Does Securitization Affect Bank Lending? Evidence from Bank Responses to Funding Shocks

Transcription

1 Does Securization Affect Bank Lending? Evidence from Bank Responses to Funding Shocks Elena Loutskina * First Version: November, 2004 Current Version: March, 2005 * Ph.D. Candidate, Finance Department, Carroll School of Management, Boston College, Chestnut Hill, MA 02467, Tel: (781) , fax: (617) , loutskin@bc.edu. For helpful comments and discussions I thank Thomas Chemmanur, Edward Kane, Alan Marcus, Philip Strahan, and seminar participants at Boston College. I alone am responsible for any errors or omissions.

2 Does Securization Affect Bank Lending? Evidence from Bank Responses to Funding Shocks ABSTRACT This paper studies the effect of securization on bank lending. I propose a new index of securizabily of a bank portfolio that can be thought of as a weighted average of the potential to securize loans of a given type, where the weights reflect the composion of an individual bank s loan portfolio. I use this new index to show that securization makes bank lending less sensive to cost of fund shocks because provides banks wh an addional source of funding. Securization, thus, weakens the link from monetary policy to bank lending activy. Furthermore, by allowing banks to convert illiquid loans into liquid funds, securization reduces banks holdings of liquid securies and increases their loan portfolios.

3 Does Securization Affect Bank Lending? Evidence from Bank Responses to Funding Shocks 1. Introduction Since the 1970s, the market for securized loans in the Uned States has grown to dominate the mortgage market and has become an increasingly important factor in lending to both consumers and businesses (Figure 1). In 2003, for example, $13.6 trillion of loans were securized, or about 40% of all loans outstanding. Today, the securization market exceeds the size of the corporate bond market. Despe s importance, there is ltle research on how securization has changed the behavior of banks. 1 This paper examines how securization is changing the way banks manage their funding and liquidy and how these changes have in turn altered the tradional links from monetary policy to bank loan supply. I show, first, that securization creates a new source of liquidy by allowing banks to convert illiquid, hard-to-sell loans into marketable securies. Second, by allowing banks to substute cash and securies for loans, securization reduces the sensivy of bank lending to the availabily of the external sources of funds and, thus, weakens the abily of the monetary authory to affect bank lending through open market operations. I first analyze the extent to which securization has reduced banks need to carry liquid assets to meet unexpected demands from deposors and borrowers. Since securization allows banks to transform formerly illiquid loans into tradable securies, and the cost of this transformation has been falling over time, leads us to question the tradional view of banks on-balance-sheet liquidy. I argue that the higher the extent of a bank s abily to securize s loan portfolio, the higher s flexibily to convert illiquid loans into cash which in turn might lead to a higher degree of substutabily between liquid funds and securizable loans as sources of liquidy and loan financing. 1 For analysis of the issue of changing mortgage rates under the condion of the evolving market for ABSs see e.g., Black, Garbade, and Silber (1981), Kolary, Fraser, and Anari (1998), and Heuson, Passmore, and Sparks (2000). For analysis of the role of the government sponsored enterprises (GSEs) and the effect of government subsidies to GSEs see Passmore (2004), Ambrose and Warga (2002), and Nothaft, Pearce, and Stevanovic (2002). For the analysis of the effect of securization on the efficacy of the monetary policy in influencing real output see Estrella (2002). 1

