How does bank competition affect their
stability?
An analysis of the effects of the Riegle-Neal Act of 1994
1 July 2018
Abstract
The Riegle-Neal Act of 1994 removed all restrictions on interstate banking in the US. The adoption of this law, in combination with the gradual fall in branching restrictions preceding it, provide a fitting context to investigate the debated relationship between bank competition and risk. Traditional theory proposes that competition incentivises banks to increase the riskiness of their portfolios in order to offset expected losses. More recent literature provides arguments for an opposite mechanic, wherein banks in more competitive environments charge lower rents to their customers, indirectly lowering their default risk, which would increase their likelihood of repaying their debts. This thesis contributes to the debate by estimating the Boone indicator for competition for a large sample of individual US banks, over the period of 1984-2006. The results suggest that competition is conducive to fragility in the banking sector, which provides evidence for the traditional view.
Teodor Todercan
Supervisor: Razvan Vlahu
10860959
Faculty of Economics and Business
This document is written by Teodor Todercan, 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.
Contents
1 Introduction 2
2 Theoretical Framework 3
2.1 The US banking system . . . 4
2.2 Debate . . . 5
2.3 Measuring competition . . . 6
2.3.1 The Herfindahl-Hirschman index . . . 6
2.3.2 The Lerner index . . . 7
2.3.3 The H-statistic . . . 8
2.3.4 The Boone indicator . . . 8
2.4 The effects of the RNA . . . 9
3 Methodology and data 10 3.1 Estimation of the Boone indicator . . . 11
3.2 IV analysis for estimating the Boone indicator . . . 12
3.3 Primary analysis . . . 13
3.4 Data . . . 14
4 Results 16
5 Robustness Tests 18
1
Introduction
Banks in the US have historically been prevented from branching out, both within their incorporation state and outside of it. The financial sectors of each state used to be separated by federal law due to an overall reluctance towards outside competition. In 1994 however, the Clinton administration passed the Riegle-Neal Act (henceforth RNA), which removed virtually all barriers to interstate branching. The main purpose of the law was allowing US banks to grow, as they could not compete with (much larger) financial service firms, the latter of which were unhindered by preexisting legal barriers (Matasar and Heiney, 1999). Another intention of this law was to allow banks to diversify their portfolios geographically, in order to reduce their levels of idiosyncratic risk. These risks became apparent due to the savings and loan crisis of the 1980s, which shed light on the overexposure of banks to local risks. However, an inevitable consequence of the RNA would be a substantial decrease in the number of US banks with an insufficient scale be able to survive in the new competitive environment. Opponents of the law therefore brought up concerns about the future stability of the smaller local banking markets (Giedeman, 2004).
The impact of the RNA on the risk-taking behavior of banks has been studied through various methodological techniques, but results remain ambiguous. This is mainly due to the two opposing mechanisms through which competition can affect bank risk, which have been debated in academia for a long time. On the one hand, traditional literature on the subject suggests that increased competition will cause banks to take on more risk in order to make up for the expected losses in their charter values. On the other hand, increased competition would imply that banks can no longer charge monopoly rents to their customers, therefore lowering their risk of defaulting on loans. Both sides of this argument have been extensively studied in empirical papers, but results have remained inconclusive.
This thesis therefore aims to contribute to this debate in the context of the US legislation on bank branching, which provides a fitting opportunity for empirical studies, due to the gradual removal of entry barriers. Specifically, it aims to investigate the impact of the adoption of the RNA on bank competition and subsequently, on bank portfolio risk. The main question this thesis will attempt to answer is therefore,
Did the adoption of the Riegle-Neal Act of 1994 affect the risk-taking behavior of US banks, through its impact on competition?
Answering this question would make a contribution to the field in several respects. Firstly, it would contribute to the existing knowledge-base on the effects of the RNA, by applying the Boone indicator to the topic, a recently developed model for firm competition, which has been most notably employed by van Leuvensteijn et al. (2011). Secondly, this thesis will attempt to take into consideration the possible endogeneity between bank risk and competition. Another
contribution of this thesis is therefore providing new insights to the existing debate about the relationship between bank competition and risk-taking behavior. Thirdly, the RNA is worth investigating once more, in light of the recent adoption of the Payment Services Directive 2 (henceforth PSD2) in the EU, a law which significantly lessens entry barriers for FinTech firms into the payments market. It does so by forcing all European banks to open up consumer transaction data to any payment service the customer might want to switch to (EUR-Lex, 2017). Because banks have dominated the payments industry since online banking started growing in popularity, their positions may now be challenged by a large number of new market entrants (Boot, 2017). PSD2 is similar to the RNA in that respect, because a consequence of both laws is an external shock to competition through a removal of entry barriers. This thesis will therefore provide new insights on the behavior of banks when faced with increased competition, which will be of use to policy-makers.
This thesis’ main analysis will measure competition using the Boone (2008) indicator for a large sample of US banks, over the period of 1986 to 2004. In order to remove endogeneity concerns, the indicator will be estimated using a 2SLS method, by using laws on the liber-alization of branching as instruments. The effect of competition on bank risk will then be investigated through OLS analysis.
The remainder of this paper will be structured as follows: First, the relevant existing liter-ature will be presented and discussed, focusing first on the RNA (and its historical context), then on the competition-risk debate, existing measures for competition, and subsequently on existing empirical studies and the results thereof. The thesis will proceed with a detailed description of the methodology used, along with the academic literature from which it draws insights. It will next provide a description on data sources and model specifics. After the results are presented and discussed, the paper concludes.
2
Theoretical Framework
The literature relevant for this study consists of two main bodies, concerning the Riegle-Neal Act of 1994, and the relationship between competition and bank stability. In order to put this research into the appropriate context, this section will first describe the historical background of the US banking system and provide arguments for using it as the primary setting of this study. Secondly, it will present the debate between the views about the effects of bank competition on risk along with existing empirical evidence. Thirdly, a number of measures for competition will be discussed, presenting their strengths and weaknesses, in order to explain why the Boone indicator was chosen for this analysis. The section will conclude by addressing the empirical effects of the RNA.
2.1
The US banking system
Prior to the passing of the RNA in 1994, the US banking system had been governed by the McFadden Act of 1927, which effectively blocked all banks from branching across state lines (Matasar and Heiney, 1999). Furthermore, state-level government was given full authority to decide on allowing banks to open new branches in the same state where they were in-corporated. Some bankers attempted to enter new markets (both intra- and interstate) by opening different banks, which were owned by the same bank holding company (BHC), but this loophole was closed in 1956 with the passage of the Douglas Amendment (Giedeman, 2004). As a result, the US banking system was comprised mainly of small, local institutions for most of the 20th century (Giedeman, 2004). These restrictions, however, resulted in sig-nificant difficulties for consumers and firms alike, due to the former not being able to receive banking services once they left their home state, and the latter having to open and maintain multiple deposit accounts for every state in which they operated (Brady and Purpura, 1998). In light of these difficulties, certain states began to slowly lessen the restrictions on intrastate branching. By 1970, only twelve states permitted unrestricted state-wide branching, but this trend increased significantly thereafter, with 38 states having removed their restrictions on state-wide branching by 1994 (Giedeman, 2004).
