Banks Exposure to Interest Rate Risk and The Transmission of Monetary Policy:

Banks Exposure to Interest Rate Risk and The Transmission of Monetary Policy: Augustin Landier David Sraer David Thesmar May 22, 2012 Abstract This paper offers a new test of the lending channel view, based on banks exposures to interest rate risk, or duration gap. While banks have, on average, positive levels of duration gap their assets have on average longer duration than their liabilities there is a substantial heterogeneity in the cross-section of banks in how exposed they are to interest rate risk. In a first step, we show that the sensitivity of bank profits to interest rates increases significantly with their duration gap, even when banks use interest rate derivatives. In a second step, we show that the duration gap also predicts the sensitivity of bank lending to interest rates, both for commercial & industrial loans and mortgages. Quantitatively, a 100 basis point increase in the Fed funds rate would lead a bank at the 75 th percentile of the duration gap to increase its lending by about.25 percentage points relative to a bank at the 25 th percentile of the duration gap. We conclude that banks exposure to interest rate risk is an important determinant of the lending channel. Toulouse School of Economics (e-mail: augustin.landier@tse-fr.eu) Princeton University, NBER and CEPR (e-mail: dsraer@princeton.edu) HEC and CEPR (e-mail: thesmar@hec.fr) 1

1. Introduction Understanding the channels through which monetary policy affects the economy has long been a key research topic in both macroeconomics and economic policy analysis. Two views on monetary policy transmission can be opposed. According to the money view, a decrease in reserves leads to a decrease in the issuance of demand deposits, effectively reducing households money holding and leading to an increase in the equilibrium real interest rate. According to the lending view, the reduction in reserve directly affects banks ability to supply credit to firms in the economy, thereby reducing investment and economic activity. In support of the lending view, Kashyap and Stein (2000) have argued that in the cross-section of banks, liquidity, measured by the ratio of securities to assets, plays an important role in how banks credit supply respond to monetary tightening. Similarly, Campello (2002) documents that internal capital markets in financial conglomerates lessen the impact of Fed policies on bank lending activity. In this paper, we point to another important cross-sectional determinant of the efficacy of the transmission of monetary policy: banks exposures to interest rate risk. Precisely, using bank holding company (BHC) data available quarterly from 1986 to 2011 we define the duration gap every quarter as the difference between the dollar amount of the banks assets that re-price within a year and the dollar amount of liabilities that re-price within a year, normalized by the banks total assets. The first contribution of the paper is to document that there is substantial variations in this duration gap, both in the time-series and in the cross-section. Banks typically have positive duration gap, which means that their assets have longer duration than their liabilities. However, in the crosssection, some banks appear to have a much larger exposure to interest rate risk than others: the bank at the 25 th percentile of the duration gap is almost matched its duration gap is close to 0 while the bank at the 75 th percentile of the duration gap will see 25 percent of its assets re-price in excess of its liabilities. In the aggregate, the average duration gap goes from 5% in 2009 to as much as 22% in 1993. The second contribution of this paper, also descriptive, is to show that the duration gap is in fact a meaningful concept. Hedging policies by banks, which we only imperfectly observe in the call reports, could be set so that the actual duration gap, as opposed to the accounting duration gap as defined above, is on average close to zero. This is clearly not the case in the data. We provide evidence that on average, the sensitivity of profits to interest rates increases with the duration 2

gap. For instance, a 100 basis point increase in the fed funds rate would lead a bank at the 75 th percentile of the duration gap to increase its earning by about.2% of equity relative to a bank at the 25 th percentile of the duration gap. Thus, while some banks might be hedging their exposure to interest rate risk through derivatives trading, our regression analysis leads us to conclude they only imperfectly do so. 1 In particular, hedging in the data seems to occur mostly through the active management of the duration gap by banks in the anticipation of future movements in monetary policy: in our sample, changes in the aggregate duration gap have significant forecasting power over long run future changes in the fed fund rates. However, while this holds at the aggregate level, there is again substantial variations in the cross-section in banks exposure to interest-rate risk, offering us the possibility to identify how monetary policy differentially affects those firms with a large or a small, potentially negative duration gap. This is precisely our third contribution. Using a standard two-step procedure, we show that the duration gap predicts significantly the sensitivity of bank lending to interest rates. Quantitatively, a 100 basis point increase in the fed funds rate would lead a bank at the 75 th percentile of the duration gap to increase its lending by about.25 percentage point relative to a bank at the 25 th percentile of the duration gap. These effects are robust to the estimation method (pooled regressions vs. twosteps procedure), to the inclusion of controls in the cross-sectional regressions (size and liquidity) and to the decomposition of total lending into commercial & industrial lending and mortgage lending. In the cross-section of banks, these effects are larger for smaller banks. Similarly, the role the duration gap plays on the sensitivity of banks lending to interest rates seems to be more pronounced for banks that report no hedging on their balance sheet. This last result, however, somewhat depends on the specification we use. Overall, our results suggest that the duration gap significantly affects the lending channel, and therefore establish the importance of this mode of transmission of monetary policy. Our paper is directly related to the economic literature that evaluates empirically the importance of the lending channel. The use of aggregate data to investigate how a bank loan supply shock affects the aggregate economy has mostly led to mixed results (e.g., Bernanke (1983), Peek and Rosengren (2000), Ashcraft (2006)). This might well be due to the fundamental identification challenge imposed by the use of aggregate data, i.e. one needs to separate credit supply from credit demand. To address 1 Consistently with these findings, Begeneau et al. have documented that the four largest banks in the US use the derivative market to increase their exposure to interest rate risk. 3