4 In this paper I propose a new bank-specific index of securizabily of a bank s loan portfolios (S ) and use to test the validy of my arguments. The proposed index measures each bank s potential to securize s loans in a way that captures both the composion of the bank s loan portfolio and the growth in the depth of the securization market over time. The index can be thought of as a weighted average of the potential to securize loans of a given type (based on market-wide averages), where the weights reflect the composion of an individual bank s loan portfolio. Thus, market trends generate time variation in the index, whereas differences in bank loan portfolios generate variation across instutions. Using this new measure, I find that as banks abily to securize loans has increased, their holding of liquid assets on balance sheet has decreased. 2 The magnude of this decline is not only statistically but also economically significant (Figure 2). For example, from 1976 to 2003, the percentage of total assets held as liquid securies decreased on average by 3.29 percent due to expanding market for securizable loans. This decline is equivalent to roughly 36 percent of bank capal. 3 Thus, securization seems to have increased the supply of bank lending per dollar of capal in the industry. The decline in bank holdings of liquid assets suggests that securization acts as a substute for liquidy on the balance sheet. If true, the increasing liquidy of bank loans ought to change the link from bank funding availabily (e.g. deposs) to their willingness to supply cred. The existing lerature documents that the availabily of addional internal and external sources of funds partially alleviates the effect of restrictions in availabily of funds on the supply of loans. Kashyap and Stein (2000) document that more liquid banks are less susceptible to shocks to costs of external financing than less liquid ones. Campello (2002) shows that internal capal markets also help to shield banks from the impact of funding shocks. Since securization provides banks wh an addional source of financing, one can make an argument in tune wh this lerature. 2 I use the tradional approach to measuring the on-balance-sheet liquid securies and compute this measure as the sum of securies and federal funds sold divided by total assets. 3 Note that this decline cannot be explained by any time trends such as the increase in average bank size over time, changes in the banking regulation, etc. 2

5 Consider two small banks, both facing limed abily to raise external finance. The banks are alike except that the first has more opportunies to securize s loans, perhaps because holds most of s portfolio in mortgages, whereas the second bank holds many business loans (commercial and industrial (C&I) loans) which continue to be difficult to securize. Under a funding shock, which causes them to lose cheap sources of funds (e.g., the insured deposs), each bank is likely to contract s assets by eher selling liquid assets or shrinking s loan portfolios. Wh securization, however, banks may also choose to pool and securize existing loans. Since the first bank holds more mortgages inially, has better abily to insulate s lending from these shocks through securization than the second bank. My empirical strategy rests on the idea that new lending by the first bank ought to decline less under a posive cost of funds shock than new lending by the second bank. Furthermore, this approach allows me to test how the abily of a monetary authory to shift bank loan supply has been affected by securization. 4 The second part of this paper tests the argument that securization alleviates sensivy of banks lending to the cost of funds shocks. The empirical tests follow the regression framework of Kashyap and Stein (2000), which allows me to explo both time-series and cross-sectional variation in S and s interaction wh the cost of external funds. I explo the Federal Reserve s abily to affect bank cost of funds via open market operations and proxy external costs of funds by various monetary policy indicators. 5 The results suggest that securization has indeed made total loan growth (especially growth in business loans) less sensive to monetary policy shocks. For example, a bank wh a high degree of potential access to securization (e.g., one that holds significant amount of mortgages) incurs a smaller decrease in lending under a monetary tightening than a bank wh limed access to securization (e.g., a bank focused on business lending). Figure 3 illustrates the result intuively by plotting average loan 4 Estrella (2002) argues that securization, in principle, may reduce the potency of monetary policy by insulating banks from the effects of tight funding condions. His research uses aggregate data to illustrate that the growing trend toward securization in the U.S. has weakened the extent to which a given change in monetary policy affects real output. I extend this idea by exploing not only aggregate trends but also differences across banks. 5 The Federal Reserve s abily to affect bank lending behavior via open market operations is called the bank lending view of the monetary policy transmission. During a Fed-induced tightening, banks access to information-insensive external funds such as insured deposs is constrained. For a review of this lerature, see Bernanke and Blinder (1992), Bernanke and Gertler (1995), and Kashyap and Stein (1994). The empirical evidence is shown in Kashyap and Stein (1995, 2000), Jayaratne and Morgan (2000), and Krishnan and Opiela (2000) 3

