Finx

International Review of Financial Analysis 48 (2016) 376–387

Contents lists available at ScienceDirect

International Review of Financial Analysis

Banking industry performance in the wake of the global financial crisis☆

Diptes C. Bhimjee a, Sofia B. Ramos b,⁎, José G. Dias c a Instituto Universitário de Lisboa (ISCTE-IUL), Lisboa 1649-026, Portugal b ESSEC Business School, Cergy-Pontoise 95000, France c Instituto Universitário de Lisboa (ISCTE-IUL), BRU-IUL, Lisboa 1649-026, Portugal

☆ The authors would like to thank the discussants and p Conference on International Finance (Prato, Italy), the X Spain), the Bank of Portugal's Conference on “Econome Finance”, and two anonymous referees for their valuab more specifically, we would like to thank INFINITI's Ed Kan able insights on the future development of the HRSM's the acknowledges support from BRU-UNIDE, Instituto Unive Portugal. ⁎ Corresponding author.

E-mail address: [email protected] (S.B. Ramos). 1 A fundamental distinction between the ‘subprime’ cri

the ensuing global financial crisis (as a truly global financia tional financial contagion processes) is thus respected thr

http://dx.doi.org/10.1016/j.irfa.2016.01.005 1057-5219/© 2016 Elsevier Inc. All rights reserved.

a b s t r a c t

a r t i c l e i n f o

Available online 27 January 2016

JEL classification: G01 G15 F30

Keywords: Global financial crisis International financial contagion ‘Subprime’ crisis Banking institutions Heterogeneous regime-switching model (HRSM)

This paper analyzes the performance of the banking industry both prior to and during the global financial crisis (GFC). Through the application of a panel regime-switching model designed to capture heterogeneity, our find- ings suggest that global banking performance can be grouped into two distinctive clusters, each with its own spe- cific regime dynamics. Before the crisis, a cluster of banking institutions pertaining to advanced economies stood out for its buoyant stock market performance, whereas a second cluster, mainly composed of banking indexes that belong to emerging economies, exhibited a more subdued performance. Further, this differentiation was ac- companied by low regime synchronization between the clusters. During the crisis, banking institutions behaved similarly, regime synchronization increased, and the differences in the regime dynamics vanished. Finally, the GFC constituted a highly synchronized and systemic extreme financial event, as evidenced by our findings depicting the onset of severe underlying international financial contagion processes.

© 2016 Elsevier Inc. All rights reserved.

1. Introduction

The global financial crisis of 2007–2009 (hereafter, the GFC) consti- tuted a resounding systemic failure that had a profound effect on finan- cial markets and, more specifically, on international banking institutions that operate in increasingly borderless markets.1 The banking industry has been indelibly marked by this unprecedented credit event of global magnitude, and its worldwide implications continue to be felt.

As a result of severe international financial contagion processes, the GFC ultimately affected the valuation of outstanding securitized assets worldwide and the performance of the corresponding banking institu- tions holding assets of uncertain (or even ‘toxic’) value. This global un- certainty regarding the soundness of bank balance sheets affected the valuation of the financial industry worldwide, and the negative outlook for future profitability led to severe falls in stock returns. Nevertheless,

articipants of the 12th INFINITI XII Finance Forum (Zaragoza, tric Methods for Banking and le comments and suggestions; e (Boston College) for his valu-

oretical framework. Sofia Ramos rsitário de Lisboa (ISCTE-IUL),

sis (as a localized US event) and l event associated with interna- oughout this paper.

and despite the pervasiveness of international financial contagion throughout the GFC, banks were heterogeneously affected due to their distinct balance sheet exposure to the global systemic shock. According- ly, the stock market valuation of international banking industries was very heterogeneous throughout the GFC, in view of the differing impacts of the systemic crisis on the banking systems of each country.

This paper studies the main characteristics of this pivotal element of heterogeneity in the global banking industry. It addresses the hetero- geneous stock market performance of the banking industry and, in particular, considers the impact of the GFC. The paper uses a novel methodology that was designed to capture both heterogeneity and in- ternational financial contagion processes. The sample is composed of 42 country banking indexes and encompasses a mix of emerging and advanced markets.

The paper has two main goals. First, it aims to characterize the de- gree of heterogeneity in banking industry performance in both the up- swing and downswing of the business cycle associated with the GFC. Banking industry indexes are used to capture the aggregate perfor- mance of banks because the market valuation reflects the investors' as- sessment of the future profitability of the banking industry.2

Second, given the global impact of this systemic breakdown, this study's objective is to assess the main attributes of international finan- cial contagion processes during the GFC.

2 A natural implicit assumption is that markets are “informationally efficient”; that is, investors are able to incorporate the new information coming from the crisis into asset prices.

377D.C. Bhimjee et al. / International Review of Financial Analysis 48 (2016) 376–387

Although some studies have addressed financial contagion in this period, they have focused either on stock market returns (refer to e.g., Dimitriou, Kenourgios, & Simos, 2013; Kotkatvuori-Örnberg, Nikkinen, & Äijö, 2013) or volatility spillovers between banking indus- tries (Choudhry & Jayasekera, 2014). Dimitriou et al. (2013) investigate the contagion effects of the global financial crisis in a multivariate frac- tionally integrated asymmetric power ARCH (FIAPARCH) dynamic con- ditional correlation (DCC) framework, using the stock market returns of the BRICS countries and the United States. Kotkatvuori-Örnberg et al. (2013) use data from 50 equity markets to examine both the condition- al and unconditional correlations around two major events during the financial crisis of 2007–2009. The researchers found that the Lehman Brothers' collapse resulted in a significant increase in correlations. Choudhry and Jayasekera (2014) investigate the influence of the global crisis on the spillover between the banking indexes of Europe and the United States by using a multivariate GARCH–GJR framework. Their re- sults indicate an increase in both means and volatility spillover between the major economies and the stressed EU economies from the pre-crisis to the crisis period, but minimal evidence of a significant spillover from the smaller economies to the major economies.

This paper addresses the heterogeneous performance of banking in- dustries at a global level, both before and during the 2007–2009 crisis. Our paper captures the dynamics associated with international financial contagion processes among financial institutions during the GFC. Ac- cordingly, we propose an extension of the regime-switching model, the heterogeneous regime-switching model (HRSM), to achieve this dual purpose. This model extension allows us to distinguish between the likelihood of switching between regimes among the heterogeneous sub-sets of country banking indexes under analysis. The approach also expands on the existing methodologies that do not currently allow mar- ket regimes to be incorporated in the analysis of crises. The proposed methodology further accounts for the problem of non-normality in financial returns that often occurs in emerging markets.3 The flexible modeling of observed returns using a mixture of normal distributions enables nearly any departure from normality to be captured in a straightforward manner.4 In addition, the inclusion of market regimes in the modeling of financial time series is suitable due to structural breaks, e.g., regime switching due to pro-cyclicity and the asymmetry of volatility (see, e.g., Baele (2005); Billio and Pelizzon (2003), and Kearney and Potì (2008)), is obtained endogenously within our model applications.

Our main findings are summarized as follows. Our results suggest that the banking industries depicted in our sample can be disentangled into two clusters (based on the best clustering identified by the mini- mum Bayesian information criterion (BIC)). The first encompasses advanced economies, whereas the second is chiefly composed of emerg- ing market economies. The groups are mainly distinguished because the emerging market group was not buoyant before the GFC but was subse- quently affected by it. The evidence suggests that co-movements in- crease during the GFC because we find a large synchronization in these two groups for all regimes. There are nearly no detectable differences between the groups in the regime dynamics during the crisis period.

