Tilburg University
Essays on banking and regulation
Todorov, R.I.
Publication date: 2013
Document Version
Publisher's PDF, also known as Version of record Link to publication in Tilburg University Research Portal
Citation for published version (APA):
Todorov, R. I. (2013). Essays on banking and regulation. CentER, Center for Economic Research.
General rights
Copyright and moral rights for the publications made accessible in the public portal are retained by the authors and/or other copyright owners and it is a condition of accessing publications that users recognise and abide by the legal requirements associated with these rights. • Users may download and print one copy of any publication from the public portal for the purpose of private study or research. • You may not further distribute the material or use it for any profit-making activity or commercial gain
• You may freely distribute the URL identifying the publication in the public portal Take down policy
If you believe that this document breaches copyright please contact us providing details, and we will remove access to the work immediately and investigate your claim.
Proefschrift ter verkrijging van de graad van doctor aan Tilburg
University op gezag van de rector magnificus, prof. dr. Ph. Eijlander, in
het openbaar te verdedigen ten overstaan van een door het college voor
promoties aangewezen commissie in de aula van de Universiteit
op woensdag 5 juni 2013 om 10.15 uur
door
Radomir Ivanov Todorov
Promotor:
prof.dr. Thorsten Beck
Copromotor:
dr. Olivier De Jonghe
Overige Leden:
prof. dr. Franklin Allen
prof. dr. Hans Degryse
prof. dr. Vasso Ioannidou
dr. Fabio Feriozzi
dr. Maria Fabiana Penas
I am grateful to the many people that have supported me in writing this
thesis. I thank my supervisors for their invaluable support and advice, and I hope
to continue working with them in the future. I also thank the members of
my PhD
committee, which comments have greatly benefitted my work. The
Department of Finance at CentER and the Wharton Financial Institutions
Center in Philadelphia have been great hosts for my research, and my
spe-cial thanks go to Frank de Jong, Joost Driessen, and Franklin Allen. I also
want to take this opportunity to thank my fellow PhD students, colleagues,
and friends
for the good laughs and moments we had in the past years.
Above all, I thank my family for being a constant source of unconditional
support, encouragement, and inspiration over the years. Unfortunately, my
father has passed away a month before my defense in June 2013. My dad
has been a great source of inspiration and support, and he
would have been
very happy if he could attend my
defense. I am immensely grateful to my
parents
for standing by me throughout my life and the past years, and I
gratefully dedicate this thesis to them.
Radomir Todorov
Frankfurt (Main),
May 2013
Preface……….. 1
Chapter 1. The Effect of Peers on Bank Capital……….. 4
Chapter 2. Supervising Cross-Border Banks: Theory, Evidence and Policy…….. 54
Chapter 3. Banks and Monetary Policy in Africa: Evidence from Uganda……… 95
This Ph.D. dissertation consists of three essays on multinational bank supervision, bank capital, and monetary policy. The purpose of these essays is to explore (i) the in‡uence of peer banks on bank …nancing decisions, (ii) the leniency of national supervisors in the intervention of banks with cross-border activities, and (iii) the impact of monetary policy and international credit market conditions on the provision of credit in a developing country. Below I present a brief overview of the three chapters of the thesis.
Chapter 1: The E¤ ect of Peers on Bank Capital
This chapter studies the impact of peer banks on a bank choice of capital. Previous studies suggest that banks herd in many areas of their business, which may have aggravated the extent of the global …nancial crisis. However, little is known whether bank capital decisions are in‡uenced by mimicking the capital choice of peers. The question is important because regulating capital is an essential part of any debate on bank regulation. If peer banks in‡uence bank capital, this can also lead to externalities with a signi…cant impact on the …nancial system.
The study adds to the discussion on optimal bank capital level. In the banking literature, there is no clear consensus yet about the optimal range for capital, but a common agreement that bank value decreases when capital is either very little or too much. In either case, capital sends a warning sign to investors about the degree of agency con‡icts or asymmetric information in the bank. Since market investors have to make a guess about the permissible level of capital for any bank in the group, the capital holdings of peer banks can provide a clue about the admissible range of capital. For this reason, a bank has an incentive to adjust its capital ratio to the capital level in the peer group; otherwise, it can face a higher cost of equity or debt …nancing in the markets.
In line with these arguments, I show empirically in the chapter that peer bank choices on capital have a statistically and economically signi…cant impact on bank capital. The evidence from data for publicly listed commercial banks in the US between 1971 and 2010 suggests that publicly listed banks follow the capital choices of their peers when they decide on changes in bank capital, or in other areas of …nancial policy such as the reliance on nondeposit liabili-ties. Furthermore, the evidence suggest that the impact of peers intensi…es with an increase in competition and changes in the banking environment.
Beck and Wolf Wagner)
Cross-border banking has gained importance across Europe in recent decades, as a part of a larger globalization wave in …nancial services. The problematic resolution of failing cross-border banks in Europe during the current crisis has focused attention on the discrepancy in geographic boundaries of bank activities and their supervision. The failure of the Icelandic banks, for example, or the Fortis bank have come with wide-ranging economic and political consequences. In this chapter, we study the distortions of national supervision of international banks by comparing the trade-o¤s of national versus supra-national resolution authority.
In a theoretical model, we demonstrate …rst the distorted incentives of national supervisors when deciding to intervene in failing banks with cross-border activities. We show that national supervisors’ incentives to intervene in a timely manner in a failing bank increase in the bank foreign equity share and decrease in the bank share of foreign deposits and assets.
In a second step, we provide empirical evidence consistent with the model using a sample of intervened banks during the crisis of 2007-2009. We …nd that banks with a higher share of foreign assets or deposits are subject to interventions at a higher level of fragility. Banks with a higher share of foreign equity are subject to less lenient regulatory decisions.
Finally, we add to our analysis a discussion of the current regulatory arrangements for cross-border banking in Europe and recent reforms. For example, a supranational supervisor could always improve welfare because of taking into account the e¤ects that materialise outside of each country. However, supra-national supervision might itself also be subject to imperfections and we discuss in the paper cases when the supranational supervisors may fail.
Chapter 3: Banks and Monetary Policy in Africa: Evidence from Uganda (with Thorsten Beck)
This chapter studies how changes in monetary policy and foreign interest rates a¤ect bank credit provision in Uganda between 1999 and 2005. Like in many other developing countries, the country faces conditions for the conduct of monetary policy that are uncommon for developed countries. There is a small banking sector that is dominated by foreign-owned banks. A signi…-cant share of assets and deposits at some banks are in foreign currency. Further, the institutions necessary for the ‡uent conduct of monetary policy are at an early stage of development.
particular ways of mitigating the impact of monetary policy shocks. We try to …nd answers to these questions using a unique quarterly dataset on commercial banks coming from the central bank in Uganda.
Our analysis gives no evidence at the aggregated data level of bank loans or bank assets being sensitive to monetary policy in Uganda. In contrast, we …nd preliminary evidence for the transmission of foreign …nancial shocks to the Ugandan economy. We discuss and test in a following step for the existence of a bank lending channel of monetary policy under three alternative views. Under the conventional credit view, the rise in the policy interest rate induces a decline in bank liquidity. When banks cannot get additional funds, they have to cut the supply of loans. Another view considers the role of foreign-owned banks in transmitting foreign monetary shocks in the host country. Monetary tightening in a parent bank’s country can negatively a¤ect the credit supply in the economy hosting its subsidiary. Our empirical analysis let us conclude there is no support for the conventional view of a bank lending channel in Uganda, while foreign monetary policy can have real e¤ects on the economy of the country. Changes in the foreign interest rates are found to inhibit loan growth in domestic currency and increase the growth in foreign assets.
