Master Thesis Finance
Bank risk-taking, ownership structure and monetary policy:
Evidence from the U.S.
Michiel Smeenk
",∗, supervised by Dr. Michalis Zaouras
"" Faculty of Economics and Business, University of Groningen, The Netherlands
ARTICLE INFO ABSTRACT
8 June 2017 JEL classifications: E52 G21 G32 Keywords: Bank risk-taking Monetary policy
Bank ownership structure Default risk
This paper conducts an empirical assessment of loan pricing policies, as a proxy for risk-taking, of 88 listed U.S. banks contingent on monetary policy, default risk and various bank ownership structures. The focus is on the effect varying forms of bank ownership structure have on its risk-taking profile when confronted with monetary policy changes. The results show that monetary policy easing leads to increased bank risk-taking when banks are characterized by managerial ownership, low shareholder concentration and state involvement. From the central bank’s perspective, the findings show the importance of identifying the reaction from diversely owned banks on the implementation of monetary policy.
1. Introduction
Are central banks able to effectively conduct and communicate their monetary policy when bank risk-taking incentives are different across diversely owned banks? The recent financial crisis has raised awareness of the effect that monetary policy transmission has on bank risk-taking behavior (Delis and Kouretas, 2011; Ioannidou et al., 2015; Jimenez et al., 2014). According to
* Corresponding author at: Faculty of Economics and Business, student number S2409763, Nettelbosje 2, 9747 AE Groningen, NL.
Maddaloni and Peydró (2011), the main causes of the crisis include softening of lending standards, increased securitization activities, prolonged periods of low short-term interest rates and mediocre regulations on bank capital. These developments have directly incentivized banks to increase their risk-taking behavior following low short-term interest rates by softening the lending standards and consequently materialization of risks in the economy. The prolonged low short-term interest rate environment has eased banks’ risk tolerance manifested by lower risk compensation requirements, known as the risk-taking channel of monetary policy (Borio and Zhu, 2012). Greater awareness for the link between monetary policy and bank risk-taking has hitherto not resulted in more attention for the effect of different bank ownership structures interacted with monetary actions on a bank’s risk profile. Surprisingly, since many studies have stressed the importance of ownership structure on bank risk-taking behavior (see, e.g., Jensen and Meckling, 1976; Saunders et al., 1990; Laeven and Levine, 2009). Therefore, the focus of this paper is to scrutinize the effect that monetary policy transmission with various bank ownership structures has on bank risk-taking behavior.
Borio and Zhu (2012) argue that current macroeconomic models are not sufficiently adequate to capture the effect of the risk-taking channel because of changes in the financial world that reduce the effectiveness of monetary policy. Bank risk-taking is mainly determined by its ownership structure, based on managerial stake in the bank (see, e.g., Anderson and Fraser, 2000; Saunders et al., 1990; Ferri et al., 2014), shareholder concentration and corporate governance (see, e.g., Iannotta et al., 2007; Laeven and Levine, 2009). Following Borio and Zhu (2012), the risk-taking channel is recognized as the effect of alterations in the monetary policy rates on risk perception and risk tolerance leading to changes in pricing risk of financial assets and the issuing of loans (see Section 2.1). An important paper for this research written by Paligorova and Santos (2017), measures the risk-taking channel of banks through the loan spread differential offered, simply the mark-up over costs, based on lagged values of monetary policy and firm default risk. Based on their loan spread differential theory, a similar approach is adapted to measure bank risk-taking in this paper. It elaborates on their work by including the ownership structure of banks into the analysis and looking to find the combined effect with monetary policy on bank risk-taking behavior.
here. For instance, Dell’Ariccia et al. (2011) define risk-taking as the banks’ choices that raise volatility of profits. Regarding the effects, Pathan (2009) argues that solid executive boards of banks have a positive impact on taking while high CEO power suppresses the banks’ risk-taking behavior. Furthermore, through financial innovation, the ability of households and consumers to obtain financing has loosened due to the increased capability to spread risk amongst various financial instruments and markets (Rajan, 2006). The shift towards larger risk appetite by financial institutions is strengthened by monetary policy. The yields and prices of financial assets are subjective to monetary policy, thus affecting financial institutions in their risk setting policy. Monetary policy easing1 (lowering short-term interest rates by the central bank) with near ground-level short-term interest rates affects banks’ risk-taking behavior as well. Rajan (2006) argues that banks are eager to maintain relatively high nominal rates of return acquired during preceding periods characterized by high short-term interest rates and therefore undertake excessively risky projects to achieve these returns. Furthermore, banks can understate risk levels for future investment decisions based on current conditions (Yellen, 2011; Adrian and Shin, 2009). In the opposing scenario, monetary policy tightening surges the short-term interest rate towards a higher level and consequently decreases the attractiveness of borrowing, potentially leading to underinvestment. Among other authors, Allen (1988) measures bank risk-taking by looking at the difference between the loan and deposit rates. However, this research focuses on the mark-up a bank demands from its borrowers when issuing a loan, calculated as the natural logarithm of the spread above LIBOR2, following the concept developed by Paligorova and Santos (2017).
Although the effect of ownership on bank risk-taking is discussed in many studies, (see, e.g., Laeven and Levine, 2009; Bouwens and Verriest, 2014), limited research has been done in combining the effect of monetary policy and ownership structure on bank risk-taking. A recent study by Figueira et al. (2016) examines this relation and found that ‘heterogeneity in organizational forms accounts for a differential impact on monetary policy on financial intermediaries’ risk-taking’. However, to the authors’ best knowledge, little research has been conducted on the various different measurements of a bank’s ownership structure in conjunction
1 The terms ‘monetary easing’, ‘monetary expansion’ and ‘low interest rates’ are used interchangeably throughout
the paper. The same holds for ‘monetary tightening’, ‘monetary contraction’ and ‘high interest rates’.
2 London Interbank Offered Rate. LIBOR is benchmark interest rate at which banks supply each other with
with monetary policy. Therefore, the aim of this paper is to identify this potential effect by considering different bank ownership structures.
The remainder of the paper is organized as follows. Section 2 reviews the relevant literature and explains the theoretical concepts for the analysis. Section 3 discusses various economic theories to develop testable hypotheses. Section 4 describes the data sources, methodology and the process of selecting the sample and constructing the variables. Section 5 illustrates the empirical results and their economic relevance. Section 6 will discuss alternative possibilities by reporting some robustness tests and a discussion of the limitations and implications of the results. Thereafter, section 7 concludes the paper along with the advances for future research.
2. Literature review
This section outlines the related literature on bank risk-taking and the mechanisms influencing risk-taking behavior. The first factor is to what extent monetary policy affects bank risk-taking, explained through the risk-taking channel, before advancing to the effect of different ownership structures. Thereafter, the combined effect of ownership structure and monetary policy on bank risk-taking is examined. Lastly, various additional factors are discussed.
2.1. The risk-taking channel of monetary policy
(Rajan, 2006). The third effect emanates from the ability of the central bank to actively and effectively communicate policies and its reaction function. The transparency of the central bank’s policy decisions and their credibility of actually pursuing this strategy, affects uncertainty and thus risk-taking. Furthermore, the effectiveness of the central bank’s reaction function to combatting large downside risk exposure affects risk-taking behavior (Farhi and Tirole, 2012).
