Introduction Methodology and Data Estimation results Analysis of cross-country di¤erences Conclusions

The Role of Liquidity, Risk and Economic Activityin the Global Transmission of the Financial Crisis

Alexander Chudik� Marcel Fratzscher‡

(�) European Central Bank and CIMF(‡) European Central Bank

DG ECFIN, ULB & UBC Conference "Advances ininternational macroeconomics - Lessons from the crisis"

Brussels, 23-24 July 2010

Introduction Methodology and Data Estimation results Analysis of cross-country di¤erences Conclusions

Motivation

Global reach of �nancial crisis with high degree ofheterogeneity across countries and regionsWhat are transmission mechanisms?Liquidity shocks

"liquidity squeeze" in credit markets and esp. inter-bankmarketscollapse or near-collapse of �nancial institutionsmassive central bank interventions, incl. cross-border throughswap lines

Pricing of risk & risk appetitehigh leverage of �nancial institutions, though not in EMEsdeleveraging and "�ight-to-safety" phenomena, esp. out ofEMEs; into US treasuries

Real economy shocksde-coupling vs re-couplingcollapse in trade, esp. for more open EMEs

Introduction Methodology and Data Estimation results Analysis of cross-country di¤erences Conclusions

Motivation

First step: Understanding the transmission mechanism

How does the transmission of shocks di¤er across economies?Has the transmission mechanism changed in the crisis?

Second step: Explaining the heterogeneity in the globaltransmission

Potential transmission channels

real and �nancial exposureidiosyncratic: country macro fundamentalsstrength of domestic institutions

Introduction Methodology and Data Estimation results Analysis of cross-country di¤erences Conclusions

Methodology

Challenge of identi�cation of liqudity shocks, risk shocks andreal activity shocks

Financial market perspective

response of money markets as proxy for impact on �nancialconditionsresponse of equity markets as proxy for impact on real economy

Global VAR (in�nite-dimensional VARs) of Chudik andPesaran (2009, 2010) allows addressing:

large dimension of VAR - for 26 economies - via concept ofneighborhood e¤ectsall variables treated as endogeneousrestrictions allow for rich spatial and temporal interactionsamong variablesidenti�cation of various shocks, with additional sign restrictionsdistinguish US shocks from shocks to other economies

Introduction Methodology and Data Estimation results Analysis of cross-country di¤erences Conclusions

Main �ndings �transmission

1 Liquidity shocks key driver during crisis

primarily for advanced economiesfor both money markets and equity markets

2 Risk shocks and real activity shocks also more important incrisis

mainly for EMEs

3 E¤ect on advanced economies more via �nancing conditionsvs on EMEs more via real economy channel

4 Further cross-economy heterogeneity

Europe experienced highest increase in exposure to US shocksin crisisLatin America and CEEC mainly via risk shocks, Emerg. Asiavia liquidity

Introduction Methodology and Data Estimation results Analysis of cross-country di¤erences Conclusions

Main �ndings �channels

Bayesian Averaging of Classical Estimates (BACE) approach(Sala-i-Martin et al. 2004)

cross-sectional averaging with OLSto deal with large number of potential determinants

1 During tranquil periods:

real and �nancial exposure to US more relevant forunderstanding heterogeneity in transmission of US shocks

2 During crisis:

domestic fundamentals, risk and quality of institutions morerelevante.g. FX reserves, sovereign country ratingreal and �nancial exposure less relevant

Introduction Methodology and Data Estimation results Analysis of cross-country di¤erences Conclusions

Literature �Crisis, global transmission

Current �nancial crisis

Focus on US policy responses (Calomiris 2008, Taylor 2009)Role of liquidity (Adrian et al. 2009, Heider, Hoerova andHolthausen 2009)Financial constraints rather than demand in US (Tong & Wei2008)Little on global transmission (e.g. IMF 2009 on �nancial stresstransmission)

Crisis and role of contagion for transmission

Contagion and related channels (Bae et al. 2003, Karolyi 2003,De Gregorio and Valdes, 2001, Dungey et al. 2004, 2005)Transmission channels (Forbes and Rigobon 2002, Forbes andChinn 2004, Bekaert, et al. 2005)Time-varying global market integration (Bekaert & Harvey1995, 2000)

Introduction Methodology and Data Estimation results Analysis of cross-country di¤erences Conclusions

Literature �Methodology

The methodological approach of the paper links to a broadliterature focusing on Global VAR (GVAR) models.

