1
The Outperformance of G-SIBs during the Chinese
Stock Market Crash: An Event Study Analysis
Ramon Vendelbos
Thesis Supervisor: Dr F López-de-Silanes
University of Amsterdam
Master Business Economics: Finance Master Thesis, July 2016
2
Statement of Originality
This document is written by Student Ramon Vendelbos who declares to take full responsibility for the contents of this document.
I declare that the text and the work presented in this document is original and that no sources other than those mentioned in the text and its references have been used in creating it.
The Faculty of Economics and Business is responsible solely for the supervision of completion of the work, not for the contents.
3
Abstract
This paper analyses the financial markets reaction to the new regulations for global systematically important banks (G-SIBs) during times of financial turmoil. The stock market’s reaction to the regulatory changes designed to address the too-big-to-fail problem for G-SIBs are examined during the largest stock market crashes of the Chinese stock market crash. Bank capital theories, risk management theories and theories on the impact of regulatory changes form hypotheses and suggest a positive effect of the new regulations on stock price returns during times of financial distress. This study applies an event study methodology and cross-sectional analyses to a sample of the world’s 75 largest banks for which we examine whether the stock prices of G-SIBs reacted significantly and differently from those of their peers during times of financial turmoil. Both the event study and the cross-sectional analyses find evidence for outperformance of G-SIBs during the Chinese stock market crash. Further, we find that financial markets react differently to bank characteristics for G-SIBs than to those of their peers. Bank characteristics such as size, performance, capital adequacy and complexity are found to be important in explaining G-SIBs their cumulative abnormal returns during times of financial distress
4
Table of content
1. Introduction ... 6
2. Literature review... 8
2.1 An introduction of G-SIBs ... 8
2.4 Risk Management theories ... 13
2.5 The impact of regulatory changes ... 13
2.6 Hypotheses ... 14
3. Data and Methodology ... 15
3.1 Data ... 15
3.2 Event study methodology ... 16
3.2.1 Event of interest ... 16
3.2.2 Event window and estimation period ... 16
3.2.3 Selection criteria ... 17
3.2.4 Calculating the Abnormal Return ... 18
3.2.5 Calculating the Cumulative Abnormal Return (CAR) ... 19
3.2.6 Calculating the average and the standard deviation of the CARs ... 20
3.2.7 The Mackinlay (1997) multiday t-test statistic ... 21
3.2.8 The Cowan (1992) multi-day sign z-statistic ... 21
3.3 Cross-sectional analyses methodology ... 22
3.4 Descriptive statistics ... 23
4. Results ... 25
4.1 Event study results ... 25
4.2. Cross-sectional analysis results ... 27
4.2.1 The outperformance of G-SIBs ... 28
4.2.2 The stock market reactions to bank characteristics for G-SIBs and their peers ... 30
5. Conclusion and Discussion ... 34
6. References ... 37
7. Appendices ... 40
A. Variable construction ... 40
B. Descriptive statistics of subsamples ... 41
C. Robustness: Event study results ... 43
D. Robustness: cross-sectional analysis ... 46
5
D.2 Robustness: the stock market reactions to bank characteristics for G-SIBs and their
peers ... 48
E. Robustness of the multiple linear regression model ... 50
E.1 Checking for normality of the residuals ... 50
E.2 Checking for multicollinearity ... 51
6
1. Introduction
In November 2011 the Financial Stability Board (FSB) published a policy consisting of measures to address the systematic and moral hazard risks of systematically important financial institutions (SIFI’s). Besides, the identification of SIFI’s, the publication of the Financial Stability board also identified a list of global systematically important banks (G-SIBs), based on a methodology developed by the Basel Committee on Banking Supervision. The G-SIBs are more closely monitored, have to comply to G-SIB regulatory rules and are required to hold higher levels of bank capitalization. Those measures should solve moral hazards and too-big-too-fail problems of the world’s most important banks. Moreover, the measures should increase the stability of the G-SIBs during future financial turmoil and prevent them from future bankruptcy or government bail-outs (Financial Stability Board, 2015). To test if these regulatory measures have the intended effect, such as FSB regulators hope for. We examine if G-SIBs outperformed their peers during the Chinese stock market crash. The Chinese stock market crash is interesting to investigate, since it is the first large stock market crash after the designation of G-SIBs causing international financial turmoil. Furthermore, we compare these findings with results of the recent financial crisis, to test the effect of the regulatory requirements and the designation as a G-SIB.
Since large banks are only designated as G-SIBs upwards of 2011 and regulators are still engaged in designing and implanting G-SIB regulation, little prior research has been performed on this topic. Moreover, the lack of prior research on this specific topic can be explained by the novelty of the topic, since the Chinese stock market crash occurred just a few months ago and this crash is the first one to bring the G-SIB regulation and their impact into practice. Nevertheless, Moenninghoff et al. (2013) examined the designation effect of G-SIBs and find a positive effect on stock returns. Further, Bongini, Nieri and Pellagatti (2015) investigated the direct impact of this new G-SIB regulation and find that market did not react univocally, although they did discriminate on the level of capital and estimated the probable effects correctly. In contrast, Abreu and Gulamhussen (2013) find no effect of the G-SIB regulation on stock price performance. These studies only studied the direct effect of the regulatory changes for G-SIBs and do not provide any evidence for the effect of these regulatory changes during times of financial turmoil and their impact on stock market investors. Therefore, the purpose of this study is to close this gap by examining the effects of these regulatory requirements during times they are implanted for, namely times of financial distress.
7
impacted bank performance during the Chinese stock market crash and contributes to the existing literature in several relevant ways. First of all, together with Abreu and Gulamhussen (2013) and Bongini, Nieri and Pellagatti (2015) this one of the few studies that tries to quantify the positive effects related to the G-SIB designation. Secondly, unlike previous studies, this study focuses on the effects of G-SIB regulation during times of financial turmoil. Thirdly, this study contributes to the on-going debate on the relationship between bank capitalization and stock price performance.
Furthermore, this study contributes to the open-ended debate concerning systematic risk by investigating the stock market reaction to the regulatory requirements of G-SIBs during times of financial turmoil. Besides, it shows the effectiveness of G-SIB regulation imposed by the Financial Stability Board in creating stability of large banks during times of financial distress, and could help in preventing future bankruptcies or bailouts. Moreover, since too-big-to-fail and moral hazard problems of large banks were one of the major causes of the recent global financial meltdown, the results of this study might help in preventing future financial crises and for the implantation of future regulation on large banks. In addition, our results could also impact investor’s decisions, corporate governance of banks, regular bank regulation and many more related decisions.
To scrutinize the impact of G-SIB regulation during the Chinese stock market crash. We formed hypothesis based on bank capitalization, risk management and regulatory theories hypotheses. Thereafter, we conduct an even study to examine if G-SIBs experienced higher abnormal returns than their peers during the Chinese stock market crash. We use similar to previous research a sample of the world’s 75 largest banks and subsamples of G-SIBs, D-SIBs and a peer group (Abreu and Gulamhussen, 2013; and Bongini, Nieri and Pellagatti, 2015). Since the Chinese stock market crash exists of multiple stock market crashes, we established a rule of only investigating one-week time periods in which the Chinese stock market dropped over 10%. This resulted in the examination of two large aftershocks of the Chinese stock market crash and we also compare those events to the largest one-week stock market crash of the recent financial crisis for robustness. Finally, we regress the cumulative abnormal returns (CARs) on bank characteristics to examine the relationships more rigorously.
