Earnings management, derivatives usage and risk-taking by
US Bank Holding Companies.
Word count: 11077
Patrick Uffels S2039206
MSc. Finance
MSc. Accountancy & Controlling
Supervisor MSc. Finance dr. P.P.M. Smid
1
Abstract
The goal of this paper is twofold. First, the relation between earnings management and bank risk-taking is researched. Second the relation between bank risk-taking and derivatives usage is studied. Based on a sample of 2247 banks and 11011 firm year observations two panel data OLS regressions with firm and period fixed effects are performed. The results indicate that earnings management and bank risk-taking are not related. Secondly there is a weak positive relation between derivatives usage and bank risk-taking.
JEL classification: G21, G31
Keywords: derivatives usage, bank risk-taking, z-score, distance to default, earnings management, loan loss provisions
1.
Introduction
1.1 Goal of paper
Certainly after the global financial crisis (GFC) of 2007 onwards, societal interest in financial institutions, in particular banks, has increased. The GFC led to a series of governmental interventions, such as (increased) deposit insurance for consumers and bail-outs of financially distressed banks. These interventions are ultimately funded by tax-payers and thus societal interest in banks has increased.
Banks are a-typical in capital structure from other firms because they are more leveraged. A relatively small adverse movement in the value of assets could cause a bank to become financially distressed. The banking industry is therefore highly regulated and the main goal of bank regulation is to sustain financial stability of the financial system, and hence of individual banks. The need for governmental interventions and bail-outs raises the question whether banks are taking on too much risk.
The goal of this paper is twofold. First, an analysis on the relation between earnings management and bank risk-taking will be conducted. Second, the relation between bank risk-risk-taking and derivatives usage will be researched.
1.2 Research questions
1.2.1 Earnings management
2
bank faces as riskiness of a bank is assessed through the accounting information it publishes. When a bank takes more risk, the probability that it becomes financially distressed also increases. The goal of banking regulators is to ensure financial stability. Regulators evaluate banks on the published accounting information in the form of annual reports, or bank regulatory forms. When banks use earnings management to conceal increased risk-taking, regulators can not accomplish their goal, because they are misled by the information that banks supply. The following research question is therefore asked:
How are earnings management and bank risk-taking related?
Supplementary to this research question, several sub questions are asked: - What is earnings management?
- Why do firms engage in earnings management?
- How do banks apply earnings management through the loan loss provision? - How can bank risk-taking influence earnings management?
- Does bank risk-taking has a negative or a positive influence on earnings management?
1.2.2 Derivatives usage
Over the past few decades trading in derivative instruments has increased. Along with the increase in derivatives usage the Financial Accounting Standards Board (FASB) introduced an accounting standard where derivative instruments had to be valued on the balance sheet at fair value. Several scandals occurred where derivative instruments were involved like Barings bank and Long Term Capital Management (Jorion, 2000).
There is substantial evidence in academic literature that derivative instruments are used by firms to hedge (reduce) risks (Hentschel & Kothari, 2001; Guay & Kothari, 2003).The aforementioned literature has focused on non-financial companies. Some papers have focused on derivatives usage at banks and have found that banks use derivatives for hedging risks (Kamau, et al., 2015; Tóth, 2014).
Banks’ usage of derivatives is a topic that attracts attention of society. Warren Buffet has called derivatives “financial weapons of mass destruction” (Berkshire Hathaway, 2002, p. 15). Joseph Stiglitz compares the role that derivatives played in the financial world to the nuclear meltdown in Fukushima and says: “Experts in both the nuclear and finance industries assured us that new technology had all but eliminated the risk of catastrophe. Events proved them wrong: not only did the risks exist, but their consequences were so enormous that they easily erased all the supposed benefits of the systems that industry leaders promoted.” (The Guardian, 2011). The terms “meltdown”, “weapons of mass destruction” and the reluctance of experts to assess or inform on the risks of derivatives embody the view of society towards derivative instruments and the risks that are involved when banks use derivatives. From this view the following question arises:
How are bank risk-taking and derivatives usage related?
In addition to this research question, several sub questions are asked: - What is a derivative instrument?
3
- Does derivatives usage cause bank risk, or does bank risk causes derivatives usage? - How are bank risk-taking and derivatives usage related?
1.3 Sample & research methods
In this paper US Bank Holding Companies (BHCs) are studied from 2004 to 2013. In total 2247 banks are sampled yielding 11,011 firm-year observations. Two specifications are estimated with a panel data ordinary least squares (OLS) fixed effects regression. The results show no relation between earnings management and bank risk-taking. And a weak positive link between derivatives usage and bank risk-taking.
1.4 Contribution to existing literature & practical use
This paper contributes to existing literature by analyzing the relationship between z-score and discretionary loan loss provisions. As Gombola et al. (2016) establishes a relation between leverage and discretionary loan loss provisions, it is of interest to ascertain whether the distance to default is related to the extent of earnings management applied by banks. The results are of interest to regulators, policy makers, auditors and other stakeholder of banks, as they gain more insights in how they should approach financial statements of banks and assess the risks they are involved in.
Furthermore, this paper contributes to existing literature by addressing the skepticism of society towards derivatives usage of banks and the risks involved.
1.5 Structure of the paper
4
2.
Literature review
2.1 Earnings management and bank risk-taking 2.1.1 What is earnings management?
The application of earnings management can be explained by the agency theory developed by Jensen & Meckling (1976). Ownership and management of the firm are separated, where the principal (owner) hires an agent (management) to manage the firm. An agency problem arises because the incentives of the principal and the agent are misaligned, the owner’s incentive is to maximize firm value and the incentive of the agent is to maximize his or her own wealth. Incentives can be aligned by making managerial compensation dependent upon firm value. The principal is unable to monitor all the actions of the agent, hence an information asymmetry exists between the principal and the agent. Monitoring mechanisms such as a board of directors and financial statement audit by an independent auditor are legally required in most countries in order to reduce the information asymmetry between the principal and the agent. Even when monitoring mechanisms are present, managers can act in their own interest at the expense of the owner. Within the boundaries of the law and applicable financial reporting standards, managers have discretion in preparing financial reports and structuring transactions. In their literature review Healy & Wahlen (1999) define earnings management as: “Earnings management occurs when managers use judgment in financial reporting and in structuring transactions to alter financial reports to either mislead some stakeholders about the underlying economic performance of the company or to influence contractual outcomes that depend on reported accounting numbers.”
2.1.2 Why do firms apply earnings management?
In the definition of Healy & Wahlen (1999) a distinction can be made between opportunistic earnings management (‘judgement in financial reporting’) and real earnings management (‘structuring transactions’). Academic literature on earnings management by banks shows that the loan loss provision is used to apply opportunistic earnings management (Chang, et al., 2008). Loan loss provisions can have a large impact on earnings as it forms 0.5% of total assets in the sample of this paper. Furthermore, banks have discretion in determining the size of the loan loss allowance (loan loss reserve). Bank managers use their discretion in determining loan loss provisions to manage earnings and bank capital (Ahmed, et al., 1999; Beatty, et al., 1995; Chang, et al., 2008). Lastly, loan loss provisions influence earnings as well as capital. As the banking industry is highly regulated, loan loss provisions can be used to avoid capital adequacy regulations.
