just started in the post-
Crisis period for Korea, whereas it had been a feature of the labour markets of Western
countries since the 1980s. Future studies should examine the validity of these hypotheses
for Korea when more data are available. The benefit of this is that if the validity of the
hypotheses for Korea can be established with a higher degree of certainty, then the micro-
policies derived from the findings on the factors that motivate mobility can be implemented
with greater confidence. The year 1998 appears to have been a structural break and the
economy appears to be at a significant turning point. The traditional monetary and fiscal
policies are deemed insufficient. So a policy combination of micro- and macro-policies for
sectoral mobility could be an innovative tool in the new millennium.
There appears to be a dearth of studies for this type of research for Asia, possibly the
consequence of a lack of longitudinal data available for research. Some studies, like Prasad
(1997) who examined the mobility-unemployment relationship for the manufacturing sector
in Japan using informal graphical techniques, are informative but appear to fall short of the
rigor required in research that is to lead to the development of policy. In the case of Korea,
the KLIPS is the first panel study for labour issues, so it is a useful starting point. The
study of sectoral mobility could be extended to the NIEs (Japan, Hong Kong, Singapore
and Taiwan) and the rest of Asia so that the standard of research can be more aligned with
that of the Western countries. This is attainable if datasets like the KLIPS become
available for research in other developing countries. If similar research on sectoral mobility
can be undertaken for other countries in Asia, the lessons learnt from the West, and the
benefits of a new microeconomic policy via sectoral mobility for Asia, could be immense.
363
Endnotes:
1. Since the nature of unemployment was not expounded in the empirical application, it is not mentioned here.
2. The other methods are not mentioned owing to their unsuitability for the current work, namely the ζ-U co-
movement approach and U-V argument for the ADH, and computing contemporaneous correlations between
labour reallocation measures and the average value of foregone production for the RTH. In the case of the
former, there is an absence of a direct assessment of predicted mobility on unemployment. In the latter, the
seemingly non-existent correlations make the method not worthwhile.
3. The chapter suggested that separate analyses might be undertaken for males and females. Since it was not
undertaken owing to the findings obtained for the overall study, it was not mentioned.
4. No distinction in the mobility behaviour of the overall, male or female labour groups is made here.
5. Where no references are provided alongside the findings of the current study, it means that there was no
study that adopted the relevant variable.
6. In Cheon and Jung (2004), wages are more performance-based rather than seniority-based. However, this
change is only partial and the Korean employment system retains traits which separates it from the West, such
as honorary retirement, and seniority-based wages, especially for blue-collar workers in large companies and
union members.
7. According to the empirical result, the target should be to reduce permanent incomes in high-wage sectors.
However, owing to wage inflation and individuals‟ constant demand for higher wages, a lowering of the
lifetime incomes is not desired in the longer term. Hence the next best alternative is to maintain permanent
income levels in the already high-wage sectors.
8. According to the empirical results, to reduce mobility, the target should be to lower the GDP growth of the
old sector, and raise the GDP growth of new sector. Since a lower GDP growth is undesirable, and given the
phenomenon of the jobless growth hypothesis in Korea, the better alternative is to cater for multi-sector
production by raising labour productivity.
364
REFERENCES
Abraham, K. (1983) „Structural/Frictional vs. Deficient Demand Unemployment: Some
New Evidence‟, American Economic Review, vol. 73(4), pp. 708-724.
Abraham, K. (1987) „Help-wanted Advertising, Job Vacancies and Unemployment‟,
Brookings Papers on Economic Activity, vol. 1987(1), pp. 207-243.
Abraham, K. and Katz, L. (1986) „Cyclical Unemployment: Sectoral Shifts or Aggregate
Disturbances?‟, Journal of Political Economy, vol. 94(3), pp. 507-522.
Addison, J. (1997) „The U.S. Labour Market: Structure and Performance‟, article in
„Structural Changes and Labor Market Flexibility: Experience in OECD Countries‟,
edited by Horst Siebert, Mohr, Tübingen, pp. 187-222.
Addison, J. and Portugal, P. (1989) „Job Displacement, Wage Changes and Duration of
Unemployment‟, Journal of Labor Economics, vol. 7(3), pp. 281-301.
