Financial market stability—A test

Int. Fin. Markets, Inst. and Money 19 (2009) 506–519

Contents lists available at ScienceDirect

Journal of International Financial Markets, Institutions & Money j ou rn al h om ep ag e: w w w . e l s e v i e r . c o m / l o c a t e / i n t f i n

Financial market stability—A test Dirk G. Baur a, Niels Schulze b a b

Dublin City University, The Business School, Collins Avenue, Glasnevin, Dublin, Ireland European Commission – JRC, TP 361, Via Fermi 2749, 21027 Ispra (VA), Italy

a r t i c l e

i n f o

Article history: Received 23 January 2008 Accepted 26 June 2008 Available online 16 July 2008 JEL classification: C22 C51 G15 Keywords: Financial stability Systematic risk Contagion Quantile regression

a b s t r a c t This paper proposes a definition for financial market stability and an econometric test. It analyzes the impact of systematic and systemic shocks on developed and emerging market stock indices in normal and extreme market conditions. Financial market stability is defined as a constant impact of systematic shocks in normal and extreme market situations. Empirical results show that the impact of systematic shocks is significantly larger in extreme market conditions than in normal conditions for emerging markets. In contrast, the relationship is stable for developed markets. Hence, only developed markets meet an essential condition for financial market stability. © 2008 Elsevier B.V. All rights reserved.

1. Introduction The crises in East Asia, Russia and Brazil in 1997 and 1998 have raised concerns about the adequate functioning and the stability of the financial system. This has triggered a large literature on contagion which attempts to measure and explain the spread of country-specific shocks across markets. In contrast, the literature on the stability of financial markets is relatively scarce compared to the contagion literature even though the topics are closely linked to each other. Moreover, there is no common definition of ‘financial stability’ which resembles the situation at the time of the Asian crisis when there was no standard definition for ‘contagion’. This paper is motivated by the observation that there is no generally accepted definition of financial stability and no respective econometric test. Moreover, there is no study explicitly analyzing the role of systematic shocks in crises periods or in extreme market conditions. This gap in the literature is surprising since shocks of the systematic risk component can also affect individual financial

E-mail addresses: [email protected] (D.G. Baur), [email protected] (N. Schulze). 1042-4431/$ – see front matter © 2008 Elsevier B.V. All rights reserved. doi:10.1016/j.intfin.2008.06.003

D.G. Baur, N. Schulze / Int. Fin. Markets, Inst. and Money 19 (2009) 506–519

507

markets and the financial system as a whole. Examples for an increased importance of systematic shocks are manifold. The impact of the sub-prime crisis in the US on global stock markets is a recent example showing that idiosyncratic shocks are more systematic in nature today than they were in the past. Studies defining or analyzing financial stability are rare. An overview of definitions for financial stability can be found in Schinasi (2004). Other definitions are given by the IMF (IMF, 2003) and the ECB (Padoa-schioppa, 2003). Official (central bank) statements that there remains no consensus on how financial stability should be defined can be found in Federal Reserve Board (2006) and Bank of England (2004). Academic papers support this statement (e.g. see Cihak, 2007; Goodhart, 2006; Oosterloo and de Haan, 2004). Most of the contributions define financial stability in a broad sense, that is, these definitions include the entire financial system and its linkages to the real sector. In contrast, this paper aims to analyze financial stability in a stock market context thereby advocating a more narrow definition of financial stability, that is, financial market stability. The study contributes to the literature by proposing a definition of financial market stability and an econometric testing framework. The framework tests the impact of systematic and systemic shocks on financial markets and can assess whether a country exhibits financial market stability or not. The literature on contagion or financial crises in general usually focuses on the spread or transmission of country-specific or idiosyncratic shocks. Statements of policy makers and regulators suggest that this focus is not appropriate or sufficient in a financial stability context. A broader focus is consistent with definitions put forward by the IMF (2003) and the ECB through Padoa-schioppa (2003) that financial stability is the constant propagation of shocks to the financial system across markets. Since ‘shocks to the financial system’ comprise more than one or two markets, an analysis of a type of shock that is affecting more than one country simultaneously seems more adequate. A natural candidate is a systematic risk component given by an index comprising certain regions of stock market indices or firms. An analysis of country-specific shocks would be too narrow in a financial stability context. Moreover, such country-specific shocks or idiosyncratic shocks have been analyzed extensively in the contagion literature while no particular attention has been paid to the role of systematic shocks. The empirical results for a selection of emerging and developed markets show that the impact of systematic shocks varies considerably and is significantly larger in highly volatile regimes for some emerging markets. In contrast, most developed markets exhibit a constant dependence on systematic risk. Hence, only those developed markets meet an essential condition for financial market stability. There are a number of explanations for this finding. First, developed stock markets are larger in terms of market capitalization and liquidity. Second, developed markets are better regulated. Third, investor’s portfolios are better diversified and thus more effective in absorbing shocks. Fourth, emerging markets might be more exposed to global investors’ sentiment and therefore are potentially the first to be affected when markets start to fall. It is thus possible that emerging markets suffer more from phenomena like flight-to-quality and contagion. The paper is structured as follows: Section 1 presents several definitions of financial stability and proposes a model to test the degree of financial market stability. Section 2 presents the empirical results obtained by a quantile regression analysis. Section 3 provides several different model specifications to check the robustness of our approach. Section 4 contains a simulated case study that aims to explain the potential sources of an increasing dependence in extreme market conditions. The conclusions summarize our findings. 2. Financial market stability There is no clear consensus of what is ‘financial stability’ and people seem to find it more convenient to define financial instability (see Padoa-schioppa, 2003). Interestingly, 3 years after the speech of ECB representative Padoa-Schioppa, officials of the Federal Reserve Board in the US (Federal Reserve Board, 2006) confirm this statement with ‘The term ‘financial instability’ is often poorly defined’. Moreover, many definitions are relatively broad either in terms of the conditions of financial stability or with regard to the effects of financial stability. For example, a Financial Stability Report of the ECB