Both the event study and the cross-sectional analyses find strong evidence for outperformance of G-SIBs during the second large aftershock of the Chinese stock market crash, whereas only the event study methodology finds evidence for outperformance of G-SIBs during the first aftershock. Furthermore, we find that financial markets react differently to bank characteristics for G-SIBs than to those of their peers. Bank characteristics such as size,
8
performance, capital adequacy and complexity are found to be important in explaining the CARs of G-SIBs during times of financial distress. Whereas, size, performance and leverage are found to be positively related to the CARs of other large bank during times of financial turmoil. Finally, we found that banks with more exposure to China performed worse during the Chinese stock market crash. All our results holds to several robustness tests as provided in sections C and D of the Appendices.
The paper is organized as follows. Section 2 will provide a literature review on G-SIBs, interconnectedness of markets and bank shares, bank capital theories, risk management theories and the impact of regulatory changes. Thereafter, we derive hypotheses from those theories. In section 3 the used data and the employed methodology for the analysis will be defined. Next, section 4 will provide both the event study and cross-sectional analysis results. Finally, section 5 wraps up with a conclusion and discussion.
2. Literature review
In this section we will introduce literature corresponding to our paper and form hypotheses based on the literature background. The first part of the literature review is a paragraph that gives an introduction to the existing literature on G-SIBs. Thereafter, the second paragraph explains the relationship between the Chinese stock market crash and the impact on large banks. Subsequently, the third and fourth paragraph contain theories based on the capitalization of banks and risk management, respectively. Next, paragraph five will provide an overview of the literature on the impact of regulatory changes. Finally, the last paragraph provides an overview of the theory and forms hypotheses based on the discussed theories.
2.1 An introduction of G-SIBs
To solve the too-big-to-fail problem for large banks the Financial Stability Board developed in 2011 a method to identify global systematically important banks (G-SIBs). Using this method the Financial Stability Board and the Basel Committee on Banking Supervision publish a yearly updated list of G-SIBs. The banks on this list are more closely monitored, need to submit an updated Emergency Resolution plan every year and are required to hold a minimum total capital adequacy ratio from March, 2018 according to Basel III. In addition, G-SIBs have also been allocated to buckets corresponding to the required level of additional loss absorbency and are
9
required to regularly perform resolvability assessments (Financial Stability Board, 2015). Besides the designation of G-SIBs, some countries also designated domestic systematically import banks (D-SIBs) to address the too-big-to-fail problem for banks that have an important impact on the domestic financial system and economy. The D-SIBs have to apply to regulatory frameworks as imposed by the country for which the bank is systematically important (Bank for International Settlements, 2012).
Several studies investigated the financial markets reaction to the new regulatory changes for SIFI’s and G-SIBs. First of all, Bongini, Nieri and Pellagatti (2015) find that financial markets react differently to these new regulations. However, the financial markets did discriminate on the level of capital and estimated the probable effects of the new requirements for SIFIs correctly. Abreu and Gulamhussen (2013) did a similar study regarding the market’s reaction on the Financial Stability Boards’ announcement of G-SIBs. They find no abnormal returns and argue that these reforms had little effect on stock market investors’ believes. However, in contrast to Bongini, Nieri and Pellagatti (2015) they used a smaller sample and currently more information is known with respect to the identification process of G-SIBs, the new capital requirements of Basel III and the updated lists of G-SIBS. Moreover, Moenninghoff et al. (2015) also found that the designations of banks as G-SIB had a positive effect on their stock returns.
2.2 Interconnectedness of the Chinese stock market and bank shares
The interconnectedness of financial markets has always been very important and interesting for financial economists as well as investors (Bartram and Duffey, 2001). The opening-up of China and the on-going economic reforms have stimulated the development of the Chinese financial markets (Chang et al., 2007).Nowadays, the Chinese stock market has become one of the most important stock markets in the world and has integrated with other major foreign stock markets (Laurence et al., 1997).
Experiences of international stock market crashes before the Chinese stock market crash have already shown that a catastrophic event in China could easily harm other stock markets around the world (Wang et al., 2011). Moreover, according to Guo and Huang (2010) the speculative capital inflow in China has affected their real estate and stock markets. Besides driving up property prices the huge capital inflows also increased volatilities in the China’s financial markets, ending up in the Chinese stock market crash which affected stock exchanges
10
and shares all around the world. Choi and Wang (2015) find for example co-movement between the Chinese and U.S. aggregate stock returns. This relationship can also be observed by looking at the returns of the largest stock indices of those two countries as presented in Figure 1. One can observe from Figure 1 that the aftershocks of the Chinese stock market crash directly influenced the S&P 500 returns and caused the S&P 500 to drop over 10%.
Figure 1.
Banks have a traditional and dominant place in the Chinese financial system and they are the most important source of funds for the Chinese stock market. Besides, banks are found to be very important for the efficiency of the Chinese financial system (Groenewold et al., 2003). In addition, Luo et al. (2010) find a causal relationship between Chinese banks and the Shanghai Stock Exchange (SSE) Composite index. They also find that the Chinese stock markets integrated with other stock markets, which was in part driven by the Chinese banks. According to Irresberger et al. (2015) stock performance of international banks is driven by investors’ irrational market-wide crisis sentiment, which can lead to a devaluation of bank stocks irrespective of any changes in idiosyncratic or macroeconomic fundamentals. This effect is found to be the strongest in the absence of bailout guarantees. Furthermore, De Long et al. (1990) find that share prices are affected by uninformed noise traders who base their expectations on irrational investors’ sentiment rather than rational information. The rational and irrational effects of the Chinese stock markets on bank’s their share prices are clarified by Figures 2 and 3. One can observe that the Shanghai Stock Exchange SE Composite and the
1800 1850 1900 1950 2000 2050 2100 2150 0 1000 2000 3000 4000 5000 6000
mrt-14 jun-14 sep-14 dec-14 apr-15 jul-15 okt-15 jan-16 mei-16 aug-16
S&P500 IN U.S . DO LL A R S ($) SS E S E COMPOS IT E IN CHIN ES E Y U A N ( ¥)
S&P 500 and SSE SE Compositie Returns
Shanghai SE Composite S&P 500 1stAftershock (17-08-2015) Start of Chinese stock market crash (12-06-2015)
2nd Aftershock
11
bank indexes behave very similar after the first aftershock on 17 August 2015. Moreover, the graphs even seem to be cointegrated after the first shock, exemplifying the effect of the Chinese stock market crash on banks.
Figure 2. Figure 3. 0 20 40 60 80 100 120 0 1000 2000 3000 4000 5000 6000
mrt-14 jun-14 sep-14 dec-14 apr-15 jul-15 okt-15 jan-16 mei-16 aug-16
M SC I W O R LD B A N KS IN U.S D O LL A R S ( $) SS E SE C O M P O SIT E IN C H INES E YUAN (¥)
SSE SE Composite and MSCI World Bank Index Returns
SSE SE Composite MSCI World Banks Index
0 500 1000 1500 2000 2500 3000 3500 0 1000 2000 3000 4000 5000 6000
mrt-14 jun-14 sep-14 dec-14 apr-15 jul-15 okt-15 jan-16 mei-16 aug-16
N A SD A Q B A N KS IN U.S D O LL A R S ( $) SS E SE C O M P O SIT E IN C H INES E YUAN (¥)
SSE SE Composite and Nasdaq Bank Returns
SSE SE Composite Nasdaq Banks 2nd Aftershock (17-08-2015) 1st Aftershock (30-12-2015) 1st Aftershock (17-08-2015) 2nd Aftershock (30-12-2015)
12
2.3 Bank capitalization during crises
Demirguc-Kunt et al. (2013) examined the relation between a better capitalization of banks and their stock price performance during the financial crisis. They find that a stronger capitalization of banks resulted in a better stock market performance, especially for larger banks. Furthermore, Beltratti and Stulz (2012) find a positive relation between Tier 1 capital and bank performance for larger banks during the recent financial crisis. However, according to Berger and Bouwman (2009) banks with a higher degree of capitalization performed better during the crisis in the early 1990s, but not during the recent financial crisis. Moreover, in contrast to Demirguc-Kunt et al. (2013), Akhigbe et al. (2012) found a negative relation between bank capital and stock returns. These contradicting findings could be explained by a U-shaped relationship between bank capital and risk-taking (Calem and Rob, 1999). The former papers of Demirguc-Kunt et al. (2013), Beltratti and Stulz (2012) and Berger and Bouwman (2009) show that a higher degree of capitalization can result in better stock performance and suggest a positive reaction of the markets to the capital requirements for G-SIBs. However, since Akhigbe et al. (2012) had contrary findings and Calem and Rob (1999) find a U-shaped bank capital and risk-taking relation, this study will also examine if the additional capital requirements have the intended effect as stated by Demirguc-Kunt et al. (2013). Besides investigating the effect of the regulatory reforms, this study will thus also investigate the relationship between bank capital and share price performance and support one of the two theories as stated above.