Furthermore, our results indicate differentiated or heterogeneous response dynamics emanating from the performance of the banking in- dustries. Shehzad and De Hann (2013) also observe this heterogeneity, as they find that stock prices of banks in emerging countries were less affected by the systemic shock than were the corresponding prices of their counterparts in developed economies. Our findings are similar, although they have been endogenously determined by applying the HRSM in a global macro-financial setting.

Moreover, the pattern found is in accordance with the stylized fact that crises are typically associated with temporary changes in

3 See, for instance, Harvey (1995) or Susmel (2001). 4 See, for example, Dias and Wedel (2004) and McLachlan and Peel (2000) on the use of

mixture models to address unobserved heterogeneity.

fundamentals. Beltratti and Stulz (2012) argue that the expectations as- sociated with bank stock returns were significantly different before and after the GFC. Before the crisis, stock markets favored banking business strategies that involved financial innovation-related products. Subse- quently, the onset of the GFC shifted market expectations in favor of more conservative banking business strategies that promoted staple products.

The current work helps clarify the process of contagion during the global financial crisis. Our paper offers a complementary view because it focuses on the heterogeneity across cycles, which has not been ad- dressed in previous work.

The findings have important asset management implications, as de- veloped by Ahmad, Bhanumurthy, and Sehgal (2015), because changes in correlations imply changes in portfolio weights. The results suggest that the origin of the crisis was located in developed countries, but quickly spread to emerging countries, which highlights the systemic na- ture of the crisis.

These findings are important on two aspects: first, they are essential when assessing the performance of banking industries throughout crisis episodes. Second, they may be useful in the subsequent design, by cen- tral banking institutions, of coordinated support policies for the banking industry, particularly in the aftermath of systemic episodes that severely constrain the banking industry. The results are also of interest for regula- tors and demand measures to prevent contagion in such an important area.

Section 2 presents a short description of regime-switching models and fully depicts our methodology, the heterogeneous regime- switching model (HRSM). Section 3 presents preliminary consider- ations regarding the data set. Section 4 describes the empirical findings pertaining to the model applications encompassing the GFC. Finally, Section 5 summarizes our main findings.

2. Methodology

This section introduces the statistical framework, the heterogeneous regime-switching model (HRSM), which is based on an extension of panel regime-switching models.

Hamilton (1989) was the first to show how regime-switching models (RSM) can be useful in macroeconomic data modeling by allowing non-linear instead of linear stationary processes. RSM has be- come very popular because it captures the ‘turning points’ in a given economic time series as discrete regime shifts in the behavior of the time series. This behavior is naturally connected to the existence of dra- matic breaks (or discontinuities) in the economic time series and is often associated with the occurrence of financial crises and economic cycles (Bhar & Hamori, 2004). Therefore, these models are suited to an- alyzing and characterizing both the ‘turning points’ and abrupt changes (discontinuities) that occur in economic and financial time series that are affected by the occurrence of extreme, but reversible, financial events such as the GFC.

Heterogeneous regime-switching models are an extension of the Markov-switching model and were initially developed by Dias, Vermunt, and Ramos (2008) and Ramos, Vermunt, and Dias (2011). This model can also be viewed as an extension of the hybrid model in- troduced by Dias and Ramos (2014), which estimates a panel regime switching model and then, on the posterior probabilities, applies heuris- tic cluster analysis to identify the hierarchical structure of the stock mar- ket. The HRSM enables the statistical estimation of regime-switching models based on the similarity of the dynamics associated with each ho- mogeneous group (or cluster), i.e., clusters and regimes are estimated simultaneously. A model with S groups is denominated HRSM-S. To achieve this estimation, essentially, two types of clustering are assumed. Each underlying time series is both assigned to a specific cluster and modeled as a regime-switching model within each cluster.

Let yit represent the return, at time t, of each country banking index contemplated in our sample, where i∈{1, …,n} and t∈{1, …,T}, with

Table 1 Characterization of the sample banking industries (year 2010).

Countries Bank deposits to GDP (%)

Bank branches per 100,000 adults

Bank credit to bank deposits (%)

Foreign banks among total banks (%)

GDP per capita

Argentina (AR) 19 13 67 33 Australia (AU) 94 31 127 42 36,175 Austria (OE) 97 11 125 11 40,099 Belgium (BG) 104 45 88 43 37,745 Brazil (BR) 49 44 105 38 5,618 Canada (CN) 24 39 36,467 Chile (CL) 37 17 168 43 8,610 China 51 235 21 2,870 Czech Republic (CZ) 62 22 67 14,640 Denmark (DK) 55 41 9 47,792 Finland (FN) 62 15 149 22 39,698 France (FR) 80 42 140 5 35,216 Germany (BD) 114 16 92 14 37,146 Greece (GR) 103 40 123 22 21,894 Hong Kong (HK) 300 24 59 73 31,329 Hungary (HN) 47 17 82 11,109 India (IN) 60 10 76 12 1,032 Ireland (IR) 108 29 197 85 46,424 Israel (IS) 82 19 105 23,224 Italy (IT) 84 66 133 10 30,788 Japan (JP) 208 34 50 1 36,296 Luxembourg (LX) 339 89 55 96 81,565 Malaysia (MY) 119 11 89 40 6,319 Mexico (MX) 25 14 70 39 8,085 Netherlands (NL) 132 23 153 45 43,675 Norway (NW) 11 2 64,590 Pakistan (PK) 29 8 69 42 748 Peru (PE) 28 47 80 67 3,575 Philippines (PH) 51 8 55 11 1,403 Poland (PO) 46 32 74 10,038 Portugal (PT) 116 66 151 37 19,240 Russian (RS) 36 35 107 20 6,386 Singapore (SG) 114 10 81 50 34,758 South Africa (SA) 59 10 119 24 5,911 Spain (ES) 158 97 133 8 26,191 Sweden (SD) 53 23 1 44,878 Switzerland (SW) 133 52 116 21 58,140 Taiwan (TA) 24 18,572* Thailand (TH) 94 11 99.5 24 3,164 Turkey (TK) 49 18 85 39 7,834 United Kingdom (UK) 25 58 39,472 United States (US) 80 35 66 32 43,961

Source: World Bank; *Chen and Liu (2013) for Taiwan. GDP per capita is at constant prices in USD.

