Loan Loss Provisioning and the Impact
of Monetary Policy on Bank Lending
R.J. (Ruben) Mulder1
UNIVERSITY OF GRONINGEN Faculty of Economics and Business
MSc Business Administration, Specialization Finance
Supervisor: prof. dr. K.F. Roszbach March 2012, Groningen
Abstract
This paper investigates the existence of cross-sectional differences in the way banks respond to changes in the monetary policy stance, while it separates banks into different asset size classes. It adds to existing literature by introducing loan loss provisioning as bank-specific characteristic that might explain the impact of monetary policy on bank lending. It uses a two-step methodology, as proposed by Kashyap and Stein (2000), and examines lending behavior at the individual bank level to separate loan demand from loan supply shifts. The dataset yields 433,234 observations from U.S. insured commercial banks, that covers the period between 1995Q1 and 2010Q4. This paper concludes that within the class of small banks, high loan loss provisioning banks tend to be more sensitive to changes in the monetary policy stance than low loan loss provisioning banks. The overall findings support the idea of many researchers to take loan loss provisioning as an integral part of the risk-based minimum capital framework, to enhance the effectiveness of monetary policy.
JEL classification: E44; E51; E52; G21
Keywords: Monetary-transmission mechanism; bank lending channel; loan loss provisioning
1
Contents
1. Introduction ... 3
2. Related literature ... 5
3. Hypothesis development ... 9
3.1 Loan loss provisioning ... 9
3.2 Hypothesis ... 13
4. Data ... 14
5. Methodology... 18
5.1 The two-step regression approach ... 18
5.2 Potential endogeneity bias ... 20
6. Results ... 21
7. Robustness checks ... 24
7.1 The one-step regression approach ... 24
7.2 The “quasi” instrumental variables approach ... 27
8. Conclusion ... 29
References ... 31
Appendix A ... 33
Appendix B ... 34
1. Introduction
During the financial crisis of 2007 - 2010, a significant number of banks were rescued by governmental institutions by way of capital injections to ensure that banks continue to supply loans. One reason for this governmental support is that banks did not set enough loan loss provisions and did not set aside sufficient capital to cover expected and unexpected loan losses. An important concern for policymakers is that when banks’ balance sheets are weakened during downturns, banks become more conservative and therefore may reduce loan supply.
This paper investigates the existence of cross-sectional differences in the way banks respond to changes in the monetary policy stance. It uses individual balance sheet data from 11,785 U.S. insured commercial banks. Monetary policy affects economic activity and inflation through several channels, which are collectively known as the monetary-transmission mechanism. This paper discusses two different channels, assuming that changes in the monetary policy stance are followed by significant movements in aggregate bank lending volume. First, the
bank lending channel theory relies on an imperfect market for bank debt, where the impact of monetary policy
on bank lending depends on certain bank-specific characteristics. Second, the bank capital channel theory relies on an imperfect market for bank equity, where capital has an impact on banks’ ability to absorb shocks to bank profits. When it is too costly to raise new equity, banks tend to reduce loan supply during a monetary policy tightening, in order to meet regulatory minimum capital requirements.
While many researchers focus on liquidity and capital as determinants of bank lending, this paper analyses loan
loss provisioning as bank-specific characteristic that might explain bank lending behavior. The federal banking
regulators in the U.S. require banks to build up adequate loan loss reserves, to absorb expected loan losses embedded in the loan portfolio. Under current accounting standards, a bank’s estimate of inherent losses in the loan portfolio are based on an “incurred loss model”, that requires that one or more default events have to be occurred, before a loan loss reserve can be established. Many politicians and regulators have criticized these accounting standards, since they permit banks to recognize loan losses fairly late in the credit cycle. Comptroller of the Currency, John C. Dugan, argued in his speech before the Institute of International Bankers on March 2th 20092, that existing accounting standards have amplified the economic crisis:
“Rather than being counter-cyclical, loan loss provisioning has become decidedly pro-cyclical, magnifying the impact of the downturn”.
By delaying the recognition of expected loan losses, loan loss reserves might become inadequate to absorb loan losses during downturns. When greater provisioning is required, bank capital has to absorb both expected and unexpected loan losses. Subsequently, the amount of retained earnings that are available to supplement the required level of regulatory capital is reduced, which might have a negative impact on banks’ loan supply.
2
This paper investigates cross-sectional differences in the way banks respond to changes in the monetary policy stance, while it separates banks into different asset size classes. It uses a two-step methodology, as proposed by Kashyap and Stein (2000), to examine bank lending behavior at the individual bank level and to separate loan demand from loan supply shifts. The dataset yields 433,234 observations from U.S. insured commercial banks, that covers the period between 1995Q1 and 2010Q4. This paper adds to existing literature by introducing loan loss provisioning as a bank-specific characteristic, that might explain the impact of monetary policy on bank lending. The hypothesis is that within the class of small banks, bank lending is more sensitive to changes in the monetary policy stance for low loan loss provisioning banks. The results from this paper contradict the hypothesis. Furthermore, Appendix C provides evidence for the existence of the “traditional” bank lending channel, that reinforces the argument that banks’ condition affects the impact of monetary policy on bank lending.
2. Related literature
The Federal Reserve in the U.S. implements monetary policy by performing open market operations, for example by imposing reserve requirements. Through its open market operations, the Federal Reserve exercises considerable control over the demand for and supply of balances that depository institutions hold at the Federal Reserve Banks, and in this way it influences the federal funds rate. According to the Federal Reserve Act, the objectives of monetary policy are to promote effectively the goals of maximum employment, stable prices, and moderate long-term interest rates.
The monetary-transmission mechanism describes the way monetary policy affects economic activity and inflation. Bernanke and Blinder (1992) find evidence that monetary policy affects the composition of bank assets, and show that the federal funds rate is extremely informative about future movements in real macroeconomic variables. A monetary policy tightening, as indicated by an increase in the federal funds rate, depresses the economy to the extent that some borrowers are dependent on bank loans.
Kashyap and Stein (1995) introduce the bank lending channel theory, that supports the existence of cross-sectional differences in the way banks respond to changes in the monetary policy stance. The critical assumption for the bank lending channel to hold, is that a monetary policy tightening should be followed by a drop in insured deposits3, while the reduction in relatively cheap insured deposits cannot be completely offset by alternative uninsured forms of liabilities. When the Federal Reserve drains reserves from the system and reserve requirements are binding, a monetary policy tightening reduces the extent to which banks can accept reservable deposits, such as insured deposits. Moreover, if the market for non-reservable bank liabilities is imperfect, some banks are unable to raise more costly uninsured forms of liabilities, that will lead those banks to reduce loan supply.
Previous literature find evidence that a monetary policy tightening has a negative impact on aggregate loan supply. Kashyap and Stein (2000) investigate whether the impact of monetary policy is stronger for small banks with illiquid balance sheets, and use quarterly observations from every U.S. insured commercial bank. They suppose that during a monetary policy tightening, some banks are not able to find alternative uninsured liabilities and must shrink their assets. They suggest that more liquid banks can relatively easily protect their loan portfolio by drawing down some liquid assets, while small banks are least likely to be able to frictionlessly raise uninsured liabilities. Their results show that within the class of small banks, bank lending is more sensitive to changes in the monetary policy stance for those banks with the least liquid balance sheets.