In addition, bank size and capitalization are essential factors when considering bank regulations, especially in times of turmoil in the financial markets. On the one hand, larger banks benefit more from competition. On the other hand, a higher capital ratio is also beneficial for banks that operate in collusive markets and bank capital is of extreme importance in creating stability for large banks under competition (Cajueiro et al., 2012). Taking into account this literature, one would expect large banks to hold higher levels of capital. However, according to Demsetz and Strahan (1997) larger bank holding companies tend to have lower capital ratios, which results in that they are not less risky despite of their lower idiosyncratic risk gained by diversification advantages.
13
2.4 Risk Management theories
Risk management is a very important aspect of bank performance. Consistent with the risk culture hypothesis, Fahlenbrach et al. (2012) find that banks who performed poorly during the 1998 crisis also performed worse during the recent financial crisis. According to Fahlenbrach et al. some aspect of bank’s their business model and risk culture makes them more sensitive to crisis. Illustrating the fact that short-term funding, the amount of leverage and growth are important aspects of poor bank performance during both crisis, they show that the risk management of bank is an important characteristic of a bank’s their share price performance, especially during financial turmoil.
In addition, Ellul and Yerramilli (2013) developed a risk measurement index that measures the strength and independence of bank holding companies. They find that a higher risk measurement index is related to better operating and stock performance during the financial crisis, lower tail risk and a lower level of nonperforming loans .This implies that risk management and higher returns are positively related, which is also found by Aebi et al. (2012). Taking into account the findings Fahlenbrach et al. (2012) and Aebi et al. (2012), one would expect a positive stock market reaction to the imposed regulatory requirements for G-SIBs.
2.5 The impact of regulatory changes
A report of the Organization for Economic Co-operation and Development (OECD) argues that the recent financial crisis is mainly caused by failures and weaknesses in corporate governance arrangements (Kirkpatrick, 2008). Moreover, the National Commission on the Causes of the Financial and Economic Crisis in the United States found the same cause for the financial crisis and conclude that the failure of corporate governance at SIFI’s were a major cause of the global financial meltdown (The Financial Crisis Inquirey Report, 2011). The role of corporate governance in the performance of banks has also be emphasized in academic studies such (Diamond and Rajan, 2009) and (Bebchuk and Spamann, 2010). The general findings are that banks with poor corporate governance took excessive risk, resulting in large losses and even banking failures during the crisis. Since banks were able to take excessive risk, lax regulation can be seen as an indirect cause of the crisis. Therefore, in the aftermath of the crisis there has been an enhanced focus on regulatory changes and their impact during crises in the academic literature (see, for instance, Dooley, Folkerts-Landua and Garber (2009); Stiglitz (2010); Beltratti and Stulz (2012)).
14
Schäfer et al. (2013) examined the market’s reaction of stock prices and credit defaults swaps to regulatory reforms between 2009 and 2011. They find that these regulatory reforms successfully lowered the bailout expectations, especially for SIFIs. Since regulatory reforms are found to have a strong on stock prices of SIFIs, their study suggests that stock markets take those regulatory changes into account and that these effects are even larger for systematically important financial institutions. Moreover, Beltratti and Stulz (2012) find that large banks which are more regulated performed better during the recent financial crisis.
2.6 Hypotheses
In lines with Beltratti and Stulz (2012); Demirguc-Kunt et al. (2013); Ellul and Yerramilli (2013); and Schäfer et al. (2013) we expect an outperformance of G-SIBs during the Chinese stock market crash. According to these studies, the higher required level of bank capital and the other imposed regulatory reforms should have a positive effect on the performance of the G-SIBs during times of financial turmoil, based on this literature we derive the following hypothesis:
H1: G-SIBs had an abnormal return relative to their non G-SIB peers during the Chinese
stock market crash. (H0 : abnormal return > 0)
In addition, the academic literature provides an on-going debate on the relationship between bank capitalization and stock price performance. On the one hand, Demirguc-Kunt et al. (2013) find that better capitalized banks their stock prices performed during crises. On the other hand Akhigbe et al. (2012) find a negative relations between bank capital and stock price returns. Furthermore, Berger and Bouwman (2009) also find a positive relationship in the early 1990s crisis, but not in the recent financial crisis. Whereas, Beltratti and Stulz (2012) find a positive relation between Tier 1 capital and bank performance. Therefore, we formed a second and third hypothesis to examine which theory of bank capitalizations during crises can be supported.
H2: Banks with more high quality capital (Tier 1) had an abnormal return during the Chinese
stock market crash.
H3: G-SIBs with more high quality capital (Tier 1) had an abnormal return during the
15
3. Data and Methodology
This section starts with an overview of the data and the construction of the samples in the first paragraph. Thereafter, the second paragraph describes the employed event study methodology and the third paragraphs explains the cross-sectional analysis that we run to explain the impact of bank characteristics on the CARs. Finally, this section concludes with the descriptive statistics and the construction of the variables.
3.1 Data
The dataset consists of the world’s largest banks. To retrieve the data, we run the following search on the Compustat IQ Bank database: include all banks whose stocks are listed and whose assets exceed $20 billion for the year 2015. However, some G-SIBs were not included in this sample and therefore these G-SIBs were manually added to the sample. As a result, we retrieved similar to Bongini, Nieri and Pellagatti (2015) a sample of the world’s 75 largest banks listed over stock exchanges all around the world. We divided this sample in three subsamples of G-SIBs, D-SIBs and a peer group to perform analyses within the designations of the largest banks. The subsample of G-SIBs consists of 29 G-SIBs and the subsamples of D-SIBs and the peer group consist of 30 and 16 observations, respectively.
The corresponding stock price data has been retrieved from Datastream using the equity price (P, adjusted-default) and to collect price indexes of the market portfolios of the corresponding stock markets the datatype price index (PI) is used from the Datastream database. Thereafter, we computed the expected returns using the market model and the abnormal returns by subtracting the expected returns from the actual returns. Finally, we accumulated the abnormal returns to get the cumulative abnormal returns.
Furthermore, data for the cross-sectional analysis which will follow after the event study has been retrieved using the Wharton Research Data Services (WRDS). Using WRDS, the accountancy data has been retrieved from the Compustat IQ Bank database. Additionally, we hand-collected any missing data using financial reports and the data for the Exposure to China variable using the consolidated statistics on the website of the Bank for International Settlements. The variable construction is described in words in section 3.4 and in formulas by Table 5 in section A of the appendices.