2.1.3 How do banks apply earnings management through the loan loss provision?
5
The loan loss reserve is periodically adjusted through the loan loss provision and the net off. The net charge-off consists of a gross charge-charge-off on the loan portfolio, which are the all the loans in one period that are considered to be uncollectable. Furthermore, the net charge-off consists of recoveries on loans that were charged-off earlier. The recoveries (positive amount) are added to the gross charge-off (negative amount) and form the net charge-off. In general, the loan loss provision adds a positive amount of expected loan loss to the loan loss reserve and the net charge-off subtracts a realized amount of loan loss from the loan loss reserve. Loan loss provisions are however allowed to take a negative value when the loan loss allowance needs to be adjusted downward.
Earnings management research generally use a variation on the Jones (1991) model to measure discretionary accruals. The thought behind the Jones (1991) model is that a part of the accruals is non-discretionary and is not used for earnings management, because it fits the normal course of business and it is used correctly in the accounting sense, namely that through the use of the accrual the accounting information reflects the ‘true’ performance of the firm. Jones (1991) also recognizes a discretionary part of accruals, which is used for earnings management, and the measurement of the discretionary accruals or abnormal loan loss provisions in this paper will be the measure for earnings management.
2.1.4 How can bank risk-taking influence earnings management?
Gombola et al. (2016) argue that loan loss provisions are used to avoid regulatory scrutiny, and state that loan loss provisions can be viewed as a risk management tool. The risk of regulatory consequences arises when capital adequacy regulations are violated. Earnings management can be used to influence earnings and regulatory capital. So earnings management can be used to avoid regulatory consequences of risk-taking behavior of banks. The empirical result of Gombola et al. (2016) on the relation between leverage and earnings management is significantly positive. So an increase in leverage, measured as tier 1 capital divided by risk-weighted assets (RWA), leads to an increase in earnings management, measured by abnormal loan loss provision. The results of Gombola et al. (2016) imply that banks with a low capitalization engage less in earnings management. Gombola et al. (2016) expect a negative relation between leverage and earnings management and explain the difference in sign of the relation by theorizing that low capitalized banks are subject to increased regulatory oversight and hence can not engage in earnings management.
To reduce earnings volatility, banks overstate loan loss provisions in good times, and understate loan loss provisions in bad times, which is known as income smoothing (Perez, et al., 2008). A bank can smooth income over time when the reserves that are built up in good times are high enough to absorb the losses in bad times. When the losses in bad times are higher than the reserve available, banks have no opportunity to manage earnings and conceal the risks that are present. Therefore, banks apply less earnings management when bank risk is higher. From the aforementioned theories and empirical evidence, the following hypothesis is formulated:
H1: Bank risk has a negative impact on earnings management.
2.2 Bank risk-taking and derivatives usage 2.2.1 What is a derivative instrument?
6
instruments are used for risk management within the firm by reducing (hedging) or increasing (speculating) risk. Hedging is the procedure by which a firm reduces its dependence on external factors affecting firm value (Smith & Stulz, 1985). By speculating an investor takes a position in the market based on a belief in an informational advantage over other market participants, or it must benefit from economies of scale in transaction costs for arbitrage opportunities (Géczy, et al., 1997).
2.2.2 Why do banks use derivatives?
In perfect capital markets risk management using derivatives would not add value to the firm because investors can manage firm-specific risk through diversification (Modigliani & Miller, 1958). However, in imperfect capital markets there are incentives for firms to use derivatives. Smith & Stulz (1985) theorize that firm can increase firm value by hedging to incur a tax benefit, by lowering cost of debt and/or lowering financial distress costs. Furthermore, firms are inclined to hedge when it is beneficial for the manager’s compensation (Smith & Stulz, 1985). These theoretical incentives for hedging are confirmed by empirical evidence by e.g. Nance et al. (1993) and Berkman & Bradbury (1996). Specifically for banks, evidence on hedging has also been shown by Brewer III et al. (2014), who argue that through the use of interest rate derivatives economic stability of banks is increased which implies lower risk.
2.2.3 Does derivative usage cause bank risk-taking or vice versa?
Smith & Stulz (1985) argue that financial distress costs can be lowered using derivatives. Smith & Stulz (1985) provide an example where a firm can avoid debt covenants to become binding, because the debt covenant is based on accounting information and the accounting information can be altered by using derivatives. Income volatility can be lowered using derivatives for hedging to prevent a firm from becoming financially distressed and avoid debt covenants from becoming binding. On average BHCs hold 46.3% (calculated as average loans divided by average total assets from table 2) of their assets in loans. As the loans as a large portion of total assets, total risk can be effectively increased by assuming riskier new loans. When a bank does not have the ability to use derivatives this increased total risk can only be lowered by assuming less risky new loans, so the risk of the total loan portfolio lowers. When a bank has access to derivatives it can assume riskier loans and use derivatives to share that risk with counterparties of the derivatives contracts. So systemic risk of the banking industry could increase, but the individual bank risk decreases. Therefore, banks use derivatives to influence risk-taking and not vice versa.
Based on the aforementioned theories and empirical evidence the following hypothesis is formed:
7
3.
Methodology
In this section, the specifications for testing the hypotheses will be described. Followed by an explanation for the construction of the variables. Lastly the regression analysis and data selection will be discussed.
3.1 Earnings management and bank risk-taking The following model is estimated
𝐴𝐵𝑆𝐷𝐿𝐿𝑃𝑖𝑡 = 𝛼0+ 𝛼1𝑍𝑆𝐶𝑂𝑅𝐸𝐸𝑀𝑖𝑡+ 𝛼2𝐵𝐼𝐺4𝑖𝑡+ 𝛼3LIQ𝑖𝑡+ 𝛼4SIZE𝑖𝑡+ 𝑢𝑖+ 𝑣𝑡+ 𝜀𝑖𝑡 (1)
ABSDLLPit is discretionary loan loss provisions, residual of specification (2)
ZSCOREEMit is distance to default, measure for bank risk-taking
BIG4it is dummy variable, takes value of 1 when auditor is a big 4 auditor and 0 otherwise
LIQit is liquidity is scaled by beginning of year book value of total assets
SIZEit is natural logarithm of total assets
ui is the firm fixed effects term
vt is the year fixed effects term
εit is the error term
The model is estimated using panel data ordinary least squares with fixed year and fixed firm effects.
3.1.1 Variables
3.1.1.1 Earnings management
I start by estimating the dependent variable. To construct discretionary earnings management (ABSDLLP), a variation on the Jones (1991) model developed by Kanagaretnam et al. (2010) will be employed. This model separates non-discretionary loan loss provisions and discretionary loan loss provisions. Discretionary loan loss provisions are used by bank managers for earnings management. The residuals of specification (2) represent discretionary loan loss provisions. In construction the earnings management variable, the test results will be discussed
Following from Kanagaretnam et al. (2010), I construct the following specification:
𝐿𝐿𝑃𝑖𝑡= 𝛼0+ 𝛼1𝐿𝐿𝐴𝑖,𝑡−1+ 𝛼2𝑁𝑃𝐿𝑖,𝑡−1+ 𝛼3Δ𝑁𝑃𝐿𝑖𝑡+ 𝛼4𝐿𝐶𝑂𝑖𝑡+ 𝛼5Δ𝐿𝑂𝐴𝑁𝑆𝑖𝑡+ 𝛼6𝐶𝑂𝑀𝑖𝑡+ 𝛼7𝑀𝑂𝑅𝑇𝑖𝑡+ 𝛼8𝑂𝑇𝐻𝐸𝑅𝑖𝑡+ 𝑢𝑖+ 𝑣𝑡+ ε𝑖𝑡
(2)
LLPit is loan loss provision scaled by beginning of year book value of total assets
LLAi,t-1 is loan loss allowance scaled by beginning of year book value of total assets lagged 1 year
NPLi,t-1 is non-performing loans scaled by beginning of year book value of total assets lagged 1 year
∆NPLit is change in non-performing loans scaled by beginning of year book value of total assets
LCOit is loan charge-off scaled by beginning of year book value of total assets
∆LOANSit is change in total loans scaled by beginning of year book value of total assets
8
MORTit is mortgage loans outstanding scaled by beginning of year book value of total assets
OTHERit is other loans outstanding scaled by beginning of year book value of total assets
ui is the firm fixed effects term
vt is the year fixed effects term
εit is the error term
Following Kanagaretnam et al. (2010) the following expectations for the variables are formed. The coefficient of LLA is expected to be negative, because loan loss provisions are used to adjust the loan loss allowance to an adequate level. When LLA is relatively high, the upward adjustment through the loan loss provision is not needed, therefore a high LLA has a negative influence on LLP. A higher performing loan ratio and change in non-performing loans would cause the loan loss provision to increase to adjust the loan loss allowance for the expected losses on the non-performing loans. Therefore, the coefficients of NPLt-1 and ∆NPL are expected to be positive.