378 D.C. Bhimjee et al. / International Review of Financial Analysis 48 (2016) 376–387

S and K being the number of clusters and regimes, respectively. Let f(yi;ψ) be the probability density function associated with the banking index return rate pertaining to country i. The HRSM-S is given by

f yi; ψð Þ ¼ ∑S wi¼1

∑ K

zi1¼1 ∑ K

zi2¼1 … ∑

K

ziT ¼1 f wi; zi1; …; ziTð Þf yijwi; zi1; …; ziTð Þ ð1Þ

The right side of Eq. (1) indicates that the underlying model struc- ture is typical of a mixture model consisting of the time-constant latent variable wi and T realizations of the time-varying latent variable zit. The observed data density f(yi;ψ) is obtained by marginalizing over the latent variables and is provided by the total probability theorem in which the marginal probability is obtained by the sum of conditional probabilities over the partitions {wi, zi1, … , ziT}. Furthermore, in view of the Markov assumption for the sequence {zi1, … , ziT}, the term (wi, zi1, … , ziT) of Eq. (1) can be further transformed into.

f wi; zi1; …; ziTð Þ ¼ f wið Þf zi1jwið Þ∏Tt¼2 f zitjzi;t�1; wi � �

ð2Þ

where f(wi) essentially represents the probability of a specific country's banking index belonging to a given cluster w, the multinomial parame- ter λw=P(Wi = w), f(zi1 | wi) represents the initial-regime probability, and f(zit | zi , t - 1, wi) represents the latent transition probability of the

Markov process in cluster wi. Moreover, the observed index return value depends solely on the regime that is applicable at that specific chrono- logical point, i.e., response yit is independent of the returns at other mo- ments (this is known as the local independence assumption). Simultaneously, the observed value is also independent of regimes at other times. These assumptions can be formulated as follows:

f yijwi; zi1; …; ziTð Þ ¼ ∏ T

t¼1 f yitjzitð Þ ð3Þ

where the probability density that a particular observed index return value at time t conditional on the regime in place at that chronological point, f(yit | zit), is assumed to follow a univariate Gaussian density function.

In addition, the parameters of the HRSM-S are estimated using max- imum likelihood (ML) estimation, where the log-likelihood function is

↕ψ; yð Þ ¼ ∑ni¼1logf yi; ψð Þ: ð4Þ The expectation–maximization algorithm can subsequently be

employed to solve this maximization problem. Nevertheless, it should be noted that the application of the expectation–maximization algo- rithm requires both a lengthy computational effort and a cumbersome computer storage capacity. To circumvent these computational issues,

6

15

278 Argentina

694

2298 Australia

159

1226 Austria

132

2557 Belgium

44

765 Brazil

665

2653 Canada

0

5 Chile

1

10 Czech Rep.

127

856 Denmark

123

551 Finland

281

1538 France

131

918 Germany

680

4584 Greece

270

930 Hong Kong

3

37 Hungary

24

370 India

119

11225 Ireland

23

111 Israel

908

4284 Italy

0

4 Japan

492

1914 Luxembourg

221

754 Malaysia

1527

7854 Mexico

247

6782 Netherlands

19

128 Norway

2

69 Pakistan

19

191 Peru

4

20 Philippines

22

149 Poland

87

482 Portugal

6

329 Russia

180

553 Singapore

142

896 South Africa

205

767 Spain

105

499 Sweden

341

1404 Switzerland

Jan02 Mar04 May06 Jun08 Aug10 2

8 Taiwan

Jan02 Mar04 May06 Jun08 Aug10 2

10 Thailand

Jan02 Mar04 May06 Jun08 Aug10 77777

1234370 Turkey

Jan02 Mar04 May06 Jun08 Aug10 2461

18859 United Kingdom

Jan02 Mar04 May06 Jun08 Aug10 325

1885 United States

Fig. 1. Time series of country banking indexes (in USD). Source of underlying data: Datastream; China is not included in this figure due to unavailability of full data.

379D.C. Bhimjee et al. / International Review of Financial Analysis 48 (2016) 376–387

a special variant of the expectation–maximization algorithm, the Baum–Welch algorithm, has been advanced in the literature and allows the abovementioned maximization problem to be more easily solved (Dias et al., 2008).

Furthermore, the choice of the appropriate number of clusters (S) and regimes (K) is traditionally based on the analysis of statistical infor- mation criteria. For example, to identify the number of clusters (S), the Bayesian information criterion (BIC) value was employed. Thus, the most appropriate pair of (S, K) values is selected for all of our model ap- plications until the optimized (i.e., minimized) value of the BIC is achieved.

3. Data

The sample is composed of countries' banking indexes5 and includes a diversified set of developed and emerging market banking industries. This sample diversity facilitates our analysis of the heterogeneity of the regime dynamics associated with the impact of the GFC on different rep- resentative banking systems and institutions worldwide.

5 Datastream uses the Industry Classification Benchmark (ICB) to classify firms in indus- tries; ICB was jointly created by FTSE and Dow Jones. The indexes have several layers, de- pending on the level of detail of the industry that the user wants. To be more specific, the ICB classifies the aggregated market into 10 industries, divides them into 19 supersectors, and then subdivides them into 41 sectors, for a total of 114 subsectors. We use the Banks sector. According to Datastream, indexes are calculated on a representative list of stocks for each market. The number of stocks for each market is determined by the size of the market. The sample covers a minimum of 75–80% of total market capitalization. Sector and market ag- gregations are weighted by market value. More details regarding the indices can be found in: A Guide to the Industry Classification Benchmark, 2012 in http://www.icbenchmark.com Datastream Global Equity Indices User Guide: http://extranet.datastream.com/data/ Equity%20indices/Index.htm

Table 1 presents the sample of countries6 and certain indicators re- lated to the size and development of the banking sector. The economic importance of banks is high in countries such as Hong Kong, Japan, and Luxembourg, where we find a large ratio of bank deposits to gross domestic product (GDP). If we examine their retail presence, we con- clude that the number of bank branches per 100,000 adults is high in Luxembourg, Italy, Portugal, and Spain.

Foreign banks are largely important in Luxembourg, Ireland and Hungary, whereas in countries such as Japan, Norway or Sweden, they have small importance. Bank credit to bank deposits is high in countries such as Ireland or China. Finally, the last column presents GDP per capita, which ranges from 748 USD in Pakistan to 81,565 in Luxembourg.

The collected indexes7 depict the stock market valuation of the most significant financial institutions of any given country (or jurisdiction);

The country selection is contingent on data availability from our data source. More- over, the methodology requires a “balanced panel”. Further, we excluded specific coun- tries that present little liquidity, i.e., for which zero returns consequently repeat. We note also that these indexes are related to either countries or special administrative re- gions (Hong Kong). These disentangle neither cross-border ownership of banks nor the di- rect influence of the shadow banking system; we thank an anonymous referee for noting this. Due to data availability constraints, Chinese banking institutions are only included in our model applications for the 2007–2010 period; the corresponding findings are included in the Supplementary Appendix, which is available upon request.

7 These indexes represent a proxy for the performance of each country's banking indus- try; the main advantage of using these indexes resides in the comparability among the se- ries; accordingly, the paper's framework involves the global macro-financial architecture and the propagation of international financial contagion processes through the global banking industry during the preceding decade. Regrettably, the influence of the shadow banking system cannot be properly addressed by our data because the highly complex ‘vertical slicing’ of the credit intermediation process of traditional banks into a highly spe- cialized network of credit-related specialisms is performed by non-bank financial inter- mediaries, which are typically not publicly quoted (Pozsar, Adrian, Ashcraft, & Boesky, 2012). However, the impact of the role of these specialized financial intermediaries is assessed through international financial contagion, the latter being very observable in our analysis of the GFC. We want to thank an anonymous referee for highlighting this issue.

Table 2 Summary statistics of banking index returns. This table reports descriptive statistics of banking indexes (weekly), namely the mean, standard deviation (Std. Deviation), skewness, and kurtosis. Returns are computed as the first dif- ferences of the logarithm of prices and presented in percentage. The last column presents the Jarque–Bera test of normality and respective p-values. All returns are in USD. Sample period is from January 2, 2002 to August 25, 2010.