The existence of the risk-taking channel is confirmed by Angeloni et al. (2015), by showing the negative effect of monetary rates on the banks’ eagerness to take risks, particularly evident regarding funding. Their time series evidence is based on a common VAR reinforced with multiple bank risk proxies to support the relation by altering the banks’ funding structures and riskiness of their assets. According to Paligorova and Santos (2017), monetary policy easing reduces the perceived riskiness of bank loans and increases the ability of risky borrowers to acquire funding. In addition, banks with intrinsically higher risk tolerance in their lending standards are more likely to charge riskier borrowers lower premiums in times of monetary easing relative to monetary tightening. Their findings are supported by providing evidence from U.S. banks based on the corporate loan market. The authors use the Federal Reserve System’s SLOOS3 to measure bank management’s perceived changes in lending standards for commercial and industrial loans.
In another study on lending standards, Maddaloni and Peydró (2013) show that low short-term interest rates augment the possibility to acquire loans for households and corporates based on bank lending standards from the Euro-area and U.S. Their analysis provides evidence that lending standards are down-sized particularly by low short-term monetary rates comparative to low long-term rates. Furthermore, through the risk-taking channel of monetary policy, banks’ increased risk tolerance eases the provision of loans and thereby accumulates significant risk on their assets. An interesting finding from the paper is the existence of an enforcing effect on relaxing lending standards in the case of prolonged monetary easing, called the ‘too low for too long’ effect. This effect is observed when lending standards tighten in response to tighter monetary rates in the preceding period, concluding that lending standards require an adjustment period.
However, many other factors significantly impact the relation between monetary easing and bank risk-taking, e.g. banks’ ability to adjust capital structures raises incentive to increase leverage and risk in the event of lower interest rates (Dell’Ariccia et al., 2013). Furthermore, the evidence in Delis and Kouretas (2011) shows the importance of bank specific characteristics on
its risk-taking profile, especially the degree of non-traditional bank operations and capitalization. Banks with higher equity capital and lower off-balance sheet items are less prone to excessive risk-taking in response to low interest rates. In addition, Valencia (2014) argues that if the bank is equity financing constraint, the size of the monetary shock is decisive in explaining the effect on bank risk-taking. Modest contractions in the interest rate lead to diminished risk-taking behavior and considerable shocks to elevated risk-taking. If the equity financing constraint is abandoned, monetary easing unambiguously leads to excessive bank risk-taking.
2.2. Ownership structure and bank risk-taking
A vast amount of literature has examined the relation between all types of ownership structures and bank risk-taking. Most studies confirm that the organizational form of a corporation or bank affects its risk-taking behavior. According to an early study conducted by Saunders et al. (1990) on stock ownership and bank risk-taking, stockholder controlled banks, wherein managers hold a relatively large proportion of the stocks and act as shareholder value maximizers, are more induced to take on additional risk relative to banks that are managerially controlled. The authors argue that the stockholders’ capability to enhance the value of their equity call options is through accumulating the riskiness of the underlying bank’s assets. Therefore, if managers hold a relatively large proportion of stocks in the banking firm, managers will strive for value maximization and mount their risk profile. Secondly, the authors show that periods of deregulation relative to more regulated policy periods amplify risk-taking behavior, and stockholder controlled banks increase their risk profile more compared to managerial controlled banks. However, they face considerable critique on their empirical analysis, as cross-sectional differences are not taken into account (Mullins, 1992). In continuation of their research many other authors (Bouwens and Verriest 2014; Pathan 2009) have found statistically significant results showing the positive effect of managerial ownership on risk-taking behavior.
bank’s ownership structure and risk-taking behavior empirically. Their study differs from Saunders et al. (1990) as they assess the effect of bank regulation on its risk-taking behavior based on the comparative power shareholders have in the ownership structure. However, their analysis complements the work by Saunders et al. (1990) as the findings confirm that stockholder controlled banks with relatively powerful owners are induced to take additional risks. Their empirical results are in line with established theory that large equity holders require higher risk-taking and persuasively use their power to influence management.
Their findings are concurred by Ferri et al. (2014), among others, who mention the gravity of managerial stake in the firm to incentivize firm value enhancement. Stakeholder banks are therefore not mainly striving for (short-term) value increases. Stakeholder banks are less likely to raise additional debt to acquire a higher return on equity (Ayadi et al., 2009). Furthermore, these banks face relatively large barriers, compared with shareholder banks, in raising external capital (Ayadi et al., 2010). All in all, these results imply that stakeholder banks have a lower tendency to increase their risk profile to benefit equity holders. These findings are confirmed by Bouwens and Verriest (2014), who show the risk-taking behavior of managers with a high stake in the bank and compare their behavior with banks that employ managers with a low stake. They find that high stake managers are less prone to excessive risk-taking as their exposure to syncretic risk is larger. Another interesting finding suggests that large external equity groups are able to influence directors in their risk-taking behavior.
Iannotta et al. (2007) distinguish three kinds of bank ownership structures, next to privately owned and government-owned banks, they incorporate mutual banks to their analysis. Based on a sample from large European banks, they conclude that the portfolio of government-owned banks consists of relatively weaker quality loans and face a greater threat of insolvency than other types of banks. These banks are able to dodge indirect costs of their inferior asset quality and lower profitability of intermediation activities due to assurances of the government. Furthermore, according to their results, mutual banks perform best on loan quality and asset risk minimization compared with private and public banks. In conclusion, a bank’s ownership structure influences risk-taking, where public sector banks are the least risk averse.
1992 and 2000, compensation based on stock options generally follows deregulation. They find that the construction of managerial compensation incentivizes risk-taking through the possibility of acquiring stock option rewards as a growing percentage of total compensation. Their findings are in line with other studies as stock options for performance compensation increases the manager’s stake in the bank and thus induces additional risk-taking as mentioned earlier.
According to Shleifer and Vishny (1986), monitoring of management is intensified with ownership concentration through improved corporate control. Shareholders in banks with diffused ownership, have lower incentive to monitor management. On the contrary, with concentrated ownership, large shareholders are heavily invested and therefore shirking costs will mainly be borne by them, raising incentives to monitor management. Therefore, Shleifer and Vishny (1986) state that concentration of ownership reduces bank risk-taking. However, other studies (see, e.g., Gomes and Novaes 1999, 2005) have advocated that conflicts of interests could arise between minority and majority shareholders. In addition, they state that the efficiency of the decision-making process could be hampered due to the presence of multiple controlling shareholders. Furthermore, in the financial sector, regulation increases the effectiveness of managerial discipline and intensified regulation could replace ownership monitoring (see, e.g., Demsetz and Lehen, 1985; Elyasiani and Jia, 2008).
2.3. The combined effect of ownership structure and monetary policy on bank risk-taking
The current literature on the separate and joint effect of the ownership structure of banks and monetary policy transmission on bank risk-taking is rudimentary. The bank’s ownership structure can be examined from several perspectives, i.e. state- and privately owned banks, shareholder- and stakeholder oriented banks.
