Empirics of Vertical FDI and off-shoring Lessons 3 and 4 Giorgio Barba Navaretti Gargnano, June, 11-14 2006
Dec 19, 2015
Empirics of Vertical FDI and off-shoring
Lessons 3 and 4
Giorgio Barba Navaretti
Gargnano, June, 11-14 2006
Objectives
• OBJECTIVES– Examine if VFDI is indeed a relevant mode of
investment
– Examine the effects of VFDI (host and home)
– Offshoring of services: a different issue?
Standard test of the VFDI model (industry i host country j):
FDIij = β1Tij + β2ScaleEcoi + β3MKT SIZEj + β4RelFactEndj + β5FactIntensi + εij
If β4<0 VFDI rejected
But it is the interaction between factor inensities in i and factor endowments in j that matters
The model should instead be specified asF DIij = β1Tij + β2ScaleEcoi + β3MKT SIZEj + β4RelF actEndj ++β5F actIntensi + β6RelF actEndj F actIntensi + εij, (8)∗
It’s the coefficient β6 that matters. Test if β6>0
Testing for the relevance of the VFDI model: the issue
Example: the proximity-concentration trade-off(US investments ,Brainard, 1997)
hijj
hij
hijij
jhh
hij
hij
hij
Xtarifffreightpwgdp
taxplantscalefirmscaleEXPAS
EXP
)ln()ln()ln()ln(
)ln()ln()ln(ln
7654
3210
D ependent variable hijY : Share o f expo rts in
affiliate supply + expo rts
hfirmscale -0 .2726 (-4 .7)
hplantscale 0 .1345 (2 .7)
jtax -0 .569 (-1 .79)
pwgdp ij 0 .296 (3 .75)
hijfreight -0 .2717
(-4 .6) h
ijtariff -0 .3707 (-7 .4)
A djusted R 2 0 .233 N umber o f o bservatio ns 1035 O L S estimates. t sta tistics in brackets S ource: B ra inard (1997), tab le 1 , column 4 .
Testing HFDI vs HFDI
ijjiji
jijijiij
XSKSKGDPGDP
SKSKGDPGDPGDPGDPAS
))((
)()()(
4
32
210 (6.3)
•Facts in favour of the predominance of HFDI:•Dominant negative effect of trade costs•Weak evidence on the importance of relative factor endowments•Affiliates sales mostly directed to the local market
•Need to estimates the relative importance of the two (Carr et al. (2001), Markusen and Maskus (2002) – cross country data of activities of US MNEs
Horizontal:
Vertical:
RESULTS: support HFDI but note only country variables
Testing jointly country and sector specific factors
Yeaple, 2003 Sales of US MNEs
Testing jointly country and sector specific factors
Effects of Fragmenting production:reminding the predictions
NOTE: INTRASECTORAL RATHER THAN INTERSECTORAL EFFECTS
• Skill mix and skill premium– Home: increases, almost uncontroversially– Host: ambiguous
• Scale and productivity – Home: ambiguous
• Scale – Negative: Relocation of production and labour substitution (VFDI) Substitute export (HFDI)– Positive: Gain market share Product complementarity (export of final and intermediate goods).
• Productivity– Factor mix – Technological sourcing
Effects of fragmenting: evidence on skill mix
• Home: most evidence based on imported intermediates and sectoral level data – no info on source of imported inputs:
– Feenstra and Hanson,1996, 1999 offshoring could account for about 15 percent of the observed increase in the relative wage of non-production workers in the US during the 1979-1990 period.