Besides liquidity and bank asset size, bank capital might also influence the way banks respond to changes in the monetary policy stance. Empirical literature distinguishes two theories that explain the role of bank capital in the monetary-transmission mechanism. Both theories derive from a failure of the Modigliani and Miller’s (1958) theorem. In a Modigliani-Miller world, capital markets are assumed to be perfect, and therefore, banks would
3
always be able to raise funds in order to finance lending opportunities. The two theories that explain the role of bank capital in the monetary-transmission mechanism can be described as follows.
First, the bank lending channel theory relies on an imperfect market for bank debt. Kishan and Opiela (2000) hypothesize and find evidence that small and undercapitalized banks may not be able to offset a drain in insured deposits by selling large time deposits. They suppose that the market for bank debt is not frictionless, in that investors ask a premium for uninsured liabilities. Bank capital might affect banks’ ability to raise uninsured liabilities, when it provides investors a signal about a bank’s creditworthiness. Assuming that low-capitalized banks are more exposed to asymmetric information problems, an increase in banks’ capital base has a positive impact on the capacity to limit the effect of a drop in insured deposits on bank lending.
Second, the bank capital channel theory relies on an imperfect market for bank equity. Gambacorta and Mistrulli (2004) argue that during a monetary policy tightening, banks are subject to interest rate risk because bank assets typically have a longer maturity than bank liabilities. Due to this maturity mismatch, banks bear a loss when the short-term interest rate rises, which reduces profits and then capital. Moreover, assuming that the market for bank equity is imperfect, low-capitalized banks are forced to reduce loan supply during a monetary policy tightening, in order to meet regulatory minimum capital requirements and to prevent themselves from further capital adequacy concerns.
Van den Heuvel (2009) examines the role of bank lending in the monetary-transmission mechanism. He develops a dynamic model of optimal bank lending that incorporates an imperfect market for bank equity and the presence of risk-based capital requirements. He relies on the bank capital channel theory, and finds that shocks to bank profits, such as loan defaults, have a persistent impact on bank lending. Moreover, he concludes that even when the capital base is greater than minimum capital requirements, low-capitalized banks are more responsive to monetary policy shocks.
This paper departs from above described literature, by introducing loan loss provisioning as differentiating bank-specific characteristic, that might influence the impact of monetary policy on bank lending. Banks use loan loss provisions (non-cash costs) to adjust loan loss reserves that should reflect expected loan losses embedded in the loan portfolio. Ahmed et al. (1999) find evidence that loan loss provisions have a significant impact on profits and thus capital, and it reflect meaningful changes in the expected quality of a bank loan portfolio.
Loan loss provisioning in the U.S. is regulated by GAAP, based on the Statement of Financial Accounting Standards (FAS) No. 5. It requires banks to recognize loans as impaired, when there is objective evidence that one or more loss impairment events have occurred that will have a negative impact on the estimated future cash flow of that loan. These standards are based on an “incurred loss model”, that supports objective evidence of impairment events, while it does not allow banks to provision for loan losses based on expected future impairment events. Many policymakers suppose that under current accounting standards, banks are permitted to recognize loan losses fairly late in the credit cycle, which might lead to an inadequate assessment of expected loan losses. In April 2009, the Financial Stability Forum argued that:
“Under the current accounting requirements of an incurred loss model, a provision for loan losses is recognized only when a loss impairment event or events have taken place that are likely to result in non-payment of a loan in the future. Identification of the loss event is a difficult and subjective process that results in a range of practice and, potentially, a failure to fully recognize existing credit losses earlier in the credit cycle.”
It is generally recognized that loan loss reserves should absorb expected loan losses, while bank capital should provide a buffer against unexpected loan losses. Cavallo and Majnoni (2001) suggest that an inadequate assessment of expected loan losses leads to under-provisioning and to a shortage of loan loss reserves during periods associated with increased credit impairments and increased default events. When loan loss reserves need to grow fast during downturns, bank capital has to offset both expected as well as unexpected loan losses, and the amount of retained earnings that are available to supplement the required level of regulatory capital is reduced.
Jiménez and Saurina (2006) develop a regulatory framework based on a countercyclical forward-looking
provisioning approach, that measures the credit risk profile of a bank’s loan portfolio along the business cycle.
The idea is that reserves are build up to cover expected or potential loan losses on a business cycle that might happen over a period longer than the average life of loans. According to Balla and McKenna (2009), such a countercyclical provisioning approach would have smoothed bank income through the business cycle, and the need to provision for loan losses would have been significantly lower during the financial crisis of 2007 - 2010. Beatty and Liao (2011) suggest that the extent to which the recognition of expected losses is delayed, affects the ability of loan loss reserves to cover expected loan losses. They find that during downturns, a reduction in bank lending is lower for banks with smaller delays in expected loan loss recognition.
3. Hypothesis development
3.1 Loan loss provisioning
This paper analyzes loan loss provisioning as bank-specific characteristic that might explain the impact of monetary policy on bank lending. According to Nichols et al. (2009), loan loss reserves should reflect the total amount of expected future loan losses embedded in the loan portfolio, and are the amounts of earnings retained rather than distributed to shareholders. In accounting terms, loan loss reserves are a contra-asset account, used to reduce the value of total gross loans reflected on a bank’s balance sheet. Loan loss reserves are built up through loan loss provision costs, reflected on a bank’s income statement.
The conceptual framework of credit risk management supposes that loan loss reserves should absorb expected loan losses, while bank capital serves to absorb unexpected loan losses. Figure 1 illustrates that when actual loan losses exceed the expected value, credit risk capital should absorb unexpected loan losses to avoid default up to a certain level of confidence (up to 99 percent of all cases in Figure 1). According to Resti and Sironi (2007), the confidence interval selection is determined by a bank’s degree of risk aversion. By selecting a high confidence level, a bank obtains a greater degree of protection, in the sense that the probability of occurrence of excess losses is reduced.
FIGURE 1 Probability density function loan losses
Expected loan losses Unexpected loan losses Loan loss reserves
Credit risk capital
Loan losses
FIGURE 2 Non-performing loans coverage ratio
NOTES: Figure 2 shows the aggregate level of the non-performing loans coverage ratio for all U.S. insured commercial banks between 1992 and 2010. LLR = Aggregate level of loan loss reserves, NPL = Aggregate level of non-performing loans. Loan loss reserves refer to the allowance for loan and lease losses. Non-performing loans refer to assets past due 90 days or more, plus assets placed in nonaccrual status. Source: FDIC’s Consolidated Reports of Condition and Income.
FIGURE 3 Loan loss reserves versus non-performing loans ratio
NOTES: Figure 3 shows the aggregate level of the loan loss reserves ratio versus non-performing loans ratio, for all U.S. insured commercial banks between 1992 and 2010. LLR/TL = Ratio of the aggregate level of loan loss reserves against total gross loans, NPL/TL = Ratio of the aggregate level of non-performing loans against total gross loans. Loan loss reserves refer to the allowance for loan and lease losses. Non-performing loans refer to assets past due 90 days or more, plus assets placed in nonaccrual status. Total gross loans refer to total loans and lease financing receivables, net of unearned income. Source: FDIC’s Consolidated Reports of Condition and Income.