6 growth during a tight and loose monetary regime for banks wh high and low S. One can see that during the period of monetary tightening (when the banks cost of funds increases), banks wh wide access to securization exhib significantly higher business loan growth than banks wh limed access to the securization market. 6 Securization thus seems to alleviate the effect of the monetary policy on loan supply, and this weakening varies across banks. 7 My statistical model also allows me to formally compare how an increase in the federal funds rate would affect two otherwise identical banks (equal-size, same level of on-balance-sheet liquidy). I assume that the bank wh a limed access to the securization market has the securizabily of s loan portfolio (S ) at the 10 th percentile of the distribution, and the bank wh a wide access to the securization market has S at the 90 th percentile. My results indicates that under a monetary tightening of 100 basis points in the federal funds rate a bank wh wide access to the securization market would experience 0.7% to 1.3% smaller decline in total loans than a bank wh limed access to the securization market, while the first bank s business lending would decline almost 5.25% less. In other words, the abily to securize their existing loans insulates banks willingness to supply cred from a monetary policy induced shock to the availabily of external financing. 8 Securization, thus, changes the nature of banking and weakens the abily of monetary authory to affect bank lending activy. The remainder of the paper is organized as follows. Section 2 describes the structure and magnude of the market for securized loans as well as possible channels of s influence on banks operations. Section 3 describes data and sample selection. Section 4 presents the intuion and methodology behind the bank-specific index of securizabily of a bank loan portfolio proposed in this paper. Section 5 presents the empirical tests and results for the hypothesis of substutabily between liquid funds and 6 Note that in contrast to the multivariate panel model, this simple univariate comparison does not control for loan demand; hence, loan growth is higher during the period of tightening than during the period of loosening. 7 This result is consistent wh Estrella (2002) who argues that securization, in principle, may reduce the potency of monetary policy by insulating banks from the effects of tight funding condions. His research uses aggregate data to illustrate that the growing trend toward securization in the U.S. has weakened the extent to which a given change in monetary policy affects real output. 8 For comparison, two equal-sized banks wh the same access to the securization market but wh levels of on-balance-sheet liquidy around 10 th and 90 th percentiles of the level of liquidy distribution will have 0.4% to 0.7% loan growth differential four quarters after a 100 basis points increase in the federal funds rate. 4

7 securizable loans on banks balance sheets. Section 6 describes empirical evidence for the argument that securization alleviates sensivy of banks lending to the cost of funds shocks. Section 7 concludes the paper. 2. The Securization Market In the last quarter of a century cred markets have been significantly influenced by securization, an increasingly important way that capal flows through the economy. Securization is part of an enormous wave of the financial innovations that lowers the costs of moving funds. It fosters the flow of funds and allows more efficient redistribution of funds in the economy. Securization is a process of creating new financial instruments by pooling the cash flows from a number of similar assets such as mortgages or cred cards accounts, and putting them into a separate legal enty (or special purpose vehicle, SPV) often wh some addional implic or explic guarantee or extra collateral. Creating this separate SPV isolates the cash flow generating assets and/or collateral so that the secury is not a general claim against the issuer, ust against those assets. The pooling process results in a diversified portfolio of cash flows that can be further stripped and repackaged based on various characteristics (e.g., the prepayment behavior), thereby reducing the need to monor each underlying payment stream. 9 The maor contributors to the development of bank loan securization are the Government- Sponsored Enterprises (GSEs) that were created by the US Congress to provide stabily and ongoing assistance to the secondary market for residential mortgages and to promote access to mortgage cred and home ownership in the US. GSEs are the enormous force that simplifies and fosters securization by being the largest buyer of mortgages in the US. Fannie Mae and Freddie Mac, combined, purchase almost one-half of all conventional single-family mortgage loans originated each year. The US economy has seen an enormous expansion of the securization market. Table 1 presents the amount of loans outstanding and loans securized for six loan categories over the sample period 1976:I to 9 For detailed discussion of the securization process and the role of SPVs see Groton and Souleles (2004). 5

