A dynamic factor model to assess the real time state of the Spanish industry using confidence indicators Ángel Cuevas Research and Analysis Unit Ministry.

A dynamic factor model to assess the real time A dynamic factor model to assess the real time state of the Spanish industry using confidence state of the Spanish industry using confidence

indicators indicators

Ángel CuevasResearch and Analysis Unit

Ministry of Industry, Energy and Tourism of Spain (MINETUR- prevMITYC)

[email protected]

SUBSECRETARÍA DE INDUSTRIA, ENERGÍA Y TURISMO

SECRETARÍA GENERAL TÉCNICA

Subdirección General de Estudios, Análisis y Planes de Actuación

EU workshop on business and consumer surveys (BCS)

Brussels, 15th-16th November 2012

BACKGROUNDInterest in improving the measurement of the

state of the industrial business cycle for purposes of:Anticipation of adverse situations.Evaluation and implementation of economic and

industrial policies.Improvement in databases of economic indicators.Advances in econometric methods for time series.Development of computer tools.Stock and Watson (1991, 2002), Gayer and Genet

(2006), Angelini et al. (2008) Camacho and Perez-Quirós (2009), Cuevas and Quilis (2011).

Objective: Multivariate modeling of a broad and representative set of monthly indicators of industrial activity, with the purpose of prediction, analysis and monitoring and forecasting of macroeconomic aggregates (industrial GVA).

2

Inputs

Preprocessing

Dynamic factor model

{Treatment unbalanced

panel}

3

Applications

Selection of indicators

• High frequency indicators (monthly).

• Must provide a synthetic measure of the Spanish industrial activity.

• They must be available promptly.• They must be correlated with the

reference series: Industrial Production Index (IPI)

4

5


• Prove the correlation with the IPI: Cross-correlation with the growth

signal of SAC series. Cyclical Analysis: Butterworth

(band-pass) + classification of the turning points (Bry-Boschan).

6


Variable SourceStart date

Release date

Unit

Industrial Confidence Indicator Ministry of Industry / EC 1990 01 t-2 days balance of repliesPMI Industry Markit Economics 1998 05 t+1 day balance of replies

Car Registrations General Directorate of Traffic 1990 01 t+1 day unitsElectricity Consumption Spanish Electricity Network 1990 01 t+1 day million Kw/h

Manufacture of cars Ministry of Industry 1994 01 t+25 days unitsManufacture of commercial and industrial

vehicles Ministry of Industry 1994 01 t+25 days units

Consumption of dieselPetroleum Products

Corporation1990 01 t+30 days thousand of metric tons

I ndustrial Production Index National Statistical Institute 1990 01 t+35 days volume index

Large Companies Sales. Industry Tax State Agency 1995 01 t+35 days deflated value indexTurnover Index in Industry National Statistical Institute 2002 01 t+50 days deflated value index

New Orders Index in Industry National Statistical Institute 2002 01 t+50 days deflated value index

Leading indicators are highlighted in yellow

7

Leading indicators

Cross-correlation: y-o-y rates/differences

8

Leading indicators

Cross-correlation: m-o-m rates/differences

9

Leading indicators

Cross-correlation: q-o-q rates/differences(quarterly frequency)

10

Leading indicators

Cyclical Analysis: ICI

11

Leading indicators

Cyclical Analysis: Car registrations

12

Leading indicators

Cyclical Analysis: PMI

Inputs

Preprocessing



panel}

13

Applications

Preprocessing

• The series are adjusted for seasonal and calendar effects (if such effects are significant).• Logarithmically transformed.• Regular differences.

• The above variables are standardized.

14

sac1t,i

sac1t,i

sact,isac

t,it,i x

xx)xlog()B1(z

sact,it,i x)B1(z (Soft)

Inputs

Preprocessing



panel}

15

Applications

16

ft

z1,t z2,t z3,t

u1,t u2,t u3,t

Static factor model

1(B) 2(B) 3(B)

e1,t e2,t e3,t

(B)

at

Common dynamic

Idiosyncratic dynamics

17

Dynamic factor model: complete representation

et

t

t

t

t

t

U

f

ttt

Q0

01,

0

0Niid~

e

a

e

a

U

f

)B(0

0)B(

ULfZ

Factor model: dynamics

18

k..1idiagQ

k..1iB1diag)B(

BBBB1)B(

i,ie

1,iU

44

33

221f

19

Factor model: estimation (Nº. factors)

0 2 4 6 8 10 120

0.5

1

1.5

2

2.5

3

3.5

4

SCREE PLOT

Number

Eig

enva

lue

20

Dynamic factor model: estimation (f)

• The common factor and its standard deviation are estimated by Kalman filter, adjusting the dynamic factor model to state space representation.