The adequacy of loan loss reserves to cover expected loan losses is generally measured by the non-performing loans coverage ratio (the ratio of loan loss reserves against performing loans). Figure 2 shows the non-performing loans coverage ratio for all U.S. insured commercial banks between 1992 and 2010. Data come from the FDIC’s Consolidated Reports of Condition and Income4, that combine all individual bank accounting data into an “aggregate balance sheet”. Figure 2 indicates that loan loss reserves are higher than non-performing loans at any point between 1993 and 2007. During the downturn of 2007 - 2010, the aggregate level of the non-performing loans coverage ratio becomes historically low. Figure 3 shows that during this period, loan loss reserves have to grow fast to deal with the exacerbation in the level of non-performing loans. Moreover, Figure 3 suggests that during downturns, the level of loan loss reserves tends to increase more slowly than the level of non-performing loans.
4
Reports of Condition and Income are provided by the Federal Deposit Insurance Corporation (FDIC), and used by the bank regulatory agencies in their bank monitoring activities.
Current accounting standards permit banks to use environmental and other qualitative factors to identify expected loan losses. Therefore, the timing of a decision to identify a loan loss for balance sheet purposes, might be an extremely subjective assessment. Ahmed et al. (1999) suppose that bank managers have incentives to use loan loss provisions to manage earnings and regulatory capital, as well as to “signal” private information about future earnings. They conclude that capital management is an important determinant of loan loss provisions, if specific loan loss provisions can be used to reduce the level of Tier-1 capital5. Cavallo and Majnoni (2001) observe that the allocation of bank’s operating profits among income taxes, dividends and retained earnings is a highly subjective decision, and the definition of loan loss provisions is affected by a host of agency problems. They adopt an agency approach on loan loss provisioning and find evidence that the protection of outsiders’ claims over bank profits (claims from minority shareholders and fiscal authority) has a negative impact on the level of loan loss provisioning.
From a supervisory point of view, banks should maintain a minimum level of loan loss provisioning based upon a pre-defined supervisory loan grading scheme. In practice, banks identify loan losses by categorizing loans based upon their credit quality, where non-performing loans reflect to loans in one of the lower credit quality categories. An incorrect loan grading scheme leads to inadequate loan loss provisioning and to a distortion in a bank’s balance sheet. Therefore, the adequacy of loan loss provisioning is only as good as the methodology used to estimate incurred loan losses embedded in the bank loan portfolio. In the context of loan loss provisioning, transparency and disclosures about a bank’s risk profile and risk management process can stimulate banks to maintain adequate levels of loan loss reserves and to recognize loan losses in a timely manner. Therefore, many researchers promote to make loan loss provisioning an integral part of risk-based capital regulations, to prevent capital adequacy concerns and to enhance financial stability.
5
3.2 Hypothesis
This paper suggests that loan loss provisioning might play a role in the monetary-transmission mechanism. The level of loan loss provisioning can be seen as the extent to which banks are able to absorb expected loan losses during a monetary policy tightening. When loan loss reserves become inadequate to absorb expected loan losses, bank capital has to offset both expected as well as unexpected loan losses. Therefore, the amount of retained earnings that is available to supplement the required level of regulatory capital is reduced. Due to financial market imperfections, a reduction in a bank’s capital base might have a negative impact on bank lending. This paper hypothesizes that low loan loss provisioning banks are more likely to be confronted with capital adequacy concerns, and therefore, those banks are more responsive to changes in the monetary policy stance than high loan loss provisioning banks. Moreover, it suggests that small banks in particular have the most difficulty with finding alternative uninsured liabilities during a monetary policy tightening, and therefore, it hypothesizes that the above described mechanism works exclusively within the class of small banks.
The hypothesis can be defined as follows:
H1: Within the class of small banks, bank lending is more sensitive to changes in the monetary policy stance for low loan loss provisioning banks.
Equation (1) reflects the hypothesis, and exploits both the cross-sectional derivative ∂L / ∂LLP , as well as the time-series derivative ∂L / ∂M :
∂ L / ∂LLP ∂M > 0 (1)
where L is a measure of bank i’s lending activity, LLP is a measure of the level of loan loss provisioning and M is a monetary policy indicator. An increase in the level of M corresponds to a tighter monetary policy.6
Equation (1) captures two intuitions, depending on the order in which one takes the derivatives. First, the degree to which bank lending depends on the level of loan loss provisioning is sensitive to the changes in the monetary policy stance. Second, the degree to which bank lending is sensitive to changes in the monetary policy stance depends on the level of loan loss provisioning.
This paper examines lending behavior at the individual bank level to distinguish loan demand from loan supply shifts. Furthermore, Appendix C investigates the role of liquidity in the monetary-transmission mechanism, to find out whether the “traditional” bank lending channel holds for this study. Finding evidence for the existence of the bank lending channel, could reinforce the argument that banks’ condition affects the impact of monetary policy on bank lending. Section 4 discusses the dataset.
6
4. Data
The individual bank level data come from the FDIC’s Reports of Condition and Income (also known as “Call Reports”). The dataset includes quarterly report data on the population of 11,785 U.S. insured commercial banks. Those banks are relevant for this study, since their activities are concentrated on loans and deposits. The dataset covers the period between 1993Q4 and 2010Q4, and yields 657,832 observations.
This paper imposes a few filters on the dataset. It eliminates bank observations in which the dependent variable (total gross loans) is more than five standard deviations from its mean, in order to delete banks that are involved in mergers that create discontinuities in the surviving banks’ balance sheets. It would be better to use merger files in eliminating those banks, however, that would fall outside the scope of this paper. It eliminates banks that do not give information on the independent variables, that are needed to conduct this study. Moreover, it requires banks to have at least five consecutive quarters of bank lending data, since the regressions include lagged values of the dependent variable. After imposing these filters, the dataset yields 433,234 observations. For the regressions that include commercial and industrial loans (C&I loans) as dependent variable, banks that focus on a negligible amount of C&I loans are eliminated. The C&I loans sample yields 394,526 observations, after omitting any bank for which C&I loans constitutes less than 5 percent of their total gross loan portfolio.
This paper uses total gross loans (panel A), as well as the subcategory C&I loans (panel B), as bank lending variable L . Kashyap and Stein (2000) argue that one reason for examining both is that the results for total gross loans might be influenced by asset composition effects. Assuming that C&I loan demand and real estate loan demand move differently over the business cycle, banks that tend to engage primarily in C&I loans might systematically hold different levels of loan loss provisioning. This could bias the estimated quantity ∂ L / ∂LLP ∂M .
This paper uses the ratio of securities plus federal funds sold against total assets as a measure of the liquidity variable LIQ . This ratio captures the extent to which banks have a buffer stock of liquid assets that they can draw down during a monetary policy tightening. It does not include any cash item, since cash is for a large part made up of required reserves that cannot be draw down freely. However, the results do not change when any cash item is included in the liquidity base.
This paper obtains macro-economic data from Datastream Advance 5.1. It uses the federal funds rate as a monetary policy indicator. Ashcraft (2006) argues that the federal funds rate is a good monetary policy indicator, which can be illustrated by a strong negative correlation between the federal funds rate and the ratio of insured deposits against total short-term liabilities. Kashyap and Stein (2000) use three different measures of the monetary policy stance, namely the Boschen-Mills (1995) index, the federal funds rate, and the Bernanke and Mihov (1998) indicator. They find that the federal funds rate has the most explanatory power.
The dataset is divided into three subsamples based on banks’ asset size. Consistent with previous literature, this paper defines small banks below the 75th percentile, medium banks between the 75th and 90th percentile, and large banks above the 90th percentile.7 Because of an extremely skewed nature of the asset size distribution, an overwhelming bank majority belongs to the small asset size class.