Akyüz, Y. (2000) „Causes and Sources of the Asian Financial Crisis‟, UNCTAC Geneva,
Paper presented at the host Country Event, Symposium on Economic and Financial
recovery in Asia, UNCTAD X Bangkok, 17 Feb 2000.
Alogoskoufis, G. and Smith, R. (1991) „On Error Correction Models: Specification,
Interpretations, Estimation‟, Journal of Economic Surveys, vol. 5(1), pp. 97-128.
Altonji, J. and Ham, J. (1990) „Variation in Employment Growth in Canada: The Role of
External, National, Regional and Industrial Factors‟, Journal of Labor Economics,
vol. 8(1), S198-S236.
Andersen, T. (1997) „Structural Changes and Barriers in the Danish Labour Market‟, article
in Structural Changes and Labor Market Flexibility: Experience in OECD Countries,
edited by Horst Siebert, Mohr, Tübingen, pp. 123-149.
Anderson, J. (1979) „Determinants of Bargaining Outcomes in the Federal Government of
Canada‟, Industrial and Labor Relations Review, vol. 32(2), pp. 224-241.
Antolin, P. and Bover, O. (1997) „Regional Migration in Spain: The Effect of Personal
Characteristics and of Unemployment, Wage and House Price Differentials using
pooled Cross-sections‟, Oxford Bulletin of Economics and Statistics, vol. 59(2), pp.
215-235.
Arnott, R., Hosios, A. and Stiglitz, J. (1988) „Implicit Contracts, Labor Mobility and
Unemployment‟, American Economic Review, vol. 78(5), pp. 1046-1066.
Ashenfelter, O. and Pencavel, J. (1969) „American Trade Union Growth: 1900-1960‟,
Quarterly Journal of Economics, vol. 83(3), pp. 434-448.
365
Austria, M. and Martin, W. (1995) „Macroeconomic Instability and Growth in the
Philippines, 1950-1987: A Dynamic Approach‟, Singapore Economic Review, vol.
40(1), pp. 65-81.
Bain, G. and Elsheikh (1976) Union Growth and the Business Cycle, Basil Blackwell,
Oxford.
Barro, R. J. (1977) „Unanticipated Money Growth and Unemployment in the United
States‟, American Economic Review, vol. 67(2), pp. 101-115.
Barro, R. J. (1980) „Inter-temporal Substitution and the Business Cycle‟ NBER working
paper no. 490, National Bureau of Economic Research.
Barrows, G. (2004) „Isolating Trade‟s Effects on Growth‟, paper submitted for the 2004
Moffatt Prize in Economics, Economics at About.Com (http://economics.about.com).
Bartel, A. (1989) „Where do the New U.S. Immigrants Live?‟, Journal of Labor
Economics, vol. 7 (4), pp. 371-391.
Bartel, A. and Koch, M. (1991) Internal Migration of U.S. Migrants’, Immigration, Trade
and the Labor Market, University of Chicago Press, United States.
Bartram, D. (2000) „Japan and Labor Migration: Theoretical and Methodological
Implications of Negative Cases', International Migration Review, vol. 34 (1), pp. 5-
32.
Becker, G. (1964) Human Capital: A Theoretical and Empirical Analysis with Special
Reference to Education, New York and London: Columbia University Press, 1964.
Becker, G. (1993) „Nobel Lecture: The Economic Way of Looking at Behaviour‟, Journal
of Political Economy, vol. 101(3), pp. 385-409.
Beggs, J. and Chapman, B. (1988) „Labor Turnover Bias in Estimating Wages‟, Review of
Economics and Statistics, vol. 70(1), pp. 117-123.
Bernanke, N. (1983) „Non-monetary Effects of the Financial Collapse in the Propagation of
the Great Depression‟, American Economic Review, vol. 73(3), pp. 257-293.
Bernstein, I. (1954) „The Growth of American Unions‟, American Economic Review, vol.
44(3), pp. 301-318.
Blackburn, M, and Neumark, D. (1992) „Unobserved Ability, Efficiency Wages and Inter-
industry Wage Differentials‟, Quarterly Journal of Economics, vol. 107(4), pp. 1421-
1436.
Blan, D. (1991) „Search for Non-wage Job Characteristics: A Test of the Reservation Wage
Hypothesis‟, Journal of Labor Economics, 1991, vol. 9(2), pp. 186-205.