16
3.2 Event study methodology
To test the outperformance of G-SIBs with respect to other banks during the Chinese stock market crash, an event study will be performed. An event study is a commonly used method to measure the effects of a specific economic event on the stock price of a firm. The event study will thus measure the effect of the Chinese stock market crash on the value of the firm. The event study will be performed using the methodology of Mackinlay (1997), and the following steps will describe in detail how to perform an event study using his methodology.
An event study starts with defining the event of interest and determining the event window and estimation period. The event window is the period in which the effects of the event on the share price are examined and the estimation period is the period in which the parameters of the model are estimated. Thereafter, an event study determines the selection criteria to define the included firms and events. The third step of an event study is to calculate the abnormal return, the abnormal return is the actual ex-post return of a share minus the expected return of the share for a given day. Subsequently, the cumulative abnormal returns can be computed by accumulating the abnormal returns over the event window. Finally, we can test the statistical significance of the cumulative abnormal returns by using a t-test.
3.2.1 Event of interest
The event of interest is the Chinese stock markets crash. However, the Chinese stock market crash exist of multiple crashes of the Chinese stock market over a longer period of time. Therefore, only the most severe crashes have been chosen as event windows and will be examined.
3.2.2 Event window and estimation period
The event window is customary defined to be larger than the specific period of interest. As a result of the larger event window the periods surrounding the event will also be examined. In practice, the expanded event window is used to capture the price effects of the event after the stock market closes and to capture price effects of the event before it is known the general public, such as insiders’ information trading (Mackinlay, 1979). The following figure present
17
the time horizon of an event study, where t0 to t1 represents the estimation window, the event window is the period of time between t1 to t2 and the post-event window is from t2 to t3.
Estimation Event Post-event
window window window
t0 t1 0 t2 t3
Figure 4. Timeline for an event study
Typically, an estimation window and an event window do not overlap, this provides that the parameters of the market model are not influenced by the event. The estimations window usually consist of a period of one year or 250 trading days (Mackinlay, 1997). Therefore, this study will use an estimation window of 250 days before the event window as recommended by Mackinlay (1997).
In this study two different event windows are examined. The first event window is from 17 to 25 August and the second event is from 30 December to 7 January. Since the stock price returns of banks during a market crash are examined, we know the exact dates on which the Chinese stock market crash affected the market and therefore we do not have an specific event day and an expanded event window, but an event window that consist of the exact period of the most severe crashes of the Chinese stock market crash.
3.2.3 Selection criteria
The Chinese stock market crash, as mentioned before, exist of multiple crashes over a longer period of time. Since these crashes differ in the degree of intensity, only the most severe crashes will be examined. The impact of these most severe crashes affected stock markets around the world in a time period of about one week. Therefore, the following selection criteria rule has been established:
All crashes resulting in a 10% drop of the Chinese stock market in a one-week time period will be examined.
18
This selection criteria results in two one-week periods to be examined, namely 17 to 25 August 2015 and 30 December 2015 to 7 January 2016. In addition, we will also test the largest one-week stock market crash to compare with. This stock market crash was from 1 to 9 October 2008 and shows the impact of the G-SIB designation. Additionally, to examine the performance of the world’s largest banks during this period. We included only those banks who were publicly listed and whose assets exceeded $20 billion. This resulted in a sample of the world’s 75 largest banks, which in according to Bongini, Nieri and Pellagatti (2015) account for about 65% of global banks’ assets.
3.2.4 Calculating the Abnormal Return
The abnormal return is the actual ex-post return of a share minus the expected returns of that share on a given day. The following formula presents the formula used to calculate the abnormal return:
𝐴𝑅𝑖𝑡 = 𝑅𝑖𝑡− 𝐸(𝑅𝑖𝑡|𝑋𝑡) (1)
Where:
- 𝐴𝑅𝑖𝑡 is the abnormal return for bank i at time t - 𝑅𝑖𝑡 is the actual return for bank i at time t
- 𝐸(𝑅𝑖𝑡|𝑋𝑡) is the expected return for bank i at time t based on the market model
The ex-post actual returns for all banks have been retrieved from DataStream. Further, the expected returns are computed using the market model. The market model is a commonly used model to estimate the expected returns in an event study (Knapp, 1990). Brown and Warner (1980) find that a methodology based on the market model works for a wide variety of conditions. Moreover, multiple factor models have limited gains with respect to the market model due to the small marginal explanatory power of the added parameters and a negligible reduction in the variance (Mackinlay, 1997). In addition, according to Binder (1998) the market model is also able to provide reliable expected returns. Moreover, a major benefit of the market model over the constant mean return model is that it removes a portion of the return that is
19
related to variation in market returns (Mackinlay, 1997). In lines with the literature the market model will also be used in this study and is represented by the following formula:
𝑅𝑖𝑡 = 𝛼𝑖+ 𝛽𝑖𝑅𝑚𝑡+ 𝜀𝑖 (2)
Where:
- 𝑅𝑖𝑡 is the return for bank i at time t
- 𝑅𝑚𝑡 is the return of the market portfolio at time t - 𝛼𝑖 is the intercept of the market model for bank i - 𝛽𝑖 is the slope of the market model for bank i - 𝜀𝑖 is the error term of the market model for bank i
To calculate the expected return for bank i on day t, we estimated the parameters of the market model by Ordinary Least Squares regressions. The stock price return of the bank 𝑅𝑖𝑡 and the return of the market portfolio 𝑅𝑚𝑡 have been collected using Datastream. The error term of the market model 𝜀𝑖 and the variance of this error term are considered to be zero (Mackinlay, 1997). After collecting and estimating the parameters of the market model, the expected returns can be calculated by filling in the formula. Subsequently, the final step is to calculate the abnormal return by subtracting the expected return of the actual return.
3.2.5 Calculating the Cumulative Abnormal Return (CAR)
To calculate the CAR, all abnormal returns of the event window are accumulated as given by the following formula:
𝐶𝐴𝑅𝑖(𝑡1, 𝑡2) = ∑𝑡𝑡=𝑡2 1𝐴𝑅𝑖𝑡 (3)
20
- 𝐶𝐴𝑅𝑖 is the cumulative abnormal return of bank i for the event window (𝑡1, 𝑡2) - 𝐴𝑅𝑖𝑡 is the abnormal return of stock i at time t
- (𝑡1, 𝑡2) is the event window where 𝑡1 is the first day of the event window and 𝑡2 the last day.
3.2.6 Calculating the average and the standard deviation of the CARs
After calculating the CARs for every single bank, the average of those CARs is computed using the following formula:
𝐶𝐴𝑅 ̅̅̅̅̅̅(𝑡1, 𝑡2) = 1 𝑁∑ 𝐶𝐴𝑅 𝑡2 𝑡=𝑡1 (4) Where:
-𝐶𝐴𝑅̅̅̅̅̅̅(𝑡1, 𝑡2) is the average of the CARs of the banks for the event window (𝑡1, 𝑡2). - 𝐶𝐴𝑅 is the cumulative abnormal return of bank i for the event window (𝑡1, 𝑡2) - 𝑁 is the number of observations
- (𝑡1, 𝑡2) is the event window where 𝑡1 is the first day of the event window and 𝑡2 the last day.