7
TABLE 1 Average values of the main balance sheet items
All banks Small banks Medium banks Large banks (<75%) (75% - 90%) (>90%)
Number of observations 433,234 317,732 70,788 44,714
Number of banks 11,785 8,786 1,796 1,203
Mean assets ($ millions) 1,092 92 352 9,367
As a fraction of total assets
Total gross loans 0.633 0.623 0.664 0.659
Real estate loans 0.395 0.379 0.461 0.405
Farm loans 0.052 0.065 0.018 0.009
Commercial and industrial loans 0.102 0.097 0.110 0.126
Loans to individuals 0.075 0.075 0.063 0.095
Cash 0.050 0.052 0.045 0.047
Securities 0.242 0.250 0.223 0.215
Federal funds sold 0.037 0.040 0.029 0.029
Loan loss reserves 0.010 0.010 0.010 0.011
Non-performing loans 0.009 0.009 0.009 0.009
TABLE 2 Descriptive statistics on the main variables
All banks Mean St. dev. Min. Max.
∆log (L ) 0.008 0.024 -0.165 0.181
LIQ ¹ 0.028 0.014 0.000 0.099
LLP ² 0.008 0.041 0.000 1.457
Small banks (<75%) Mean St. dev. Min. Max.
∆log (L ) 0.008 0.023 -0.165 0.181
LIQ ¹ 0.029 0.015 0.000 0.096
LLP ² 0.009 0.042 0.000 1.425
Medium banks (75% - 90%) Mean St. dev. Min. Max.
∆log (L ) 0.011 0.023 -0.165 0.181
LIQ ¹ 0.025 0.013 0.000 0.088
LLP ² 0.007 0.040 0.000 1.457
Large banks (>90%) Mean St. dev. Min. Max.
∆log (L ) 0.011 0.028 -0.165 0.181
LIQ ¹ 0.024 0.014 0.000 0.099
LLP ² 0.006 0.036 0.000 1.443
Macro-economic variables Mean St. dev. Min. Max.
∆M -0.078 0.476 -1.430 0.640
∆GDP 0.025 0.029 -0.089 0.080
NOTES: Table 2 shows some descriptive statistics on the main variables per asset size class, using the period 1995Q1 - 2010Q4. It uses the filtered dataset from the total gross loans sample that eliminates outliers and banks that do not provide information on the independent variables, needed to conduct this study. The dataset is divided into three subsamples, based on banks’ asset size. Small banks are defined below the 75th percentile, medium banks between the 75th and 90th percentile, and large banks above the 90th percentile, where percentiles are worked out on mean values. L refers to bank i’s lending activity, LIQ is a measure of liquidity, LLP is a measure of loan loss provisioning, M is a monetary policy indicator, and GDP is a measure of economic activity.
¹ For scaling purposes, actual ratios are divided by 10. ² For scaling purposes, actual ratios are divided by 1,000.
5. Methodology
This paper uses a two-step regression approach, based on Kashyap and Stein (2000), to estimate the quantity ∂ L / ∂LLP ∂M for banks in different asset size classes. It differs from the approach of Kashyap and Stein (2000) in the following way. Kashyap and Stein (2000) are interested, among others, in the small-bank/big-bank differential that measures whether the effect will be stronger for small banks, while this paper focuses on the role
of loan loss provisioning in the monetary-transmission mechanism. Besides that, it examines liquidity to find
evidence for the “traditional” bank lending channel theory, that could reinforce the argument that banks’ condition affects the impact of monetary policy on bank lending (see Appendix C).
5.1 The two-step regression approach
The two-step regression approach exists of running a sequence of cross-sectional regressions (first-step), and then using the estimated coefficients on the bank-specific variable in a time-series regression (second-step).
Equation (2) represents the first-step regression, that exists of running a series of cross-sectional regressions separately for each asset size class and each time period t:
∆log (L ) = c + ∑ α ∆log (L ) + β LIQ + γ LLP + ∑ θ D + ε (2)
where the log change in L is regressed against a constant c , four lags of the dependent variable, a liquidity variable LIQ , a loan loss provisioning variable LLP , and a Federal Reserve-district dummy variable D . The first-step regression includes lagged values of the dependent variable to account for heterogeneity across banks regarding lending behavior. The Federal Reserve-district dummy variable serves as a geographic control.
Equation (3) represents the so-called “univariate” specification from the second-step regression, that exists of running a purely time-series regression for each asset size class, where the estimated γ coefficients from Equation (2) are used as dependent variable:
= c + ∑ ∅ ∆M + δTIME + (3)
where γ is regressed against a constant c , some (lagged) values of the change in the monetary policy indicator M , and a time trend variable TIME that accounts for a possible linear time trend.
The key coefficient from this regression is ∅ = ∑ ∅ , that measures to what extent the degree to which bank lending depends on the level of loan loss provisioning is sensitive to changes in the monetary policy stance. This paper hypothesizes that within the class of small banks, the degree to which bank lending depends on the level of loan loss provisioning is positively correlated with the degree of a tightening of monetary policy. Therefore, the signs of the ∅ coefficients on monetary policy should be positive.
Equation (4) represents the so-called “bivariate” specification from the second-step regression, that includes also a measure of economic activity:
= c + ∑ ∅ ∆M + ∑ ∆GDP + δTIME + (4)
where GDP is a measure of economic activity, defined by the growth rate of real GDP. The coefficient = ∑ measures to what extent the degree to which bank lending depends on the level of loan loss provisioning is sensitive to changes in the level of economic activity. Section 5.2 discusses the relevance of including economic activity into the second-step regression. The parameters of the second-step regressions will be estimated using the Zellner’s Seemingly Unrelated Regression (SUR) model.8
The two-step regression approach attempts to disentangle loan demand from loan supply shifts by examining lending behavior at the individual bank level. Moreover, it assumes that when all banks in the dataset are confronted with the same demand shock at a given point in time, then differences in lending behavior across banks reflect exclusively differences in the banks’ loan supply function. More specifically, it assumes that the cross-sectionally estimated coefficient measures differences in the banks’ loan supply function, that can be attributed to differences in the level of loan loss provisioning.
8
This paper examines both the role of liquidity as well as the role of loan loss provisioning in explaining bank lending
behavior. Therefore, it uses a SUR model to account for the likely correlation between and , resulting from the fact that
5.2 Potential endogeneity bias
Kashyap and Stein (2000) argue that an endogeneity bias could arise when the first-step regressions include some endogenous explanatory variables. This section describes the main endogeneity issues.
The first-step regressions deliver estimates of the level of γ that are potentially biased, which could be either positive or negative. This paper is interested in the correlation between γ and M , and therefore a bias in the level of γ should not be an issue. However, there may be some endogenous influences on LLP that are more problematic, in that they lead to a bias in the estimated ∅ coefficients on monetary policy in the second-step regressions. The endogenous nature of loan loss provisioning is in particular true when there is an endogenous link between the level of LLP and the cyclical sensitivity of loan demand. The following two stories illustrate that the potential endogeneity bias could be either positive or negative.
First, the “heterogeneous risk aversion” story assumes that some banks are inherently more conservative than others. Therefore, conservative banks will be systematically less exposed to cyclically sensitive borrowers, as well as tend to protect themselves by holding larger values of LLP . So their might exist a negative correlation between LLP and the cyclical sensitivity of loan demand, and therefore, the correlation between LLP and bank lending is expected to be relatively low during upturns, and relatively high during downturns. This can lead to a bias in which the estimated ∅ coefficients on monetary policy in the second-step regressions are too positive.