366
Blanchard, O. and Diamond, P. (1989) „The Beveridge Curve‟, Brookings Papers on
Economic Activity, vol. 1989(1), pp. 1-60.
Blanchard, O. and Quah, D. (1989) „The Dynamic Effects of Aggregate Demand and
Supply Disturbances‟, American Economic Review, vol. 79(4), pp. 655-673.
Blank, R. (1985) „An Analysis of Workers‟ Choice Between Employment in the Public and
Private Sectors‟, Industrial and Labor Relations Review, vol. 38(2), pp. 211-224.
Bleany, M. (1992) „Sectoral Rigidities, Wage Inertia and Expectations‟, Australian
Economic Papers, vol. 31(59), pp. 303-310..
Blinder, A. (1973) „Wage-Discrimination: Reduced Form and Structural Variables‟,
Journal of Human Resources, vol. 8(4), pp. 436-455.
Blinder, A. (1976) „On Dogmatism in Human Capital‟, Journal of Human Resources, vol.
11(1), pp. 8-22.
Bradshaw, Y. and Naonan, R. (1997) „Urbanization, Economic Growth and Women‟s
Labour Force Participation. A Theoretical and Empirical Reassessment‟, Cities in the
Development World: Issues, changes and policy, Oxford University Press 1997.
Brainard, S. and Cutler, D. (1993) „Sectoral Shifts and Cyclical Unemployment
Reconsidered‟, Quarterly Journal of Economics, vol. 108(1), pp. 219-243.
Brown, P. and Browne, M. (1962) „Earnings in Industries of the United Kingdom, 1948-
1959‟, The Economic Journal, vol. 72(287), pp. 517-549.
Brown, R., Moon, M. and Zoloth, B. (1980) „Occupational Attainment and Segregation by
Sex‟, Industrial and Labor Relations Review, vol. 33(4), pp. 506-517.
Booth, A. (1983) „A Reconsideration of Trade Union Growth in the United Kingdom‟,
British Journal of Industrial Relations, vol. 21(3), pp. 377-391.
Borjas, G. (1980) „The Relationship Balance Between Wage and Weekly Hours of Work:
The Role of Division Bias‟, Journal of Human Resources, vol. 15(3), pp. 409-423.
Borjas, G. (1982) „The Politics of Employment Discrimination in the Federal Bureaucracy‟,
Journal of Law and Economics, vol. 25(2), pp. 271-299.
Borjas, G. (1990) „The Intergenerational Mobility of Immigrants‟, mimeo, University of
Chicago, 1990.
Borjas, G. (1994) „The Economics of Immigration‟, Journal of Economic Literature, vol.
32(4), pp. 1667-1717.
Borland, J., Hirschberg, J. and Lye, J. (1998) „Earnings of Public Sector and Private Sector
Employees in Australia: Is There a Difference?‟, Economic Record, vol. 74(224), pp.
36-53.
367
Borland, J. and Ouliaris, S. (1994) „The Determinants of Australian Trade Union
Membership‟, Journal of Applied Econometrics, vol. 9(4), pp. 453-468.
Bound, J. and Krueger (1991) „The Extent of Measurement Error in Longitudinal Earnings
Data: Do Two Wrongs Make a Right?‟, Journal of Labor Economics, vol. 9(1), pp. 1-
24.
Bull, C. and Jovanovic, B. (1988) „Mismatch versus Derived Demand Shift as Causes of
Labour Mobility‟, Review of Economic Studies, vol. 55(1), pp. 169-175.
Burdett, K. (1978) „A Theory of Employee Job Search and Quit Rates‟, American
Economic Review, vol. 68(1), pp. 212-220.
Bureau of Labour Market Research, „Structural Change and the Labour Market‟, Research
Report no. 11, Australian Government Publishing Services, Canberra, 1987.
Burgess, J. and Green, R. H. (2000) „Is Growth the Answer? in S. Bell (ed.) The
Unemployment Crisis in Australia: Which Way Out?, Cambridge University Press,
Cambridge, pp. 125-148.
Byrne, J. (1975) „Occupational Mobility‟, Monthly Labor Review, vol. 98, pp. 53-59.