GVAR was proposed by Pesaran, Schuermann and Weiner(2004). Since then, it has been developed further and used invarious applications (Pesaran et al., 2006, Dees et al., 2007,Pesaran, Smith and Smith, 2007, among others)

Methodological foundations for the speci�cation of countrymodels were developed recently by Chudik and Pesaran (2010)and later extended by Pesaran and Chudik (2010) to allow fordominant units.

Introduction Methodology and Data Estimation results Analysis of cross-country di¤erences Conclusions

Methodology

We follow strategy of Global VAR literature (2 steps):1 Estimation of country-speci�c models of small dimension2 Solving estimated country models in one large Global VAR

We follow Chudik and Pesaran (2010) and Pesaran andChudik (2010) to design individual country models.

Our starting point is the following high-dimensional VARmodel augmented with common factors,

xt = α+Φxt�1 + Γft + ut , and ut = Rεt , (1)

where Φ is a large k � k matrix of coe¢ cients,ut = (u01t , ...,u

0Nt )

0 is an k � 1 vector of reduced form errors,ft is m� 1 vector of (strong) unobserved common factors,and Γ is the corresponding k �m matrix of factor loadings.

Introduction Methodology and Data Estimation results Analysis of cross-country di¤erences Conclusions

All coe¢ cients in system (1) cannot be estimated due to curseof dimensionality (large number of endogenous variables).

Pesaran and Chudik propose economically intuitive solution tothe curse of dimensionality based on concept of neighborhoode¤ects.

We are very generous on the possibilities of spatio-temporallinkagesWe allow for US dominance in �nancial markets, other sourcesof strong cross section dependence besides the US in�uence,and local neighborhood e¤ects.

Our methodology treats all variables as endogenous. We donot rely on some of the restrictive assumptions in factormodel literature (e.g. the assumption that unboundedeigenvalues cannot rise at a rate slower than N).

Introduction Methodology and Data Estimation results Analysis of cross-country di¤erences Conclusions

Identi�cation of global shocks

Global shocks enter the vector residuals in the US marginalmodel (featuring domestic variables and foreign cross sectionaverages), but additional restrictions are needed if one wantsto distinguish between US and foreign global shocks withnon-US origin

To accomplish this, we suppose that the US shocks come �rst

Within the set of US shocks, we aim to distinguish between aUS macro surprise shock, a stock market shock, an interestrate shock, a risk aversion shock and a liquidity shock. Wecombine sign restriction and partial ordering approaches toachieve identi�cation

Partial ordering of US shocks: (Group 1) a US macro surpriseshock, (Group 2) risk aversion shock and a liquidity shock, and(Group 3) a stock market shock and an interest rate shock

Introduction Methodology and Data Estimation results Analysis of cross-country di¤erences Conclusions

Summary of sign restrictions.

i1t r1t vixt tedt newst ı̄t r̄t

VIX shock � � + . . . .

TED shock + � . + . . .

US interest rate shock + � . . . + �US stock market shock + + . . . + +

Introduction Methodology and Data Estimation results Analysis of cross-country di¤erences Conclusions

Data

26 economies �open advanced economies and EMEsPre-crisis/tranquil period (1 Jan 2005 - 6 Aug 2007) vs crisisperiod (7 Aug 2007 - July 2009)Weekly frequency � trade-o¤ speed of transmission vs.non-overlapping trading times:

Liquidity - TED spreadRisk - VIXEquity returns (MSCI, LC)Money markets (3M interbank)Real activity - US macro news

unweighted aggregate across GDP, IP, retail sales,NAPM/ISM, non-farm payroll employment, unemployment,consumer con�dence, workweek(a) normalizing by their standard deviation over the sampleperiod, (b) then by aggregating by weekNote: macroeconomic news are exogenous by de�nition(Andersen et al. 2003, Ehrmann & Fratzscher 2007)

Introduction Methodology and Data Estimation results Analysis of cross-country di¤erences Conclusions

Stock market indices

Introduction Methodology and Data Estimation results Analysis of cross-country di¤erences Conclusions

VIX and TED spread

Introduction Methodology and Data Estimation results Analysis of cross-country di¤erences Conclusions

US macro surprise shocks

Introduction Methodology and Data Estimation results Analysis of cross-country di¤erences Conclusions

Estimation �transmission

Variance decomposition

gauge how much of total variation in equity markets andmoney markets can be accounted for by various shocks

Generalised impulse respons functions (GIRF)

sensitivity of equity markets and money markets to a speci�cshock of a given magnitude