The coefficient of loan charge-offs is expected to be positive as the amount of loan charge-offs signals the current ability of the bank to recover non-performing loans. When loan charge-off is higher, the ability of the bank to recover loans is low, thus expected losses should be higher as well, thus a positive relation between LCO and LLP. A prediction for the sign of ∆LOANS is not provided by Kanagaretnam et al. (2010), because the quality of the loans is not observable, thus the sign could be negative when the quality of the new loans is high and could be positive when the quality of the new loans is low. No expectations are given for the variables COM, MORT and OTHER.
The specification is estimated using paned data OLS with fixed effects. Table 1 shows the regression results of specification (2). Model (1) in table 1 is first estimated using pooled OLS to perform a White test. The null hypothesis of a White test is that the sample has no heteroscedasticity. The result of the White test is a F-statistic of 1765.1850 with a significance at the 1% level. I conclude that the data is heteroskedastic and will apply White diagonal coefficient covariance method to account for the heteroscedasticity in the panel data regression. A redundant fixed effects test is conducted, the result is a F-statistic of 1.9879 with a significance at the 1% level. To account for multicollinearity the variables LLAt-1 and LCO are dropped, the correlation coefficients are stated
in table 4. LLAt-1 and LLP have a correlation coefficient of 56.33% which is too high. Furthermore, LCO and
LLP have a correlation coefficient of 80.69%, which is very high. After dropping the variables from the specification model (2) in table 1 is estimated.
Again a pooled OLS is regressed to perform a White test, the result is a F-statistic of 347.4043 with a
9
The signs of NPLt-1 and ∆NPL are positive and consistent with the expectations. The explanatory power of model
(2) in table 1 is 66.29%, this is consistent with Kanagaretnam et al. (2010), who reported an adjusted R2 value of
66.32%. Bank managers thus have about 34% of loan loss provisions at their discretion.
Loan loss provisions can be used by bank management to increase or decrease earnings. A positive (negative) residual implies an understatement (overstatement) of the loan loss provision. Both over- and understatement of loan loss provisions are considered to be earnings management. In this paper no distinction is made between income increasing and income decreasing earnings management. Therefore, the absolute value of the residuals of specification (2) represents the measure for earnings management (ABSDLLP).
Table 1 Results Loan loss provision regression (2)
The dependent variable is LLP for the two models. The redundant fixed effects test result is for both cross-section and period effects, these fixed effects are both applied to the two models. Model (1) contains the full specification as presented in specification (2). Due to multicollinearity issues, LLAt-1 and LCO are dropped from the final specification used for further
analysis. Model (2) contains the final specification. LLP is loan loss provision scaled by beginning of year book value of total assets. LLAt-1 is loan loss allowance scaled by beginning of year book value of total assets lagged 1 year. NPLt-1 is
performing loans scaled by beginning of year book value of total assets lagged 1 year. ΔNPL is change in non-performing loans scaled by beginning of year book value of total assets . LCO is loan charge-off scaled by beginning of year book value of total assets. ΔLOANS is change in total loans scaled by beginning of year book value of total assets. COM is commercial loans outstanding scaled by beginning of year book value of total assets. MORT is mortgage loans outstanding scaled by beginning of year book value of total assets. OTHER is other loans outstanding scaled by beginning of year book value of total assets.
(1) (2)
Coef. t stat. Coef. t stat.
Constant 0.0070 *** 6.7881 0.0033 *** 0.0002 LLAt-1 -0.4079 *** -10.7455 n.a. NPL t-1 0.0874 *** 6.5479 0.1980 *** 13.2758 ΔNPL 0.1277 *** 13.6605 0.1817 *** 16.4094 LCO 0.8126 *** 16.7424 n.a. ΔLOANS 0.0191 *** 5.0000 -0.0016 -1.3336 COM -0.0023 -1.0494 0.0003 0.1015 MORT -0.0075 *** -4.0309 -0.0058 *** -4.2819 OTHER 0.0031 ** 2.0037 0.0131 *** 4.9243 Adjusted R2 86.47% 66.29% F-stat. (p-value) 31.7636 *** 10.4737 *** Durbin-Wason stat. 2.0573 2.0123 Observations 11090 11090
Redundant fixed effects test 1.9879 ***
10
3.1.1.2 Bank risk-taking
The independent variable is ZSCOREEM, which is calculated as follows:
𝑍𝑆𝐶𝑂𝑅𝐸𝐸𝑀 = 𝐸𝐵𝑃𝑇 + 𝑇𝐼𝐸𝑅1 𝜎𝐸𝐵𝑇𝑃
(3)
Z-score is the number of standard deviations of return on assets (ROA) that a company is to default. Where default is the situation where ROA plus capitalization (CAP) is lower than zero. It is a measure of financial distress, where a higher value implies a lower risk of default, and thus lower bank risk. ROA is calculated as net income divided by book value of total assets. The loan loss provision influences all components of z-score. First it affects the ROA, because loan loss provision influences the net income of a bank. Secondly, capitalization is often measured as book value of total equity over book value of total assets. Loan loss provisions influences capitalization through the earnings and through the loan loss allowance. Earnings influence retained earnings which is part of capitalization, and loan loss allowance is part of total equity of a bank. Lastly, the standard deviation of ROA might be affected by loan loss provisions, as the ROA is influenced by loan loss provisions. To account for the influence of the loan loss provisions on z-score, the earnings before provisions and taxes (EBPT) will be employed in calculating ROA. So ROA is calculated as EBPT over book value of total assets. Secondly capitalization will be calculated as the tier 1 capital ratio (TIER1), as bank tier 1 capital does not include loan loss allowances.
3.1.1.3 Control variables
Control variables are added to specification (1), to control for factors that influence earnings management. By adding the control variables, I prevent omitted variables from distorting the coefficient of ZSCOREEM.
The first control variable is BIG4. DeAngelo (1981) argues that auditor size has a positive impact on audit quality, as big auditors have a lot to lose when they do not provide high audit quality. A higher audit quality implies less earnings management, as the auditor corrects the applied earnings management. For banks, Kanagaretnam et al. (2010) finds that big four auditors are associated with lower earnings management. A dummy variable is added to the specification which takes the value of 1 when a bank is audited by Deloitte, EY, KPMG or PwC in a firm year and zero otherwise. From the theory and the empirical evidence, I expect the variable to have a negative impact on earnings management (ABSDLLP). Secondly liquidity (LIQ) is added as a control variable to specification (1). Gombola et al. (2016) provides empirical evidence that liquidity and earnings management are negatively related. Therefore, I expect LIQ to be negatively related to ZSCOREEM. Lastly SIZE is added as a control variable. Gombola et al. (2016) argues that larger banks have better diversified loan portfolios which result in lower risk and hence a higher incentive to apply earnings management. In accordance with Gombola et al. (2016) I add SIZE to specification (1) to control for size. I expect SIZE to have a positive impact on earnings management.