508

D.G. Baur, N. Schulze / Int. Fin. Markets, Inst. and Money 19 (2009) 506–519

(ECB, 2005) describes financial stability as facilitating economic processes and exhibiting the ability to prevent adverse shocks from having inordinately disruptive impacts.1 Despite the many definitions of financial stability and financial instability and the fact that there remains no consensus on how it should be defined, we focus on two definitions put forward by the ECB through Padoa-schioppa (2003) and the IMF (2003). ‘Financial stability is a condition whereby the financial system is able to withstand shocks without giving way to cumulative processes, which impair the allocation of savings and investments opportunities and the processing of payments in the economy.’ (Padoa-schioppa, 2003, page 2). The second definition on financial system instability can be found in the IMF report on financial stability (IMF (2003), page 63): ‘The degree to which shocks to the financial system are amplified and propagated across markets or across institutions is a key element of financial system instability.’ These definitions clarify that financial stability can be viewed as a condition in which the exposure to common shocks is not increasing. Thus shocks are not amplified. The definitions leave some space for interpretation. In particular, there are two specific aspects. The definitions neither clarify the type of shocks nor the question in which situations (e.g. time periods) these shocks determine financial stability or instability. We focus on systematic shocks since we interpret ‘shocks to the financial system’ as shocks that affect the entire financial system and are thus systematic or systemic. Systemic shocks differ from systematic shocks in terms of the severity and frequency of shocks. Systematic shocks are frequent and not extreme while systemic shocks are infrequent and extreme (see Das and Uppal, 2004). Systemic shocks are thus an element of systematic shocks. The focus on extreme market situations is consistent with the contagion literature where the correlation of markets in normal and extreme situations is analyzed. Moreover, it is also in line with the ECB statement that ‘an emphasis should be on extreme market situations’ (ECB, 2005, page 150). Therefore, we examine the propagation and potential amplification of systematic and systemic shocks in normal and in extreme market conditions. The focus on systematic shocks can also be justified by the extensive contagion literature that analyzes idiosyncratic shocks and mainly disregards the role of system-wide shocks in crisis periods. In addition, the role of such risks compared to idiosyncratic risks is likely to increase due to higher cross-market linkages and interdependencies of financial markets. Hence, we propose an econometric model that analyzes the effect of systematic shocks on individual financial markets in normal and extreme market conditions. 2.1. Econometric framework In a first step, one has to obtain a proxy for ‘shocks to the financial system’ as defined by the IMF (IMF, 2003). We use a systematic risk component such as a regional or global stock market index as a proxy for ‘shocks to the financial system’ since systematic shocks cover several countries and not only one. This systematic risk component is denoted as ft . In order to construct innovations in systematic risk we first regress ft on a constant and an adequate number of lags to generate residuals that are serially uncorrelated. The innovations in systematic risk denoted as ft∗ are then equal to the residuals of the respective regression. In order to circumvent a priori definitions of normal and extreme market situations, we employ a quantile regression model that provides estimates of the effect of systematic shocks on financial markets in any condition represented by the conditional quantiles of the return of the market under investigation. The model can be written as follows: rit = ai + bi ft∗ + vit ,

Qr (|ft∗ ) = ai () + bi ()ft∗

(1)

ft∗

where rit is the market return at t for country i, denotes the systematic shock as described above and vit represents the idiosyncratic shock of market i at time t. Qr ( | ft∗ ) denotes the -th conditional quantile of rit , assumed to be linearly dependent on ft∗ . The model is estimated with the quantile

1 The most recent financial stability report (ECB, 2007) contains an extended definition but is not fundamentally different to former definitions (ECB, 2005).

D.G. Baur, N. Schulze / Int. Fin. Markets, Inst. and Money 19 (2009) 506–519

509

regression method and can thus assess the impact of ft∗ on different conditional quantiles of rit , that is, different market conditions (see Koenker and Bassett, 1978 or Buchinsky, 1998 for an introduction to quantile regression).2 If bi is stable, that is, constant over all quantiles, a fundamental condition of financial market stability is fulfilled. On the contrary, if bi increases for lower quantiles, the market under investigation is more exposed to shocks originating from the financial system in negative market conditions than in its normal state. The latter implies that there is an amplification of systematic shocks on the financial market and therefore no financial market stability. It is noteworthy that an ordinary least squares (OLS) regression only provides an estimate for the conditional mean but not for different quantiles. Thus, OLS cannot be used to analyze whether systematic shocks exhibit a constant or a varying impact on individual markets. Finally, we can write a definition that is directly associated with the econometric model above as follows: Financial market stability is the constant (stable) propagation of systematic shocks on a financial market in normal and extreme market conditions. 2.2. Link to contagion Our definition of financial stability is based on systematic shocks and can thus be clearly distinguished from the contagion literature that focuses on idiosyncratic shocks. There is, however, a direct link to the contagion literature which also analyzes differences in the propagation mechanism in normal and extreme market conditions (e.g. see Forbes and Rigobon, 2002).3 An advantage of our approach is that the quantile regression model can detect and test for heteroscedasticity in the data (see Koenker and Basset, 1982). A simulation study confirms that an increased variance of the systematic shock does not lead to an increased estimate of the propagation (or amplification) of such shocks (see Section 4). Moreover, the fact that the quantile regression model provides estimates for a set of different conditional quantiles avoids an ad hoc definition of certain market conditions or crises periods in contrast to the common approach in the contagion literature in which a crisis period is defined a priori (see Baur and Schulze, 2005). Finally, the quantile regression approach provides information about a potentially asymmetric impact of systematic shocks in good (upper quantiles) or bad (lower quantiles) market conditions in the country under investigation. Again, the ‘traditional’ contagion literature does not distinguish between such asymmetries (e.g. see Forbes and Rigobon, 2002; an exception is Bae et al., 2003). 3. Empirical analysis 3.1. The data We use daily (close-to-close) continuously compounded index returns (log differences) of twenty international stock markets calculated in US dollars. The data is provided by Morgan Stanley Capital International Inc. (MSCI). Furthermore, seven regional stock indices are analyzed: Emerging Markets Asia, Emerging Markets Latin America, Emerging Markets Global, Europe, North America and World. The indices span a time period of approximately 10 years from April 1997 until July 2007. The number of observations is T = 2668. A look at the standard deviations of the series shows that there is considerable variation in the risk of the markets and indices. The highest variation can be seen in Indonesia and Russia and the lowest