At the same time, changes began to take place in interstate banking regulation. In 1978, Maine started allowing entry by out-of-state BHCs, as long as the new entrant’s state reciprocated the law (Strahan, 2003). Subsequently, other states followed, such that all states but Hawaii had passed similar laws by 1993 (Strahan, 2003). It is important to mention that these laws, although allowing BHCs to acquire existing banks in other states, did not allow assets from banks in different states to be merged together, implying that BHCs still had to run separate operations for every state (Giedeman, 2004).
The adoption of the Riegle-Neal Act in 1994 removed all remaining restrictions to in-terstate branching in the US, allowing them to expand not only by means of mergers and acquisitions, but also by opening new (de novo) branches in any state. The passage of this law brought about significant change in the American banking sector as for the first time, banks were permitted to operate widespread nation-wide networks (Strahan, 2003). The law allowed each state to opt out of this piece of legislation, and only Montana and Texas chose to do so, although both reversed their decisions in 1999 (Giedeman, 2004).
The post-1970’s US banking system therefore provides a fitting context for analyzing this thesis’ main question, because different states modified their legal branching restrictions at different points in time. This makes it possible to analyze the effects of the RNA on states that had different degrees of entry restrictions prior to the adoption of the law. This will be discussed in more detail in section 4.
2.2
Debate
The traditional literature on the relationship between banking competition and risk stipulates that in more concentrated markets, banks benefit from high charter values, part of which they stand to lose when faced with more competition. This argument has been known as the
franchise value hypothesis and it has been extensively investigated in the academic literature.
One of the most prominent studies to argue in favor of this hypothesis was done by Keeley (1990), who modeled the relationship between market power and bank risk incentives and then tested it on a sample of 150 BHCs in the US. The study starts from the idea that banks were highly anticipated to exhibit moral hazard problems once the deposit insurance system went into place in 1933, with the establishment of the Federal Deposit Insurance Corporation. The author, however, points to the fact that the amount of insurance payouts remained low and stable over the following 50 years, as well as the tendency of banks to hold more capital than required. Keeley (1990) then emphasizes the fact that both the number of bank failures and the level of insurance expenses skyrocketed in the early 1980’s. He argues that this wave of defaults was caused by factors other than deposit insurance, considering the relative stability of the banking system during the prior 50 years, when deposits had also been insured. The author then conjectures that the increase in bank portfolio risk may have been caused by changes to the level of competition faced by banks. As explained in the previous subsection, most US states opened their financial markets to out-of-state BHCs during the 1980’s, and Keeley (1990) argues that this wave of legal liberalization may have posed significant threats to the charter values of existing banks. Because bank owners could not sell their charters once declared insolvent, they gained higher incentives not to go bankrupt and therefore increase their portfolio risk in order to offset the expected losses on their charters (Keeley, 1990). The author hypothesizes that these incentives ultimately lead to a rise in the number of bank failures due to imprudent behavior, which triggered deposit insurance payments by the FDIC. He then builds a theoretical model that captures the potential causal relationship between the level of competition in a banking sector and the charter values of participating banks, and the relationship between the charter values of banks and the number of bank failures. After testing the model on a sample of US BHCs, Keeley (1990) finds evidence to support his hypothesis, attributing part of the increase in bank failures and FDIC insurance payouts to the decrease in bank charter values caused by increased competition.
On the other side of the debate, there are numerous studies that reject the franchise value hypothesis. Mishkin (1999) for example, argues that concentrated banking markets can lead to too-big-to-fail concerns, the damaging effects of which can dominate the potentially stabilizing outcomes of low competition. One of the most prominent recent studies to provide an opposing risk-incentive mechanism to the traditional one was done by Boyd and De Nicolo
(2005). The authors point to the mixed empirical results on the matter, arguing that many of them use proxies for risk that are not directly related to the probability of default, or weak measures for competition. Their main argument against Keeley’s (1990) view is that it ignores competition on loan markets. In other words, they do not deny the predictions of the franchise value hypothesis. Rather, they argue that a contradicting, and sometimes dominant, risk-incentive process occurs on the asset side of balance sheets. More specifically, the authors claim that although decreased competition leads to higher interest earned in deposit markets, it also causes banks to charge monopoly rates on their loans. This in turn indirectly implies higher default probabilities for borrowers, reinforced by potentially higher moral hazard on their part, as they would raise their own risk when faced with higher rent costs (Boyd and De Nicolo, 2005). The authors subsequently develop a model in which banks increase loan rates when their market power is high, and as a result, borrowers choose riskier projects. The empirical study by Boyd et al. (2006) indicates that the probability of default increases with more concentration in the sector.
This debate represents the core of this paper. As will be discussed in more detail below, numerous empirical studies have provided evidence for both sides of the debate. This could be explained by the fact that a variety of empirical measures was employed in these studies. Recent developments in the field of industrial organization (IO) provide yet novel ways to measure the much-debated effects of this risk-stability relationship. This will be the focus of the next subsection.
2.3
Measuring competition
In the existing empirical literature, various methods are used for analyzing the competition-risk relationship in the banking sector. This subsection will examine a number of them in order to illustrate the existing measures of competition, as well as to discuss their benefits and drawbacks. At the same time, it will elaborate on why the Boone indicator for competition was chosen for the purpose of this particular research.
A widespread problem in banking literature is that competition in the financial sector cannot be measured directly, due to the unavailability of the prices of single banking products (van Leuvensteijn et al., 2011). As a result, various attempts have been made to estimate it indirectly.
2.3.1 The Herfindahl-Hirschman index
The Herfindahl-Hirschman index of market concentration (HHI) has been used extensively as a proxy for competition (Degryse, 2009). However, Claessens and Laeven (2004) have proven concentration ratios to be generally poor proxies for competition due to their contradictory predictions across different studies. They argue that rather than describing competition,
concentration affects bank performance through other, independent channels. Furthermore, Jimenez et al. (2013) find evidence for a nonlinear relationship between the HHI and bank risk. This suggests that banking systems are generally more stable for both low and high levels of market concentration than for mid-range values of the HHI. Another criticism of this measure for concentration relates to the fact that it can only be calculated on the basis of geographical borders, thereby assuming that they coincide with the borders of banking markets (Berger et al., 2009; Degryse, 2009). On the other hand, Craig and Dinger (2013) show that the HHI is significantly correlated with their measure of competition (deposit rates), and therefore use it as an instrument for estimating the level of competition.