this concern, Kashyap and Stein (2000) and Campello (2002) exploit variations in banks access to liquid assets or to internal capital market to better identify the lending view. Our paper is another attempt in that direction, which focuses on the duration gap, i.e. the exposure of banks to interest rate risk. Finally, this study relates to the surprisingly small literature on interest rate risk in banking and corporate finance (Begeneau, Piazzesi, Schneider, 2012, Chava and Purnanandam, 2007, and Purnanandam, 2007). The rest of the paper is organized as follows. Section 2 presents the data. Section?? shows the relationship between banks duration gap and the sensitivity of their profits to variations in interest rates. Section 4 analyzes the role of the duration gap on the elasticity of banks lending policy to interest rates. Section 6 concludes. 2. Descriptive statistics and Data 2.1. Duration Gap We measure the net exposure of banks to interest rate risk by using schedule HC-H from the quarterly Consolidated Financial Statements for Bank Holding Companies (FR Y-9C). These reports have to be filed with the FED by all US bank holding companies with total consolidated assets of $500 million or more. Schedule HC-H is specifically dedicated to interest sensitivity of the balance sheet. It reports variables that are determinants of the banks overall exposure to short-term interest rate risk. On the asset side, there is a unique variable (bhck3197), equal to Assets that reprice or mature within one year. The overall exposure to interest risk of the liability side is composed of four elements: Long-term debt that reprices within one year (bhck3298); Long-term debt that matures within one year (bhck3409); Variable-rate preferred stock (bhck3408); and Interest-bearing deposit liabilities that reprice or mature within one year (bhck3296), such as certificates of deposits. Empirically, the latter is by far the most important determinant of the liability-side sensitivity to interest rates. We scale all these variables by assets, and report summary statistics in Table 1. We define a variable called MATCH that is equal to the difference between interest-sensitive assets and interest sensitive liabilities, as a fraction of total assets. This variable measures the banks net exposure to short-term interest risk. Summary statistics are based on the universe of all US bank holding companies with total consolidated assets of $1Bil or more reporting quarterly Consolidated Financial Statements (Files FR Y-9C) between 1986 and 2010. 4

2.2. The Use of Derivatives to Hedge Interest Risk Banks can neutralize some of their net exposure to interest risk by using interest derivatives. Starting in 2005, schedule HC-L from the quarterly Consolidated Financial Statements for Bank Holding Companies reports the notional amounts in interest derivatives contracted by banks. Five kinds of derivative contracts are separately reported: Futures (bhck8693), Forwards (bhck8697), Written options that are exchange traded (bhck8701), Purchased options that are exchange traded (bhck8705), Written options traded over the counter (bhck8709), Purchased options traded over the counter (bhck8713), and Swaps (bhck3450). We scale all these variables by assets, and report summary statistics in Table 2. Swaps turn out to be overwhelmingly the prevalent form of hedge used by banks. However, knowing the notional amounts of derivatives held by each bank does not allow us to compute the impact on balance sheets interest sensitivity. Thus we just consider whether a bank has some interest hedges or not by constructing a dummy variable (at the quarter-bank level) called HEDGED that is equal to one if a bank has a positive notional amount of interest derivatives of at least one of the seven kinds of derivatives mentioned above. 63% of the banks in our sample have a positive amount of interest hedges. This fraction is monotonic in bank size, varying from 38% for the lowest decile to 99% for the highest decile. 3. Interest Rate Risk and Profits In this Section, we check that the sensitivity of profits to interest rate movements depends on our measure of duration gap. This Section serves as a validation of our measure of duration gap, but also shows that hedging, although present in the data, is limited. This suggests that banks may seek to adjust their interest rate exposure so as to benefit from variations in interest rate. Consistently with this, we propose evidence that banks seem to time their exposure. 3.1. Duration Gap and Profits We first check that the sensitivity of profits to interest rate movements depends on duration gap. We use the same specification as in Kashyap and Stein (2000). We first run the following regression: k=4 P rofit it = α k.gap it 1. fedfunds t k + controls it 1 + ɛ it (1) k=0 5

for bank i in quarter t. k=4 k=0 α k is the cumulative effect of interest rate changes, given interest rate exposure on bank i s balance sheet. Banks with large gap it 1 have more interest rate sensitive assets than interest rate sensitive liabilities. As a result, cash flows should increase when interest rate go up. So, as long as banks do not hedge interest risk too much, we expect k=4 k=0 α k > 0. The list of controls is the following: calendar time dummies, four lags of P rofit it and four lags of fedfunds t. We also control for the fact that large banks, or well capitalized banks, may react differentially to changes in interest rates. So we also control for log(assets it 1 ) and equity it 1 /assets it 1, as well as interactions with all four lags of fedfunds t. All in all, we are measuring the impact of duration gap on profits beyond the effects of sheer size or leverage. We report the results in Table 3. Columns 1-5 use (quarterly) change in interest income normalized by lagged assets, as the dependent variable. Column 6-10 uses change in earnings normalized by lagged book equity. Columns 1 and 6 report regression results on the total sample. The bottom panel reports the cumulative impact of an interest rate increase, k=4 k=0 α k and the p-value of the F-test of statistical significance. Both interest income and total earnings exhibit a similar, strongly significant, sensitivity to interest rate and duration gap. For instance, the cumulative effect on earnings is 0.0084. This means that, if we compare a bank at the 25 th percentile of gap (approximately 0) and a bank at the 75 th percentile of gap (approximately 0.25), and if the economy experiences an increase by 100bp in fed funds rate, earnings in the latter bank will increase by about.2% of equity more than earnings of the former. Columns 2-3 and 7-8 split the sample into large and small banks. Large banks correspond to the 100 largest BHC each date, small banks are the rest. The size of the impact of gap-induced cash flows is the same for both categories of banks. Finally, columns 4-5 and 9-10 split the sample into banks that have some notional exposure on interest rate derivatives and banks that report zero notional exposure. This sample split reduces the period of estimation to 1995-2011. It is important to note that notional derivative exposure may not mean that the bank is hedging its maturity mismatch. For instance, derivatives may be used to hedge exposure resulting from fixed income trading positions, or to only partially offset the gap. Consistently with the idea that notional exposure captures hedging behavior imperfectly, we find strong and statistically significant effects for both categories of banks. The impact of duration gap on interest income is slightly smaller for banks with some derivative exposure, but the effect is not statistically significant. In non-reported regressions, we further restrict the sample to BHCs whose notional interest rate derivative exposure exceeds 10% of total assets (some 4,000 observations): 6