2 Other applications of quantile regression to finance can be found, e.g. in Bassett and Chen (2001), Barnes and Hughes (2002), Engle and Manganelli (2004), Baur and Schulze (2005) and Taylor (2008). 3 Quantile regression could also be utilized to test for contagion. If bi () is interpreted as a measure of co-movement, a comparison of co-movement in normal market conditions (e.g. bi (50)) and extreme negative market conditions (e.g. bi (1)) could serve as a test for contagion.

510

D.G. Baur, N. Schulze / Int. Fin. Markets, Inst. and Money 19 (2009) 506–519

Fig. 1. Quantile regression results for Asian markets. The figure shows the varying impact of a systematic shock (MSCI Emerging Markets Asia) on 10 Asian markets. Most countries exhibit a significantly higher exposure to systematic shocks in lower and upper quantiles compared to intermediate quantiles. Exceptions are Korea, Taiwan and India.

standard deviation is given in the UK and the US. The skewness is minimal for Argentina (around −1.4) and maximal for Malaysia (around +1.2) and the kurtosis is most pronounced for Argentina, Malaysia and Indonesia. 3.2. Tests for financial market stability This section starts by presenting the estimation results for the Asian markets and the varying propagation of systematic shocks onto these markets. Secondly, the results for the four Latin American countries are presented followed by the findings for Russia and South Africa. The systematic shocks that are used for the analysis depend on the markets. We use the Emerging Markets Index Asia for the Asian markets, Emerging Markets Latin America for Argentina, Brazil, Chile and Mexico and the global Emerging Markets Index for Russia and South Africa. Finally, we present the estimation results for developed markets where the systematic shock is derived from innovations in the MSCI World index. Fig. 1 shows the coefficient estimates for the systematic shock (MSCI Emerging Markets Asia) on the return of the market under investigation. The graph illustrates that most Asian countries exhibit a larger coefficient in the extreme conditional lower and higher quantiles than in the middle quantiles leading to a u-shaped pattern for most countries. Exceptions are India, Korea and Taiwan that exhibit a relatively stable relationship. The differential impact of the systematic shock on these countries is not surprising since Korea and Taiwan could be categorized as developed countries and have relatively mature financial markets. Interestingly, Hong Kong and Singapore exhibit a pattern more similar to emerging markets than to developed markets. Table 1 reports the associated coefficient estimates with t-statistics, p-values and R2 for three exemplarily chosen countries, Hong Kong, Korea and Malaysia. It can be seen that in particular Malaysia exhibits a pronounced u-shaped pattern of the coefficients across the quantiles, while Korea has a relatively stable coefficient measuring the propagation of systematic shocks. Similarly to the Asian markets, we performed the according regressions for the other emerging markets. The estimates for Argentina range from 0.86 (1% quantile) over 0.77 (median) to 0.96 (99%