2.3.2 The Lerner index
Another body of literature measures market competition by using the Lerner index, which is based on the margin of price over marginal costs (Berger et al., 2009). Traditional IO literature suggests that market competition drives prices down towards marginal costs, such that the larger the mark-up of prices over marginal cost are, the lower the level of competition in a market must be. The indicator developed by Lerner (1934) captures this effect. The first advantage of the Lerner index over concentration indicators is that it is calculated at the firm-level, therefore eliminating the need for defining geographical markets. Second, it can be calculated for distinct products, which allows for investigating competition on different markets (Beck et al., 2013). On the other hand, because the Lerner index relies on the efficiency of individual firms within a market, a higher index value is not necessarily associated with a lack of competition (van Leuvensteijn et al., 2011). For example, more efficient banks may have a higher markup of prices over costs due to the exploitation of scale economies, which would lead to an increase in the industry’s average Lerner index, contrary to expectations (van Leuvensteijn et al., 2011). Jimenez et al. (2013) estimate the index for a variety of bank-specific interest rates in Spain (including those on deposits and loans), and their results support the traditional franchise value hypothesis, but only in the loan market. Their results do not seem to prove the predictions made by Boyd and De Nicolo (2005), and the authors speculate that market power over deposits may allow banks to behave more aggressively in the loan market. Berger et al. (2008) argue that a potential explanation for these findings is the focus of the study on portfolio risk (as measured by non-performing loans), without investigating overall bank risk. They also speculate that competition can be endogenous, because riskier banks with higher expected returns may gain incentives to increase their market power. The authors therefore revisit the issue, but with two differences from Jimenez et al. (2013): (i) they measure the effect of competition on the default risk of banks rather than their portfolio risks, and (ii) they account for the possible endogeneity of market power. The authors employ a 2SLS technique to estimate the Lerner index, yet
their results remain relatively unchanged. They find evidence to support the prediction that increased competition leads to less bank risk, but insufficient proof for this effect’s dominance over the franchise value hypothesis.
2.3.3 The H-statistic
Panzar and Rosse (1987) propose a different measure for market competition, the H-statistic, which captures the extent to which changes in bank input prices are reflected in their rev-enues. The authors argue that under perfect competition, an increase in factor input prices will increase a bank’s marginal costs and total revenues by the same amount. Under monopoly conditions, however, they claim that an increase in input prices only raises marginal costs, causes a decrease in (equilibrium) output, and therefore decreases revenues. The H-statistic distinguishes between three levels of competition: (i) monopoly or perfect collusion for neg-ative values of H, (ii) monopolistic competition for values between 0 and 1, and (iii) perfect competition if H is equal to 1. An important assumption made when calculating this indicator is that the market is in long-term equilibrium (Degryse, 2009). Claessens and Laeven (2004) apply the Panzar and Rosse (1987) index to the banking sectors of 50 countries and find that the market structure in most countries can be described by monopolistic competition. Fur-thermore, their study investigates potential determinants of competition as measured by the H-statistic and finds evidence that fewer entry barriers and activity restrictions are conducive to more banking competition. Their results also reinforce the inability of concentration ratios such as the HHI to predict competition.
2.3.4 The Boone indicator
A novel way of measuring competition has been introduced by Boone (2008). The author begins from the idea that inefficient firms (i.e. those firms with higher marginal costs) inevitably stand to lose part of their profits, and that this loss is reinforced the higher market competition is. The goal of his 2008 study is to develop a measure of competition based on this assumption. The intuition underlying this estimator for competition has certain similarities with the one behind the HHI, specifically that a fall in entry barriers will reduce market concentration, which is likely to increase competition (Degryse, 2009). However, Boone (2008) further develops upon standard measures of concentration by considering the idea that market incumbents may behave more aggressively when faced with increased competition. In other words, it is possible that increased competition will reallocate outputs from inefficient firms to the more efficient ones (which already benefit from high levels of output), and thus increase market concentration (van Leuvensteijn et al., 2011). As long as this reallocation process holds, the Boone estimator remains valid, regardless of the source of the change in the level of competition (Schaeck and Cihak, 2014). In order to capture this effect, Boone (2008)
employs the elasticity of profits with respect to marginal costs (profit elasticity). Intuitively, as the market becomes more competitive, the profits of more efficient banks increase relative to those of less efficient ones, for given levels of efficiency of individual banks (Schaeck and Cihak, 2014). The theoretical model of Boone (2008) can be found in Appendix A.
A considerable advantage of the Boone indicator over the Panzar and Rosse (1987) H-statistic is that it can focus on individual sub-markets for distinct products, rather than the industry as a whole (van Leuvensteijn et al., 2011). Furthermore, an increase in the H-statistic cannot unambiguously be associated with more competition, whereas an increase in the Boone indicator can (Degryse, 2009). On the other hand, this measure is limited by its assumptions of product homogeneity between distinct firms, as well as its disregard for incentives to innovate (van Leuvensteijn et al., 2011). An additional possible drawback of the Boone indicator is that there is no benchmark for its interpretation other than comparing it to the values found in other markets.
The first empirical study to apply the Boone indicator to the banking sector was done by van Leuvensteijn, Bikker, van Rixtel, and Sorensen (2011), who investigate the competition in the bank loan markets of several countries. Their results suggest that competition varies considerably across countries, which they attribute to characteristics of individual banking sectors, such as changes in the regulatory environment of banks.
The Boone indicator is the best fitting measure of competition for the purpose of this study, mainly thanks to its ability to distinguish between markets for different products, as well as to unambiguously capture an increase in the level of competition. However, a drawback of using the Boone indicator in the banking sector is that bank products are difficult to classify according to IO principles. For example, different strands of banking literature have classified deposits as both input products1, and output products2. This issue
will be revisited in section 4.
2.4
The effects of the RNA
A considerable number of studies has investigated the effects of the Riegle-Neal Act on the US banking industry, many among which have focused specifically on the elements relevant to this thesis. For example, Matasar and Heiney (1999) provide an early glimpse at the first market trends initiated by the RNA. They find that in the first two years after the adoption of the law, the total number of FDIC-insured institutions declined from 10,227 to 9,807, whereas the total number of bank branches in the the country increased by almost 2000 in the same time frame. The authors show that the largest banks have grown at the expense of smaller banks, and they attribute this to the need of large banks to grow enough be able to
1See Freixas and Rochet (1998). 2See Berger and Humphrey (1997).
compete on an international level. Similar results were found by Strahan (2003), who found that market shares of small banks have decreased post-RNA.
Another goal of the RNA was allowing banks to diversify their geographic risk. Aguirre-gabiria et al. (2016) show that although the RNA expanded the possibilities for geographic diversification greatly, banks were reluctant to do so. Their results, consistent with those of Matasar and Heiney (1999), suggest that mainly large banks expanded outside of their home state with the intention to grow rather than reduce deposit risk. The authors also provide evidence that US banks largely remained overexposed to idiosyncratic local risks up until the crisis of 2007-2008, despite having the opportunity to diversify. They attribute this to high merger costs, which only allowed larger banks to expand. On the other hand, Goetz et al. (2016) provide evidence of a negative relationship between the geographic expansion of BHCs and their overall risk, but no evidence that geographic diversification improved loan quality. The authors therefore infer that the observed risk benefits of geographic expansion are due to a reduction in local idiosyncratic risk.