even for this smaller sample, the duration gap effect remains strongly significant and has the same order of magnitude. 3.2. Duration Gap Management and Market Timing The above results suggest that most banks remain exposed to interest rate risk, that they do not hedge fully. One possibility is that this exposure is managed actively by banks. If this is the case, and if banks can predict future monetary policy, we expect BHCs to adjust their duration gap in expectation of future interest rate movements. Figure 8 suggests that banks are able to forecast future monetary policy. Duration gap systematically increases about 2-3 years before fed fund rates increase. The average bank thus starts increasing its asset exposure to variable rates a few years before monetary policy tightens, so as to boost their profits. This is consistent with banks forecasting future monetary policy and timing the market. In Table 4, we test this econometrically by regressing the average duration gap on future fed fund rates, using quarterly data. Since both times series have a unit root (as is apparent from the Figure - DF test not reported), we use first differences. We also adjust the standard error estimate with the Newey-West methodology. Consistently with Figure 8, we find that interest rate exposure of banks predicts fed funds rates 8 to 16 quarters ahead. The t-stats are large, from 2.8 to 3.9, considering the small number of observations. At the 3-4 year horizon, the R 2 of the regression is about 20%. Interest rates are not a random walk: Current long-term rates are informative about changes to future short-term rates. A natural question is thus whether the aggregate duration gap is predicting changes in interest rates beyond the expected changes to those rates embedded in the yield curve. To answer this question, we construct for each quarter t the expected one year treasury rate expected to prevail in one year (quarter t+4) and two years (quarter t+8). We note these rates E t R t+i,t+i+4, i = 4, 8. We construct these anticipated rates from Treasury rates extracted from the Federal Reserve H15 report: We retrieve the constant maturities treasury rates at each time t for maturities equal to one and two years (i = 12, 24), and express it as a quarterly return: r t,i. We then define for i = 4, 8: E t R t+i,t+i+4 = (1 + r t,i+4) i+4 (1 + r t,i ) i 1 The unexpected shock (as of expectations at time t) to R t+i,t+i+4 is given by the gap between 7

the realization of the treasury rate and its anticipation at time t: (R E t R) t+i,t+i+4 In Table 5, we test this econometrically by regressing future expected and realized yearly Treasury rates, one year and two years ahead (time t+4 and t+8), on the average duration gap at time t. We also regress the corresponding forecast error ((R E t R) t+i,t+i+4 ) on the average duration gap. As in the previous table, we use first differences and adjust the standard error estimate with the Newey-West methodology. Consistently with the previous table, we find that interest rate exposure of banks predicts fed funds rates 8 to 12 quarters ahead. This predictability applies to the expected as well as the unexpected component of future rates, albeit slightly weakly. 4. Duration Gap, Interest Rate Risk and Lending We have established that interest rate movements affect profits when the duration gap is larger. A natural next step is to ask whether these shocks to bank earnings affect their lending policy. If banks are financially constrained, a decrease in cash flow should negatively affect lending, even in positive NPV loans. We take the exact same econometric specification as Kashyap and Stein (2000), and run the following regression: k=4 Loans it = α k.gap it 1. fedfunds t k + controls it 1 + ɛ it (2) k=0 for bank i in quarter t. As before, k=4 k=0 α k is the cumulative effect of interest rate changes, given interest rate exposure on bank i s balance sheet. Banks with large gap it 1 have earnings that decrease less when interest interest rate go up. Thus, they should scale down their lending less / lend more. As before, long as banks do not hedge interest risk too much, we expect k=4 k=0 α k > 0. As controls, like before, we use calendar time dummies, four lags of P rofit it and four lags of fedfunds t. We also control for the fact that large banks, or well capitalized banks, may react differentially to changes in interest rates (Kashyap and Stein, 1995). So we also control for log(assets it 1 ) and equity it 1 /assets it 1, as well as interactions with all four lags of fedfunds t. Hence, again, we are measuring the impact of duration gap on profits beyond the effects of sheer size or leverage. We report the results in Table 6, we report the results of this regression, separately for C&I loan 8