A third view of the bank lending channel stresses the role of …nancial dollarization in bank deposits. Under this view, banks can o¤set any liquidity shortage during monetary tightening by converting foreign currency deposits in the domestic currency. Being less a¤ected by mone-tary shocks, those banks should have a higher loan growth than other banks during monemone-tary tightening. We …nd support for the role of deposit dollarization in the transmission of monetary policy shocks in the banking system. We also provide evidence of reserve requirements having a signi…cant impact on bank asset composition.
Chapter 1
The Effect of Peers on Bank Capital
Abstract: I analyze if bank peers in‡uenced bank decisions on …nancial policy and Tier 1
capital among publicly listed US commercial banks between 1971 and 2010. The results show that capital decisions of peer banks are an economically signi…cant determinant of capital. Banks adjust capital ratios in response to capital decisions of other banks in their peer group. Similarly, banks make adjustments to Tier 1 capital and nondeposit liabilities depending on the choice made by their peers. There is also evidence suggesting that the impact of peer banks on bank capital decisions is stronger when competition among banks intensi…es.
1.1 Introduction
This paper studies the impact of peer banks on a bank choice of capital.1 Previous studies
suggest that banks herd in many areas of their business, which may have aggravated the extent of the global …nancial crisis (Freixas (2010), Farhi and Tirole (2012), Bon…m and Kim (2012)). However, little is known whether bank capital decisions are in‡uenced by the capital choice of peers. The question is important because regulating capital is an essential part of any debate on bank regulation. If peer banks in‡uence bank capital, this can also lead to externalities with a signi…cant impact on the …nancial system. Therefore, better understanding of peer e¤ects among banks is necessary for an e¤ective bank capital regulation.
The study also adds to the discussion on optimal bank capital level. In the banking literature, there is no clear consensus about the optimal range for capital. However, there is a common agreement that bank value decreases when capital is either very little or too much. In either case, capital sends a warning sign to investors about the degree of agency con‡icts or asymmetric information in the bank (e.g. Berger et al. (1995), Acharya et al. (2013)). Since market investors have to make a guess about the permissible level of capital for any bank in the group, the capital holdings of peer banks can provide a clue about the admissible range of capital. Therefore, a bank has an incentive to adjust its capital ratio to the capital level in the peer group; otherwise, it can face a higher cost of equity or debt …nancing in the markets.
1In the area of research in banking, the capital ratio traditionally refers to the ratio of equity to assets.
In addition to bank capital, banks could also adjust the mix of debt financing or regulatory capital in response to peer levels. Market investors could get additional information from these variables about bank financial health. Holding capital above the minimum level can be costly except out of precautionary motives not to violate the minimum capital requirement. The same line of reasoning can apply to the ability of banks to rely on non-deposit funding as well. Deposits offer a cheap source of financing to banks, also subject to deposit insurance. Because bank ability to attract non-deposit debt relative to its peers depends on financial health and soundness, a high share of non-deposit debt can lower the cost of debt financing.
In the analysis of bank peer interactions, I use data for publicly listed commercial banks in the US between 1971 and 2010. A challenge comes from disentangling the simultaneity between actions of an individual and the actions of peers in the peer group. This type of simultaneity is commonly known as the reflection problem as discussed by Manski (1993, 2000). An instrumental variable approach can help in resolving the problem. It requires a variable that is correlated with peer characteristics but not with the characteristics of any individual member of the group. For isolating the impact of peer financial policies on bank capital, I use the idiosyncratic shocks to stock returns of peer banks in an IV analysis. My estimations show that the peer equity shock has a significant impact on peer capital, but it is unlikely to explain the capital of a bank excluded from the peer group. The approach follows Leary and Roberts (forthcoming), which evidence suggest that this instrument can adequately address the endogeneity issues in a peer effects analysis on publicly listed firms.
In addition, the paper brings into discussion what should be the proper definition of a bank peer group. The theoretical literature predicts that bank size and business focus are a joint measure of bank success (Strahan (2008)). Forming peer groups in the paper follows this prediction reflecting a standard approach among practitioners and supervisors when comparing peer banks (FFIEC (2008)). First, in every time period banks are sorted in two peer groups having a diversified or regional focus. The banks in each peer group are ranked according to total assets and assigned to their quartile of the asset distribution. In a robustness check, I also follow a less general approach. I assign banks according to their geographic location in one of the 12 Federal Reserve Bank districts in the first step. Next, I repeat the same procedure in forming the peer groups. The estimated models yield results that are similar in economic and statistical significance under both definitions of the peer groups. The findings support the view of banks being perceived by investors as a homogeneous group nationwide among investors while bank location seems to be of secondary importance.
Finally, the results in the paper are of relevance to bank regulation. Mimicking peer capital poli-cies can have negative externalities when many banks engage (Acharya and Yorulmazer (2008), Freixas (2010), Wagner (2011), Bonfim and Kim (2012), Farhi and Tirole (2012)). Peer influence can weaken financial stability and heighten the spread of contagion. For example, systemically important banks may have an incentive to adopt similar business strategies. If they fail, this behavior increases the chance of a collective bailout. Allen et al. (2011) show that similar portfolio decisions by banks can lead to a higher systemic risk by making bank defaults more correlated. Similarly, when many banks engage in following peer capital decisions, systemic risk can increase if it leads to reducing capital buffers. While microprudential policy can address this practice at individual banks, the possible engagement of many banks simultaneously creates concern at a macroprudential level.2However, the possible impact of peers on bank capital remains untouched by any regulatory policy.
To the best of my knowledge, my empirical study is the first one providing evidence of interactions among banks and their peers having an effect on bank capital holding. The study is close to Leary and Roberts (forthcoming), but both studies differ along several dimensions. While Leary and Roberts (forthcoming) analyzes peer capital structure decisions as a determinant of firm capital structure, this study places a focus on analyzing peer effects in a regulated industry. Further, the empirical analysis takes into account the specific nature of bank business developing empirical hypotheses in line with the existing theoretical and empirical models of bank capital (Berger et al. (1995), Heider and Gropp (2010), Acharya et al. (2013)). Different factors explain bank capital structure and firm capital structure. Finally,
Leary and Roberts (forthcoming) uses firm industry as the criterion for assigning peer groups, while this study uses size and degree of diversification as within-industry criteria for peer group formation.
The paper continues as follows. The next two sections include a discussion on why peer effects could have an impact on bank capital and offer a brief review of the literature. Section 1.4 presents the data and methodology. It also describes the peer group formation, and addresses the identification of peer effects. Section 1.5 presents the estimated models, and section 1.6 contains the conclusion.
