Bennett S. LeBow College of Business

Transcription

1 Bennett S. LeBow College of Business Drexel E-Repository and Archive (idea) Drexel University Libraries The following item is made available as a courtesy to scholars by the author(s) and Drexel University Library and may contain materials and content, including computer code and tags, artwork, text, graphics, images, and illustrations (Material) which may be protected by copyright law. Unless otherwise noted, the Material is made available for non profit and educational purposes, such as research, teaching and private study. For these limited purposes, you may reproduce (print, download or make copies) the Material without prior permission. All copies must include any copyright notice originally included with the Material. You must seek permission from the authors or copyright owners for all uses that are not allowed by fair use and other provisions of the U.S. Copyright Law. The responsibility for making an independent legal assessment and securing any necessary permission rests with persons desiring to reproduce or use the Material. Please direct questions to archives@drexel.edu

2 The Cyclical Properties of Banks Price-Cost Margins Roger Aliaga-Díaz and María Pía Olivero June 27 Abstract When price-cost margins vary endogenously in response to aggregate shocks, their variation becomes an additional channel through which such shocks can affect economic activity. This was first recognized by Rotemberg and Woodford (1991 and 1992) and its implications have been widely studied in goods markets. Bernanke and Gertler (1989) and Bernanke, Gertler and Gilchrist (1998) study this accelerator effect in financial markets in theoretical frameworks. However, the cyclical behavior of price-cost margins in the market for credit has been given only scant empirical attention so far. The literature still lacks empirical evidence on this cyclicality as a necessary condition for the existence and importance of this financial accelerator. In this paper we attempt to start filling this gap and study the cyclical behavior of banks price-cost margins in the United States banking sector, using time series quarterly data for the period and applying the methodology proposed by den Haan (2) to study the comovement between variables. We find empirical evidence in support of the countercyclicality of margins, and are able to show that it is robust to controlling for the effects of credit risk and the term structure of interest rates. As a mechanism for the propagation of macroeconomic shocks, this cyclical behavior can have important implications for both stabilization policy and RBC theory. Keywords: Bank margins, time varying markups, business cycles, financial accelerator. JEL Classification: E32, E44, G21 We are grateful to Alastair Hall, John Seater, Martín Uribe and participants at the 26 Midwest Macroeconomics Meetings and the 26 Southern Economic Association Conference for helpful comments. Department of Economics, LeBow College of Business, Drexel University. 1

3 1 Introduction After the seminal contributions by Rotemberg and Saloner (1986) and Rotemberg and Woodford (1991 and 1992) an extensive body of theoretical and empirical literature has explored the endogenous variation of price-cost margins in response to aggregate shocks (Gali (1994), Ravn, Schmitt-Grohé and Uribe (26), Lebow (1992), Chevalier and Scharfstein (1995 and 1996), Galeotti and Schiantarelli (1998) and Bloch and Olive (21), among others). This literature focuses on goods markets and looks at how endogenous price-cost margins can become an additional channel through which such shocks affect the economy. For financial markets, Bernanke and Gertler (1989) and Bernanke, Gertler and Gilchrist (1996 and 1998) study the role of an endogenous external finance premium (the difference between the cost of funds raised externally and the opportunity cost of funds internal to the firm) as an amplifier of business fluctuations. In their principal-agent model, the borrowers net worth acts as a source of output dynamics as it is inversely related to the agency cost and the external finance premium of financing real capital investment. Aggregate shocks are exacerbated in this framework as a result. Therefore, to the extent that borrowers net worth is procyclical, this theory predicts countercyclical external finance premia. With banks marginal costs of funds considered a good proxy for firms opportunity costs of internal funds 1, evidence on the countercyclicality of banks price-cost margins (calculated basically as the difference between the interest rate on loans and the marginal cost of funds for banks) provides some evidence on the necessary conditions for the exis- 1 In Bernanke, Gertler and Gilchrist (1998) this opportunity cost is the risk-free interest rate obtained by households on their savings. 2

4 tence of the financial accelerator in Bernanke and Gertler (1989) 2. So far, the literature on the empirical evidence for the financial accelerator focuses on firms and looks at two of its implications: first, the flight to quality in credit extension and second, the differences in real activity between firms with different levels of agency costs 3. This work pays only scant attention to the cyclical properties of margins themselves. The banking literature also suggests that there is an important cyclical component in banks price setting behaviour and lending standards, and argues that this could have important policy implications for bank regulation 4. This literature has looked at these changes in lending standards but, in general, without looking at the role of macroeconomic activity 5. Despite this influential literature, empirical research has not devoted much attention to the cyclical behavior of margins in credit markets. Therefore, the goal of our paper is 2 The countercyclicality is a necessary but not sufficient condition for the existence of the accelerator. The fact that margins are countercyclical does not guarantee that they exacerbate the effect of aggregate shocks on economic activity. Moreover, the reason for the countercyclicality observed in the data might be different from that advocated by Bernanke and Gertler (1989) in their theoretical model. 3 Several empirical studies support the hypothesis that in recessions credit flows away from borrowers more subject to agency costs. Previous work also finds important cross-sectional differences between borrowers more and less subject to agency costs in how real economic activity responds to adverse shocks. For details on this strand of the literature see Gertler and Gilchrist (1994), Bernanke, Gertler and Gilchrist (1996) and Oliner and Rudebusch (1995, 1996a, 1996b). 4 See Weinberg (1995) and references therein. 5 An exception is Lown et al (2) who use a VAR analysis augmented with macroeconomic variables. They find that lenders set their standards based on their own lending capacity and their expectations, and that standards are relatively exogenous to macroeconomic developments. Lown et al (2) measure standards by the percentage of loan officers participating in the Federal Reserve s Senior Loan Officer Opinion Survey who report tightening standards for commercial loans to large and medium-sized firms. 3

