Financial Market Linkages During Crises Werner Barthel Introduction The General Model Related Models Spillover and Contagion Data Model specification Estimation Results Conclusion Financial Market Linkages During Crises An empirical approach to contagion and spillover effects Werner Barthel December 2, 2008
41
Embed
Crises Werner Barthel Introduction Financial Market ...jin.cao.userweb.mwn.de/contagion.pdf · Financial Market Linkages During Crises Werner Barthel Introduction The General Model
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
Financial MarketLinkages During
Crises
Werner Barthel
Introduction
The GeneralModel
Related Models
Spillover andContagion
Data
Modelspecification
Estimation
Results
Conclusion
Financial Market Linkages During CrisesAn empirical approach to contagion and spillover
effects
Werner Barthel
December 2, 2008
Financial MarketLinkages During
Crises
Werner Barthel
Introduction
The GeneralModel
Related Models
Spillover andContagion
Data
Modelspecification
Estimation
Results
Conclusion
Overview
I This talk is based on "‘Unraveling Financial MarketLinkages During Crises"’ by Mardi Dungey andVance L. Martin in Journal of Applied Econometrics,2007.
I Related work:I Source for the definition of contagion and spillovers:
"‘Contagion: Monsoonal Effects, Spillovers, andJumps between Multiple Equilibria"’ by Paul Massonin IMF Working Paper, 1998.
I More detailed description of the simulation estimator:"‘A Multifactor Model of Exchange Rates withUnanticipated Shocks: Measuring Contagion in theEast Asian Currency Crises"’ by Mardi Dungey andVance L. Martin in Journal of Emerging MarketFinance, 2004.
Financial MarketLinkages During
Crises
Werner Barthel
Introduction
The GeneralModel
Related Models
Spillover andContagion
Data
Modelspecification
Estimation
Results
Conclusion
Overview
I Empirical model of multiple asset classes acrosscountries
I Application to linkages between currency and equitymarkets during the East Asian crisis
I Two asset classes in each country: equities andcurrencies
I Countries of the crises region as well as developedcountries are included.
I Financial market linkages during periods of financialcrises are formally specified.
I Contagion and Spillover effectsI Type of the empirical model:
I Latent factor model with common and idiosyncraticfactors
I Autocorrelation structure of the factorsI GARCH structure of autocorrelation residuals
Financial MarketLinkages During
Crises
Werner Barthel
Introduction
The GeneralModel
Related Models
Spillover andContagion
Data
Modelspecification
Estimation
Results
Conclusion
Factor models in general
I Factor models are a special case of the state spacerepresentation of multivariate time series.
I We assume that a set of K observed variables Ytdepends linearly on N unobserved common factorsFt and on individual or idiosyncratic components ut ,where N < K .
I Yt = LFt + utI where L is a (K × N) matrix of factor loadings.I This can be seen as the measurement equation in
the state space terminology.I In our model the observed returns of equity and
currency markets in the different countries will bedriven by unobserved common factors andidiosyncratic components.
Financial MarketLinkages During
Crises
Werner Barthel
Introduction
The GeneralModel
Related Models
Spillover andContagion
Data
Modelspecification
Estimation
Results
Conclusion
Classification of common factors andidiosyncratic components
I The idiosyncratic component has impacts upon aparticular asset market within a particular country
I Russian bond default (August 1998)I Shocks to the market factor impact upon a specific
asset class within a group of countriesI East Asian Crisis (1997-98), assumed to have
started as currency crisesI Shocks to the country factor impact upon all asset
classes of a countryI The effects of entering IMF negotiations, such as for
Argentina in 2001-2I Shocks to the global factor (Wt) impact upon all
asset classes across all countriesI No specific crises cited in the paper, but US interest
rates may be an example of having such generalimpacts.
Financial MarketLinkages During
Crises
Werner Barthel
Introduction
The GeneralModel
Related Models
Spillover andContagion
Data
Modelspecification
Estimation
Results
Conclusion
The general modelI Build on standard latent factor finance modelsI N + 1 countries (i) each with J asset markets (j)I Latent factor model for demeaned returns Ri,j,t :
I → parsimonious representation of modelling a largenumber of potential linkages between asset marketsthrough the asset market factor Ms,t and itsunanticipated term εs,t
Financial MarketLinkages During
Crises
Werner Barthel
Introduction
The GeneralModel
Related Models
Spillover andContagion
Data
Modelspecification
Estimation
Results
Conclusion
Country source of the shock
I Shock is located in a particular country (Ck ,t )I Remember: Ck ,t = ρC,kCk ,t−1 + ζk ,t
I We augment equation (1) once again by a spilloverand a contagion term:
I As with the asset market source, this specificationcan be extended further by including feedbackbetween countries.