Callagham, G. (1997) Flexibility, Mobility and the Labour Market, Ashgate Publishing Ltd.
Carrington, W. and Zaman, A. (1994) „Interindustry Variation in the Costs of Job
Displacement‟, Journal of Labor Economics, vol. 2(2), pp. 243-275.
Carruth, A. and Disney, R. (1988) „Where Have Two Million Trade Union Members
Gone?‟, Economica, vol. 55(217), pp. 1-19.
Chaison, G. and Dhavale, D. (1992) „The Choice Between Union Membership and Free-
Rider Status‟, Journal of Labor Research, vol. 13(4), pp. 355-369.
Chan, W. (1996) „Intersectoral Mobility and Short-Run Labor Market Adjustments‟,
Journal of Labor Economics, vol. 14(3), pp. 454-471.
Chang, Jiyeun and Yang, Sukyung (2007) „Non-standard Employment from the Social
Exclusion Perspective‟, working paper, Korea Labor Institute, Jun 2007.
Chang, Y., Jaeryang, N. and Changyong, R. (2004) „Trends in unemployment rates in
Korea: A search-matching model interpretation‟, Journal of the Japanese and
International Economies, 18(2), pp. 241-263.
Cheon, Byung-You and Jung, Ee-hwan (2004) „Economic Crisis and the Changes in the
Employment Systems in East Asia: The Cases of Japan and Korea‟, working paper,
Korea Labor Institute, Jun 2004, pp. 1-27.
Cheon, Byung You and Kim, Sungteak (2004) „Labor Market Policies to Meet Challenges
in the Korean Labor Market after the 1997 Financial Crisis‟, paper prepared for
368
International Conference on International Perspectives on Labor Market Institutions,
Seoul, Korea, Jul 19-20 2004, pp. 1-49.
Cho, Sang-Wook (2005) „Household Wealth Accumulation and Portfolio Choices in
Korea‟, Job Market paper, University of Minnesota, Oct 2005, pp. 1-43.
Choo, H. (1989) „The Asian Newly Industrialising Economies (NIEs): Are Economic
Miracles really Miraculous?‟, Singapore Economic Review, vol 34(1), pp. 2-12.
Chew, S. (1990) „Brain Drain in Singapore: Issues and Prospects‟, Singapore Economic
Review, Oct 1990, vol. 35(2), pp. 55-77.
Chillemi, O. and Gui, B. (1997) „Team Human Capital and Worker Mobility‟, Journal of
Labor Economics, vol. 15(4), pp. 567-85.
Chirco, A. and Colombo, C. (1996) „The AD-AS Model and Its Solutions: Some New
Results in a Disequilibrium Perspective‟, Australian Economic Papers, vol. 35(67),
pp. 263-281.
Chiswick, B. (1991) „Speaking, Reading and Earnings among Low-skilled Migrants‟,
Journal of Labor Economics, vol. 9(2), pp. 149-170.
Chiswick, B., Le, A. and Miller, P. (2008) „How Immigrants Fare Across the Earnings
Distribution in Australia and the United States‟, Industrial and Labor Relations
Review, vol. 61(3), pp. 353-373.
Chiswick, B., Lee, Y. L. and Miller, P. (2005) „Longitudinal Analysis of Immigrant
Occupational Mobility: A Test of the Immigrant Assimilation Hypothesis‟,
International Migration Review, vol. 39(2), pp. 332-353.
Chiswick, B. and Miller, P. (1985) „Immigrant Generation and Income in Australia‟,
Economic Record, vol. 61 (173), pp. 540-573.
Chiswick, B. and Miller, P. (1994) „The Determinants of Post-Immigration Investments in
Education‟, Economics of Education Review, vol. 13(2), pp. 163-177.
Choy, K. (1999) „Sources of Macroeconomic Fluctuations in Singapore: Evidence from A
Structural VAR model‟, Singapore Economic Review, Apr 1999, vol. 44(1), pp. 74-
98.
Christie, V. (1992) „Union Wage Effects and the Probability of Union Membership‟,
Economic Record, vol. 68(200), pp. 43-56.
Chuma, J. (1997) „Structural Change and the State of the Labour Market in Japan‟, article
in „Structural Changes and Labor Market Flexibility: Experience in OECD
Countries‟, edited by Horst Siebert, Mohr, Tübingen, pp. 257-289.