Country Mean Median Std. deviation Skewness Kurtosis Jarque–Bera test

[both adjusted for bias] statistics p-value

Argentina (AR) 0.145 0.455 5.695 −0.938 8.474 605.80 0.000 Australia (AU) 0.153 0.546 4.049 −1.000 10.382 1059.51 0.000 Austria (OE) 0.235 0.735 5.457 −0.784 7.122 351.07 0.000 Belgium (BG) −0.171 0.555 7.132 −1.185 11.455 1398.17 0.000 Brazil (BR) 0.408 0.770 5.919 −1.538 12.732 1890.95 0.000 Canada (CN) 0.218 0.385 3.510 −0.255 5.763 141.19 0.000 Chile (CL) 0.295 0.501 3.239 −1.169 12.716 1811.47 0.000 Czech Rep. (CZ) 0.383 0.393 5.714 −0.588 8.437 558.70 0.000 Denmark (DK) 0.086 0.455 4.798 −0.728 10.466 1047.40 0.000 Finland (FN) 0.217 0.460 4.723 −1.515 16.157 3310.61 0.000 France (FR) 0.018 0.524 5.583 −0.057 6.915 275.67 0.000 Germany (BD) −0.047 0.299 5.519 −0.896 9.300 776.30 0.000 Greece (GR) −0.112 0.428 5.383 −0.699 6.793 294.66 0.000 Hong Kong (HK) 0.028 0.132 3.348 −0.477 11.774 1410.62 0.000 Hungary (HN) 0.266 1.063 7.465 −1.320 11.541 1449.62 0.000 India (IN) 0.596 0.690 5.809 0.316 6.312 203.94 0.000 Ireland (IR) −0.438 0.134 11.672 −0.253 29.401 12,675.42 0.000 Israel (IS) 0.176 0.326 4.362 −0.109 5.780 138.94 0.000 Italy (IT) −0.024 0.247 4.475 −0.446 6.105 187.33 0.000 Japan (JP) −0.007 0.000 4.582 −0.049 4.423 35.67 0.000 Luxembourg (LX) 0.064 0.278 3.129 −0.414 8.534 565.11 0.000 Malaysia (MY) 0.255 0.284 2.656 −0.120 5.147 82.95 0.000 Mexico (MX) 0.257 0.554 3.986 −0.929 12.572 1723.72 0.000 Netherlands (NL) −0.487 0.283 8.462 −8.634 132.343 310,152.74 0.000 Norway (NW) 0.198 0.587 6.170 −0.908 11.909 1498.24 0.000 Pakistan (PK) 0.471 0.642 5.356 −0.894 6.583 289.65 0.000 Peru (PE) 0.457 0.191 3.175 0.705 10.457 1042.45 0.000 Philippines (PH) 0.182 0.150 3.527 −0.004 5.511 112.41 0.000 Poland (PO) 0.241 0.599 5.492 −1.349 10.166 1063.42 0.000 Portugal (PT) −0.139 0.142 3.892 −0.661 6.135 208.41 0.000 Russia (RS) 0.753 0.983 6.636 −0.625 13.104 1878.83 0.000 Singapore (SG) 0.155 0.082 3.684 0.453 10.366 996.53 0.000 South Africa (SA) 0.324 0.466 5.451 −0.644 7.348 370.92 0.000 Spain (ES 0.047 0.291 4.618 −0.398 6.032 176.32 0.000 Sweden (SD) 0.114 0.531 5.122 −0.599 7.356 368.01 0.000 Switzerland (SW) 0.012 0.139 4.925 −0.236 6.579 233.91 0.000 Taiwan (TA) 0.178 0.206 4.363 0.019 5.265 91.27 0.000 Thailand (TH) 0.308 0.252 4.525 −0.088 3.760 10.35 0.006 Turkey (TK) 0.398 0.912 7.048 −0.596 4.698 77.19 0.000 United Kingdom (UK) −0.146 0.105 5.129 −0.691 13.647 2089.84 0.000 United States (US) −0.159 0.017 4.903 −0.167 12.091 1498.63 0.000

380 D.C. Bhimjee et al. / International Review of Financial Analysis 48 (2016) 376–387

they have been extracted from the Datastream database using a weekly frequency and are in United States dollars (USD) to facilitate interna- tional comparisons. The original source of data is the Industry Classifica- tion Benchmark (ICB), which collectively depicts more than 8000 banks that provide a broad range of financial services (including retail bank- ing, loans and money transmissions).

Accordingly, the beginning point of our indexes is 2002 (more spe- cifically, January 2), and the end-point of our data is August 25, 2010. The starting date was chosen for two reasons. First, this date is the year subsequent to the occurrence of the previous global financial crisis, the 2001 ‘dot-com’ crisis. Second, our choice concurs with the beginning of the upward trajectory of the ‘subprime’ crisis, which was a related business cycle that led directly to the present crisis.8 Further, this ample time frame allows us to have a broad overview of the crisis. That is, the time interval between 2002 and 2010 encompasses not only the upward phase of the ‘subprime’ global business cycle prior to the occurrence of the GFC but also the ensuing downward phase.

8 For example, in the case of the United States, the epicenter of the present crisis, the of- ficial business cycle dating committee, the National Bureau of Economic Research (NBER), dated this upward phase associated with the ‘subprime’ cycle as between November 2001 and December 2007; however, the ‘subprime’ crisis was dated as between December 2007 and June 2009 (National Bureau of Economic Research, 2010).

Fig. 1 portrays the data used herein and describes the global evolution of the stock market valuation of each banking industry in the countries included in our sample, considering our adopted timeline.

Moreover, Table 2 provides summary statistics pertaining to the country banking data collected. In addition to presenting the standard descriptive statistics associated with each country's banking index, the table presents the respective results for the Jarque–Bera statistic. The re- sults indicate that the null hypothesis of normality can be safely rejected. The mean returns are negative for the Netherlands, Ireland, Belgium, the United Kingdom, and the United States and are very high in countries such as Russia, India, Pakistan, and Peru. The standard devi- ation is the lowest in the United States and is very high in countries such as Russia, Turkey, and Brazil.

4. Empirical results

4.1. Banking indexes in the 2002–2010 period

This section presents the results of the model. The model selection criterion (BIC) identifies the existence of heterogeneity (S) and a multi-regime (K) framework simultaneously. The minimization of the BIC criterion, which is equal to 104,031.14, yields an optimal solution of two clusters and three regimes (S=2,K =3). That is, our findings

Table 3 Estimated prior probabilities, posterior probabilities and modal classes for HRSM-2. This table reports the classification of banking indexes by clusters. Prior probabilities pro- vide the size of each cluster or group and posterior probabilities express the evidence that a given stock market belongs to a given cluster. The maximum posterior probability indi- cates the assignment to the modal cluster.