growth (columns 1-5) and for total lending growth (columns 6-10). As before, we run regressions on the whole sample (columns 1 and 6), split the sample into large and small banks, and split the sample between banks with some interest rate derivatives and banks without. Focusing on total lending growth, we find results that are statistically significant at 1%, except for large banks. The size of the effects is significant. If we compare a bank at the 25 th percentile of gap (approximately 0) and a bank at the 75 th percentile of gap (approximately 0.25), and if the economy experiences an increase by 100bp in fed funds rate, total loans in the latter bank will grow by about.4 percentage points more than the former. This has to be compared with a sample average loan growth of about 1.8%. Effects are similar when we restrict attention to C&I loans, except that they are not significant any more for banks with some derivative exposure. They remain significant for banks with no derivative, small banks, and on the overall sample. It is important to note that these regressions control for the fact that large, or less well capitalized, banks have other reasons to be less sensitive to increases in interest rate. In the third and fourth rows of the bottom panel, we report the sum of the coefficients on log(assets it 1 ). fedfunds t k and their statistical significance. Consistently with intuition, large banks decrease their lending less when the Fed raises the fed funds rates (the coefficient is positive). On C&I loans, the estimated effect is statistically significant (Kashyap and Stein, 1995, report a similar result on commercial bank data over the 1976-1993); but on total loans we find no significant impact (it even goes in the wrong direction). The impact of size on the reaction to monetary policy has the same order of magnitude as the duration gap. If we compare banks at the 25 th and banks at the 75 th percentile of the size distribution (log of assets equal to 14.2 vs 15.9), and consider a 100bp increase in fed funds rates, the smaller bank will reduce its C&I lending by 0.3% more. So duration gap explains movements C&I lending as well as bank size, and it explains movements of total lending better. Turning to the role of capitalization, rows 5 and 6 in the bottom panel of Table 6 report the sum of the coefficients on equityit 1 assets it 1. fedfunds t k. Estimates are insignificant and go in the wrong direction: better capitalized banks tend to reduce their lending more when interest rates increase. This counterintuitive result does not come from the fact that equity is correlated with size (negatively) or with duration gap (positively): in unreported regressions, we have tried specifications including the interactions term with equity only, and the coefficient remained negative. In Table 7, we include a third control: the effect of liquid assets. Kashyap and Stein (2000) have found evidence that increases in interest rates have less of an impact on lending when banks have 9

enough liquidity to buffer the adverse effect of interest rates on reservable liabilities and on their balance sheets. We include this control separately because it greatly reduces our sample size: the liquidity to asset ratio is only available starting in 1993. In spite of this dramatic drop in power, our results resist well: they remain statistically significant at 5% in for all banks, small banks, and banks without derivatives, for both C&I and total loans (and statistically significant at 1% in 5 out of 6 specifications). Liquidity does not, however, come in significant in these regressions, and has the wrong sign: banks with more liquid assets tend to reduce their lending more when interest rates increase. The discrepancy with Kashyap and Stein s results comes from the fact that we are using BHC data, instead of commercial bank data. Our main reason for doing so is that the aggregated duration gap is not consistently provided in commercial bank data: the variable change, and has to be constructed manually. 5. Increases vs decreases in interest rates Is the duration gap concentrated around rate increases or rate decreases? Table 8 re-runs the specification in equation 2 splitting dates between moments of fed funds increases and decreases. We find that all of the action is concentrated around moments when interest rates decrease, i.e. it is banks with variable rate liabilities that tend to take advantage of accomodating monetary policy to lend more. 5.1. Robustness In Appendix A, we provide estimates using an alternative approach also used in the literature (Kashyap and Stein, 2000, Campello, 2002). We proceed in two steps. First, we run, separately for each quarter, the following regression: X it = γ t gap it 1 + controls it + ɛ it (3) where X it is a cash flow or lending LHS variable.controls it include: X it 1,..., X it 4, log(assets it 1 ), equity it 1 assets it 1. From this first step, we obtain a time-series of X to gap sensitivity γ t. In our second step, we regress γ t on change in fed funds rate and four lags of it, as well as four quarter dummies: 10

k=4 γ t = α k. fedfunds t k + quarterdummies t + ɛ it (4) k=0 Again, we expect that k=4 k=0 α k > 0: in periods where interest rates increase, high duration gap firms tend to make more profits, or lend more. We report the results using the new methodology in Tables 9, 10 and 11. Results are a little bit weaker using this approach, but have the same order of magnitude. Results on profits and cash flows are still all significant at the 1% level of significance, and have the same order of magnitude. Results on lending, controlling for size and leverage, but not for liquidity, remain significant at the 1 or 5% level for total lending growth. They become a bit weaker, albeit still significant at the 5% level for the whole sample, for C&I loans. Controlling for liquidity reduces the sample to 1994-2011 (BHC data do not report liquidity holdings before 1994), so it reduces the sample size by a third. Significance weakens, but duration gap effects on total lending remains statistically significant at the 5% level for the whole sample and small firms, as well as firms with some interest rate derivative exposure. This alternative estimation procedure provides estimates with very similar orders of magnitude. 6. Conclusion [To be completed] 11

7. References Ashcraft (2006), New Evidence on the Lending Channel, Journal of Money, Credit and Banking Bernanke (1983), Nonmonetary Effects of the Financial Crisis in Propagation of the Great Depression, American Economic Review Begeneau, Piazzesi, Schneider, 2012, The Allocation of Interest Rate Risk in the Financial Sector, WP Campello, 2002, Internal Capital Markets in Financial Conglomerates: Evidence From Small Bank Responses to Monetary Policy, Journal of Finance Chava and Purnanandam, 2007, Determinants of the floating-to-fixed rate debt structure of firms, Journal of Financial Economics Kashyap and Stein, 1995, The Impact of Monetary Policy on Bank Balance Sheets, Carnegie- Rochester Conferences on Public Policy Kashyap and Stein, 2000, What Do A Million Observations on Banks Say About the Transmission of Monetary Policy?, American Economic Review Peek and Rosengren, 2000. Collateral Damage: Effects of the Japanese Bank Crisis on Real Activity in the United States, American Economic Review Purnanandam, 2007, Interest Rate Derivatives at Commercial Banks: An Empirical Investigation, Journal of Monetary Economics 12

8. Figures Figure 1: Aggregate Duration Gap and Fed Fund Rates Fed Fund Rates 0 2 4 6 8 10 1985 1990 1995 2000 2005 2010 Date.05.1.15.2.25 Average Duration Gap Fed Fund Rates Average Duration Gap 13