Table 1 Quantile regression results for Asian markets Coefficient

t-Statistics

p-Value

R2

Korea

Coefficient

t-Statistics

p-Value

R2

Malaysia

Coefficient

t-Statistics

p-Value

R2

bi Constant

0.8430 −0.0368

5.03 −14.57

0.00 0.00

0.1902

bi Constant

1.3178 −0.0582

6.02 −10.05

0.00 0.00

0.1776

bi Constant

1.0778 −0.0428

8.77 −15.77

0.00 0.00

0.2811

q2

bi Constant

0.8421 −0.0292

7.31 −28.18

0.00 0.00

0.1860

bi Constant

1.4605 −0.0396

6.33 −11.74

0.00 0.00

0.2146

bi Constant

0.9902 −0.0353

18.74 −27.74

0.00 0.00

0.2575

q5

bi Constant

0.7594 −0.0200

15.19 −34.74

0.00 0.00

0.1901

bi Constant

1.5072 −0.02

16.67 −22.44

0.00 0.00

0.2928

bi Constant

0.8226 −0.0243

11.27 −19.26

0.00 0.00

0.1899

q10

bi Constant

0.7573 −0.0134

16.50 −22.58

0.00 0.0

0.1937

bi Constant

1.4947 −0.0154

20.89 −18.12

0.00 0.00

0.3331

bi Constant

0.6398 −0.0146

11.55 −21.64

0.00 0.00

0.1368

q50

bi Constant

0.6112 0.0003

20.90 1.51

0.00 0.13

0.1619

bi Constant

1.3581 0.0006

59.50 2.03

0.00 0.04

0.3622

bi Constant

0.4014 0.0000

17.34 −0.04

0.00 0.97

0.0839

q90

bi Constant

0.7041 0.0139

17.13 35.31

0.00 0.00

0.1782

bi Constant

1.3845 0.0166

28.46 21.88

0.00 0.00

0.3250

bi Constant

0.6289 0.0144

13.83 20.74

0.00 0.00

0.1118

q95

bi Constant

0.7125 0.0199

13.66 35.20

0.00 0.00

0.1773

bi Constant

1.4306 0.0248

14.62 18.62

0.00 0.00

0.3064

bi Constant

0.8340 0.0241

9.53 18.80

0.00 0.00

0.1483

q98

bi Constant

0.6831 0.0285

5.81 17.00

0.00 0.00

0.1750

bi Constant

1.4425 0.0413

9.55 10.93

0.00 0.00

0.2435

bi Constant

1.1118 0.0381

10.27 18.15

0.00 0.00

0.2075

q99

bi Constant

0.7919 0.0365

4.21 16.90

0.00 0.00

0.1678

bi Constant

1.5602 0.0560

12.52 12.03

0.00 0.00

4.210

bi Constant

1.3535 0.0514

8.47 11.01

0.00 0.00

0.2520

This table reports estimation results for three Asian markets. This table illustrates the impact of systematic shocks (innovations of MSCI EM Asia) for different market conditions of Hong Kong, Korea and Malaysia. In particular the Malaysian market exhibits a pronounced u-shaped impact of systematic shocks across the quantiles. The coefficient estimates range from 1.078 in the 1% quantile to 0.401 in the 50% quantile and 1.354 in the 99% quantile.

D.G. Baur, N. Schulze / Int. Fin. Markets, Inst. and Money 19 (2009) 506–519

Hong Kong q1

511

512

D.G. Baur, N. Schulze / Int. Fin. Markets, Inst. and Money 19 (2009) 506–519

Fig. 2. Quantile regression results for developed markets. The figure shows the varying impact of a systematic shock (MSCI World) on three European countries (France, Germany and the UK), the US and Japan. The countries exhibit a relatively stable exposure to systematic shocks in lower and upper quantiles compared to intermediate quantiles.

quantile). The analogous coefficients for Brazil are 1.2, 1.29 and 1.40, for Chile 0.42, 0.42 and 0.46 and for Mexico 0.95, 0.91 and 1.01. These estimate show that from the South American markets only Argentina exhibits some economically significant variation in the parameter estimates. The propagation of systematic shocks is relatively stable for Brazil, Chile and Mexico. In contrast, the estimates for Russia and South Africa show a pronounced u-shaped pattern across the quantiles. The estimates range from 1.98 (1% quantile) over 1.19 (median) to 2.24 (99% quantile) for Russia and from 1.20 (1%) over 0.84 (median) to 0.94 (99%) for South Africa. Finally, Fig. 2 presents the estimates for developed markets’ exposure to shocks derived from the MSCI World index. The graphs show that there is some variation in the coefficient estimates across the quantiles, more pronounced for France and Germany and with a downward trend from the lowest quantiles towards the highest quantiles. In contrast, the coefficients for the UK and the US are relatively stable. The coefficient estimates reported in the tables and figures show the propagation of systematic shocks to countries’ stock markets in different market conditions. The estimates indicate whether the propagation is stable or varying and also show whether the impact of systematic shocks increases in market turmoil. The latter implies a focus on extreme quantiles, especially low quantiles representing negative returns. Extreme positive shocks might also occur in crisis periods or periods of market turmoil but such events usually are not initiated by an extreme positive shock. For an evaluation of financial market stability it is thus appropriate to set the focus on extreme quantiles on the left side of the conditional distribution. We propose a test for financial market stability that evaluates whether the propagation of systematic shocks is constant or whether the propagation varies across different quantiles. Therefore, we test whether the estimated impact on the lower conditional quantiles (1, 2 and 5%) is statistically different from the coefficient for the conditional median. The joint test is an F-test with three degrees of freedom. The outcome is tabulated in Table 2 for all developed and emerging markets of our sample. Table 2 shows that all developed markets exhibit financial market stability. For emerging markets in Asia, the test results indicate financial market stability only for China, Korea and Taiwan. The other Asian

D.G. Baur, N. Schulze / Int. Fin. Markets, Inst. and Money 19 (2009) 506–519

513

Table 2 Test for financial market stability Joint test q1 = q2 = q5 = q50

q1 = q50

F(3, 2663)

F(1, 2663)

Prob > F

q2 = q50 Prob > F

F(1, 2663)

q5 = q50 Prob > F

F(1, 2663)

Prob > F

Financial market stability?

Developed markets Systematic shock: World Germany 1.35 France 1.05 Japan 0.32 UK 0.20 US 0.88

0.26 0.37 0.81 0.90 0.45

1.95 0.16 0.09 0.08 1.26

0.16 0.69 0.77 0.78 0.26

0.52 0.14 0.56 0.14 2.20

0.47 0.71 0.45 0.71 0.14

2.55 2.44 0.69 0.58 2.26

0.11 0.12 0.41 0.45 0.13

YES YES YES YES YES

Emerging markets Systematic shock: EM Asia China 0.42 Hong Kong 4.73 India 4.12 Indonesia 7.74 Korea 1.17 Malaysia 38.35 Philippines 10.41 Singapore 3.47 Taiwan 0.27 Thailand 3.32

0.74 0.00 0.01 0.00 0.32 0.00 0.00 0.02 0.84 0.02

1.03 2.33 0.01 6.46 0.0 72.52 23.54 6.83 0.31 7.94

0.31 0.13 0.91 0.01 0.96 0.00 0.00 0.01 0.58 0.00

0.52 6.31 1.71 18.60 0.14 80.0 24.45 4.20 0.73 8.31

0.47 0.01 0.19 0.00 0.71 0.00 0.00 0.04 0.39 0.00

0.68 14.11 3.79 12.41 2.48 36.48 23.16 7.02 0.08 6.80

0.41 0.00 0.05 0.00 0.12 0.00 0.00 0.01 0.77 0.01

YES NO NO NO YES NO NO NO YES NO

Systematic shock: emerging markets index Russia 4.83 0.00 5.03 South Africa 6.11 0.00 17.41