Financial MarketLinkages During
Crises
Werner Barthel
Introduction
The GeneralModel
Related Models
Spillover andContagion
Data
Modelspecification
Estimation
Results
Conclusion
Data description
I The above model is now applied to testing thetransmission between currency and equity marketsduring the East Asian financial crises of 1997-98
I Sample period: 2 July 1997 to 31 August 1998 (304daily observations)
I Indonesia, Korea, Malaysia, and Thailand : includedto identify the linkages amongst financial marketswithin the same geographical region which aredirectly exposed to the crises.
I US and Australia: included to identify transmissionmechanisms to countries outside the East Asianregion
I US: identify common shocks as well as identifyingnumeraire shocks as all exchange rates aredenominated in terms of the US dollar
Financial MarketLinkages During
Crises
Werner Barthel
Introduction
The GeneralModel
Related Models
Spillover andContagion
Data
Modelspecification
Estimation
Results
Conclusion
Contemporaneous structure
Figure: Correlation matrices of equity and currency returns
Financial MarketLinkages During
Crises
Werner Barthel
Introduction
The GeneralModel
Related Models
Spillover andContagion
Data
Modelspecification
Estimation
Results
Conclusion
Autocorrelation structure
I Preliminary identification of the autocorrelationstructure
I VAR(1) to VAR(5) containing all eleven returns seriesare estimated.
I AIC and HIC are minimized for a lag length of one.
Financial MarketLinkages During
Crises
Werner Barthel
Introduction
The GeneralModel
Related Models
Spillover andContagion
Data
Modelspecification
Estimation
Results
Conclusion
Conditional volatility structure
I Preliminary analysis of the volatility structure of theeleven return series
I GARCH(1,1) models for each of the series:
rt = µt + ut
ut =√
htzt
ht = α0 + α1u2t−1 + β1ht−1
ut ∼ N (0,ht)
zt ∼ i .i .d .(0,1)
I SIC is minimized for five out of eleven return seriesby the GARCH(1,1) model.
Financial MarketLinkages During
Crises
Werner Barthel
Introduction
The GeneralModel
Related Models
Spillover andContagion
Data
Modelspecification
Estimation
Results
Conclusion
Model specification (1)I East Asian crisis is characterized as a cross market
crises.I Two asset markets (equity (q) and currency (x)) in
I Distribution of country and idiosyncratic factors
Ci,t ∼ N (0,1) i = 0,1, . . . ,5
ui,t ∼ N (0,1)
Financial MarketLinkages During
Crises
Werner Barthel
Introduction
The GeneralModel
Related Models
Spillover andContagion
Data
Modelspecification
Estimation
Results
Conclusion
Unconditional volatility decomposition
Figure: Based on the specifications of Ri,q,t , Ri,x,t , and theirfactor dynamics
Financial MarketLinkages During
Crises
Werner Barthel
Introduction
The GeneralModel
Related Models
Spillover andContagion
Data
Modelspecification
Estimation
Results
Conclusion
Estimation approach
I The unobserved factors in a latent factor model areusually extracted using a Kalman filter.
I However, the authors claim that the Kalman filter isinconsistent due to the nonlinearities arising from theGARCH structure of the model. Gourieroux andMonfort (1994)
I An Estimation through maximum likelihoodprocedures needs multidimensional integrals that arebeyond the scope of standard numerical methods.
I The authors thus propose a simulation estimator.
Financial MarketLinkages During
Crises
Werner Barthel
Introduction
The GeneralModel
Related Models
Spillover andContagion
Data
Modelspecification
Estimation
Results
Conclusion
Simulation estimator - basic idea
I Simulate the latent factor model for an initial set ofstarting parameters to generate a set of simulatedreturns.
I The simulated returns are then calibrated with theactual returns via a set of moment conditions.
I Moment conditions:I Represent an auxiliary modelI Should capture the empirical characteristics of the
data (contemporaneous correlations amongstreturns, autocorrelations in both the means and thevariances of returns)
I Used to identify the parameters of the underlyingmodel
I See Dungey and Martin (2004)
Financial MarketLinkages During
Crises
Werner Barthel
Introduction
The GeneralModel
Related Models
Spillover andContagion
Data
Modelspecification
Estimation
Results
Conclusion
Indirect parameter estimates (1)
Figure: Parameter estimates with standard errors inparentheses
I Standard errors are in general relatively large.I A joint test confirms a correct specification.