Second, the “rational buffer-stocking” story assumes that all banks have the same risk aversion, but some have more opportunities to lend to cyclically sensitive customers than others. In this case, those banks that are systematically more exposed to cyclically sensitive borrowers, protect themselves by holding larger values of LLP . So their might exist a positive correlation between LLP and the cyclical sensitivity of loan demand, and therefore, the correlation between LLP and bank lending is expected to be relatively high during upturns, and relatively low during downturns. This can lead to a bias in which the estimated ∅ coefficients on monetary policy in the second-step regressions are too negative.
To some degree, one can find out the direction of the endogeneity bias by including economic activity into the second-step regressions. Under the “heterogeneous risk aversion” story, the sign of the coefficient on economic activity should be negative, while under the “rational buffer-stocking” story it should be positive. The intuition is as follows. Suppose that less conservative banks are more exposed to cyclically sensitive (risky) borrowers, and an increase in economic activity favors risky borrowers relatively more. Then under the “heterogeneous risk-aversion” story, an increase in economic activity has a more positive effect on the lending of
low loan loss provisioning banks. In contrast, under the “rational buffer-stocking” story, an increase in economic
activity has a more positive effect on the lending of high loan loss provisioning banks.
6. Results
TABLE 3 Main estimates from the first-step regressions
Panel A: Total gross loans
Dependent variable = ∆log (L ) γ
All banks 0.00601 (0.39463) Small banks (<75%) 0.00553 (0.40517) Medium banks (75% - 90%) 0.00943 (0.43370) Large banks (>90%) 0.00524 (0.58698)
Panel B: C&I loans
Dependent variable = ∆log (L ) γ
All banks 0.01227 (0.44067) Small banks (<75%) 0.01229 (0.44111) Medium banks (75% - 90%) 0.00925 (0.48722) Large banks (>90%) 0.01794 (0.48320)
NOTES: Table 3 shows the average values of the main estimates from the first-step regressions as represented by Equation (2). Each specification also includes four lagged values of the dependent variable, a liquidity variable and a Federal Reserve-district dummy variable. P-values are in parentheses, and calculated by estimating the coefficient covariance using the Ordinary Least Square (OLS) method. The dataset is divided into three subsamples, based on banks’ asset size. Small banks are defined below the 75th percentile, medium banks between the 75th and 90th percentile, and large banks above the 90th percentile, where percentiles are worked out on mean values.
¹ Average values for the period 1995Q1 - 2010Q4.
Second, the level of the γ coefficients depends on whether total gross loans or C&I loans are used as dependent variable. Therefore, it might be well reasonable to argue that the results from the total gross loans sample are influenced by asset composition effects, such that banks that tend to engage primarily in C&I lending hold systematically different levels of loan loss provisioning.
TABLE 4 Main estimates from the second-step regressions
Panel A: Total gross loans “Univariate” “Bivariate”
Dependent variable = ∅ ∅ All banks -0.00252 -0.00699 ** 0.15295 ** (0.36950) (0.02140) (0.02040) Small banks (<75%) -0.00180 -0.00578 * 0.14607 ** (0.53840) (0.08130) (0.04270) Medium banks (75% - 90%) -0.01462 -0.01638 0.06506 (0.70790) (0.72320) (0.94840) Large banks (>90%) -0.00557 -0.03759 1.34645 (0.90370) (0.49190) (0.25710)
Panel B: C&I loans “Univariate” “Bivariate”
Dependent variable = ∅ ∅ All banks -0.01328 * -0.01933 ** 0.22518 (0.09940) (0.04420) (0.28060) Small banks (<75%) -0.01037 -0.01952 * 0.39236 * (0.24290) (0.05940) (0.08110) Medium banks (75% - 90%) 0.00111 -0.02854 1.19136 (0.98510) (0.68290) (0.43250) Large banks (>90%) 0.04891 -0.01095 2.29756 (0.63640) (0.92550) (0.36660)
NOTES: Table 4 shows the main estimates from the second-step regressions as represented by Equations (3) and (4). The dependent variable is derived from Equation (2). Each specification also includes a linear time trend variable. P-values are in parentheses, and calculated by estimating the coefficient covariance using the Seemingly Unrelated Regression (SUR) method, that accounts for heteroscedasticity and contemporaneous correlation in the error terms across equations. The dataset is divided into three subsamples, based on banks’ asset size. Small banks are defined below the 75th percentile, medium banks between the 75th and 90th percentile, and large banks above the 90th percentile, where percentiles are worked out on mean values. Period: 1995Q1 - 2010Q4.
¹ Sum of coefficients: ∅ = ∑ ∅ and = ∑ .
Table 4 presents the main results from the second-step regressions, i.e. the sum of the ∅ coefficients on monetary policy and the sum of the coefficients on economic activity. It shows both the results from the “univariate” specification, as well as the results from the “bivariate” specification. Considering the results from the “univariate” specification, within the full sample of panel B, the estimated ∅ coefficient on monetary policy is significantly negative, with corresponding p-value of 9.9 percent. Considering the results from the “bivariate” specification, the statistical significance of the main coefficients on monetary policy becomes stronger. Within the full sample, the estimated ∅ coefficients on monetary policy are significantly negative with corresponding p-values of 2.1 percent (panel A) and 4.4 percent (panel B). Within the class of small banks, the estimated ∅ coefficients on monetary policy are also significantly negative, with corresponding p-values of 8.1 percent (panel A) and 5.9 percent (panel B).
In contrast to the hypothesis9, the results suggest that the degree to which bank lending depends on the level of loan loss provisioning is negatively correlated with the degree of a tightening of monetary policy. Accordingly, the results show evidence that within the class of small banks, the impact of the level of loan loss provisioning on bank lending is weaker during a monetary policy tightening. Alternatively, it follows that within the class of small banks, bank lending tends to be more sensitive to changes in the monetary policy stance for high loan loss provisioning banks.
It is important to mention that the results may be ascribed to an endogeneity bias induced by the “rational buffer-stocking” story. Section 5.2 explains that under this story, the correlation between the level of loan loss provisioning and the cyclical sensitivity of loan demand is positive. When the more cyclical sensitive borrowers are associated with high loan loss provisioning banks, an increase in economic activity has a more positive effect on the lending of those banks than that of low loan loss provisioning banks. Table 4 shows that within the class of small banks, the signs of the coefficients on economic activity are significantly positive, which is in line with the “rational buffer-stocking” story. Therefore, the estimated ∅ coefficients on monetary policy may be biased towards being too negative, and it becomes reasonable to argue that this paper is failing to reject the null-hypothesis, even when it is false.
Furthermore, Table 4 shows that within the class of medium and large banks, the main coefficients on monetary policy and economic activity are not significant. This result is supportive to the argument that small banks in particular may not be able to offset a drain in reservable deposits during a monetary policy tightening, since small banks have more difficulty with raising external finance than medium and large banks. Therefore, the results suggest that exclusively within the class of small banks, loan loss provisioning tends to play a role in the monetary-transmission mechanism.
Appendix B provides full details of the main results from the “bivariate” specification. Appendix C shows evidence for the existence of the “traditional” bank lending channel, that supports the argument that banks’ condition affects the impact of monetary policy on bank lending. Section 7 describes some robustness checks.