1.2 Context
There are several explanations in the literature on herding in bank business. One view explains it with managers’ concerns for a reputation, while the other with the tension between bank managers and bank shareholders (e.g. Rajan (1995), Zwiebel (1995)). Scharfstein and Stein (1990) offer a model explaining the behavior of mimicking others with the goal of attaining a good image. The incentive follows from a belief that a reputation improves by mimicking peers whose better payoffs are the result of a better business status or high financial performance. Freixas (2010) argues that bank managers can be under pressure by bank owners to follow the trends in a particular market if this will boost short-run profits. The tension can motivate managers to follow market trends even if their view is pessimistic. A long-run investment focus can lead to a more stable return on equity, but it could also decrease bank profits in comparison to competitors. Therefore, bank owners can hold managers liable for a weak performance. Meanwhile, they are less likely to be liable for a focus on short-term profits if a market-wide shock causes bank losses. In this case, bank owners and markets will be more forgiving since aggregate shocks are largely unavoidable by any bank.
In the empirical research on bank capital, there is no evidence of peer effects. Theory, however, suggests that markets can use the capital of a bank to assess how it fares with asset management and financial health (Flannery and Rangan (2008)). Bank capital can also give information about the degree of agency conflicts between shareholders and managers within the bank, or between shareholders and creditors. An ambiguity about the optimal bank capital level still prevails in the literature. However, there is a consensus that bank value decreases when capital is very little or too much (Dewatripont and Tirole (1994), Berger et al. (1995), Flannery and Rangan (2008), Acharya et al. (2013)).3 In this case, one could argue that peer bank actions can influence bank capital decisions in more than one way. Facing
3
uncertainty about the admissible long-term capital ratio of any bank, investors can use the capital of peer banks to make a guess. Deviations from the peer average capital could be a useful criterion for judging the impact of stakeholders’ conflicts on future bank prospects. The credit market - a bank’s major line of business, for example - features a high degree of asymmetric information letting investors extract limited information about bank loan quality. Thus, banks have an incentive to conform to peer discipline getting capital ratios closer to the average capital in their group. Otherwise, they may face a higher cost of debt or equity financing. This hypothesis supports the common view of banks facing two capital requirements - one by the market and one by the regulator (Berger et al. (1995), Flannery and Rangan (2008)).4 It should be also stressed that banks can adjust capital ratios without issuing equity. For example, when banks experience a period of high profits but bank management does not increase dividend payouts or share repurchases proportionally, the level of capital can increase. In a similar way, bank capital can increase when the share price appreciates.5
The importance of peer banks in the choice on bank capital is also consistent with the existing theories about bank capital structure in a frictional world. Take first the tax advantage of debt and the expected costs of financial distress, which is among the overriding departures of the Modigliani and Miller (1958) propositions. Under this assumption, the market capital "requirement" is the capital ratio at which the tax disadvantage of debt offsets the expected costs of financial distress. Bank owners prefer debt financing because of the tax deductibility of interest payments. However, higher leverage reduces capital and creditors can count on fewer funds from shareholders in case of an insolvency. With a decrease in capital, the expected costs of financial distress, such as the cost of bankruptcy and loss in bank value, rise. In turn, the capital requirement imposed by the market rises, too. Banks can anticipate that peer average capital helps the market in assessing how they fare compared with peers because of the opaqueness of bank activities. Thus, banks have incentives to adjust capital ratios considering peer bank characteristics along with bank ones. In this way, peer capital can influence the long run capital target of a bank.6
4Berger et al. (1995, p. 3) broadly define the market capital requirement as the capital ratio that maximizes the value of the bank in the absence of regulatory capital requirements, but in the presence of the rest of the regulatory structure (e.g. safety net, etc.). The market capital requirement embeds the notion of an optimal market capital ratio as a target for each bank in the long run in the absence of regulatory capital requirements.
5Flannery and Rangan (2008, p. 5) discuss in a greater detail how other bank capital adjustments can take place: “. . . For example, many BHCs sold their headquarters building in the late 1980s, booked a capital gain, and then leased it back from the purchaser. A bank can also “cherry-pick” its securities portfolio, realizing the gains on appreciated securities while postponing recognition of unrealized losses. Loan provisioning provides another (notorious) avenue for troubled banking firms to boost their book capital.”
6
Peer capital can also have a role in resolving asymmetric information problems. Agency conflicts be-tween creditors and shareholders set an upper limit on the admissible bank capital (Berger et al. (1995)). These conflicts can shift wealth from creditors to bank owners by investing in excessively risky projects with negative net present values. Creditors adjust interest rates to account for this possibility.7 Publicly available information can help creditors assess the risk of an asset portfolio, but bank balance sheets provide limited asset quality information (Merton (1977)). Thus, banks have an incentive of committing to market discipline by raising capital ratios to signal a better asset quality. Supporting evidence by Flannery and Rangan (2008) shows indeed that market discipline lowers the probability of default. It also becomes more important for stock market investors when regulatory support is withdrawn.8 While raising capital ratios signals bank soundness, the deviation from the peer level can additionally show how closely aligned are the interests of debt holders and shareholders in the bank compared to other banks.
Furthermore, agency conflicts between shareholders and managers impose a lower limit on admissi-ble capital level. Bank owners cannot oversee manager’ actions effectively but they can use leverage as a motivation for bank management to increase bank value ((Jensen and Meckling (1976), Calomiris and Kahn (1991), Diamond and Rajan (2000)). More leverage exerts a pressure on bank managers to make better investments and try to avoid bankruptcy. Deviations from peer capital can signal how a bank fares in resolving such conflicts compared to other banks. Peer capital therefore can serve as a benchmark for the lower limit of admissible capital levels to shareholders and managers.
1.3 Related literature
This paper adds to the literature on capital structure, bank capital, and social interactions among eco-nomic agents. Gropp and Heider (2010) find similarities in the capital structures of large non-financial firms and banks suggesting they have common determinants. The paper also challenges the common assumption of bank regulation as the overriding departure from the Modigliani-Miller (1958) irrelevance propositions. Its results are in line with model predictions that capital requirements are not necessarily binding (Flannery (1994), Myers and Rajan (1998), Diamond and Rajan (2000), Allen et al. (2011)).
The literature has aimed to explain why banks have positive levels of capital. Some studies point to precautionary motives to avoid liquidity shortages ((Blum and Hellwig (1995), Ayuso et al. (2004), Barth et al. (2005), Bolton and Freixas (2006), Peura and Keppo (2006), Peura and Keppo (2006), Van den
7
See Jensen and Meckling (1976) for general results and Acharya et al. (2012) for recent empirical evidence about risk shifting in the case of banks during the global financial crisis.
8
Heuvel (2008), Berger et al. (2008), Brewer et al. (2008)). Other studies stress on the positive impact of bank competition on capital buildups (for example, Barrios and Blanco (2003), Cihak and Schaeck (2008)). Mehran and Thakor (2009) and Allen et al. (2011) explain positive capital levels by market discipline coming from the asset side of the balance sheet. Flannery and Rangan (2008) identify either regulatory pressure, periods of unusually high profitability, or market discipline as determinants of the capital buildups at US banks.