8 2003:IV. Consider, for example, home mortgages, if in 1976:I the amount of securized home mortgages was $27.7 billion, then by the end of 2003 the total amount of securized home mortgages reached $4.25 trillion. At the same time the amount of home mortgages outstanding grew almost 15 times from $489 billion to $7,283 billion. In 1976 neher commercial mortgages, nor C&I loans, nor consumer cred were securizable types of loans. By the end of 2003 one can observe $294 billion worth of securized commercial mortgages, $104 billion worth of securized C&I loans, and $658 billion worth of securized consumer cred. Through the years C&I loans remain least securizable loan category. Figure 1 shows how the aggregate, economy-wide share of securized loans in total loans outstanding has been changing over the years for six loan categories. The share of securized home mortgages climbed from around 5% in 1976:I to almost 60% in In 2003:IV 20% of the commercial mortgages outstanding, 6% of C&I loans, 30% of consumer cred was securized. If I consider the aggregate lending characteristics I will observe only 2.2% of the total loans outstanding securized in 1976:I, and 40% of the total loans outstanding securized in 2003:IV. These numbers once again confirm the economic significance of the securization market. This quickly expanding financial market has significantly affected the nature of banking and, in particular, bank lending activy. Securization plays an important role in bank s operations. It provides banks wh a new source of financing their investment opportunies. Today banks can fund new loans by securizing them (or other outstanding loans). In addion, securization changes the tradional view on the depos instutions assets liquidy. Loan portfolios that were considered to be too cumbersome and expensive to sell years ago are becoming more and more liquid. In this paper I concentrate on these two implications of securization. The effect of securization, however, is not limed to two effects I discuss in this study. There exists number of other channels through which securization significantly affects the nature of banking. First, securization provides an opportuny for banks to hold more diversified loan portfolios, thus, protecting them against local economic shocks. Although deregulation has eliminated most of the legal restraints on geographic segmentation, many banks continue to originate loans in the regions or industries where they 6

9 have a superior knowledge of the market condions. Wh securization these loans can be bundled wh others and bought and sold all around the country. Money can flow from the regions wh excess deposs to the regions wh the unsatisfied loan demand. Second, securization gives banks an addional flexibily in terms of the matury of their assets. Now banks can adust their portfolios so that the matury of the assets matches the matury of the deposs more closely. Finally, allows banks to easily move assets off their balance sheets and provides a means to migate the regulatory capal requirements. All these factors together have significantly reshaped the way the depos instutions do business and are interesting to look at in the later studies. 3. Data and Sample Selection 3.1. Bank-Level Data Bank-level data come from the Federal Reserve s Report of Condion and Income ( Call Reports ) submted by insured banks each quarter. I compile a dataset wh quarterly income statements and balance sheet information for all reporting banks over the period 1976:I through 2003:IV. This dataset presents a number of challenges in terms of creating a consistent time series. Appendix I describes the construction of the key series in detail. When analyzing the dataset I first exclude all the bank-quarters wh missing information on total assets, total loans, and liquid funds. I then use data from the Federal Reserve National Information Center (NIC) to identify bank mergers and exclude banks in any quarter in which they go through a merger. Specifically, I exclude the acquiring bank in the quarters before and after a merger. To prevent the possibily of outliers driving the results, I eliminate all bank-quarters wh asset growth over the last quarter in excess of 50 percent, those wh total loan growth exceeding 100 percent, those wh total loans-to-asset ratio below 10 percent, and those wh the share of cred card loans in the loan portfolio above 50 percent. 10 The final dataset contains 1,344,696 bank-quarters. Addionally I 10 The cred card loans have experienced a significant increase in securization over last years. I, however, can account for cred card loans level of securization only as part of the consumer cred securization. Hence, I believe that my securizabily index do not carefully capture the degree of securizabililty of loan portfolios for banks heavily involved in the cred card business. Consequently, I exclude these banks from the considered sample to avoid possible distortions due to unobservable degree of securizabily of the cred card accounts. 7