9 The hypothesis can be defined as follows: ∂ L / ∂LLP ∂M > 0. Remember that this equation captures two intuitions,
7. Robustness checks
7.1 The one-step regression approach
Kashyap and Stein (2000) argue that the two-step regression approach probably errs on the side of being “over-parameterized”. Moreover, the two-step regression approach imposes a constraint of a linear and constant relationship between monetary policy and bank lending, which might be extremely restrictive. Therefore, this paper considers an alternative one-step regression approach, as proposed by Kashyap and Stein (2000), to impose more structure on the dataset. This approach will examine both the cross-sectional and time-series dimensions of the dataset into a single one-step regression.
Equations (5) and (6) represent the one-step regressions within a respectively “univariate” and “bivariate” specification:
∆ log(L ) = c + ∑ α ∆log (L ) + ∑ σ ∆M + ∑ θ D + δTIME + ∑ ρ QUARTER +
LIQ β + ∑ ∅ ∆M + LLP γ + ∑ ∅ ∆M + ε (5)
∆ log(L ) = c + ∑ α ∆log (L ) + ∑ σ ∆M + ∑ π ∆GDP + ∑ θ D + δTIME +
∑ ρ QUARTER + LIQ β + ∑ ∅ ∆M + ∑ GDP + LLP γ + ∑ ∅ ∆M +
∑ GDP + ε (6)
TABLE 5 Main estimates from the one-step regression approach
Panel A: Total gross loans “Univariate” “Bivariate”
Dependent variable = ∆log (L ) ∅ ∅
All banks -0.00531 ** -0.00426 -0.01552 (0.03170) (0.12930) (0.79000) Small banks (<75%) -0.00565 ** -0.00542 * 0.03378 (0.03640) (0.08710) (0.59380) Medium banks (75% - 90%) -0.01479 ** -0.01078 * -0.13641 (0.01970) (0.09880) (0.44270) Large banks (>90%) 0.00530 0.01018 0.13376 (0.69090) (0.47390) (0.66740)
Panel B: C&I loans “Univariate” “Bivariate”
Dependent variable = ∆log (L ) ∅ ∅
All banks -0.01644 ** -0.01217 -0.00794 (0.02680) (0.15200) (0.96350) Small banks (<75%) -0.01660 * -0.01648 0.12778 (0.05840) (0.11050) (0.53100) Medium banks (75% - 90%) -0.04417 ** -0.03243 * -0.68403 (0.01490) (0.08300) (0.17220) Large banks (>90%) 0.04513 * 0.05402 ** -0.23140 (0.05670) (0.03980) (0.96710)
NOTES: Table 5 shows the main estimates from the one-step regression approach as represented by Equations (5) and (6). Each specification also includes lagged values of the dependent variable, (lagged) values of the monetary policy indicator, (lagged) values of the economic activity indicator within the “bivariate” specification, a Federal Reserve-district dummy variable, a time trend variable, a quarter dummy variable, and the bank-specific variables liquidity and loan loss provisioning. P-values are in parentheses, and calculated by estimating the coefficient covariance using the Ordinary Least Square (OLS) method. The dataset is divided into three subsamples, based on banks’ asset size. Small banks are defined below the 75th percentile, medium banks between the 75th and 90th percentile, and large banks above the 90th percentile, where percentiles are worked out on mean values. Period: 1995Q1 - 2010Q4.
¹ Sum of coefficients: ∅ = ∑ ∅ and = ∑ .
Table 5 presents the main results from the one-step regression approach. Within the class of small and medium banks, the signs of the estimated ∅ coefficients on monetary policy are negative, with corresponding p-values between 1.5 and 11.1 percent. The results are basically consistent with the results from the two-step regression approach. However, within the class of large banks in panel B, the signs of the estimated ∅ coefficients on monetary policy become significantly positive, with corresponding p-values of 5.7 and 4.0 percent. These results make clear that within the class of small and medium banks, high loan loss provisioning banks tend to be more sensitive to changes in the monetary policy stance. Within the class of large banks, the opposite might be true in that low loan loss provisioning banks tend to be more sensitive to changes in the monetary policy stance.
The one-step regression approach seems to be a more powerful test than the two-step regression approach, since the prior imposes more structure on the data, and therefore, it does not eliminate much of the variation in the dataset. However, there might exist some econometric problems, when the assumptions about the error terms of OLS regression do not hold. First, the loan loss provisioning variable may have some endogenous influences, and therefore the regressors may be correlated with the error terms. Second, the inclusion of lagged dependent variables gives rise to autocorrelation, in that the covariance between the error terms over time might be unequal to zero. Third, the error terms might contain fixed effects, that consist of unobserved bank-specific effects.
To deal with the above described econometric problems, this paper implements an instrumental variables approach that controls for endogeneity of the explanatory variables and allows for the inclusion of lagged values of the dependent variable as regressors. Unfortunately, it seems to be very difficult to find valid instruments that are uncorrelated with the error terms10. Besides that, Kashyap and Stein (2000) indicate that much of the variation in the explanatory variables will be lost by adopting a level fixed effect, and the remaining bank-specific variation in the explanatory variables is contaminated by the kind of endogeneity that is most difficult to address. Therefore, Section 7.2 introduces a “quasi” instrumental variables procedure.
10
7.2 The “quasi” instrumental variables approach
If there is an endogenous link between loan loss provisioning and the cyclical sensitivity of loan demand, then it would be meaningful to have an instrument for the bank-specific variable that is uncorrelated with loan cyclicality. However, as described before, it seems to be very difficult to find such an obvious truly exogenous instrument. Therefore, Kashyap and Stein (2000) introduce a “quasi” instrumental variables approach, that regresses the bank-specific variables against a plausible observable measure of loan cyclicality, and uses the residuals from this regression as instruments.
Equation (7) represents the “zero-step” regression, where for each asset size class, loan loss provisioning is regressed against an observable measure of loan cyclicality:
LLP = µRE + πFA + ρCI + σIN + δTIME + ε (7)
TABLE 6 Main estimates from the second-step regressions, using a “quasi” instrumental variables approach.
Panel A: Total gross loans “Univariate” “Bivariate”
Dependent variable = ∅ ∅ All banks -0.00260 -0.00689 ** 0.14438 ** (0.36870) (0.02690) (0.03270) Small banks (<75%) -0.00223 -0.00588 * 0.13072 * (0.44710) (0.07750) (0.07070) Medium banks (75% - 90%) -0.00189 -0.01539 0.56316 (0.95990) (0.72550) (0.55420) Large banks (>90%) 0.01337 -0.02954 1.79445 (0.77780) (0.59730) (0.13950)
Panel B: C&I loans “Univariate” “Bivariate”
Dependent variable = ∅ ∅ All banks -0.01319 * -0.01863 * 0.19941 (0.09930) (0.05110) (0.33650) Small banks (<75%) -0.01058 -0.01897 * 0.36085 (0.22850) (0.06520) (0.10640) Medium banks (75% - 90%) -0.00963 -0.02738 0.71064 (0.87480) (0.70490) (0.65100) Large banks (>90%) 0.02702 -0.01892 1.65037 (0.79820) (0.87230) (0.51870)
NOTES: Table 6 shows the main estimates from the second-step regressions as represented by Equations (3) and (4). The dependent variable is derived from Equation (2), where the residuals from Equation (7) are used as the bank-specific variable. P-values are in parentheses, and calculated by estimating the coefficient covariance using the Seemingly Unrelated Regression (SUR) method, that accounts for heteroscedasticity and contemporaneous correlation in the error terms across equations. The dataset is divided into three subsamples, based on banks’ asset size. Small banks are defined below the 75th percentile, medium banks between the 75th and 90th percentile, and large banks above the 90th percentile, where percentiles are worked out on mean values. Period: 1995Q1 - 2010Q4.
