What Do a Million Observations on Banks Say About the Transmission of Monetary Policy?

Transcript

1 What Do a Million Observations on Banks Say About the Transmission of Monetary Policy? * NIL J EREMY C. S TEIN ASHYAP AND KK A By We study the monetary-transmission mechanism with a data set that includes quarterly observations of every insured U.S. commercial bank from 1976 to 1993. We find that the impact of monetary policy on lending is stronger for banks with less liquid balance sheets—i.e., banks with lower ratios of securities to assets. More- over, this pattern is largely attributable to the smaller banks, those in the bottom 95 percent of the size distribution. Our results support the existence of a “bank lending channel” of monetary transmission, though they do not allow us to make precise statements about its quantitative importance. ( JEL E44, E52, G32) In this paper, we use a new and very big data seek to learn here is whether there are also cross-sectional differences important set to address an old and very basic question, in the way that banks with varying characteristics respond namely: how does monetary policy work? With an almost 20-year panel that includes quarterly to policy shocks. data on every insured commercial bank in the In particular, we ask whether the impact of United States—approximately 1 million bank- monetary policy on lending behavior is stronger for banks with less liquid balance sheets, where quarters in all—we are able to trace out the liquidity is measured by the ratio of securities to effects of monetary policy on the lending be- havior of individual banks. It is already well assets. It turns out that the answer is a resound- known that changes in the stance of monetary ing “yes.” Moreover, the result is largely driven policy are followed by significant movements in by the smaller banks, those in the bottom 95 bank lending volume (Ben S. percent of the size distribution. aggregate Bernanke and Alan S. Blinder, 1992); what we This empirical exercise is best motivated as a test of the so-called “bank lending view” of mon- etary transmission. At the heart of the lending view is the proposition that the Federal Reserve * Kashyap: Graduate School of Business, University of can, simply by conducting open-market opera- Chicago, 1101 East 58th Street, Chicago, IL 60637, Federal Reserve Bank of Chicago, and National Bureau of Eco- tions, shift banks’ loan supply schedules. For nomic Research; Stein: Sloan School of Management, Mas- example, according to the lending view, a contrac- sachusetts Institute of Technology, E52-434, 50 Memorial tion in reserves leads banks to reduce loan supply, Drive, Cambridge, MA 02142, and National Bureau of thereby raising the cost of capital to bank- Economic Research. This paper is a revision of our June 1997 NBER working paper entitled “What Do A Million dependent borrowers. Importantly, this effect is on Banks Have to Say About the Transmission of Monetary top of any increase in the interest rate on open- 1 Policy?” Research support was provided by the National market securities such as Treasury bills. Science Foundation and the Finance Research Center at The lending view hinges on a failure of the MIT. We are grateful for the generous assistance of the 2 Modigliani-Miller (M-M) proposition for banks. Federal Reserve Bank of Chicago, particularly Nancy An- drews and Pete Schneider, as well as for the comments and suggestions of two anonymous referees, and seminar par- ticipants at numerous institutions. Thanks also to Maureen 1 O’Donnell, Melissa Cunniffe, and Svetlana Sussman for See Kashyap and Stein (1994) for a detailed discussion help in preparing the manuscript. Finally, we are deeply as to why the debate over the lending channel is of practical and policy relevance. indebted to our team of research assistants—John Leusner, 2 Burt Porter, Brian Sack, and Fernando Avalos—for their The lending channel also requires: (1) some borrowers extraordinarily thoughtful and tireless work on this project. who cannot find perfect substitutes for bank loans; and (2) John was tragically killed in an accident on March 5, 1996, imperfect price adjustment. See Bernanke and Blinder and we dedicate this paper to his memory. (1988). 407

2 JUNE 2000 THE AMERICAN ECONOMIC REVIEW 408 When the Fed drains reserves from the system, it Blinder (1992) find that a monetary contraction is followed by a decline in aggregate bank lend- reservable compromises banks’ ability to raise ing. This is consistent with the lending view, but forms of finance, such as insured transaction de- non- also admits another interpretation: activity is posits. But it cannot constrain banks’ use of liabilities, such as large-denomination being depressed via standard interest-rate ef- reservable fects, and it is a decline in loan rather demand, CD’s. In an M-M world, banks are indifferent at that drives the results. In an the margin between issuing transactions deposits than loan supply, and large CDs, so shocks to the former do not effort to resolve this ambiguity, Kashyap et al. (1993) show that while a monetary contraction affect their lending decisions. All that monetary commercial reduces bank lending, it increases policy can do in this textbook setting is alter the paper volume. This fact would seem to suggest amounts of deposits (aka “money”) and CDs (aka 3 an inward shift in loan supply, rather than an “bonds”) outstanding. inward shift in loan demand. However, others So if there is to be an active lending channel, it have argued that it is not decisive either: per- must be that banks cannot frictionlessly tap unin- haps in recessions there is a compositional shift, sured sources of funds to make up for a Fed- induced shortfall in insured deposits. Stein (1998) with large firms faring better than small ones, and actually demanding more credit. Since most develops this argument, observing that many commercial paper is issued by large firms, this classes of bank liabilities which escape reserve 5 could explain the Kashyap et al. (1993) results. requirements are not covered by deposit insur- ance, and hence are potentially subject to adverse- Moving away from the aggregate data, a num- selection problems and the attendant credit ber of researchers have used micro data to test the rationing. For example, if there is adverse selec- cross-sectional implications of the lending view. tion in the market for large uninsured CDs, a bank One prediction is that tight money should pose a that loses a dollar of insured deposits will not raise special problem for small firms, which are more likely to be bank-dependent. And indeed, several a full dollar of new CD financing to offset this papers find that contractions in policy intensify loss. As a result, its lending is likely to decline. reservability Thus if there is a link between the liquidity constraints in the inventory and invest- 6 But again, while ment decisions of small firms. and of bank liabilities, the M-M the- insurability this is consistent with the lending view, there is orem can break down, and open-market opera- tions can matter for bank lending. another interpretation: what Bernanke and Gertler (1995) call a “balance sheet channel,” whereby These theoretical arguments notwithstanding, tight monetary policy weakens the creditworthi- the M-M benchmark might lead one to be skep- tical of the empirical importance of the lending ness of small firms, and hence reduces their ability any channel, particularly in the current, deregulated to raise funds from external provider, not just banks. environment where banks have a wide range of Our premise here is that, to make further nonreservable liability instruments at their dis- posal. And while a great deal of relevant evi- progress on this difficult identification problem, one has to examine lending behavior at the dence has been produced in the last several individual bank level. As discussed above, the years, it can be argued that previous studies have not completely overcome the fundamental theory ultimately rests on the idea that banks but very difficult problem of disentangling loan- cannot frictionlessly tap uninsured sources of funds to make up for a Fed-induced shortfall in supply effects from loan-demand effects. Con- sequently, the empirical case in support of a lending channel has not been viewed as airtight. 5 See Stephen Oliner and Glenn D. Rudebusch (1996). A quick literature review highlights the iden- the within Kashyap et al. (1996) rebut by noting that even 4 tification problems that arise. Bernanke and class of the largest firms, commercial paper rises relative to bank lending after a monetary contraction. Sydney Ludvig- son (1998) provides further evidence that financing “mix” 3 results like those of Kashyap et al. (1993) are not an artifact For an articulation of this M-M view, see Christina D. of compositional effects. Romer and David H. Romer (1990). 6 4 See, e.g., Gertler and Hubbard (1988), Robert E. For detailed surveys, see Kashyap and Stein (1994), Carpenter et al. (1994), Gertler and Simon Gilchrist (1994), Bernanke and Mark Gertler (1995), Stephen G. Cecchetti and Kashyap et al. (1994). (1995), and R. Glenn Hubbard (1995).

3 KASHYAP AND STEIN: THE TRANSMISSION OF MONETARY POLICY VOL. 90 NO. 3 409 insured deposits. But if this is true, then the to frictionlessly raise uninsured finance. This effect of monetary policy on lending should be leads us to our second hypothesis, which is that 3 more pronounced for some banks than for ≠ ≠ ≠ M B ≠ / L 0. Simply put, the . SIZE it t it it effect that we are interested in should be stron- others. gest for small banks. One would expect the Consider two small banks, both of which face limitations in raising uninsured external finance. largest banks to have an easier time raising uninsured finance, which would make their The banks are alike, except that one has a much more liquid balance sheet position than the lending less dependent on monetary-policy shocks, irrespective of their internal liquidity other. Now imagine that these banks are hit by 8 positions. a contractionary monetary shock, which causes The rest of the paper proceeds as follows. Sec- them both to lose insured deposits. In the ex- tion I describes our data set. Section II lays out the treme case where they cannot substitute at all baseline econometric specification, and discusses towards other forms of finance, the asset sides of their balance sheets must shrink. But the potential biases and other pitfalls. Section III pre- more liquid bank can relatively easily protect its sents our main results, and Section IV follows with a range of robustness checks. Section V loan portfolio, simply by drawing down on its assesses the quantitative importance of the results, large buffer stock of securities. In contrast, the and Section VI concludes. less liquid bank is likely to have to cut loans significantly, if it does not want to see its secu- I. Data Sources and Choice of Variables rities holdings sink to a dangerously low level. This logic leads to our first hypothesis: for Bank-Level Data A. banks without perfect access to uninsured 2 sources of finance, ≠ ≠ 0, where M B ≠ , L / it t it L Our sources for all bank-level variables are the is a bank-level measure of lending activity, it B Report of Condition and Income Consolidated is a measure of balance sheet strength, and it M (known as the Call Reports) that insured banks is a monetary-policy indicator (with higher t values of M submit to the Federal Reserve each quarter. With corresponding to easier policy). t This hypothesis exploits both cross-sectional the help of the staff of the Federal Reserve Bank of Chicago, we were able to compile a large data and time-series aspects of the data, and can be thought of in two ways, depending on the order set, with quarterly income statement and balance in which one takes the derivatives. Looking first sheet data for all reporting banks over the period at the cross-sectional derivative ≠ 1976Q1–1993Q2, a total of 961,530 bank- L — ≠ B / it it 9 quarters. which captures the degree to which lending is This data set presents a number of chal- —the hypoth- t liquidity constrained at any time lenges, particularly in terms of creating consistent esis is that these constraints are intensified dur- time series, as the definitions often change for ing periods of tight money. Alternatively, variables of interest. The Appendix describes ≠ the construction of our key series in detail, looking first at the time-series derivative L / it - and notes the various splices made in an effort to ≠ M —the sensitivity of lending volume to mon t 10 —the hypothesis is that i ensure consistency. etary policy for bank this sensitivity is greater for banks with weaker 7 balance sheets. 8 In testing this first hypothesis, we focus on - In Kashyap and Stein (1995), we test the related hy 2 pothesis that ≠ the smaller banks in our sample, based on the L 0: the lending of large , SIZE ≠ M ≠ / it t it banks should be less sensitive to monetary shocks than that idea that these banks are least likely to be able of small banks. Although the evidence strongly supports this hypothesis, there are alternative interpretations— e.g., large banks lend to large customers, whose loan demand is 7 - less cyclical. The tests we conduct below control for this, by Michael Gibson (1996) finds that the impact of mone have size classes. tary policy is stronger when banks in the aggregate focusing on differences in balance sheets within 9 lower securities holdings. His approach exploits the time- These data are available on the internet at: www. frbchi.org/rcri/rcri_database.html. series variation in bank balance sheets, while we use the 10 cross-sectional variation. Also somewhat related is John C. In addition to these splices, we also further cleaned the Driscoll (1997), who shows that state-level shocks to banks’ data set by eliminating any banks involved in a merger, for that quarter in which the merger occurs. deposits affect their loan supply.

