The Bank Lending Channel: A FAVAR Analysis

Villanova School of Business Economics Working Paper # 4 The Bank Lending Channel: a FAVAR Analysis Chetan Dave Scott J. Dressler University of Texas at Dallas Villanova University Lei Zhang University of Texas at Dallas April 2009 Abstract We examine the role of commercial banks in monetary transmission in a factoraugmented vector autoregression (FAVAR). A FAVAR exploits a large number of macroeconomic indicators to identify monetary policy shocks, and we add commonly used lending aggregates and lending data at the bank level. While our results suggest that the bank lending channel (BLC) is stronger than previously thought, this feature is not robust. In addition, our results indicate a di¤use response to monetary innovations when individual banks are grouped according to asset sizes and loan components. This suggests that other bank characteristics could improve the identi…cation of the BLC. Keywords: Bank Lending Channel; FAVAR; Monetary Policy JEL: E51, E52, C32 Corresponding author. Address: University of Texas at Dallas; School of Economic, Political and Policy Sciences; 800 W. Campbell Rd.; Richardson, TX 75080. Phone: (972) 883-2306. Fax: (972) 883-6486. Email: cdave@utdallas.edu. 1. Introduction Since Bernanke and Blinder’s (1992) observation that signi…cant movements in aggregate bank lending volume follow changes in the stance of monetary policy, the bank lending channel (henceforth, BLC) has been a prominent mechanism in the literature on monetary transmission. The BLC focuses on the balance sheets of commercial banks and assumes that insured, reservable deposits and other forms of external loan …nance (e.g. time deposits, CDs, etc.) are not perfect substitutes due to the higher costs of acquiring the latter. A monetary contraction resulting in less reservable deposits should therefore result in a decrease in the contemporaneous supply of loans. Building upon the initial intuition for the BLC, the literature has since stressed crosssectional di¤erences among commercial banks’ balance-sheets as well as loan components. Kashyap and Stein (1995, 2000) considered bank assets and liquidity positions as aggregating criteria and …nd that increases in the Federal funds rate are followed by signi…cant declines in lending volume for the smallest (in terms of assets) and least liquid banks.1 Den Haan et al. (2007) consider loan components aggregated across banks and …nd that real estate and consumer loans sharply decline in response to a monetary contraction while commercial and industrial (C&I) loans increase.2 While Perez (1998), Ashcraft (2006), and others have suggested the irrelevance of the BLC in monetary transmission, Kashyap and Stein (1995, 2000) and Den Haan et al. (2007) provide evidence for its existence. This evidence is not without its limitations. For example, the common use of the Federal funds rate as the monetary policy instrument may not result in an appropriate identi…cation of monetary policy innovations. In addition, aggregating bank lending across either asset categories or loan components may be contaminating the true responses of individual banks who are responding to both bank-speci…c and aggregate sources of ‡uctuations. It should 1 Kishan and Opiela (2000) further …nd that banks with the weakest capital positions are the most responsive to monetary policy. 2 The authors suggest that the perverse response of C&I loans could still be consistent with the BLC due to a bank’s preference for the relative safety and term of a C&I loan rather than a longer-term asset (such as a real estate loan). 2 be noted that most contributions to the literature which either supports or refutes the BLC are in some way subjected to these limitations. The goal of this paper is to empirically put these limitations to the test. We examine the lending response of commercial banks in a factor-augmented vector autoregression (FAVAR). A FAVAR, which combines standard structural VAR methods with factor analysis, exploits a large number of time series and summarizes the information into a relatively small set of estimated indexes (i.e. factors). It also has many desirable properties for an analysis of the BLC. First, utilizing a large data set of macroeconomic variables like those used by central banks is important when properly identifying monetary policy innovations. Bernanke et al. (2005) argue that the measurement of policy innovations is likely to be contaminated by limiting the analysis to a small number of comprehensive macroeconomic variables.3 Second, one does not need to take a stand on speci…c observables (such as industrial production or real GDP) which need to correspond to theoretical concepts (such as economic activity) because a FAVAR summarizes these concepts using large amounts of economic information. Finally, a FAVAR provides impulse responses for every variable in the conditioning set, as well as a decomposition of their individual ‡uctuations into those due to aggregate factors and speci…c innovations. Our FAVAR framework considers the large set of macroeconomic indicators used by Bernanke et al. in their identi…cation of monetary policy shocks, and extends this data by appending a variety of commercial-bank lending variables. First, total loan growth and growth in loan components are aggregated up to the total banking system (as in Bernanke and Blinder, 1992 and Den Haan et al., 2007) as well as up to groups according to asset size (as in Kashyap and Stein, 1995 and 2000).4 While these variables deliver an indication of how aggregate bank lending responds to an improved identi…cation of monetary policy shocks, 3 Imperfectly controlling for the information central bankers may have is exactly Sim’s (1992) critical interpretation of an increase in aggregate prices in response to a monetary contraction (i.e. the Price Puzzle) observed in traditional VAR analyses. 