0.03 0.00

3.53 6.43

0.06 0.01

13.05 9.56

0.00 0.00

NO NO

Systematic shock: EM Latin America Argentina 3.01 0.03 Brazil 0.94 0.42 Chile 1.39 0.24 Mexico 0.27 0.85

0.65 0.24 0.98 0.60

2.04 0.04 1.84 0.58

0.15 0.84 0.17 0.45

6.83 0.17 1.85 0.74

0.01 0.68 0.17 0.39

NO YES YES YES

0.21 1.39 0.00 0.28

This table presents the test statistics for the equality of coefficients. It is tested whether the propagation of systematic shocks is equal in different market conditions (quantiles) with a focus on negative returns (lower quantiles). The null hypothesis is that the coefficient estimates of the 1, 5, 10 and 50% quantiles are equal. A rejection of the null hypothesis (at the 1% level of significance) implies that there is no financial market stability.

markets do not meet the criteria for financial market stability. The largest F-statistics are obtained for Indonesia, Malaysia and the Philippines leading to a clear rejection of the null hypothesis of ‘financial market stability’. The null hypothesis is also rejected for Russia and South Africa as well as for Argentina in the sample of Latin American countries. In contrast, Brazil, Chile and Mexico exhibit financial market stability. 3.2.1. Systematic and idiosyncratic shocks The importance of systematic shocks in explaining the stock market index returns in different market conditions is not directly related to the test for financial market stability. However, it is important for an evaluation of the sources of financial market stability or instability. If systematic shocks become more important in extreme market conditions, the financial system plays a bigger role per se compared to country-specific shocks. In the context of contagion, this means that in periods of market turmoil or high volatility, country-specific shocks become less important than systematic shocks. If, on the other hand, idiosyncratic shocks become more important, it would imply that country-specific shocks are in fact the driver of extreme events. Moreover, it would justify the focus on idiosyncratic shocks as prevalent in the literature on contagion. Fig. 3 presents the average goodness-of-fit measure (pseudo R2 ) for all types or regions of markets across 99 quantiles. The figure shows that the R2 -value increases for all emerging markets in the lower and upper quantiles compared to intermediate quantiles (around the median). However, the

514

D.G. Baur, N. Schulze / Int. Fin. Markets, Inst. and Money 19 (2009) 506–519

Fig. 3. Importance of systematic shocks and idiosyncratic shocks. The figure shows the average (pooled) R2 (goodness of fit) across 99 quantiles for four types of countries, (i) Asian markets, (ii) Latin American markets, (iii) developed markets, and (iv) Russia and South Africa. The plotted R2 shows a u-shaped pattern for all types except for developed markets.

increase is more pronounced in the lower quantiles than in the upper quantiles. Developed markets do not exhibit such a pronounced u-shaped pattern. In the extreme lower quantiles the numbers even decrease. The findings are evidence for an increasing importance of systematic shocks for emerging markets and a relatively constant importance for developed markets. For the latter group, idiosyncratic shocks tend to play a bigger role in periods of market turmoil. This outcome also implies that developed markets are more efficient in extreme market conditions than emerging markets since the ratio of country-specific information to systematic information is higher in developed markets where informational environments allow market participants to acquire information and act quickly and inexpensively upon it.4 Our results also imply that diversification is more effective for portfolios comprising developed markets than emerging markets since the ratio of diversifiable risk stays relatively constant in extreme market conditions. 3.2.2. Asymmetric shock propagation It is not only interesting whether the propagation of systematic shocks increases in crisis periods or extreme market conditions but also whether there is an asymmetric impact of systematic shocks in negative and positive market conditions.5 In order to analyze this question we test the hypothesis that the propagation is equal in extreme lower and higher quantiles (1 and 99%, 2 and 98% and 5 and 95%). Table 3 presents the test results and shows that the propagation is symmetric for most stock markets except Germany, France and the US for a test of equality of the 5 and 95% quantiles and for South Africa for all tested pairs of quantiles. Interestingly, the Asian and Latin American countries do not exhibit any statistically significant asymmetry.

4 Morck et al. (2000) and Bries et al. (2007) demonstrate that the ratio of idiosyncratic risk in (more efficient) developed markets is higher than in emerging markets. 5 The role of asymmetric correlations for equity portfolios is analyzed by Ang and Chen (2002). Yuan (2005) also examines asymmetries with a focus on financial crises.

D.G. Baur, N. Schulze / Int. Fin. Markets, Inst. and Money 19 (2009) 506–519

515

Table 3 Test for asymmetry q1 = q99

q2 = q98

q5 = q95

F(1, 2663)

Prob > F

F(1, 2663)

Prob > F

Germany France Japan UK US

1.58 1.63 0.99 0.24 0.09

0.21 0.20 0.32 0.62 0.76

1.31 1.52 0.04 0.48 1.77

0.25 0.22 0.84 0.49 0.18

China Hong Kong India Indonesia Korea Malaysia Philippines Singapore Taiwan Thailand

1.11 0.09 0.03 0.00 0.10 0.19 0.08 0.72 0.68 0.05

0.29 0.77 0.86 0.97 0.76 0.66 0.77 0.40 0.41 0.82

0.10 0.58 0.67 0.14 0.00 0.20 0.75 0.11 0.75 0.00

Russia South Africa

0.19 3.50

0.66 0.06

Argentina Brazil Chile Mexico

2.05 0.89 1.11 1.36

0.15 0.35 0.29 0.24

F(1, 2663)