2.3.1. State- and privately owned banks
monetary tightening, state-owned banks have greater ability to offset reduced loan demand by collecting supplementary deposits as households have the insurance that governments support these banks through regulations (e.g. increase participation in bank capital or raise deposit guarantee) in case of financial distress. Their analysis continues on Cecchetti and Krause (2001), who investigate the relation between government participation in the financial sector and varying economic policy outcomes based on a sample of developed and developing countries. They argue that as the proportion of state-owned banks falls, central bankers have more instruments to lower inflation and stabilize output volatility. In other words, the effectiveness of monetary policy is particularly dependent on the proportion of privately owned banks as their responsiveness to changes in the interest rates are more pronounced.
2.3.2. Share- and stakeholder oriented banks
2.4. Other factors influencing bank risk-taking
Besides monetary policy and ownership structure, various other factors could explain bank risk-taking behavior. In the literature, competition in the banking sector plays a major role on bank risk-taking behavior. Increased competition has a negative impact on the bank’s solvency and disturbs the financial stability of the sector. As a result, banks are threatened to lose franchise value and therefore have greater incentive to pursue riskier strategies (see, e.g., Keeley, 1990; Besanko and Thakor, 1993).
According to Boyd and Nicoló (2005), current literature states that banks faced with increased competition, and consequently deteriorating moral hazard, have a tendency to alleviate their risk-taking standards. However, they argue that the relation between the number of bank competitors and risk-taking behavior is negative, known as the risk-shifting effect. Boyd and Nicoló (2005) state that increasing competition in the market leads to a reduction of the loan rates, resulting in lower default probabilities and hence safer banks. In extension of the Boyd and Nicoló (2005) model, Martínez-Miera and Repullo (2010) combine the two views presented above to show their individual credibility by allowing for a nonlinear relation between market concentration and bank risk-taking behavior. They mention the existence of a ‘margin’ effect as a consequence of their nonlinear model causing bank revenues to decline due to lower interest payments. In competitive banking markets, where the margin effect is stronger than the risk shifting paradigm, high competition has detrimental effects on bank solvency and increases default risk. On the contrary, in relatively concentrated markets, bank default risk is negatively related to
increasing competition.
3. Hypothesis development
In the following section four hypotheses are developed regarding the relation between both monetary policy and ownership structure on bank risk-taking. The first hypothesis is commonly applied and refers to the risk-taking channel of monetary policy:
H1: Monetary policy easing increases bank risk-taking.
The intuition behind this hypothesis is that lower short-term interest rates lead to increased risk-taking. Several studies that are based on European and U.S. bank data, find evidence using expected default frequencies (Altunbas et al., 2010), lending surveys by banks (Maddaloni and Peydró, 2011) and loan ratings (Dell’ Ariccia et al., 2017). Based on research by Borio and Zhu (2012) and Dell’ Ariccia et al. (2013), the expected coefficient of monetary policy on bank risk-taking is negative.
The following three hypotheses state the effect of various forms of ownership structure on bank risk-taking. In addition, these hypotheses incorporate the effect of the interaction of monetary policy and ownership structure on bank risk-taking. Previous research has mainly focused on their separate effects and uses a different measure of ownership structure. Although this interaction effect has recently been examined (see, e.g., Laeven and Levine, 2009; Figueira et al., 2016), there is ample room to expand on their research. Particularly, if the findings show a significant relation, it would be intriguing to discover the central bank’s ability to modify monetary actions to increase implementation efficiency in the financial sector.
H2: Managerially owned banks charge lower loan spreads.
According to Shleifer and Vishny (1986), highly concentrated banks are induced to take less risk than more diffusely owned banks due to better corporate control. Furthermore, Iannotta et al. (2007) mention that increased concentration of ownership is associated with lower asset risk. However, other authors have argued the reverse effect of ownership concentration due to the occurrence of conflicts of interest between small and large shareholders (see, e.g., Gomes and Novaes 1999, 2005). In addition, they state that the efficiency of the decision-making process can be hampered due to the presence of multiple controlling shareholders. Despite arguments provided above, most authors follow Shleifer and Vishny (1986) and a positive link is hypothesized between diffused ownership and increased risk-taking.
H3: Low shareholder concentration induces banks to take more risk.
Cecchetti and Krause (2001) describe the varying effect monetary policy has on banks that are either state- or privately owned. Following their analysis, central bank monetary policy actions have a relatively stronger impact on privately owned banks compared to state-owned banks. Thus, highlighting the importance of state involvement in a bank’s ownership structure. Furthermore, state-owned banks have lower incentives to generate profits and pursue more social initiatives. Therefore, banks characterized by state ownership are expected to react less extreme to monetary policy changes and are more stable. Furthermore, Shleifer and Vishny (1997) argue that bureaucrats control state-owned banks and strive for politically motivated decisions instead of shareholder wealth maximization. Concomitant with this argument, Iannotta et al. (2013) show, based on Western European countries, the similar effect of government ownership on the riskiness of the portfolio of especially commercially oriented banks. This leads to the following prediction:
H4: Banks with higher state involvement are more risk averse than privately owned banks.
4. Data and methodology
4.1. Data sources
The data originate mainly from two sources, namely Orbis bank focus and the Thomson Reuters Datastream. The Orbis bank focus database contains detailed information from U.S. and non-U.S. financial institutions. Furthermore, the Thomson Reuters Datastream is used to access a wide array of macroeconomic and financial data of financial institutions.
Following Paligorova and Santos (2017), who use the all-in drawn loan spread over LIBOR, an equivalent proxy for bank risk-taking is employed in this paper. Their study is based on thousands of commercial and industrialized loans taken out by U.S. companies that are supplied by U.S. banks between 1990 and 2008. For instance, if a bank employs a relatively low spread above LIBOR, it signals a low risk profile. The authors have chosen this pre-crisis period to estimate clearly the effect of monetary policy changes on bank risk-taking and avoid periods of constant artificially low interest rates set by the Federal Reserve as a response to the eruption of the financial crisis. In this study, a slightly different approach is taken, wherein the Orbis bank focus database is accessed to collect quarterly balance sheet data of listed U.S. banks from 2000 to 2008, to construct a variable of the loan spread differential. Furthermore, the Orbis bank focus database provides data to group banks into categories based on different forms of ownership structure. In addition, the database is used to construct the default risk variables as well as many of the set of control variables.
The Thomson Reuters Datastream provides economic and equity data of the banks. In addition, the Thomson Reuters Datastream provides general economic trend variables (e.g. LIBOR, federal funds rate and GDP growth).
4.2. Methodology
time-series regression for 88 U.S. banks over the nine-year period 2000-2008. Furthermore, the several specifications (1) – (4) in table 2 and 3 are restrictive forms of Eq. (1).