There are a growing number of studies on the dynamic behavior of bank capital as well. These studies assume that banks, like any other firms, maintain an optimal target ratio for capital in a world without frictions. In the presence of frictions, shocks create deviations from the desirable target levels, and banks face adjustment costs to reach their desired state of capital holdings. Flannery and Rangan (2006) use a partial-adjustment model of firm leverage to show that firms have target capital structures. They also find that the typical firm closes about one-third of the gap between its actual and its target debt ratios each year. Most studies in banking show that banks close to the regulatory minimum requirements exhibit a faster adjustment (Jacques and Nigro (1997), Rime (2001), Berger et al. (2008), Memmel and Raupach (2010), Jokipii and Milne (2011)). De Jonghe and Oztekin (2012) expand this analysis in a thorough cross-country study to show that a bank’s ability to adjust its capital ratio is influenced by macroeconomic conditions, herding behavior and bank regulatory and institutional characteristics. They find that banks can make faster capital structure adjustments in countries with better supervision, stricter capital requirements, more developed capital markets and high inflation. The results suggest that in times of crises, banks adjust their capital structure significantly faster, while in normal times, adjustment is substantially slower if banks must increase their capital ratio.
or performance. Zwiebel (1995) and a survey among CEOs by Graham and Harvey (2001) support these findings.9 There is also evidence of how social interactions can lead to financial runs. In the models of Diamond and Dybvig (1983) and Bernardo and Welch (2004), an action by an agent can affect the payoffs to other agents by the action itself. In a bank run, for example, a payoff externality arises since any depositor will have an incentive to withdraw deposits if she expects others to do it as well.
There are also models on bank herding. Acharya and Yorulmazer (2008) show that profit maximizing bank owners have an incentive to herd with other banks because of the risk of information contagion. Bad news about a given bank such as increases in loan-loss reserves can exert an impact on the cost of borrowing of other banks when their loan portfolios share a common factor. Bank herding by investing in correlated investments offers a way to minimize the impact of bad news about other banks on the cost of borrowing and future profits. Rajan (1995) motivates herding among banks by short-termism and reputational concerns. Bank managers in the model are rational but aim to maximize short-run profits. They are also concerned about the market view of their abilities. Managers anticipate that the stock market is more forgiving for bank losses during a banking crisis and the consequences for a bank reputation are milder when all banks perform poorly. Therefore, it is optimal for managers to coordinate jointly on investment decisions and herd by lending to negative NPV projects. Recently, Wagner (2011) and Farhi and Tirole (2012) also point to the bank benefits of adopting the same risky strategies as other banks. Banks anticipate that authorities have a little choice but to intervene if their collective actions lead to a financial crisis.
The empirical evidence on bank peer effects and herding is scarce. A focus of previous research is on bank business activities and risk management.10 Jain and Gupta (1987), Barron and Valev (2000) and Uchida and Nakagawa (2007) find evidence of herding among banks in the US and Japanese loan markets. Recently, Bonfim and Kim (2012) provide evidence of herding in managing liquidity risks among the largest banks in the US and Europe. Jain and Gupta (1987)’s study uses data from the US to test if bank herding was a main trigger of the debt crisis between 1982 and 1984. Many loans to governments of developing countries precede this crisis and anecdotal evidence suggests that smaller banks may have followed larger peer banks in making the same decisions of lending to certain countries.
9
In their survey, Graham and Harvey (2001) report that about one third of CEOs consider the behavior of competitors to be an important factor in financial decision making.
10
Jain and Gupta (1987) find weak evidence of herding as a cause of the crisis. Barron and Valev (2000) use a theoretical model to examine if it is optimal for investors to wait before investing until they receive information about the investments of others. According to the model, less wealthy investors gain the least from using only current information, so it is optimal to wait until the wealthier make a first move. Using data for the 1982-1984 debt crisis, Barron and Valev (2000) find empirical support for the tendency of smaller banks (followers) to follow larger banks (leaders) in granting credit to foreign governments. Uchida and Nakagawa (2007) give evidence of herding in lending among Japanese banks that can explain the increasing number of nonperforming loans after the bust of the asset bubble in the early 1990s. The paper also shows that herding varies in degree with the highest peak during the bubble period in 1987.
Banks can also herd in areas such as risk management. They can engage in risk taking strategies jointly instead of optimizing liquidity risks individually because of a higher chance of a collective bailout. Bonfim and Kim (2012) find such evidence for the largest banks in the US and Europe. The authors argue that this herding could have a negative impact on the banking system by increasing systemic risk. While liquidity risks are addressed usually at a microprudential level, the collective actions by banks change it to a macroprudential one. By herding, banks obviously ignore the socially optimal level. Further, Ratnovski (2009) shows that banks choose suboptimal liquidity in risk management by herding as long as they expect that other banks will do the same. In this way, herding contributes to a new component of liquidity risk that is currently overseen by regulation.
1.4 Data and methodology 1.4.1 Methodology
The model follows Leary and Roberts (forthcoming) in taking a general form as:
yijt= 0+ y ijt 1+ 0X ijt 1+ 0Xijt 1+ 0 j+ 0ci+ 0ct+ uijt: (1)
for a bank i in peer group j at time t. ci, ctand jdenote respectively bank level fixed effects, time fixed
effects, and peer fixed effects. The model assumes correlation in the residuals uijtof a given bank. Due
to the presence of heteroscedasticity, the model uses robust standard errors clustered at the bank level. The simultaneity between yijtand y ijt 1is known as the reflection problem (Manski (1993)). Section
1.4.4 describes in details the instrument variable and its use for resolving the reflection bias.
previous studies (Gropp and Heider (2010)).11 The vector yijt includes the book capital ratio and the
market capital ratio as alternative measures of bank financial policy. The book capital ratio is of relevance to regulators since there are capital requirements on book capital only. Capital market participants use the market capital ratio to assess the prospects of the bank including the probability of default and the expected costs of financial distress. As a result, bank managers have to manage jointly the levels of book and market capital so they can comply with both requirements.12 On the left-hand side, I also include either the bank Tier 1 capital ratio as a proxy of risk-weighted capital, or nondeposit liabilities as a measure of bank debt composition.
Every peer variable is the average value among all banks in a quarter within a peer group excluding bank i ’s observation. y ijtis the measure of bank peers’ financial policies defined as the average effect
of a peer group j’s financial decisions on bank i in period t. Averaging of the variables aggregates the most relevant information in peer characteristics. By that, the possible impact of nonlinearities in peer interactions on the empirical results is minimized. The channels by which peers can exert an impact on bank financial policy include either peer capital levels or balance sheet characteristics, or both. For example, it could be the profits of bank A or its bank capital that can influence the financial policy of bank B. Therefore, the set of control variables includes measures of bank profitability, market-to-book ratio, collateral, and bank size for an individual bank (Xijt 1) , and those of bank peers (X ijt 1). A
dummy variable controls for dividend payouts. It takes the value of one if a bank has paid a dividend in a previous quarter, and zero otherwise. The model also includes the ratio of loan loss provisions to total assets as a proxy for asset risk. Appendix A lists definitions for the key variables in the empirical analysis.
1.4.2 Data
The data come from two sources. Bank balance sheet data for publicly listed US commercial banks are obtained from the Standard & Poor’s (S&P) Compustat database. Next, the dataset is merged with the stock price database of the Center for Research in Security Prices (CRSP). The data for financial firms other than commercial banks are filtered from the combined dataset. All variable distributions are winsorized two-side at the one percentile level to minimize the possible impact of outliers. Any
11
Gropp and Heider (2010) find evidence of similarities in the set of cross-sectional, time-varying determinants of the capital structure of non-financial firms and large, publicly listed banks.