4 JUNE 2000 THE AMERICAN ECONOMIC REVIEW 410 Table 1 examines balance sheets for banks of as a precaution, we have also reproduced our different sizes. There are two panels, corre- main results using the holding-company ap- 13 proach. As it turns out, nothing changes. sponding to the starting and ending points of In terms of the specific variables required for our sample. In each, we report data on six our regressions, we make the following choices. size categories: banks below the 75th percentile L First, for the lending volume variable by asset size; banks between the 75th and 90th ,weuse it percentiles; banks between the 90th and both total loans, as well as at the most commonly 95th percentiles; banks between the 95th and studied subcategory, commercial and industrial 98th percentiles; banks between the 98th and (C&I) loans. One reason for examining both is the 99th percentiles; and banks above the 99th concern that any results for total loans might be percentile. influenced by compositional effects. For example, Whether one looks at the data from 1976 or it may be that C&I loan demand and real estate loan demand move differently over the business 1993, several patterns emerge. On the asset side, cycle. If, in addition, banks that tend to engage small banks hold more in the way of securities, 11 primarily in C&I lending have systematically dif- This is what one would and make fewer loans. B ferent levels of liquidity expect to the extent that small banks have more than banks that tend to it specialize in real estate lending, this could bias our trouble raising external finance: they need bigger 2 14 estimates of ≠ A countervailing buffer stocks. On the liability side, the smallest ≠ / L B ≠ . M it it t drawback of focusing on just C&I loans is that banks have a very simple capital structure—they are financed almost exclusively with deposits and some banks do only a negligible amount of C&I 15 Thus in the regressions that use C&I business. common equity. In contrast, the larger banks make less use of both deposits and equity, with the lending, we omit any banks for which the ratio of difference made up by a number of other forms of C&I to total loans is less than 5 percent. This screen leads us to drop approximately 7 percent of borrowing. For example, the largest 2 percent of 16 banks make heavy use of the federal funds market our sample. to finance themselves; the smallest banks do vir- For the balance sheet variable B , we use the it tually no borrowing in the funds market. Given ratio of securities plus federal funds sold to total 17 that federal funds are unsecured borrowing, this The intuition is as described above: banks assets. again fits with the existence of financing frictions: with large values of this ratio should be better able small banks are less able to use instruments where to buffer their lending activity against shocks in the availability of external finance, by drawing on credit risk is an issue. The numbers in Table 1, as well as the base- line regression results below, reflect balance 13 An al- individual bank level. sheet data at the It should not be too surprising that the results are robust in this way, since the vast majority of all banks are ternative approach would be to aggregate the stand-alones, and even large holding companies are typi- balance sheets of all banks that belong to a cally dominated by a single bank. See Allen N. Berger et al. single bank holding company. This latter ap- (1995). 14 proach makes more sense to the extent that bank - This is just a specific version of the general proposi tion that B might be endogenously linked to the cyclical holding companies freely shift resources among it sensitivity of loan demand. We discuss this issue in Section the banks they control as if there were no II, subsection B2, below, and argue that the bias is likely to 12 A priori, it is not obvious which is boundaries. make our tests with total loans too conservative. 15 the conceptually more appropriate method, so This problem is even more pronounced with other subcategories, e.g., agricultural loans. 16 Even after this filter, there are some extreme values of 11 loan growth in our sample. To ensure that our results are not In Panel A, for 1976Q1, we report data for domestic driven by these outliers—which could be data errors—we loans only. This is because prior to 1978, figures for inter- national loans are not available, although such loans implic- drop any further observations for which loan growth is more itly show up in total bank assets. Consequently, for the very than five standard deviations from its period mean. How- largest category of banks—the only ones with significant ever, our results are not sensitive to either of these screens. 17 international activities—we are understating the true ratio of We do not include cash in the numerator, because we suspect that cash holdings largely reflect required reserves, loans to assets in 1976. 12 which cannot be freely drawn down. However, our results Joel Houston et al. (1997) present evidence that shocks to one bank in a holding company are in fact par- are very similar if cash is added to our measure of B . See it the NBER working paper version (1997) for details. tially transmitted to others in the same holding company.

5 KASHYAP AND STEIN: THE TRANSMISSION OF MONETARY POLICY VOL. 90 NO. 3 411 ALANCE S HEETS FOR B ANKS OF D IFFERENT S IZES T 1—B ABLE Panel A: Composition of Bank Balance Sheets as of 1976 Q1 Between Between Between Between Above 75th and 90th and Below 98th and 95th and 99th 99th 90th 98th 75th 95th percentile percentile percentile percentile percentile percentile 431 144 2,157 10,784 Number of banks 719 144 1,341.45 247.73 119.14 32.82 Mean assets (1993 $ millions) 10,763.44 556.61 3,964.55 Median assets (1993 $ millions) 28.43 1,228.66 508.06 239.00 112.63 0.56 Fraction of total system assets 0.13 0.09 0.06 0.09 0.07 Fraction of total assets in size category 0.13 0.22 0.09 0.09 0.10 Cash 0.12 0.29 0.15 Securities 0.34 0.33 0.32 0.27 0.04 0.03 0.04 0.04 0.05 Federal funds lent 0.05 0.53 0.54 0.41 Total domestic loans 0.52 0.53 0.53 0.18 0.09 0.20 Real estate loans 0.17 0.19 0.17 0.16 0.17 0.17 C & I loans 0.10 0.13 0.15 0.15 0.14 0.16 0.15 Loans to individuals 0.15 0.06 0.84 0.87 0.81 0.89 0.90 0.90 Total deposits Demand deposits 0.25 0.31 0.30 0.30 0.31 0.33 0.55 0.51 0.33 Time and savings deposits 0.59 0.60 0.59 0.10 0.12 0.14 0.14 0.16 Time deposits . $100K 0.07 0.04 0.08 0.02 0.07 Federal funds borrowed 0.00 0.01 0.00 0.01 0.01 Subordinated debt 0.00 0.00 0.00 0.01 0.02 0.06 0.01 0.01 0.01 Other liabilities 0.07 0.07 0.07 0.05 Equity 0.08 0.08 Panel B: Composition of Bank Balance Sheets as of 1993 Q2 Between Between Between Between 90th and Below 98th and Above 95th and 75th and 99th 95th 99th 75th 90th 98th percentile percentile percentile percentile percentile percentile 560 336 113 8,404 Number of banks 1,681 112 1,072.57 165.81 44.42 Mean assets (1993 $ millions) 3,366.01 380.14 17,413.41 3,246.33 920.78 9,297.70 Median assets (1993 $ millions) 38.59 362.75 155.73 0.11 Fraction of total system assets 0.10 0.08 0.06 0.10 0.55 Fraction of total assets in size category 0.07 0.07 Cash 0.05 0.05 0.05 0.09 0.29 0.25 0.22 Securities 0.34 0.32 0.27 0.04 0.04 0.03 0.04 0.04 0.04 Federal funds lent Total loans 0.60 0.59 0.53 0.56 0.60 0.59 0.34 0.30 0.25 0.21 Real estate loans 0.30 0.33 0.11 0.13 0.10 0.18 C & I loans 0.09 0.12 0.12 0.09 0.10 Loans to individuals 0.14 0.10 0.17 0.79 0.69 0.85 0.76 Total deposits 0.88 0.87 0.24 0.26 0.19 Transaction deposits 0.26 0.26 0.25 0.25 0.24 0.21 0.22 0.21 0.17 Large deposits 0.01 0.02 0.02 Brokered deposits 0.00 0.00 0.01 0.02 0.06 0.10 0.09 Federal funds borrowed 0.01 0.04 0.00 0.00 0.02 0.00 0.00 Subordinated debt 0.00 0.03 0.05 0.06 Other liabilities 0.01 0.02 0.13 0.09 0.08 0.09 0.08 0.07 0.10 Equity

6 JUNE 2000 THE AMERICAN ECONOMIC REVIEW 412 their stock of liquid assets. Of course, as in all of reading of FOMC documents, John Boschen and Leonard Mills each month rate Fed policy as the liquidity-constraints literature, we must be being in one of five categories: “strongly ex- - B aware that is an endogenous variable. We dis it pansionary,” “mildly expansionary,” “neutral,” cuss the potential biases this might cause, as well as our approach to controlling for these biases, “mildly contractionary,” and “strongly contrac- below. tionary,” depending on the relative weights that Finally, we need to decide on cutoffs in they perceive the Fed is putting on inflation versus 21 order to assign banks to size categories. Be- We code these policy stances as unemployment. 2 cause of the extremely skewed nature of the 2 2, 1, 0, 2 respectively. 1, and Our second measure is the federal funds rate, size distribution, an overwhelming majority which has been advocated by Robert Laurent of the banks in our sample are what anyone would term “small,” by any standard. (Recall (1988), Bernanke and Blinder (1992), and Marvin from Table 1 above that even banks between Goodfriend (1993). However, it should be noted that as the Fed’s operating procedures have varied the 90th and 95th percentiles have average 18 In assets of below $400 million in 1993.) over time, so too has the adequacy of the funds rate as an indicator. Both conventional wisdom as the end, we choose to use three categories: the well as the formal statistical analysis of Bernanke smallest one encompasses all banks with total and Mihov (1998) suggests that the funds rate may assets below the 95th percentile; the middle one includes banks from the 95th to 99th be particularly inappropriate during the high- percentiles, and the largest one has those volatility Volcker period, which fits within the 19 banks above the 99th percentile. first half of our sample period. Motivated by this observation, we also work B. Measures of Monetary Policy with a third measure of monetary policy, that developed by Bernanke and Mihov (1998). A prerequisite for all our tests is a good They construct a flexible VAR model that nests previous VARs based on more specific assump- indicator of the stance of monetary policy M . t tions about Fed operating procedures—i.e., Unfortunately, there is no consensus on this their model contains as special cases either topic—indeed, a whole host of different indica- funds-rate targeting (Bernanke and Blinder, tors have been proposed in the recent litera- 20 1992) or procedures based on nonborrowed re- ture. Therefore, rather than trying to argue for serves (Christiano and Martin Eichenbaum, a single best measure, we use three different 1992; Steven Strongin, 1995). The Bernanke- ones throughout. While not an exhaustive list, Mihov methodology can be used to calculate these three do span the various broad types of either high-frequency monetary-policy methodologies that have been employed. shocks, or an indicator of the overall stance of policy. Our first measure, which represents the “narra- We focus on the latter construct, as it is more tive approach” to measuring monetary policy, is 22 the Boschen-Mills (1995) index. Based on their appropriate for the hypotheses we are testing. The series we use is exactly that shown in Figure III of their paper (p. 899). 18 To get an idea of how small a $400-million bank is, note that regulations restrict banks from having more than 15 percent of their equity in a single loan. Thus a bank with 21 $400 million in assets and a 6-percent equity ratio cannot The other well-known indicator in this vein is the make a loan of more than $3.6 million to a single borrower. so-called “Romer date” variable (Romer and Romer, 1989). 19 However, there are only three Romer dates in our sample, - We experimented with further subdividing the small and two—August 1978 and October 1979 —are so close est category— e.g., looking only at those banks below the 75th percentile— but did not discern any differences that they are not completely independent observations. Moreover, their zero-one nature further limits the informa- amongst the subcategories. We also tried using an expanded definition of the largest category—all banks above the 98th tion in the series. The Boschen-Mills index, which embod- percentile— but this also made no significant difference to ies a finer measure of the stance of policy, is more appropriate for the high-frequency experiment we are con- our results. 20 ducting. See Bernanke and Ilian Mihov (1998) and Lawrence 22 Christiano et al. (2000) for a recent discussion of the liter- Even if a contraction in policy is partially anticipated ature on measuring monetary policy and for further refer- by banks, it should still have the cross-sectional effects that we hypothesize. ences.