3.1.2 Bank risk-taking and derivatives usage
Hypothesis 2 will be tested using the following specification:
𝑍𝑆𝐶𝑂𝑅𝐸𝐷𝐸𝑅𝑖𝑡 = 𝛼0+ 𝛼1𝐷𝐸𝑅𝑖,𝑡−1+ 𝛼2𝑁𝑃𝐿𝑖,𝑡−1+ 𝛼3𝐸𝐹𝐹𝑖,𝑡−1+ 𝛼4𝐿𝐼𝑄𝑖𝑡+ 𝛼5𝑆𝐼𝑍𝐸𝑖𝑡+ 𝛼6𝐶𝑅𝐼𝑆𝐼𝑆𝑖 + 𝛼7𝐷𝐸𝑃𝐼𝑁𝑆𝑖+ 𝑢𝑖+ 𝑣𝑡
11
ZSCOREDERit is the proxy for bank risk-taking, measured as distance to default
DERi,t-1 is the proxy for derivatives usage, measured as absolute value of fair value of derivative
assets minus fair value of derivative liabilities
NPLi,t-1 is a proxy for asset quality, measured as ratio of non-performing loans over total assets
EFFi,t-1 is a proxy for efficiency, measured as total revenue over total expenses.
LIQit is a proxy for liquidity, measured as liquid assets over total assets
SIZEit is a proxy for size, measured as natural logarithm of total assets
CRISISt is a dummy variable that takes a value of 1 in the years 2008, 2009 and 2010 and 0
otherwise
DEPINSt is a dummy variable for the increased deposit insurance scheme by the US government.
Takes value of 1 for the years 2008 to 2013 and 0 otherwise. ui is the firm fixed effects term
vt is the year fixed effects term
εit is the error term
The variables DER, NPL and LIQ are scaled by beginning of the year book value of total assets. The model will be estimated using panel data OLS with fixed year and fixed firm effects.
3.1.2.1 Bank risk-taking
For hypothesis 2, ZSCOREDER is the dependent variable, and will be calculated as follows:
𝑍𝑆𝐶𝑂𝑅𝐸𝐷𝐸𝑅 = 𝑅𝑂𝐴 + 𝐶𝐴𝑃 𝜎𝑅𝑂𝐴
(5)
Specification (4) tests whether derivatives usage (DER) influences bank risk-taking (ZSCOREDER). As DER is measured at fair value, the gains and losses on derivatives transactions influence net income. Therefore, the measurement of DER directly influences ROA, which is an important component in calculating ZSCOREDER. The data on derivatives gains and losses is not available from WRDS, so adjustments to ROA can not be made. To address the direct influence of DER on ZSCOREDER, a robustness test will be conducted where DER is measured as the notional amount of derivative assets minus notional amount of derivative liabilities.
Capitalization (CAP) will be measured as the book value of total equity scaled by total assets. Following Lepetit & Strobel (2013) the standard deviation of ROA is calculated based on all firm year observations.
3.1.2.2 Derivatives usage
12
derivative assets, the proxy for derivative usage will be negative. When the proxy for derivative usage is negative, this would imply that the bank uses less derivatives. A negative net position in derivatives does not necessarily imply less use of derivatives. The bank could hedge exposures that leads to a negative net position in derivatives. Also the bank could speculate leading to a net negative fair value of derivatives. To measure derivatives usage the net position in fair value of derivatives is adjusted to the absolute value of the net position in derivatives.
As discussed earlier, there are many measures proposed by academics for derivatives usage. A robustness test on derivatives usage will be included, by measuring derivatives usage by notional principal amount of total
derivatives outstanding.
Endogeneity
The relation between derivatives and bank risk-taking is endogenous, or reverse causal. Derivatives usage can cause bank risk-taking or vice versa. A method for dealing with the endogeneity issue is to lag the explanatory variable, which is derivatives usage in this paper. The thought behind this method is that prior year derivatives usage is a proxy for current year derivatives usage. But, as pointed out, the relation between current year derivatives usage and current year risk-taking is endogenous. The relation between prior year derivatives usage and current year risk can not be reverse causal as current year risk can not influence prior year derivatives usage. Therefore, the reverse causality issue is exempted from the analysis.
To address the endogeneity issue related to the relation between derivatives usage and bank risk-taking I will lag the derivative variable by one year (Park & Kim, 2015; Delis & Kouretas, 2011).
3.1.2.3 Control variables
In order to arrive at the specification for testing hypothesis 2, the following variables will be added to control for omitted variable bias. In the US the federal reserve bank uses a CAMEL rating system to assess a bank’s overall condition (Federal Reserve Bank, 1996). CAMEL is an abbreviation for capital adequacy (C), asset quality (A), management capability (M), earnings (E) and liquidity (L). The banks are evaluated by rating each of the aforementioned aspects of a bank, subsequently leading to a single rating. Key components of ZSCOREDER are capitalization (CAP), which is a measure for capital adequacy, and return on assets (ROA), which is a measure for earnings. Together with the control variables, all components of the CAMEL rating are addressed.
13
securities, as these securities proved to be liquid during the financial crisis. Bank size is often controlled for in previous academic papers (Rodríguez, et al., 2015; Li & Yu, 2010). A control for bank size will be added and measured as the logarithm of beginning total assets (SIZE). As larger banks are more capable of managing risks the expected relation between SIZE and ZSCOREDER is positive.
As the recent financial crisis clearly has had an influence on bank risk, a control variable for this period will be added to the model. Following Li & Marinč (2014) this paper considers the years 2007 to 2010 to be crisis years. Because the specification is estimated using lagged variables to address the endogeneity issues the crisis years will start in 2008. A dummy variable will take the value of 1 during the crisis years (2008,2009 and 2010) and 0 in the other years (2004 to 2007 and 2011 to 2013). The expected influence of the crisis on ZSCOREDER is negative, as the crisis is expected to lower the distance to default.
The US has a deposit insurance scheme to protect depositors. The presence of a deposit insurance scheme introduces moral hazard problems that cause banks to increase risk-taking (Anginer, et al., 2014). The Dodd-Frank act of 2008 increased the limit of deposit insurance from $100,000 to $250,000. Anginer et al. (2014) also document that the presence of a deposit insurance scheme in crisis years reduces bank risk-taking. Therefore, I expect that the increase of the deposit insurance limit in the US in 2008 has a positive effect on the z-score.
3.2 Data selection
The sample consists of bank holding companies (BHC) in the US in the period from 2004 to 2013. BHCs are sampled, because derivatives usage is concentrated at the highest level of the bank holding structure (Brewer III, et al., 2014). BHCs are obligated to file call reports to the federal reserve each quarter. These FR Y-9C reports are available from the Wharton Research Data Services (WRDS) database. Call reports contain all the
information needed to construct the variables. In this paper, the annual information is used instead of quarterly. The income sheet statements are reported cumulatively each quarter, so information in the fourth quarter equals the annual information. Therefore, I delete all information from quarter 1 to 3.
14
4.