– Falk and Koebel 2002, Strauss-Kahn 2004, Hijzen, Görg and Hine 2005, Geishecker and Gorg, 2004
• Limited evidence with firm level data– Marin on Austria and Germany: high skilled activities get offshored (questionable)– Barba Navaretti, Bertola, Sembenelli (2006): share of skilled workers rises,
Criscuolo 2006– Whithn MNES skill premium rises (Slaughter, 2000, Hansson, 2001, Head and
Ries, 2002
• Host– Feenstra and Hanson, (1997) impact of FDI on the demand for skills in
maquiladoras in Mexico. FDI account for over fifty percent of the increase in the share of skilled labour in total wages in the late 1980s
Effects of fragmenting: evidence on skill mix 1
Geishecker and Gorg, 2004
• Estimating demand for skills when firms MNEs (IS IT REALLY FRAGMENTATION?):– Slaugther (2000) on US– Hansson (2001) on Sweden– Head and Ries (2002) on Japan
• Estimate short run labour demands derived from translog cost functions
kt5
kt4k
t
kt
3St2Ut10kSt MNEYln
Y
KlnwlnwlnSH
Effects of fragmenting: evidence on skill mix 2
Effects of fragmenting: evidence on skill mix 3Table 9.2: Offshore production and skill upgrading in Japanese manufacturing MNEs
Dependent Variable: log of non-production share of the wage bill )SH( kSt
Unit of observation
Industries
Firms
Method First differences Industry fixed effects
Firm fixed effects
(1) (2) (3) (4)
)Y
Kln(
kt
kt
- 2.49***
(0.38) -1.81*** (0.35)
-7.92*** (0.17)
-4.10*** (0.13)
)Yln( kt
-3.83*** (0.47)
-3.51*** (0.45)
0.86*** (0.08)
-3.18*** (0.16)
ktMNE
-1.14 (1.02)
-1.81 (1.20)
1.11*** (0.23)
3.01*** (0.18)
Residual change
0.07 (0.16)
0.16 (0.19)
6.76*** (0.53)
12.52*** (0.32)
N 1584 1584 19,845 19,845 R2 0.08 0.06 0.154 0.262 Root Mean Square Error
1.008 1.183 11.782 5.58
Source: Head and Ries (2002), Table 3
• Survey (Karsten Bjerring Olsen, 2006)– No clear patterns as to how
offshoring/outsourcing affects productivity, depends os sector and firm level characteristics
• Different ways of looking a the matter (industry or firm level evidence??)
• Key methodological issue:=> Define the right counterfactual
Evidence on other firm level effects of fragmenting
Estimating the effect of fragmentation: Methodological
issues
Time
Average performance
t
NATIONALs
MNEs
Time
Average performance
t
NATIONALs
SWs (Switching firms)
MNEs
Estimating the effect of fragmentation: Methodological issues
Time
Average performance
t
NATIONALs
SWs (Switching firms)
MNEs
Benchmark: hypothetical trajectory if switching firms had not invested
Estimating the effect of fragmentation: Methodological issues
Evidence on other firm level effects of fragmenting
• Barba Navaretti, Castellani and Disdier 2006
• Specific question: do firms improve performance at home by investing abroad?– Define the right counterfactual: what would have
happened if firms had not invested abroad?– Investing in DCs vs. LDCs
Propensity score• The effect of investing on performance is:
where 1 denotes performance after the investment and 0 the hypothetical performance if firms had not invested
• But the last term in unobservable we need to find an
observable counterfactual: untreated firms
• Propensity score matching computes and finds non-investing firms with (almost) identical ex ante
probability of investing
)|1( 1, tiit XSWP
1 01 1
1 01 1
ˆ ( | 1)
( | 1) ( | 1)
t t it
t it t it
E y y SW
E y SW E y SW
Estimators• Standard matching estimator (SM):
• Difference-in-difference estimator (DID):
• DID accounts for further unobserved differences in ex ante performance growth, which were not accounted by matching
• Multiple treatment: firms can switch both in DCs and LDCs:
1 01 1
ˆATT t ty y
1 1 0 01 1 1 1ˆ ( ) ( )DID t t t ty y y y
Counterfact.