7 KASHYAP AND STEIN: THE TRANSMISSION OF MONETARY POLICY 413 VOL. 90 NO. 3 OLICY P F M EASURES OF 1. M IGURE ONETARY Figure 1 plots the three measures in levels. tions. In general, all of the correlations appear to (Throughout the paper, we invert the funds rate be lower—in many cases substantially so—in the first part of the sample, which we date as for comparability with the other two measures.) running from 1976Q1 to 1985Q4. For example, As can be seen, they all seem to contain broadly similar information. All three indicate that mon- the correlation of quarterly changes in the Boschen-Mills and Bernanke-Mihov indicators etary policy was very tight following the Fed’s is only 0.02 in the first part of the sample, but change in operating procedures in October 1979; all three suggest a relatively loose stance rises to 0.36 in the second part. Apparently, given the enormous volatility during the Vol- of policy in the period 1985–1986; and all three capture the common wisdom that policy was cker period, it is harder to get an unambiguous reading of the stance of monetary policy, even if tightened again in 1988, before being eased once more beginning in late 1989. one uses measures other than the funds rate. In Table 2 documents the statistical correlations light of this concern, we check below to see how our results hold up across subperiods; one among the three measures. Overall, the numbers might expect a priori that they would be more confirm the visual impressions from Figure 1, with some qualifications. In levels, the pair- clear-cut and consistent in the more recent data. wise correlations are all moderately high— II. Econometric Specification between 0.58 and 0.71— over the full sample. The lowest of these correlations is that between the funds rate and the Bernanke-Mihov mea- A. The Two-Step Regression Approach sure. However, this correlation remains rela- Again, our basic goal is to measure the quan- tively stable when we look at annual and 2 ≠ tity quarterly changes. In contrast, the correlation of L ≠ M , for banks in different size / ≠ B it t it classes. In doing so, one important choice is the Boschen-Mills index with the other two how tightly to parametrize our model. As a measures is much reduced when we look at baseline, we opt for a flexible specification, higher-frequency changes. This is due to the which we implement with a two-step procedure. discrete nature of the Boschen-Mills index, In the first step, we run the following cross- which at higher frequencies effectively intro- sectional regression duces measurement error into this indicator of for each size separately : the log change in t class and each time period monetary policy. The table also looks at subsample correla- L B against (i) four lags of itself; (ii) ; and it 1 2 it

8 JUNE 2000 414 THE AMERICAN ECONOMIC REVIEW T 2—C EASURES ORRELATIONS OF ABLE M the estimated coefficient on B , which we 1 it 2 OF M ONETARY OLICY P - b denote by . As discussed earlier, this coeffi t cient can be thought of as a measure of the Correlation of: intensity of liquidity constraints in a given size Quarterly Annual class at time t . Levels changes changes In the second step of our procedure, we take A. Full sample (76Q1–93Q2) b for each size class the ’s, and use them as the t 0.382 0.219 1. Boschen-Mills/Federal 0.608 dependent variable in a purely time-series re- funds gression. We consider two variants of this time- 0.416 2. Boschen-Mills/ 0.710 0.099 series regression. In the first, “univariate” Bernanke-Mihov 0.483 0.486 0.580 specification, the right-hand-side variables in- 3. Federal funds/ Bernanke-Mihov clude: (i) the contemporaneous value and four B. 1st-half sample (76Q1–85Q4) lags of the change in the monetary measure M , t 0.318 0.233 1. Boschen-Mills/Federal 0.514 24 as well as (ii) a linear time trend: funds 0.018 2. Boschen-Mills/ 0.293 0.665 Bernanke-Mihov 4 3. Federal funds/ 0.526 0.476 0.471 1 u b (2) 5 h f D M 1 d TIME 1 . O j t t 2 j t t Bernanke-Mihov 5 j 0 C. 2nd-half sample (86Q1–93Q2) 0.414 0.647 1. Boschen-Mills/Federal 0.733 funds 2. Boschen-Mills/ 0.361 0.844 0.734 In the second, “bivariate” specification, we also Bernanke-Mihov add the contemporaneous value and four lags of 0.730 0.567 3. Federal funds/ 0.871 25 real GDP growth to the right-hand side: Bernanke-Mihov Annual changes are defined as the change between Notes: 4 the level of a variable in a certain quarter and the level four (3) M D 1 h 5 f b O j t t 2 j quarters before that. The sign of the federal funds rate has j 5 0 been inverted to preserve the convention in the paper that a higher level of the monetary-policy measure reflects a looser policy. 4 GDP D g 1 O j t 2 j j 5 0 (iii) a Federal Reserve-district dummy variable 23 That is, we esti- (i.e., a geographic control). u . TIME d 1 1 t t mate: L ~ log (1) ! D In either case, our hypothesis is that, for the it smallest class of banks, an expansionary im- 4 M pulse to should lead to a reduction in b — t t i.e., the sum of the ’s should be negative. f b ! a B D log 1 L 5 ~ O t tj it it 2 j 2 1 As an alternative to this method, we also try j 5 1 in Section IV, subsection A, below a more 12 1 C 1 ́ . FRB O ik it kt 24 The time trend turns out to be borderline significant in k 5 1 some cases, and insignificant in others. If it is deleted from the specification, nothing changes significantly. We discuss The key item of interest from this regression is one potential economic interpretation of the time trend below. 25 We also experimented with including four lags of the 23 dependent variable b For the smallest size class, we also tried replacing the to the right-hand side. However, t conditional on the real GDP lags being already in the Federal Reserve-district dummies with state-level dummies, regression, this adds nothing further—the lagged dependent to get a tighter geographic control. This made no difference. variables are always insignificant, and have no substantive Nor did using more complex lag specifications, including, impact on any of the other coefficient estimates. e.g., quadratic lagged-lending terms.

9 KASHYAP AND STEIN: THE TRANSMISSION OF MONETARY POLICY VOL. 90 NO. 3 415 tightly parametrized one-step, interactive spec- b Biases in the Level of 1. .—First and most t L obviously, the first-step regression delivers es- ification, where we run the change in it B timates of the against: (i) of b level that are potentially ; (ii) the change in M ; and (iii) t 2 1 t it B biased. In principle, this bias could be either . In this M interacted with the change in it t 2 1 case, the tests center on the interaction coeffi- positive or negative, but in a banking context, a natural story goes as follows. Because of demo- cients. What distinguishes the two-step ap- allows for a different macro proach is that it graphic factors, some banks have an advantage shock in each period for each Federal Reserve at deposit taking, but few good lending oppor- tunities. Rather than make bad loans, these This makes it harder to explain away district. banks have portfolios that are skewed towards our results based on unobserved loan-demand variability. For example, the two-step specifica- securities. If the weak lending opportunities are only imperfectly controlled for by past loan tion prevents us from taking credit for any de- cline in lending that is common to all banks in growth, there may be a tendency for high values of B the Chicago district in a given quarter, even if to be associated with slow growth of it 1 2 L all these banks have similarly weak balance —i.e., b will be biased downward. it t However, the key point to note is that biases in sheets. As will become clear, the trade-off rel- b ative to the one-step method is that this poten- of level the are in and of themselves not an t issue, since our hypothesis centers on the tially sacrifices a great deal of statistical power. corre- b lation One control that we do not adopt is a bank- of . Indeed, if the with M variation in only t t B level fixed effect. There are two reasons for this. across banks arose from the specific link it sketched above—that some banks have fewer First, we would lose much of the variation in lending opportunities and hence hold more in se- our explanatory variable— 67 percent of the B total variation in curities—there would be no reason to expect a is eliminated by bank fixed it 26 Second, we worry that the remaining spurious correlation between b effects. and , and our M t t B tests would be wholly uncontaminated. within-bank variation in is contaminated by it the kind of endogeneity that is most difficult to 27 Biases in the Correlation of This is not to say that there are no b 2. address. . and M t t endogeneity issues with respect to across-bank —Unfortunately, there may be other endogenous variation in B B influences on that are more problematic, in that , but as we argue momentarily, it it these can be dealt with to some degree. coefficients f they lead to a bias in the estimated M on in the second-step regression. Generally t Potential Biases and Other Pitfalls B. speaking, this happens when there is an endoge- B nous link between cyclical sensitivity and the it Before turning to the results, we highlight a of loan demand. In principle, the bias can go either number of issues that could pose problems. The way. First, consider what might be called the “heterogeneous risk aversion” story, wherein cer- single biggest source of concern is that in our first-step regression—like in all of the liquidity- tain banks are inherently more conservative than constraints literature—we use an endogenous others. Conservative banks tend to protect them- B right-hand-side variable in selves both by having larger values of B - , as well . This endogene it it as by shunning cyclically sensitive customers— ity can take a number of different forms, some of which are more troubling for us than others. i.e., there is a negative correlation between B and it the cyclical sensitivity of loan demand. This can lead to a bias in which the estimated effect of M t 26 - This is after accounting for the time/geographic dum b on is too negative. Thus we may be biased t mies. towards being too aggressive, rejecting the null 27 For example, consider a bank with a that is only 20 B it hypothesis even when it is true. t percent at time 2 1, but that spikes up to 25 percent at Alternatively, consider the “rational buffer- time t . A fixed-effects model would deem the bank unusu- B t ally liquid at time (although its value of is still lower stocking” story, in which all banks have the it than most banks’). But the shock may just reflect a surge in same risk aversion, but some have more oppor- bank profits due to improved borrower performance. So if tunities to lend to cyclically sensitive customers we now see the bank lending more, it would be wrong to than others. In this case, those banks with more credit a strong balance sheet—rather it may just be an increase in loan demand. cyclically sensitive customers will rationally