12
missing observations for all variables in the model are also removed. Additionally, banks are classified as either diversified or regional according to the Global Industry Classification Standard system (GICS). Appendix B lists the GICS criteria that determine if a bank is diversified or regional according to the focus of its business activities. The final sample contains 36; 741 bank-quarter observations for 909 publicly listed banks during the period 1971 to 2010. Capital regulation is introduced in the US in the 90’s, so estimations on regulatory capital use observations from a subsample of the data. The subsample contains
22; 233 bank-quarter observations for 805 banks between 1990 and 2010. To assure consistency, the
estimations of the empirical models are performed on the full sample of the data and on the subsample. In both cases, the regressions yield results that are qualitatively similar.
Table 1 lists summary statistics for the variables in the empirical model. A typical bank in the cross-section has a book capital ratio of 8 percent and market capital ratio of 12 percent. Out of total debt, nondeposit liabilities make up a share of only 15 percent. Banks keep a buffer of excess capital that is on average 6 percent above the permissible level. They are frequent dividend payers and the payout takes place in 83 percent of all cases. Besides, the typical bank in the sample gives out 30 percent of its net income as a dividend.
In addition, Table 1 documents wide variations in bank size and other characteristics in the cross sec-tion. The market capital ratio is between 1.2 percent and 30 percent. Bank holdings of Tier 1 regulatory capital vary between 5.6 percent and 25.1 percent. The variation across banks in the holdings of nonde-posit liabilities is large as well. While a bank has on average about 15 percent of nondenonde-posit liabilities, nondebt holdings vary between zero and 43 percent among banks. The largest difference across banks is in size. On average, a bank has assets of about 9 billion. The largest one in the dataset has more than 1300 billion dollars in a given quarter, and the smallest one, about 10 million.
1.4.3 Which are the peers of banks?
The peer group definition is in line with the criteria of the Federal Financial Institutions Examination Council in the United States (FFIEC) distinguishing among banks according to asset size.13 Established in 1979, the FFIEC main task is to develop and apply uniform standards, and report forms for the federal examination of financial institutions. It is an interagency developing those standards on bank analysis for the Board of Governors of the Federal Reserve System (FRB), the Federal Deposit Insurance Cor-poration (FDIC), the National Credit Union Administration (NCUA), the Office of the Comptroller of the Currency (OCC), and others. The agency reports publicly available data on bank peer groups once their call reports become known. Market investors with interest in banks have access to this data and information about the bank peer groups’ definitions. Every bank has also access to the information about its peer group and its peers.
The use of bank size and degree of diversification as peer criteria follows predictions by theory and empirical studies such as Jain and Gupta (1987). Theory predicts that banks become larger and more diversified in the long run (Diamond (1984), Strahan (2008)). Larger banks are of systemic importance and face a higher probability of a (collective) bailout during a financial crisis. Similarly, larger and more diversified banks tend to fare better during a financial crisis relative to smaller and regional banks. Therefore, size and the degree of diversification exert an impact on the probability of a bank default; which further impacts the cost of bank capital and debt.
The within-group variation according to bank size also assures that the orthogonality condition be-tween the instrument and any bank specific peer average of the right-hand side variables in the first stage of 2SLS regressions is satisfied. At this stage, linear dependence can arise because of the joint presence of peer average variables and time fixed effects.
When forming peer groups, banks are sorted first as diversified or regional. Next, banks are ranked within each peer group according to size and then banks are split into subgroups according to the quartiles of the asset distribution. In total, a bank can be in one of eight groups. Peer groups are newly formed in every time period so a bank can change a peer group from one period to the other, for example because of a bank merger or acquisition. On average, a bank in the sample has 40 peers in any time period, while the number of peers in a given quarter varies between 2 and 159. Table C1 in Appendix C provides statistics on the dispersion in bank capital within the peer groups. The average standard deviation of the book capital ratio in the peer groups is about 1.5 percent, and 4.5 percent for the market capital ratio.
13
In addition, the Panels of Table C1 show that regional banks outnumber diversified banks across groups. Further, there is less dispersion of book capital compared to market capital within and across peer groups. The standard deviation of book capital increases with the quartiles of the asset distribution, while there is a varying trend for market capital.
1.4.4 Identification
The key empirical challenge in any analysis of peer effects is about identification. Banks can simulta-neously influence one another. Endogeneity between the dependent variable yijt and the peer variable
y ijt 1 causes a bias broadly known as the "reflection problem" (Manski (1993, 2000)). The bias re-quires a proper instrument to isolate the peer group impact on any member from the individual impact of a member on the entire group. Leary and Roberts (forthcoming) use the idiosyncratic component of stock returns when analyzing the impact of peer corporate firms on the capital structure of firms.14 Peer equity shocks can be a valid instrument because it is unique for every firm. It is unshared with a firm excluded from the group and uncorrelated with its characteristics.
I use the same instrument in the analysis of peer effects among banks. I estimate average monthly idiosyncratic returns as the difference between realized and expected returns estimated by a Fama-French (1993) factor-pricing model. The model conditions bank monthly total return Rit on a standard set of
factor loadings:
Rit= 0+ Mit (RMt RFt) + SM Bit SM Bt+ HM Lit HM Lt+ uit: (2)
The set of factors includes the excess market return on the broad market index ( RMt RFt), the
small minus big portfolio return ( SM Bt), and the high minus low portfolio return (HM Lt).15
I run rolling regressions over a window that requires at least 24 time periods to generate time varying betas for every bank in each period t. After estimating predicted returns, the idiosyncratic stock return for a bank i in period t is extracted as:
b
uit = Rit Rcit: (3)
14Leary and Roberts (2013) study the instrument properties and find evidence that the exclusion condition holds and the relevance condition is satisfied.
15
I annualize the equity shock by a factor of 12 to come to the average annual return.16 In regressions, the equity shock is taken in a second lag. Table 1 shows that the equity shock as a residual of a regression is on average only close to zero. It is expected since the estimate is a conditional average.
The exclusion condition may not be necessarily satisfied because it implies that the instrument affects the capital of bank i only via its effect on the capital levels of peer banks. Since banks are subject to systemic risk in contrast to non-financial firms, a liquidity shock to a systemically important bank can easily span over other banks via contagion triggering a major repercussion in the banking system. A shared unobserved common factor among banks can be eliminated by including time fixed effects. However, financial systems are marked with a high degree of interdependence (Allen and Babus (2009)). It is likely that bank stock returns can have a common systemic determinant linked to their asset portfolios as well (Acharya and Yorulmazer (2008)). If this common factor is not eliminated by constructing bank idiosyncratic stock returns, the exclusion condition can be violated. The instrument will directly affect not only peer banks but also bank i.