5 to study this cyclical behavior in the United States and to document a fact about business cycles that has received very little attention before. As a by-product, this paper attempts to start filling the gap between theoretical and empirical research by providing some evidence on the necessary conditions for the existence of the financial accelerator. We use time series quarterly data for the period and apply the methodology proposed by den Haan (2) to measure the comovement between variables. To our knowledge, so far the only empirical evidence on this cyclicality is Asea and Blomberg (1998). They use a Markov switching panel model and show that banks change their lending standards systematically over the cycle. In their framework the pricing of loans depends on the state of the loan market, which is assumed to switch between two states. They find evidence consistent with the markup over the banks cost of funds increasing with costs in high-risk states (recessions). However, according to their results, banks charge a constant markup over their cost of funds during low-risk states (expansions). Also related to our study are the working papers by Dueker and Thornton (1997) and Chen, Higgins and Mason (25). Based on a model where switching costs for borrowers are combined with risk-averse bank management, Dueker and Thornton (1997) study the hypothesis that the markup of the bank prime lending rate over the marginal cost of funds for banks (measured by the rate on certificates of deposits) is countercyclical and asymmetric in its response to recessionary and expansionary shocks 6. They find evidence that in cyclical downturns banks opt for a relatively high price-cost margin. The weekly frequency of their study restricts their choice of the business cycle indicator. Thus, they use the spread between the commercial paper rate and the Treasury bill rate and the slope of 6 Their novel time-series application of a conditionally heteroscedastic ordered probit model is crucial to study the asymmetric behavior of markups. 4

6 the yield curve as proxies for aggregate economic conditions. They do not study the effect of default risk or the term structure of interest rates on the cyclical behavior of margins. Chen, Higgins and Mason (25) estimate translog flexible function specifications for banks costs and profit efficiency in the US economy, and they find evidence for a substantial element of procyclicality in efficiency. Relative to the previous literature our contribution is threefold. First, we use a methodology different from Asea et al. so that the state of the economy is measured by a continuous variable instead of assuming a Markov switching structure. Also, they focus on survey data on commercial and industrial (C&I) loans while we use population data from the Report of Condition and Income on all types of loans. Second, we use the methodology in den Haan (2) which has been proposed as a tool to characterize the comovement between variables that is robust to detrending methods. den Haan s approach captures important information on the system dynamics that is lost when using the standard correlation approach. Third, we assess whether there is any explanatory power left to business cycles per se in the cyclicality of margins after accounting for changes in banking regulation, credit risk, the term structure of interest rates and monetary policy. Ex-ante, these could be suggested as the main, and maybe the only, determinants of the cyclical behavior of margins. In most cases we find that margins are negatively correlated with aggregate economic activity and further, that the sign and statistical significance of the correlation is robust to the inclusion in our specifications of these potentially exclusive determinants of the cyclical behavior of margins. Our results are consistent across alternative definitions for margins and business cycle measures. We believe the countercyclicality of margins in the market for credit has important 5

7 implications for both macroeconomic theory and stabilization policy, in particular as a mechanism for the propagation of macroeconomic shocks. With price-cost margins in the market for credit being countercyclical, a financial accelerator seems to be operating in the United States economy. In bad times, countercyclical margins make credit become more expensive relative to economies with constant margins. Firms may as a result delay investment and production decisions and recessions may be made even worse. Further research should assess whether this fact provides additional grounds for stabilization policy in economies where the countercyclicality of margins is stronger. The paper proceeds as follows. Section 2 presents the data and some preliminary evidence on the cyclical properties of margins. Section 3 describes the econometric methodology. Section 4 presents the estimation results for a reduced form model of margins. The last section concludes and outlines some directions for further research. The appendices contain a detailed description of the data as well as the results of some robustness checks performed on the model. 2 The Data and Preliminary Evidence on the Cyclicality of Margins This study uses time series quarterly data for the period Balance sheet and income data for banks is taken from the Call Reports on Condition and Income, available from the Federal Reserve Bank of Chicago for all banks regulated by the Federal Reserve System, the Federal Deposit Insurance Corporation, and the Comptroller of the Currency. We decided to start our series in the first quarter of 1984 given that our measures of margins 6