Results
The results section is divided into three subsections. First, an overview of the descriptive statistics is given and discussed in table 1. Followed by the correlation matrices on the variables in tables 2 to 4. Lastly the results of the panel data regressions are discussed in table 5 and 6.
4.1 Variable descriptive statistics
Table 1 contains the descriptive statistics on all variables used in the different specifications. On average, banks hold total assets with a value 12 bln US dollars. The median total assets held by banks is 750 mln US dollars, which indicates a right skewed distribution. The average loan loss provision is 0.5% of total assets, where 0.3% of total assets is discretionary loan loss provision. The average ZSCOREEM is 6.699, so the average BHC is 6.699 standard deviations from default.
The average absolute value of net derivatives is 0.0% of total assets, so BHCs have a low exposure to their derivative position. The notional principal amount is 3.5% of total assets.
4.2 Correlations matrices
4.2.1 Earnings management and bank risk-taking
In table 3 the correlation matrix for the variables of specification (1) is given. Problems of multicollinearity could exist when the correlation coefficient is higher than 50%. The correlation between BIG4 and SIZE is 60.33%. Because of the high correlation, SIZE is dropped from the final model in estimating specification (1). Other correlation coefficients are all lower than 15% and do not cause multicollinearity issues.
4.2.2 Loan loss provision
In table 4 the correlation matrix for the variables of specification (2) is given. LLAt-1 and LCO are dropped
because of the correlation between LLAt-1 and LLP (56.33%), LLAt-1 and LCO (64.55%) and LCO and LLP
(80.69%). The other correlation coefficients of the variables do not cause multicollinearity issues.
4.2.3 Bank risk-taking and derivatives usage
15 Table 2: Descriptive statistics
Total assets, total equity and total loans are measured in thousands of US dollars ($).
ABSDLLP is the absolute value of discretionary loan loss provisions. ZSCOREEM is the proxy for bank risk-taking and measured as earnings before provisions and taxes (EBPT) plus tier1 capital over the standard deviation of EBPT. BIG4 is a dummy variable that takes a value of 1 when the auditor of the bank is Deloitte, EY, KPMG or PwC and 0 otherwise. LIQ stand for liquidity and is measured by liquid assets. SIZE is the natural logarithm of total assets. LLP is loan loss provision scaled by beginning of year book value of total assets. LLAt-1 is loan loss
allowance scaled by beginning of year book value of total assets lagged 1 year. NPLt-1 is non-performing loans scaled by beginning of year
book value of total assets lagged 1 year. ΔNPL is change in non-performing loans scaled by beginning of year book value of total assets . LCO is loan charge-off scaled by beginning of year book value of total assets. ΔLOANS is change in total loans scaled by beginning of year book value of total assets. COM is commercial loans outstanding scaled by beginning of year book value of total assets. MORT is mortgage loans outstanding scaled by beginning of year book value of total assets. OTHER is other loans outstanding scaled by beginning of year book value of total assets. ZSCOREDER is the proxy for bank risk-taking, and is measured as return on assets (ROA) plus capitalization (CAP) over the standard deviation of ROA. DERt-1 is derivatives usage measured as the absolute value of fair value of derivative assets minus fair
value of derivative liabilities scaled by beginning of year book value of total assets. NPLt-1 stands for beginning of the year non-performing
loans scaled by beginning of year book value of total assets. EFFt-1 is the proxy for efficiency and is measured by total revenues over total
expenses. CRISIS is a dummy variable that takes a value of 1 for the years 2008, 2009 and 2010 and 0 otherwise. DEPINS is a dummy variable that takes a value of 1 for the years 2008 to 2014 and 0 for the years 2004 to 2007. LIQ and SIZE are used in testing both hypothesis 1 and 2, for those two variables the number observations is the maximum number available.
Mean Median Maximum Minimum σ Observations
16
Table 3: Correlation matrix: Hypothesis 1, specification (1)
The correlation coefficients related to BIG4 are estimated on a sample with 8454 observations and 1798 banks over 9 years. The correlations of ABSDLLP, ZSCOREEM, LIQ and SIZE are estimated on a sample with 11,010 observations and 2247 banks over 10 years. ABSDLLP is the absolute value of discretionary loan loss provisions. ZSCOREEM is the proxy for bank risk-taking and measured as earnings before provisions and taxes (EBPT) plus tier1 capital over the standard deviation of EBPT. BIG4 is a dummy variable that takes a value of 1 when the auditor of the bank is Deloitte, EY, KPMG or PwC and 0 otherwise. LIQ stand for liquidity and is measured by liquid assets. SIZE is the natural logarithm of total assets.
ABSDLLP ZSCOREEM BIG4 LIQ SIZE
ABSDLLP 1 ZSCOREEM -0.0255 *** 1 BIG4 -0.0026 0.0182 * 1 LIQ -0.0560 *** 0.0065 -0.0727 *** 1 SIZE 0.0947 *** -0.0114 0.6033 *** -0.1304 *** 1
17
Table 4: Correlation matrix: Loan loss provision regression, specification (2)
LLP is the loan loss provision. LLAt-1 is the loan loss allowance at the beginning of the year. NPLt-1 stands for non-performing loans. ΔNPL equals the change in non-performing loans.
LCO is the charge-off on the loan portfolio. ΔLOANS is the change in total loans. COM stand for commercial loans. MORT is the portion of the loan portfolio that are mortgage backed. OTHER is the loans that are not commercial and not mortgage. All variables are scaled by beginning of year book value of total assets. The number of observations used in calculating the correlation coefficients is 11090, with 2243 bank holding companies over 10 periods.
LLP LLA t-1 NPL t-1 ΔNPL LCO ΔLOANS COM MORT OTHER
18
Table 5: Correlation matrix: hypothesis 2, specification (4)
ZSCOREDER is the proxy for bank risk-taking, and is measured as return on assets (ROA) plus capitalization (CAP) over the standard deviation of ROA. DERt-1 is derivatives usage
measured as the absolute value of fair value of derivative assets minus fair value of derivative liabilities scaled by beginning of year book value of total assets. NPLt-1 stands for beginning of
the year non-performing loans scaled by beginning of year book value of total assets. EFFt-1 is the proxy for efficiency and is measured by total revenues over total expenses. LIQ is the proxy
for liquidity measured by liquid assets. SIZE is the natural logarithm of total assets. CRISIS is a dummy variable that takes a value of 1 for the years 2008, 2009 and 2010 and 0 otherwise. DEPINS is a dummy variable that takes a value of 1 for the years 2008 to 2014 and 0 for the years 2004 to 2007.
ZSCOREDER DER t-1 NOT t-1 NPL t-1 EFF t-1 LIQ SIZE CRISIS DEPINS
19
4.3 Results of regression
4.3.1 Earnings management and bank risk-taking
Table 6 contains the results of the panel data OLS regression of specification (1). Model (1) in table 5 represents the results of specification (1) including all variables. First a White test was conducted on the pooled sample. The results of the white test for model (1) is F-statistic 9.0554 which is significant at the 1% level. I therefore reject the null hypothesis of the White test that heteroscedasticity is not present. The model is estimated using White diagonal coefficient covariance method, to adjust the model for heteroscedasticity. Next the model is estimated using a panel data OLS regression with both cross-section and period fixed effects. A redundant fixed effects test is conducted, I reject the null hypothesis that fixed effects are redundant with a F-statistic of 3.7101 which is significant at the 1% level. Because of multicollinearity issues I dropped SIZE and estimated model (2) in table 5.