Additionally, to perform a t-test we also need the standard deviation of the CARs. To calculate the standard deviation of the CARs we used the following formula:
𝐷(𝐶𝐴𝑅̅̅̅̅̅̅(𝑡1, 𝑡2)) = 1 𝑁∑ 𝛿𝑖(𝑡1, 𝑡2) 𝑁 𝑖=1 (5) Where:
- 𝜎𝑖2 is the standard deviation of the return of bank i - 𝑁 is the number of observations
21
3.2.7 The Mackinlay (1997) multiday t-test statistic
The first test that we use to test if the average of the cumulative abnormal returns is statistically significant and larger than zero is the Mackinlay (1997) t-test statistic. This t-test will be performed using the hypotheses and test-statistic as stated below:
𝐻0 : 𝐶𝐴𝑅̅̅̅̅̅̅(𝑡1, 𝑡2) = 0 𝐻1 : 𝐶𝐴𝑅̅̅̅̅̅̅(𝑡1, 𝑡2) > 0
𝑡 = 𝐶𝐴𝑅̅̅̅̅̅̅(𝑡1, 𝑡2)
𝑆𝐷(𝐶𝐴𝑅̅̅̅̅̅̅(𝑡1, 𝑡2)) (6)
3.2.8 The Cowan (1992) multi-day sign z-statistic
To verify the robustness of our event study results, we use in comparison with Abreu and Gulamhussen (2013) a second test static based on a different methodology, namely the Cowan multi-day sign z-statistic. The Cowan (1992) multi-day sign z-statistic examines if the number of stocks with positive CARs during the event window is equal to the fraction of stocks with positive CARs during the estimation period. First of all, we have to estimate the fraction of stocks with positive CARs during the estimation period using the following formula:
𝑃̂ = 1 𝑁∑ 1 𝑀 𝑁 𝑖=1 ∑𝑡𝑡=𝑡1 0𝑆𝑗𝑡 (7) Where:
- 𝑁 is the number of observations
- (𝑡0, 𝑡1) is the estimation window where 𝑡1 is the first day of the estimation window and 𝑡2 the last day.
- 𝑀 is the number of days used in the estimation window - 𝑆𝐽𝑡 = {1 𝑖𝑓 𝐴𝑅𝑗𝑡 > 0
22
The Cowan (1992) multi-day sign z-statistic uses normal approximation to the binomial distribution with the parameter (Cowan, 1992). The z-statistic uses the same hypotheses as the Mackinlay (1997) t-test statistic and is represented by the following formula:
𝑍𝐺 = 𝑤 − 𝑁𝑃̂
√𝑁𝑃̂(1 − 𝑃̂) (8)
Where:
- 𝑤 is the number of stock with positive CARs during the event window - 𝑁 is the number of observations
- 𝑃̂ is the fraction of stocks with positive CARs during the estimation period.
3.3 Cross-sectional analyses methodology
To examine the cumulative abnormal returns of G-SIBs and their peers more rigorously, a cross-sectional analysis will be performed to explore the outperformance of G-SIBs and to investigate the impact of bank characteristics on the CARs. First of all, we regress the dummy variables G-SIB and D-SIB on the CARs, where G-SIBs measures the performance of G-SIBs and D-SIB is used as a control variable for the banks that are classified as D-SIBs. These regressions are performed to analyse the performance of G-SIBs and D-SIBs with respect to their peers. Thereafter, we add similar to Abreu and Gulamhussen (2013) and Bongini, Nieri and Pellagatti (2015) bank characteristics such as size, performance, capital adequacy and complexity to the model. The supplementary measures will further analyse the influences of bank characteristics on the CARs. Subsequently, we analyse the influences of bank characteristics on the full sample, a subsample of G-SIBs and a subsample consisting of non-G-SIB and non-D-SIB peers. These analysis provide a comparison between the influences of bank characteristics on G-SIBs, their peers and the full sample. As a result, the subsample regressions might show if there are different stock market reactions to bank characteristics for G-SIBs in times of financial turmoil. All regressions will be performed on the first large
23
aftershock of the Chinese stock market crash from 17 to 25 August 2015 and the second large aftershock from 30 December 2015 to 7 January. Additionally, we will also perform regressions for the largest one-week stock market crash of the recent financial crisis to compare results before and after the designations of G-SIBs.
3.4 Descriptive statistics
This paragraph provides the descriptive statistics of the full sample and the construction of the variables based on the existing literature. In addition, section A of the appendices provides a table with the formulas used to construct the variables and section B of the appendices presents the descriptive statistics of the subsamples that will be used for the performed regressions in section 4.2.2.
Table 1
The sample includes the world’s 75 largest banks with returns available from Datastream, with Tier 1 capital ratio higher than 5%, exposure to China and total assets larger than $20 billion as of 2015. The CARs, ROA, Exposure to China and Tier 1 are given in decimals and all variables are winsorized on the 2.5% level. Bank characteristics are computed using data from 2015, the year in which the Chinese stock market crash started. CAR172015 is the CAR in the week of 17 to 25 August 2015, CAR302015 represents the CAR in the week of 30
December 2015 to 7 January 2016, CAR012008 is the CAR in the week of 1 to 9 October 2008, G-SIB is a dummy variable equal to one
where the bank is a global systematically important bank, D-SIB is a dummy variable equal to one where the bank is a domestic
systematically important bank, ROA is the bank’s return on assets calculated as net income divided by total assets, Tier 1 is the ratio of Tier 1 capital to risk-weighted assets, Banksize is the natural logarithm of the bank’s total assets, Exposure to China measures the consolidated exposure to China, by nationality of reporting bank, retrieved from the Bank for International Settlements database. Calculated as
consolidated positions on counterparties resident in china divided by consolidated foreign claims, by nationality of reporting bank. Leverage is a measure of financial leverage and is the ratio of total assets to equity. For a measure of potential agency cost we use Tobin’s Q, defined as market capitalization plus book value of debt divided by total assets. As a proxy for liquidity, the ratio of liquid assets to total assets is used. Finally, as a proxy for asset quality, we use the ratio of loan loss reserves to total loans.
Observations Mean Standard deviation Minimum Maximum
CAR172015 75 0.022 0.0782 -0.075 0.337 CAR302015 75 -0.014 0.029 -0.072 0.066 CAR012008 72 0.091 0.119 -0.087 0.397 G-SIB 75 0.387 0.490 0.000 1.000 D-SIB 75 0.213 0.412 0.000 1.000 ROA 75 0.007 0.005 -0.007 0.014 Banksize 75 12.257 1.776 9.946 14.896 Tier 1 75 0.126 0.025 0.077 0.191 Exposure to China 75 0.0788 0.210 0.000 0.993 Leverage 75 11.880 4.296 4.841 22.024 Tobin’s Q 75 0.393 0.466 0.088 2.595 Liquidity 75 0.085 0.0756 0.012 0.318 Asset Quality 75 0.015 0.032 0.001 0.155
24
Table 1 presents the descriptive statistics of the variables which are used in the cross-sectional analysis. The first three variables CAR172015, CAR302105 and CAR012008 are the dependent variables and represent the bank’s cumulative abnormal return with respect to the market portfolio for the different event windows. Every dependent variable corresponds to a one-week event window of stock returns during a stock market crash, where CAR172015 and CAR302015 represent aftershocks of the Chinese stock markets crash from 17 to 25 August 2015 and 30 December 2015 to 7 January 2016, respectively. CAR012008 is used for robustness and represents the largest stock market crash during a one-week period of stock returns in the recent financial crisis, which was from 1 to 9 October 2008. One can observe that the minimum and maximum of the CARs differ a lot, which can be explained by the fact that these are CARs in times of financial turmoil. Furthermore, it is interesting to notice that the mean of CAR172015 and CAR012008 is positive and shows outperformance of the world’s largest banks with respect to the market portfolio, whereas CAR302105 is negative and shows underperformance.
The variables GSIB and DSIB are dummy variables and equal to 1 if the bank is classified as a global systematically important bank or domestic systematically important bank, respectively. The means of the variables show that 38.67% of the banks are classified as G-SIBS, 21.33% are classified as D-SIBs and that 40.10% of the sample exists of peers.