Asymmetry? Prob > F

q1 = q99

q2 = q98

q5 = q95

3.07 5.45 0.36 0.10 4.78

0.08 0.02 0.55 0.75 0.03

NO NO NO NO NO

NO NO NO NO NO

YES YES NO NO YES

0.75 0.44 0.41 0.71 0.96 0.65 0.39 0.74 0.39 0.97

0.19 0.92 0.56 1.49 0.51 1.08 2.52 2.06 0.02 0.04

0.66 0.34 0.46 0.22 0.48 0.30 0.11 0.15 0.88 0.85

NO NO NO NO NO NO NO NO NO NO

NO NO NO NO NO NO NO NO NO NO

NO NO NO NO NO NO NO NO NO NO

0.31 3.73

0.58 0.05

0.19 13.57

0.66 0.00

NO YES

NO YES

NO YES

0.03 0.08 0.10 1.67

0.86 0.78 0.75 0.20

0.35 0.00 0.19 0.47

0.56 0.97 0.66 0.49

NO NO NO NO

NO NO NO NO

NO NO NO NO

This table presents the test statistics and probabilities for equality of extreme lower and upper quantiles. For each country, the equality of three pairs of quantiles is examined, that is, the 1 and 99% quantile, the 2 and 98% quantile and the 5 and 95% quantile. The results show that the null hypothesis of ‘no asymmetry’ cannot be rejected in most cases. Exceptions are Germany, France and the US for the 5 and 95% quantiles and South Africa for all extreme quantiles.

4. Robustness checks In this section we discuss a number of specification issues and report quantitative and qualitative differences of estimates obtained with alternative model specifications compared to the results reported in the previous section. First, we include an interaction term calculated as the product of a crisis dummy for the Asian financial crisis and the systematic shock in order to examine whether the u-shaped pattern is mainly due to the Asian crisis. The crisis dummy is equal to one from 15 October to 15 November 1997 and zero otherwise (see Forbes and Rigobon, 2002 for a similar crisis period definition). The inclusion of the interaction term leads to a positive coefficient estimate for the additional regressor for all Asian markets except India, China and Korea. These three countries exhibit a negative loading with the systematic shock in the lower and upper quantiles. The coefficient estimates measuring the exposure to systematic shocks do not change significantly and the u-shaped pattern for the estimates remains. Interestingly, the coefficient estimates are relatively heterogeneous for developed markets concerning their sign and homogeneous regarding their insignificance. The coefficient estimates are positive for the US and negative for Germany, France and the UK. Japan exhibits a negative loading in the lower quantiles and mainly positive values in the other quantiles. A positive coefficient implies that the impact of systematic shocks increased during the Asian crisis while a negative coefficient implies a decreased impact. Thus, only the US was affected by a greater role of systematic shocks in the crisis period compared to previous periods. Second, an augmentation of the model with lagged systematic or lagged idiosyncratic shocks yields no economically or statistically different results than the basic model. The u-shaped and the stable pattern of the propagation of systematic shocks for emerging and developed markets, respectively, are sustained. Furthermore, the estimation results with lagged dependent or independent variables show some characteristics that are worth mentioning. Lagged systematic and individual market returns are significant for some quantiles for both emerging and developed markets. However, the coefficient

516

D.G. Baur, N. Schulze / Int. Fin. Markets, Inst. and Money 19 (2009) 506–519

estimates are more pronounced and economically significant for emerging market. The estimates for the lagged dependent variables show that negative and positive country-specific returns exhibit some significantly higher degree of autocorrelation in extreme market conditions than in normal conditions. This suggests that positive or negative market sentiment is more persistent than normal market conditions. Moreover, the results for a specification with lagged systematic shocks illustrate that there is a delayed or incomplete contemporaneous propagation of systematic shocks on individual financial markets.6 We do not report these results in more detail since the test results for financial market stability do not change with the inclusion of lagged dependent or independent variables. Third, we include a GARCH term with the aim to control for heteroscedasticity, different volatility regimes or extreme systematic shocks in different market conditions. For the Asian markets, the inclusion of such an estimated GARCH term as an additional regressor leads to larger coefficient estimates in the lower quantiles converging to a zero difference for quantiles above the median. Thus, the u-shape becomes more pronounced for the left tail of the distribution. The coefficient estimates for the GARCH term exhibit an s-shape with negative values for lower quantiles and positive values for the upper quantiles. Not surprisingly, the estimates imply that extreme negative and positive market conditions are associated with higher volatility regimes than normal or average market conditions. Finally, the average R2 -value increases by approximately 0.1 for each quantile. The results for the Latin American markets show that the propagation of systematic shocks exhibits a u-shaped pattern when the GARCH term is included compared to a relatively constant propagation across quantiles in the specification without the volatility estimate. R2 increase by approximately 0.05. Developed markets show more distinct coefficient estimates in general but the changes are less homogeneous than for emerging markets. The coefficients in extreme lower quantiles are slightly higher for the US, the UK and France, lower for Germany and considerably higher for Japan and the other Asian markets. The coefficient for the GARCH term also exhibits a pronounced s-shaped pattern as found for emerging markets. Furthermore, the average R2 increases significantly in the tails and the difference is around 0.1 compared to the model without the GARCH term. The difference is monotonically decreasing towards zero for middle quantiles (30–70%). Fourth, the number of bootstrap repetitions is varied in order to examine the impact on the test results of financial market stability. The test results reported above are based on 20 bootstraps. As a robustness check we increased the number of bootstraps from 20 to 200, 500 and 1000. The qualitative results do not change with a higher number of bootstrap repetitions. Interestingly, there is no clear trend regarding the evolution of the standard errors with higher repetitions. In some cases (markets and quantiles), standard errors increase and in other cases, the standard errors decrease with increasing repetitions. Fifth, we perform the analysis with the stock indices denominated in other currencies than the US dollar to account for any potential impacts of changes in the exchange rate. The outcomes confirm our previously obtained results. As an example, the respective 2, 50 and 98% quantile coefficients with innovations from EM Asia as regressor are 0.973, 0.569 and 1.081 for Malaysia, but 1.267, 1.275 and 1.280 for Korea if all values are denominated in JPY instead of USD. Finally, we regress all emerging markets on the shocks of the World index as done for the developed markets. The coefficients for a series of quantiles are listed in Table 4. The estimates reveal that Latin American countries exhibit a relatively high exposure to these types of shocks with an average coefficient estimate around one across all markets. Interestingly, this exposure is also relatively stable across all quantiles. The larger estimates for Latin American markets compared to the other emerging markets can be explained with different trading hours in Asia and Latin America. Since Latin America’s trading hours partly overlap with developed markets a higher loading to this factor is expected. The highest exposure to global shocks has Russia in the lower quantiles with a coefficient above two. South Africa’s exposure is significantly lower around one. Both countries also exhibit the u-shaped pattern of exposure but more pronounced for Russia.