)](Q,0R),,(F),L(H[

Q0

0R,

0

0Niid~

W

V

WFSS

VHSZ

UF

t

t

t1tt

ttt

21

Dynamic factor model: loadings

Loadings Lead/ LagTurnover Index in Industry 0,90 0Industrial Production Index 0,83 0

New Orders Index in Industry 0,83 0Large Companies Sales. Industry 0,73 0

Consumption of diesel 0,71 0Manufacture of commercial and industrial

vehicles 0,43 0

PMI Industry 0,39 3Electricity Consumption 0,36 0

Manufacture of cars 0,36 0Industrial Confidence Indicator 0,31 3

Car Registrations 0,24 3

Inputs

Preprocessing



panel}

22

Applications

23

Estimation with an unbalanced panel

1 2 3 4 5 6 7 8123

......

......

......

......

T1

T2

Indicator

Obs

erva

tion

Longitudinal panel: initial estimate of the common factor

Cross-section panel

Dark grey: observedLight grey: non observed

24


t 1 2 3 4 5 6 7 8 Static Dynamic123456789

1011121314

FactorI ndicator

25


BALANCED PANEL: Using OLS (Stock-Watson)

t 1 2 3 4 5 6 7 8 Static Dynamic123456789

1011121314

FactorI ndicator

26


BALANCED PANEL: Using Kalman Filter

t 1 2 3 4 5 6 7 8 Static Dynamic123456789

1011121314

I ndicator Factor

27


REFINED BALANCED PANEL (KF). Repeat until convergence is achieved.

t 1 2 3 4 5 6 7 8 Static Dynamic123456789

1011121314

FactorI ndicator

28


FINAL BALANCED PANEL AND FORECASTS: Using Kalman Filter.

t 1 2 3 4 5 6 7 8 Static Dynamic123456789

1011121314151617

FactorI ndicator

Inputs

Preprocessing



panel}

29

Applications

30

Industrial GVA and dynamic factor

31

Industrial GVA and dynamic factor

Cross-correlation: y-o-y rates

Cross-correlation: q-o-q rates

32

Forecasting and interpolation of industrial GVA

Benchmarking method: Chow-Lin, Fernández

Forecasting performance,

2003:Q1 – 2012:Q1 RMSE

ARIMA 2,755

DFM + Bench. 1,461

ISI (MEC) + Bench. 1,707

33

Markov swithching model

34

ConclusionsIt has developed a coincident indicator of Spanish

industrial activity, trying to exploit all possible information from various related monthly indicators.

The presence of leading indicators is critical in order to project the factor and anticipate the evolution of industrial activity in real time.

These leading indicators are ICI, PMI and car registrations and they have a lead of three months.

The methodology allows not only to estimate this factor, but also get individual predictions in a multivariate context of all the indicators included in the model.

With the estimated factor there are various options for use:

Perform a real time prediction of IGVATranslate its variations into probabilities of recession

This work can be extended in many directions: transfer functions, more sophisticated Markov switching models, etc.

35

Angelini, E., Camba-Méndez, G., Giannone, D., Reichlin, L., Runstler, G (2008) “Short-term forecasts of Euro area GDP growth”. CEPR Discussion Paper n. 6746.

Camacho M, Pérez-Quirós G (2010) “Introducing the Euro-STING: Short Term Indicator of Euro Area Growth. Journal of Applied Econometrics.

Cuevas, A. & Quilis, E.M. (2011) “A factor analysis for the Spanish economy”. SERIEs Journal of the Spanish Economic Association.

Gayer, C. & Genet, J.(2006) “Using Factor Models to Construct Composite Indicators from BCS Data - A Comparison with European Commission Confidence Indicators”. Economic Papers N.240, European Commission.

Kim, C.-J. & Nelson, C.R. (1999) “State-Space Models with Regime Switching”, The MIT Press.

3535

References

Thanks for your attentionThanks for your attention

Ángel CuevasResearch and Analysis Unit

Ministry of Industry, Energy and Tourism of Spain (MINETUR)

[email protected]

SUBSECRETARÍA DE INDUSTRIA, ENERGÍA Y TURISMO

SECRETARÍA GENERAL TÉCNICA

Subdirección General de Estudios, Análisis y Planes de Actuación

EU workshop on business and consumer surveys (BCS)

Brussels, 15th-16th November 2012

A dynamic factor model to assess the real time state of the Spanish industry using confidence indicators Ángel Cuevas Research and Analysis Unit Ministry.

Documents

unbalanced panel slide

ici slide

pmi slide

yellow slide

soft slide

y ratesdifferences slide

markov swithching model

complete representation