8 pre-1984 exhibit a significantly higher volatility than in the post-1984 period 7. Also, for some of our margin measures the data start in There is an important change in the Call Reports between 1987 and We have built consistent time series for this study following Kashyap and Stein (1997). Based on these bank-level data, we construct aggregate series for price-cost margins as weighted averages over the banks individual margins, with the weights given by each bank s share in total loans. Although admittedly there are some heterogeneities across banks (mainly depending on their size and geographical location), we want to study a macroeconomic issue and to present a fact about US business fluctuations that the RBC literature can build upon. This literature does not take into account these sources of bank heterogeneity. This is also important when addressing our second goal of providing some empirical evidence on the necessary conditions for the existence of the Bernanke and Gertler (1989) financial accelerator. Three alternative business cycle measures are used in this study: Gross Domestic Product, the unemployment rate and total loans. Two remarks are worth making regarding the inclusion of loans as a business cycle indicator. First, the total loans measure corresponds to the sum of loans supplied by all banking institutions, and not average loans for each bank in the sample. Therefore, loans are a good proxy for the aggregate level of economic activity. Second, adding loans as an alternative measure is useful because they may be even more sensitive to the cycle than GDP. It is also conjectured here that loans may reflect more closely than GDP the behavior of aggregates that are key to study the cyclicality 7 This high volatility of interest rates and margins might be explained in part by the Federal Reserve s change in operating target during the early 198s, and we want to prevent our data from being affected by changes in the policy rule followed by the monetary authority. 7

9 of banks price-cost margins. This is because these aggregates, such as investment and production, depend critically on bank financing. Figure 1 plots GDP together with loans. Six alternative definitions are used here for margins. Margin 1 is calculated as the difference between the ratio of interest income on loans to total loans and the ratio of interest expenses on deposits to total deposits. Margin 2 is calculated as the ratio of the difference between interest income and expenses to banks assets. Margin 3 is calculated as the ratio of the difference between interest income and expenses to loans. Thus, margin 1 more closely measures an interest rate spread, whereas margins 2 and 3 are strictly net interest margins ( NIMs ). NIMs are calculated as the ratio of the difference between interest revenues and interest expenses to assets. The fourth measure is given by the spread between the bank prime and Treasury bill rate, the fifth is the spread between the bank prime and the federal funds rate, and the sixth is the spread between the lending rate taken from the Survey of Terms of Business Lending and the federal funds rate 8. Following Nabar, Park and Saunders (1993) and Feinberg (21), among others, we use the Treasury bill and the federal funds rates as proxies for the cost of funds for commercial banks. The idea is that when the Federal Reserve boosts short-term interest rates by raising the federal funds rate, it makes it harder and more expensive for banks to raise funds. The reader is referred to Appendix A for details on variable definitions and sources. 8 The Survey of Terms of Business Lending collects data on gross loan extensions made during the first full business week in the middle month of each quarter. The authorized size for the survey is 348 domestically chartered commercial banks and 5 U.S. branches and agencies of foreign banks. The sample data are used to estimate the terms of loans extended during that week at all domestic commercial banks and at all U.S. branches and agencies of foreign banks. 8

10 Figures 2-4 plot the price-cost margin measures against each of the three indicators of economic activity. To provide a first insight on the cyclicality of margins Table 1 shows the sample unconditional correlations between alternative measures of margins and business cycle indicators. All variables are detrended using the Hodrick-Prescott filter 9. In most cases the contemporaneous correlations are negative and significant (positive and significant when the unemployment rate is used as a business cycle indicator), providing some evidence on the countercyclicality of margins. However, these correlations must be taken as non-conclusive evidence since they may be affected by changes in banking regulation, seasonality of the data and more importantly, credit risk, maturity mismatches between banks assets and liabilities and monetary policy. 3 Empirical Methodology There are two reasons why the unconditional correlations presented in Section 2 may not be appropriate for studying the comovement of variables at business cycle frequencies. First, they fail to capture important information about the dynamic aspects of the comovement of variables. Second, given that unconditional correlation coefficients are defined only for stationary series, the data must be first transformed accordingly, and this makes both the magnitude and sign of the correlations very sensitive to the detrending methods used to obtain stationary series. In light of these considerations, Rotemberg (1996) argues that to study the relationship 9 Correlations for band-pass filtered data show a similar pattern and are available from the authors upon request. 9

11 between the expected growth rates of two variables it is better to work with the part of the expected growth rates that is not autocorrelated, and that revisions in expected growth rates meet this criterion. Thus, he uses a VAR to construct expected changes and revisions in expected changes of the variables in the VAR. He proposes a short-run statistic given by the correlation coefficient of one-period ahead forecast errors, and a long-run statistic given by the correlation between the revisions in the expected changes. However, den Haan (2) argues that looking only at these two statistics still provides limited information on the comovement of variables. He shows that two economies with different dynamics for prices and GDP still have the same value for the two statistics proposed by Rotemberg (1996). Therefore, den Haan (2) proposes an alternative approach also based on VAR estimates. The idea is to use the correlations of VAR forecast errors at various horizons to measure the comovement between variables. To address these issues when studying the comovement between banks net interest margins and aggregate economic activity, we use the methodology proposed by den Haan (2). An advantage of this approach is that it captures important information about the dynamic aspects of the comovement of variables. Impulse response functions also provide this type of information about the dynamics, but the identifying assumptions needed to estimate impulse response functions are often ad-hoc and the results depend on the particular type of assumptions made. Furthermore, the methodology we use is appealing because it does not require assumptions about the order of integration of the variables either. There is no need for detrending as working with stationary variables is not a requirement. The vector of variables included in the VAR can contain any combination of stationary processes and processes that are integrated of arbitrary order. The only restriction is that the number of lags is large enough 1