The independent variable ZSCOREEM is not statistically significant, a relation with ABSDLLP is not proven. BIG4 is also not significant, therefore a relationship between BIG4 and ABSDLLP is not proven. The link between earnings management and auditor size is widespread in academic literature, it is surprising that in the sample used in this paper no relation is proven between BIG4 and earnings management. LIQ has a positive coefficient and is significant at the 5% level. A negative sign was expected based on the empirical proof by Gombola et al. (2016).
Table 6 Results regression specification (1)
ZSCOREEM is the proxy for bank risk-taking and measured as earnings before provisions and taxes (EBPT) plus tier1 capital over the standard deviation of EBPT. BIG4 is a dummy variable that takes a value of 1 when the auditor of the bank is Deloitte, EY, KPMG or PwC and 0 otherwise. LIQ stand for liquidity and is measured by liquid assets. SIZE is the natural logarithm of total assets. In model (1), all the variables are included. Because of multicollinearity issues BIG4 is deleted, resulting in model (2) which is the final model for testing hypothesis 1.
(1) (2)
Coef. t stat. Coef. t stat.
Constant 0.0105 ** 2.1312 0.0028 *** 18.8376 ZSCOREEM -0.0000 -1.4437 -0.0000 -1.2632 BIG4 0.0003 0.8800 0.0003 0.7495 LIQ 0.0026 ** 2.1870 0.0024 ** 2.1665 SIZE -0.0013 -1.5826 n.a. Adjusted R2 36.99% 36.95% F-stat. (p-value) 3.7429 *** 3.7395 *** Durbin-Wason stat. 2.4515 2.4482 Observations 8454 8454
Redundant fixed effects test 3.7101 ***
20
The effect is economically small as an increase of 1% in liquid assets over total assets leads to an increase in income increasing or income decreasing earnings management of 0.0024%.
4.3.2 Bank risk-taking and derivatives usage
Table 7 contains the results of the panel data OLS regression of specification (4) that tests hypothesis 2.
First a White test is conducted on the pooled sample of model (1) in table 7. The F-statistic is 43.3413 which is significant at the 1% level. The null hypothesis of no heteroscedasticity is rejected and White diagonal
coefficient covariance method is applied in estimating the panel data OLS regression. Next, model (1) is estimated using panel data OLS with firm fixed effects. Period fixed effects are not applied as these effects are perfectly collinear with CRISIS and DEPINS. A redundant fixed effects test is performed, resulting in a
F-Table 7 Results bank risk-taking specification
The dependent variable is ZSCOREDER. Both models are regressed using cross section fixed effects. Period fixed effects are not applied since the dummy variables CRISIS and DEPINS are perfectly collinear with period dummies. ZSCOREDER is the proxy for bank risk-taking, and is measured as return on assets (ROA) plus capitalization (CAP) over the standard deviation of ROA. DERt-1 is derivatives usage measured as the absolute value of fair value of derivative assets minus fair
value of derivative liabilities scaled by beginning of year book value of total assets. NPLt-1 stands for beginning of the year
non-performing loans scaled by beginning of year book value of total assets. EFFt-1 is the proxy for efficiency and is
measured by total revenues over total expenses. LIQ is the proxy for liquidity measured by liquid assets. SIZE is the natural logarithm of total assets. CRISIS is a dummy variable that takes a value of 1 for the years 2008, 2009 and 2010 and 0 otherwise. DEPINS is a dummy variable that takes a value of 1 for the years 2008 to 2014 and 0 for the years 2004 to 2007.
(1) (2)
Coef. t stat. Coef. t stat.
Constant -11.5069 * -1.9585 -11.2742 * -1.9175 DERt-1 -32.6722 ** -2.1565 n.a. NOTt-1 n.a. 0.2546 1.5743 NPL t-1 -41.2801 *** -13.7116 -41.3139 *** -13.7147 EFFt-1 1.3935 ** 1.9897 1.3960 ** 1.9910 LIQ 3.2406 ** 2.3658 3.2411 ** 2.3665 SIZE 2.7768 *** 2.7490 2.7342 *** 2.7042 CRISIS -0.5719 *** -5.4698 -0.5712 *** -5.4690 DEPINS -0.2380 -1.3552 -0.2397 -1.3628 Adjusted R2 70.44% 70.43% F-value 12.6443 *** 12.6403 *** Durbin-Watson stat. 1.3789 1.3806 Observations 11011 11011 Fixed effects
Firm fixed effects 9.9773
***
Firm fixed effects 9.9672
***
21
statistic of 9.9773 which is significant at the 1% level. I reject the null hypothesis that firm fixed effects are redundant and apply firm fixed effects in estimating specification (4). When there is no serial autocorrelation present, the Durbin-Watson statistic equals 2. There is positive autocorrelation when the Durbin-Watson statistic is between 0 and 2, and negative autocorrelation when the Durbin-Watson statistic is between 2 and 4. The Durbin-Watson statistic of table 7 model (1) is 1.3789 which indicates that there is positive autocorrelation present in the dataset, it is however at an acceptable level. Some autocorrelation is expected as the variables are likely to depend on the prior year value on average. For instance, a large bank with high total assets will, on average, remain a large bank with high total assets in the following year.
DERt-1 has a negative coefficient, which is contrary to the expected positive sign. The coefficient is significant at
the 5% level. The economic impact is high, as a 1% increase in derivatives usage leads to a decrease of 0.3267 in ZSCOREDER. The coefficient of NPLt-1 is negative and statistically significant. This is result is according to the
expected negative sign. The economic impact is high, a 1% increase in non-performing loans leads to a decrease in ZSCOREDER of 0.4128. The sign of the coefficient of EFFt-1 is as expected positive and also statistically
significant. An increase in efficiency leads to an increase of distance to default. LIQ is also positively related zo ZSCOREDER and statistically significant. Higher liquidity leads to lower financial distress risk, which is as expected. The CRISIS dummy variable is, as expected, negatively related to ZSCOREDER and statistically significant. DEPINS is negatively related to ZSCOREDER but not statistically significant. The increase of the deposit insurance scheme in 2008 did not severely influence the financial distress risk of banks.
The explanatory value of the model is high, 70.44%. The measures for the CAMEL rating explain 70.44% of the variation in ZSCOREDER.
Robustness test
As a robustness test specification (4) is estimated with the prior year total notional principal amount of
derivatives outstanding as a proxy for derivatives usage. The coefficient of NOTt-1 is statistically not significant.
22
5.
Conclusion
5.1 Findings
This paper revolves around two research questions. I begin by relating the results to the research questions and answer the main research questions.
5.1.1 Earnings management and bank risk-taking
First the influence of bank risk on earnings management has been researched. The results of the analysis does not provide proof of a relation between the bank’s distance to default and discretionary loan loss provisions. An explanation could be that increased bank risk is not an incentive to engage in earnings management for banks. When a bank becomes financially distressed due to increased risk taking, earnings management could be used to avoid regulatory consequences. Since there is not a relation between risk and earnings management, bank managers are either not interested in earnings management when the bank is riskier, or not able to engage in earnings management. Income smoothing requires a bank to overstate loan loss provisions in good times and build up a reserve to use this reserve and understate loan loss provisions in bad times. When the increased risk of a bank has caused the reserves to be too low to further smooth income, bank management is unable to engage in earnings management through loan loss provisions.
5.1.2 Bank risk-taking and derivatives usage
The second question regards derivatives usage and its influence on bank risk-taking. The analysis shows a weak negative relation between derivatives usage and bank risk. A higher use of derivatives causes bank risk to increase. Because the result is weak and not robust to another proxy for derivatives usage, no relation is established between bank risk-taking and derivatives usage. Although the analysis is not definitive, it does provide an indication that derivatives usage by banks could lead to risk-taking and instead of risk reduction. If there is a relation between derivatives usage and bank risk-taking it is of interest to bank management, banking regulators, shareholders, auditors and society. Future research could use a more in depth view on which derivatives are used for risk-taking and could expand the research to other countries to see the global effects of derivatives usage on bank risk-taking.