Treatment
Non switching
Switching in LDCMultiple treatment
Switching in DC
Data• France: 2002 version of the database “Enquêtes filiales” constructed by the
Direction of Foreign Economic Relations of the French Ministry of the Economy, Finance and Industry– First time investors between 95 and 2000 (80 in LDCs and 91 in DCs)
• Italy: Reprint for information on Italian multinationals (stock and newly established subsidiaries)– First time investors between1993 and 2001 (174 in LDCs and 95 in DCs)
• Amadeus database of Bureau Van Dijck– Balance sheet and employment data– Information on counterfactual
Descriptive stat. on national and switching firms (mean)
National firms
Firms switching Firms switching
to LDC to DC
Italy France Italy France Italy France
N. obs. 17,219 28,645 174 80 95 91
N. of employees 71 89 142 241 304 326
Turnover 15'831 21'411 30'468 80'125 69'754 94'614
TFP 1.6 1.2 2.2 1.9 3 2.0
Value added per employee 50.1 44.4 61.8 58.9 70.9 69.4
Cost of labour per employee 29.8 32.0 29.4 37.7 33.6 41.4
Age 22.1 24.9 24.2 31.8 27.4 25.6
ROI 6.5 6.7 6.1 7.1 7.5 8.0
Current ratio 1.3 1.5 1.3 1.6 1.3 1.7
Probability of switching for French and Italian firms. Multinomial logit
France Italy
Switching to LDC
Log TFPi, t-1 1.577*** (0.421) 2.001*** (0.264)
Log Nb. Employees i, t-1 0.524*** (0.138) 0.078 (0.106)
Log Cost of labour per employeei, t-1 0.949 (0.644) -1.299*** (0.417)
Log Agei, t-1 0.326** (0.140) 0.256** (0.117)
Return on investments i, t-1 0.013 (1.312) -3.841*** (1.033)
Current ratio i, t-1 -0.050 (0.146) -0.319** (0.160)
Switching to DC
Log TFPi, t-1 1.336*** (0.396) 2.170*** (0.401)
Log Nb. Employees i, t-1 0.520*** (0.117) 0.495*** (0.141)
Log Cost of labour per employeei, t-1 1.176** (0.565) -1.703*** (0.635)
Log Agei, t-1 -0.090 (0.118) 0.323** (0.152)
Return on investments i, t-1 -0.443 (1.196) -2.056 (1.543)
Current ratio i, t-1 -0.010 (0.119) -0.186 (0.191)
Number of obs 28816 17488
Pseudo R2 0.2567 0.1923Asterisks denote significance at 1% (***), 5% (**) and 10% (*). Intercept and sector, regional and year dummies not reported
Descriptive stat. on switching firms and matched controls (mean)
CFT to Sw. Sw.
CFT to Sw. Sw.
CFT to Sw. Sw.
CFT to Sw. Sw.
to LDC to LDC to LDC to LDC to DC to DC to DC to DC
Italy France Italy France
N. obs. 161 71 87 82
N. of empl. 89 115 226.9 207.8 298.6 278.1 386.4 274.5
Turnover 19'838 26'702 68'760.7 70'435.6 59’703 63’080 106'859.9 84'030.4
TFP 1.8 2.1 1.7 1.7 2.4 2.6 1.9 2.0
Labour prod 52.2 62.2 54.9 55.8 58.5 63.5 63.7 69.7
Wage 28.5 29.3 37.9 37.6 32.6 33.1 41.2 41.1
Age 22 23.8 33.0 31.6 33.4 26.4 32.5 26.2
ROI 6.7 6 8.1 6.9 7.4 7.7 7.9 8.0
Current ratio 1.3 1.3 1.6 1.6 1.3 1.4 1.7 1.7
The effect of investing abroad on performance at home: France vs Italy France Italy
Effect sw. in LDC Effect sw. in DC Effect sw. in LDC Effect sw. in DC