Log loan spread(i,t) = a + b1 Mpolicy(t-1) + b2 Default risk(i,t-1) + b3 Ownership(i) + b4 Mpolicy(t-1) x
Default risk(i,t-1) + b5 Mpolicy(t-1) x Ownership(i) + e(i,t)4 (1)
The dependent variable for bank i at time t is bank risk-taking proxied by the log loan spread differential. It is measured as the spread above LIBOR, a benchmark interest rate at which banks supply each other with short-term loans, to be interpreted as a bank’s cost of acquiring funding, specified to the 3-months delayed United States Interbank LIBOR for this research. The log of the loan spread is a direct measure of a bank’s willingness to supply loans at a deliberately chosen interest rate. This interest rate reflects the bank’s risk-taking behavior because a low spread over LIBOR signals higher risk-taking behavior dependent on the stance of monetary policy and default risk. Therefore, if borrowers are charged less during monetary policy easing, banks require lower risk premiums on their loans. Therefore, the variable is found to be an impeccable proxy for bank risk-taking.
The key coefficient of interest from regression Eq. (1), b5, measures the interaction between
monetary policy and the forms of bank ownership structure. These interaction terms capture the effect interest rate differentials have on bank risk-taking as a result of different organizational forms. These ownership dummies include, managerial ownership, diffused ownership and state ownership. These three ownership variables are chosen in particular because several authors have stressed the importance of managerial stake (see, e.g., Anderson and Fraser, 2000; Saunders et al., 1990; Ferri et al., 2014), ownership concentration (Shleifer and Vishny, 1986) and ownership structure (see, e.g., Iannotta et al., 2007; Laeven and Levine, 2009) to measure bank risk-taking. The effect of the interaction term, Mpolicy(t-1) x Ownership(i), is described in Section 5.
Next, banks can increase their credit supply for other reasons than changes in their risk profile. Therefore, bank-fixed effects are included to account for these issues, which are depicted in table 3. In addition, bank fixed effects diminish conflicts of unobserved heterogeneity at the
4 where Mpolicy
(t-1) = the federal funds rate and Romers in quarter t-1, Default risk(i,t-1) = Z-score and Default
bank level that could explain bank behavior in selecting loan spreads. Secondly, for all the specifications, Hausman tests are performed to check for correlated random effects (Appendix A).
4.3. Set of control variables
The following section describes the set of control variables and their predicted effect on loan rates. The variable Mpolicy(t-1) is the monetary policy situation one quarter before a loan is
issued. The main components of the monetary policy variable are the federal funds target interest rate (FF rate) and Romer and Romer (2004)’s monetary policy shock variable (Romers). One quarter lagged values are taken into account for the period before the loan is originated. Firstly, the target interest rate set by the Federal Reserve is the overnight interest rate at which depository financial institutions lend balances to other financial institutions through the Federal Reserve. According to Piazzesi and Swanson (2008), these target rates encompass the actual stance of monetary policy just after the FOMC5 meeting, however, other anticipated and unanticipated externalities of monetary policy could already have influenced loan rates. Therefore, they mention that futures rates from the Federal Reserve are predictors of the market’s expectations of the monetary policy stance. Secondly, a variable of monetary policy shocks is implemented into the model to avoid this problem, calculated as the sum of monthly fluctuations from expected rates. The approach is based on Romer and Romer (2004)6, who analyze the internal forecasts (inflation and real activity) of the Federal Reserve against its intentions of altering the federal funds rates around FOMC meetings. Their findings show strong evidence of the implications monetary policy has on inflation and output. The shock variable removes a significant share of the endogenous movements that occur between macroeconomic trends and the federal funds rate. Intuitively, the monetary policy shock variable should have weak correlation with economic events to measure more thoroughly the actual effect of the monetary policy shock instead of simultaneous movements with economic events. Appendix B shows the correlation matrix of economic variables, where the
5 Federal Open Market Committee.
6 The data are acquired from the American Economic Association website. To acquire data for my full sample
monetary policy shock variable is not highly correlated with any of the other main economic variables, showing a maximum correlation of -0.541 with market volatility.
The variable default risk is measured along two dimensions as done by Paligorova and Santos (2017). First of all, a dynamic model of bankruptcy filings is employed to identify default probabilities (Campbell et al., 2008). Secondly, Altman (1968)’s Z-score is computed from balance sheet data to measure the likelihood of bankruptcy based on financial ratios. Campbell et al. (2008) estimate their model based on bankruptcy filings utilizing various new equity market and accounting variables to measure financial distress empirically in two dissimilar ways. They examine unequivocal bankruptcy filings according to Chapter 7 or 11 of the bankruptcy code7. In addition, the authors rely on the principal credit rating agencies to come up with a broader definition of failure by looking at default ratings, financially driven de-listings and bankruptcies. The default probability variable solely has negative values due to the high negative coefficients from the bankruptcy model, lower values implying a diminished probability of bankruptcy. Campbell et al. (2008) assemble all the data available up to and including a certain year, e.g. 1990, and then calculate fitted probabilities of failure at the beginning of January the next year, i.e. 1991. All the input measures (leverage, profitability, cash, ROE, volatility, relative size, market-to-book ratio and stock price) are constructed for the full sample of U.S. banks to calculate default probability.
The second model used to predict bankruptcy is the Z-score test, originally based on Altman (1968). This test uses statistical techniques to estimate corporate bankruptcy within the following two years. Altman (1968) finds a large number of significant variables in explaining bankruptcy from income statement and balance sheet data. Subsequently he clusters the variables into five ratio categories; i.e. leverage, solvency, liquidity, profitability and activity, before selecting five variables that have the highest overall predictive power on bankruptcy. The coefficients are constructed by iteratively running computer models to estimate the highest contribution of the full profile. These coefficients are standardized, time-consistent and provide firm sustainability estimates and therefore the index is still commonly used by researchers. The original Z-score is used by numerous authors in defining firm bankruptcy (see, e.g., Hillegeist et
7 The U.S. Bankruptcy code is the source of bankruptcy law. Chapter 7 governs the process of liquidation and
al., 2004; Bharath and Shumway, 2008; Paligorova and Santos, 2017). However, the original Z-score test was mainly constructed for manufacturing firms, whereas the revised application (Altman, 1983; 1993) of the test predicts the financial distress status for non-manufacturing industries and is commonly used to assess public-sector banks (see, e.g., Vaziri et al., 2012; Pradhan, 2014; Kokkoris et al., 2016). The revised Z-score model is computed according to the following formula:
Z-score = 6.56A + 3.26B + 6.72C + 1.05D, (2)
where A is working capital / total assets, B is retained earnings / total assets, C is earnings before interest and taxes / total assets, D is book value of equity / total liabilities and Z is the overall index. Z-score contains positive values, where higher values reflect greater solvency and lower liquidity risk, thus reduced bank default risk. To ensure robustness to bank bankruptcy measures, both measures of default risk are included, lagged one quarter before origination of the loan to prevent contemporaneous correlations.
shares, thus virtually reflecting non-state ownership. This dummy variable, ‘State involvement’, does not explicitly consider full state control but rather examines the level of involvement.
In addition, several time-variant loan-, bank- and economic specific control variables are included in the regression that impact the loan spread requested by banks to avoid omitted-variable bias. The goal is to incorporate both idiosyncratic (i.e. loan and bank controls) and systemic (i.e. economy controls) elements, to fully assess the bank’s risk portfolio and individual risk-taking motives.