9. Tables Table 1: Summary Statistics: Balance Sheet Sensitivity to Interest Rates mean sd p25 p75 count Gap it 1 0.134 0.186 0.016 0.252 35362 Interest Sensitive Liabilities 0.302 0.150 0.201 0.383 35362 Interest Sensitive Assets 0.437 0.149 0.343 0.532 35642 Short Term Liabilities 0.291 0.151 0.189 0.372 35638 Variable Rate Long Term Debt 0.010 0.025 0.000 0.009 35515 Short Maturity Long Term Debt it 1 0.001 0.006 0.000 0.000 35490 Preferred Stock it 1 0.000 0.002 0.000 0.000 35378 Summary statistics are based on the quarterly Consolidated Financial Statements (Files FR Y-9C) between 1986 and 2010 restricted to US bank holding companies with total consolidated assets of $1Bil or more. The variables are all scaled by total consolidated assets (bhck2170) and are defined as follows: Interest Sensitive Liabilities =(bhck3296+bhck3298+bhck3409+bhck3408)/bhck2170; Interest Sensitive Assets=(bhck3197)/bhck2170; Short Term Liabilities=bhck3296/bhck2170; Variable Rate Long Term Debt=bhck3298/bhck2170; Short Maturity Long Term Debt=bhck3409/bhck2170; Prefered Stock=bhck3408/bhck2170 Table 2: Summary Statistics: Derivatives Hedges of Interest Rate Risk mean sd p25 p75 count Futures 0.023 0.154 0.000 0.000 35547 Forward Contracts 0.037 0.263 0.000 0.003 35563 Written Options (Exchange Traded) 0.009 0.076 0.000 0.000 35531 Purchased Options (Exchange Traded) 0.012 0.116 0.000 0.000 35529 Written Options (OTC) 0.029 0.187 0.000 0.003 35557 Purchased Options (OTC) 0.030 0.183 0.000 0.000 35582 Swaps 0.196 1.557 0.000 0.043 46070 HEDGED 0.615 0.487 0.000 1.000 35526 Summary statistics are based on Schedule HC-L of the quarterly Consolidated Financial Statements (Files FR Y-9C) between 2005 and 2010 restricted to US bank holding companies with total consolidated assets of $1Bil or more. Schedule HC-L is not available prior to 2005. The variables report notional amounts in each kind of derivatives at the bank holding-quarter level and are all scaled by total consolidated assets (bhck2170). Variables are defined as follows: Futures contracts = bhck8693/bhck2170; Forward contracts = bhck8697/bhck2170; Written options (exchange traded) = bhck8701/bhck2170; Purchased options (exchange traded) = bhck8705/bhck2170; written options (OTC) = bhck8709/bhck2170; Purchased options (OTC) = bhck8713/bhck2170; Swaps=bhck3450/bhck2170. HEDGED is a dummy equal to one if a bank has a positive notional amount in any of the seven types of interest hedging derivatives in a given quarter. 14

Table 3: Interest Rate Shocks and Cash-Flows interestit ROEit All Small Big No Hedge Some Hedge All Small Big No Hedge Some Hedge Gapit 1 fedfundst.0039***.0042***.0023.0055***.0032**.0053***.0054***.0069***.005**.0071*** (4.4) (4.1) (1.3) (3.8) (2.4) (4.3) (3.9) (2.7) (2.4) (3.4) Gapit 1 fedfundst 1.0034***.0033***.0044***.0035***.0037***.0025*.0032** -.00082.0045*.0024 (4) (3.3) (2.8) (2.7) (2.7) (1.7) (2) (-.26) (1.9) (.98) Gapit 1 fedfundst 2 -.00028 -.00049.0007 -.00079 -.00042.0006.00024.00099 -.00054.0012 (-.38) (-.57) (.46) (-.61) (-.4) (.4) (.13) (.39) (-.2) (.47) Gapit 1 fedfundst 3.00091.001 -.00018 -.00059.0005.00065.0007.0018.0028 -.00067 (1.1) (1.1) (-.12) (-.42) (.45) (.46) (.43) (.58) (1.1) (-.32) Gapit 1 fedfundst 4 -.00058 -.00076.00023.0012 -.0013 -.00072 -.0015.0018 -.0015 -.00027 (-.86) (-.94) (.2) (1.1) (-1.4) (-.63) (-1.1) (.8) (-.7) (-.16) N 27971 20596 7375 8160 12512 25798 18976 6822 7605 11518 ar2.11.12.078.12.087.22.22.22.23.22 Sum of coefficients.0074.0072.0074.0089.0057.0084.008.011.01.0097 p-value 0 0 0 0 0 0 0 0 0 0 All regressions include calendar time dummies, 4 lags of the depend variable, as well as gapit 1 trend, log(assetsit 1) and log(assetsit 1) fedfundst k for all k = 0, 1, 2, 3, 4. 15

Table 4: Aggregate Duration Gap and the Fed Fund rates fedfunds t+1 fedfunds t+4 fedfunds t+8 fedfunds t+12 fedfunds t+16 Gap t 1 1.5 11 82*** 135*** 142*** (.33) (.69) (2.8) (3.5) (3.9) R 2.00085.0041.095.2.23 Observations 96 93 89 85 81 Forecasting regressions of changes in the Fed Fund rates at the 1 quarter, 1 year, 2 years, 3 years and 4 years horizon on past quarter changes in the aggregate duration gap. The aggregate duration gap is defined as the equal-weighted average of gap it 1. Robust standard errors are presented in column 1. Newey-West standard errors allowing for 3 (resp. 7, 11 and 15) lags are presented in column 2 (resp. 3, 4 and 5) Table 5: Aggregate Duration Gap and Unexpected vs. Expected Shocks to One Year Treasuries Realizations Expectations Unexpected Shock R t+4,t+8 R t+8,t+12 E tr t+4,t+8 E tr t+8,t+12 (R E tr) t+4,t+8 (R E tr) t+8,t+12 Gap t 1 23 85*** 20** 36*** 2.5 49** (1.4) (3.2) (2.6) (4) (.18) (2.4) R 2.018.12.05.11.00033.061 Observations 97 94 97 97 97 94 Forecasting regressions of changes in realized and expected one year Treasuries on past quarter changes in the aggregate duration gap. R t+i,t+i+4 is the one year treasury rate prevailing at quarter t + i. ER t+i,t+i+4 is the expected one year treasury rate expected to prevail in i quarters from t based on the yield curve at time t, for i = 4, 8. Changes ( ) are vis-a-vis quarter t 1. The aggregate duration gap is defined as the equal-weighted average of gap it 1. Robust standard errors are presented in column 1. Newey-West standard errors allowing for 7 (resp. 11) lags are presented in column 1,3,5 (resp. 2,4,6). 16