Furthermore, Exposure to China controls for the bank’s consolidated exposure to China during the Chinese stock market crash. This variable is the ratio of consolidated positions on counterparties resident in China to consolidated foreign claims, by nationality of reporting banks. The minimum and maximum have a wide range since the four Chinese G-SIBs mainly invest domestically, whereas some peers from other countries have negligible exposure to China.
In addition, just as Bongini et al. (2015) and Abreu and Gulamhussen (2013) this studies analyses characteristics of G-SIBs and their peers. The focus is on variables that capture different bank characteristics such as size, performance, capital adequacy and complexity. Banksize measures the size of the bank and is defined as the natural logarithm of total assets. The performance measures are capital adequacy, profitability, liquidity and asset quality. Tier 1 and Leverage are measures of capital adequacy and denote the regulatory Tier 1 capital definition and the ratio of total assets to equity, respectively. ROA is a proxy for profitability and represents the return on assets, calculated as net income divided by total assets. As a proxy for Liquidity, we use the financial liquidity ratio of liquid assets to total assets. Asset quality is a proxy of the quality of the bank’s assets and is the ratio of loan loss provisions to total loans.
25
For complexity, we use a measure of agency, namely Tobin’s Q. The variable is defined as the sum of market capitalization and book value of debt to total assets and indicates potential agency cost if Tobin’s Q is lower than 1 (Abreu and Gulamhussen, 2013). Table 1 shows that on average the large banks do experience these agency costs, however, a maximum of 2.595 indicates that some of the banks in our sample do not experience agency costs.
4. Results
In this section we will provide our event study results and the results of the cross-sectional analyses. First of all, paragraph one will provide the event study results which examine the size and significance of the CARs. Thereafter, paragraph 2 will investigate the outperformance of G-SIBs and the different market reactions to bank characteristics for G-SIBs, a peer group of non-G-SIBs and non-D-SIBs and the full sample.
4.1 Event study results
Table 2 presents the event study results and examines the outperformance of G-SIBs compared to a subsample of D-SIBs, a subsample of non-G-SIB and non-D-SIB peers and the full sample. The table reports the abnormal returns and their statistical significance for the three different event windows. The first column shows the bank sample consisting of the world’s 75 largest banks, whichever divided in three subsamples of G-SIBs, D-SIBs and their peers. The corresponding number of observations represented in the second column. The third column represents the average cumulative abnormal return in percentages and the final two columns represents two multi-day test statistics based on two different methodologies to test the statistical significance of the CARs similar to the study of Abreu and Gulamhussen (2013). The fourth column presents Mackinlay’s (1979) multi-day t-statistic and the fifth column gives the p-value of Cowan (1992) multi-day sign z-statistics. The results of table 2 are found to be robust to the robustness tests performed in section C of the Appendices.
Table 2
The market reaction during the three events. The following table reports the mean absolute returns over the three event periods, whereas the first two events are aftershocks of the Chinese stock market crash and the third event is the largest crash during the recent financial crisis and is used as a robustness check. The statistical significance of the abnormal returns are tested using the Mackinlay (1997) multi-day t-statistic and the Cowan (1992) multi-day sign z-statistic. The results of both the test statistics are presented in column 4 by a t-value and in column 5 by a p-value. For all events the abnormal returns have been examined over a one-week period, in which the stock market crash occurred.
26
(Table 2 continued)
Panel A: CAR172015 (17 - 25 August, 2015)
Bank sample N Mean CAR (%) t-statistic Cowan Generalized Sign Test
Full sample 75 2,094% 2,118** 0.610
G-SIBS 29 0,851% 1,607 0.021**
D-SIBS 16 7,470% 2,047** 0.875
Peer Group 30 0,309% 0,268 0.179
Panel B: CAR302015(30 December 2015 - 7 January 2016)
Bank sample N Mean CAR (%) t-statistic Cowan Generalized Sign Test
Full sample 75 -1,373% -3,940*** 0.003**
G-SIBS 29 -0,264% -0,444 0.509
D-SIBS 16 -1,406% -2,296** 0.107
Peer Group 30 -2,392% -4,593*** 0.000***
Panel C: CAR012008 (1 - 9 October, 2008)
Bank sample N Mean CAR (%) t-statistic Cowan Generalized Sign Test
Full sample 72 9,602% 6,780*** 0.000***
G-SIBS 28 8,959% 4,149*** 0.001***
D-SIBS 16 7,045% 1,952* 0.309
Peer Group 27 10,089% 4,493*** 0.001***
*,** and *** respectively denote significance at a 10%, 5% and 1% level.
Panel A of Table 3 presents the variable CAR172015 which is the average of the CARs during the first aftershock of the Chinese stock market crash from 17 to 25 August 2015. During this period we find positive averages of the CARs for all the samples, especially the sample of D-SIBs shows large positive CARs with respect to the market portfolio. We find outperformance of G-SIBs using the Cowan (1992) multi-day sign z-statistics on a 5% level, however this is not confirmed by the Mackinlay (1979) multi-day t-test statistic which finds evidence for outperformance D-SIBs and the full sample. Nevertheless, Panel A suggests that both G-SIBs and D-SIBs outperformed their peer group during the first large aftershock of the Chinese stock market crash.
Panel B presents the variable CAR302015 which is the average of the CARs during the second aftershock of the Chinese stock market crash from 30 December 2015 to 7 January 2016. This panel is the most interesting, since during this crash the Mean CARs are negative which suggest that bank stock prices underperformed the market portfolio and that this crash had more impact on banks. Panel B shows that the full sample underperformed the stock market based on
27
a 1% significance level of the Mackinlay (1979) multi-day t-test statistic and a 5% significance level of the Cowan (1992) multi-day sign z-statistics. However, from the subsamples one can notice that there is no statistical evidence that G-SIBs underperformed the market, whereas there is strong evidence of underperformance of their peers. The results finds from both the test statistics that the peer group underperformed by 2.392% on a 1% significance level. Therefore, Panel B finds strong evidence for outperformance of G-SIBs with respect to their peers. Further, it suggests that D-SIBs performed worse than G-SIBs, although they outperformed the peer group.
Panel C shows the event study results of the variable CAR012008 which represents the mean CARs during the largest one-week period stock market crash of the recent financial crisis. Panel C is used as a robustness test to examine the performance of G-SIBs with respect to their peers in the recent financial crisis when they were not yet classified as G-SIBs. Panel C finds evidence for outperformance of the full sample, G-SIBs and their peers by both test statistics on 1 % significance level. Further, the D-SIB sample is only found to outperform the market portfolio by Mackinlays (1979) t-test statistic. The mean CAR for the peer groups is even the largest among the samples which suggest that the peer group slightly outperformed both the G-SIBs and D-G-SIBs in the recent financial crisis.
All in all, Table 3 finds outperformance of G-SIBs with respect to their peers during the aftershocks of the Chinese stock market crash. However, it suggest slightly underperformance of G-SIBs to their peers during the recent financial crisis, which is argumentative since G-SIBs had only to be classified as G-SIBs upwards of 2011. In addition, these results indicate that the designation as a G-SIB and the corresponding regulatory requirements can have a positive effect on stock price performance and help to stabilize the bank’s stock price returns during a stock market crash which confirms the first hypothesis.
4.2 Cross-sectional analysis results
This paragraph examines the cumulative abnormal returns more rigorously. In section 4.2.1, we provide in addition to the event study results a cross-sectional analysis to investigate the outperformance of G-SIBs and test the first hypothesis. Subsequently, section 4.2.1 presents a cross-sectional analyses to examine the different market reactions to bank characteristics for G-SIBs, a peer group of non-G-SIBs and non-D-SIBs and the full sample. The presented results
28
in both sections 4.2.1 and 4.2.1 hold to several robustness performed in section D and E of the Appendices.