Two key loan-specific variables are used, namely log loan loss provisions (L llp)8 and the log of the loan amount (L amount). The loan loss provisions variable is a non-cash expense that banks utilize to consider future losses on borrowers’ loan defaults. According to Laeven and Majnoni (2001), loan loss provisions can enhance the overall capital regulatory framework and thereby explain differences in the loan prices demanded by banks. The other loan-specific variable entails the size of the loan and has an ambiguous effect on loan spreads. Large sized loans are generally riskier but economies of scale in monitoring and processing the loans allows for a risk reduction (Paligorova and Santos, 2017).
Next, the bank-specific control variables are discussed. Following Paligorova and Santos (2017), who incorporate some of these bank controls into their analysis, the log of the assets (L assets) is included to control for banks’ size. Furthermore, the return on assets (ROA) controls for improvements in the financial position, incentivizing banks to decrease the loan spread. On the contrary, the volatility of the return on assets (ROA vol) and the charge offs on net loans (Charge-offs), measured as a certain amount of debt severely delinquent borrowers are unable to repay, capture the fluctuation in funding costs. Another bank control variable is the capital-to-assets ratio (Capital) and a negative relation is expected. According to Boot et al. (1993) low capitalized banks misuse their informational advantage and charge higher rates to borrowers that are liquidity constraint and permanently require bank funding. In addition, liquidity (Liquidity) is expected to be negatively related to loan spread as banks with liquid assets can easily acquire funding. In line with the liquidity analysis, the quarterly subordinated debt to assets ratio (Subdebt) is expected to negatively impact loan spreads. According to Figueira et al. (2016), profit before taxes over total assets (Profitability) and the cost-to-income ratio (Efficiency) have a positive effect on the overall
risk-taking profile of the bank. In addition, they argue that it is important to control for the bank’s liability structure by looking at the total deposits to liabilities ratio (Deposits). To conclude with the bank control variables, asset securitization is proxied by the off-balance sheet items to total assets ratio (Securitization).
Lastly, general macroeconomic and financial control variables are included as they have a bearing on bank’s risk-taking behavior. Therefore, quarterly economic growth figures (GDP growth) and the S&P500 market index volatility (Market vol) are included in the analysis. Stock market improvements could alleviate risk burdens and will lead to additional risk-taking. Furthermore, including the yield difference between five-year and one-year zero coupon bonds (Slopeyc), controls for modifications of short-term rates. Finally, the quarterly spread on AAA- versus BBB-bonds (L yield spd) controls for risk preferences of investors.
4.4. Sample characterization
The full sample in this paper consists of 116 U.S. banks9 compiled from the top 500 largest banks based on assets. The banks incorporated in the sample have sufficient quarterly data availability from the first quarter of 2000 until the fourth quarter of 2008, obtained from the Orbis bank focus database. Furthermore, all banks are listed on the two major U.S. stock exchanges, NYSE and NASDAQ10, thereby dropping six banks from the original group. To increase the accuracy of the test, several banks are excluded from the sample due to data availability regarding the construction of the default probability variable. Furthermore, five banks have disproportionally high Z-score values and loan loss provisions due to their large size compared to the other banks in the sample. Lastly, two banks demand unrealistically excessive loan spreads from their borrowers, thus these outlying points are erroneous and the banks are dropped from the analysis, resulting in a restricted sample of 88 U.S. banks. The full sample analysis is included in the Appendix as a robustness check (see Section 6.2; Appendix C1). To mitigate the effects of outliers, all the variables are winsorized at the 1% level. This technique transforms statistics through limiting extreme values from the data set at both ends. All regressions have a slight leptokurtic distribution,
9 The panel selection is based on the largest banks that have a certain assets base and are geographically dispersed.
These criteria aid to ensure that the sample corresponds to a substantial fraction of the magnitude of the banking system.
which is common when analyzing financial data11. Table 1 below presents the sample characteristics of all the variables included in the regression analysis.
Table 1
Summary statistics of all banks in the restricted sample from 2000 to 2008. The table presents means, medians, maximums, minimums, and standard deviations of all variables. FF rate and GDP growth reflect percentages, while all the bank controls with ‘(%)’ are ratios and have to be multiplied by 100 to reflect percentages. See Section 2.3 for the definitions of the control variables.
Mean Median Maximum Minimum Std. Dev. Obs.
L loan spd 5.768 5.808 6.416 4.789 0.412 3168 FF rate (%) 3.236 2.875 6.500 0.250 1.880 3168 Romers -0.035 -0.001 0.476 -0.947 0.337 2992 Loan controls Loan spd (bsp) 346.382 332.875 611.786 120.180 132.220 3168 L llp 7.343 7.313 12.630 0.000 2.270 3168 L amount 14.822 14.485 18.619 13.051 1.240 3168 Bank controls Z-score 1.013 0.999 1.606 0.629 0.188 3168 Default prob -11.554 -11.484 -8.323 -15.354 1.388 3168 L assets 15.275 14.911 19.007 13.628 1.205 3168 ROA (%) 0.016 0.017 0.038 -0.025 0.008 3168 ROA vol 0.206 0.092 1.928 0.000 0.321 3168 Subdebt (%) 0.004 0.000 0.040 0.000 0.010 3168 Charge-offs (%) 0.001 0.001 0.005 -0.000 0.001 3168 Liquidity (%) 0.058 0.048 0.216 0.016 0.036 3168 Capital (%) 0.129 0.125 0.263 0.068 0.034 3168 Profitability (%) 0.004 0.004 0.008 -0.006 0.002 3168 Deposits (%) 0.814 0.831 0.978 0.422 0.106 3168 Efficiency (%) 0.595 0.593 0.936 0.349 0.100 3168 Securitization (%) 0.213 0.182 0.724 0.009 0.134 3168 Economy controls L yield spd 0.082 -0.005 0.844 -0.496 0.399 3168 Slopeyc 0.761 0.757 2.473 -0.402 0.887 3168 GDP growth (%) 1.809 2.188 6.693 -5.976 2.408 3168 Market vol 20.330 19.110 39.810 11.400 7.157 3168
The median of the loan spread differential (Loan spd) measured in basis points is 332.875, implying that the median loan supplied by a bank to borrowers has a mark-up over costs of 3.33%. Several reasons can explain particularly high values of the loan spread differential, for instance, to combat increased uncertainty due to intensified interest rate volatility in the market. The minimum value is 120.180 implying a slight mark-up of 1.2%, logically banks will not lend money below their own cost of acquiring the funds. Secondly, banks face the risk of borrowers who default on
11 I have performed a Jarque-Bera test for normal distribution. Also after restricting the sample by eliminating
their loans, namely credit risk.12 A third source for relatively high loan spreads might have to do with liquidity risk, therefore banks impose liquidity premiums on loans to avoid becoming cash constraint (Prisman, Slovin, and Sushka, 1986) and having to borrow at higher costs (Angbazo, 1997). Lastly, banks protect themselves, by holding certain levels of capital, against expected and unexpected credit risk (Saunders and Schumacher, 2000). However, holding capital is relatively expensive compared to debt, which affects the bank’s behavior in setting its
loan prices.