4 Following Kashyap and Stein (1995, 2000), we consider the asset groups to be banks with assets within the 95th percentile or less (small), banks with assets within the 95th and 99th percentile (medium), and banks with assets within the 99th percentile or more (large). 3 we also consider a large amount of lending data at the individual-bank level. This allows us to disentangle the ‡uctuations in bank-level lending data which are due to aggregate macroeconomic factors (such as a change in monetary policy) from those that are due to bank-speci…c conditions. To our knowledge, our analysis is the …rst to consider purely disaggregated lending data within the same framework as their commonly used aggregates and provides a comparison between the responses of individual and aggregate lending in response to monetary policy. Our …ndings suggest a stronger BLC than previously thought when examining data aggregated up to the banking sector and asset groups. In particular, we …nd that total, C&I, and individual consumer loan growth all signi…cantly decline after a monetary policy contraction for the entire banking sector as well as bank groups according to asset size. While this suggests that the BLC e¤ects more than just the smallest banks, this result weakens when employing post-1984 data. Our results also suggest that the individual, bank-level responses to a monetary policy innovation are quite di¤use. There are almost as many banks who increase lending as those who decrease, and this result remains if we control for bank groups and loan components. A main reason for these varied responses is that macroeconomic ‡uctuation explain on average between 8 and 22 percent of the variation in individual bank lending for the banks within our sample. Therefore, most of the variation in individual bank-lending re‡ect bank-speci…c shocks to which the banks immediately respond. Nonetheless, when considering lending aggregates comprised of only those banks we observe individually, there are signi…cant declines for all bank groups in one or more loan components, and these declines remain when employing post-1984 data. Our analysis indicates that while particular measures of the BLC are strengthened by our FAVAR framework, the large degree of heterogeneity observed in the individual-bank responses cast doubt on the notion that the BLC is stronger for banks based on asset size or loan components. This does not imply that other banking characteristics might prove more 4 suitable to di¤erentiate banks which have a di¤erent BLC e¤ect. For example, Cetorelli and Goldberg (2009) show that the degree of globalization of a commercial bank could matter for the BLC because globalized banks can activate foreign capital markets to insulate themselves from domestic liquidity conditions. Evidence of this type will prove useful to target the important features of intermediation in monetary transmission and provide theorists with a set of crucial features of banking to incorporate into their environments. The rest of the paper is organized as follows. Section 2 outlines the formulation and estimation of the FAVAR. Section 3 discusses the data sets used. Section 4 presents our empirical results by …rst detailing the impulse responses of loan aggregates to a monetary policy shock, and then examining the characteristics of disaggregated loan data. Section 5 concludes. 2. The FAVAR Our implementation of the FAVAR follows Bernanke et al. (2005). A general description of the framework is as follows. Assume the economy is a¤ected by a vector Ct of common components which a¤ect all variables in the data set. For example, we assume that a measure of the stance of monetary policy is considered to be a common component, and we follow the literature and assume that this stance is measured by the Federal funds rate (Rt ). The remaining shared dynamics of each data series are captured by a K 1 vector of unobserved factors Ft , where K is relatively small. These unobserved factors capture ‡uctuations in general economic concepts such as economic activity, aggregate prices, credit conditions, etc., that cannot be easily represented by a few time series but rather are re‡ected in a wide range of economic variables. We assume that the joint dynamics of Ft and Rt are given by Ct = (L) Ct 5 1 + t (1) where Ct0 = [Ft0 Rt ] and (L) is a conformable lag polynomial of in…nite order which may contain a priori restrictions as in the structural VAR literature. The error term t is i.i.d. with zero mean and covariance matrix Q. While (1) is a VAR in Ct , it cannot be directly estimated because the factors comprising Ft are unobserved. However, since these factors are