In the years prior to the adoption of the RNA, there had been widespread concern about banking markets becoming too competitive for small firms to survive after the law went into effect. Giedeman (2004) investigates whether this fear had been founded, by measuring the deviation in HHI growth post-1994, as compared to its pre-1994 trend. Using a sample of 1500 randomly selected US cities, the author finds that the RNA had no significant impact on the pre-1994 trend in HHI change, suggesting that the local banking market concentration had been determined by factors prior to the full removal of interstate barriers.
Akhigbe and Whyte (2003) study the direct effects of the RNA on the volatility of BHC stock returns. They provide evidence that in the first year after the adoption of the RNA, the systematic risk of banks declined on average. These results are not driven by changes in market risk, since market returns do not change significantly over the sample period. The authors also show that banks incorporated in those states that had more restrictions on interstate banking in place prior to 1994 experience a significant decrease in risk. They find the opposite to be true for banks incorporated in states with the most liberal branching laws, suggesting that the competitive environment of each market plays an important role in these findings.
3
Methodology and data
The empirical analysis of this paper will be executed in three steps. The first step is to calcu-late the quarterly Boone indicator in all state-level markets, by following the methodologies of Schaek and Cihak (2014), van Leuvensteijn et al. (2011), and that of Berger et al. (2009). Secondly, the indicator will be estimated by means of 2SLS. The third and final step will present this paper’s primary analysis by investigating the relationship between this indicator
and risk. The remainder of this section will present detailed descriptions of the methodology used in each step. A list of definitions for all variables can be found in Appendix B.
An important note about applying the Boone indicator to the banking sector is the calculation of the marginal costs of bank products, as it requires distinguishing between input products and output products. Whereas most literature that uses bank marginal costs takes a “single output approach” (i.e. does not distinguish between loans, securities, or deposits)3, van Leuvensteijn et al. (2011) calculate marginal costs for bank loans only. In
order to answer this paper’s research question, the main analysis will also employ a “single output approach” to calculate overall bank competition. The approach of van Leuvensteijn et al. (2011) will also be included, but only as a robustness measure.
3.1
Estimation of the Boone indicator
The analysis begins with the empirical specification of the Boone (2008) model (explained in more detail in Appendix A)
πit = α + βln(mcit) (1)
where πit measures the variable profits of bank i at time t, mcit represents marginal costs,
and β is the Boone indicator. However, instead of using profits πit as the dependent variable,
Van Leuvensteijn et al. (2011) replace it with the loan market share of bank i at time t. Their reasoning in doing so is that profits are by definition the product of market share and profit margins, and as a result, any efficiency gains (lower marginal costs) will lead to higher market shares. The empirical specification of the model then becomes
ln(si) = α + βln(mci) (2)
with market share defined as si = piqi/Pj6=ipjqj. The authors also specify the relationship
in log-linear terms to deal with heteroscedasticity4.
Because marginal costs cannot be observed directly, they will be calculated by employing a trans-log cost function (TCF), which yields more precise estimates than approximating them by average variable costs5 (van Leuvensteijn et al., 2011). To that end, by using individual
3See Berger et al. (2009); De-Ramon and Straughan (2016); Schaek and Cihak (2016).
4Equation (2) will provide valuable information about robustness, as it can disentangle the effect of loan
competition from total competition. It will therefore be run in two separate specifications, for different dependent variable: (i) total bank profits, and (ii) loan market share.
bank data, a TCF can be estimated for each state, of the following form6: ln(Cit) = α0+ α1ln(Qit) + α2 2 ln 2(Q it) + 3 X j=1 βjln(Wj,it) + 3 X j=1 γjln(Wj,it)ln(Qit)+ +1 2 3 X k=1 3 X j=1 δkjln(Wk,it)ln(Wj,it) + + 1 2λE 2 it+ Φ 0X it+ εit (3)
where Citis the total cost of bank i at time t, Qitis its total output7, and Wj,itrepresents three
input prices: labor (W1,it), physical capital (W2,it), and fixed capital (W3,it). Furthermore, Eit
is bank capital and Xit is a vector of period fixed effects. The standard errors are clustered
at the bank level8. The α,β, γ, δ, λ, and Φ parameters will be estimated through regression
analysis. An assumption of cost functions is that input costs are linearly homogeneous. In order to ensure that this is the case, both the total costs and the input prices are divided by
W3,it.
The marginal costs of bank i at time t, ∂Cit/∂Qit, are therefore defined as
mcit = Cit Qit α1+ 2 × α2ln(Qit) + 3 X j=1 γjln(Wj,it) (4)
3.2
IV analysis for estimating the Boone indicator
After having estimated the Boone indicator for bank competition, it is necessary to address the potential endogeneity of bank risk. Because evidence has been found that a bank’s risk level can also determine its efficiency 9, this analysis will employ a 2SLS method to remove the potential reverse causality between the measures of risk and the Boone indicator. The instruments used are dummy variables which describe whether a bank operates within a state that has legally liberalized its branching restrictions. More precisely, D1,it is equal to 1 if the
home state of bank i has no restriction on intrastate branching at time t, and 0 otherwise. Similarly, D2,it and D3,itrefer to restrictions on inter state branching and the implementation
of the RNA, respectively. Because the individual states adopted the three sets of laws at dates different from each other, it is easy to construct the sample in this manner.
The Boone (2008) model proposes that the only relevant sources of increased competition
6This formulation was reproduced directly from the study of De Ramon and Straughan (2016), who
incorporate time trends, bank capital, and control variables into the TCF expression given by Berger et al. (2009), as suggested in the original text.
7The total output of banks is used in order to estimate its marginal costs, due to the fact that deposits
can be considered both input and output goods in the activities of banks. As previously shown by empirical literature however (e.g. Matasar and Heiney, 1999; Aguirregabiria et al., 2016), the post-RNA expansion of banks was mainly driven by growth. Consequently, this study expects marginal costs to affect the ability of banks to compete on deposits, as well as loans.
8See Appendix B for a list of proxies used for these variables. 9See Fiordelisi et al., (2011); Barra et al., (2016); Delis et al., (2017);
are increased product homogeneity, and a fall in entry costs. The liberalization of banking markets is viewed as an external shock to the latter effect, and therefore the instrumental variables can be expected to capture it. Although these instruments may be expected to directly affect bank risk as well (through the channel of geographic diversification poten-tial), previous literature has largely proven this not to be the case (Matasar and Heiney, 1999; Aguirregabiria et al. (2016)). To address the evidence that BHC risk decreased as a consequence of geographic expansion, as found by Goetz et al. (2016), the primary anal-ysis will control for this effect by including the total number of subsidiaries per BHC as a diversification proxy.