Continuing with loan controls, the mean of the log-transformed loan loss provisions is 7.343 whereas the median is slightly lower with a value of 7.313. In the unrestricted sample, counterintuitively, a few banks utilized negative loan loss provisions which can be explained by perfect income smoothing. Laeven and Majnoni (2001) argue that bank earnings are to a lesser extent affected by varying losses on credit over the cycle. Therefore, loan loss provisions can balance differences in actual credit losses and average losses on credit by taking both positive and negative values dependent on the cyclical phase. However, after dropping various banks from the sample, the loan loss provision variable only has non-negative values. The loan loss provisions variable is log transformed to account for its large size distribution, where values of zero are log transformed after adding a small constant (+1). The log transformed loan size variable has a mean of 14.822 with a slightly higher median of 14.485.
Subsequently, the focus is shifted to bank controls that are calculated from bank balance sheet data. The banks from the sample have a mean log bank assets of 15.275 and a median of 14.911. The mean of charge-offs is 0.1% while the minimum value of charge-offs is potentially negative as written off debt could unexpectedly be repaid in a future accounting period, reducing the current net charge-offs value. The liquidity and capital ratios are averaged at about 6% and 13%, while the mean profitability ratio is particularly low at 0.4%, presumably due to the large assets base of the banks. Subordinated debt as a funding proxy only accounts for 0.4% and return on assets averages around 1.6%. Banks hold a relatively large amount of deposits resulting in a deposits to liabilities ratio of 81.4%. Also, efficiency and securitization give reasonable results of 59.5% and 21.3%, which is in line with other studies (Figueira et al., 2016).
12 In addition, Maudos and Fernández de Guevara (2004) suggest that high correlation between interest and credit
Lastly, the economic control variables are discussed. The log of the yield spread between primary and secondary bonds gives a mean value 0.082 log points, and negative values are the result of taking the log of small differences in the bond rates. The slope of the yield curve is 0.761, with generally higher yield to maturity for five-year zero coupon bonds as investors expect to be compensated for bearing more interest rate risk with long-term instruments. The negative values are due to two contiguous phases in the sample period, lasting for a couple of quarters. Uncertainty about interest rates increases could cause the yield for short-term bonds to exceed the yield for long term bonds for a brief period (Schrøder and Stampe Sørensen, 2010). Furthermore, the quarterly GDP growth has averaged around 1.81%, with flat peaks and a sharp downturn at the wake of the financial crisis. Furthermore, the market volatility index mean value is 20.33. The federal funds target rate had a mean of 3.24% over the sample period. The monetary policy shock variable shows on average stronger monetary easing than forecasted. Furthermore, year-fixed effects are included to account for annual effects on loan rate differences. Finally, to measure bank risk-taking behavior specifically through loan rates offered over time, based on macroeconomic fluctuations and the underlying bank fundamentals, the standard errors are robust and clustered at the bank-year level.
5. Empirical results
The purpose of this section is to report the results from the analysis, give the economic interpretation of the main findings and discuss the implications regarding the hypotheses. Specifications (1) – (4) in tables 2 and 3 provide the relation among monetary policy, default risk, bank ownership structures and other control variables on a quarterly basis over the nine-year sample period from 2000 to 2008.
Table 2
Base results: Loan Spread Differential. The dependent variable is the natural logarithm of the loan spread over LIBOR at origination. FF rate is the federal funds target short-term interest rate. Romers is the monetary policy shock variable based on Romer and Romer (2004). Default prob is the dynamic panel logit model based on Campbell et al. (2008). Z-score measures the possibility of bankruptcy within the following two years. FF rate and Romers interacted with Z-score and Default prob give the interaction terms between the federal funds rate and monetary policy shock with the proxies for default risk. Ownership x Mpolicy are the interaction terms between the proxies for monetary policy and the dummy variable, e.g. equaling 1 if the bank’s manager is also a shareholder. Similar descriptions apply to the dummy variables of diffused ownership, meaning low shareholder concentration, and state ownership. All the other variables are loan-, bank- and economic specific control variables. Robust standard errors (clustered at the bank-year level) are reported in parentheses. Significance level of 1% at ***, 5% at ** and 10% at *. The federal funds rate in column (1) has a p-value of 0.1172, reflecting near significance at the 10% level. (1) (2) (3) (4) FF rate -0.027 -0.050** (0.017) (0.022) Z-score -0.031 -0.146*** (0.042) (0.024) FF rate x Z-score -0.031*** (0.012) Default prob 0.007 0.010*** (0.006) (0.004)
FF rate x Default prob 0.001
(0.002)
Romers 0.386*** 0.249***
(0.094) (0.059)
Romers x Default prob 0.028***
(0.008) Romers x Z-score -0.186*** Bank ownership (0.056) Managerial ownership -0.034** -0.028** 0.026*** 0.026*** (0.014) (0.014) (0.008) (0.008) Managerial ownership x FF rate 0.016*** 0.014***
(0.005) (0.005)
Diffused ownership -0.020 -0.026 0.076*** 0.069***
(0.025) (0.026) (0.012) (0.012) Diffused ownership x FF rate 0.028*** 0.031***
(0.007) (0.007)
State involvement -0.029** -0.021 -0.017* -0.030***
(0.014) (0.015) (0.009) (0.009) State involvement x FF rate -0.001 0.001
(0.005) (0.005)
Managerial ownership x Romers 0.071*** 0.081***
(0.022) (0.022)
Diffused ownership x Romers 0.017 -0.008
(0.036) (0.036)
State involvement x Romers 0.045** 0.032*
Table 2 (continued) (0.033) (0.032) (0.031) (0.032) ROA -1.037 0.300 1.749 0.330 (1.775) (1.700) (1.595) (1.676) ROA vol 0.021 0.027* 0.034** 0.028** (0.014) (0.014) (0.014) (0.013) Subdebt -2.470*** -2.204*** -2.077*** -2.355*** (0.696) (0.700) (0.672) (0.671) Charge-offs 42.150*** 41.402*** 50.068*** 50.346*** (6.039) (6.060) (6.308) (6.302) Liquidity -0.051 -0.024 -0.002 -0.030 (0.117) (0.117) (0.113) (0.113) Capital 0.041 -0.573*** -0.631*** 0.058 (0.188) (0.183) (0.183) (0.190) Profitability 31.777*** 27.929*** 20.062*** 24.775*** (7.304) (6.880) (6.698) (7.174) Deposits 0.052 0.052 0.063 0.055 (0.046) (0.046) (0.045) (0.045) Efficiency 0.028 0.027 -0.016 -0.004 (0.062) (0.063) (0.062) (0.061) Securitization -0.184*** -0.217*** -0.180*** -0.145*** (0.045) (0.045) (0.044) (0.044) Economy controls L yield spd 0.469*** 0.461*** 0.048 0.073* (0.029) (0.030) (0.039) (0.038) Slopeyc 0.211*** 0.210*** 0.273*** 0.276 (0.016) (0.016) (0.010) (0.010) GDP growth 0.013*** 0.013*** -0.004 -0.003 (0.003) (0.003) (0.003) (0.003) Market vol -0.015*** -0.015*** 0.007*** 0.007*** (0.001) (0.001) (0.001) (0.001)
Year-fixed effects Yes Yes Yes Yes
R-squared 0.794216 0.792140 0.809735 0.811558
Observations 3168 3168 2992 2992
Table 3
Loan spread differential fixed effects model. Robust standard errors (clustered at the bank-year level) are reported in parentheses. Significance level of 1% at ***, 5% at ** and 10% at *. The dummy variables are omitted from the regression to estimate the bank fixed effects model. For all specifications, the various control variables as well as year-fixed effects are included in the regression but excluded from the table for space considerations. (1) (2) (3) (4) FF rate -0.013 -0.048** (0.015) (0.019) Z-score 0.177*** 0.077** (0.052) (0.038) FF rate x Z-score -0.039*** (0.010) Default prob -0.014** -0.018*** (0.006) (0.004)