12 to guarantee that the vector of innovations is not integrated. Therefore, in this study we restrict the minimum number of lags to two. The VAR we estimate is shown in equation (1). L X t = α + µ 1 t + µ 2 t 2 K2 3 + β l X t l + δ i R it + θ i Q it + ɛ t (1) l=1 i=1 i=1 where t denotes time, L is the total number of lags included in the equations, and X t is a vector of variables and includes the margin measure, the business cycle indicator and some other macroeconomic variables that are conjectured to be endogenously determined with the aggregate level of economic activity and margins. The R matrix includes dummy variables to control for two important regulatory changes that took place in the United States banking sector during the period covered by this study. First, in 1994 the Riegle-Neal Interstate Banking and Branching Efficiency Act repealed the Douglas Amendment. It allowed national banks to operate branches across state lines after June 1, Second, the Gramm-Leach-Bliley Act (GLBA) enacted in November of 1999 increased the number of activities allowed for banks and their holding companies (before 1999 commercial banks were prevented from expanding into a wide range of financial services such as investment banking) 1. The Q matrix includes dummy variables to control for seasonality in the quarterly data. Stacking the VAR(L) system in equation (1) into a VAR(1) we obtain 1 Notice that the fact that our sample period starts in 1984 avoids the need to control for the Depository Institutions Deregulation and Monetary Control Act of 198, which eliminated the deposit interest rate ceilings imposed by Regulation Q and increased the limit of deposit insurance by the FDIC from $4, to $1, per account. 11

13 Ξ t = BΞ t 1 + α + µ 1 t + µ 2 t 2 + K2 i=1 δ i R it + 3 i=1 θ i Q it. + ɛ t. (2) where Ξ t = X t X t 1. X t L+1 and B = β 1 β 2... β L 1 β L I I Following Seppala and Xie (25) and defining Ξ ue t+k = Ξ t+k E t Ξ t+k, the K-period ahead forecast error, then it can be shown that the variance-covariance matrix of the forecast error is computed as K 1 V ar(ξ ue t+k) = B j j= Ω (B j ) (3) where Ω = V ar(ɛ t ). Denoting by n the number of endogenous variables in the system, the n n upper block of the V ar(ξ ue t+k) matrix is the variance-covariance matrix for the K-period ahead forecast error of X t+k. We denote by COV (K) the covariance between the K-period ahead forecast errors of the two variables of interest (i.e. a measure of margins and a business cycle indicator), and by COR(K) their correlation coefficient. If the series for both variables are stationary, then the correlation coefficient of the forecast errors will converge to the unconditional correlation coefficient of the two series as 12

14 K goes to infinity. When some of the series are not stationary, the correlation coefficient may not converge but it can still be estimated consistently for a given forecast horizon. Thus, no assumption is needed about the order of integration of the series (den Haan, (2)). In general, if the correlation coefficient between two series of forecast errors is positive, then these two series are said to comove. Working with forecast errors implies working with correlations conditional on the time t information set on the dynamics of the system. After a shock, the correlation coefficient stabilizes when the variables in the system reach a stationary equilibrium. With this interpretation in mind and given that according to the NBER, cycles last between 4 and 32 quarters 11, we calculate COR(K) for all K between 1 and 32 quarters. We use the Schwarz s Bayesian information criterion to determine the number of lags for each VAR system. As a robustness check, we also used the Akaike information criterion, and results are not qualitatively different. However, we decided not to use this criterion given that the optimal number of lags is too large and we cannot afford to lose information with our relatively small sample. Using the estimated VAR we constructed bootstrapped confidence bands for COR(K) based on 25 replications of the system. Arguably and because of the way in which margins are measured in this paper, default or credit risk, the term structure of interest rates and monetary policy could be suggested as the main (and maybe the only) determinants of the cyclical behavior of margins. Therefore, 11 According to the NBER, from 1854 to 21 the U.S. economy has gone through 32 cycles with an average duration of 55 months, and in the Post World War II period, through 1 cycles with an average duration of 67 months. Cycles have shown a duration anywhere from 17 to 128 months (or 4 to 32 quarters). 13

15 in a second step we want to assess whether there is any explanatory power for the cyclicality of margins left to business cycles per se after accounting for these potential explanations. To do so we include additional endogenous variables in the X matrix. Subsection 3.1 discusses the bases for these additions. 3.1 Multivariate Specification for the Vector Auto-Regression Default or Credit Risk The net charge-off rate is used as a measure of the degree of default or credit risk in the economy. This rate is defined as loan charge-offs 12 net of loan recoveries as a percentage of total loans. Optimally chosen margins should be enough to cover the cost of increasing banks capital when risk exposure increases. Thus, an increase in the economy s default rate on loans should imply an increase in the margin charged by commercial banks. If, as expected, a higher credit risk is associated with periods of declining economic activity, risk is a very important candidate to explain the countercyclical behavior of margins. Moreover, credit risk could fully explain the cyclicality of margins. If so, no explanatory power would be left to business cycles per se after accounting for the effect of risk. Alternative regressions were run with the delinquency rate 13 and the Baa-Treasury bond spread 14 used as measures of default risk. Results for this robustness check are presented in Appendix B. 12 Loan charge-offs are the value of loans removed from banks books and charged against loss reserves. 13 According to the Federal Reserve s definition, delinquent loans and leases are those past due thirty days or more and still accruing interest as well as those in non-accrual status. 14 This spread has been suggested as a useful indicator of the default risk prospects on private debt. 14