Cluster 1 Cluster 2 Cluster

Prior probabilities 0.296 0.704

Posterior probabilities Argentina (AR) 0.997 0.003 1 Australia (AU) 0 1 2 Austria (OE) 0.001 0.999 2 Belgium (BG) 0 1 2 Brazil (BR) 0.955 0.046 1 Canada (CN) 0 1 2 Chile (CL) 0 1 2 Czech Rep. (CZ) 0.996 0.004 1 Denmark (DK) 0 1 2 Finland (FN) 0 1 2 France (FR) 0 1 2 Germany (BD) 0 1 2 Greece (GR) 0.044 0.956 2 Hong Kong (HK) 0 1 2 Hungary (HN) 0.959 0.041 1 India (IN) 0.98 0.02 1 Ireland (IR) 0 1 2 Israel (IS) 0.976 0.025 1 Italy (IT) 0 1 2 Japan (JP) 0.03 0.97 2 Luxembourg (LX) 0 1 2 Malaysia (MY) 0 1 2 Mexico (MX) 0.005 0.995 2 Netherlands (NL) 0 1 2 Norway (NW) 0.002 0.998 2 Pakistan (PK) 1 0 1 Peru (PE) 0 1 2 Philippines (PH) 0 1 2 Poland (PO) 1 0.001 1 Portugal (PT) 0 1 2 Russia (RS) 0.993 0.008 1 Singapore (SG) 0 1 2 South Africa (SA) 0.998 0.002 1 Spain (ES) 0 1 2 Sweden (SD) 0 1 2 Switzerland (SW) 0 1 2 Taiwan (TA) 0.976 0.024 1 Thailand (TH) 0.95 0.05 1 Turkey (TK) 0.072 0.929 2 United Kingdom (UK) 0 1 2 United States (US) 0 1 2

Table 5 Estimated cluster-specific probabilities of regimes, regime occupancy for each cluster, re- gime transition and sojourn time. P(Z | W) represents the proportion of banking indexes in each regime for each cluster. Re- maining rows report transition probabilities between regimes. Standard errors are report- ed in round brackets. Sojourn time represents the time expected for banking indexes to exit a given regime.

Cluster 1 Cluster 2

Regime 1

Regime 2

Regime 3

Regime 1

Regime 2

Regime 3

P(Z|W) 0.091 0.711 0.198 0.090 0.400 0.510 (0.021) (0.030) (0.029) (0.015) (0.028) (0.034)

Regime 1 0.931 0.069 0.001 0.939 0.061 0.000 (0.015) (0.016) (0.004) (0.010) (0.010) (0.000)

Regime 2 0.008 0.950 0.042 0.014 0.962 0.024 (0.002) (0.012) (0.011) (0.002) (0.004) (0.004)

Regime 3 0.001 0.148 0.851 0.000 0.019 0.981 (0.003) (0.030) (0.029) (0.000) (0.002) (0.002)

Sojourn time (weeks) 14.388 19.920 6.693 16.420 26.110 52.632

381D.C. Bhimjee et al. / International Review of Financial Analysis 48 (2016) 376–387

suggest that there are two distinct clusters of countries operating under the different dynamics of three very distinctive regimes (thus, in addi- tion to the two end-of-spectrum bull and bear market regimes, there is an intermediate regime). Both the composition of clusters and the characteristics of regime profiles have been endogenously determined within each model application. That is, the latter are determined by each of the models' empirical applications.

Table 3 summarizes the results pertaining to the estimated prior class probabilities (the cluster dimension), the posterior probabilities associated with the distribution of the banking industries across the

Table 4 Estimated marginal probabilities of regimes, mean returns and variances – 2002–2010. This table reports the estimated marginal probabilities of regimes. P(Z) is the average proporti regimes. The last three column is the variance of the returns in each regime. Standard errors a

P(Z) Return (m

Regime 1 Regime 2 Regime 3 Regime 1 Regime

0.091 0.492 0.417 −1.850 0.250 (0.012) (0.031) (0.034) (0.349) (0.055)

two clusters (reflecting the degree of membership to each cluster), and the respective modal cluster. The estimated prior class probabilities are 0.296 (cluster 1) and 0.704 (cluster 2), which reflect the fact that the first cluster is significantly smaller than the second and that banking in- dustries are unevenly distributed across these two clusters for the 2002–2010 period. The estimated posterior cluster probabilities reflect the degree of membership associated with each of the clusters, and these probabilities are conditional on the observed data. The modal cluster column ascribes each country to a specific cluster, considering these probabilities. Thus, cluster 1 is composed of the following 12 banking indexes: Argentina, Brazil, Czech Republic, Hungary, India, Israel, Pakistan, Poland, Russia, South Africa, Taiwan, and Thailand, i.e., the group is formed mainly by emerging markets. Conversely, clus- ter 2 is composed of the following 29 indexes: Australia, Austria, Belgium, Canada, Chile, Denmark, Finland, France, Germany, Greece, Hong Kong, Ireland, Italy, Japan, Luxembourg, Malaysia, Mexico, Netherlands, Norway, Peru, Philippines, Portugal, Singapore, Spain, Sweden, Switzerland, Turkey, the United Kingdom, and the United States.

The regime's profile and respective dynamics are described in Table 4. There is a 0.091 (regime 1), 0.492 (regime 2) and 0.417 (regime 3) probability that the banking indexes may be in one of the three re- gimes. Regime 1 exhibits negative returns and a high degree of volatility (−1.85 and 177.17, respectively); regime 2 exhibits positive returns as- sociated with a much lower degree of volatility (0.25 and 22.64, respec- tively); and regime 3 exhibits the highest (positive) returns with the lowest volatility (0.43 and 4.94, respectively). That is, regime 1 is mark- edly associated with bear market dynamics, regime 2 is associated with mild bull market dynamics, and regime 3 is associated with buoyant bull market dynamics. The results are in accordance with the common ac- knowledgement of the presence of asymmetric volatility in financial markets, i.e., volatility is very likely to be higher when the performance of financial markets is faltering and lower when the performance is buoyant. The regime heterogeneity described herein is in accordance with previous studies (such as Guidolin and Timmermann (2007)),

on of markets in each regime over time. The next three columns are the log returns of the re reported in round brackets.

ean) Risk (variance)

2 Regime 3 Regime 1 Regime 2 Regime 3

0.434 177.170 22.641 4.943 (0.029) (8.169) (0.649) (0.146)

382 D.C. Bhimjee et al. / International Review of Financial Analysis 48 (2016) 376–387

which incorporate and validate regime heterogeneity across much lon- ger time frames.

Let P(Z | W) represent the estimated probability that each cluster's set of country banking industries is in a given regime, conditional on the specificities of each individual cluster. The results in Table 5 suggest that the banks associated with cluster 1 have a 0.091 probability of being in a bearish environment, a 0.711 probability of being in a mild bull environment, and a 0.198 probability of being in a bullish environ- ment. Similarly, the banking indexes associated with cluster 2 have a 0.09 probability of being in a bearish environment, a 0.40 probability of being in a mild bull environment, and a 0.51 probability of being in a bullish environment. That is, banking industries in the latter cluster have a higher probability of operating under more bullish financial con- ditions during the adopted time frame. This is remarkable because the probability of operating in a recessionary regime is practically the same (approximately 0.09) for both clusters. The main difference be- tween these clusters concerns the incidence of heterogeneity in the dominant regimes for each cluster. The dominant regime for cluster 1 countries is the mild bull regime 2, whereas that for cluster 2 countries is the bull market regime 3. That is, banking institutions in emerging market economies typically operate in a mild bull environment (0.711), but banking institutions in advanced economies typically oper- ate under a bullish environment (0.51). This distinction in the dominant regime for each cluster is properly ascertained through the HRSM model applications.