3.3
Primary analysis
To test the implications that bank competition has on risk, the primary analysis of this paper will take the form of
BankRiskit = α + βCompetitiont+ BankControlsit+ StateControlst+ εit (5)
where BankRiskit is a measure of the stability of bank i at time t, Competitiont is the
2SLS estimate of the Boone indicator, and the bank- and state-specific control variables are included as specified by Schaek and Cihak (2014)10. All explanatory variables are lagged by
one period. In this model, β is the parameter of interest, as it will capture the relationship between market competition and bank risk.
The level of bank risk will be estimated using the Z-score, which measures by how many standard deviations the ROA can decrease before a bank defaults. Although most empirical studies on this topic also employ the ratio of non-performing loans (NPL) as a proxy for bank risk, this thesis will leave it out. A common issue with the NPL is that there is no clear threshold for when a certain loan should be declared as non-performing, which makes it highly subjective on the judgement of the bank’s management (Lastra et al., 2016). This is expected to be especially true in the pre-internet era, when the lack of information on individual loans was even more severe than now11.
The primary goal of this analysis is to answer the research question by testing the null hypothesis H0 : Bank competition has no effect on stability. The sign of the coefficient of Competition will provide the answer to the question. It is possible that the coefficient is
pos-itive, which will be associated with a negative competition-risk relationship, as hypothesized by Keeley (1990). On the other hand, if the coefficient is negative, it will serve as evidence
10For a complete list of the variables and their measurement, refer to Appendix B
11See Meeker and Gray (1987) for an overview of the NPL measure during the first years after banks
Table 1: Summary statistics of the translog cost function variables
Variable Obs. Mean Std. Dev. Min. Max
Costs (log) 953,912 7.041 1.283 0.693 16.742
Assets (log) 953,944 11.071 1.317 4.718 20.902
Equity (log) 953,939 8.682 1.293 0.693 18.52
Price of labor (log) 952,829 -5.524 0.395 -13.765 -0.424 Price of physical capital (log) 949,448 -0.847 0.994 -8.003 11.414 Price of fixed capital (log) 951,809 -4.555 0.535 -12.536 7.809 Estimated marginal costs 948,157 0.019 0.025 0.001 10.248
that higher competition is conducive to stability, as proposed by Boyd and De Nicolo (2005).
3.4
Data
The main database used in this study was provided by Drechsler et al. (2017), who com-bine quarterly call reports from WRDS into one consistent time series, which includes both balance sheet and income statement data. Their data covers all FDIC-insured banks over the period between 1984 and 2013. However, in order to avoid foreseen distortions of the lead-up to the 2008 financial crisis, this paper’s sample stops at 2006. This database was subsequently complemented with GDP and unemployment measures, retrieved from the Bu-reau of Economic Analysis12 and the Bureau of Labor Statistics13, respectively. The dates
at which various banking restrictions were removed for each state are provided by Strahan (2003). In order to exclude any unreliable observations, the data was cleaned, according to the following criteria: assets, loans, equity, deposits, and all measures of revenue and expenses should be positive14; the deposits-to-assets- ratio should be less than 0.98, the loans-to-assets ratio should be less than 1. Furthermore, in order to remove any outliers, all financial ratios are winsorized using the standard procedure, at the 1% level. The final data set contains 954,003 quarterly bank observations of 18,373 individual banks from all 51 states. Table 1 presents the summary statistics of the variables used in the process of estimating marginal costs. Table 2 presents those of the main analysis.
12Link to the BEA website. 13Link to the BLS website.
Table 2: Summary statistics of the main analysis variables
Variable Obs. Mean Std. Dev. Min. Max.
Z-score (log) 947,337 4.738 1.085 1.899 7.098
Boone indicator 954,003 -1.414 0.717 -7.119 1.533
Asset growth 935,630 0.022 0.061 -0.109 0.331
No. of BHC subsidiaries 954,003 3.553 10.093 1 94
Loan loss provisions / Total assets 954,003 0.001 0.002 0 0.0141 Total market assets (log) 954,003 18.034 1.041 12.971 21.368
HHI deposits 954,003 0.064 0.077 0.005 0.915
HHI loans 954,003 0.087 0.091 0.007 0.927
GDP (log) 954,003 8.858 0.361 8.272 9.551
Unemployment rate 954,003 0.057 0.017 0.021 0.154
Instruments
Unrestricted intrastate branching 954,003 0.754 0.430 0 1 Restricted interstate branching 954,003 0.821 0.382 0 1 Unrestricted interstate branching (RNA) 954,003 0.337 0.473 0 1
Figure 1: Yearly averages of the Boone indicator and Z-score
The graph presents the yearly mean values of both variables, on a national level. In order to facilitate interpretation, the graph presents the negative values of the Boone indicator because higher absolute values are associated with higher competition.
4
Results
Following the procedure described above, the Boone indicator was calculated separately for every state. In order to observe the development of competition over time, a different speci-fication of equation (1)15 was run with year dummy variables, where the level of competition
in every year was given by the coefficient of the interaction term. Figure 1 presents the devel-opment of the national level of competition in the banking sector along with the develdevel-opment in the overall riskiness of banks. The graph suggests a downward trend in competition until 1994 (the year of adoption of the RNA), a period in which individual states were making agreements with each other to open up their financial markets. Competition seems to sta-bilize until the implementation of the RNA in 1997 and increases thereafter, until the year 2000. The subsequent decrease in the level of competition coincides with the crash of the Dot-Com bubble. On the other hand, the average Z-score exhibits a steady upward trend, suggesting that bank stability improves over time. There is also a slight dip in average bank stability around 2000, which may be explained by the same crisis. The period between 2002 and 2003 shows that competition as well as stability start increasing, which may be explained by the raise in the issuance of mortgages as well as mortgage-backed securities, before the crisis of 2008.
Table 3 presents the results of this paper’s main analysis, which employs a 2SLS regression with robust standard errors clustered at the state level. Model (1) presents the coefficients of a simple OLS regression, whereas model (2) employs instrumental variables. Both models measure bank stability with the natural logarithm of the Z-score, and competition with the Boone indicator. Because the Boone indicator is calculated as the relationship between profits and marginal costs, it is expected to largely have negative values. It is therefore an inverse measure of competition. The first stage F-test of model (2) shows that the instruments used to estimate the Boone indicator are relevant. Furthermore, both models include control variables for bank size, risk-preferences, diversification, and asset quality16. To control for
market and macroeconomic effects, the HHI, total market size, GDP, and unemployment rate have been included in the model. All explanatory variables have been lagged by one period in order to allow for the fact that banks adjust their portfolio risk retroactively. In both specifications, bank size and GDP predict higher stability on average, whereas risk-appetite, asset quality, market concentration, and the rate of unemployment are associated with lower Z-scores. The coefficient of the No. of BHC subsidiaries is negative, which suggests that banks owned by a BHC tend to be riskier, the larger the BHC.