interpreted as representing forces a¤ecting many economic variables, one can potentially use a large set of observed “informational” series to infer something about them. Let Xt denote the N 1 vector of these informational variables, where N is relatively large. It is assumed that the Xt is related to all common components according to (2) Xt = Ct + et where is an N (K + 1) matrix of factor loadings. The N 1 vector et contains the zero- mean, series-speci…c components that are uncorrelated with Ct , but allowed to be serially correlated and weakly correlated across indicators. Equation (2) re‡ects that Ct represents pervasive forces which drive the common dynamics of Xt . Conditional on Rt , the Xt are thus noisy measures of the underlying unobserved factors Ft . Bernanke et al. note that the implication of Xt depending only on current factors is not restrictive in practice, as Ft can be interpreted as including arbitrary lags of the fundamental factors. Estimation of the above model involves a two-step principal component approach. In the …rst step, principal components are extracted from Xt to obtain consistent estimates of the common factors. In the second step, the Federal funds rate is added to the estimated common factors and the data set is used to estimate (1). In particular, estimation of our model follows Boivin et al. (2009), who slightly di¤er from the estimation described by Bernanke et al. insofar that it is assumed that Rt is one of the factors in the …rst-step. This guarantees that the latent factors recover common dynamics not captured by the Federal funds rate.5 5 See Boivin et al. (2009) for details. 6 3. The Data Our data set is a balanced panel of 1512 quarterly series from 1976:1 to 2005:3. The …rst 111 series are macroeconomic indicators originally considered in the initial FAVAR analysis of Bernanke et al., and also used by Boivin et al. (see appendix for details).6 Included in these series are several measures of industrial production, price indices, interest rates, employment as well as other key macroeconomic and …nancial variables, which have been found to contain information useful to capture the state of the economy and identify monetary policy. The remainder of our data set includes several variables constructed using loan information for individual commercial banks taken from the Consolidated Report of Condition and Income (Call Reports) that all insured banks submit to the Federal Reserve. For each commercial bank, data on total loans, total C&I, total real estate loans, and individual loans were collected following the detailed instructions on forming consistent time series attributable to Kashyap and Stein (2000). For each quarter, we used total asset holdings of the commercial banks to assign each bank into one of three possible size categories: banks with total assets below the 95th percentile (small banks), banks with total assets between the 95th and 99th percentile (medium banks), and banks with total assets above the 99th percentile (large banks). To retain comparability with previous studies of the BLC, we use these asset categories to construct a disaggregation of the commercial banking data. In particular, we use the lending data to construct loan growths for all bank components aggregated up to the entire sector as well as the three asset groups. However, since the FAVAR framework can handle large amounts of data, we also keep individual banks separate and use the asset size categories to determine if there are any common movements in banks that di¤er across this characteristic. In order to arrive at a manageable data set for our FAVAR analysis, we had to apply several …lters on the individual bank-level data. In particular, our balanced panel of commercial banks initially consisted of 4743 individual banks. Of these, 219 banks were removed because 6 We are grateful to Marc Giannoni for providing us with this data. 7 their bank size was not consistent throughout the sample. The resulting data set consisted of 18 large banks, 24 medium banks, and 4482 small banks. Since the small banks are still too numerous, the data set we settled on to estimate the FAVAR consists of a random selection of 10 percent of the small bank population.7 We then used these banks to construct time series for their loan growths in the exact same way as in the loan aggregates. In addition, in order to directly compare the individual bank responses with some aggregate measure of their response, we also constructed loan aggregates similar to those above for all banks but only using the banks we observe in our bank-level data set. 4. Estimation Results We estimate the above system (1) and (2) for four di¤erent FAVARs which di¤er in the type of bank lending (total, C&I, real estate, and individual). For each FAVAR, the data set Xt consisted of the macroeconomic indicators as well as the aggregate and bank-level lending data for each particular loan category. We chose the size of factors Ft for each FAVAR after some experimentation to ensure that our conclusions are not a¤ected by additional latent factors.8 All models use 4 quarterly lags in estimating (1). The …rst subsection focuses on the response of aggregated lending data to a monetary policy shock, while the second subsection focuses on the characteristics and behavior of the disaggregated lending data. 4.1. Aggregated Lending Following Bernanke et al. and Boivin et al., we assume that the Federal funds rate may respond to contemporaneous ‡uctuations in estimated factors, but that none of the latent common components can contemporaneously respond to monetary policy shocks. This is 7 The estimation was conducted for several di¤erent samples of small banks to ensure robustness of the results. 