10 JUNE 2000 THE AMERICAN ECONOMIC REVIEW 416 choose to insulate themselves against the instrumental-variables procedure to purge it B from our estimates. Unfortunately, to do this greater risk by having higher values of . it properly requires creating an instrument for B Now the direction of the bias is reversed—there it will be a positive influence on our key that is uncorrelated with loan cyclicality—a dif- ficult task. Still, we can at least make a partial coefficients—and we will tend to be too conser- vative, failing to reject the null hypothesis even effort, by regressing B against any plausible it observable measures of loan cyclicality, and when it is false. using the residuals from this regression as our A priori, the latter story strikes us as more instruments. For example, it seems reasonable plausible, in that it can be easily told within the 28 to posit that some categories of loans are on context of a fully rational model. Nonetheless, average more cyclically sensitive than others. In it is obviously important for us to ascertain which of the stories is of more relevance in the this spirit, we can regress a bank’s B against its it data. Fortunately, there are a couple of distinct ratio of C&I to total loans, its ratio of mortgages ways to do so. The first emerges out of the to total loans, etc., and use the residuals as bivariate version of the second-step regression. instruments. The results from these “quasi-IV” tests are virtually identical to those from the If the heterogeneous risk aversion story is true, 29 baseline specifications that we report below. the g coefficients on GDP growth should be negative. In contrast, if the rational buffer- 3. Disentangling the Direct Effects of Mon- ’s should be positive. g stocking story is true, the —While etary Policy vs. Bank Capital Shocks. The intuition is straightforward. Under the het- our focus is on the narrow question of how erogeneous risk-aversion story, an increase in GDP favors riskier borrowers, who are affiliated open-market operations work, there are other mechanisms that can generate similar effects on with less conservative banks, who in turn have bank lending. In particular, a growing literature B lower values of . Thus an increase in GDP it argues that lending will be constrained by has a more positive impact on the lending of B low- - banks’ equity capital, which in turn can be banks, which implies a negative coeffi it cient in a regression of b impacted by a wide variety of shocks— changes on GDP. t 30 As a second method of deducing the direction in interest rates, real estate values, etc. From the perspective of this literature, one caveat is of the bias, one can look to the results for the largest banks. In the limiting case where there that our results may not be capturing the work- ings of the lending channel, but rather an indi- are no capital-market frictions facing these f coefficients on rect capital-shock effect. According to this banks, any nonzero M in the t story, tight money simply raises rates and sup- second-step regression must reflect the direction ’s for the largest banks are of the bias. If the presses economic activity, causing banks to ex- f negative, this supports the heterogeneous risk- perience loan losses and reductions in capital. This in turn leads weaker banks to cut back on aversion story, while if they are positive, this new lending. favors the rational buffer-stocking story. Thus, Fortunately, it is possible to disentangle the two while it may seem counterintuitive, the evi- alternatives. The capital-shock story implies two dence will be more strongly in favor of our on the opposite signs hypothesis if we get the f ’s for large and small banks. As will be seen 29 shortly, both pieces of evidence point to the For the sake of brevity, we do not tabulate the results of the quasi-IV regressions here, but they can be found in rational buffer-stocking story. So if anything, the NBER working paper version (Kashyap and Stein, our tests for the small banks are probably biased 1997). towards being too conservative. 30 This literature shares with our work the broad theme Ideally, in addition to just figuring out the that banks face costs of external finance, but the emphasis is direction of the bias, we would also devise an on frictions in the equity market, as opposed to the market for uninsured bank debt. See Bengt Holmstro ̈ m and Jean Tirole (1997) for a model; Katherine Samolyk (1994), Joe Peek and Eric Rosengren (1995, 1997), Houston et al. 28 (1997), and Ruby P. Kishan and Timothy P. Opelia (2000) Although the heterogeneous risk-aversion story might be justified by appealing to agency effects that vary in for examples of recent empirical work; Steven A. Sharpe (1995) for a survey. strength across banks.

11 KASHYAP AND STEIN: THE TRANSMISSION OF MONETARY POLICY 417 VOL. 90 NO. 3 T QUATIONS E STIMATION OF (1), (2), TEP -S WO 3—T ABLE E predictions about the bivariate version of the OEFFICIENTS ON (3): S AND C UM OF second-step regressions. First, adding GDP OLICY M ONETARY NDICATOR I -P growth (or any proxy for activity) should diminish - the importance of the monetary measure M . Sec Panel A: C&I Loans t coefficients on GDP growth should be g ond, the Univariate Bivariate negative. As will be seen, neither prediction is 1. Boschen-Mills borne out, suggesting that our results are not 0.0438 2 , 95 2 0.0131 31 driven by capital-shock effects. (0.0188) (0.0187) 0.0339 2 95–99 0.0094 (0.0303) (0.0401) III. Baseline Results . 99 0.0960 0.1411 (0.0428) (0.0661) 2 0.1398 2 0.1542 Small-Big Tables 3 and 4 present the results of our (0.0449) (0.0611) second-step regressions. Table 3 gives a com- 2. Funds rate pact overview of all the specifications, showing 0.0267 2 0.0151 95 , 2 (0.0089) (0.0071) only one number (with the associated standard 2 95–99 0.0097 0.0066 f error) from each regression: the sum of the (0.0112) (0.0137) coefficients on the relevant monetary indicator. 99 0.0795 0.1175 . (0.0281) (0.0314) The table is divided into two panels: Panel A for Small–Big 2 0.1327 0.1062 2 C&I loans, and Panel B for total loans. In each (0.0296) (0.0376) panel, there are 12 test statistics. First, we test 3. Bernanke-Mihov 2 1.8633 2 95 , 0.5269 f ’s is negative six ways whether the sum of the (1.2463) (1.0933) for the “small” banks—those in the bottom 95 95–99 3.3461 0.7345 percent of the size distribution. The six tests (2.1119) (2.1853) 4.7862 7.5911 . 99 correspond to our univariate and bivariate spec- (2.3927) (3.5220) ifications for each of the three monetary indica- 6.6495 2 Small–Big 8.1181 2 (3.3966) (3.0215) tors. Second, we test in the same six ways ’s is lower for the f whether the sum of the small banks than for the “big” banks—those in Panel B: Total Loans the top 1 percent of the size distribution. Univariate Bivariate As can be seen from Panel A of Table 3, the 1. Boschen-Mills overall results for C&I loans are strong. Con- 0.0179 , 95 2 0.0044 2 sider first the results for the small banks. In all (0.0120) (0.0110) 2 95–99 0.0167 0.0129 six cases, the point estimates are negative, con- (0.0118) (0.0236) sistent with the theory. Moreover, in two of six 99 0.0516 0.0921 . cases, the estimates are significant at the 2.0- (0.0522) (0.0373) Small–Big 0.0965 2 0.0695 2 percent level or better; in two others, the stan- (0.0348) (0.0464) -values that are around 9.0 dard errors imply p 2. Funds rate 32 percent. , 2 2 95 0.0088 0.0046 (0.0037) (0.0049) Next, turn to the small-bank/big-bank dif- 95–99 2 0.0040 2 0.0126 ferentials. In every case, the estimate for the (0.0079) (0.0060) so that these differen- positive, big banks is . 0.0460 0.0258 99 (0.0152) (0.0188) tials are in absolute value than the larger 0.0506 2 0.0346 2 Small–Big (0.0182) (0.0174) 3. Bernanke-Mihov 31 We are not claiming that bank capital does not affect 2 95 , 0.1926 0.7827 lending— only that it does not explain away our results. (0.5344) (0.5780) Indeed, the positive time trend in b that shows up in some 1.1191 95–99 2 0.2849 t (0.7766) (1.1178) regressions may reflect the well-documented bank-capital . 99 3.6558 6.7373 problems of the late 1980’s and early 1990’s. 32 (1.4636) (2.5209) In all cases, the tables display robust standard errors 2 3.8484 2 5.9545 Small–Big that account for heteroskedasticity and serial correlation. (2.2180) (1.5971) Moreover, when comparing the small and big bank esti- mates, the standard errors also account for the correlation of Standard errors are in parentheses. Note: the residuals across these two equations.

12 JUNE 2000 THE AMERICAN ECONOMIC REVIEW 418 D 4—T TEP E STIMATION OF E QUATIONS (1), (2), AND (3): F ULL -S ETAILS WO T ABLE Panel A: Money Measure: Change in Boschen-Mills Monetary-policy indicator Change in GDP 0123401234 C&I loans Univariate , 0.0074 2 0.0066 2 0.0138 2 0.0137 2 0.0023 95 2 2 (0.0048) (0.0056) (0.0074) (0.0041) (0.0044) ( R 0.1357) 5 0.0088 2 0.0172 2 0.0226 2 0.0139 0.0111 95–99 2 (0.0139) (0.0132) (0.0211) (0.0132) (0.0140) ( R 5 0.1259) 2 0.0193 0.0105 0.0329 99 0.0002 0.0722 . 2 2 (0.0114) (0.0220) (0.0315) (0.0205) (0.0189) R ( 0.0988) 5 Bivariate 95 2 0.0037 2 0.0008 2 0.0064 2 0.0054 0.0031 0.6259 0.2955 0.7707 0.9165 0.2920 , 2 (0.2665) (0.3831) (0.3127) (0.2489) (0.0057) (0.0046) (0.0033) (0.0076) (0.0047) (0.5292) ( R 0.3404) 5 0.9987 0.0151 2 0.0149 2 0.0033 0.0213 0.0088 0.7424 0.6632 0.5668 1.2774 95–99 2 2 (0.0177) (1.2009) (0.8997) (0.0133) (0.0148) (0.6125) (0.0107) (1.0185) (0.5464) (0.0133) R ( 0.2086) 5 1.2170 99 0.0141 0.0100 0.0317 . 0.0910 2 0.8454 2 3.8951 4.2036 3.1755 0.0225 2 (2.4012) (2.8855) (1.0001) (1.1834) (0.0088) (0.0175) (0.0201) (0.0146) (0.0185) (1.4284) ( R 0.2589) 5 Panel B: Money Measure: Change in Federal Funds Rate Monetary-policy indicator Change in GDP C&I loans 0123401234 Univariate 95 2 0.0069 2 0.0077 2 0.0062 2 0.0054 2 0.0005 , 2 (0.0012) (0.0012) (0.0020) (0.0022) (0.0019) ( R 0.2868) 5 2 0.0018 2 95–99 2 0.0027 2 0.0056 0.0065 0.0030 2 (0.0013) (0.0054) (0.0021) (0.0038) (0.0058) R ( 0.0834) 5 . 99 0.0118 0.0140 0.0235 0.0142 0.0161 2 (0.0063) (0.0075) (0.0042) (0.0075) (0.0082) R ( 5 0.0958) Bivariate 0.0024 0.0053 2 0.0059 2 0.0040 2 0.0023 2 0.5664 2 0.0587 0.4089 1.0498 0.5039 , 95 2 (0.0021) (0.3866) (0.2334) (0.0014) (0.4193) (0.0024) (0.0026) (0.0021) (0.5115) (0.3923) ( R 5 0.4526) 0.7481 0.0015 0.0002 2 0.0036 95–99 1.0847 0.0006 2 0.0294 1.3017 1.2965 0.0110 2 (0.8977) (0.0045) (0.9570) (1.1979) (0.0034) (0.6423) (0.0052) (0.0063) (0.0028) (0.7850) ( R 0.1870) 5 2 99 0.0113 0.0236 0.0250 . 1.2667 0.0146 0.7602 2 5.1621 7.5245 3.3015 0.0431 2 (2.1000) (0.0103) (3.4558) (1.6353) (0.0068) (0.0051) (1.6629) (0.0093) (0.0133) (1.8068) R ( 0.3442) 5 Panel C: Money Measure: Change in Bernanke-Mihov Monetary-policy indicator Change in GDP C&I loans 0123401234 Univariate , 95 2 0.0237 0.2379 2 0.2093 2 1.2548 2 0.6135 2 (0.4173) (0.3317) (0.2980) (0.3094) (0.3736) R ( 5 0.1440) 95–99 0.9354 2 0.3920 2 1.5933 0.6255 1.1588 2 (0.5546) (0.6414) (0.9313) (1.0750) (0.5074) ( R 5 0.0887) 99 0.0938 0.5484 0.6991 1.7404 1.7045 . 2 (1.5339) (1.7450) (1.3596) (1.0018) (1.0032) R ( 5 0.0309) Bivariate 95 0.1069 0.1053 2 0.1351 2 0.6614 0.0574 0.6555 0.1692 0.8708 0.8861 0.3600 , 2 (0.2782) (0.6661) (0.3316) (0.2609) (0.4119) (0.2394) (0.3824) (0.2422) (0.3275) (0.2957) ( R 0.3535) 5 95–99 1.4562 0.8915 2 0.1876 2 0.6774 1.8634 1.5949 0.5113 1.3196 0.6149 1.089 2 (0.6306) (0.5117) (0.8721) (1.3575) (0.4651) (0.8073) (0.8733) (0.8566) (0.5309) (0.4627) R ( 0.2043) 5 1.8398 4.2408 3.8538 2 0.8679 . 99 1.1890 0.7249 2 0.3289 2.6111 3.3949 2.5876 2 (0.9098) (2.7145) (1.8625) (0.9044) (1.2244) (1.0684) (1.0878) (1.2061) (3.9985) (1.8175) R ( 5 0.1781) corresponding figures for the small banks in the 5.0-percent level or better; indeed, four of -values are well below 1.0 percent. the isolation. Moreover, each of the six small- p bank/big-bank differentials is significant at These results help us begin to discriminate