Table 2 presents results from a test of validity for the instrument. The peer average equity shock is fit on the set of control variables used in regressions in order to gauge the degree of statistical correlation between the covariates and the instrument. In Column 1, Table 2, the instrument is regressed on the set of control variables used in the baseline model. None of the control variables except for dividends and asset risk is found to contain information about the peer average of the idiosyncratic stock return. However, the coefficient estimates of those variables are of a very low economic significance despite being statistically significant. A standard deviation increase in asset risk, for example, is associated with an increase in the peer average of idiosyncratic stock returns by about half a basis point. Finally, there is no information that the instrument carries any information about future values of control variables in Column 2 except for the asset risk variable. However, in comparison to its estimate in column 1, it further decreases in statistical and economic significance. Overall, the results in Table 2 suggest that the peer average idiosyncratic stock return passes the test of a weak instrument and can be used in IV analysis.17 The feature of opaqueness in bank balance sheets can further stress the usefulness of peer equity shocks
16The use of compounding instead of annualizing yields qualitatively similar results as well.
as an instrument. Unlike the case of corporate firms, it is difficult for the market to assess the risk profile of every bank from its balance sheet. Therefore, the equity shocks to peers are likely to contain less information about any bank excluded from the peer group.
1.5 Results
1.5.1 A basic model of bank capital structure
The results in Table 3 offer insights about the determinants of bank capital for publicly listed US commer-cial banks. The capital ratios are fit on a standard set of bank characteristics used in previous empirical research (e.g. Gropp and Heider (2010)). These results also offer a test of the validity of the Modigliani-Miller (1958) propositions.18 There is a view in the banking literature that deposit insurance is the main overriding departure of the Modigliani-Miller (1958) propositions about the irrelevance of capital struc-ture in the case for banks. In this case, the control variables for bank characteristics should enter the model with either statistically or economically insignificant coefficient estimates. Otherwise, it can be concluded that deposit insurance may not be the first-order determinant of bank capital structure relative to other determinants.19 Appendix D summarizes expectations about the sign on control variables that enter the models in Table 3. Following Gropp and Heider (2010), their predicted effects differ under two alternative views of bank capital structure. The first column in the table in Appendix D shows the expected signs of the control variables based on documented evidence for non-financial firms in the exist-ing corporate finance literature, or the so called market view (e.g. Rajan and Zexist-ingales (1995), Frank and Goyal (2004)). The second column lists expectations under an alternative view of banks as risk averse firms aiming to build a reserve of funds so they can prevent a raise in equity on short notice.20 Under the buffer view, banks that have either lower market-to-book ratios, pay less dividends, or are less profitable have higher book or market capital ratios in consequent periods. Besides, banks increase the share of nondeposit liabilities when they face a lower cost of raising debt under both views.
The first column in Table 3 presents the estimated model on bank book capital ratio and its determi-nants. The control variables that enter regressions except for the categorical dummies are scaled to have a mean with the value of zero and a standard deviation with the value of one so their economic effect
18
The test is possible since the capital ratio is an inverse function of bank leverage, i.e., the ratio of bank debt to total assets. 19
See Gropp and Heider (2010) for a more thorough discussion on this point. One can argue that the determinants of bank capital structure correspond to the same determinants of non-financial firms with the only difference that they reflect the unique nature of banks. If regulation is the overriding departure of the Modigliani and Miller (1958) propositions, one should expect these control variables to have no explanatory power.
20
on the dependent variable can be easily compared across table columns. All explanatory variables are standardized to compare their economic effect on the dependent variable in a percentage change.21 The market-to-book ratio enters the model with insignificant coefficient estimate. It suggests that the growth opportunities of publicly listed commercial banks have no impact on the book levels of bank capital. The results show that larger banks tend to keep a higher leverage, and tend to rely less on equity capital. Similarly, increases in collateral levels have a negative effect on the book capital ratio. Holding all other factors constant, the book capital also increases with bank profits. It is positively associated with regular dividend payouts, but negatively associated with changes in asset risk, as proxied by loan loss provisions. An increase in the asset portfolio risk by one standard deviation leads to a decrease of 0.06 percent in the quarterly bank capital. Column 2 in Table 3 presents estimates of the model on the market capital ratio. The results are similar to the ones in the model on the book capital ratio. In addition, the coefficient estimate of the market-to-book ratio enters statistically significant at the one percent level. Holding other factors constant, a standard deviation increase in the market-to-book ratio leads to an increase by 3.7 percent in the market capital ratio.
The standard determinants of bank capital structure can also explain the relative composition of bank debt. Column 3, Table 3 lists estimates of a model about the bank holdings of nondeposit liabilities. In contrast to non-financial firms, bank liabilities include deposits as well as short-term or long-term debt. Because larger banks can access easier debt financing at a lower cost, results show that those banks tend to have a higher share of nondeposit liabilities. Additionally, more profitable banks and banks with lower market-to-book ratios tend to increase their nondeposit holdings in consequent quarters, ceteris paribus. Further, neither asset risk, nor the regular payment of dividends, is found to be a significant determinant of the relative composition of bank debt. The model on the Tier 1 regulatory capital is presented in the last column of Table 3. More profitable banks, regular dividend payers or banks with higher levels of collateral are found to increase the level of regulatory capital while it tends to decrease at larger banks. Also, the market-to-book ratio and asset risk are found to have no impact on the regulatory capital.
The model on the market capital ratio has the best fit among the models in Table 3 with an adjusted R-squared of about 0.80, while the model on regulatory capital has the lowest explanatory power. In addition, the estimates in columns 1 and 2 have a greater match with the market view on leverage rather than with the buffer view. These results are similar to the findings of Gropp and Heider (2010) for the largest US and EU banks. The current study, however, uses the sample of all publicly listed banks in the US and covers a much larger sample period. There is also support to the findings by Gropp and Heider
21
(2010) on the secondary importance of regulation as the overriding departure from the Modigliani-Miller (1958) proposition. While most control variables in the models of book and market capital in Table 3 enter with statistically significant coefficient estimates, there are large differences in economic signifi-cance. A standard deviation increase in profits, for example, is associated with a 0.06 percent increase in the book capital ratio, ceteris paribus, while one standard deviation of collateral decreases the capital ratio by 0.60 percent. Additionally, a standard deviation in log (size) leads to a decrease in the book ratio about 0.39 percent. The difference in economic significance points out that bank regulation may not be the only overriding departure for the case of banks.
1.5.2 A reduced-form peer model
Panel A of Table 4 shows the estimates of the baseline peer model in its reduced form. Each column corresponds to a model that aims to explain a different measure of bank capital or bank debt composi-tion. Across columns, the variable in use is indicated on the top of each column. Other estimates from regressions such as the F-statistics, the number of observations, or the number of banks are given in the bottom of each column. The F-statistics corresponds to the Cragg-Donald Wald test statistic for a weak instrument with its range of critical values being estimated by Stock and Yogo (2005). The table also reports the adjusted R-squared for brevity though it provides no useful information about the fit of a 2SLS model (see Wooldridge (2010)). The regressions are run under the assumption that the dependent variable in each regression is explained by a set of bank level characteristics and a set of characteristics of bank peers. The empirical model also includes panel fixed effects and time fixed effects. The time dummy variables control for unspecified macroeconomic and financial market factors with an impact on bank capital, while bank fixed effects have been shown to be important in previous studies (Flannery and Rangan (2006), Lemmon, Roberts and Zender (2008), Gropp and Heider (2010)). The definitions for explained and explanatory variables can be found in appendix A.