8 Our FAVARs for Total and Real Estate loans required 5 latent factors, while C&I and Industrial loans required 4. 8 the FAVAR extension of the standard recursive identi…cation of monetary policy shocks in conventional VARs, which has been used for instance by Den Haan (2007). Note that in contrast to VARs, the macroeconomic indicators (Xt ) are allowed to contemporaneously respond to monetary shocks. We can therefore disentangle monetary policy shocks from the other macroeconomic shocks. The responses of our lending aggregates to an unexpected (25 basis point) increase of the Federal Funds rate are illustrated in Figure 1. Each panel illustrates the response of a particular loan component for loans aggregated across all banks as well as the three asset groups. A diamond indicates that the impulse response at that particular time horizon is signi…cantly di¤erent than zero at the 90 percent level. As the …gure indicates, there are signi…cant and persistent declines in Total, C&I, and Individual loans in response to a monetary policy shock for all bank groups (including the total). This result suggests that the BLC is actually stronger than previously reported under this identi…cation of monetary policy. Previous analyses only …nd a signi…cant BLC in either the smallest (asset-wise) or least liquid banks. Only aggregate real estate lending illustrates a BLC for the smallest banks exclusively. Our result of a stronger BLC, however, is not a consistent feature of the data. Due to the evidence of widespread instability in many macroeconomic series, a change in monetary policy, and a decline in overall macroeconomic volatility around 1984, we re-estimated our FAVARs using post-1984 data. Figure 2 illustrates a large reduction in the signi…cance of the impulse responses. In fact, some loan components actually increase (albeit, insigni…cantly) after a monetary contraction. The only loan component which retains a signi…cant BLC is Individual loans, which accords with the results of Den Haan (2007) who …nd the largest BLC e¤ect in aggregate consumer loans.9 9 It should be noted that our results for real estate loans also mimic the results of Den Haan (2007), but fail to be signi…cant. 9 4.2. Disaggregated Lending This section turns to an analysis of the bank-speci…c lending data which was used along with the loan aggregates for estimating the system (1) and (2). For all loan growth series considered, (2) implies xit = 0 i Ct (3) + eit ; where xit is the quarterly change in loan growth for bank i. The ‡uctuations for all banks due to the macroeconomic factors are represented by the the common components Ct which have a di¤use e¤ect on the individual banks due to di¤erences in i, while the bank-speci…c ‡uctuations are captured by eit . We detail some summary statistics on the average volatility of loan components and their corresponding aggregates in Table 1. It should be noted that the corresponding aggregates are not the aggregates discussed in the previous section, but the aggregation of banks that appear in our bank-level data set. This comparison serves to illustrate potential aggregation e¤ects among the individual banks. The …rst column of Table 1 suggests a large amount of average volatility in our banklevel lending data. This volatility is decomposed into volatility stemming from common macroeconomic and speci…c factors, and the R2 statistic measures the fraction of the variance in aggregate lending explained by the common components. The results suggest that loan ‡uctuations stemming from aggregate or common shocks make up a very small amount of the average volatility in our lending data. For example, when considering banks with assets less than the 95th percentile, bank-speci…c shocks account on average for 76 percent of their ‡uctuations in total lending and as much as 92 percent of their ‡uctuations in lending components. When comparing these volatilities with their corresponding aggregates, one …nds a large reduction in volatility due to a large reduction in the bank-speci…c component. The R2 statistics now state that 75 percent or more of the ‡uctuations in these variables are attributable to ‡uctuations in macroeconomic components. While this comparison is the 10 most stark for the smallest bank group, similar comparisons for all bank groups and all loan components report a reduction in volatility due to aggregating the bank-level data as well as an increased R2 . Quite naturally, these results suggests that disturbances arising at the individual bank level tend to cancel each other out when aggregating. Returning focus to the bank-level data, Figure 3 illustrates a strongly positive correlation between the macroeconomic and bank-speci…c components of lending volatility. While the …gure considers all banks, this positive relationship between the volatility