4.2.1 The outperformance of G-SIBs
In this section, we test in addition to the event study, the first hypothesis and examine the outperformance of G-SIBs by performing a cross-sectional analysis on the CARs using dummy variables to indicate the designation of the banks and control variables to control for bank characteristics. Table 3 reports the performed regressions for the different event windows on the full sample of the world’s 75 largest banks. Regressions 1 to 4 show the cross-sectional analysis on the CARs during the Chinese stock market crash, whereas regressions 5 and 6 presents this analysis for the largest stock market crash of the recent financial crisis to test the robustness of the results. The Exposure to China variable has been left out in regression 5 and 6, since Exposure to China is not a relevant variable for explaining CARs during the recent global financial meltdown.
Table 3
The following table examines the outperformance of G-SIBs and the influence of bank characteristics on the CARs. In columns 1, 3 and 5 we regress the CARs on the dummy variables G-SIB and D-SIB, these regressions measure the performance of G-SIBs and D-SIBs with respect to their peers. Subsequently, in columns 2, 4 and 6 we add bank characteristics the regression model to further explore the CARs. The sample includes the world’s 75 largest banks with returns available from Datastream, with Tier 1 capital ratio higher than 5%, exposure to China and total assets larger than $20 billion as of 2015. The CARs, ROA, Exposure to China and Tier 1 are given in decimals and all variables have been winsorized on 2.5% level. Bank characteristics are computed using data from 2015, the year in which the Chinese stock market crash started. CAR172015 is the CAR in the week of 17 to 25 August 2015, CAR302015 represents the CAR in the week of 30 December 2015 to 7
January 2016, CAR012008 is the CAR in the week of 1 to 9 October 2008, G-SIB is a dummy variable equal to one where the bank is a global
systematically important bank, D-SIB is a dummy variable equal to one where the bank is a domestic systematically important bank, ROA is the bank’s return on assets calculated as net income divided by total assets, Tier 1 is the ratio of Tier 1 capital to risk-weighted assets, Exposure
to China measures the consolidated exposure to China, by nationality of reporting bank, retrieved from the Bank for International Settlements
database. Calculated as consolidated positions on counterparties resident in china divided by consolidated foreign claims, by nationality of reporting bank. Leverage is a measure of financial leverage and is the ratio of total assets to equity. For a measure of potential agency cost we use Tobin’s Q, defined as market capitalization plus book value of debt divided by total assets. As a proxy for liquidity, the ratio of liquid assets to total assets is used. Finally, as a proxy for asset quality, we use the ratio of loan loss reserves to total loans.
Variable CAR172015 CAR302015 CAR012008
(1) (2) (3) (4) (5) (6) G-SIB 0.002 -0.005 0.022*** 0.010 -0.014 0.041 (0.15) (-0.18) (3.02) (1.12) (-0.45) (0.93) D-SIB 0.069** 0.074* 0.009 0.004 -0.038 -0.015 (2.01) (1.80) (1.24) (0.56) (-0.96) (-0.36) Tier 1 0.468 0.021 -1.153* (1.55) (0.16) (-1.78) ROA 0.672 1.852* --0.979 (0.38) (1.70) (-0.24) Exposure to China -0.652 -0.240
29 (Table 3 continued) (-1.11) (-0.50) Tobin's Q 0.002 -0.006 -0.037 (0.08) (-0.48) (-0.96) Leverage 0.001 0.004*** -0.004 (0.39) (4.12) (-0.97) Liquidity -0.168 -0.039 -0.0326 (-1.31) (-1.14) (-0.15) Asset Quality 0.357* 0.071 0.0918 (1.69) (0.46) (0.20) Constant 0.007 -0.040 -0.025*** -0.066*** 0.104*** 0.293*** (0.59) (-0.72) (-5.81) (-2.55) (4.54) (3.08) Observations 75 75 75 75 72 72 R2 0.1296 0.1984 0.1182 0.4015 0.0146 0.1081 Prob > F 0.1264 0.0055 0.0132 0.0000 0.6316 0.0057
*,** and *** respectively denote significance at a 10%, 5% and 1% level.
In the first regression of Table 3, the dummy variables G-SIB and D-SIB are regressed on the dependent variable CAR172015. Both the variables G-SIB and D-SIB are positive which indicates that G-SIBs and D-SIBs outperformed their peers. However, the first regression only finds statistical evidence on a 5% level for outperformance of D-SIBs by 6.9%.
In the second regression, the bank characteristics are added to the regression model, just as in the studies of Abreu and Gulamhussen (2013) and Bongini, Nieri and Pellagatti (2015). The dummy variable D-SIB keeps significant after adding the bank characteristics which indicates that D-SIBs outperformed their peers during the aftershock of the Chinese stock market crash from 17 to 25 August 2015. Further, regressions two also finds a positive and significant relation between Asset Quality and the cumulative abnormal returns. This indicates that bank with a higher capacity of catching up losses outperformed during the first aftershock of the Chinese stock market crash.
In the third regression, the dependent variable changes to CAR302015 to examine the performance of the banks during the second large aftershock of the Chinese stock market crash. The regression examines again the relation between the CARs and the designation of the bank. The main variable SIB is found to positive and significant on a 1% which suggests that G-SIBs had on average 2.2% higher CARs during the second aftershock of the Chinese stock market crash and confirms the first hypothesis. Further, the results also shows on average higher CARs for D-SIBs, although there is no statistical evidence of outperformance.
Regressions 4 adds again the bank characteristics to the model, as a result the variables G-SIB and D-SIB are still positive, although not significant. The model shows that return on
30
assets are important in explaining the CARs, since we find a large coefficient of ROA which is significant on a 10% level. This is also economically argumentative since firm performance is known to affects investors’ sentiment and stock prices (Fama, 1970). Further, regressions 4 finds a positive relation between CAR302015 and Leverage. Since Leverage is the ratio of total assets to equity, the results find that a lower level of book leverage is related to higher CARs and support the risk culture hypotheses of Fahlenbrach et al. (2012).
In regression five, one can observe that both the variables G-SIB and D-SIB are negative which indicates underperformance of G-SIBs and D-SIBs during the recent financial crisis, although we find no statistical evidence for this result. Again interpreting these robustness results one should take into account that G-SIBs and D-SIBs were not yet designated back in 2008. Therefore, one could argue that the designation as a G-SIB or D-SIB and the corresponding requirements have a positive effect on the bank’s stock returns during stock market crashes.
Finally, in the sixth regression the variable G-SIB turns positive after adding the bank characteristics to the model. However, we find no statistical evidence of outperformance of G-SIBs during the largest stock market crash of the recent financial crisis. Additionally, we find relevance of Tier 1 capital in explaining the CARs during this stock market crash. The results of regression 6 show a large negative coefficient which is significant on a 10% level indicating that Tier 1 capital had a negative effect on stock performance during the recent financial crisis. This finding is in line with Akhigbe et al. (2012) who also find a negative relation between Tier 1 capital and stock returns during the recent financial crisis, but in contrast to findings of Demirguc-Kunt et al. (2013) who find a positive relationship.