6 The delayed propagation of systematic shocks can also be due to non-synchronous trading relevant for the European markets and Japan and the innovations of the MSCI World index.

D.G. Baur, N. Schulze / Int. Fin. Markets, Inst. and Money 19 (2009) 506–519

517

Table 4 World index as systematic factor Quantile 0.02

0.5

0.10

0.25

0.50

0.75

0.90

Germany France Japan UK USA

1.3136 1.0924 0.5560 0.8320 1.2204

1.3521 1.1327 0.5269 0.8376 1.1851

1.3586 1.1238 0.5133 0.8487 1.1528

1.3073 1.1241 0.4948 0.8340 1.1449

1.2580 1.0638 0.4831 0.8694 1.1393

1.2144 1.0415 0.5531 0.8710 1.1413

1.1698 1.0124 0.5431 0.8344 1.1362

0.95 1.1890 0.9936 0.5864 0.8682 1.1206

0.98 1.1651 0.8867 0.5885 0.7323 1.1372

China Hong Kong India Indonesia Korea Malaysia Philippines Singapore Taiwan Thailand

0.4510 0.5346 0.4237 0.8776 0.6035 0.5845 0.4461 0.6141 0.2681 0.4155

0.4197 0.5899 0.3535 0.4310 0.4553 0.3438 0.3170 0.5700 0.2889 0.5184

0.4896 0.5682 0.3553 0.3738 0.5271 0.2571 0.1782 0.4728 0.2493 0.4338

0.3541 0.4786 0.2518 0.2817 0.5727 0.1397 0.1640 0.4700 0.2744 0.3522

0.2954 0.58990 0.1623 0.2139 0.4279 0.0818 0.1405 0.4379 0.2094 0.2951

0.4347 0.5040 0.2297 0.2443 0.5441 0.1344 0.1603 0.4601 0.3151 0.3717

0.5099 0.5075 0.2854 0.2366 0.6587 0.2013 0.1398 0.4671 0.2384 0.4324

04534 0.5592 0.3559 0.3231 0.7610 0.4926 0.2494 0.4590 0.3443 0.6571

0.3043 0.8423 0.1915 0.8433 0.7729 0.4142 0.2247 0.7297 0.2624 0.8114

Russia South Africa

2.1322 0.9665

1.7616 0.9153

1.2251 0.8597

1.0036 0.7427

0.7391 0.6577

0.8326 0.6612

1.0339 0.7714

1.3011 0.6655

1.7853 0.9055

Argentina Brazil Chile Mexico

0.9459 1.3238 0.7397 1.4326

1.0218 1.2513 0.7204 1.3162

1.0071 1.2962 0.6533 1.2918

0.8843 1.2787 0.5590 1.1556

0.8640 1.2713 0.5239 1.1120

0.9224 1.2114 0.6032 1.0927

0.9624 1.1868 0.6074 1.1882

1.0061 1.2165 0.6801 1.2616

1.1767 1.3923 0.7609 1.4536

This table reports estimation results for all markets with the World index as regressor. It can be seen that the exposure to global shocks varies significantly over the countries in our sample. Moreover, the estimated coefficients show significant patterns across the depicted conditional quantiles.

The u-shaped pattern is also found for the emerging Asian markets but with a significantly lower exposure than reported in previous sections. The coefficient estimates measuring the exposure to emerging market shocks are approximately 0.5 higher on average across all quantiles compared to the exposure to global shocks. 5. Simulation study In order to clarify our approach, we present a small example with simulated values. We arbitrarily set the length of the dataset to 4000 and assume a ‘crisis period’ between observation 1801 and 2200.7 Then we construct a time series εt following an EGARCH(1,1)-process and drawn from a student tdistribution with five degrees of freedom (according to Campbell et al., 2000 and Bae et al., 2003 daily stock returns can be well described by t-distributions with degrees of freedom ranging between 3 and 6). The EGARCH-parameters (a = 0.01, b = −0.05, c = 0.15 and d = 0.97) are taken from an estimation on the Emerging Markets Asia index time series of our dataset. In the same way, we (independently) construct a time series vt and subsequently generate a market return ft = εt and a country return r1t = ft + vt . During the crisis period, we consider five different cases: (i) a benchmark model with no change in the propagation mechanism, (ii) an increased volatility of the underlying factor, (iii) an increased volatility of the country return, (iv) an increased volatility of both the factor and the country, and (v) an increased transmission mechanism. Fig. 4 present the average estimated quantile regression market coefficients from 100 repetitions for the five cases, respectively. It can be seen that the benchmark model and the settings with increased volatility show rather flat curves of the coefficient b(). In contrast, the fifth model (increased propagation) exhibits a pronounced u-shape similar to the results obtained for emerging markets in the

7 Splitting this period in several shorter crisis periods does not alter the results since quantile regression is not based on specific periods but utilizes the entire sample for the estimation.