10 separate the sample into two groups: large banks and small banks. I measure the size of a bank as a log of the real total assets. I assign bank-quarter to the group of small banks if s real total assets are in the bottom 75% of the size distribution, and to the group of large banks if s real total assets are in the top 5% of the size distribution. Table 2 presents summary statistics of various balance sheet ems for the obtained sample. It reports the means and medians for the full sample, and for the sub-samples of small and large banks. It also presents how composion of banks balance sheets changes from starting to ending point of my sample of bank-quarters. When comparing small and large banks one can see that small banks tend to hold more liquid assets (34.8% versus 25.63% of total assets) and less loans (54.85% versus 60.94% of total assets) in their portfolios relative to large banks. This is expected because small banks have more trouble raising external finance and they need a bigger liquidy buffer as protection against shocks to the availabily of funds. On the liabily side, small banks are mostly financed by deposs (88% of total assets) and equy (9.58%), in contrast to large banks who use deposs and equy to a smaller extent (79.89% and 6.32% correspondingly). When comparing statistics for bank-quarters in 1976 and in 2003 several patterns emerge on the asset side of banks balance sheets. First, over time the level of on-balance-sheet liquidy fell significantly not only for small banks (from 34.3% to 26.1%), but also for large banks (from 27.5% to 24%). Second, the on-balance-sheet liquidy differential between large and small banks decreased dramatically. This might be attributed to increasing availabily over time of the external financing to small banks and the evolution of the market for securized loans. Today there is less need for small banks to maintain thick liquidy buffers if they can easily obtain funds by securizing their loan portfolios. Third, the share of loan portfolios in total assets increased for both small and large banks. Finally, there is a significant decrease in the share of business loans (C&I loans) in bank loan portfolios. This decrease might be caused by the development of the commercial paper market as well as the unk bonds market Since the emergence of the commercial paper market in 1970, many large businesses swched from banks lines of cred to the commercial paper to finance their working capal. 8

11 3.2. Monetary Policy Proxies To proxy monetary policy I use three different monetary policy indicators: (i) the federal funds rate (Fed Funds); (ii) the difference between the rates paid on six-month prime-rated commercial papers and 180-day Treasury bills (Paper-bill); and (iii) the Strongin measure of monetary policy (Strongin). These indicators of monetary policy are constructed using time series data available from the Federal Reserve and are described in details in Appendix II. 12 All policy measures are transformed so that increases in their levels represent Fed tightening. They are also normalized to have the same standard deviation. 4. Measuring the Bank-Level Securizabily of a Loan Portfolio (S ) In this paper I propose a new index of securizabily of a bank s loan portfolio. The individual bank portfolio structure and economy wide securization are the crucial factors to be considered in constructing the index. Consider two banks: Bank A and Bank B. Assume that Bank A holds 80% of s loan portfolio in home mortgages and 20% in C&I loans, whereas, Bank B holds 20% of s loans in home mortgages and 80% in C&I loans. Obviously, Bank A has wider access to the securization market than Bank B, since home mortgages have been more securizable than C&I loans over the years. Following this intuion, I construct an index to proxy each bank s potential to securize s loans ( S ) in a way that captures both the composion of a bank s loan portfolio and the growth in the depth of the securization market over time. The proposed measure is computed as follows: S = 6 = 1 Economy-wide Securized Loans of Type at Time t Economy-wide Total Loans Outstanding of Type at Time t * Share of Type Loans in Bank i Portfolio at Time t (1) The index can be thought of as a weighted average of the potential to securize loans of a given type (based on market-wide averages), where the weights reflect the composion of an individual bank s loan portfolio. Thus, the market trends generate time variation in the index, whereas the differences in bank loan portfolios generate variation across instutions. I construct this measure by breaking the loan portfolio down into six categories: (i) home mortgages, (ii) multi-family residential mortgages, 12 For detailed discussion of these three proxies see Kashyap and Stein (2000) and Bernanke and Blinder (1992). 9

12 (iii) commercial mortgages, (iv) consumer cred, (v) business loans not secured by real estate (commercial and industrial loans), and (vi) farm mortgages. The index can be computed using marketlevel data from the U.S. Flow of Funds and individual bank-level data on loans from the Reports of Income and Condion. 13 Table 2 presents average S for the full sample of bank-quarters, large and small banks sub-samples, and for the beginning and ending points of the considered sample. The average securizabily of a bank loan portfolio in my sample is 10.89%. Over the sample period of 1976 to 2003 the average S has increased from about 1.7% to roughly 24.5%. This increase in the securizabily of a loan portfolio is economically significant, which once again calls for careful analysis of securization effects on a bank s operations. 5. Securization and the On-Balance-Sheet Liquidy In this section I present the analysis of the substutabily hypothesis between securized loans and liquid funds on banks balance sheets. I propose the set of empirical predictions and test them in univariate and multivariate frameworks Testable Predictions Securization is a process of creating liquid financial instruments out of assets that could be too cumbersome or expensive to sell individually. Consider the extreme case when a bank can securize the existing loans as easily as can convert liquid funds into cash. In this case, there is no need for this bank to hold liquid assets because liquid securies offer less return than loan intermediation. When facing a new lending opportuny and/or deposs whdrawal, this bank will convert a necessary amount of the existing loans into cash. The process of securizing loans, however, is time consuming and costly. Consequently, a bank needs to hold enough liquid funds to finance the new lending opportunies as well as the potential losses in deposs that might occur during this time span. In the absence of the market for 13 Appendix III discusses in detail the construction of the economy-wide time-series components of formula (1). Appendix I provides exact definions of a bank loan portfolio components. 10