13 KASHYAP AND STEIN: THE TRANSMISSION OF MONETARY POLICY VOL. 90 NO. 3 419 4 ABLE —Continued. T Panel D: Money Measure: Change in Boschen-Mills Change in GDP Monetary-policy indicator 0123401234 Total loans Univariate 0.0045 2 0.0124 0.0023 0.0012 2 0.0045 , 95 2 2 (0.0040) (0.0033) (0.0021) (0.0035) (0.0044) R ( 5 0.2346) 2 0.0066 95–99 0.0005 2 0.0121 0.0028 0.0024 2 (0.0064) (0.0046) (0.0090) (0.0073) (0.0049) R ( 0.0692) 5 . 0.0238 0.0121 0.0213 0.0202 0.0218 2 99 2 (0.0118) (0.0077) (0.0210) (0.0155) (0.0140) ( R 5 0.1208) Bivariate 0.0020 0.0024 2 0.0099 0.0049 0.0051 2 2 0.3639 0.2310 0.0841 0.4924 0.1157 , 95 2 (0.1818) (0.3089) (0.3651) (0.0036) (0.0040) (0.0022) (0.0044) (0.0042) (0.2315) (0.2037) R ( 0.3216) 5 0.0068 0.0019 0.0087 0.0070 2 0.0077 0.2406 1.4108 0.2038 0.9510 2 0.2656 95–99 2 (0.3785) (0.4722) (0.0055) (0.0033) (0.5489) (0.0047) (0.0037) (0.2614) (0.2796) (0.0064) R ( 0.2847) 5 2 0.0182 0.0153 0.0244 0.0364 0.0341 99 0.6884 2 1.7162 2.6482 1.6745 . 1.1544 2 (1.0135) (0.8612) (1.6450) (0.0115) (0.0095) (0.0153) (0.0132) (0.0110) (0.9052) (0.5291) ( R 5 0.2880) Panel E: Money Measure: Change in Federal Funds Rate Monetary-policy indicator Total loans Change in GDP 0 123401234 Univariate 95 2 0.0033 2 0.0032 2 0.0015 2 0.0001 2 0.0006 , 2 (0.0010) (0.0009) (0.0011) (0.0013) (0.0011) ( R 0.1607) 5 2 0.0042 2 0.0038 2 95–99 2 0.0018 2 0.0005 0.0023 2 (0.0016) (0.0021) (0.0025) (0.0016) (0.0018) R ( 0.0769) 5 . 99 2 0.0041 0.0044 0.0102 0.0068 0.0085 2 (0.0037) (0.0037) (0.0036) (0.0045) (0.0056) ( R 0.1084) 5 Bivariate 0.0758 0.3422 0.0020 0.1688 0.3096 , 95 2 0.0025 2 0.1830 2 0.0004 0.0003 0.0000 2 (0.0015) (0.3703) (0.0012) (0.0017) (0.0015) (0.4556) (0.0018) (0.3502) (0.3213) (0.2251) ( R 0.2202) 5 0.0021 0.0002 0.0014 2 0.0024 2 0.0011 2 0.0999 1.7631 0.0265 0.5797 2 0.1162 2 95–99 2 (0.0021) (0.5732) (0.0011) (0.0011) (0.0023) (0.0018) (0.5349) (0.4222) (0.4076) (0.3354) R ( 5 0.2665) 99 2 0.0021 0.0034 0.0101 0.0115 0.0231 0.8456 2 0.0946 2 2.9525 3.7976 2.2407 . 2 (0.0056) (0.9924) (0.0030) (0.0038) (0.0054) (0.0745) (0.0057) (1.0891) (0.6442) (1.9395) ( R 0.3450) 5 Panel F: Money Measure: Change in Bernanke-Mihov Monetary-policy indicator Change in GDP Total loans 0123401234 Univariate 95 0.2491 0.0389 0.0671 2 0.4285 2 0.1192 , 2 (0.1805) (0.2014) (0.3583) (0.1698) (0.2176) ( R 0.1254) 5 0.4467 0.4244 0.1109 2 1.1053 2 0.1616 95–99 2 (0.4164) (0.2725) (0.2800) (0.3769) (0.3610) R ( 0.1280) 5 2 0.7211 1.0204 2.3224 0.2335 0.8005 . 99 2 (0.8896) (0.7678) (1.3088) (0.6189) (0.8930) ( R 5 0.1406) Bivariate 95 0.3874 0.0854 , 2 0.1057 0.2799 0.6243 0.3872 0.138 0.2682 0.1502 0.1358 2 (0.1916) (0.3757) (0.3109) (0.1781) (0.2625) (0.1331) (0.2239) (0.2215) (0.2974) (0.1333) R ( 0.2420) 5 95–99 0.5088 0.5294 0.3072 2 0.5354 0.309 0.4486 1.3427 0.0998 0.6261 2 0.0366 2 (0.2425) (0.4909) (0.2571) (0.2064) (0.5315) (0.3127) (0.3746) (0.2477) (0.2580) (0.3455) ( R 0.3040) 5 . 99 0.0652 1.2694 2.0188 1.079 2.3049 2.585 0.9495 2 1.4611 1.6821 1.439 2 (0.9654) (0.7247) (1.3014) (2.0710) (1.0846) (0.4983) (0.5772) (1.0308) (0.3927) (0.6407) R ( 0.3068) 5 Notes: Standard errors are in parentheses. All regressions also contain a time trend, which is not shown. f between the two types of endogeneity effects the sum of the ’s for the big banks is always positive is supportive of the rational buffer- that might be biasing our estimates for the stocking story. small banks. As discussed above, the fact that

14 JUNE 2000 THE AMERICAN ECONOMIC REVIEW 420 This suggests that the magnitude of the ’s f in other categories, such as long-term mort- gages. If this is so, the effects that we are from the small-bank regressions might be understating looking for will emerge more clearly with the effects of monetary policy b C&I loans. on . Taking the logic further, one might be t tempted to argue that the effects of monetary Table 4 presents the details of the individual regressions that make up Table 3. There are six policy would be more accurately measured by panels, A through F, one for each combination the small-bank/big-bank differentials. How- of loan type and monetary indicator. Most of the ever, some caution is probably warranted on ’s patterns are similar across panels, so it is in- f this latter point. Not only is the sum of the structive to focus first on just one—Panel B, for for the big banks positive in all our specifi- cations, in most cases the estimates are sur- C&I loans and the federal funds rate—for which prisingly large, often several times (in the estimates are the most precise. A couple of salient facts emerge. First, while we reported in absolute value) the size of the corresponding Table 3 only the sums of the five f coefficients negative estimates for small banks. It may (lags 0 through 4), we can now look at all the well be reasonable to ascribe these large individual ’s, and see that the sums are not f positive values to a strong bias induced by hiding any erratic behavior. In fact, for the rational buffer-stocking, and to posit that the small-bank category, every single one of the bias has the same sign for big and small ’s is negative in the univariate spec- f individual banks. It is more of a leap to claim that the bias is of the same size for big and small ification, and all but one are negative in the bivariate specification. Moreover, in both cases, banks, which is what one must believe if one the implied response of is to use the small-bank/big-bank differentials b to a monetary shock t to explicitly quantify the effects of monetary has a plausible hump shape for the small banks, b with the coefficients increasing over the first policy on . Given that the implied bias is so t couple of lags and then gradually dying down. large for the big banks, and given that we do not have a precise understanding of why this Second, in the bivariate versions of the specifi- might be so, care should be taken not to cations, the g coefficients on GDP are for the most part positive. Again, this is consistent with overinterpret the small-bank/big-bank differ- entials in this regard. the rational buffer-stocking story, and thus gives us yet another reason to think that our In Panel B, with total loans, the point esti- estimates for the small-bank category err on the mates generally go in the same direction as in 33 Panel A—five of six estimates for the small- side of conservatism. bank category are negative, and all six for the Comparing across the different panels in Ta- big-bank category are positive. But the magni- ble 4, one can get an idea of how well the second-step regressions fit with the different tude of the small-bank estimates is typically indicators. The federal funds rate clearly has the only about one-third to one-quarter that of the corresponding values in Panel A. Consequently, most explanatory power of our three measures. For example, Panel B tells us that with C&I only 4 of the total of 12 test statistics are sig- -values are loans, the univariate second-step regression for p nificant at 2.0 percent; three other small banks that uses the funds rate achieves an below 10.5 percent. 2 R Why are the results for C&I loans stronger of 29 percent. In the bivariate specification 2 than those for total loans? There are at least that adds GDP, the R rises to 45 percent. Considering that the left-hand-side variable in two possible explanations. First, this outcome is to be expected based on the rational buffer- this regression is just a noisy proxy for the degree of banks’ liquidity constraints, these stocking story. If this story is correct, our numbers strike us as quite remarkable. estimates are generally too conservative, and the conservatism will be more pronounced for total loans, since aggregation across loan cat- 33 There is another reason why the coefficients on GDP egories of different cyclicality exacerbates might be positive: an increase in activity raises loan de- any bias. Second, and more simply, it may be mand, and liquid banks are more able to accommodate their that because of their short maturity, banks can customers—i.e., increased demand makes banks’ liquidity constraints more binding. adjust C&I volume more readily than volume