4.2.2 The stock market reactions to bank characteristics for G-SIBs and their peers This section provides a cross-sectional analysis to test the second and third hypothesis. The analysis scrutinizes the effects of bank characteristics on the CARs for the different event windows and subsamples. In the regressions presented in Table 4, we regress the bank characteristics on the full sample, a subsample consisting of 29 G-SIBs and a subsample consisting of 30 non-GSIB and non-D-SIB peers. Furthermore, the bank characteristic Bank size is added to the regression model. Since several studies emphasize the relevance of bank size in explaining bank performance (Beltratti and Stulz, 2012; Schäfer et al., 2013 and Kane, 2000). However, we were not able to add this variable in the regressions of table 3, since the G-SIBs are the largest banks in the world. Therefore, this would result in a high correlation
31
between G-SIBs and bank size, which biases the regression results. So dividing the full sample in subsamples solves this problem and we can still measure the size of effect on the CARs.
Table 4
This table examines the effect of bank characteristics on the CARs for the full sample, a subsample of G-SIBs and subsample consisting of non-G-SIB and Non-D-SIB peers. In addition, the bank characteristic Banksize is added to the regression to measure the impact of the size on the CARs. In Columns 1, 4 and 7 the bank characteristics are regressed on the G-SIB subsample consisting of 29 G-SIBs. Columns 2, 4 and 6 provides the same regression for the peer group consisting of 30 non-G-SIB and non-D-SIB peers. Finally, columns 3, 6 and 9 represent the impact of bank characteristics on CARs for the full sample. The full sample includes the world’s 75 largest banks with returns available from Datastream, with Tier 1 capital ratio higher than 5%, exposure to China and total assets larger than $20 billion as of 2015. The CARs, ROA, Exposure to China and Tier 1 are given in decimals and all variables have been winsorized on the 2.5% level. Bank characteristics are computed using data from 2015, the year in which the Chinese stock market crash started. CAR172015 is the CAR in the week of 17 to 25 August 2015, CAR302015 represents the CAR in the week of 30 December 2015 to 7 January 2016, CAR012008 is the CAR in the week of 1 to 9 October
2008, G-SIB is a dummy variable equal to one where the bank is a global systematically important bank, D-SIB is a dummy variable equal to one where the bank is a domestic systematically important bank, ROA is the bank’s return on assets calculated as net income divided by total assets, Tier 1 is the ratio of Tier 1 capital to risk-weighted assets, Banksize is the natural logarithm of the bank’s total assets, Exposure to
China measures the consolidated exposure to China, by nationality of reporting bank, retrieved from the Bank for International Settlements
database. Calculated as consolidated positions on counterparties resident in china divided by consolidated foreign claims, by nationality of reporting bank. Leverage is a measure of financial leverage and is the ratio of total assets to equity. For a measure of potential agency cost we use Tobin’s Q, defined as market capitalization plus book value of debt divided by total assets. As a proxy for liquidity, the ratio of liquid assets to total assets is used. Finally, as a proxy for asset quality, we use the ratio of loan loss reserves to total loans.
Variable CAR172015 CAR302015 CAR012008
G-SIBs Peer group Full sample G-SIBs Peer group Full sample G-SIBs Peer group Full sample (1) (2) (3) (4) (5) (6) (7) (8) (9) Tier 1 0.166 -0.062 0.462 0.306* -0.010 0.008 -1.268 -1.544 -0.981 (0.78) (-0.08) (1.05) (1.94) (-0.04) (0.06) (-1.32) (-1.09) (1.52) Banksize 0.013 0.002 0.005 0.025*** -0.002 0.006*** -0.016 -0.117* -0.003 (1.43) (0.06) (0.60) (3.60) (0.14) (2.62) (-0.38) (-1.95) (-0.27) ROA -1.707 0.272 1.747 1.322 -2.018 1.556** -0.782 -5.401 -1.430 (1.39) (0.05) (0.63) (1.46) (-1.27) (2.00) (-0.14) (-0.58) (-0.35) Exposure to China -0.612 -0.834 0.192 -1.522*** 0.871 -0.169 (-1.33) (-0.20) (0.19) (-4.48) (0.66) (-0.61) Tobin's Q -0.011 -0.136 -0.008 0.016* -0.076 -0.010 -0.028 -0.195 -0.034 (-0.89) (-0.75) (-0.27) (1.79) (-1.33) (-1.16) (-0.74) (-0.92) (-0.87) Leverage 0.002 -0.005 0.001 0.002*** 0.001 0.003*** -0.001 -0.016 -0.002 (1.64) (-0.76) (0.35) (2.37) (0.24) (3.62) (-0.31) (-1.33) (-0.42) Liquidity -0.013 0.358 -0.225 0.109** -0.092 -0.055 -0.006 0.633 0.095 (-0.18) (0.72) (-1.58) (2.09) (-0.58) (-1.38) (-0.02) (0.75) (0.45) Asset Quality -0.035 0.741 0.272 -0.029 0.112 0.048 1.211* 0.780 0.105 (-0.23) (1.25) (0.87) (-0.26) (0.59) (0.54) (1.85) (0.75) (0.23) Constant -0.198 0.074 -0.103 -0.418*** -0.033 -0.118 0.505 1.717*** 0.2857*** (-1.35) (0.20) (-1.02) (-3.87) (-0.27) (-4.14) (0.76) (2.45) (2.34) Observations 29 30 75 30 29 75 28 27 72 R2 0.534 0.208 0.061 0.7975 0.423 0.448 0.281 0.275 0.519 Prob > F 0.027 0.698 0.827 0.000 0.106 0.000 0.3604 0.347 0.085
32
In regressions 1, 2 and 3 of Table 4, we regress the CARs during the first aftershock of the Chinese stock market crisis on bank characteristics for the different subsamples. The regressions find no statistical evidence for explaining the CARs by the bank characteristics. However, as stated in hypothesis 3, regression 1 finds a positive relation between the performance of G-SIBs and the level of Tier 1 capital during the Chinese stock market crash, although it is not significant. In addition, one can observe that the added variable Banksize has a positive but insignificant relationship with respect to the CARs. Overall, regressions 1, 2 and 3 find that bank characteristics do a poor job in explaining the CARs during the first aftershock of the Chinese stock market crash.
Nonetheless, the bank characteristics do explain the CARs very well during the second aftershock of the Chinese stock market crash as shown in regression 4, 5 and 6. Especially, the CARs of the G-SIBs are very well explained in regression 4, as one can observe that from the significance of the bank characteristics, R-squared of 0.7975 and Prob > F statistic equal to zero. A possible explanation for this phenomenon could be that banks are more affected by the second aftershock of the Chinese stock market crash than by the first aftershock. One can observe this from the CARs stated in panel A and B of Table 2. These panels show that banks outperformed the market portfolio during the first aftershock, whereas they show underperformance of the market during the second aftershock. These differences in the CARs indicate that the second aftershock had more impact on bank’s their stock price returns.
In regression 4, we find a positive and significant relation between the CAR and the Tier 1 capital. This result confirms the third hypothesis that G-SIBs with more high quality capital outperformed during the Chinese stock market crash on a 10% significance level. According to these findings, an increase of 1% in the Tier 1 capital ratio for G-SIBs is related to 0.306% higher CARs during the second aftershock. Moreover, this result supports the findings of Beltratti and Stulz (2012) and Demirguc-Kunt et al. (2013), who also find a positive relation between stock price performance and Tier 1 capital. According to these results, the enhanced capital requirements as stated in Basel III should help to stabilize the stock market performance of G-SIBs during future stock market crashes. In addition, regression 4 finds in lines with previous evidence of Schäfer et al. (2013) and Kane (2000) a positive relation between bank size and abnormal returns. Further, as expected this regression finds a negative and significant relation between the Exposure to China variable and the CARs, which is also economically argumentative since it indicates that G-SIBs who were more exposed to the Chinese financial market underperformed. Tobin’s Q which is a proxy of potential agency costs is also positive and significant which suggests that banks with lower agency costs or better performance gain