16 However, it is important to note that we do not expect the credit risk measure to fully explain the cyclical properties of margins, particularly for the case of margins 1, 2 and 3. Given the way in which we construct these margin measures, there is no reason to expect a significant effect for credit risk in this setup. Our price-cost margins use ex-post interest rates on loans, calculated using the actual income obtained by banks after accounting for bad loans. Actually, for these margin measures, one can expect a negative effect of risk on margins, given that an increase in the share of bad loans can imply a fall in the income measure used to compute ex-post margins The Term Structure of Interest Rates and Monetary Policy The cyclical behavior of price-cost margins can be explained in part by maturity mismatches between assets and liabilities in banks balance sheets. If bank assets are of longer maturity than their liabilities, when a recession drives short-term interest rates down, interest expenses for banks might fall by more than their interest income. Then, during economic downturns, ex-post margins, calculated as interest income minus interest expenses over assets, might increase as a result of this maturity mismatch. Including a measure of the term structure in this multivariate specification of the VARs is specially important given that we compute margins over the entire stock of assets and liabilities, rather than at the margin. Banks are maturity transformers and shocks to short-term interest rates may affect their margins if they are not able to adjust their returns on long-term assets. Therefore, we include the slope of the yield curve in matrix X to account for the effect of these maturity mismatches. The slope is measured here as the spread between the tenyear and the one-year rate, and it is an indication of what rate of return investors require 15

17 for taking the added risk of lending money during a longer period of time. The slope of the curve is highly countercyclical due to the fact that short-term rates increase by more than their long-term counterparts in economic upturns. Therefore, it could be argued that no explanatory power would be left to business cycles per se after including this additional regressor in the multivariate specification (i.e. it could be argued that the term structure is the only factor driving the cyclicality of banks price-cost margins). As suggested by Bernanke and Blinder (1992), the tilt of the term structure is also affected by monetary policy and therefore, including this additional regressor in the multivariate specification also allows us to control for the effects of monetary policy on margins 15. Bernanke and Blinder (1992) use the spread between the federal funds rate and the ten-year bond rate as a measure of monetary policy. The results for an alternative specification that uses the Bernanke and Blinder (1992) measure, instead of our spread between the ten-year and the one-year rate, are presented in Appendix B. It is worth noting that it is not our goal to identify exogenous shocks to the monetary policy stance and their effects on spreads, a task for which a structural specification would be needed. 15 The federal funds rate might seem a more obvious choice to control for the effects of monetary policy. However, this measure has the problem that one specific value for that rate might imply a loose policy when general market interest rates are high, as well as a tight policy when general market interest rates are low. 16

18 4 Results Table 2 shows the characteristics of the estimated VARs and the number of lags used in each case. We present the results in two steps. First, Figure 5 shows the results for a bivariate VAR where X t includes only a measure of margins and a business cycle indicator. Second, Figure 6 does it for a multivariate VAR where X t also includes a measure of credit risk and the slope of the yield curve as additional endogenous variables in the system. Both specifications include banking regulation and quarterly dummies, as well as a a quadratic time trend (see Table 2). The goal in the second step is to assess whether there is any countercyclicality of margins left after controlling for the effects of credit risk, maturity mismatches and monetary policy. The figures plot the correlation coefficients for various K forecast horizons and the 1% and 5% significance level confidence bands for the estimated correlation coefficients. The countercyclicality of price-cost margins is documented with negative (positive when the unemployment rate is used as a business cycle indicator) correlation coefficients COR(K) for most K and for all the specifications of the business cycle indicator and price-cost margins. In all cases the sign of the COR(K) coefficients shows countercyclicality. However, in some cases the bootstrapped confidence bands indicate that the estimated coefficients are not different from zero. It is still remarkable that most of the estimates are significant because as pointed out by Hamilton (1994), den Haan (2) and references therein, a large number of parameters is estimated in the VAR with not many restrictions imposed on the dynamics. This often results in estimates that are not significant at conventional 17

19 levels. Results for the second step indicate that margins are countercyclical even after controlling for the effects of credit risk, the term structure of interest rates and monetary policy (see Figure 6). This conclusion is robust to the inclusion of controls for banking regulation and seasonality in the data. It is also important to note that the pattern of significance of the coefficients is the same in both figures 5 and 6. Therefore, it is not the case that accounting for these additional effects makes estimations lose significance. We believe that showing how the countercyclicality survives to the inclusion of these controls is one of the important contributions of this paper. Furthermore, our results indicate that there is room for other explanations for the observed cyclical pattern in price-cost margins. As discussed in Section 3, it is not surprising that the inclusion of a measure of credit risk in the multivariate specification does not alter our results relative to the bivariate specification. Given the way in which margins are measured in this paper, they should not be positively related to credit risk. Moreover, it is reasonable to expect a negative effect of credit risk on our margins, as an increase in the share of bad loans can imply a fall in the interest income we use to compute margins. As a first robustness check, we used alternative measures for credit risk, namely, the delinquency rate and the Baa-Treasury bond spread. Our results are robust to these changes in specification of the multivariate VAR. The second robustness check we performed was to use an alternative measure for the term structure of interest rates, using the spread between the ten-year rate on government securities and the federal funds rate 16. Again, results are robust to this alternative definition. The reader is referred to Appendix B for this robustness analysis. 16 This is the measure suggested in Bernanke and Blinder (1992). 18