These results are not robust yet to endogeneity such as unobserved heterogeneous factors affecting all banks in the peer group. However, despite the presence of the reflection problem, they can give a first insight on the effect of peers on bank financing decisions. In Panel A of Table 4, the peer average level of the RHS variables is listed first followed by the bank-level characteristics and the characteristics of peer banks.
a positive effect on the bank capital ratios. A standard deviation increase in its value is associated on average with about 0.47 percent increase in the book capital, and about 1.5 percent increase in the market capital level. Similarly, the relative debt composition of peer banks as well as the peer average regulatory capital exert a positive impact on bank choices on nondeposit financing and the build-up of regulatory capital. A standard deviation increase in the nondeposit liabilities of peer banks leads to a 0.40 percent increase in the nondeposit debt of a representative bank in the population, and 0.18 percent in bank regulatory capital.
Panel A of Table 4 shows that average size and profits of peers can influence bank financing decisions, too. Banks tend to keep more capital when peer banks are larger, while peer profits tend to have a negative impact on capital levels. This finding could indicate that peer pressure can exert a positive impact on bank managers to maximize the return on bank assets. Banks with more profitable peers could raise bank profitability by using more capital to invest in high-returns projects. Further, banks operating in an environment with smaller or less profitable peers is associated with a decrease in the share of nondeposit bank liabilities. Because creditors determine the cost of bank debt by comparing how a bank fares relative to its peers, this finding is consistent with the idea that smaller and less profitable banks are less likely to obtain cheaper debt funding. Finally, column 4 in Panel A of Table 4 shows that the bank Tier 1 capital ratio is negatively associated with the profitability of bank peers. This finding is again consistent with the hypothesis that banks seem to deplete regulatory capital when facing fiercer competition. As a robustness test, the lagged dependent variable is also included in the model to check if the impact of peers retains statistical significance. In Panel B of Table 4, estimations yield a statistically significant coefficient estimate of the peer book average variable, thus suggesting the absence of a lagged dependent variable qualitative effect on the peer average. Its economic effect is tenfold smaller compared to the estimate of the lagged book ratio. In addition, the coefficient estimate of the market capital peer variable becomes statistically insignificant. It should be noted that the model in this form can only offer a basic insight. The model with the lagged dependent variable is misspecified because the error term is correlated with the dependent variable. As a result, the regression model can no longer yield consistent estimators for the regression parameters.
1.5.3 IV Analysis 1.5.3.1 IV Regressions
share of nondeposit liabilities. The first stage regressions are not reported except for the coefficient estimates obtained for the instrument, and the estimate of the F-statistics of the Cragg-Donald Wald test in the bottom of the table. The F-statistics across all models in columns 1-5 are well above the minimum threshold value of 10, so they pass the weak instrument test (Stock and Yogo (2005)).
The first stage regression estimates indicate that on average banks tend to increase the holdings of capital in response to uncertainty as being embedded by equity shocks to their stock. The results reported on the bottom of Panel A of Table 5 show that the average equity shock of peer banks is found to have a positive and statistically significant effect on the average book and market capital ratios of peer banks, and a negative one on the peer average holdings of nondeposit liabilities. The coefficient estimates are also economically significant. One standard deviation in peer banks’ average equity shock is associated with a 0.53 percent increase in the peer banks’ average book capital ratio, and about 3.4 percent increase in the peer average market ratio. Equity shocks seem to increase the cost of nondeposit funding to banks so consequently they rely more on deposit funds, and less so on nondeposit liabilities. The corresponding quarterly decrease in nondeposit liabilities is by 0.43 percent. Similarly, banks also tend to increase the risk adjusted capital. A standard deviation increase in the average equity shock leads to about 1.30 percent increase in the Tier 1 holdings of peer banks.
The second stage results in Panel A of Table 5 show that peer effects are an economically significant determinant of bank capital. In response to a standard deviation increase in peer banks’ average book and market capital ratios, the book capital ratio increases by about 0.77 percent (Column 1), and the market capital ratio by about 1.78 percent (Column 2), ceteris paribus. Similarly, a given bank actively increases its risk-adjusted capital ratio by approximately 0.23 percent for a standard deviation increase in the risk-adjusted ratios of peer banks (Column 4). Banks also actively adjust their debt composition, in particular their reliance on nondeposit funding in response to peers.
Table 5 also shows that the financial policy of peer banks is of high relevance to bank financial policy. Across all columns, the average peer ratio is among the most important determinants of book and market capital ratios. In comparison, other bank capital determinants are of lower economic significance. The coefficient estimates of bank-specific control variables have the expected signs shown in previous studies (Flannery and Rangan (2008), Gropp and Heider (2010)). Profits have a positive effect on bank capital, while larger banks tend to have less equity. Most of these estimates also retain significance in the 2SLS estimation alike in the OLS regressions in Table 3; therefore, the instrument adds sufficient exogenous variation to the empirical model.
highlight the importance for a bank on how it fares in comparison to rival banks. Peer size and prof-itability seem to be most important to a bank for capital decisions. A standard deviation increase in the average profits of bank peers is associated with a decrease in the quarterly book ratio by about 0.045 percent, and about 0.06 percent in bank regulatory capital. These findings suggest that banks exhaust capital levels for new investments when peers are more profitable. Further, the results suggest that the bank capital is used to compensate for a smaller size relative to larger peer banks in assessing the market cost of bank equity. Since bank size is also a measure of systemic importance, this finding is at edge with the key goal of regulation that aims at having risk-adjusted capital increase with bank size independent of the peers’ total assets. Finally, a bank of any size has higher non-deposit debt when its peer competitors are smaller in size but not when they are larger.
Peer pressure could take the form of deviations from the peer average. Banks have an incentive not to deviate much of the peer level, otherwise they may face higher costs of equity or debt. Moreover, peer pressure could be asymmetric. Banks below the peer average can differ in the speed of "catching up with the Joneses" compared to banks above the peer average level. For this analysis, I create two data subsamples. I estimate the deviation of bank book capital from the peer average sorting banks as being either above the average peer level, or banks that are below the peer average level in every time period. In a following step, I run the baseline empirical model. Results are reported in Panels B & C of Table 5. There is strong evidence that banks below the peer average level tend to increase book capital consequently to the average level. In contrast, banks above the peer average level tend to increase book capital but not as much. The coefficient estimate is statistically significant at the 10 percent level only. The economic effect is twice smaller compared to the one observed in the sample below the peer average. Therefore, there is evidence of asymmetry in bank responses to peer average levels. Results are similar for market capital and non-deposit liabilities. For example, the coefficient estimate of the peer average is statistically insignificant in the subsample of banks above the peer average market capital, but economically and statistically significant at the one percent level for the other subsample. The economic effects in the models of Tier 1 regulatory capital are of a similar magnitude.
results in Panel B. In addition, shareholder-manager conflicts give a strong incentive to banks in adjusting capital upward to the peer average level according to Panel C. This outcome can mean that banks do not use leverage as a motivation for managers to increase bank value. Rather than that, it implies that high leverage can signal a tension between shareholders and managers questioning the effective monitoring of bank business. It should be noted that the economic effect of regulatory capital is of a similar magnitude in both tables, thus there is no evidence of an asymmetry in bank responses in the model.