13 securized loans, however, while choosing the optimal level of liquidy a bank should consider potential lending opportunies and deposs whdrawals over much longer horizon, and, consequently, hold a larger posion in liquid securies. This argument suggests potential substutabily between liquid securies and securizable loans on a bank s balance sheet that should be an increasing function of degree of access to the securization market. What about large banks and small banks? Does the expanding securization market affect those two groups differently? I believe that there exists a dynamic trade-off. On the one hand, large banks have significant advantages accessing the securization market. The process of securization involves pooling a diversified loan portfolio and packaging securies and, hence, requires significant services from the government agencies (FNMA, GNMA, Freddie MAC) and/or another pooling agencies (e.g., the investment banks). Large banks usually have a much tighter relationships wh the investment banks which provides them a direct route to the derivatives markets. These long-lasting relationships are likely to decrease the costs of securization as well as the time takes to securize loans for large banks relative to small banks. In addion, small banks tend to hold less diversified loan portfolios. As a result they are forced to attract other banks and/or pooling agencies to the securization process. In contrast, large banks can issue asset-backed securies using a pool of their own loans. On the other hand, these crucial advantages, that simplify the securization process for large banks, might not be exploed by them to a full extent since large banks tend to be more efficient in managing their liquid funds even in the absence of securization. Large banks can afford to maintain less liquid assets on their balance sheet for a number of reasons. 14 First of all, large banks usually face less severe principal-agent problem while trying to raise the uninsured funds (e.g., CDs) when compared to the small banks. Second, large depos instutions have more diversified deposors base that makes the deposs whdrawals less volatile and more predictable. Finally, large banks are more likely to have an access to significant internal capal markets. 15 These factors suggest that even though large banks have a greater 14 Similar arguments are empirically tested in Kashyap and Stein (2000) and Jayaratne and Morgan (2000). 15 See, e.g., Campello (2002) and Houston, James and Marcus (1997). 11

14 degree of access to the securization market, they might not use to reduce their on-balance-sheet liquidy since they have other channels of liquidy enhancement Univariate Tests In this sub-section, I present the results of the univariate analysis of the relationship between securizabily of bank loan portfolios, levels of on-balance-sheet liquidy and bank size. Panel A of Figure 2 graphically presents the relationship between average S and average B over time. There seems to be a tendency that the evolution of the securization market coincides wh decreasing amount of liquid funds on bank balance sheets. The time-series correlation between average S and average B in my sample is and significant at 1% level. This aggregate analysis supports the substutabily hypothesis. Table 3 presents more detailed analysis of this issue. In Panel A of Table 3, I separate my sample of bank-quarters into four quartiles based on the distribution of the on-balance-sheet liquidy measure B. I then compute the average securizabily of loan portfolios S for banks in each quartile. I repeat the analysis for various sub-periods of my sample. The results suggest that banks wh more liquid funds on their balance sheets tend to have lower securizabily of their loan portfolios, which is reflected in posive and statistically significant at 1% difference in securizabily of bank loan portfolio S between banks in the least liquid quartile and banks in the most liquid quartile of my sample as well as in s various sub-samples. This evidence supports the idea that banks tend to substute liquid funds for securizable loans on their balance sheets. Further, I analyze the relationship between the level of on-balance-sheet liquidy and the size of the banks in my sample. Panel B of Table 3 presents the average liquidy measure B across size quartiles for the full sample of bank-quarters as well as for various sub-samples. The results confirm that over the years large banks tend to maintain lower levels of on-balance-sheet liquidy than small banks. Panel B of Figure 2 graphically illustrates this relationship. In the considered sample banks in the highest size quartile have around 6.6% less total assets held as liquid securies compared to lowest size quartile banks. This is equivalent to roughly 18% less liquid funds in total assets for banks in the largest size 12