of the idiosyncratic shocks (Sd (ei )) and the volatility of the common component (Sd ( 0i Ct )) would remain if we considered banks groups separately. Among the loan components, the tightest relationships are among C&I and Real Estate loans, with a weaker relationship among Individual loans and the total. All slope coe¢ cients are statistically di¤erent from zero at the 95 percent level, and are corrected for possible heteroscedasticity.10 From this perspective, banks with the highest idiosyncratic volatility also respond the strongest to macroeconomic shocks. Therefore, whatever characteristics which help banks smooth over individual shocks will also help smooth over macroeconomic shocks. Our …nal analysis of the bank-level data is to document how loan growth responds to bank-speci…c and macroeconomic disturbances. These impulse responses are illustrated in Figures 4 through 6 for large, medium, and small banks, respectively. The left panels of the …gures report the response of each of the individual banks to an adverse (one standard deviation) shock to its bank-speci…c component. The solid lines represent an unweighted average response. Across all bank groups and loan components, lending responds sharply and promptly to bank-speci…c disturbances. There is very little persistence in the response of all banks, and they quickly reach a new equilibrium. While bank-speci…c shocks rapidly shift the loan growth of individual banks to a new level, the macroeconomic shocks are quite di¤erent. The middle panels of the …gures illustrate the response of each bank group and loan component to an innovation (of minus one 10 The respective t statistics for the slope coe¢ cients are 10.5, 22.9, 26.7, and 12.4. 11 standard deviation) to its common component 0 i Ct . These …gures suggest a large amount of sluggishness in the response to macroeconomic disturbances, and this persistence is shared by all bank groups and all loan components. In particular, the …gures illustrate that all banks behave quite similarly to a common macroeconomic shock. While this exercise fails to identify a speci…c structural macroeconomic shock and instead illustrates the response to a combination of macroeconomic shocks, the response to these shocks strongly contrast with the responses to bank-speci…c shocks. We …nally turn to the e¤ects of monetary policy on our bank lending panel. The identi…cation of monetary policy shocks is accomplished in the same way as the previous section focusing on the aggregated lending data, and the results are illustrated in the third columns of Figures 4 through 6 for our three bank groups. The thick solid lines again represent an unweighted average response, while the thick dashed lines illustrate the response of the corresponding aggregated data mentioned in the discussion of Table 1. A circle indicates that the impulse response at that particular time horizon for the aggregated data is signi…cantly di¤erent than zero at the 90 percent level. Similar to the responses illustrated in Figures 1 and 2, there is a fair amount of signi…cance in the aggregated data for all bank groups. The most signi…cant declines appear to be coming from C&I and Real Estate loan components, with only a few signi…cant periods of decline for Individual loans exclusively from the small bank group. It should be kept in mind that these aggregates are not over the entire banking sector, but for the banks that made it into our bank-level panel. These banks have their own individual response to the same monetary policy shock, and are illustrated in the …gures by the thin dotted lines. A striking feature of these bank-speci…c responses is that there is a large amount of heterogeneity among banks of similar asset size, with almost as many banks increasing their lending in response to a surprise monetary contraction as there are banks decreasing their lending. This is a robust feature of the data across bank size and loan components, and suggests a rather stark discrepancy between an individual bank response and the aggregate of which it is a member. 12 4.3. 4.3.1. Robustness Results Disaggregated Lending, Post-1984 Similar to the aggregate lending results, the disaggregated bank-level data was also reestimated using post-1984 data. In contrast to the loan aggregates, the disaggregated-loan results tell a similar story to the results under the full sample. As illustrated in Table 2, we …nd much less volatility in the aggregated time series relative to the average volatility of the bank-level panel, as well as a much larger percentage of the aggregate ‡uctuations being attributable to ‡uctuations in macroeconomic components. More importantly, our impulse response analysis under post-1984 data retains much of the signi…cance of the BLC illustrated under the full sample. These are illustrated in Figures 7-9. Again, it should be noted that these aggregates are constructed using only the individual banks in our panel and therefore is not a complete picture of aggregate lending. 