518

D.G. Baur, N. Schulze / Int. Fin. Markets, Inst. and Money 19 (2009) 506–519

Fig. 4. Simulation results. The graph shows the mean coefficients of a simulation study based on 100 iterations with five different data-generating processes (benchmark model, increased volatility of the systematic shock, increased volatility of the idiosyncratic shock, increased volatility of both systematic and idiosyncratic shocks and an increased transmission of shocks during the crisis period). The graph presents the average coefficient estimates of the propagation of systematic shocks onto country-specific returns across the entire distribution (quantiles). Only an increased transmission of shocks leads to significantly different coefficient estimates across the quantiles (u-shape).

previous section. The fifth case can also be interpreted as contagion due to the increased co-movement of the systematic and the country-specific components in extreme market conditions compared to normal conditions. 6. Conclusions This paper presents a definition of financial market stability with an econometric testing framework. We find that the dependence of individual emerging markets on a regional or world index is increasing in extreme compared to normal market conditions. Together with the fact that the goodness of fit also increases in the tails of the conditional distribution, we argue that systematic shocks become more important and predictive for emerging markets in times of stress. In contrast, we do not find a similarly increased propagation mechanism for developed countries in extreme market conditions. The importance of systematic shocks is a new finding and emphasizes that a focus on idiosyncratic shocks in the analysis of financial crisis or contagion is not sufficient. The increasing importance of system-wide shocks is also interesting for investors and international portfolio diversification. The finding that systematic shocks are excessively propagated in emerging markets but not so in developed markets can explain why investors view emerging markets as one class of investment and simultaneously flee emerging markets in crisis periods. Moreover, it is important to stress that the excessive propagation of systematic shocks in emerging markets cannot be explained with a higher volatility of emerging markets compared to developed markets. Future research could apply the framework to bond markets or aim to analyze stock and bond markets simultaneously. References Ang, A., Chen, J., 2002. Asymmetric correlations of equity portfolios. Journal of Financial Economics 63, 443–494. Bae, K.-H., Karolyi, G.A., Stulz, R.M., 2003. A new approach to measuring financial contagion. Review of Financial Studies 16 (3), 717–763.

D.G. Baur, N. Schulze / Int. Fin. Markets, Inst. and Money 19 (2009) 506–519

519

Bank of England, 2004. Financial Stability Review—June 2004, Issue 16. Barnes, M.L., Hughes, A.W., 2002. A quantile regression analysis of the cross section of stock market returns, Federal Reserve Bank of Boston Working Paper No. 02-2. Baur, D., Schulze, N., 2005. Coexceedances in financial markets—a quantile regression analysis of contagion. Emerging Markets Review 6 (1), 21–43. Bassett, G.W., Chen, H.-L., 2001. Portfolio style: return-based attribution using quantile regression. Empirical Economics 26, 293–305. Bries, A., Goetzmann, W.N., Zhu, N., 2007. Efficiency and the bear: short sales and markets around the world. Journal of Finance 62 (3), 1029–1079. Buchinsky, M., 1998. Recent advances in quantile regression—a practical guide for empirical research. Journal of Human Resources 33 (1), 88–126. Campbell, R., Koedijk, K., Kofman, P., 2000. Covariance and Correlation in International Equity Returns: A Value-at-Risk Approach. Erasmus University Rotterdam. Cihak, M., 2007. Central Banks and Financial Stability: A Survey, IMF Working Paper. Das, S.R., Uppal, R., 2004. Systemic risk and international portfolio choice. Journal of Finance 59 (6), 2809–2834. Engle, R.F., Manganelli, S., 2004. CAViaR: conditional autoregressive value at risk by regression quantiles. Journal of Business and Economic Statistics 22 (4), 367–381. European Central Bank, 2005. Financial Stability Review—June 2005. European Central Bank, 2007. Financial Stability Review—June 2007. Federal Reserve Board, 2006. Thoughts on Financial Stability and Central Banking, Speech of Vice Chairman Roger W. Ferguson, Jr. on April 17, 2006. Forbes, K.J., Rigobon, R., 2002. No contagion, only interdependence: measuring stock market comovements. Journal of Finance 57 (5), 2223–2261. Goodhart, C.A.E., 2006. A framework for assessing financial stability? Journal of Banking and Finance 30, 3415–3422. International Monetary Fund, 2003. Global Financial Stability Report—Market Developments and Issues, World Economic and Financial Surveys. Koenker, R., Bassett, G., 1978. Regression quantiles. Econometrica 46 (1), 33–50. Koenker, R., Bassett, G., 1982. Robust tests for heteroscedasticity on regression quantiles. Econometrica 50 (1), 43–61. Morck, R., Yeung, B., Yu, W., 2000. The information content of stock markets: why do emerging markets have synchronous stock price movements? Journal of Financial Econometrics 58, 215–260. Oosterloo, S., de Haan, J., 2004. Central banks and financial stability: a survey. Journal of Financial Stability 1, 257–273. Padoa-schioppa, T., 2003. Central Banks and Financial Stability, European Central Bank. Schinasi, G.J., 2004. Defining Financial Stability, IMF Working Paper WP/04/187. Taylor, J.W., 2008. Using exponentially weighted quantile regression to estimated value at risk and expected shortfall. Journal of Financial Econometrics 6 (3), 382–406. Yuan, K., 2005. Asymmetric price movements and borrowing constraints: a rational expectations equilibrium model of crises, contagion, and confusion. Journal of Finance 60, 379–411.