Table 6: Interest Rate Shocks and Lending: Size and equity ratio interacted control C&I lending growth Total lending growth All Small Big No Hedge Some Hedge All Small Big No Hedge Some Hedge Gapit 1 fedfundst.0078.007.00097.011 -.0016 -.0057 -.0074 -.0012 -.0035.00016 (.95) (.73) (.067) (.77) (-.13) (-1.4) (-1.6) (-.13) (-.52) (.025) Gapit 1 fedfundst 1 -.0042 -.0082.014 -.005.0021.011**.013***.0029.015**.0027 (-.44) (-.72) (.95) (-.32) (.16) (2.5) (2.8) (.26) (2.1) (.37) Gapit 1 fedfundst 2.021**.028*** -.01.031*.017.0058.0085* -.0054.0076.001 (2.5) (2.9) (-.64) (1.9) (1.4) (1.3) (1.7) (-.59) (1.2) (.15) Gapit 1 fedfundst 3.005.0077 -.00076 -.0047.013.004.0031.0097.0041.011* (.56) (.75) (-.042) (-.3) (1.2) (.95) (.67) (.95) (.7) (1.7) Gapit 1 fedfundst 4 -.0088 -.0082 -.016 -.00082 -.026** -.0012 -.0044.0067 -.0081 -.00012 (-1.1) (-.94) (-.91) (-.056) (-2.3) (-.32) (-1.1) (.74) (-1.3) (-.02) N 28917 20927 7990 8226 12641 28835 21117 7718 8246 12497 ar2.097.089.15.076.11.2.23.14.23.19 Sum of coefficients gapit 1 fedfunds.021.027 -.011.031.0038.014.013.013.015.015 p-value.0044.0017.42.014.76 0.0047.14.019.0085 Sum of coefs Log(assetsit 1) fedfunds.0026.0027.0037.0043.0036 -.00028 -.00036 -.00031 -.0028.0008 p-value.0048.31.028.38.0043.55.79.78.19.2 Sum of coefs (equityit 1/assetsit 1) fedfunds -.075 -.058 -.21.033 -.17 -.054 -.044 -.078 -.037 -.13 p-value.28.45 0.72 0.18.31.3.19.005 All regressions include calendar time dummies, 4 lags of the depend variable, as well as gapit 1 trend, log(assetsit 1), log(assetsit 1) fedfundst k for all k = 0, 1, 2, 3, 4, equity it 1 and equity it 1 assets it 1 assets it 1 fedfundst k for all k = 0, 1, 2, 3, 4. 17

Table 7: Robustness: With KS, equity ratio, and size interacted controls C&I lending growth Total lending growth All Small Big No Hedge Some Hedge All Small Big No Hedge Some Hed Gapit 1 fedfundst.0092.0081.01.021 -.0049 -.001 -.0027.0042.0031.0023 (.96) (.74) (.53) (1.4) (-.38) (-.22) (-.56) (.34) (.47) (.35) Gapit 1 fedfundst 1 -.0061 -.0089.0037 -.014.0044.005.0081 -.0062.0097.0027 (-.6) (-.75) (.21) (-.91) (.32) (1) (1.5) (-.46) (1.4) (.38) Gapit 1 fedfundst 2.026**.031*** -.000071.034**.013.0058.01* -.012.0059 -.0011 (2.6) (2.7) (-.0033) (2) (1.1) (1.2) (1.8) (-1.1) (.94) (-.16) Gapit 1 fedfundst 3.0081.0087.0061 -.01.014.0052.0026.02* -.00052.01 (.89) (.85) (.26) (-.63) (1.3) (1.1) (.5) (1.7) (-.087) (1.6) Gapit 1 fedfundst 4 -.019** -.014 -.042* -.000031 -.028** -.0022 -.0055.0048 -.0021.00047 (-2.1) (-1.4) (-1.9) (-.002) (-2.4) (-.53) (-1.3) (.41) (-.34) (.085) Liquidassetsit 1 fedfundst.0081.015.016.065** -.028.023**.02*.025.038**.016 (.41) (.63) (.45) (2.4) (-.92) (2.2) (1.7) (1.3) (2.1) (1.3) Liquidassetsit 1 fedfundst 1 -.017 -.028 -.012 -.068**.022 -.018* -.017* -.018 -.032** -.0037 (-.88) (-1.2) (-.4) (-2.2) (.91) (-1.9) (-1.7) (-.83) (-2.3) (-.3) Liquidassetsit 1 fedfundst 2.0043.0064.0021.036 -.035 -.012 -.0067 -.034* -.01 -.016 (.2) (.26) (.049) (1) (-1.5) (-1.5) (-.76) (-1.7) (-.81) (-1.5) Liquidassetsit 1 fedfundst 3.0017.015 -.018 -.04.023 -.015* -.016*.014 -.028*** -.0074 (.082) (.58) (-.52) (-1.3) (.96) (-1.9) (-1.9) (.61) (-2.7) (-.74) Liquidassetsit 1 fedfundst 4 -.018 -.027 -.02.0078 -.026.017**.014.0077.035***.0049 (-.99) (-1.3) (-.58) (.32) (-1.1) (2.1) (1.6) (.42) (3) (.48) N 22186 16555 5631 8226 12638 22069 16612 5457 8246 12494 ar2.093.081.16.077.12.2.24.13.23.2 Sum of coefficients gapit 1 fedfunds.018.025 -.022.031 -.00075.013.013.01.016.014 p-value.046.016.2.024.95.002.0072.28.01.014 Sum of coefs Log(assetsit 1) fedfunds.0046.0044.0046.0043.0033.00019 -.000089.00055 -.003.0007 p-value 0.15.016.39.0085.71.95.64.15.25 Sum of coefs (equityit 1/assetsit 1) fedfunds -.066 -.038 -.19.035 -.18 -.049 -.033 -.086 -.034 -.12 p-value.37.64.0012.7 0.21.41.21.21.0056 Sum of coefs (liquidityit 1/assetsit 1) fedfunds -.021 -.019 -.031.00077 -.044 -.005 -.0056 -.0052.0032 -.0061 p-value.19.35.24.97.052.51.52.74.79.52 All regressions include calendar time dummies, 4 lags of the depend variable, as well as gapit 1 trend, log(assetsit 1), log(assetsit 1) fedfundst k for all k = 0, 1, 2, 3, 4, liquidassetsit 1, liquidassetsit 1 fedfundst k for all k = 0, 1, 2, 3, 4, equity it 1 and equity it 1 assets it 1 assets it 1 fedfundst k for all k = 0, 1, 2, 3, 4. 18