The coefficient of the Boone indicator is positive in both specifications, suggesting that
15The new specification becomes π
it= α +PTk=1βk1dktln(cit) +PT −1k=1 βk2dk1+ uit, where dktis a dummy
variable that equals one if k = 1 and equals zero otherwise.
16Proxied by total assets (log), asset growth, the number of subsidiaries of the BHC, and loan loss provisions
T able 3: The effect of comp etition on bank stabilit y This table presen ts the co efficien ts of the OLS and 2SLS regressions of bank stabilit y on comp etition. Mo del (1) presen ts the results of the OLS regress ion, and mo del (2) those of th e 2SLS regression. In b oth situations, the dep enden t v ariable is the Z-score (log), and the main meas ure for comp etition is the y early state-lev el Bo one indicator . The instrumen t v ariables of mo del (2) are the absence of in trastate branc hing restrictions, th e presence of bilateral in terstate banking agreemen ts, and the implemen tation of the RNA. The fi rst stage F-test is also pr esen te d in mo del (2), whic h is a measure of instrumen t exogeneit y . The standar d rule of th um b in econometric literatu re is that a first stage F-statis tic larger than 10 implies that the instrumen ts are exogenous. Standard errors are clustere d at the state-lev el, to allo w for heterogeneit y . All dep e nden t v ariables are lagged b y one p erio d. It shou ld b e noted th at the Bo one indicator is a negativ e measure of c omp etition and there fore a negativ e co efficien t implies a p ositiv e comp etition-risk relationship and vice v ers a. Dep enden t v ariable Z-score (log) Mo del OLS (1) 2SLS (2) Bo one indicator 0 . 0228 0 . 2392** (0 . 0277) (0 . 1074) T otal assets (log) 0 . 1861*** 0 . 1745*** (0 . 0131) (0 . 0149) Asset gro wth -0 . 0469 -0 . 143 (0 . 0973) (0 . 1043) No. of BHC subsidiaries -0 . 0034** -0 . 0032** (0 . 0016) (0 . 0016) Loan loss pro visions / T otal assets (%) -1 . 5271*** -1 . 5262*** (0 . 0481) (0 . 0475) Mark et as sets (log) 0 . 0119 -0 . 0058 (0 . 0382) (0 . 0439) HHI dep osits -0 . 0882 -0 . 0475 (0 . 3052) (0 . 5099) HHI loans -0 . 2568 -0 . 5262 (0 . 3238) (0 . 4452) GDP (log) 2 . 0401** 1 . 8663** (0 . 9615) (0 . 9117) Unemplo ymen t rate -3 . 4292 -3 . 21 (2 . 2031) (2 . 281) Time trend -0 . 0815 -0 . 0847* (0 . 0523) (0 . 0487) Observ ations 884,119 884,119 R 2 0 . 2323 0 . 2207 First stage F-test − 14 . 187*** ***Significan t at the 0.01 lev el. **Significan t at the 0.05 lev el. *Significan t at the 0.1 lev el.
competition is negatively related to bank risk. However, both the magnitude and significance level of the coefficient increases noticeably when employing a 2SLS methodology, which may suggest the presence of endogeneity issues in model (1). These are likely to arise because more fragile banks have more incentive to exhibit moral hazard problems, which may affect their competitive behavior. By using legal barriers to entry as instruments for the Boone indicator, these problems appear to be ameliorated. The results of the second specification imply that bank competition, as determined by (the lack of) branching restrictions, significantly reduces their stability.
These results appear to be in line with the predictions of Keeley (1990). It should be noted, however, that this analysis focuses on overall measures of competition and risk. In other words, the estimated Boone indicators express the intensity of bank competition on all outputs, and the Z-score measures overall bank soundness. Thus, the results cannot be interpreted as evidence against the competition-stability view of Boyd and De Nicolo (2005).
5
Robustness Tests
Table 4 presents the results of the two robustness checks performed. The first test uses the Boone indicator for the loan market as the relevant measure of competition. Its calculation follows the methodology of van Leuvensteijn et al. (2011), which first estimates the marginal costs of loans and then uses a modified specification of the Boone (2008) model. The main reason behind using this variable is the fact that it allows for investigating how banks stability reacts to loan market competition, in order to determine if the predictions of Boyd and De Nicolo (2005) hold for the US financial sector. Aside from the fact that a different competition measure is used, the test includes the same dependent and independent variables as employed in the main analysis, and standard errors are clustered by state in order to account for heterogeneity. The resulting coefficients of model (1) in table 4 firstly indicate the robustness of the control variables, as their values and significance levels remain largely unchanged from the ones in the main analysis above. The coefficients of the HHI (deposits) and of the competition measure have changed signs in the new test, implying that both measures are positively related to bank stability. The OLS specification provides some evidence that the effect of competition on the loan market can offset that of overall competition, in line with Boyd and De Nicolo (2005).
Model (2) presents the results of a 2SLS regression of the Boone loan market indicator on the Z-score of banks, using the same instruments as in the main analysis in the previous section. The F-statistic of the first-stage regression shows that the instruments are exogenous. Most of the resulting coefficients lose their significance, including the one of the Boone indicator. Model (2) suggests that loan competition, when estimated through instrumental variables, is detrimental to bank stability but this effect is not significantly different from
T able 4: Robustness tests This table presen ts the results of tw o robustness tests. The first substi tutes th e comp etition measure for the Bo one indicator on the loan mark et only . The second uses the P an zar and Rosse H-statistic as a pro xy for comp etition. The co efficien ts of the OLS and 2SLS regressions of bank stabilit y on comp etition are presen ted for eac h test. Mo de ls (1) and (3) pr esen t the results of the OLS regressions , and mo dels (2) and (4) those of the 2SLS regressions. The dep enden t v ariable is the Z-score (log) in all sp ecifi cations. The instru men t v ariables of mo dels (2) an d (4) are the absence of in trastate branc hing restrictions, the presence of bilateral in ters tate banking agreemen ts, and the implemen tation of the RNA. The first stage F-test is al so presen ted in mo dels (2) and (4), wh ic h is a measure of instrumen t exogeneit y . The standard rule of th um b in econometric lite rature is that a first stage F-statis tic larger than 10 implies that the instrumen ts are exogenous . Standard errors are clustered at the state-lev el, to allo w for heterogen eit y . All dep enden t v ariables are lagged b y one p erio d. It should b e noted that the Bo one indicator is a negativ e measure of comp etition and therefore a negativ e co efficien t implies a p ositiv e comp etition-risk relationship and vice-v ersa. Dep enden t v ariable Z-score (log) Z-score (log) Comp etition measure Bo one indicator for the loan mark et P anzar and Rosse H-statistic Mo del OLS (1) 2SLS (2) OLS (3) 2SLS (4) Comp etition measure -0 . 0562* 2 . 5111 -0 . 1434 2 . 551** (0 . 0335) (2 . 6947) (0 . 1562) (1 . 2637) T otal assets (log) 0 . 1904*** 0 . 0455 0 . 1879*** 0 . 1749*** (0 . 0129) (0 . 1307) (0 . 0135) (0 . 0159) Asset gro wth -0 . 0222 -0 . 6897 -0 . 0316 -0 . 1278 (0 . 0962) (0 . 7223) (0 . 0985) (0 . 1167) No. of BHC subsidiaries -0 . 0034** -0 . 0014 -0 . 0034 -0 . 0027 (0 . 0015) (0 . 0061) (0 . 0016) ( . 002) Loan loss pro visions / T otal assets (%) -1 . 5209*** -1 . 8065*** -1 . 5236*** -1 . 5894*** (0 . 0477) (0 . 3555) (0 . 0481) (0 . 071) Mark et as sets (log) 0 . 0097 0 . 1948 0 . 0139 0 . 0108 (0 . 0363) (0 . 2679) (0 . 0364) (0 . 0549) HHI dep osits 0 . 0308 -5 . 6149 -0 . 0872 -0 . 1876 (0 . 2358) (5 . 9785) (0 . 2662) (0 . 9595) HHI loans -0 . 2753 1 . 8688 -0 . 2067 -0 . 6152 (0 . 3041) (2 . 9084) (0 . 3138) (1 . 0268) GDP (log) 2 . 1137** -0 . 4177 2 . 0645** 1 . 951* (0 . 9608) (3 . 5701) (0 . 988) (1 . 179) Unemplo ymen t rate -3 . 6968 7 . 4935 -3 . 2554 -6 . 9543* (2 . 2345) (15 . 476) (2 . 1145) (3 . 9612) Time trend -0 . 0846 0 . 0698 -0 . 0808 -0 . 0882 (0 . 0518) (0 . 1924) (0 . 0538) (0 . 0626) Observ ations 884,119 884,119 884,119 884,119 R 2 0 . 2330 -0 . 2325 0 . 0968 First stage F-test -63 . 6289*** -12 . 5546 *** ***Significan t at the 0.01 lev el. **Significan t at the 0.05 lev el. *Significan t at the 0.1 lev el.