¹ Sum of coefficients: ∅ = ∑ ∅ and = ∑ .
* Significant at 10 percent confidence level; ** Significant at 5 percent confidence level; *** Significant at 1 percent confidence level.
8. Conclusion
Previous literature on the monetary-transmission mechanism find evidence that there are some cross-sectional differences in the way banks respond to changes in the monetary policy stance. This paper adds to existing literature by investigating whether loan loss provisioning influences the impact of monetary policy on bank lending. By analyzing loan loss provisioning as an additional differentiating bank specific characteristic, this paper extends the two-step methodology as proposed by Kashyap and Stein (2000).
This paper assumes that financial markets are imperfect, and that low loan loss provisioning banks are the most likely to be confronted with capital adequacy concerns. Besides that, small banks in particular may not be able to offset a drain in reservable deposits during a monetary policy tightening, since those banks have more difficulty with raising external finance than medium and large banks. Therefore, it hypothesizes that within the class of small banks, bank lending is more sensitive to changes in the monetary policy stance for low loan loss provisioning banks. The overall findings do not support the hypothesis, since they suggest that within the class of small banks, bank lending depends less on the level of loan loss provisioning during a monetary policy tightening. Alternatively, the results suggest that within the class of small banks, high loan loss provisioning banks tend to be more sensitive to changes in the monetary policy stance.
The results from this paper may be driven by an endogeneity problem, if there is an endogenous link between the level of loan loss provisioning and the cyclical sensitivity of loan demand. Moreover, this paper finds evidence that is supportive to the “rational buffer-stocking” story. Under this story, an increase in the level of economic activity has a more positive impact on the lending of high loan loss provisioning banks. Therefore, the results on the monetary policy indicator may be biased towards being too conservative, i.e. failing to reject the null-hypothesis, even when it is false. This paper attempts to absorb some of the endogenous variation in the level of loan loss provisioning by introducing a “quasi” instrumental variables approach. The results from this approach are basically consistent with previous mentioned findings, and therefore it becomes unlikely that the main results are driven by an endogeneity bias. After performing some robustness checks, this paper contradicts the hypothesis and concludes:
Within the class of small banks, bank lending tends to be more sensitive to changes in the monetary policy stance for high loan loss provisioning banks.
The overall findings may be explained by the argument of some researchers, who suppose that the aggregate impact of monetary policy on bank lending depends on the financial condition of the banking sector. They suggest that the bank capital channel becomes weaker when bank lending is tied to risk-based capital requirements and bank equity is at or below the regulatory minimum for a large fraction of banks.11 Under these circumstances, monetary policy will only be effective for those banks with sufficiently high levels of regulatory capital.
11
Accordingly, if provisions are set to cover expected loan losses and bank capital should absorb unexpected loan losses, then one might argue that the banking sector requires a more precise methodology for adequately estimating the incurred losses embedded in the loan portfolio. This will stimulate banks to provision for expected loan losses in a timely manner, which might have a positive impact on the effectiveness of monetary policy.
References
Ahmed, A.S., C. Takeda, S. Thomas. 1999. “Bank loan loss provisions: a reexamination of capital management, earnings management and signaling effects” Journal of Accounting and Economics 28: 1 - 25
Angklomkliew, S., J. George, F. Packer. 2009. “Issues and developments in loan loss provisioning: the case of Asia” BIS Quarterly Review December: 69 - 83
Arellano, M. and S. Bond. 1991. “Some Tests of Specification for Panel Data: Monte Carlo Evidence and an Application to Employment Equations.” Review of Economic Studies 58: 277 - 297
Ashcraft, A.B. 2006. “New evidence on the Lending Channel” Journal of Money, Credit, and Banking 38: 751 - 775
Balla, E. and A. McKenna. 2009. “Dynamic Provisioning: A Countercyclical Tool for Loan Loss Reserves”
Economic Quarterly 95: 383 - 418
Beatty, A. and S. Liao. 2011. “Do delays in expected loss recognition affect banks’ willingness to lend?” Journal
of Accounting and Economics 52: 1 - 20
Bernanke, B.S. and A.S. Blinder. 1992. “The Federal Funds Rate and the Channels of Monetary Transmission”
The American Economic Review 82: 901 - 921
Bernanke, B.S. and I. Mihov. 1998. “Measuring Monetary Policy” Quarterly Journal of Economics 113: 869 - 902
Boschen, J. and L. Mills. 1995. “The Effects of Countercyclical Policy on Money and Interest Rates: An Evaluation of Evidence from FOMC Documents.” Federal Reserve Bank of Philadelphia Working Paper No. 91-20
Bouvatier, V. and L. Lepetit. 2008. “Banks’ procyclical behavior: Does provisioning matter?” Int. Fin. Markets,
Inst. and Money 18: 513 - 526
Cavallo, M. and G. Majnoni. 2001. “Do Banks Provision for Bad Loans in Good Times? Empirical Evidence and Policy Implications.” World Bank Policy Research Working Paper No. 2619
Gambacorta, L. and P.E. Mistrulli. 2004. “Does bank capital affect lending behavior?” Journal of Financial
Intermediation 13: 436 - 457
Jiménez, G. and J. Saurina. 2006. “Credit Cycles, Credit Risk, and Prudential Regulation.” International Journal
of Central Banking 2: 65 - 98
Jiménez, G., S. Ongena, J.L. Peydró, J. Saurina. 2010. “Credit supply and Monetary Policy: Identifying balance-sheet channels with loan applications and granted loans” ECB working paper series no. 1179
Kashyap, A.K., J.C. Stein, D.W. Wilcox. 1993. “Monetary Policy and Credit Conditions: Evidence from the Composition of External Finance” American Economic Review 83: 78 - 98
Kashyap, A.K. and J.C. Stein. 1994. “Monetary Policy and Bank Lending” Studies in Business Cycles 29: 221 - 256
Kashyap, A.K. and J.C. Stein. 1995. “The Impact of Monetary Policy on Bank Balance Sheets.” Carnegie
Rochester Conference Series on Public Policy 42: 151 - 195
Kashyap, A.K. and J.C. Stein. 2000. “What Do a Million Observations on Banks Say About the Transmission of Monetary Policy?” The American Economic Review 90: 407 - 428
Kishan, R.P. and T.P. Opiela. 2000. “Bank Size, Bank Capital, and the Bank Lending Channel.” Journal of
Money, Credit and Banking 32: 121 - 141
Laeven, L. and G. Majnoni. 2003. “Loan loss provisioning and economic slowdowns: too much, too late?”
Journal of Financial Intermediation 12: 178 - 197
Modigliani F. and M.H. Miller. 1958. “The Cost of Capital, Corporation Finance and the Theory of Investment”
The American Economic Review 48: 261 - 297
Nichols, D.C., J.M. Wahlen, M.M. Wieland. 2009. “Publicly traded versus privately held: implications for conditional conservatism in bank accounting” Review Accounting Studies 14: 88 - 122
Resti, A. and A. Sironi. 2007. “Risk management and shareholders’ value in banking” John Wiley and Sons
Van den Heuvel, S.J. 2002. “Does Bank Capital Matter for Monetary Transmission?” Economic Policy Review May 259 - 265
Van den Heuvel, S.J. 2009. “The Bank Capital Channel of Monetary Policy.” Working paper, Federal Reserve
Appendix A
Throughout the analysis, this paper uses the following bank-specific variables as defined by the Federal Deposit Insurance Corporation (FDIC)12:
Variable Definition
Total assets The sum of all assets owned by the institution including cash, loans, securities, bank premises and other assets. This total does not include off-balance-sheet accounts. Total gross loans Total loans and lease financing receivables, net of unearned income.