4.3.2. Alternative Factor Estimations In order to verify that the number of factors in our FAVARs were reasonable, we performed robustness checks as in Bernanke et al. by re-estimating the FAVARs with an increased number of factors. The impulse responses for bank speci…c, common component and monetary policy shocks did not qualitatively change from the main results discussed above for total, C&I, or Real Estate loans. The only minor change in impulse responses was for individual loans, which displayed a slightly positive response to a contractionary monetary shock for aggregated large banks. In terms of the disaggregated data, increasing the number of factors did not qualitatively change the results reported in Table 1, in particular, the R2 calculations measuring the amount of ‡uctuation in individual bank lending attributable to macroeconomic shocks. 13 5. Conclusion This paper examined the role of commercial banks in monetary transmission in a factor- augmented vector autoregression (FAVAR). The ability of a FAVAR to exploit a large conditioning set of macroeconomic indicators when identifying monetary policy shocks, coupled with the ability to calculate impulse responses for every variable in this set, allows us to assess the bank lending channel of monetary policy using commonly considered lending aggregates and a panel of individual lending data. Our analysis delivers two results. First, our results suggest that the BLC is stronger than previously thought for loan aggregates and particular loan components. In particular, we …nd a signi…cant BLC for more bank groups than the smallest banks (asset-wise) as well as for Total, C&I, and Individual loans. While this result suggests that an improved identi…cation of monetary policy shocks uncovers a profound BLC, it is not robust when employing post1984 data. Second, the bank-level data show a di¤use response to monetary innovations. We …nd that almost as many commercial banks increase their lending in response to a monetary contraction as there are banks that decrease. This result is unchanged when considering di¤erent loan components as well as banks grouped according to asset size. However, the responses of loan aggregates using only the banks in our balanced panel still show a stronger BLC than previously thought, and this result is robust to employing post-1984 data. We believe that these results deliver two conclusions. First, the improved identi…cation of monetary policy shocks stemming from a FAVAR analysis is a useful tool for uncovering the BLC and potentially many more monetary matters. Second, the large degree of heterogeneity among commercial banks of similar asset size as well as loan components suggests that these might not be the best characteristics for di¤erentiating whether or not a bank is susceptible to the BLC. Some alternative characteristics (such as the level of global operations, geographic characteristics, etc.) are presently being proposed in the literature, and it would be interesting to see if the individual lending responses from our FAVAR would behave similarly when banks share these similar characteristics. Uncovering these impor14 tant characteristics of banking can uncover their role in monetary transmission, and provide theorists with a set of crucial features of banking to incorporate into their environments. References [1] Ashcraft, Adam (2006), “New Evidence on the Lending Channel,” Journal of Money, Credit, and Banking 38(3), 751-775. [2] Bernanke, Ben S. and Alan S. Blinder (1992), “The Federal Funds Rate and the Channels of Monetary Transmission,”American Economic Review 82(4), 901-921 [3] Bernanke, Ben S., Jean Boivin, and Piotr Eliasz (2005), “Measuring Monetary Policy: A Factor Augmented Vector Autoregression (FAVAR) Approach,”Quarterly Journal of Economics 120(1), 387-422. [4] Boivin, Jean, Marc P. Giannoni, and Ilian Mihov (2009), “Sticky Prices and Monetary Policy: Evidence from Disaggragated U.S. Data,” American Economic Review, forthcoming. [5] Cetorelli, Nicola and Linda S. Goldberg (2009), “Bank Globalization and Monetary Transmission,”mimeo. [6] Christiano, Lawrence J., Martin Eichenbaum, and Charles L. Evans (1999), “Monetary Policy Shocks: What Have We Learned and to What End?”In: Taylor J.B., Woodford, M. (Eds.), Handbook of Macroeconomics. North-Holland, Amsterdam. [7] Den Haan, Wouter J., Steven W. Sumner, and Guy M. Yamashiro (2007), “Bank Loan Portfolios and the Monetary Transmission Mechanism,”Journal of Monetary Economics 54(3), 904-924. [8] Kashyap, Anil K. and Jeremy C. Stein (1995), “The Impact of Monetary Policy on Bank Balance Sheets,”Carngie-Rochester Conference Series on Public Policy 42, 1551-195. 15 [9] Kashyap, Anil K. and Jeremy C. Stein (2000), “What do a Million Observations on Banks Say About the Transmission of Monetary Policy?” American Economic Review 90(3), 407-428. [10] Kishan, R.P. and T.P. Opelia (2000), “Bank Size, Bank Capital, and the Bank Lending Channel,”Journal of Money, Credit, and Banking 32(1), 121-141. [11] Perez, Stephen, J. (1998), “Causal Ordering and the Bank Lending Channel,” Journal of Applied Econometrics 13, 613-626. [12] Sims, Christopher (1992), “Interpreting the Macroeconomic Time Series Facts: The E¤ects of Monetary Policy,”European Economic Review 36, 975-1000. 16 Appendices 6. Data Appendix Our data set consists of 111 macroeconomic indicators, a series of loan aggregates, and a series of lending at the individual bank level. The macroeconomic indicators are the same as those used by Bernanke et al. and Boivin et al., and we refer to the data appendix of Bernanke et al. (2005) pages 416-420. Our lending data was constructed following Notes on Forming Consistent Time Series written by Anil Kashyap and Jeremy Stein and is available at the Federal Reserve Bank of Chicago website.11 7. 