Remark:

decrease in sensitivity of a particular market to a speci�c shockis not necessarily inconsistent with higher share of varianceaccounted for by that shockchanges in volatility of underlying shock

Introduction Methodology and Data Estimation results Analysis of cross-country di¤erences Conclusions

Main �ndings �transmission

1 Liquidity shocks key driver during crisis

primarily for advanced economiesfor both money markets and equity markets

2 Risk shocks and real activity shocks also more important incrisis

mainly for EMEs

3 E¤ect on advanced economies more via �nancing conditionsvs on EMEs more via real economy channel

4 Further cross-economy heterogeneity

Europe experienced highest increase in exposure to US shocksin crisisLatin America and CEEC mainly via risk shocks, Emerg. Asiavia liquidity

Introduction Methodology and Data Estimation results Analysis of cross-country di¤erences Conclusions

Variance decompositionUS macro TED VIX US US Restnews shock shock stock m. money m.

Stock MarketsPre-crisis period

US 0.00 24.34 37.13 14.73 7.63 16.18Advanced 0.35 9.63 16.10 3.19 1.80 68.93Emerging 0.44 8.23 9.24 3.47 3.13 75.50

Crisis periodUS 7.26 33.57 14.48 21.32 15.59 7.77Advanced 3.29 25.81 9.20 9.50 6.98 45.21Emerging 3.99 19.81 9.76 9.44 4.66 52.33

Introduction Methodology and Data Estimation results Analysis of cross-country di¤erences Conclusions

Variance decomposition (Ctd.)US macro TED VIX US US Restnews shock shock stock m. money m.

Money MarketsPre-crisis period

US 1.94 0.58 2.90 3.70 21.04 69.84Advanced 0.42 0.67 1.96 0.22 0.15 96.59Emerging 0.64 0.59 0.67 1.26 5.17 91.68

Crisis periodUS 0.02 29.48 14.61 21.37 19.78 14.74Advanced 0.74 9.73 6.20 4.20 0.89 78.24Emerging 1.13 4.59 3.21 3.21 2.78 85.07

Introduction Methodology and Data Estimation results Analysis of cross-country di¤erences Conclusions

Impulse response functions

Impulse response function of a shock to US TED spread, impact on stockmarkets. Dashed lines correspond to crisis period.

Introduction Methodology and Data Estimation results Analysis of cross-country di¤erences Conclusions

Impulse response function of a shock to VIX, impact on stock markets.Dashed lines correspond to crisis period.

Introduction Methodology and Data Estimation results Analysis of cross-country di¤erences Conclusions

Impulse response function of US macro news shock, impact on stockmarkets. Dashed lines correspond to crisis period.

Introduction Methodology and Data Estimation results Analysis of cross-country di¤erences Conclusions

Impulse response function of US stock market shock, impact on stockmarkets. Dashed lines correspond to crisis period.

Introduction Methodology and Data Estimation results Analysis of cross-country di¤erences Conclusions

Impulse response function of US money market shock, impact on stockmarkets. Dashed lines correspond to crisis period.

Introduction Methodology and Data Estimation results Analysis of cross-country di¤erences Conclusions

Contemporaneous impact of a shock to US TED spread on stock marketsand 25-75% bootstrap error bands. Dark/brown bars correspond to crisis

period; light/green bars to pre-crisis period.

Introduction Methodology and Data Estimation results Analysis of cross-country di¤erences Conclusions

Contemporaneous impact of a shock to VIX on stock markets and25-75% bootstrap error bands. Dark/brown bars correspond to crisis

period; light/green bars to pre-crisis period.

Introduction Methodology and Data Estimation results Analysis of cross-country di¤erences Conclusions

Contemporaneous impact of US macro news shock on stock markets and25-75% bootstrap error bands. Dark/brown bars correspond to crisis

period; light/green bars to pre-crisis period.

Introduction Methodology and Data Estimation results Analysis of cross-country di¤erences Conclusions

Analysis of cross-country di¤erences in the transmission ofshocks - Methodology

To shed light on the cross-section heterogeneity in thetransmission of US shocks to the rest of the world, weestimate the following cross-section regression

y (s)i = c (s) +K

∑`=1

β(s)` xi` + ζ

(s)i , for i = 2, ...,N,

where y (s)i is the contemporaneous impact of a US shock s(to US macro news, VIX, TED, US money market or US stockmarket) on the stock market or the money market of countryi , and xi` for i = 2, ...,N and ` = 1, 2, ...,K is the set of Kfundamentals speci�c to country i .