A potential explanation for the observed discrepancy in dominant regimes for each cluster may be attributed to the degree of exposure to the globally expansive twin bubbles in the real estate and credit de- rivative markets. This dual expansion specifically benefited the banking industries of advanced economies during the upswing of the economic cycle.9 This benefit is noticeable in the context of the financial perfor- mance of banks belonging to the most advanced economies in our sam- ple, as attested by the composition of cluster 2. Simultaneously, the same conclusion may be attained, given that cluster 2 banking indus- tries exhibit more integrated financial network structures than do their counterparts in cluster 1. For example, Allen and Carletti (2009) suggest that financial networks encompassing banking systems with more interconnected links typically shelter internationally diversified banks. This implies that banking industries in advanced economies are more interconnected, which may extend the duration of bull markets in these economies. This explanation for the discrepancy in the domi- nant regimes for each cluster warrants additional scrutiny in subse- quent research.

The transition probabilities between these three regimes for each of the clusters are also presented. Strong intra-cluster regime persistence continues to be observed during this period, with banking indexes be- longing to both clusters exhibiting very high probabilities of remaining in a given regime (with 0.931, 0.950 and 0.851 vs. 0.939, 0.962 and 0.981, respectively, for regimes 1, 2 and 3 in clusters 1 and 2). Regarding the mean sojourn time (which reflects the duration of the bear, mild bull and bull regimes, as measured in weeks), banking industries associ- ated with cluster 1 tend to take less time to emerge from any given re- gime than do their cluster 2 counterparts (14.39 vs. 16.42 weeks for regime 1, 19.92 vs. 26.11 weeks for regime 2, and 6.69 vs. 52.63 weeks for regime 3). The difference is greatest in the mean sojourn time asso- ciated with regime 3 (52.63–6.69 = 45.94 weeks), which suggests that cluster 2 banking industries tend to remain in the bull regime for more weeks (a multiple of 52.63/6.69 = 7.866 times as much) before transitioning to other regimes. This result may be explained by the prof- itability buoyancy exhibited by banks belonging to the countries in

9 The GFC was a systemic event associated with the bursting of the twin economic bub- bles in the United States real estate and the credit markets. Shiller (2008) identifies the GFC as the “deflating of a speculative bubble in the housing market that began in the United States in 2006 and has now cascaded across many other countries in the form of financial fail- ures and a global credit crunch” (p. 9).

cluster 2. The latter banking industries operated under credit and real estate asset bubble environments throughout the business cycle under scrutiny, as the cases of the United Kingdom and the United States clear- ly demonstrate.10

The synchronization of regimes across our sample set of country banking industries is also presented. The posterior probabilities de- scribed in Figs. 2 and 3 indicate a significant synchronized impact that is associated with the occurrence of the GFC across our sample. Fig. 2 de- picts the posterior probabilities of being in a given regime for cluster 1 countries. Until mid-2008, the banking indexes comprised therein were primarily alternating between regimes 2 and 3, with the former being the dominant regime of the two, notwithstanding country- specific idiosyncrasies. Therefore, during this time frame, intermediate bull and bull regimes appear to dominate over the bear regime. In addi- tion, Argentina, Brazil, and Russia experienced a crisis in 2001–2002. However, the financial impact associated with the occurrence of the present GFC appears to have been widely felt in 2008. The impact was transversally persistent and synchronized across the entire cluster. Ac- cordingly, the summer of 2008 appears to have witnessed the full onset of the impact of the GFC for the entire sample of country banking indexes. The corresponding bear regime duration varied across banking indexes, with Hungary being the worst-hit country and Argentina the least affected. The crisis subsided in 2009, although the rebound capac- ity is very distinct across the cluster. Banks in Hungary, for example, were overwhelmed by a further bear episode in 2010. Fig. 3 depicts the banking indexes of cluster 2. The results confirm that the GFC indeed constituted a systemic episode that had a persistent impact throughout the cluster's sample. For example, these attributes can be clearly discerned by the fact that banks belonging to both the United Kingdom and the United States, which were at the epicenter of the sys- temic episode under study, operated under a very bullish environment throughout the ‘subprime’ cycle. Once the systemic crisis took root in 2008, these institutions were subjected to a severe downturn that sub- sided in mid-2009. Overall, these institutions experienced a sustained asset price boom that was followed by a severe downturn. The main dif- ference between the figures is that the overall propensity to experience a bull regime for the banking industries included in cluster 2 is higher than that associated with the cluster 1 banking industries.11 Our find- ings also confirm that a high degree of financial interconnectedness is positively correlated with the development of the abovementioned twin asset price booms. Under the influence of the latter bubbles, the fi- nancial institutions that reaped the benefits of financially integrated structures were subsequently compromised by its implosion through severe financial contagion processes. As observed in both Figs. 2 and 3, the occurrence of the systemic event under study was truly global and highly synchronized.

Table 6 – Panel A shows the results for the synchronization of re- gimes. The results are aggregated by cluster for the sake of simplicity. In accordance with Dias and Ramos (2013), synchronization is mea- sured by the likelihood that the country set of banking industries shares the same regime and is quantified by their proposed logit-based corre- lation measure. This measure has the advantage of filtering out the ex- treme observations normally observed during crisis episodes. The measure is computed as

logititk ¼ log α̂itk

1 � α̂itk

� � ð5Þ

where α̂itk is the posterior probabilities of being in regime k in country i at time t.

10 A detailed analysis of the impact on international bank lending and borrowing for banks domiciled in the United Kingdom and the United States was presented by Batten and Szilagyi (2012). 11 The exception is Turkey, which exhibits high volatility across the entirety of our adopted timeline.

Fig. 2. Estimated posterior probability of the three regimes within cluster 1 (2002–2010).

383D.C. Bhimjee et al. / International Review of Financial Analysis 48 (2016) 376–387

The synchronization is quantified using the product–moment corre- lation between the logits for two time series.

The banking indexes of cluster 2 are synchronized with each other, among all regimes and in regimes 1 and 3 in particular. As expected, the synchronization of banks of cluster 1 is larger in regime 1, the bear regime, and is substantially smaller in regimes 2 and 3, which indicates that the paths are very different. The synchronization of regimes be- tween clusters 1 and 2 ranges between 0.53 in regime 1, the bear re- gime, and 0.09 in regime 2, the mild bull regime.12

4.2. Banking indexes in the global financial crisis in the 2007–2010 period

In this section, we present and discuss the findings for the period encompassing the GFC. Given space constraints, the results are set out in the Supplementary Appendix. The sample date starts on July 2007 and ends on August 2010. The start date reflects the month in which the first signs of financial distress occurred in the financial markets13

and after which certain major financial systemic failures occurred (e.g., Bear Stearns, Lehman Brothers). Furthermore, our results contem- plate the specific case of the Chinese banking index (the corresponding time series data were available for the 2007–2010 period) in addition to the banking indexes pertaining to the countries already encompassed by our 2002–2010 analysis.

In general, 2007 and 2008 were very critical for the performance of the global banking industry. Indeed, four major and resounding system- ic failures disrupted the industry, thereby aggravating the dynamics of the international financial contagion (IFC) processes. These four sys- temic examples, Bear Sterns, Lehman Brothers, Northern Rock and IKB, were all connected to the implosion of the twin real estate and

12 Country synchronization in the different regimes is presented in the Supplementary Appendix. The Supplementary Appendix is available online. 13 July 2007 witnessed a series of smaller defaults and loss warnings by US financial in- stitutions exposed to ‘subprime’ assets. As a premier US financial player, Bear Stearns pub- licly acknowledged on July 17, 2007 major losses (up to 90%) on two of its hedge funds specializing in ‘subprime’-related debt investments (Cox & Glapa, 2009).

credit market bubbles. These examples illustrate both the interconnec- tedness among the country banking indexes operating in globalized fi- nancial markets and the effects associated with international financial contagion processes.14

The optimal choice of parameter values for clusters (S) and regimes (K) indicates that the value of S is equal to one and that the value of K is equal to four. That is, the optimal result yields a sole undifferentiated and non-heterogeneous cluster that contains all banking indexes, oper- ating under the framework of four distinct regimes.