15 KASHYAP AND STEIN: THE TRANSMISSION OF MONETARY POLICY VOL. 90 NO. 3 421 (5) IV. Robustness ~ D L log ! it We have already mentioned a number of robustness checks throughout the text and 4 4 footnotes. Just to remind the reader of some of the more significant ones, our results are 5 1 ! ~ ~ log D a log D M D m ! L L O O 2 it it t 2 j j j j generally unaffected by: how we screen for j 5 1 j 5 0 outliers; whether we base our analysis on 4 4 bank banks versus holding companies; whether we include cash in our measure of L M 1 ~ log D a ! m D 5 O O j j 2 it 2 t j j liquidity; whether we use a more complex lag 5 0 j 5 1 j specification or tighter geographic controls 4 in our first-step regressions; and whether or not a time trend is included in the second- 1 1 Q GDP p TIME D O t j j 2 t step regressions. We also obtain essentially 0 5 j identical results with a “quasi-IV” approach 12 3 that purges our liquidity measure of any correlation with observable measures of QUARTER r C FRB 1 1 O O kt k k ik loan riskiness. However, there remain a cou- 1 5 1 5 k k ple of items which merit a more detailed treatment. 4 M B 1 1 TIME d D 1 f h O j 1 2 it j t 2 t S 5 j 0 An Interactive, One-Step A. Regression Approach 4 1 . 1 ́ g D GDP O t j j it 2 D As argued above, our two-step method prob- 0 j 5 ably errs on the side of being overparameter- ized. Thus we now consider a more tightly structured approach, compressing our “univari- By comparing equations (4) and (5) with equa- tions (1)–(3), one can see the main differences ate” and “bivariate” two-step models into the following one-step models respectively: between the two methods. In the two-step method, macro variation in loan growth is ab- sorbed with a separate dummy term for each of periods k Federal Reserve districts in each of t L log ~ D (4) ! it dummies). In the one-step (i.e., a total of kt k time-invariant Federal method, there are only 4 4 Reserve-district dummies, and macro effects are ! ~ 5 1 L D a m log D M O O j 2 2 j j t j it modeled much more parsimoniously as a linear j 1 0 5 5 j function of changes in monetary policy and 34 GDP. 3 Table 5 presents an overview of the estimates TIME Q 1 QUARTER r 1 O k t kt generated by the one-step approach. As can f of 5 k 1 be seen, the point estimates are generally quite close to those in Table 3. However, the standard 12 errors are much reduced, leading to more 1 B 1 C TIME d 1 h FRB O k 2 it 1 ik t S -values. This outcome is p strongly significant 1 5 k 4 34 The two-step method also allows the lag coefficients f D M 1 1 ́ O j it j 2 t D on past loan growth—the a ’s—to vary period by period, 0 5 j while the one-step method makes them time-invariant.

16 JUNE 2000 422 THE AMERICAN ECONOMIC REVIEW (4) QUATIONS E STIMATION OF T TEP -S NE 5—O ABLE E what one would expect—to the extent that we AND UM OF C OEFFICIENTS ON (5): S are willing to impose more structure, and not M ONETARY -P OLICY I NDICATOR throw away much of the variation in the data, our tests should become more powerful. Panel A: C&I Loans Bivariate Univariate 1. Boschen-Mills 95 2 0.0614 2 0.0430 , Results from Subsamples B. (0.0069) (0.0077) 0.0171 2 95–99 0.0242 Finally, we check to see how our results hold (0.0276) (0.0288) 99 0.1337 . 0.0862 up across subsamples. There are two motiva- (0.0530) (0.0581) tions for doing so. First, as noted earlier, there 0.1475 2 Small–Big 0.1767 2 are reasons to think that our monetary indicators (0.0535) (0.0586) 2. Funds rate may not be as reliable during the first part of our 0.0238 , 95 2 0.0339 2 sample period, which contains the Volcker re- (0.0041) (0.0022) 95–99 2 0.0013 0.0102 gime. Second, we would like to know if our (0.0133) (0.0147) conclusions are colored by Regulation-Q type 0.0602 . 99 0.0903 restrictions, which were still in place in the (0.0239) (0.0266) Small–Big 2 0.1141 0.0941 2 early part of our sample period. By looking only (0.0240) (0.0269) at the latter part, we can directly address this 3. Bernanke-Mihov concern. 95 2 , 1.7802 2 2.7518 (0.4567) (0.3920) In Table 6, we reproduce all the numbers in 1.3557 3.7417 95–99 Table 3 for each of two subsamples. A clear (1.5929) (1.7226) . 6.3203 3.8509 99 pattern emerges: the results are almost uni- (2.7288) (3.0582) formly stronger and more statistically signifi- Small–Big 6.6027 8.1005 2 2 cant in the second subsample, which begins in (2.7568) (3.0921) 1986Q1. For example, in spite of the reduced Panel B: Total Loans number of observations, we find that for this Bivariate Univariate later period 10 of the 12 test statistics for C&I 1. Boschen-Mills -values of 5.5 percent or p loans in Panel A have , 95 2 0.0309 2 0.0268 p lower; eight have -values below 1.0 percent. In (0.0022) (0.0022) contrast, while all but one of the C&I point 95–99 2 0.0029 0.0392 estimates for the earlier period go the right way, (0.0164) (0.0155) only four are significant at the 5.2-percent level 0.0379 . 99 2 0.0117 (0.0336) (0.0380) or better. This fits with the idea that it is harder 0.0647 0.0191 2 Small–Big 2 to get an accurate handle on monetary policy (0.0337) (0.0380) during the first half of our sample period. It 2. Funds rate also makes it clear that our earlier results are , 95 2 0.0144 2 0.0119 not in any way driven by Regulation-Q-related (0.0009) (0.0007) 0.0056 0.0013 2 95–99 factors. (0.0084) (0.0077) 99 0.0297 . 0.0509 V. Economic Significance of the Results (0.0130) (0.0141) 0.0628 Small–Big 2 0.0440 2 (0.0130) (0.0141) So far, we have focused on the statistical sig- 3. Bernanke-Mihov nificance of our estimates. Now we ask whether 0.0803 0.6628 2 95 , they imply economically interesting magnitudes. (0.1626) (0.1410) For the sake of transparency, we focus on the 0.4789 95–99 2.5306 (0.8774) estimates from the funds-rate regressions. A first (0.8378) . 1.6325 99 5.3095 step is to quantify how two equal-sized banks with (1.8224) (1.5571) different values of B respond to a shock. From it Small–Big 2 2.2953 2 5.2292 Table 3, Panel A, the most conservative estimate (1.5634) (1.8296) f of the sum of the ’s for small banks’ C&I loans Note: Standard errors are in parentheses. 2 is 0.0151. (This comes from the bivariate spec-

17 KASHYAP AND STEIN: THE TRANSMISSION OF MONETARY POLICY VOL. 90 NO. 3 423 WO -S TEP E STIMATION OF E QUATIONS (1), (2), AND (3): T 6—T ABLE AMPLE R :S UM OF C OEFFICIENTS ON M ONETARY -P OLICY I NDICATOR PLIT S S ESULTS Panel A: C&I Loans 86Q1–93Q2 76Q1–85Q4 Bivariate Univariate Univariate Bivariate 1. Boschen-Mills 2 0.0074 2 0.0074 95 , 2 0.0049 0.0756 2 (0.0201) (0.0156) (0.0106) (0.0175) 0.1271 0.0537 2 2 0.0397 95–99 0.0317 (0.0221) (0.0502) (0.0445) (0.0269) 0.1981 0.1620 99 2 0.0561 . 0.0082 (0.0857) (0.1092) (0.0388) (0.0361) 2 0.2056 2 0.1694 2 Small–Big 0.0195 2 0.0131 (0.0448) (0.0407) (0.1206) (0.0897) 2. Funds rate 2 0.0004 2 0.0260 2 0.0256 , 95 2 0.0193 (0.0133) (0.0099) (0.0070) (0.0101) 0.0030 0.0367 0.0519 95–99 0.0203 2 2 (0.0216) (0.0167) (0.0232) (0.0225) . 99 0.0899 0.1936 0.2084 0.0473 (0.0258) (0.0392) (0.0318) (0.0270) 2 0.2196 2 0.2339 0.0665 0.0903 2 Small–Big 2 (0.0294) (0.0292) (0.0456) (0.0256) 3. Bernanke-Mihov 1.6565 2 1.8269 2 2.9728 , 95 2 0.9554 (0.8072) (1.6555) (0.4844) (0.9415) 95–99 1.5482 1.2918 5.8490 8.2468 2 (3.5492) (2.2360) (2.4472) (2.9796) 2.4275 15.2314 15.1635 0.1705 2 99 . (2.6942) (3.4808) (2.6893) (2.3654) 2 17.0583 2 0.7849 2 18.1363 Small–Big 2 0.7710 (2.4477) (2.7472) (2.2147) (2.2301) Panel B: Total Loans 86Q1–93Q2 76Q1–85Q4 Bivariate Univariate Univariate Bivariate 1. Boschen-Mills 2 0.0008 95 , 2 0.0018 0.0417 0.0095 (0.0043) (0.0160) (0.0114) (0.0055) 0.0334 0.0309 2 0.0798 2 95–99 0.0204 (0.0189) (0.0106) (0.0154) (0.0189) 2 0.0265 0.1449 . 99 2 0.0800 0.1455 (0.0289) (0.0469) (0.0601) (0.0249) 0.1430 2 0.0170 0.0384 2 Small–Big 0.1447 (0.0312) (0.0268) (0.0609) (0.0473) 2. Funds rate , 2 0.0079 2 0.0135 95 2 0.0062 0.0015 (0.0062) (0.0041) (0.0048) (0.0056) 2 0.0082 0.0175 0.0076 95–99 2 0.0217 (0.0151) (0.0051) (0.0089) (0.0158) 0.0290 0.0050 99 0.1004 . 0.1121 (0.0180) (0.0265) (0.0162) (0.0129) 2 0.1083 0.0113 0.1256 Small–Big 2 0.0275 2 2 (0.0141) (0.0167) (0.0194) (0.0282) 3. Bernanke-Mihov 0.0246 0.0169 2 0.2564 95 1.6395 , 2 (0.5633) (0.3865) (0.9106) (0.3712) 0.1350 1.9791 2.3618 2 95–99 1.4103 (1.2238) (1.7311) (0.7047) (1.5434) 4.6248 12.2748 1.1932 10.0419 . 99 (3.1775) (1.6549) (1.1024) (1.3265) 2 2.9853 Small–Big 10.025 2 12.5311 2 1.2178 2 (1.0415) (1.1759) (1.7011) (2.3791) Note: Standard errors are in parentheses.