Real estate loans Loans secured primarily by real estate, whether originated by the bank or purchased. Farm loans Loans to finance agricultural production and other loans to farmers. Excludes savings
institutions filing a TFR. Commercial and
industrial loans
Commercial and industrial loans. Excludes all loans secured by real estate, loans to individuals, loans to depository institutions and foreign governments, loans to states and political subdivisions and lease financing receivables.
Loans to individuals Loans to individuals for household, family, and other personal expenditures including outstanding credit card balances and other secured and unsecured consumer loans. Cash Total cash and balances due from depository institution including both
interest-bearing and noninterest-interest-bearing balances.
Securities Total investment securities (excludes securities held in trading accounts).
Federal funds sold Total federal funds sold and securities purchased under agreements to resell in domestic offices. Includes only federal Funds sold for TRF Reporters before March 1998.
Loan loss reserves Each bank must maintain an allowance (reserve) for loan and lease losses that is adequate to absorb estimated credit losses associated with its loan and lease portfolio (which also includes off-balance-sheet credit instruments).
Non-performing loans Assets past due 90 days or more, plus assets placed in nonaccrual status.
12
Appendix B
TABLE A.1. Full details of the main estimates from the “bivariate” specification of the second-step regressions
Panel A: Total gross loans ∅
Dependent variable = j=0 j=1 j=2 j=3 j=4 j=0 j=1 j=2 j=3 j=4 All banks -0.00425 -0.00162 0.00259 0.00567 -0.00936 0.10555 0.08108 -0.04151 -0.06491 0.07273 R² = 0.30426 (0.13470) (0.59900) (0.40550) (0.05840) (0.00060) (0.00940) (0.05720) (0.33100) (0.12840) (0.08140) Small banks (<75%) -0.00024 -0.00159 -0.00028 0.00433 -0.00800 0.07927 0.04831 0.02613 -0.04984 0.04220 R² = 0.22787 (0.93790) (0.63750) (0.93330) (0.18400) (0.00640) (0.07180) (0.29690) (0.57500) (0.28410) (0.35240) Medium banks (75%-90%) -0.02002 -0.00552 0.03240 0.02139 -0.04463 -0.07719 0.06569 -0.38326 -0.57357 1.03340 R² = 0.06644 (0.64230) (0.90650) (0.49350) (0.63650) (0.26770) (0.89910) (0.91870) (0.55520) (0.37600) (0.10380) Large banks (>90%) -0.19855 0.25929 -0.05160 -0.06498 0.01825 0.39924 -0.13211 0.81948 -0.37368 0.63352 R² = 0.38333 (0.00020) (0.00000) (0.35700) (0.22630) (0.70100) (0.57960) (0.86230) (0.28710) (0.62540) (0.39700)
Panel B: C&I loans ∅
Dependent variable = j=0 j=1 j=2 j=3 j=4 j=0 j=1 j=2 j=3 j=4 All banks -0.00583 0.0130 -0.01363 0.00402 -0.01690 0.02173 0.10706 -0.00540 -0.09142 0.19320 R² = 0.18334 (0.51550) (0.18470) (0.16720) (0.66920) (0.04480) (0.86360) (0.42420) (0.96810) (0.49670) (0.14280) Small banks (<75%) -0.00513 0.02146 -0.01684 -0.00401 -0.01500 0.10053 -0.08591 0.10296 0.07018 0.20459 R² = 0.21928 (0.59510) (0.04340) (0.11370) (0.69230) (0.09740) (0.46140) (0.55150 (0.47920) (0.62810) (0.14980) Medium banks (75%- 90%) -0.00583 -0.00662 -0.02084 -0.03381 0.03856 -0.67735 0.59062 -0.01327 -0.56542 1.85677 R² = 0.09715 (0.92860) (0.92570) (0.77050) (0.62110) (0.52550) (0.46210) (0.54410) (0.98920) (0.56310) (0.05370) Large banks (>90%) -0.27395 0.23787 0.13868 -0.00724 -0.10631 4.02939 -1.23697 0.84991 0.47953 -1.82430 R² = 0.27525 (0.01340) (0.04770) (0.24830) (0.94960) (0.29720) (0.01010) (0.44880) (0.60550) (0.76980) (0.25540)
Appendix C
This paper also investigates whether the level of liquidity affects the degree to which bank lending is sensitive to changes in the monetary policy stance. Liquidity can be seen as the extent to which banks have a buffer stock of liquid assets that they can draw down during a monetary policy tightening. Based on the bank lending channel theory, this paper hypothesizes that bank lending is more sensitive to changes in the monetary policy stance for small and illiquid banks.
The hypothesis can be defined as follows:
H1: Within the class of small banks, bank lending is more sensitive to changes in the monetary policy stance for illiquid banks.
The two-step regression approach is used to estimate the quantity ∂ L / ∂LIQ ∂M for banks in different asset size classes. Equations (8) and (9) represent respectively the “univariate” and “bivariate” specifications from the second-step regressions, where the estimated β coefficients from Equation (2) are used as dependent variable:
β = c + ∑ ∅ ∆M + δTIME + (8)
β = c + ∑ ∅ ∆M + ∑ ∆GDP + δTIME + (9)
The key coefficient from these regressions is ∅ = ∑ ∅ , that measures to what extent the degree to which bank lending depends on the level of liquidity is sensitive to changes in the monetary policy stance. The tables given below show the main results. Table A.2. presents the results from the first-step regressions, Table A.3. presents the results from the second-step regressions, and Table A.4. and A.5. present the results from the robustness checks as discussed in Section 7.
The main results can be summarized as follows. Within the class of small and medium banks, the impact of the level of liquidity on bank lending is stronger during a monetary policy tightening. Alternatively, it follows that within the class of small and medium banks, bank lending tends to be more sensitive to changes in the monetary policy stance for illiquid banks. These results provide evidence that supports the existence of the “traditional” bank lending channel theory, which claims that banks’ condition affects the impact of monetary policy on bank lending.
TABLE A.2. Main estimates from the first-step regressions Panel A: Total gross loans
Dependent variable = ∆log (L ) β
All banks 0.06405 (0.12572) Small banks (<75%) 0.06914 (0.14197) Medium banks (75% - 90%) 0.05295 (0.31288) Large banks (>90%) 0.08585 (0.32963)
Panel B: C&I loans
Dependent variable = ∆log (L ) β
All banks 0.01587 (0.38955) Small banks (<75%) 0.02614 (0.41428) Medium banks (75% - 90%) -0.02687 (0.38714) Large banks (>90%) 0.04343 (0.46201)
NOTES: Table A.2. shows the average values of the main estimates from the first-step regressions as represented by Equation (2). Each specification also includes four lagged values of the dependent variable, a loan loss provisioning variable and a Federal Reserve-district dummy variable. P-values are in parentheses, and calculated by estimating the coefficient covariance using the Ordinary Least Square (OLS) method. The dataset is divided into three subsamples, based on banks’ asset size. Small banks are defined below the 75th percentile, medium banks between the 75th and 90th percentile, and large banks above the 90th percentile, where percentiles are worked out on mean values.