5.2 Limitations
There are several limitations to the research conducted as done in this paper. An important assumption in the measurement of earnings management based on the data retrieved from the call reports is that the information banks file in call reports is consistent with the financial information reported in their annual reports. Since annual reports are audited by an independent auditor and call reports are marginally audited by the FDIC.
23
References
Ahmed, A., Takeda, C. & Thomas, S., 1999. Bank loan loss provisions: a reexamination of capital management, earnings management and signaling effects. Journal of Accounting and Economics, 28(1), pp. 1-25.
Anginer, D., Demirguc-Kunt, A. & Zhu, M., 2014. How does deposit insurance affect bank risk? Evidence from the recent crisis. Journal of Banking & Finance, 48(1), pp. 312-321.
Bartram, S., Brown, G. & Conrad, J., 2011. The Effects of Derivatives on Firm Risk and Value. Journal of Financial and Quantitative Analysis, 46(4), pp. 967-999.
Beatty, A., Chamberlain, S. & Magliolo, J., 1995. Managing financial reports of commercial banks: the influence of taxes, regulatory capital, and earnings. Journal of Accounting Research, 33(2), pp. 231-261.
Berkman, H. & Bradbury, M., 1996. Empirical Evicence on the Corporate Use of Derivatives. Financial Management, 25(2), pp. 5-13.
Berkshire Hathaway, 2002. Berkshire Hathaway inc. 2002 annual report. [Online] Available at: http://www.berkshirehathaway.com/2002ar/2002ar.pdf
[Accessed 5 8 2016].
Brewer III, E., Deshmukh, S. & Opiela, T., 2014. Interest-rate uncertainty, derivatives usage, and loan growth in bank holding companies. Journal of Financial Stability, 15(1), pp. 230-240.
Chang, R.-D., Shen, W.-H. & Fang, C.-J., 2008. Discretionary loan loss provisions and earnings management for the banking industry. International Business & Economics Research Journal, 7(1), pp. 9-20.
Chen, J. & King, T.-H. D., 2014. Corporate hedging and the cost of debt. Journal of Corporate Finance, 29(1), pp. 221-245.
Chernenko, S. & Faulkender, M., 2011. The Two Sides of Derivatives Usage: Hedging and Speculating with Interest Rate Swaps. Journal of Financial and Quantitative Analysis, 46(6), pp. 1727-1754.
Choi, J., Mao, C. & Upadhyay, A., 2015. Earnings Management and Derivative Hedging with Fair Valuation: Evidence from the Effects of FAS 133. The Accounting Review, 90(4), pp. 1437-1467.
Cornett, M., McNutt, J. & Tehranian, H., 2009. Corporate governance and earnings management at large U.S. bank holding companies. Journal of Corporate Finance, 15(1), pp. 412-430.
DeAngelo, L., 1981. Auditor Size and Audit Quality. Journal Of Accounting & Economics, 3(3), pp. 183-199. Delis, M. & Kouretas, G., 2011. Interest rates and bank risk-taking. Journal of Banking & Finance, 35(1), pp. 840-855.
Federal Reserve Bank, 1996. Uniform FInancial Institutions Rating System. [Online]
Available at: http://www.federalreserve.gov/BoardDocs/press/general/1996/19961224/default.htm [Accessed 8 6 2016].
Géczy, C., Minton, B. & Schrand, C., 1997. Why firms use currency derivatives. Journal of Finance, 52(4), pp. 1323-1354.
Gombola, M., Yueh-Fang Ho, A. & Huang, C.-C., 2016. The effect of leverage and liquidity on earnings and capital management: evidence from U.S. commercial banks. International Review of Economics and Finance, 43(1), pp. 35-58.
Guay, W. & Kothari, S., 2003. How much do firms hedge with derivatives?. Journal of Financial Economics, 70(3), pp. 423-461.
Healy, P. & Wahlen, J., 1999. A review of the earnings management literature and its implications for standard setting. Accounting Horizions, 13(4), pp. 365-383.
24
Hutton, A., Marcus, A. & Tehranian, H., 2009. Opaque financial reports, r2, and crash risk. Journal of Financial Economics, 94(1), pp. 67-86.
Jensen, M. & Meckling, W., 1976. Theory of the firm: managerial behavior, agency costs and ownership structure. Journal of Financial Economics, 3(4), pp. 305-360.
Jones, J., 1991. Earnings management during import relief investigations. Journal of Accounting Research, 29(2), pp. 193-228.
Jorion, P., 2000. Risk management lessons from long-term capital management. European Financial Management, 6(3), pp. 277-300.
Kamau, P., Inanga, E. & Rwegasira, K., 2015. The usage of currency derivatives in multilateral banks. Management Research Review, 38(5), pp. 482-504.
Kanagaretnam, K., Krishnan, G. & Lobo, G., 2010. An empirical analysis of auditor independence in the banking industry. The Accounting Review, 85(6), pp. 2011-2046.
Kim, W. & Sung, T., 2005. What makes firms manage FX risk?. Emerging Markets Review, 6(1), pp. 263-288. Lepetit, L. & Strobel, F., 2013. Bank insolvency risk and time-varying z-score measures. Journal of
International Financial Markets, 25(1), pp. 73-87.
Li, L. & Yu, Z., 2010. The impact of derivatives activity on commercial banks: evidence from U.S. bank holding companies. Asia-Pacific Financial Markets, 17(1), pp. 303-322.
Li, S. & Marinc, M., 2014. The use of financial derivatives and risks of U.S. bank holding companies. International Review of Financial Analysis, 35(1), pp. 46-71.
Modigliani, F. & Miller, M., 1958. The cost of capital, corporation finance and the theory of investment. The American Economic Review, 48(3), pp. 261-297.
Nance, D., Smith, C. & Smithson, C., 1993. On the determinants of corporate hedging. Journal of Finance, 48(1), pp. 267-284.
Park, D. & Kim, J., 2015. Financial derivatives usage and monetary policy transmission: evidence from korean firm-level data. Global Economic Review, 44(1), pp. 101-115.
Paun, I. & Gogoncea, R., 2013. Interest rate risk management and the use of derivative securities. Economia: Seria Management, 16(2), pp. 242-254.
Perez, d., Salas-Fumas, V. & Saurina, J., 2008. Earnings and capital management in alternative loan loss provision regulatory regimes. European Accounting Review, 17(3), pp. 423-445.
Radic, N., Fiordelisi, F. & Girardone, C., 2012. Efficiency and risk-taking in pre-crisis investment banks. Journal Of Financial Services Research, 41(1), pp. 81-101.
Raturi, M., 2005. The use of derivatives by US insurers: empirical evidence and regulatory issues. The Journal of Risk Finance, 6(2), pp. 87-97.
Rodríguez, L., González, L., Agra, S. & Santomil, P., 2015. The effect of credit derivatives usage on the risk of european banks. Revista de Economia Mundial, 40(1), pp. 197-220.
Shrieves, R. & Dahl, D., 2003. Discretionary accounting and the behavior of Japanese banks under financial duress. Journal of Banking & Finance, 27(1), pp. 1219-1243.
Smith, C. & Stulz, R., 1985. The Determinants of Firms' Hedging Policies. Journal of Financial and Quantitative Analysis, 20(4), pp. 391-405.