18 JUNE 2000 424 THE AMERICAN ECONOMIC REVIEW T -B MALL ANK GGREGATE A OVEMENT IN 7—M ABLE S ification.) Now think of a “liquid” bank as having CCOUNTED FOR BY ANKS L ENDING A B C ONSTRAINED B 5 60.2 percent, and an “illiquid” bank as it F OUR Q ATE -R UNDS F EDERAL F FTER A A UARTERS having B - 5 20.6 percent; these numbers corre it OINTS 100 B P HOCK OF ASIS S spond to the 90th and 10th percentiles of the distribution for small banks in 1993Q2. In this Aggregate Percentage change in percentage case, four quarters after a 100-basis-point hike in change in lending due the funds rate, the level of C&I loans of the to constraints lending illiquid bank will be roughly 0.6 percent lower 35 A. C&I loans That is, if both than that of the liquid bank. 1. Using univariate, small- banks started with a level of C&I loans equal to a bank sum of phi’s 0.73 1.01 $1,000, then purely on the basis of liquidity dif- 2. Using bivariate, small- b ferences, we would predict a $6 gap between the bank sum of phi’s 3.33 0.41 3. Using univariate, small– two banks a year after the funds-rate shock. a 1.01 big bank differentials 2.90 The estimates in Table 3 are also consistent 4. Using bivariate, small– with a much larger cross-sectional effect. If b 3.62 big bank differentials 3.33 we base our calculation on the bivariate B. Total loans differential small-bank/big-bank coefficient 1. Using univariate, small- c 0.24 2.39 bank sum of phi’s of 0.1327 in Panel A of Table 3, we get a 2 2. Using bivariate, small- 5.3- percent gap in the level of C&I loans across d bank sum of phi’s 3.15 0.13 the liquid and illiquid small banks one year after 3. Using univariate, small– c the rise in the funds rate. However, it is impor- 0.95 2.39 big bank differentials 4. Using bivariate, small– tant to recall the caveat that applies to this d big bank differentials 1.39 3.15 second type of calculation: it implicitly assumes that the size of the rational buffer-stocking bias Notes: The numbers in the first column are based on the is the same for small and big banks. Given that two-step estimates reported in Table 3. we are attributing a large bias to the big banks, The numbers in the second column are drawn from Kashyap and Stein’s (1995) estimates for the “small95” and given that we do not have a detailed under- category as follows: standing of the roots of this bias, such an as- a Table 4, Panel 1; sumption may well lead us to overstate the b Table 4, Panel 2; c quantitative effects of monetary policy. Table 3, Panel 1; d Table 3, Panel 2. The preceding calculations only compare banks at extremes of the liquidity spectrum. To get an idea of the total impact of liquidity constraints across all small banks, we integrate over the dis- small-bank/big-bank differentials, the correspond- tribution of ing number is 3.62 percent. B ’s . To do this we use the actual B it it Once we have the total effect due to liquidity from 1993Q2, and assume that liquidity con- straints are binding everywhere below the 90th constraints among small banks, it can be com- B pared with aggregate movements in small-bank - percentile value of . For example, if we main it 2 0.0151 for the tain the conservative estimate of lending. Here, we draw on Kashyap and Stein ’s, we conclude that one year after the f sum of the (1995), who, using the same sample period and total C&I lending of methodology, find that in a bivariate specifica- shock to the funds rate, the tion, the aggregate all small banks is 0.41 percent lower than it would C&I lending of all small banks is reduced by 3.33 percent a year after a be if all these small banks were unconstrained. 36 Thus based Using the more aggressive estimate based on the 100-basis-point funds-rate shock. ’s, one f on our conservative estimates of the might argue that liquidity constraints “explain” 35 b that is This comes from multiplying the total change in 36 traced out over the year by the liquidity differential (0.0151 3 See Table 4, Panel 1, of Kashyap and Stein (1995), the 2 0.206) 5 0.006). To be more precise, one should (0.602 line labeled “small95”. The advantage of using these esti- account for the dynamic effects that arise from serial correlation in mates (rather than the one-step results reported here) is that the unit of observation is aggregate small-bank lending— loan growth. However, there is very little persistence in either C&I or total loan growth, so these effects are trivial. i.e., the numbers are value-weighted.

19 KASHYAP AND STEIN: THE TRANSMISSION OF MONETARY POLICY VOL. 90 NO. 3 425 12 percent of the total decline in small-bank in monetary policy matter more for the lending C&I lending subsequent to a monetary shock. of those banks with the least liquid balance Using the more aggressive estimates, this ratio sheets. The results are for the most part strongly is increased to 109 percent. Table 7 presents a statistically significant, and are robust to a wide range of variations in estimation technique. more complete set of numbers, covering both C&I and total loans, and drawing on the param- Moreover, the implied differences across banks eter estimates from both our univariate and bi- are of a magnitude that, at a minimum, one would call economically interesting. variate specifications. Although it is hard to be Unlike with the earlier evidence, it is much precise, this crude analysis would seem to imply economically noteworthy magnitudes. harder to come up with alternative, nonlending- channel stories to rationalize our results. In par- From a macro perspective, we are arguably ticular, if one wants to explain our results using not quite finished with this exercise, because a standard interest-rate channel, one has to ar- small-bank lending, as we have defined it, is gue that those banks whose customers’ loan only a fraction (about one-quarter) of total lend- demand is most sensitive to monetary policy ing. In other words, the next question one would like to answer is: “what portion of the total systematically opt to hold less in the way of liquid assets—i.e., one has to invoke the het- economywide drop in lending is due to liquidity constraints?” Unfortunately, here our evidence erogeneous risk-aversion story. Not only is this story somewhat implausible from a theoretical is of little direct use; we have been interpreting the surprisingly large positive perspective, we have been able to marshall sev- ’s for the big f banks as indicative of a bias, which leaves us eral distinct pieces of evidence which all imply that it is not borne out in the data. with no scope to measure the extent of their The bottom line is that it now seems hard to liquidity constraints. Rather than basing a fur- deny the of a lending channel of mon- existence ther set of calculations on totally arbitrary as- sumptions about big-bank constraints, we etary transmission, at least for the United States simply make the following observation. If one in our sample period. The next logical question wants a very loose lower bound, one can as- how important then becomes: quantitatively, is sume that the lending channel for aggregate economic ac- medium and big banks are com- all pletely unconstrained. In this case, the relative tivity? As we have begun to see in Section V above, this question is harder to answer defini- importance of liquidity constraints for total tively with our data set. First, while our results bank lending would be roughly one-quarter of 37 leave open the possibility that the aggregate what it is for small-bank lending. loan-supply consequences of monetary policy could be very substantial, our attempts to VI. Conclusions precisely measure this aggregate effect are ham- Previous work has uncovered a variety of pered by the large estimation biases in the big- evidence that is consistent with the existence of bank regressions. Second, even if one can make a stronger case a lending channel of monetary transmission. Unfortunately, much of this evidence also ad- that monetary shocks have a large impact on mits other interpretations. Our premise in this total bank-lending volume, there is a further missing piece to the puzzle. In particular, one paper has been that to provide a sharper test of the lending channel, one has to examine in more still needs to know the elasticity with which detail how monetary policy impacts the lending borrowers can substitute between bank and non- individual bank forms of credit on short notice. For exam- banks, as opposed to behavior of ple, if a small company is cut off from bank broadly aggregated measures of lending. Our principal conclusions can be simply lending, how much higher is the implicit cost of stated. Within the class of small banks, changes capital if it has to instead stretch its accounts payable? And what are the implications for its inventory and investment behavior? These are 37 This is also overly conservative for another reason: questions that will not be easy to answer satis- Kashyap and Stein (1995) show that small-bank lending factorily. Nonetheless, if the goal is to achieve a falls by substantially more than large-bank lending after a funds-rate shock. full and accurate picture of the role of banks in

20 JUNE 2000 THE AMERICAN ECONOMIC REVIEW 426 the transmission of monetary policy, they will 1984 it is not possible to separately add up all of the items that are now counted as investment eventually have to be addressed. securities. As an approximation we take the sum ATA PPENDIX D A of items rcfd0400 (U.S. Treasury Securities), rcfd0600 (U.S. Government Agency and Cor- porate Obligations), rcfd0900 (Obligations of Our sample is drawn from the set of all in- sured commercial banks whose regulatory fil- States and Political Subdivisions), and rcfd0380 (All Other, Bonds, Stocks and Securities). In ings show that they have positive assets. either case, we then add on Fed Funds Sold and Between the first quarter of 1976 and the second Securities Purchased Under Agreements to Re- quarter of 1993, this yields 961,530 bank- sell (rcfd 1350) to get an overall series for quarters of data. The actual number of observa- securities holdings. tions in our regressions is less, for several reasons. First, because our regressions involve The data for total loans after March 1984 come from item rcfd1400, Gross Total Loans growth rates, we lose an initial observation for each bank. Second, because mergers typically and Leases. Prior to March 1984 “Lease Financ- ing Receivables” (rcfd 2165) are not included as create discontinuities in the surviving bank’s part of total loans so the two series need to balance sheet, we also omit banks in any quar- summed to insure comparability. More impor- ters in which they are involved in a merger. tantly, in December of 1978 banks began re- These first two cuts leave us with a sample of porting their lending on a consolidated basis so 930,788 observations which could potentially that foreign and domestic loans were no longer be analyzed. Next, in order to make sure that outliers are not driving our results, we eliminate separately identified. Prior to that period the any observations in which the dependent vari- foreign data were unavailable. Since most banks had only limited foreign operations at that time, able is more than five standard deviations from its mean. In the regressions involving C&I loans this shift is relatively unimportant for the typical we further eliminate any banks for which C&I bank. However, for many of the biggest banks the change generates a noticeable discontinuity lending constitutes less than 5 percent of their in reported lending. One of the advantages of total lending. Together these filters remove our two-step regression approach is that it helps about another 67,000 bank-quarters. Finally, we limit the influence of this one-time jump in require that all the banks in our sample have four consecutive quarters of loan growth. The lending—the jump is absorbed in the constant term of the first-step regression. Nevertheless, cumulative effect of all these screens is that our basic C&I regressions use 746,179 observa- to confirm that the shift was not responsible for any of our key findings, we also reestimated our tions. For the total loan regressions we follow main regression omitting this period and found the same procedures except that we skip the check on the ratio of C&I loans to total loans, so no important changes. The data for commercial and industrial loans that our total sample size is 836,885. are taken from rcfd1600. Starting in March Our main results depend on accurately mea- 1984 the series begins to include holdings of suring a bank’s size and its lending and securi- ties holdings. Our size categories are formed by those bankers’ acceptances which are accepted sorting the banks on the basis of their total by other banks. Prior to that time only each assets— call report item rcfd2170. Although the bank’s own acceptances are included, but there is no way to create a series which is consistent total asset data are measured on a consistent basis throughout our sample, much more detail in the treatment of acceptances because a bank’s concerning bank assets and liabilities has been own acceptances are never separately reported. collected starting in March 1984, so that most of As in the total loan data, the reported level of the other asset data is measured differently be- C&I lending for large banks also shows a jump fore and after that point. in the fourth quarter of 1978. The snapshots of the data given in Table For our securities variable after March 1984 1 involve a number of other items from the call we begin with the sum of the book value of total investment securities (item rcfd0390) and assets reports. The details concerning these variables are given in the appendix to Kashyap and Stein held in trading accounts (rcfd 2146). Prior to