20 5 Concluding Remarks This study finds empirical evidence in support of the countercyclicality of banks price-cost margins for the United States banking sector, a key fact about US business fluctuations that has received very scarce attention so far. The results are robust to six definitions for the margins and to three different cycle indicators. We also find that margins are still countercyclical even after controlling for the effect of credit risk, monetary policy and the term structure of interest rates. Therefore, there is room for other explanations to the observed cyclical properties 17. Our results have interesting macroeconomic policy implications. With price-cost margins in the market for credit being countercyclical, a financial accelerator seems to be operating in the United States economy. This may provide additional grounds for stabilization policy in economies where these margins are more countercyclical. Finding potential explanations for the countercyclical behavior of margins documented in this study is left for future empirical work. Of particular interest is to assess whether countercyclical market power in banking plays any role in explaining the documented cyclical properties of margins. Our results also call for the development of general equilibrium models that can account for this cyclical behavior of margins. This theoretical framework should then be used to assess how countercyclical margins can provide a channel through which aggregate productivity shocks affect economic activity. 17 Bernanke and Gertler (1989) themselves argue that they use the principal-agent view of credit markets as one of the various possible ways of rationalizing theoretically the countercyclicality of external finance premia. 19

21 GDP and Total Loans Figure 1: Detrended Business Cycle Indicators Quarters Business cycle components of the logarithm of GDP (thin line) and total loans (thick line). The series were filtered using the Hodrick-Prescott filter. Vertical lines indicate beginning and end of NBER recessions. Source: RCI data, NIPA-BEA and NBER. 2

22 Figure 2: The Comovement Between Banks Price-Cost Margins and Output Margin 1 Margin Margin 3 Spread BP-TB Spread BP-Fed Funds Spread Survey-Fed Funds Business cycle components of the logarithm of GDP (thin line) and the margin measure indicated in the heading of each panel (thick line). The series were filtered using the Hodrick-Prescott filter. Vertical lines indicate beginning and end of NBER recessions. Source: RCI data, Board of Governors, BEA and NBER. 21

23 Figure 3: The Comovement Between Banks Price-Cost Margins and Unemployment Margin 1 Margin Margin 3 Spread BP-TB Spread BP-Fed Funds Spread Survey-Fed Funds Business cycle components of the unemployment rate (thin line) and the margin measure indicated in the heading of each panel (thick line). The series were filtered using the Hodrick-Prescott filter. Vertical lines indicate beginning and end of NBER recessions. 22 Source: RCI data, Board of Governors, BEA and NBER.

24 Figure 4: The Comovement Between Banks Price-Cost Margins and Lending Margin 1 Margin Margin 3 Spread BP-TB Spread BP-Fed Funds Spread Survey-Fed Funds Business cycle components of the logarithm of loans (thin line) and the margin measure indicated in the heading of each panel (thick line). The series were filtered using the Hodrick-Prescott filter. Vertical lines indicate beginning and end of NBER recessions. Source: RCI data, Board of Governors and NBER. 23

25 Table 1: Sample Correlations of Margins with Business Cycle Measures GDP Unemployment Rate Total Loans Margin (.269) (.934) (.257) Margin (.29) (.) (.) Margin () (.) (.) Spread BP-TB (.586) (31) (.126) Spread BP-FF (.3) () (.2) Spread survey-ff (.66) (.22) (1) p-values shown in parentheses. See Table A.1 for margin definitions. GDP: Real Gross Domestic Product. The series were filtered using the Hodrick-Prescott filter. 24

26 Table 2: Characteristics of the Estimated VARs Business cycle indicator Margin No. lags No. lags bivariate multivariate GDP (in logs) Urate Loans (in logs) GDP (in logs) Urate Loans (in logs) GDP (in logs) Urate Loans (in logs) GDP (in logs) Spr BP-TB 2 2 Urate Spr BP-TB 2 2 Loans (in logs) Spr BP-TB 3 2 GDP (in logs) Spr BP-FF 2 2 Urate Spr BP-FF 3 2 Loans (in logs) Spr BP-FF 3 2 GDP (in logs) Spr surv.-ff 3 2 Urate Spr surv.-ff 2 2 Loans (in logs) Spr surv.-ff 2 2 X t in the bivariate VAR specification = [business cycle indicator, margin]. Specification includes four endogenous variables, X t in the multivariate VAR specification = [business cycle indicator, margin, credit risk measure, slope of the yield curve]. Two banking regulation dummies, quarterly dummies, and linear and quadratic time trend terms are included as exogenous controls in all specifications. Number of lags based on Schwarz Bayesian information criterion. 25