Coef. Std. Err Coef. Std. Err Coef. Std. Err Coef. Std. Err
TFP growth
ATT 1-year 0.017 (0.037) 0.047 (0.035) 0.030* (0.020) 0.010 (0.033)
ATT 2-years 0.041 (0.038) 0.056 (0.047) 0.058* (0.030) 0.011 (0.035)
ATT 3-years 0.020 (0.057) 0.050 (0.061) 0.041 (0.035) -0.002 (0.043)
DID 1-year 0.126** (0.060) 0.001 (0.057) 0.063** (0.031) -0.075 (0.055)
DID 2-years 0.125 (0.092) 0.031 (0.068) 0.086*** (0.036) -0.074 (0.056)
DID 3-years 0.103 (0.112) 0.012 (0.079) 0.042 (0.048) -0.135* (0.081)
Value Added growth
ATT 1-year 0.000 (0.037) 0.033 (0.030) 0.017 (0.025) 0.006 (0.032)
ATT 2-years 0.022 (0.045) 0.030 (0.041) 0.063* (0.035) 0.055 (0.047)
ATT 3-years -0.009 (0.049) 0.039 (0.061) 0.045 (0.037) 0.042 (0.044)
DID 1-year 0.059 (0.078) -0.007 (0.045) 0.047* (0.031) -0.063 (0.049)
DID 2-years 0.075 (0.100) 0.003 (0.057) 0.112** (0.041) 0.010 (0.049)
DID 3-years 0.014 (0.102) 0.039 (0.073) 0.090* (0.057) -0.028 (0.057)
The effect of investing abroad on performance at home (cont.) France Italy
Effect sw. in LDC Effect sw. in DC Effect sw. in LDC Effect sw. in DC
Coef. Std. Err Coef. Std. Err Coef. Std. Err Coef. Std. Err
Turnover growth
ATT 1-year 0.034 (0.021) 0.029 (0.024) -0.009 (0.022) 0.049** (0.025)
ATT 2-years 0.094*** (0.031) 0.097*** (0.033) 0.006 (0.026) 0.025 (0.037)
ATT 3-years 0.000 (0.052) 0.121** (0.049) 0.052 (0.042) 0.068* (0.042)
DID 1-year 0.008 (0.036) -0.040 (0.039) -0.044 (0.035) 0.021 (0.039)
DID 2-years 0.062 (0.045) 0.034 (0.049) 0.006 (0.041) 0.026 (0.057)
DID 3-years -0.055 (0.079) 0.061 (0.078) 0.065 (0.055) 0.061 (0.057)
Employment growth
ATT 1-year 0.049* (0.029) 0.024 (0.018) -0.022 (0.026) -0.033 (0.033)
ATT 2-years 0.047 (0.031) 0.057** (0.027) 0.018 (0.036) 0.063* (0.035)
ATT 3-years 0.051 (0.039) 0.099*** (0.039) 0.048 (0.034) 0.040 (0.048)
DID 1-year 0.020 (0.029) -0.027 (0.024) -0.040 (0.038) -0.050 (0.050)
DID 2-years 0.029 (0.036) 0.013 (0.036) 0.027 (0.051) 0.068 (0.052)
DID 3-years 0.012 (0.053) 0.053 (0.045) 0.073 (0.058) 0.068 (0.062)
Outsourcing material inputs and services: firm level evidence based on imported inputs
Gorg, Hanley and Strobl, 2004
Off-shoring of high skilled white collars
• Is it a different issue (Markusen 2006)?– No from the North point of view: outsourcing of lower
skilled workers within an industry– South: you need to explain why a scarce factor of
production is cheap there (skilled labour)• Complementary factors• Not cheap relatively to local unskilled labour
– Trade expansion at the extensive margin and trade reversals
• Endless transfer of activities to the South (Trefler 2006):– Remember comparative advantage– Role of R&D institutions
Conclusions
• Fragmentation is an important empirical phenomenon, even when looking at FDI data
• Effects on skill premium in the North likely important
• Effects on employment and productivity ambiguous, but likely positive