21 KASHYAP AND STEIN: THE TRANSMISSION OF MONETARY POLICY VOL. 90 NO. 3 427 (1995) and are available at: www.frbchi.org/ “Monetary Policy Shocks: Evans, Charles. rcri/rcri_database.html. The only noteworthy What Have We Learned and to What End?” aspect of these items is that data on deposits was in John Taylor and Michael Woodford, eds., , 2000 (forth- reported differently before and after March of Handbook of macroeconomics coming). 1984. These different reporting conventions ex- Driscoll, John C. plain why we break out deposits into slightly “Does Bank Lending Affect Output? Evidence from the U.S. States.” different subcategories in 1976 and 1993. Working paper, Brown University, 1997. REFERENCES “Monetary Gertler, Mark and Gilchrist, Simon. Policy, Business Cycles, and the Behavior of Small Manufacturing Firms.” Berger, Allen N.; Kashyap, Anil K and Scalise, Quarterly Joseph M. Journal of Economics “The Transformation of the U.S. (2), pp. , May 1994, 109 309 – 40. Banking Industry: What a Long, Strange Trip Gertler, Mark and Hubbard, R. Glenn. “Finan- It’s Been.” Brookings Papers on Economic Fi- cial Factors in Business Fluctuations,” in Activity , 1995, (2) pp. 55–201. “Credit, . Kansas City, MO: nancial market volatility Bernanke, Ben S. and Blinder, Alan S. Federal Reserve Bank of Kansas City, 1988, American Money, and Aggregate Demand.” , May 1988 ( Economic Review pp. 33–71. Papers and “The Bank Lending Channel of Gibson, Michael. Proceedings ), 78 (2), pp. 435–39. “The Federal Funds Rate and the Monetary Policy Transmission: Evidence . Amer- from a Model of Bank Behavior that Incor- Channels of Monetary Transmission.” , September 1992, porates Long-Term Customer Relationships.” ican Economic Review 82 Working paper, Federal Reserve Board, (4), pp. 901–21. 1996. Bernanke, Ben S. and Gertler, Mark. “Inside the “Interest Rate Policy and Black Box: The Credit Channel of Monetary Goodfriend, Marvin. the Inflation Scare Problem: 1979 –1992.” Policy Transmission.” Journal of Economic (4), pp. 27– 48. 9 , Fall 1995, Perspectives Federal Reserve Bank of Richmond Eco- (1), pp. “Measuring nomic Quarterly , Winter 1993, 79 Bernanke, Ben S. and Mihov, Ilian. 1–24. Quarterly Journal of Eco- Monetary Policy.” Holmstro , August 1998, nomics 113 (3), pp. 869 –902. “Financial ̈ m, Bengt and Tirole, Jean. Intermediation, Loanable Funds, and the Real Boschen, John and Mills, Leonard. “The Effects of Quarterly Journal of Economics Countercyclical Policy on Money and Interest Sector.” , Rates: An Evaluation of Evidence from FOMC August 1997, 112 (3), pp. 663–91. Houston, Joel; James, Christopher and Marcus, Documents.” Working Paper No. 91-20, Fed- David. eral Reserve Bank of Philadelphia, 1995. “Capital Market Frictions and the Role of Internal Capital Markets in Banking.” Carpenter, Robert E.; Fazzari, Steven M. and Petersen, Bruce C. , November Journal of Financial Economics “Inventory Investment, (2), pp. 135– 64. 46 1997, Internal-Finance Fluctuations, and the Busi- “Is There a ‘Credit Channel’ Hubbard, R. Glenn. ness Cycle.” Brookings Papers on Economic , 1994, (2) pp. 75–122. Federal Reserve Bank for Monetary Policy?” Activity , May–June 1995, “Distinguishing Theories (3), 77 Cecchetti, Stephen G. of St. Louis Review of the Monetary Transmission Mechanism.” pp. 63–77. Federal Reserve Bank of St. Louis Review , Kashyap, Anil K and Stein, Jeremy C. “Monetary Policy and Bank Lending,” in N. Gregory (3), pp. 83–97. 77 May–June 1995, Christiano, Lawrence and Eichenbaum, Martin. Mankiw, ed., . Chicago: Uni- Monetary policy “Identification and the Liquidity Effect of a versity of Chicago Press, 1994, pp. 221–56. Monetary Policy Shock,” in Alex Cukierman, . “The Impact of Monetary Policy on Zvi Hercowitz, and Leonardo Leiderman, Bank Balance Sheets.” Carnegie-Rochester Political economy, growth, and business eds., , June Conference Series on Public Policy 1995, 42, pp. 151–95. cycles . Cambridge, MA: MIT Press, 1992, pp. 335–70. . “What Do a Million Banks Have to Say About the Transmission of Monetary Policy?” Christiano, Lawrence; Eichenbaum, Martin and

22 JUNE 2000 THE AMERICAN ECONOMIC REVIEW 428 National Bureau of Economic Research (Cam- , Journal of Money, Credit, and Banking Be.” August 1995, bridge, MA) Working Paper No. 6056, 1997. 27 (3), pp. 625–38. “The International Transmission of Fi- Kashyap, Anil K; Stein, Jeremy C. and Lamont, . Owen. nancial Shocks: The Case of Japan.” Ameri- “Credit Conditions and the Cyclical Be- can Economic Review havior of Inventories.” Quarterly Journal of , September 1997, 109 , August 1994, (4), pp. 495–505. 87 (3), pp. 565–92. Economics “Does Romer, Christina D. and Romer, David H. Kashyap, Anil K; Stein, Jeremy C. and Wilcox, David W. “Monetary Policy and Credit Con- Monetary Policy Matter? A New Test in the Spirit of Friedman and Schwartz,” in Olivier ditions: Evidence from the Composition of External Finance.” Jean Blanchard and Stanley Fischer, eds., American Economic Re- NBER macroeconomics annual: 1989 (1), pp. 78 –98. view , March 1993, 83 . Cam- “Monetary Policy and Credit Condi- . bridge, MA: MIT Press, 1989, pp. 121–70. . “New Evidence on the Monetary tions: Evidence from the Composition of Ex- Brookings Pa- Transmission Mechanism.” ternal Finance: Reply.” American Economic 86 , 1990, (1), pp. pers on Economic Activity (1), pp. 310 –14. Review , March 1996, “Bank 149 –98. Kishan, Ruby P. and Opelia, Timothy P. “Banking Conditions and Samolyk, Katherine A. Size, Bank Capital and the Bank Lending Regional Economic Performance: Evidence Channel.” Journal of Money, Credit, and Journal of of a Regional Credit Channel.” Banking, February 2000, 32 (1), pp. 121– 41. Monetary Economics , October 1994, 34 (2), “An Interest Rate-Based Indi- Laurent, Robert. pp. 259 –78. Economic Per- cator of Monetary Policy.” Sharpe, Steven A. “Bank Capitalization, Regula- spectives, Federal Reserve Bank of Chicago , tion, and the Credit Crunch: A Critical Re- 12 (1), pp. 3–14. January–February 1988, view of the Research Findings.” Working Ludvigson, Sydney. “The Channel of Monetary Paper No. 95-20, Federal Reserve Board, 1995. Transmission to Demand: Evidence from the Stein, Jeremy C. Journal of Market for Automobile Credit.” “An Adverse Selection Model , August 1998, of Bank Asset and Liability Management Money, Credit, and Banking with Implications for the Transmission of 30 Pt. 1, (3), pp. 365– 83. Rand Journal of Econom- Monetary Policy.” “Mon- Oliner, Stephen and Rudebusch, Glenn D. (3), pp. 466 – 86. 29 , Autumn 1998, ics etary Policy and Credit Conditions: Evidence of from the Composition of External Finance: Identification “The Steven. Strongin, , Comment.” American Economic Review Monetary Policy Disturbances: Explaining Journal of Monetary 86 (1), pp. 300 – 09. the Liquidity Puzzle.” March 1996, Economics , June 1995, 35 (3), pp. 463– “The Capital Peek, Joe and Rosengren, Eric. 97. Crunch: Neither a Borrower nor a Lender

Related documents

435 441 458 467r e

435 441 458 467r e

WT/DS435/R, WT/DS441/R WT/DS458/R, WT/DS467/R 28 June 2018 Page: (18 - 1/884 4061 ) Original: English AUSTRALIA CERTAIN MEASURES CON CERNING TRADEMARKS, – PACKAGING IONS AND OTHER PLAIN GEOGRAPHICAL I...

More info »
oldnew 11.dvi

oldnew 11.dvi

C ́edric Villani O ptimal transport, old and new June 13, 2008 Springer Berlin Heidelberg NewYork Hong Kong London Milan Paris Tokyo

More info »
VENUE | S3L X System Guide

VENUE | S3L X System Guide

VENUE | S3L-X System Guide and VENUE | Software 4.5

More info »
VENUE Profile Guide

VENUE Profile Guide

VENUE Profile Guide and VENUE Software 2.9 9322-62809-00 REV A 08/10

More info »
HDP book

HDP book

High-Dimensional Probability An Introduction with Applications in Data Science Roman Vershynin University of California, Irvine March 25, 2019 https://www.math.uci.edu/ rvershyn/ ~

More info »
7Sketches

7Sketches

Seven Sketches in Compositionality: An Invitation to Applied Category Theory ? Brendan Fong David I. Spivak (Last updated: October 12, 2018)

More info »
ed146562.tif.pdf

ed146562.tif.pdf

DOCUMENT RESUME RD 146 562 CS 003 773 AUTHOR Brown, Ann L. TITLE Knowing When, Where, and Now to Remember: A Problem of Netacogaition. Technical Report No. 47. INSTITUTION Bolt, Beranek and Newman, In...

More info »
The 9/11 Commission Report

The 9/11 Commission Report

Final FM.1pp 7/17/04 5:25 PM Page i THE 9/11 COMMISSION REPORT

More info »
complete

complete

LEGACY PROJECT A century-old feud over tribal fishing ignited brawls along Northwest rivers in the 1960s. Roughed up, belittled, and handcuffed on the banks of the Nisqually River, Billy Frank Jr. eme...

More info »
MDS 3.0 RAI Manual v1.16 October 2018

MDS 3.0 RAI Manual v1.16 October 2018

Centers for Medicare & Medicaid Services Long-Term Care Facility Resident Assessment Instrument 3.0 User’s Manual Version 1.16 October 2018

More info »
MC

MC

THE IWASAWA MAIN CONJECTURES FOR GL 2 CHRISTOPHER SKINNER AND ERIC URBAN 1. Introduction In this paper we prove the Iwasawa-Greenberg Main Conjecture for a large class of elliptic curves and modular f...

More info »
tsa4

tsa4

i i “tsa4_trimmed” — 2017/12/8 — 15:01 — page 1 — #1 i i Springer Texts in Statistics Robert H. Shumway David S. Sto er Time Series Analysis and Its Applications With R Examples Fourth Edition i i i ...

More info »
book0.dvi

book0.dvi

Mining of Massive Datasets Jure Leskovec Stanford Univ. Anand Rajaraman Milliway Labs Jeffrey D. Ullman Stanford Univ. c Copyright © 2010, 2011, 2012, 2013, 2014 Anand Rajaraman, Jure Leskovec , and J...

More info »
markovmixing

markovmixing

Markov Chains and Mixing Times David A. Levin Yuval Peres Elizabeth L. Wilmer University of Oregon E-mail address [email protected] : URL : http://www.uoregon.edu/~dlevin Microsoft Research, Universi...

More info »
bordism

bordism

BORDISM: OLD AND NEW DANIEL S. FREED ttheUniversityofTexasatAustin What follows are lecture notes from a graduate course given a in Fall, 2012. The first half covers some classical topics in b ordism,...

More info »
Pro Tools Reference Guide

Pro Tools Reference Guide

® Pro Tools Reference Guide Version 9.0

More info »
networks-book

networks-book

Networks, Crowds, and Markets: Reasoning about a Highly Connected World David Easley Jon Kleinberg Dept. of Economics Dept. of Computer Science Cornell University Cornell University Cambridge Universi...

More info »
iu teich 1 zu iri.pdf

iu teich 1 zu iri.pdf

̈ INTER-UNIVERSAL TEICHM ULLER THEORY I: CONSTRUCTION OF HODGE THEATERS Shinichi Mochizuki February 2019 The present paper is the first in a series of four papers, the Abstract. Teichm ̈ uller theory ...

More info »
F:\MySites\InterLand\Math\Cours

F:\MySites\InterLand\Math\Cours

CLASS FIELD THEORY J.S. Milne

More info »