zero.
The goal of the second robustness check is to study if the results of the main analysis hold for another measure of competition, the Panzar and Rosse H-statistic. The reason for choos-ing this proxy is that it is also based on elasticity measures, similar to the Boone indicator. It is calculated separately for every state, and includes yearly interacted variables in order to estimate its progression over time. An important potential issue with the H-statistic is that it assumes long-run equilibrium in the market, which may not be the case over the sample pe-riod, given the dynamic nature of banking activity laws. The test is ran in two specifications, OLS (1) and 2SLS (2). The first model confirms the negative relationship between competi-tion and stability shown in the main analysis, as well as the effects of the control variables. However, when estimating the H-statistic on the legal entry barrier variables, its coefficient suggests the opposite relationship, with a significant value. This result could be caused by a number of factors. For example, the index’s main assumption of long-term equilibrium in combination with the 2SLS regression on instruments that represent equilibrium-breaking points in time could invalidate the coefficients of model (4). In this situation, the results of the OLS regression of the H-statistic are likely more informative, despite possible endogeneity issues.
6
Conclusion
This thesis aims to investigate the relationship between bank competition and risk-taking, in the context of the Riegle-Neal Act of 1994, which removed all restrictions to interstate banking. The adoption of this piece of legislation, and the banking environment leading up to it provide a fitting context for this study. By using a sample of 18,373 individual US commercial banks over the period of 1984-2006, this study estimates the Boone indicator for every state-year competition. The proxy for bank risk is the Z-index, which is a firm-level measure describing its stability. The Z-index is then regressed on the Boone indicator through a 2SLS method, which includes several bank and market control variables. Three instrumental variables are used, namely the absence of intrastate branching restrictions, the presence of bilateral interstate banking agreements, and the implementation of the RNA.
The results suggest that competition reduces the stability of individual banks. This effect is reinforced when applying a 2SLS regression. This finding is in line with the predictions of Keeley (1990) who proposed the franchise value theory of bank competition. In order to investigate the potentially beneficial effects of competition on the loan market (as proposed by Boyd and De Nicolo (2005)), the regression is ran with the Boone indicator for loan competition as dependent variable. The results of the OLS regression suggest that higher loan competition is conducive to higher stability. However, the 2SLS estimation negates this effect, indicating that loan competition is negatively related to bank stability.
To determine the robustness of these findings, the Boone indicator is substituted for the Panzar and Rosse H-statistic. The robustness checks confirm the results of the main analysis when an OLS regression is used, but contradict them under a 2SLS method. This need not mean the analysis is not robust, since the H-statistic assumes markets are in equilibrium, which is unlikely to have been the case while the RNA was being implemented. Given these outcomes, then, the null hypothesis that competition has no effect on bank risk is rejected, as competition was shown to have a significantly negative influence on stability in the financial sector.
This paper’s results suggest that the RNA has affected the stability of the US banking system. However, more research is needed to determine the exact effects it had on loan com-petition. This study provides potentially important policy implications for financial sectors. They show that loosening entry barriers in banking markets can give them incentives to take on more risk. Therefore, policy-makers would be wise to keep a close eye on bank risk-taking behavior whenever policies are implemented that aim to reduce entry barriers. Closer cooperation between antitrust enforcers and financial regulators could be of value in such situations. Future research should focus on the effects on risk of the Payment Services Direc-tive 2, the EU legislation that opens up the payments market to non-financial institutions, in order to determine if European banks exhibit similar behavior to their US counterparts. Furthermore, more studies should focus on the distinction between competition on different banking products. Differentiating between them could help reach conclusions in the debate about its potential effects on stability.
Appendix A. The Boone (2008) indicator model
Consider a banking market where each bank i produces one product (or portfolio of products)
qi, which faces the demand curve
p(qi, qj6=i) = a − bqi− d
X
i6=j
qj (6)
and has constant marginal costs mci. The variable p denotes price, a stands for market size,
b denotes the price elasticity of demand, and d represents the extent to which consumers
view different products in a market as substitutes. Furthermore, assume that a > mci and
0 < d ≤ b. Next, the bank maximizes its profits πi = (pi − mci)qi with respect to qi. The
resulting first-order condition for equilibrium is given by
a − 2bqi− d
X
i6=j
qj− mci = 0 (7)
In a market where N banks produce positive levels of output, the N first-order conditions given by (2) are calculated as
qi(mci) =
(2b/d − 1)a − (2b/d + N − 1)mci+Pjmcj
[2b + d(N − 1)](2b/d − 1) (8)
Equation (3) provides the relationship between output and marginal costs. The expression
πi = (pi− mci)qi shows that profits depend on marginal costs in a quadratic way. Profits πi
are defined as variable profits that exclude entry costs ε. A bank will therefore only enter the market if, and only if, πi ≥ ε in equilibrium.
Competition in this industry can only increase for two reasons: (i) when the (portfolio of) products of different banks become closer substitutes, i.e. d increases (assuming that d < b) and (ii) when entry costs ε decline.