FF rate x Default prob -0.001
(0.001)
Romers 0.357*** 0.244***
(0.076) (0.050)
Romers x Default prob 0.027***
(0.007)
Romers x Z-score -0.170***
(0.045) Managerial ownership x FF rate 0.015*** 0.013***
(0.004) (0.004) Diffused ownership x FF rate 0.018*** 0.017***
(0.006) (0.006)
State involvement x FF rate 0.002 0.001
(0.004) (0.004)
Managerial ownership x Romers 0.071*** 0.069***
(0.017) (0.018)
Diffused ownership x Romers -0.028 -0.028
(0.029) (0.028)
State involvement x Romers 0.045*** 0.028*
(0.017) (0.016)
Loan controls Yes Yes Yes Yes
Economy controls Yes Yes Yes Yes
Bank controls Yes Yes Yes Yes
Year-fixed effects Yes Yes Yes Yes
Observations 3168 3168 2992 2992
R-squared 0.868 0.868 0.890 0.888
5.1. Results for monetary policy on bank risk-taking
A lower federal funds rate signals expansionary monetary policy by the central bank mainly through open market operations13. In columns (1) and (2), the variable ‘FF rate’ states the moving
process of the quarterly federal funds interest rate reflecting the macroeconomic situation14. The results in table 2 are not consistent with the first hypothesis that monetary policy easing increases bank risk-taking as the estimation results prove the opposite. As can be seen from tables 2 and 3, the stand-alone coefficients of the federal funds rate are negative and partially significant. Another coefficient of interest in Eq. (1), the interaction term b4, measures whether banks adjust loan
spreads for different indicators of monetary policy and bank default risk. The results provide analogous conclusions stating that the federal funds target rate has an opposing effect when interacted with these two different measures of default. In column (1), the interaction between the federal funds target rate and Z-score is negative and statistically significant, thus implying that as the federal funds rate decreases, i.e. monetary policy easing, the mark-up over costs rises implying less risk-taking. On the contrary, column (2) shows the interaction effect of the federal funds rate and (Campbell et al., 2008)’s probability of default measure (Default prob), which is not significantly related to the loan spread differential, where insignificance is also observed in the full sample specification (Appendix C1). The results suggest that banks reduce risk-taking when faced with a decreasing federal funds rate. According to Paligorova and Santos (2017), this effect insinuates that periods of monetary policy easing coincide with a stagnating economic situation characterized by larger risk premiums and higher spreads on credit. The estimates, in terms of economic relevancy, imply that as the standard deviation of the federal funds rate decreases by one, the corresponding loan spread differential increases by 9.4% (-0.050 x 1.880). Thus, monetary expansion reduces bank risk-taking.
Intuitively, one would expect a high correlation between the LIBOR, which is embedded in the log loan spread calculation, and the federal funds interest rate as LIBOR is based on five currencies including the U.S. dollar which moves in line with monetary policy objectives of the U.S. Federal Reserve. Therefore, to examine if multicollinearity issues are present, the most widely used diagnostic for multicollinearity, variance inflation factors (VIF),15 are calculated for the independent predictors. The R2 used (0.791) in the regression indicates the coefficient of
14 Paligorova and Santos (2017) state the high correlation between the macroeconomic environment and monetary
policy.
15 VIF j =
%
%&'() . VIF measure the proportion of the variance a specific variable j in the regression shares with the rest
determination without including the predictor of interest (federal funds rate). The variance inflation factor gives a value of 2.67, which is lower than both rules of thumb, thus concluding that there is no multicollinearity problem. From the correlation matrix in Appendix A, can be seen that especially these two predictor variables are highly correlated, therefore the other estimates of regression coefficients are reliable and stable.
Fig. 1 below plots the movements of both the federal funds target rate and the monetary policy shock variable over the sample period. The correlation coefficient between both rates is 0.237 (Appendix A), reflecting a weak relation, thereby removing endogeneity problems that occur between macroeconomic trends and the federal funds rate. The p-value (0.000) indicates a statistically significant correlation coefficient at the 1% level, concluding that the correlation is different from zero.
Fig. 1. Monetary policy movements. The figure shows the movements across the sample period of both proxies for
monetary policy, the federal funds interest rate (solid blue line) and the monetary policy shock (striped red line) based on Romer and Romer (2004).
implying that larger positive shocks increase the loan spread differential, while negative shocks increase risk-taking, consonant with the first hypothesis. This finding is consistent with the research conducted by Jiménez et al. (2014), who argue that lower interest rates have a risk dampening effect as credit risk from outstanding loans declines. If the monetary policy shock decreases by one percentage point quarterly, thus additional to the expected change, the loan spread differential reduces by 24.9% or 38.6%, dependent on the regression specification.
The results show a positively significant relation between the monetary policy shock and default probability variable and a negatively significant relation with Z-score. Paligorova and Santos (2017) find the opposite effect and their results show that the relation between default risk and loan spread becomes weaker whenever prolonged periods of monetary easing occur. However, this paper analyzes bankruptcy probabilities and insolvency from the lenders’ perspective (i.e. the banks) whereas Paligorova and Santos (2017) analyzed the firms to whom the bank issued its loans. The results suggest that during monetary policy easing, banks, ceteris paribus, are more eager to reduce their risk-taking behavior by increasing the spread above funding costs charged to borrowers. Possibly, the improved credit position of borrowers due to the lower interest rates, might lead to higher loan demand, incentivizing banks to increase the spread to increase profitability. Furthermore, with stable monetary policy interest rates, improved financial position of the banks increases their risk-taking behavior and lowers the loan spread differential.
However, external factors could explain the effect of monetary policy shocks caused by adverse macroeconomic conditions. For instance, during periods of recession or weak economic growth, borrowers might demand less credit and this increases default risk for the bank. Therefore, with higher interest rates, the investment regime tightens and banks require higher margins on their loans to compensate for losses in other divisions.
5.2. Results for managerial ownership on bank risk-taking
interest rate, thus monetary easing, incentivizes banks characterized by directors and managers who are also shareholder to take additional risk by lowering the loan spread differential. This is in line with Chen et al., (2006) who state that executive boards possessing large proportions of the stock show additional incentive to take risk, which is amplified during monetary policy easing. One standard deviation reduction in the federal funds interest rate (Table 3, column 2) decreases the loan spread differential by 2.44% (0.013 x 1.880). Similarly, the loan spread differential declines by 2.39% (0.337 x 0.071) when the monetary policy shock variable is accounted for. These results hold for both proxies of bank default risk.