369
Clark, T. (1998) „Employment Fluctuations in U.S. Regions and Industries: The Roles of
National, Region-specific, and Industry-specific Shocks‟, Journal of Labor
Economics, vol. 16(1), pp. 202-229.
Clive, B. and Jovanovic, B. (1988) „Mismatch versus Derived-Demand Shift as Causes of
Labour Mobility‟, Review of Economic Studies, vol. 55(1), pp. 169-175.
Coe, D. (1990) „Insider-Outsider Influences on Industry Wages (Evidence from fourteen
Industrialised Countries), Empirical Economics, vol. 15(2), pp. 163-183.
Conlon, R. (1992) „Determinants of Manufacturing Industry Exports: A Comparative Study
of Australia, republic of Korea, Singapore and Taiwan‟, Australian Economic Papers,
vol. 31(59), pp. 427-442.
Cotton, J. (1988) „On the Decomposition of Wage Differentials‟, Review of Economics and
Statistics, vol. 70(2), pp. 236-243.
Coulson, N. and Robins, R. (1987) „A Test of the First-Difference Transformation in Time
Series Models‟, Review of Economics and Statistics, vol. 69(4), pp. 723-726.
Cox, D. (1971) „The Effect of Geographic and Industry Mobility on Income: A Further
Comment‟, Journal of Human Resources, vol. 6(4), pp. 525-527.
Creedy, J. (1974) „Labour Mobility, Earnings and Unemployment. Selected Papers – Inter-
regional Mobility: A Cross-section Analysis‟, Scottish Journal of Political Economy,
vol. 21(1), pp. 41-53.
Creedy, J. and Thomas, R. (1982) The Economics of Labor, Butterworths, London.
Cutler, D., Glaeser, E. and Vigdor, J. (1999) „The Rise and Decline of the American
Ghetto‟, Journal of Political Economy, vol. 107(3), pp. 455-506.
Dalziel, P. (1993) „Classical and Keynesian Unemployment in a Simple Disequilibrium
AS-AD Framework‟, Australian Economic Papers, vol. 32(60), pp. 40-52.
Darby, M., Haltiwanger, J. and Plant, M. (1985) „Unemployment Rate Dynamics and
Persistent Unemployment under Rational Expectations‟, American Economic Review,
vol. 75(4), pp. 614-637.
Davidson, C., Martin, L. and Matusz, S. (1988) „The Structure of Simple General
Equilibrium Models with Frictional Unemployment‟, Journal of Political Economy,
vol. 96(6), pp. 1267-1293.
Davis, S. (1987) „Fluctuations in the Pace of Labor Reallocation‟, Carnegie-Rochester
Conference Series on Public Policy 27, pp. 335-402.
Diamond, P. (1981) „Mobility Costs, Frictional Unemployment and Efficiency‟, Journal of
Political Economy, vol. 89(4), pp. 798-812.
370
Dickens, W. and Katz, L. (1987) „Inter-Industry Wage Differences and Industry
Characteristics‟, article in Unemployment and the Structure of Labor Markets, edited
by Lang, K. and Leonard, J., Basil Blackwell, Oxford, pp. 48-89.
Disney, R. (1990) „Explanations of the Decline in Trade Union Density in Britain: an
Appraisal‟, British Journal of Industrial Relations, vol. 28(2), pp. 165-177.
Dixon, R. (1982) „The Rate of Exploitation and the Wage Share as Weighted Sums of
Sectoral Measures‟, Australian Economic Papers, vol. 21(39), pp. 421-424.
Doeringer, P. and Piore, M. (1971) Internal Labour Markets and Manpower Analysis,
Lexington Books, D.C. Heath, Lexington , Mass.
Doiron, D. and Riddell, W. (1994) „The Impact of Unionisation on Male-Female Earnings
Differences in Canada‟, Journal of Human Resources, vol. 29(2), pp. 504-534.
Doucouliagos, C. (1997) „The Aggregate Demand for Labour in Australia: A Meta-
Analysis‟, Australian Economic Papers, vol. 36(69), pp. 224-242.
Duetsch, L. (1974-1975) „Structural Performance and the Net Rate of Entry into