In the Supplementary Appendix, we present the table with the four regimes. Regime 1 is associated with a strong bearish framework (se- vere negative returns of −4.671 coupled with a very high volatility of 377.092); regime 2 is associated with a mild bearish environment (mild negative returns of −0.982 with a low volatility of 17.124); re- gime 3 is associated with a subdued bearish framework (low negative returns of −0.068 associated with a medium volatility of 55.227); and regime 4 is associated with a strong bull environment (high returns of 0.936 coupled with a very low volatility of 10.46). Conversely, P(Z) is the average probability that banking indexes are in a specific regime; it is very high for intermediate regime 3 (0.348) and 2 (0.305), followed by the bullish regime 4 (0.284). The average probability of operating under the contractionary regime 1 is 0.063.

For a comparison with the results of the previous section, we sepa- rate banking indexes into the two previous clusters and analyze their re- gime synchronization. Panel B of Table 6 presents the synchronization during the crisis period. We find a larger synchronization among the countries of cluster 2 than among the countries of cluster 1 because the latter group is composed primarily of emerging markets that had previously exhibited segmentation. The synchronization of regimes among the countries of cluster 2 is similar to that observed for the whole sample period. Overall, a large synchronization both between

14 The strength of IFC is clearly observable in the work of Cho, Hyde, and Nguyen (2015), who use a large sample of approximately 31,000 firms across 31 markets to document the GFC's global reach compared with other, less powerful financial episodes.

Fig. 3. Estimated posterior probability of the three regimes within cluster 2 (2002–2010).

384 D.C. Bhimjee et al. / International Review of Financial Analysis 48 (2016) 376–387

and within clusters is noticeable in all regimes. The absence of distinct synchronization is in accordance with the presence of a sole cluster of countries.

To obtain a better understanding of the dynamics of banking indus- tries during the crisis, Figs. 4 and 5 display the regime dynamics, with certain marked subperiods based on the official timelines provided by the Federal Reserve Board of St. Louis (2009) and the Bank for Interna- tional Settlements (Filardo et al., 2010). These studies separate the time- line of the GFC into four phases. Phase 1 spans from August 1, 2007 to

Table 6 Synchronization of regimes. This table presents the association between banking indexes based on the posterior probability average that excludes a country's synchronization with itself.

Panel A: 2002–2010

Regime 1 Regime

Cluster 1 Cluster 2 Cluster

Cluster 1 0.52 0.23 Cluster 2 0.53 0.70 0.09

Panel B: 2007–2010

Regime 1 Regime 2

Cluster 1 Cluster 2 Cluster 1 Cluster

Cluster 1 0.65 0.58 Cluster 2 0.67 0.71 0.60 0.64

September 15, 2008, and is described as “initial financial turmoil”. Phase 2 is defined as “sharp financial market deterioration” (September 16, 2008–December 31, 2008), phase 3 is described as “macroeconomic deterioration” (January 1, 2009 until March 31, 2009) and phase 4 is a phase of “stabilization and tentative signs of recovery” (post-crisis peri- od, until the end of the sample period). In the figures, we again separate the clusters of banking indexes. We confirm that the regime dynamics have many resemblances. During phases 2 and 3, i.e., from September 16, 2008 to March 31, 2009, most banking industries are in a high

of being in the same regime (see Eq. 5). Average synchronization is an equally weighted

2 Regime 3

1 Cluster 2 Cluster 1 Cluster 2

0.30 0.43 0.36 0.65

Regime 3 Regime 4

2 Cluster 1 Cluster 2 Cluster 1 Cluster 2

0.29 0.56 0.33 0.40 0.58 0.63

Fig. 4. Estimated posterior probability of the three regimes within cluster 1 (2007–2010). The lines mark the following dates: September 15, 2008, December 31, 2008, and March 31, 2009, which correspond to different phases of the global financial crisis. The phases are initial financial turmoil – August 1, 2007–September 15, 2008; sharp financial market deterioration – September 16, 2008–December 31, 2008; macroeconomic deterioration – January 1, 2009–March 31, 2009; and stabilization and tentative signs of recovery – April 1, 2009 to the end of the sample period. Source: Federal Reserve Board of St. Louis (2009) and the Bank for International Settlements (Filardo et al., 2010).

385D.C. Bhimjee et al. / International Review of Financial Analysis 48 (2016) 376–387

volatility regime with negative means. In phase 4, despite signs of re- covery in the majority of banks, in some cases, e.g., Hungary, Belgium, Greece, France, and Spain, episodes of slumping back into the crisis re- gime remain; the extreme case is Ireland, which never leaves the crisis regime. In contrast, the banking industry indexes in Thailand, Taiwan, Japan, Malaysia, and Singapore appear to be the least affected and re- cover at a much faster pace.

5. Concluding remarks

The GFC of 2007–2009 was a systemic breakdown of unprecedented proportions affecting financial markets and institutions, particularly among banking institutions worldwide. The application of a novel panel regime-switching methodology, the HRSM, has unearthed a framework of heterogeneous banking industry performance that is cap- tured within a global macro-financial setting. Our findings are summa- rized in the following paragraphs.

First, heterogeneous global banking industry performance is appro- priately captured by the HRSM. Each of the clusters operated under dis- tinctive regime dynamics. These mutually exclusive regimes are clearly identifiable with bull, intermediate and bear financial regime dynamics, thus adding a deeper regime granularity to our findings.

Second, the inter- and intra-cluster synchronization patterns of the performance of the banking industries comprising our sample indi- cate the severity of the underlying international financial contagion pro- cesses that are at work in a post-crisis environment. The results further reveal that the onset of the GFC may be associated with a loss of hetero- geneity due to the impact of a transversal common shock (the GFC).

The heterogeneity in our findings concurs with the results of Ehrmann, Fratzscher, and Mehl (2009) and Shehzad and De Hann (2013). The former confirms the existence of a set of heterogeneous eq- uity market responses to the GFC by focusing on both banking and non- banking segments rather than on macro-aggregates; the latter finds that stock prices of banks in emerging countries were less affected by the systemic shock than the corresponding prices of their counterparts

in developed economies by focusing on individual banks rather than macro-aggregates.

According to Beltratti and Stulz (2012), a possible explanation for the differing performance of stock returns of the large banking indexes comprising our sample may reside in a combination of factors. The latter may involve the role of regulation, the quality of banking governance, and the specificities of the balance sheets of important banking institu- tions. Regime changes after the GFC appear to be in accordance with the over-hauling of the expectations associated with bank stock returns be- fore and after the crisis. Before the crisis, stock markets favored banking business strategies involving innovative financial related products. The onset of the crisis may have then shifted market expectations in favor of more conservative banking business strategies promoting staple products (Beltratti & Stulz, 2012).

Furthermore, the existence of large-scale banking operations involv- ing securitization lines may have strained the transmission channels to the real economy (for example, by constraining the availability of credit, once liquidity pressures set in). At a global macroeconomic level, het- erogeneous performance within the banking industry may have caused advanced economies overtly dependent on sophisticated credit chan- nels (i.e., of securitized extraction) to succumb more perniciously to the effects of the GFC than emerging market economies did, and this heterogeneity has been innovatively depicted in our model applications.