The empirical model for estimating competition within the market is therefore
πit = α + βln(mcit) (9)
Appendix B. Definitions of regression variables
Variable name Proxy
Main
analysis
Z-score Defined as Z = (ROA + E/A)/σROA, where ROA is return on assets, E/A is the ratio of equity to assets, and σROA is the standard deviation of the return on assets. This study calculates the latter as a quarterly standard deviation for a window of 3 years.
Bank size Log(Total assets)
Risk preference Period growth rate of total assets
No. of BHC subsidiaries Total number of subsidiaries owned by the BHC which owns the bank. If the bank is not part of a BHC, this variable is equal to one.
Asset quality Loan loss provisions / Total assets
Market size Log(Total assets)
Market structure HHI (deposits); HHI (loans)
GDP Log(GDP)
Rate of unemployment As retrieved from the Bureau of Labor Statistics
Time trend Current quarter - First quarter of the sample
T
ranslog
cost
function
Variable profits Total operating profits / Total assets
Total costs Interest expenses + Salaries + Other operational expense
Loan market share Total bank loans / Total market loans
Total output Total operating income / Total assets
Loan output Interest income from loans / Total assets
Labor costs Salaries / Total assets
Physical capital Other operational expenses / Fixed assets
Fixed Capital Interest expense / Total deposits
References
[1] Aguirregabiria, V., Clark, R., & Wang, H. (2016). Diversification of geographic risk in retail bank networks: evidence from bank expansion after the Riegle-Neal Act. The RAND Journal of Economics, 47(3), 529-572.
[2] Akhigbe, A., Whyte, A. M. (2003). Changes in market assessments of bank risk following the Riegle–Neal Act of 1994. Journal of banking & finance, 27(1), 87-102.
[3] Barra, C., Bimonte, G., & Zotti, R. (2016). On the relationship among efficiency, capital-ization and risk: Does management matter in local banking market? Applied Economics, 48(41), 3912-3934.
[4] Beck, T., De Jonghe, O., & Schepens, G. (2013). Bank competition and stability: cross-country heterogeneity. Journal of financial Intermediation, 22(2), 218-244.
[5] Berger, A. N., & Humphrey, D. B. (1997). Efficiency of financial institutions: Interna-tional survey and directions for future research. European journal of operaInterna-tional research, 98(2), 175-212.
[6] Berger, A. N., Klapper, L. F., & Turk-Ariss, R. (2009). Bank competition and financial stability. Journal of Financial Services Research, 35(2), 99-118.
[7] Boone, J. (2008). A new way to measure competition. The Economic Journal, 118(531), 1245-1261.
[8] Boot, A.W.A (2017). The Future of Banking: From Scale & Scope Economies to Fintech, European Economy - Banks, Regulation and The Real Sector, Vol. 2, 2017, p.77-95. [9] Boyd, J. H., & De Nicolo, G. (2005). The theory of bank risk taking and competition
revisited. The Journal of finance, 60(3), 1329-1343.
[10] Brady, H. M., & Purpura, M. V. (1998). The Riegle-Neal Amendments Act of 1997: The Impact of Interstate Branching on the Dual Banking System. NC Banking Inst., 2, 230. [11] Claessens, S., & Laeven, L. (2004). What drives bank competition? Some international
evidence. Journal of Money, credit, and Banking, 36(3), 563-583.
[12] Craig, B. R., Dinger, V. (2013). Deposit market competition, wholesale funding, and bank risk. Journal of Banking & Finance, 37(9), 3605-3622.
[13] De Nicoló, M. G., Jalal, A. M., & Boyd, J. H. (2006). Bank risk-taking and competition revisited: New theory and new evidence (No. 6-297). International Monetary Fund.
[14] De-Ramon, S., & Straughan, M. R. (2016). Measuring competition in the UK deposit-taking sector.
[15] Degryse, H., Kim, M., & Ongena, S. (2009). Microeconometrics of banking: methods, applications, and results. Oxford University Press, USA.
[16] Delis, M., Iosifidi, M., & Tsionas, M. G. (2017). Endogenous bank risk and efficiency. European Journal of Operational Research, 260(1), 376-387.
[17] Drechsler, I., Savov, A., & Schnabl, P. (2017). Banking on deposits: Maturity transfor-mation without interest rate risk.
[18] EUR-Lex (2017). Revised rules for payment services in the EU. Retrieved January 16, 2018, from http://eur-lex.europa.eu/legal-content/EN/LSU/?uri=CELEX%3A32015L2366
[19] Fiordelisi, F., Marques-Ibanez, D., & Molyneux, P. (2011). Efficiency and risk in Euro-pean banking. Journal of Banking & Finance, 35(5), 1315-1326.
[20] Freixas, X., & Rochet, J. C. (2008). Microeconomics of banking. MIT press.
[21] Giedeman, D. C. (2004). The Riegle-Neal Act and local banking market concentration. International Advances in Economic Research, 10(4), 328-336.
[22] Goetz, M. R., Laeven, L., & Levine, R. (2016). Does the geographic expansion of banks reduce risk?. Journal of Financial Economics, 120(2), 346-362.
[23] Jiménez, G., Lopez, J. A., & Saurina, J. (2013). How does competition affect bank risk-taking?. Journal of Financial stability, 9(2), 185-195.
[24] Keeley, M. C. (1990). Deposit insurance, risk, and market power in banking. The Amer-ican Economic Review, 1183-1200.
[25] Laeven, L., & Levine, R. (2007). Is there a diversification discount in financial conglom-erates?. Journal of financial economics, 85(2), 331-367.
[26] Lastra, R., Bholat, D., Markose, S., Sen, K., & Miglionico, A. (2016). Non-Performing Loans: Regulatory and Accounting Treatments of Assets. Bank of England, Staff Work-ing Paper No. 594.
[27] Matasar, A. B., Heiney, J. N. (1999). The changing landscape of american banking: the impact of Riegle-Neal. International Advances in Economic Research, 5(1), 93-107. [28] Meeker, L. G., & Gray, L. (1987). A note on non-performing loans as an indicator of
[29] Mishkin, F. S. (1999). Financial consolidation: Dangers and opportunities. Journal of Banking & Finance, 23(2-4), 675-691.
[30] Schaeck, K., Cihak, M. (2014). Competition, efficiency, and stability in banking. Finan-cial Management, 43(1), 215-241.
[31] Schaeck, K., Cihak, M., & Wolfe, S. (2009). Are competitive banking systems more stable?. Journal of Money, Credit and banking, 41(4), 711-734.
[32] Strahan, P. E. (2003). The Real Effects of US Banking Deregulation. The Federal Reserve Bank of St. Louis, Review, 85.
[33] Van Leuvensteijn, M., Bikker, J. A., Van Rixtel, A. A., & Sørensen, C. K. (2011). A new approach to measuring competition in the loan markets of the euro area. Applied Economics, 43(23), 3155-3167.