Table 8: Increases vs decreases in Fed funds rates: with equity ratio, and size interacted controls C&I lending growth Total lending growth All Small Big No Hedge Some Hedge All Small Big No Hedge Some Hedge Gapit 1 ( fedfundst) + -.023 -.027 -.01 -.099** -.012 -.019* -.023*.00028 -.046** -.0067 (-1.2) (-1.1) (-.27) (-2) (-.36) (-1.8) (-1.9) (.011) (-2.1) (-.26) Gapit 1 ( fedfundst 1) +.035.028.046.17** -.026.015.019 -.000048.043.011 (1.3) (.85) (1.1) (2.1) (-.52) (1.1) (1.2) (-.0016) (1.3) (.39) Gapit 1 ( fedfundst 2) +.012.012.0075 -.12*.021.018.016.039.01 -.013 (.5) (.41) (.17) (-1.7) (.36) (1.5) (1.1) (1.4) (.31) (-.47) Gapit 1 ( fedfundst 3) + -.041 -.029 -.074.039 -.041 -.016 -.016 -.011 -.0042.035 (-1.6) (-.93) (-1.6) (.66) (-.75) (-1.2) (-1) (-.41) (-.14) (1.3) Gapit 1 ( fedfundst 4) +.0096.012.016 -.0021.0039.011.0099.0093 -.0089 -.005 (.46) (.47) (.45) (-.047) (.11) (1.1) (.84) (.41) (-.44) (-.24) Gapit 1 ( fedfundst).013.014 -.0017.023.0064 -.0019 -.0026 -.0037.003.0015 (1.3) (1.3) (-.1) (1.4) (.46) (-.39) (-.46) (-.36) (.38) (.2) Gapit 1 ( fedfundst 1) -.0079 -.01.0082 -.0095.0092.011**.014*** -.0008.015**.0022 (-.77) (-.83) (.47) (-.61) (.68) (2.2) (2.7) (-.06) (2.1) (.29) Gapit 1 ( fedfundst 2).027***.035*** -.0096.041**.023*.0049.009 -.016.0077.00052 (2.7) (3.1) (-.55) (2.3) (1.8) (1) (1.6) (-1.5) (1.1) (.07) Gapit 1 ( fedfundst 3).014.015.014 -.0029.02*.0062.0051.011.0049.0085 (1.4) (1.3) (.66) (-.17) (1.7) (1.4) (1) (1) (.74) (1.2) Gapit 1 ( fedfundst 4) -.0081 -.0081 -.02.0018 -.022* -.0029 -.0061.0045 -.0057 -.0013 (-.94) (-.82) (-1.1) (.11) (-1.8) (-.68) (-1.4) (.4) (-.88) (-.2) N 28917 20927 7990 8226 12641 28835 21117 7718 8246 12497 ar2.097.089.15.076.12.2.23.14.23.19 Sum of coefficients gapit 1 ( fedfunds) + -.007 -.0035 -.015 -.015 -.054.0098.0049.038 -.0058.022 p-value.71.87.73.65.078.3.65.096.7.14 Sum of coefs gapit 1 ( fedfunds).038.046 -.0096.054.037.017.019 -.0042.025.011 p-value.0011.0011.72.014.023.0097.0072.8.0098.26 All regressions include calendar time dummies, 4 lags of the depend variable, as well as gapit 1 trend, log(assetsit 1), log(assetsit 1) fedfundst k for all k = 0, 1, 2, 3, 4, liquidassetsit 1, liquidassetsit 1 fedfundst k for all k = 0, 1, 2, 3, 4, equity it 1 and equity it 1 assets it 1 assets it 1 fedfundst k for all k = 0, 1, 2, 3, 4. 19