Introduction Methodology and Data Estimation results Analysis of cross-country di¤erences Conclusions

We have relatively limited number of countries, yet thepotential set of country fundamentals is large (We havecompiled K = 14 fundamentals)

Therefore we follow Bayesian Averaging of Classical Estimates(BACE) approach of Sala-i-Martin et. al (2004), which wasoriginally developed to analyze determinants of growth.

This approach combines the averaging of estimates acrossmodels estimated by classical ordinary least squares (OLS)and is particularly useful for understanding which of the largeset of determinants (if any) might play a role empirically.

Introduction Methodology and Data Estimation results Analysis of cross-country di¤erences Conclusions

Country Fundamentals

MacroeconomicOpenness, �nancial integration, rating notches, reserves as a share of GDPunemployment, growth, current account as a share of GDPQuality of institutionsICRG institutional measures: political category index, �nancial category index,economic category indexBilateral exposure to UStrade exposure, �nancial debt exposure, �nancial equity exposure

Introduction Methodology and Data Estimation results Analysis of cross-country di¤erences Conclusions

Cross section regression results: Posterior probabilities of variablerelevance and posterior means.

Crisis period, impact on stock pricesUS shock: vix ted macro news

openness 19% (0.07) 22% (0.13) 21% (-0.08)�nancial int. 19% (-0.01) 20% (-0.01) 20% (-0.01)trade exposure 57% (-2.32) 44% (-1.98) 46% (1.45)equity exposure 29% (-1.44) 51% (-2.38) 25% (0.41)�nancial debt exposure 40% (2.59) 29% (1.60) 81% (-2.61)rating notches 59% (0.11) 23% (0.03) 27% (-0.03)icrg- political 59% (-0.05) 24% (0.01) 35% (0.02)icrg- �nancial 24% (0.02) 28% (-0.04) 20% (0.01)icrg - economic 39% (0.06) 26% (0.04) 72% (-0.06)market cap 22% (0.00) 31% (-0.00) 18% (0.00)reserves 19% (-0.00) 20% (0.00) 21% (0.00)unemployment 18% (-0.01) 25% (0.03) 19% (-0.01)growth 21% (0.04) 19% (0.02) 26% (0.04)current account 26% (0.01) 34% (-0.03) 29% (-0.01)

Introduction Methodology and Data Estimation results Analysis of cross-country di¤erences Conclusions

I. Trade openness (Dark/brown lines correspond to countries abovemedian; light/green bars to the group below median. Dotted linescorrespond to the crisis period. Impact of VIX and US macro news

shocks on stock markets.)

Introduction Methodology and Data Estimation results Analysis of cross-country di¤erences Conclusions

II. Rating notches (Dark/brown lines correspond to countries abovemedian; light/green bars to the group below median. Dotted linescorrespond to the crisis period. Impact of VIX and US macro news

shocks on stock markets.)

Introduction Methodology and Data Estimation results Analysis of cross-country di¤erences Conclusions

III. Political institutions (Dark/brown lines correspond to countries abovemedian; light/green bars to the group below median. Dotted linescorrespond to the crisis period. Impact of VIX and US macro news

shocks on stock markets.)

Introduction Methodology and Data Estimation results Analysis of cross-country di¤erences Conclusions

IV. Reserves (Dark/brown lines correspond to countries above median;light/green bars to the group below median. Dotted lines correspond tothe crisis period. Impact of VIX and US macro news shocks on stock

markets.)

Introduction Methodology and Data Estimation results Analysis of cross-country di¤erences Conclusions

Conclusions

Focus on global transmission of �nancial crisis, acrossadvanced economies and EMEs

Global VAR approach; 26 economies and 2 �nancial marketsegments

Objective to better understand role of three distinct types ofshocks as culprits:

a tightening in liquidity conditions and credit markets �>mattered more for advanced, esp. in Europea severe re-pricing of risk and �ight of investors into safe assetclasses �> EMEs, esp. CEECa strong and synchronous collapse of economic activity �>EMEs, esp. Latin America & CEEC

Factors accounting for cross-country di¤erences

role of country-speci�c fundamentals versus external exposure

Introduction Methodology and Data Estimation results Analysis of cross-country di¤erences Conclusions

Implications

Complexity of global transmission of the crisis; cannot bereduced to a single dimension

Countries were not innocent bystanders, but severity oftransmission not only related to real and �nancial exposure...

But to a substantial extent also to domestic macroeconomicfundamentals and institutions

Role for economic policy, though controversy about speci�cs(e.g. self-insurance/reserves vs. institutions)