The Guardian, 2011. Meltdown: not just a metaphor. [Online]
25
Tóth, K., 2014. The effect of derivative financial instruments on bank risks, relevance and faithful
Appendix I: Variables description, source and calculation
The source of Total Assets is FR Y-9C, with data code BHCK2170. When a range of data codes is given, the sum of these values is used in the calculation, for example BHCK(5524 to 5526) equals the sum of items BHCK5524, BHCK5525 and BHCK5526.
Variable Description Source Data code (BHCK) Deflator Calculation
LLP Loan Loss Provision FR Y-9C 4230 Total Assetst-1
𝐵𝐻𝐶𝐾4230 𝑇𝑜𝑡𝑎𝑙 𝐴𝑠𝑠𝑒𝑡𝑠𝑡−1 LLAt-1 Loan Loss Allowance (beginning of
year) FR Y-9C 3123 Total Assets
𝐵𝐻𝐶𝐾3123 𝑇𝑜𝑡𝑎𝑙 𝐴𝑠𝑠𝑒𝑡𝑠 NPLt-1 Non-performing Loans (beginning of
year) FR Y-9C 5524-5526 Total Assets
𝐵𝐻𝐶𝐾(5524 𝑡𝑜 5526) 𝑇𝑜𝑡𝑎𝑙 𝐴𝑠𝑠𝑒𝑡𝑠
ΔNPL Change in Non-performing Loans FR Y-9C 5524-5526 Total Assetst-1
𝐵𝐻𝐶𝐾(5524 𝑡𝑜 5526) − 𝐵𝐻𝐶𝐾(5524 𝑡𝑜 5526)𝑡−1 𝑇𝑜𝑡𝑎𝑙 𝐴𝑠𝑠𝑒𝑡𝑠𝑡−1
LCO Loan Charge-off FR Y-9C 4635 Total Assetst-1
𝐵𝐻𝐶𝐾4635 𝑇𝑜𝑡𝑎𝑙 𝐴𝑠𝑠𝑒𝑡𝑠𝑡−1
LOANS Total Loans FR Y-9C 2122 Total Assetst-1
𝐵𝐻𝐶𝐾2122 𝑇𝑜𝑡𝑎𝑙 𝐴𝑠𝑠𝑒𝑡𝑠𝑡−1
ΔLOANS Change in Total Loans FR Y-9C 2122 Total Assetst-1
𝐵𝐻𝐶𝐾2122 − 𝐵𝐻𝐶𝐾2122𝑡−1 𝑇𝑜𝑡𝑎𝑙 𝐴𝑠𝑠𝑒𝑡𝑠𝑡−1
COM Commercial loans FR Y-9C 1766 Total Assetst-1
𝐵𝐻𝐶𝐾1766 𝑇𝑜𝑡𝑎𝑙 𝐴𝑠𝑠𝑒𝑡𝑠𝑡−1
MORT Mortgage Loans FR Y-9C 1410,1590 Total Assetst-1
𝐵𝐻𝐶𝐾1410 + 𝐵𝐻𝐶𝐾1590 𝑇𝑜𝑡𝑎𝑙 𝐴𝑠𝑠𝑒𝑡𝑠𝑡−1
OTHER Other Loans FR Y-9C 2122, 1766, 1410,
1590 Total Assetst-1
𝐵𝐻𝐶𝐾2122 − 𝐵𝐻𝐶𝐾1766 − 𝐵𝐻𝐶𝐾1410 − 𝐵𝐻𝐶𝐾1590 𝑇𝑜𝑡𝑎𝑙 𝐴𝑠𝑠𝑒𝑡𝑠𝑡−1
ROA Return on Assets FR Y-9C 4340 Total Assetst-1
𝐵𝐻𝐶𝐾4340 𝑇𝑜𝑡𝑎𝑙 𝐴𝑠𝑠𝑒𝑡𝑠𝑡−1
CAP Total equity capital FR Y-9C 3210 Total Assetst-1
𝐵𝐻𝐶𝐾8274 𝑇𝑜𝑡𝑎𝑙 𝐴𝑠𝑠𝑒𝑡𝑠𝑡−1
ZSCOREDER
Calculated as ROA plus CAP divided by the standard deviation of ROA. The standard deviation is calculated based on the entire sample.
FR Y-9C 4340, 3210 n.a. 𝑍𝑆𝐶𝑂𝑅𝐸 = 𝑅𝑂𝐴 + 𝐶𝐴𝑃
Appendix I: Variables description, source and calculation (continued)
The source of Total Assets is FR Y-9C, with data code BHCK2170. When a range of data codes is given, the sum of these values is used in the calculation, for example BHCK(5524 to 5526) equals the sum of items BHCK5524, BHCK5525 and BHCK5526.
ABSDLLP
Absolute value of residuals of specification (1). Proxy for earnings management
n.a. n.a. n.a. n.a.
DERt-1 Absolute value of fair value of
derivatives (assets minus liabilities) FR Y-9C 8733-8756 Total Assets
|𝑃𝑜𝑠𝑖𝑡𝑖𝑣𝑒 𝐹𝑉 𝑑𝑒𝑟𝑖𝑣𝑎𝑡𝑖𝑣𝑒𝑠 − 𝑁𝑒𝑔𝑎𝑡𝑖𝑣𝑒 𝐹𝑉 𝑑𝑒𝑟𝑖𝑣𝑎𝑡𝑖𝑣𝑒𝑠| 𝑇𝑜𝑡𝑎𝑙 𝐴𝑠𝑠𝑒𝑡𝑠
NOTt-1
Notional principal amount of derivatives contracts the BHC is involved in. Proxy for derivatives usage robustness test.
FR Y-9C 8723-8732 Total Assets 𝐵𝐻𝐶𝐾(8725 𝑡𝑜 8732)
𝑇𝑜𝑡𝑎𝑙 𝐴𝑠𝑠𝑒𝑡𝑠𝑡−1
NPL Non-performing loans over beginning
total assets FR Y-9C 5524-5526 Total Assetst-1
𝐵𝐻𝐶𝐾5524 + 𝐵𝐻𝐶𝐾5525 + 𝐵𝐻𝐶𝐾5526 𝑇𝑜𝑡𝑎𝑙 𝐴𝑠𝑠𝑒𝑡𝑠𝑡−1
EFFt-1
Total interest income plus total non-interest income (total revenues) over total interest expenses plus total non-interest expenses (total expenses). Lagged one year for endogeneity concerns.
FR Y-9C 4107,4079,4073,4093 n.a. 𝐵𝐻𝐶𝐾4107 + 𝐵𝐻𝐶𝐾4079
𝐵𝐻𝐶𝐾4073 + 𝐵𝐻𝐶𝐾4093
LIQ Liquid assets FR Y-9C 0010, 1293, 1298 Total Assetst-1
𝐵𝐻𝐶𝐾0010 + 𝐵𝐻𝐶𝐾1293 + 𝐵𝐻𝐶𝐾1298 𝑇𝑜𝑡𝑎𝑙 𝐴𝑠𝑠𝑒𝑡𝑠𝑡−1
SIZE Log of total beginning assets n.a. n.a. n.a. n.a.
CRISIS
Dummy variable, takes value of 1 if observation year is 2007,2008,2009 or 2010; 0 in other years
n.a. n.a. n.a. n.a.
DEPINS
Dummy variable, takes value of 1 for the period 2008 to 2013 because of the increased deposit insurance limit. Takes value of 0 in the years preceding 2008.