Prolonged periods of monetary easing could trigger more risk-taking by managers as they are required to meet a certain standard for a nominal return target, increasing the overall riskiness of the portfolio, including risk-taking on loans (Brunnermeier, 2001; Rajan, 2006). During monetary policy easing, as measured by both proxies, the overall exposure to credit risk falls and therefore managers’ value at risk decreases and consequently incentivizes them to take more risk by lowering the loan spread differential. This finding is in line with hypothesis 2, that managerial ownership positively affects bank risk-taking when confronted with lower interest rates. The results suggest that managerial bank ownership, discretely, alleviates risk-taking standards when controlled for the federal funds rate as the proxy for monetary policy, whereas, a significantly positive effect is reported on the log loan spread differential when the monetary policy shock variable is taken into account. This is an interesting contradiction that emerges from the results. A possible explanation could be that managers are more preservative when shocks are taken into account. These shocks occur unexpectedly and could potentially disrupt the bank’s business depending on its distress situation and severity of the shock.
5.3. Results for diffused ownership on bank risk-taking
bank’s performance and thus communicate with the shareholders about the future growth path the bank should pursue. The concentration of ownership is of paramount importance as managers are incentivized to act in a certain manner according to shareholder motives. Therefore, it is interesting to identify the effect of diffused ownership on bank risk-taking.
The interaction between diffused ownership and the federal funds rate is significantly positive in both column (1) and (2) indicating that diffusely owned banks are eager to increase risk-taking when the monetary policy rate decreases, thereby providing evidence for hypothesis 3. The managers of diffused banks have equal responsibility for bank performance but lower accountability towards their shareholders because they all hold low percentages of the shares and these managers have less to fear from controlling stakeholders compared to concentrated banks. In other words, intensified corporate control due to high ownership concentration reduces bank risk-taking (Shleifer and Vishny, 1986).
When incorporating monetary policy shocks into the analysis, diffused ownership, on a stand-alone basis, has a risk-reducing effect on the loan spread differential, increasing the loan spread with about 7% for banks with a shareholder base exceeding 100. Iannotta et al. (2007) argue that increasing concentration of bank ownership is accompanied with improving loan quality. Higher loan quality accumulates costs associated with issuing loans as the monitoring and credit underwriting costs mount. Therefore, diffused ownership suppresses loan quality and consequently costs fall, leading inferior borrowers to receive loans easier which incentivizes banks to increase the loan price, assuming LIBOR remains constant. This could potentially explain the positive effect of the loan spread, contradicting with earlier findings without taking into account monetary policy shocks.
5.4. Results for state ownership on bank risk-taking
This seems counterintuitive, and contrary to hypothesis 4, but state- or governmental interference in the bank could influence the selection process of borrowers applying for a loan. When borrowers are presumed to be safer, i.e. lower risk of default, the bank can adjust the loan spread differential downwards as lower risk premiums on the loans are suitable (Maddaloni and Peydró, 2011).
The positive significant coefficient on the interaction between monetary policy shock and state involvement implies that unexpected additional monetary expansion reduces the loan spread differential. This could be a response to the overall reduction in the risk portfolio of the bank when stronger monetary policy easing occurs (Bouwens and Verriest, 2014). As mentioned by Cecchetti and Krause (2001), banks with state involvement are less prone to follow monetary policy transmission when implemented. If the monetary shock decreases by 1 percentage point in a quarter, thus additional to the expected change, the loan spread differential drops by 24.9% or 38.6% dependent on the regression specification (see Section 5.1), compared to 3.2% to 4.5% when banks are partially owned by the state or government.
5.5. Loan-, bank- and economic specific control variables
spread have significant effects across multiple specifications implying that a higher yield spread between five-year and one-year zero coupon bonds is associated with wider loan spreads. The yield spread between AAA- and BBB rated bonds has a positive and significant effect across all specifications on the loan spread differential, implying that as BBB rated bonds become riskier and/or AAA rated bonds less risky, banks lower their risk exposure on loans. Aforementioned, low correlation between monetary policy shock and the macroeconomic environment is a vital condition to measure the independent effect on loan spreads.
All in all, the results with year and bank fixed effects show that monetary expansion through the federal funds rate reduces bank risk-taking compared to tightening while the loan spread differential goes down with expansionary shocks. These findings are robust to the several proxies for monetary policy and default risk. Ceteris paribus, across all specifications with significant estimates, monetary easing reduces the loan spread differential implying more risk-taking by the bank whenever the manager is also a shareholder, is characterized by diffused ownership and has higher state involvement.
6. Robustness tests and discussion
This section includes several robustness tests to evaluate the robustness of the main findings. The robustness tests will address the main regression Eq. (1) as well as alternative explanations of risk-taking incentives by banks. In addition, a discussion of the implications and limitations of the paper in relation to the research question is provided.
6.1. Full regression specification
full analysis. Suggesting that monetary policy easing, through a reduction in the federal funds rate, decreases bank risk-taking likely caused by the adverse macroeconomic situation prior to the decision to lower the interest rate. The results are robust for the monetary policy shock variable. Regarding the other important dummy variables concerning several bank ownership characteristics, the results are nearly perfectly aligned with the unrestricted analysis. However, the individual effect of managerial ownership provides two alternative interpretations in table 2 compared to the full regression (see Section 5.2). Only the interaction of state involved banks with the federal funds rate loses significance due to the elimination of banks. The results for the interactions of Z-score with monetary policy are in line with those presented in the benchmark specifications.
6.2. Alternative indicators of monetary policy
Many studies (see, e.g., Jiménez et al., 2014; Maddaloni and Peydró, 2011), have employed the LIBOR while several other proxies exist that effectively measure monetary policy. Therefore, this robustness check aims to assess the sensitivity of the results by replacing the LIBOR with the central bank’s official policy rate, computed as the quarterly average of the daily central bank rate (Figueira et al., 2016). The results presented in Appendix C2 indicate high similarities with the bank fixed effects model of table 3. However, a striking feature of the results show the change in sign for the monetary policy shock variable for specification (3), whereas inferences about the coefficient of Romers in specification (4) are meaningless due to loss of significance. This suggests that the results are partially driven by the chosen proxy for monetary policy and especially monetary policy shocks are dependent on which variable is used. Therefore, precise knowledge regarding the uncertainty of varying monetary policy measures on bank risk-taking is crucial to design a good policy rule. Overall, a uniform result arises, implying a negative relation between monetary expansion and bank risk-taking if the central bank’s official policy rate is implemented.
6.3 Overall results
take less risk by adjusting the loan spread differential upwards for borrowers, whereas the other federal funds rate estimations are insignificant. On the contrary, more severe monetary contraction compared to expectations decreases risk-taking behavior, as implied by the positive and significant coefficients of the Romers variable in specifications (3) and (4) of all four specifications. These findings are robust when incorporating different bank default risk variables as well as three bank ownership dummies. Furthermore, all estimations suggest that managerially owned banks take more risk during periods of monetary easing, robust to using alternative proxies for monetary policy. In addition, all specifications imply that diffusely owned banks take more risk when the federal funds rate decreases, however, monetary policy shocks appear to have no impact on the loan spread offered by diffusely owned banks. Banks wherein state or government parties own a small share up to full control, are generally more risk loving during monetary easing.
6.4 Alternative measure of bank risk-taking