27 Figure 5: Bivariate VARs: Correlations of Forecast Errors A- Business cycle indicator: GDP Margin 1 Margin Margin 3 Spread BP-TB Spread BP-FF Spread Surv.-FF Quarters 26

28 B- Business cycle indicator: Unemployment Rate Margin 1 Margin Margin 3 Spread BP-TB Spread BP-FF Spread Surv.-FF Quarters 27

29 C- Business cycle indicator: Bank Loans Margin 1 Margin Margin 3 Spread BP-TB Spread BP-FF Spread Surv.-FF Quarters The x-axis measures the forecast horizon in quarters. Estimated correlation coefficients (thick line). 5%, 95% (thin lines), 1% and 9% (dashed line) bootstrapped confidence bands using a one-sided test. 28

30 Figure 6: Multivariate VARs: Correlations of Forecast Errors A- Business cycle indicator: GDP Margin 1 Margin Margin 3 Spread BP-TB Spread BP-FF Spread Surv-FF Quarters 29

31 B- Business cycle indicator: Unemployment Rate Margin 1 Margin Margin 3 Spread BP-TB Spread BP-FF Spread Surv-FF Quarters 3

32 C- Business cycle indicator: Bank Loans Margin 1 Margin Margin 3 Spread BP-TB Spread BP-FF Spread Surv-FF Quarters The x-axis measures the forecast horizon in quarters. Estimated correlation coefficients (thick line). 5%, 95% (thin lines), 1% and 9% (dashed line) bootstrapped confidence bands using a one-sided test. 31

33 References [1] Angelini, P. and N. Cetorelli (23), The Effects of Regulatory Reform on Competition in the Banking Industry, Journal of Money, Credit and Banking, Vol. 35 (5), October 23, [2] Asea, P. and B. Blomberg (1998), Lending Cycles, Journal of Econometrics, Vol. 83, pp [3] Bernanke, B. and A. Blinder (1992), The Federal Funds Rate and the Channels of Monetary Transmission, American Economic Review, September, Vol. 82, pp [4] Bernanke, B. and M. Gertler (1989), Agency Costs, Net Worth and Business Fluctuations, American Economic Review, March, Vol. 79, pp [5] Bernanke, B., M. Gertler and S. Gilchrist (1996), The Financial Accelerator and the Flight to Quality, Review of Economics and Statistics, Vol. 78, pp [6] Bernanke, B., M. Gertler and S. Gilchrist (1998), The Financial Accelerator in a Quantitative Business Cycle Framework, NBER Working Paper No. 6455, March. [7] Bloch, H. and M. Olive (21), Pricing Over the Cycle, Review of Industrial Organization, Vol. 19, [8] Chen, Y., E. Higgins and J. Mason (25), Is Bank Efficiency Cyclical? The Relationship Between Economic and Financial Market Conditions and Bank Performance, Drexel University working paper, January. [9] Chevalier, J. and D. Scharfstein (1995), Liquidity Constraints and the Cyclical Behavior of Markups, The American Economic Review, Vol. 85(2), Papers and Proceedings of the Hundredth and Seventh Annual Meeting of the American Economic Association, Washington DC, January 6-8, May 1995, [1] Chevalier, J. and D. Scharfstein (1996), Capital-Market Imperfections and Countercyclical Markups: Theory and Evidence, The American Economic Review, Vol. 86(4), September 1996,

34 [11] Demirgüç-Kunt, A., Laeven, L. and Levine, R. (24), Regulations, Market Structure, Institutions, and the Cost of Financial Intermediation, Journal of Money, Credit, and Banking, June 24, Vol. 36, Iss. 3, pp [12] den Haan, W. (2), The Comovement Between Output and Prices, Journal of Monetary Economics, Vol. 46, pp [13] Dueker, M. and D. Thornton (1997), Do Bank Loan Rates Exhibit a Countercyclical Mark-up?, working paper series Federal Reserve Bank of St. Louis, A. [14] Edwards, S. and C. Vegh (1997), Banks and Macroeconomic Disturbances Under Predetermined Exchange Rates, Journal of Monetary Economics, Vol. 4. pp [15] Feinberg, R. (21), The Competitive Role of Credit Unions in Small Local Financial Services Markets, The Review of Economics and Statistics, Vol. 83(3), pp [16] Flannery, M. (1981), Market Interest Rates and Commercial Bank Profitability: An Empirical Investigation, Journal of Finance, December, pp [17] Galeotti, M. and F. Schiantarelli (1998), The Cyclicality of Markups in a Model with Adjustment Costs: Econometric Evidence for US Industry, Oxford Bulletin of Economics and Statistics, Vol. 6 (2), [18] Gali, J. (1994) Monopolistic Competition, Business Cycles, and the Composition of Aggregate Demand, Journal of Economic Theory, Vol. 63(1), [19] Gertler, M. and S. Gilchrist (1994), Monetary Policy, Business Cycles, and the Behavior of Small Manufacturing Firms, The Quarterly Journal of Economics, Vol. 19(2). pp [2] Hamilton, J. (1994), Time Series Analysis, Princeton University Press, Princeton, NJ. [21] Hannan. T. (1997), Market Share Inequality, the Number of Competitors and the HHI: An Examination of Bank Pricing, Review of Industrial Organization, 12, [22] Hannan, T. and A. Berger (1991), The Rigidity of Prices: Evidence from the Banking Industry, American Economic Review, volume 81(4), pp