11 Tables and Figures http://www.chicagofed.org/economic_research_and_data/commercial_bank_data.cfm 17 Table 1: Volatility and Persistence of Bank Lending Disaggregates Aggregates xit Common Speci…c R2 xit Common Speci…c Total Loans All 5.82 2.54 5.23 0.22 5.47 2.51 4.86 Large 9.04 2.64 8.63 0.10 6.03 2.69 5.39 Medium 6.01 2.09 5.58 0.16 2.10 1.42 1.55 Small 5.67 2.56 4.95 0.24 1.30 1.23 0.43 R2 0.21 0.20 0.46 0.89 C&I Loans All 18.38 Large 10.30 Medium 9.90 Small 19.20 4.69 3.19 3.18 4.84 17.68 9.71 9.28 18.50 0.08 0.12 0.14 0.08 6.97 7.38 2.99 2.13 3.20 3.34 2.02 1.84 6.19 6.58 2.21 1.07 0.21 0.20 0.46 0.75 Real Estate All 9.06 Large 10.66 Medium 7.56 Small 9.07 2.82 3.14 2.44 2.82 8.54 10.14 7.13 8.55 0.13 0.10 0.12 0.13 3.78 4.24 1.99 1.23 2.33 2.59 1.16 1.09 2.97 3.35 1.61 0.58 0.38 0.37 0.34 0.78 Individual All 11.79 Large 14.80 Medium 9.04 Small 11.82 3.73 3.32 2.91 3.78 11.02 14.33 8.48 11.03 0.13 0.08 0.15 0.13 4.30 4.88 2.91 1.94 1.31 1.35 2.00 1.86 4.09 4.69 2.11 0.55 0.09 0.08 0.47 0.92 18 Total Loans C&I Loans 0 0 -0.2 -0.2 -0.4 -0.4 Total Large Medium Small -0.6 0 4 8 -0.6 -0.8 12 16 0 Real Estate Loans 4 8 12 16 12 16 Individual Loans 0 0.2 -0.2 0 -0.4 -0.6 -0.2 -0.8 -0.4 -1 0 4 8 12 16 0 4 8 Figure 1: Impulse responses of lending aggregates (in %) to an identi…ed monetary policy shock. The monetary shock is a surprise of 25 basis points in the Federal funds rate. The label Total stands for each loan component aggregated across all banks, while sizes stand for aggregates according to asset size. A diamond indicates a signi…cant distance from zero at the 90 percent con…dence level. 19 Total Loans C&I Loans 0.1 0 0 -0.2 -0.1 -0.4 -0.2 -0.6 Total Large Medium Small -0.3 -0.4 0 4 8 -0.8 -1 12 16 0 Real Estate Loans 4 8 12 16 12 16 Individual Loans 0 0.2 -0.2 0 -0.4 -0.6 -0.2 -0.8 -0.4 -1 0 4 8 12 16 0 4 8 Figure 2: Post-1984: Impulse responses of lending aggregates (in %) to an identi…ed monetary policy shock. The monetary shock is a surprise of 25 basis points in the Federal funds rate. The label Total stands for each loan component aggregated across all banks, while sizes stand for aggregates according to asset size. A diamond indicates a signi…cant distance from zero at the 90 percent con…dence level. 20 Total Loans 30 C&I Loans 80 sd(e ) = 3.07 + 0.80 * sd(λ' C) i sd(e ) = 6.00 + 2.47 * sd(λ' C) i 2 2 R = 0.19 i i R = 0.53 60 i sd(e ) i sd(e ) 20 40 10 20 0 0 5 10 sd(λ' C) 15 0 20 0 10 i 30 i Real Estate 60 20 sd(λ' C) Ind. Loans 60 sd(e ) = -0.11 + 3.06 * sd(λ' C) i i 2 R = 0.59 i i sd(e ) 40 sd(e ) 40 sd(e ) = 7.35 + 0.97 * sd(λ' C) i i 2 R = 0.23 20 0 20 0 5 10 0 15 sd(λ' C) i 0 10 20 sd(λ' C) 30 40 i Figure 3: Standard deviation (in %) of bank-speci…c and macroeconomic (common) components of bank lending. The solid line denotes the cross-section regression line. 21 22 16 16 8 12 16 0 12 4 8 12 16 16 0 1 -2 -1 0 -1 0 4 8 12 Ind. Loans: Bank-Specific 16 -15 -10 -5 0 0 4 8 12 Ind. Loans: Common Component 16 -1 0 1 0 0 0 0 8 12 8 12 8 12 4 8 12 Ind. Loans: Monetary Shock 4 RE Loans: Monetary Shock 4 C&I Loans: Monetary Shock 4 Total Loans: Monetary Shock 16 16 16 16 Figure 4: Estimated impulse responses of lending variables (in %) for large banks to a bank-speci…c shock eit of one standard deviation (left column), a common shock 0i Ct of one standard deviation (middle column), and an identi…ed monetary policy shock (right column). The monetary shock is a surprise of 25 basis points in the Federal funds rate. The thick solid lines represent unweighted average responses, while the thick-dashed lines represent the response of aggregate lending for large banks. -20 -10 0 -20 8 RE Loans: Common Component 4 16 -2 4 12 -15 0 8 C&I Loans: Common Component 4 -0.5 0 0.5 -1 -10 0 0 0 Total Loans: Common Component -10 -5 0 RE Loans: Bank-Specific -15 12 -15 8 -10 -10 4 -5 -5 0 0 0 C&I Loans: Bank-Specific -15 12 -15 8 -10 -10 4 -5 -5 0 0 Total Loans: Bank-Specific 0 23 16 16 16 16 0 0 0 0 8 12 8 12 8 12 4 8 12 Ind. Loans: Common Component 4 RE Loans: Common Component 4 C&I Loans: Common Component 4 Total Loans: Common Component 16 16 16 16 -1 0 1 -1 0 1 -1 0 1 -1 0 1 0 0 0 0 8 12 8 12 8 12 4 8 12 Ind. Loans: Monetary Shock 4 RE Loans: Monetary Shock 4 C&I Loans: Monetary Shock 4 Total Loans: Monetary Shock 16 16 16 16 Figure 5: Estimated impulse responses of lending variables (in %) for middle banks to a bank-speci…c shock eit of one standard deviation (left column), a common shock 0i Ct of one standard deviation (middle column), and an identi…ed monetary policy shock (right column). The monetary shock is a surprise of 25 basis points in the Federal funds rate. The thick solid lines represent unweighted average responses, while the thick-dashed lines represent the response of aggregate lending for middle banks. -15 12 -15 8 -10 -10 4 -5 -5 0 0 0 Ind. Loans: Bank-Specific -15 12 -15 8 -10 -10 4 -5 -5 0 0 0 RE Loans: Bank-Specific -15 12 -15 8 -10 -10 4 -5 -5 0 0 0 C&I Loans: Bank-Specific -15 12 -15 8 -10 -10 4 -5 -5 0 0 Total Loans: Bank-Specific 0 24 16 16 