These findings are fundamental to understand the significant role played by international financial contagion processes in the aftermath of a powerful systemic shock such as the GFC, as the paper exposes the high degree of cross-country transversality and the simultaneity of global banking contagion across the sample. These findings contribute to the proper understanding of how international financial contagion processes work and the corresponding implications thereof to the for- mulation of macro- and micro-prudential policies. Directions for future study should also address the following topics: first, the link between international financial contagion and trade interconnectedness and the corresponding impact on real economies throughout systemic

Fig. 5. Estimated posterior probability of the three regimes within cluster 2 (2007–2010). The lines mark the following dates: September 15, 2008, December 31, 2008, and March 31, 2009, which correspond to different phases of the global financial crisis. The phases are initial financial turmoil – August 1, 2007–September 15, 2008; sharp financial market deterioration – September 16, 2008–December 31, 2008; macroeconomic deterioration – January 1, 2009–March 31, 2009; and stabilization and tentative signs of recovery – April 1, 2009 to the end of the sample period. Source: Federal Reserve Board of St. Louis (2009) and the Bank for International Settlements (Filardo et al., 2010).

386 D.C. Bhimjee et al. / International Review of Financial Analysis 48 (2016) 376–387

crises, and second, the use of more inclusive data (e.g., aggregate data indexes comprising the shadow banking industries of the countries in- cluded in the sample) that is currently unavailable.

Appendix A. Supplementary Appendix

Supplementary data to this article can be found online at http://dx. doi.org/10.1016/j.irfa.2016.01.005.

References

Ahmad, W., Bhanumurthy, N., & Sehgal, S. (2015). Regime dependent dynamics and European stock markets: Is asset allocation really possible? Empirica, 42(1), 77–107.

Allen, F., & Carletti, E. (2009). The Roles of Banking Institutions in Financial Systems. In A. Berger, P. Molyneux, & J. Wilson (Eds.), The Oxford Book of Banking (pp. 37–57). Ox- ford: Oxford University Press (chapter 2).

Baele, L. (2005). Volatility spillover effects in European equity markets. Journal of Financial and Quantitative Analysis, 40(2), 373–401.

Batten, J.A., & Szilagyi, P.G. (2012). International banking during the global financial crisis: U.K. and U.S. perspectives. International Review of Financial Analysis, 25, 136–141.

Beltratti, A., & Stulz, R.M. (2012). Why did some banking institutions perform better dur- ing the credit crisis? Journal of Financial Economics, 105(1), 1–17.

Bhar, R., & Hamori, S. (2004). Hidden Markov Models – Applications to Financial Econom- ics. Advanced Studies in Theoretical and Applied Econometrics (Volume 40). Dordrecht: Kluwer Academic Publishers.

Billio, M., & Pelizzon, L. (2003). Volatility and shocks spillover before and after EMU in European stock markets. Journal of Multinational Financial Management, 13, 323–340.

Chen, P. -F., & Liu, P. -C. (2013). Bank ownership, performance, and the politics: Evidence from Taiwan. Economic Modelling, 31, 578–585.

Cho, S., Hyde, S., & Nguyen, N. (2015). Time-Varying Regional and Global Integration and Contagion: Evidence from Style Portfolios. International Review of Financial Analysis, 42, 109–131.

Choudhry, T., & Jayasekera, R. (2014). Returns and volatility spillover in the European banking industry during global financial crisis: Flight to perceived quality or conta- gion? International Review of Financial Analysis, 36, 36–45.

Cox, J., & Glapa, L. (2009). Credit Crisis Timeline. Working Paper. The University of Iowa Center for International Finance and Development.

Dias, J.G., & Ramos, S.B. (2014). The aftermath of the subprime crisis: A clustering analysis of world banking sector. Review of Quantitative Finance and Accounting, 42(2), 293–308.

387D.C. Bhimjee et al. / International Review of Financial Analysis 48 (2016) 376–387

Dias, J.G., & Ramos, S.B. (2013). The dynamics of stock markets cycles in the euro zone. Economic Modelling, 35, 320–329.

Dias, J.G., & Wedel, M. (2004). An empirical comparison of EM, SEM and MCMC perfor- mance for problematic Gaussian mixture likelihoods. Statistics and Computing, 14(4), 323–332.

Dias, J.G., Vermunt, J.K., & Ramos, S.B. (2008). Heterogeneous Hidden Markov Models. In P. Brito (Ed.), COMPSTAT2008. Proceedings in Computational Statistics (pp. 373–380). Heidelberg: Physica/Springer Verlag.

Dimitriou, D., Kenourgios, D., & Simos, T. (2013). Global financial crisis and emerging stock market contagion: A multivariate FIAPARCH–DCC approach. International Review of Financial Analysis, 30, 46–56.

Ehrmann, M., Fratzscher, M., & Mehl, A. (2009). What Has Made the Financial Crisis Truly Global?, Working Paper. May: European Central Bank.

Federal Reserve Board of St. Louis (2009). The Financial Crisis: A Timeline of Events and Pol- icy Actions.

Filardo, A., George, J., Loretan, M., Ma, G., Munro, A., Shim, I., & Zhu, H. (2010). The inter- national financial crisis: Timeline, impact and policy responses in Asia and the Pacific. BIS Papers, 52, 21–82.

Guidolin, M., & Timmermann, A. (2007). Asset allocation under multivariate regime switching. Journal of Economic Dynamics & Control, 11(31), 3503–3544.

Hamilton, J.D. (1989). A new approach to the economic analysis of nonstationary time se- ries and the business cycle. Econometrica, 57(2), 357–384.

Harvey, C.R. (1995). Predictable risk and returns in emerging markets. Review of Financial Studies, 8, 773–816.

Kearney, C., & Potì, V. (2008). Have European stocks become more volatile? An empirical investigation of idiosyncratic and market risk in the euro area. European Financial Management, 14(3), 419–444.

Kotkatvuori-Örnberg, J., Nikkinen, J., & Äijö, J. (2013). Stock market correlations during the financial crisis of 2008–2009: Evidence from 50 equity markets. International Review of Financial Analysis, 28, 70–78.

McLachlan, G., & Peel, D. (2000). Finite Mixture Models. New York: John Wiley & Sons. National Bureau of Economic Research (2010). Announcement of June 2009 Business Cycle

Trough/End of Last Recession, Dated September, the 20th, 2010. Pozsar, Z., Adrian, T., Ashcraft, A., & Boesky, H. (2012). Shadow Banking, Federal Reserve

Bank of New York Staff Report N° 458. Federal Reserve Bank of New York. Ramos, S.B., Vermunt, J.K., & Dias, J.G. (2011). When markets fall down: Are emerging

markets all the same? International Journal of Finance and Economics, 16, 324–338. Shehzad, C.T., & De Hann, J. (2013). Was the 2007 crisis really a global banking crisis?

North American Journal of Economics and Finance, 24, 113–124. Shiller, R. (2008). The Subprime Solution: How Today's Global Financial Crisis Happened, and

What to Do About It. Princeton: Princeton University Press. Susmel, R. (2001). Extreme observations and diversification in Latin American emerging

equity markets. Journal of International Money and Finance, 20(7), 971–986.