35 [23] Ho, T., and A. Saunders (1981), The Determinants of Bank Interest Margins: Theory and Empirical Evidence, Journal of Financial and Quantitative Analyses, Vol. 16, [24] Kashyap, A. and J. Stein (1997), What Do a Million Banks Have to Say About the Transmission of Monetary Policy?, NBER working paper series, working paper 656, June [25] Lebow, D. (1992), Imperfect Competition and Business Cycles: An Empirical Investigation, Economic Inquiry, Vol. 3(1), January 1992, [26] Lerner, E. (1981), Discussion: The Determinants of Bank Interest Margins: Theory and Empirical Evidence, Journal of Financial and Quantitative Analysis, Vol. 16, pp [27] Lown, C., D. Morgan and S. Rohatgi (2), Listening to Loan Officers: The Impact of Commercial Credit Standards on Lending and Output, FRBNY Economic Policy Review, July, pp [28] Mester, L. and A. Saunders (1995), When Does the Prime Rate Change?, Journal of Banking & Finance, Vol. 19 (5), pp [29] Nabar, P., S. Park and A. Saunders (1993), Prime Rate Changes: Is There an Advantage in Being First?, Vol. 66(1), pp [3] Neumark, D. and S. Sharpe (1992), Market Structure and the Nature of Price Rigidity: Evidence from the Market for Consumer Deposits, Quarterly Journal of Economics, 17(2), pp [31] Oliner, S. and G. Rudebusch (1995), Is There a Bank Lending Channel for Monetary Policy?, Federal Reserve Bank of San Francisco Economic Review, 2, pp [32] Oliner, S. and G. Rudebusch (1996a), Monetary Policy and Credit Conditions: Evidence from the Composition of External Finance: Comment, American Economic Review, Vol. 86(1), pp [33] Oliner, S. and G. Rudebusch (1996b), Is There a Broad Credit Channel for Monetary Policy?, Federal Reserve Bank of San Francisco Economic Review, 1, pp

36 [34] Ravn, M., S. Schmitt-Grohé, M. Uribe (26), Deep Habits, Review of Economic Studies, Vol. 73, pp [35] Rotemberg, J. (1996), Prices, Output and Hours: An Empirical Analysis Based on a Sticky Price Model, Journal of Monetary Economics, Vol. 37, pp [36] Rotemberg, J. and G. Saloner (1986), A Supergame-Theoretic Model of Price Wars During Booms, The American Economic Review, Vol. 76 (3), June 1986, [37] Rotemberg, J. and M. Woodford (1991), Markups and the Business Cycle, in NBER Macroeconomics Annual, Vol. 6, O. Blanchard and S. Fischer eds., Cambridge, MA, MIT Press. [38] Rotemberg, J. and M. Woodford (1992), Oligopolistic Pricing and the Effects of Aggregate Demand on Economic Activity, Journal of Political Economy, 1, [39] Rotemberg, J. and M. Woodford (1996), Real Business Cycle Models and the Forecastable Movements in Output, Hours and Consumption, American Economic Review, Vol. 86(1), March, pp [4] Santos, J. and A. Winton, Bank Loans, Bonds and Information Monopolies Across the Business Cycle, working paper, December 26. [41] Saunders, A. and L. Schumacher (2), The Determinants of Bank Interest Rate Margins: an International Study, Journal of International Money and Finance, Vol. 19 (6), [42] Seppala, J. and L. Xie (25), The Cyclical Properties of the Term Structure of Interest Rates, unpublished manuscript. [43] Weinberg, J. (1995), Cycles in Lending Standards?, Federal Reserve Bank of Richmond Economic Quarterly, Vol. 81/3, Summer 1995, pp [44] Wong, K. (1997), On the Determinants of Bank Interest Margins Under Credit and Interest Rate Risks, Journal of Banking and Finance, Vol. 21, pp

37 Appendix A: Data Time series were constructed taking into account the Notes on forming consistent time series. These are provided with the Call Reports on Condition and Income data in the Federal Reserve Bank of Chicago web site and based on Kashyap and Stein (1997). In addition, the data were cleaned to avoid the results from being affected by outliers and other obvious data problems. First, observations for which total assets or total loans are zero or missing were deleted. Second, banks in US territories were dropped from the database. Since there are very few banks in each territory, concentration measures are significantly higher there than in the continental US, which might have a significant impact on banks profit margins. Third, banks interest income, expenses and chargeoffs and recoveries are all measured as cumulative year to date totals. Therefore, the appropriate adjustment was made to get the corresponding values for each quarter. Thus, banks for which there is no data in at least one of the four quarters in a given year were not included in the computation of the margin and of the net charge-off rate in that year. Finally, net interest margins are based on individual bank-level data as described in Table A.1. Margin measures were obtained by computing the weighted average over the banks, with the weights given by each bank s share in total loans. Since a few very significant outliers were detected for the margin measures, only margins falling into the interval defined by the [2nd-99th] percentiles were used to compute the average. 36