16 16 0 0 0 0 8 12 8 12 8 12 4 8 12 Ind. Loans: Common Component 4 RE Loans: Common Component 4 C&I Loans: Common Component 4 Total Loans: Common Component 16 16 16 16 -2 0 2 -2 0 2 -2 0 2 -1 0 1 0 0 0 0 8 12 8 12 8 12 4 8 12 Ind. Loans: Monetary Shock 4 RE Loans: Monetary Shock 4 C&I Loans: Monetary Shock 4 Total Loans: Monetary Shock 16 16 16 16 Figure 6: Estimated impulse responses of lending variables (in %) for small banks to a bank-speci…c shock eit of one standard deviation (left column), a common shock 0i Ct of one standard deviation (middle column), and an identi…ed monetary policy shock (right column). The monetary shock is a surprise of 25 basis points in the Federal funds rate. The thick solid lines represent unweighted average responses, while the thick-dashed lines represent the response of aggregate lending for small banks. 12 -20 8 -40 4 -10 -20 0 0 0 Ind. Loans: Bank-Specific 12 -20 8 -40 4 -10 -20 0 0 0 RE Loans: Bank-Specific 12 -20 8 -40 4 -10 -20 0 0 0 C&I Loans: Bank-Specific 12 -20 8 -20 4 -10 -10 0 0 Total Loans: Bank-Specific 0 Table 2: Volatility of Bank Lending, Post-1984 Disaggregates Aggregates xit Common Speci…c R2 xit Common Speci…c Total Loans All 5.73 2.28 5.13 0.19 3.37 1.79 2.85 Large 9.32 2.52 8.93 0.08 3.55 1.84 3.03 Medium 5.75 1.90 5.40 0.12 2.18 1.49 1.59 Small 5.58 2.29 4.96 0.20 1.10 1.02 0.43 R2 0.28 0.27 0.47 0.84 C&I Loans All 17.53 Large 10.31 Medium 9.76 Small 18.28 5.06 3.34 3.50 5.22 16.68 9.66 9.02 17.41 0.10 0.14 0.15 0.09 3.98 4.14 3.03 2.09 2.15 2.18 2.11 1.66 3.35 3.52 2.18 1.26 0.29 0.28 0.48 0.63 Real Estate All 8.30 Large 11.54 Medium 7.24 Small 8.23 2.33 2.97 1.90 2.33 7.90 11.10 6.95 7.82 0.11 0.08 0.09 0.11 3.91 4.28 2.00 0.96 2.20 2.38 1.08 0.63 3.23 3.56 1.68 0.73 0.32 0.31 0.29 0.43 Individual All 11.65 Large 16.02 Medium 9.39 Small 11.60 3.68 3.71 2.64 3.73 10.85 15.48 8.94 10.78 0.13 0.07 0.12 0.13 4.01 4.35 2.73 1.56 1.02 1.02 1.36 1.38 3.87 4.23 2.37 0.72 0.07 0.05 0.25 0.78 25 26 12 16 12 16 0 0 0 0 8 12 8 12 8 12 4 8 12 Ind. Loans: Common Component 4 RE Loans: Common Component 4 C&I Loans: Common Component 4 Total Loans: Common Component 16 16 16 16 -1 0 1 -3 -2 -1 0 1 -2 -1 0 1 -1.5 -1 -0.5 0 0.5 0 0 0 0 8 12 8 12 8 12 4 8 12 Ind. Loans: Monetary Shock 4 RE Loans: Monetary Shock 4 C&I Loans: Monetary Shock 4 Total Loans: Monetary Shock 16 16 16 16 Figure 7: (Post-1984) Estimated impulse responses of lending variables (in %) for large banks to a bank-speci…c shock eit of one standard deviation (left column), a common shock 0i Ct of one standard deviation (middle column), and an identi…ed monetary policy shock (right column). The monetary shock is a surprise of 25 basis points in the Federal funds rate. The thick solid lines represent unweighted average responses, while the thick-dashed lines represent the response of aggregate lending for large banks. 12 -20 -20 8 -10 4 16 -10 0 8 Ind. Loans: Bank-Specific 4 0 0 -20 -10 0 0 -15 -10 -5 0 RE Loans: Bank-Specific -15 16 -15 12 -10 -10 8 -5 -5 4 0 -15 0 0 8 C&I Loans: Bank-Specific 4 -10 -10 0 -5 -5 -15 0 Total Loans: Bank-Specific 0 27 16 16 0 0 8 12 4 8 12 Ind. Loans: Bank-Specific 4 16 16 -15 -10 -5 0 -20 -10 0 0 0 0 0 8 12 8 12 8 12 4 8 12 Ind. Loans: Common Component 4 RE Loans: Common Component 4 C&I Loans: Common Component 4 Total Loans: Common Component 16 16 16 16 -1 0 1 -1 0 1 -2 -1 0 -1 -0.5 0 0.5 0 0 0 0 8 12 8 12 8 12 4 8 12 Ind. Loans: Monetary Shock 4 RE Loans: Monetary Shock 4 C&I Loans: Monetary Shock 4 Total Loans: Monetary Shock 16 16 16 16 Figure 8: (Post-1984) Estimated impulse responses of lending variables (in %) for middle banks to a bank-speci…c shock eit of one standard deviation (left column), a common shock 0i Ct of one standard deviation (middle column), and an identi…ed monetary policy shock (right column). The monetary shock is a surprise of 25 basis points in the Federal funds rate. The thick solid lines represent unweighted average responses, while the thick-dashed lines represent the response of aggregate lending for middle banks. -20 -10 0 -15 -10 -5 0 RE Loans: Bank-Specific -15 12 -15 8 -10 -10 4 -5 -5 0 0 0 C&I Loans: Bank-Specific -15 12 -15 8 -10 -10 4 -5 -5 0 0 Total Loans: Bank-Specific 0 28 16 16 16 12 8 12 8 12 12 8 12 16 -2 0 0 0 0 8 12 8 12 8 12 4 8 12 Ind. Loans: Monetary Shock 4 RE Loans: Monetary Shock 4 C&I Loans: Monetary Shock 4 Total Loans: Monetary Shock 16 16 16 16 Figure 9: (Post-1984) Estimated impulse responses of lending variables (in %) for small banks to a bank-speci…c shock eit of one standard deviation (left column), a common shock 0i Ct of one standard deviation (middle column), and an identi…ed monetary policy shock (right column). The monetary shock is a surprise of 25 basis points in the Federal funds rate. The thick solid lines represent unweighted average responses, while the thick-dashed lines represent the response of aggregate lending for small banks. 8 -20 4 -40 0 0 -2 0 2 -2 -10 4 16 16 0 2 -1 -20 Ind. Loans: Common Component 4 RE Loans: Common Component 4 16 2 0 8 C&I Loans: Common Component 4 0 1 0 16 0 0 0 Total Loans: Common Component 0 Ind. Loans: Bank-Specific 12 -20 8 -40 4 -10 -20 0 0 0 RE Loans: Bank-Specific 12 -20 8 -40 4 -10 -20 0 0 0 C&I Loans: Bank-Specific 12 -20 8 -20 4 -10 -10 0 0 Total Loans: Bank-Specific 0