A. Time-Series Regressions 20

Table 9: Time Series Approach: Profits γ Interestt γ ROEt All Small Big No Hedge Some Hedge All Small Big No Hedge Some Hedge fedfundst.0033***.0037***.0016.0023.0039**.0052***.0049***.0074**.0083***.0049** (3.1) (3.1) (.89) (1.7) (2.4) (4.1) (3.3) (2.6) (4.5) (2.1) fedfundst 1.0027**.0015.006***.0023.0029.0022.003*.0003.0012.0038 (2.2) (1.1) (2.9) (1.5) (1.6) (1.5) (1.7) (.089) (.57) (1.5) fedfundst 2 -.00018 -.00037.00039.00069 -.00092 -.0017 -.00083 -.0046 -.00073 -.00091 (-.14) (-.27) (.19) (.45) (-.5) (-1.2) (-.48) (-1.4) (-.36) (-.34) fedfundst 3.0014.0024* -.0012.00012.00014.0023.00068.0054 -.00022.0024 (1.1) (1.7) (-.58) (.079) (.079) (1.6) (.4) (1.6) (-.11) (.93) fedfundst 4 -.00041 -.00092.00056 -.00088.00017 -.00037 -.00041.00034.0015 -.0012 (-.39) (-.77) (.32) (-.65) (.1) (-.3) (-.28) (.12) (.85) (-.53) N 93 93 93 63 63 93 93 93 63 63 ar2 Sum of coefficients fedfundst k.0067.0063.0073.0046.0062.0077.0073.0089.01.009 p-value 0 0 0 0 0 0 0.0019 0 0 In columns 1-5, we use change in interest income (divided by assets) as the dependent variable. In columns 6-10, we focus on change in Net Income (divided by book equity). All results come from a two-step procedure. First, for each quarter from 1986 to 2011, we regress the depend variable on four lags of itself, log(assetsit 1) and the duration gap gapit 1. We obtain a time series of coefficients on gapit 1, called γ X where X is the relevant dependent variable. We then regress γ X on fedfundst k for all k = 0, 1, 2, 3, 4 and four quarter dummies. 21

Table 10: Time Series Approach: Lending Growth with Size, Equity controls γc&ilendinggrowtht γt otallendinggrowtht All Small Big No Hedge Some Hedge All Small Big No Hedge Some Hedge fedfundst.012.013 -.013.0021.028 -.0026 -.0054 -.0036.0011.00023 (1.1) (1) (-.71) (.15) (1.2) (-.48) (-.97) (-.31) (.13) (.024) fedfundst 1 -.0004 -.008.032 -.0022 -.0066.0073.0069.014.0022.015 (-.03) (-.54) (1.5) (-.14) (-.26) (1.1) (1.1) (1) (.25) (1.4) fedfundst 2.01.013 -.019.014.012.0063.012* -.015 -.006.0089 (.76) (.89) (-.9) (.88) (.45) (.98) (1.8) (-1.1) (-.66) (.83) fedfundst 3.0048.01.0029.021 -.008 -.00032 -.0017.0065.013 -.00095 (.37) (.68) (.14) (1.3) (-.32) (-.05) (-.26) (.48) (1.5) (-.09) fedfundst 4 -.0036.00078 -.01 -.03**.0071.0029.0012.0097.0024 -.006 (-.32) (.062) (-.57) (-2.2) (.32) (.54) (.21) (.82) (.3) (-.64) N 93 93 93 63 63 93 93 93 63 63 ar2 Sum of coefficients fedfundst k.023.029 -.0065.004.032.014.013.011.013.017 p-value.04.02.7.76.14.012.018.33.095.065 In columns 1-5, we use growth of C&I loans as the dependent variable. In columns 6-10, we focus on total loans growth. All results come from a two-step procedure. First, for each quarter from 1986 to 2011, we regress the depend variable on four lags of itself, log(assetsit 1), equity it 1, and the duration gap gapit 1. We obtain assets it 1 a time series of coefficients on gapit 1, called γ X where X is the relevant dependent variable. We then regress γ X on fedfundst k for all k = 0, 1, 2, 3, 4 and four quarter dummies. 22

Table 11: Time Series Approach: Lending Growth with Size, Equity and KS controls γc&ilendinggrowtht γt otallendinggrowtht All Small Big No Hedge Some Hedge All Small Big No Hedge Some Hedge fedfundst.019.015 -.0046 -.0022.033.0012 -.0023 -.00058.0016.0061 (1.6) (1.1) (-.22) (-.16) (1.5) (.2) (-.41) (-.037) (.19) (.63) fedfundst 1 -.012 -.013.0011 -.0019 -.013 -.0013 -.0017.012.0027.0095 (-.9) (-.83) (.046) (-.12) (-.53) (-.19) (-.27) (.72) (.29) (.89) fedfundst 2.02.02.0075.014.023.011.02*** -.026 -.0047.0091 (1.5) (1.3) (.31) (.89) (.93) (1.5) (3.2) (-1.5) (-.49) (.84) fedfundst 3.009.0091.014.025 -.014.0036 -.00018.02.013 -.0046 (.67) (.59) (.57) (1.6) (-.57) (.53) (-.029) (1.1) (1.4) (-.44) fedfundst 4 -.017 -.0065 -.04* -.036**.0044 -.0018 -.0039.0018.00096 -.0025 (-1.4) (-.46) (-1.9) (-2.6) (.2) (-.3) (-.7) (.12) (.12) (-.27) N 67 67 67 63 63 67 67 67 63 63 ar2 Sum of coefficients fedfundst k.019.025 -.022 -.0014.034.012.012.0074.014.018 p-value.097.059.27.92.1.033.023.61.09.055 In columns 1-5, we use growth of C&I loans as the dependent variable. In columns 6-10, we focus on total loans growth. All results come from a two-step procedure. liquidity First, for each quarter from 1993 to 2011, we regress the depend variable on four lags of itself, log(assetsit 1), it 1 equity, it 1, and the duration gap assets it 1 assets it 1 gapit 1. We obtain a time series of coefficients on gapit 1, called γ X where X is the relevant dependent variable. We then regress γ X on fedfundst k for all k = 0, 1, 2, 3, 4 and four quarter dummies. 23