In Panels D and E of Table 5, the baseline model is subject to robustness tests. Placebo tests replacing the peer average book and market capital values with the average capital values across randomly assigned peer groups tend to show that peer groups in the empirical analysis are well-defined. The analysis includes a simulation of 240 times of the model to find the simulated distribution of book capital and market capital. The F-statistics of the estimations reported in Panel D of Table 5 are on average below the value of 10. Thus, results do not pass the weak validity test for the instrument in majority and provide support that the peer groups are well defined.
Panel E of Table 5 addresses concerns with identification coming from the construction of the instru-ment. There are several assumptions for addressing adequately the identification problem. The results from Table 2 in the paper show that the instrument carries no significant current and future information about bank characteristics. The first-stage F statistics in the 2SLS analysis help judge whether the in-strument passes the weak validity tests. Further, the use of time fixed effects controls for unobserved common factors in any given period of time.
An omitted common factor in the Fama-French model however can have an impact on the estimated models. Panel E of Table 5 reports estimated models after including an additional factor loading in the model and requiring at least 36 time periods of available data. The unweighted average return on the banking sector excluding bank i in a given period t, BN Kit (R it RFt) is considered to absorb any
remaining commonality between bank stock returns. The estimated models in Panel E yield qualitatively similar results. The analysis gives weak evidence of omitted variables as a determinant of the estimated effects.
1.5.3.2 Peer effects and regulatory minimum
for those banks increases since the market and the regulator anticipate that the probability of default, and by that the expected costs of financial distress, get higher, too.
Two dummy variables are constructed therefore and interacted with the variables of interest. The dummy Close takes the value of one if a bank has Tier 1 capital that is below 6.5 percent, and zero otherwise. The second dummy variable, Close 80s, defines if a bank is close to the stock market "re-quirement." It takes the value of one if the bank is below 7 percent in holdings of primary capital (i.e. at most 1.5 percent above the minimum of 5.5 percent), and zero otherwise.22 Estimations are carried out on a subsample of the dataset starting in 1991 when Tier 1 capital requirements were introduced in the US. The estimates in column 1 of Table 6 give no evidence that the peer average book capital level has any effect on the book value of equity when a bank approaches the minimum capital threshold. The results in Column 2, however, show that those banks still actively manage their market capital ratios. Ceteris paribus, a standard deviation increase in the average market equity of peers leads to about 0.82 percent increase in the bank market capital ratio. Finally, Panel C shows that the variables of interest retain statistical significance when estimation is applied on the same subsample but using the baseline model from Table 5 without interactions with the categorical variables.
The findings suggest that adjusting book and market capital ratios by mimicking other banks is of different importance to banks when they are close to crossing the regulatory and market thresholds. On the one hand, supervisors in the US use book values to judge the financial condition of a given bank because of the need to apply uniform criteria for publicly traded and privately held banks (Berger et al. (1995), Flannery and Rangan (2008)). On the other hand, stock market investors determine the cost of equity using market equity, and therefore banks are subject to two constraints. When being close to exhausting bank capital, the results show it is still important for banks to adjust the market capital ratio so it does not deviate much from the peer level. In contrast, adjusting the book capital ratio to the peer level tends to be of secondary importance.
1.5.3.3 Peer Effects and Changes in the Banking Environment
The results in Panel A of Table 5 and Table 6 show that peer capital decisions influence bank choice of capital and non-deposit debt. We should also expect mimicking of peer capital to differ across periods
of crises, changes in market structure, or recessions. All these events can have an impact on bank performance and competition as well as on peer interactions. The models in Table 7 analyze if indeed mimicking of peers vary significantly with environmental changes along these dimensions.
In Panel A of Table 7, the peer average capital variable in the estimated models is additionally interacted with a categorical variable controlling for the event of a bank crisis or a market crisis occurring in the US. The information on crises comes from Berger and Bouwman (2010). The bank crisis periods include the credit crunch from early 1990s and the subprime mortgage banking crisis from 2007 to 2009. There are also three market-related crises. The first one was in 1987 when the stock market crashed, while the second one was in 1998 when the Russian government defaulted on its debt. The third market crisis occurred in 2001 with the dot.com bubble burst and September 11 terrorist attack. Appendix A provides definitions and sources of all event variables.
The results in Panel A give limited evidence that peer effects vary differently in crisis and non-crisis periods. Only the models of market capital and Tier 1 capital yield first stage F-statistics estimates above the value of 10, thus passing the instrument validity test. Among models, only the coefficient estimate of peer market capital enters statistically significant (Column 2 of Panel A). This lets us conclude that peers’ impact on bank market capital is much lower during crisis periods than non-crisis ones. The findings are consistent with the idea of banks preferring to focus in the first place on stabilizing their balance sheets and capital positions in bad times. In Panels B and C, the estimated models make a further distinction between stock market or banking crisis periods. The results from these models give no evidence that the impact of peers is significantly different in stock crisis periods compared to other times. However, peer banks continue exerting an impact on bank market capital and nondeposit liabilities during a banking crisis. Yet, their impact is of a lower degree letting us conclude that mimicking becomes less important when the banking sector is hit by aggregate negative shocks.
peers. Thus, we expect to find that the effect of peers on bank financing decisions vary with the degree of competition. The results in Panel A show that intrastate branch deregulation is found to have a positive impact on the book and market capital of banks compared to the periods when branch restrictions are still in place. Further, the impact of intrastate deregulation on regulatory capital and the holdings of nondeposit liabilities is a positive one as well. Peer capital seems also to have a differential impact on bank book and market capital in recessions. According to the estimates in Panel B of Table 8, peer capital has a positive impact on book capital during economic slowdowns but a negative one on market capital. This finding is consistent with the conjecture that book capital is a relevant measure of bank soundness during recessions for the stock market.
In Panel A of Table 9, the peer average variables are interacted with a dummy variable controlling for the passing of the Gramm–Leach–Bliley Act (GLB) in 1999. The GLB Act repeals parts of the Glass-Steagall Act from 1933 not allowing any financial institution to have activities in the areas of banking, insurance and investment banking. The results from these models show that the enactment of the GLB Act had no impact on book capital adjustments in response to peer banks. However, peer effects in market capital and non-deposit liabilities are found to be not as strong following the enactment. In Panel B of Table 9, the peer variables are also interacted with a dummy controlling for the interstate branch deregulation that came into effect in 1994 with the Riegle-Neal Interstate Banking and Branching Efficiency Act. The results from these models show that the peer effects on book capital become stronger, and the peer effects on market capital become weaker in the period after interstate deregulation. However, the event has had no impact on the mimicking of peers in the area of Tier 1 regulatory capital and non-deposit liabilities. It could be argued that both laws have removed the constraints to banks in terms of geographic and business-type of diversification, which seems to be of relevance to stock markets. The removal of constraints among those dimensions seems to have decreased the incentives of banks to adjust market capital levels to the average level of their peers.
1.5.3.4 Leaders vs. Followers
The theory on herding suggests that managers tend to adopt the business strategies of their more successful peers. Success can be measured by a better business image that results from a history of higher payoffs (Scharfstein and Stein (1990)). To analyze if there are leader banks in the sample that are followed by the rest in making capital choices, the model is extended to differentiate between banks with a leading position in the market for deposits or loans, and other banks.23 More precisely, the deposit and loan shares of all banks in a given peer group are estimated in the first step. Second, the banks with