15 quartile relative to the banks in the lowest size quartile. The results are economically and statistically significant at the 1% level. The evidence is consistent wh the argument that large banks are more efficient in managing their liquid funds and, hence, extract less benef from their wide access to the securization market Multivariate Tests The univariate analysis presented in the previous sub-section suggests that the amount of liquid assets on banks balance sheets is negatively correlated wh the securizabily of bank loan portfolios and bank size. The level of on-balance-sheet liquidy, however, might also be affected by other factors. In the multivariate analysis I attempt to incorporate and adust for all factors potentially affecting the level of the liquid securies maintained by banks. To explain the variabily in the level of on-balance-sheet liquidy I use the following set of independent variables: (i) securizabily of a bank loan portfolio (S ); (ii) bank size measured by the log of real total assets; (iii) bank reputation measured by the ratio of letters of cred to total assets. The control variables are the percentage of deposs in total assets, the level of capalization measured as the ratio of equy capal to total assets, and the share of non-performing loans in the total loan portfolio. In addion, I include (a) year dummies to account for changes in the regulation and the business cycle differences, and (b) bank-specific fixed effects to account for the internal differences across banks (e.g., managerial preferences, availabily of lending opportunies, etc.). 16 According to the hypotheses of substutabily between liquid funds and securizable loans on a bank s balance sheet I expect the coefficient of securizabily of a bank loan portfolio to be negative. I also anticipate the coefficients of bank size and reputation to be negative following the argument that large, more reputable banks are generally able to maintain less liquid funds. Furthermore, since large banks and more reputable banks already have lower sustainable liquidy due to a number of reasons apart from securization, I expect the effect of securizabily of a bank loan portfolio on liquidy level to 16 To ensure qualy estimation of the bank specific fixed effects I require each bank in my sample to have at least 20 quarters of data available. 13

16 decrease wh increasing bank size and/or reputation. Thus, I anticipate the coefficient on S *Size (or S *Letters of Cred) to be posive. There exists a potential endogeney between on-balance-sheet liquidy and the securizabily of a bank loan portfolio due to the abily of banks management to choose the on-balance-sheet liquidy level and structure of a bank loan portfolio simultaneously. Furthermore, banks that prefer more liquid assets are likely to have both more liquid funds and more securizable loan portfolio (which can be achieved by, e.g., issuing more mortgages and less C&I loans). Hence there exists a potential posive bias in the relationship between B and S. To adust for endogeney I adopt two approaches. First, I implement the ordinary regression wh lagged independent bank-specific variables. Second, I use instrumental variable approach. I construct the instrumental variable for S using the methodology similar to the securizabily index formula (1) where instead of variable over time bank loan portfolio composion I use the average loan portfolio structure of an individual bank over the first four quarters available in my sample. 17 This fixed portfolio structure captures an individual bank loan specialization and at the same time eliminates the source of endogeney as well as the effect of securization on the composion of a bank loan portfolio. The constructed instrumental variable captures the changes in the securizabily index for a bank that does not change s loan portfolio structure in response to changing depth of the securization market. Furthermore, since I control for the bank specific fixed effects in the instrumental variable regressions the coefficient of S are driven by the whin bank variation in the instrumental variable I consider the average loan portfolio structure for the first four bank-quarters available for each bank in my sample to alleviate the effect of securization on a bank loan portfolio composion. A bank loan portfolio structure is less likely to be affected by securization in early bank-quarters than in recent years. For robustness I also construct the instrumental variable using the average bank portfolio composion over all available quarters for each individual bank. The results of the instrumental variable regressions wh this instrument are similar the case presented above. 18 Consider an example. Assume in 1976:I Bank A has 80% of s loans in home mortgages and 20% in C&I loans, whereas Bank B holds 20% in mortgages and 80% in home mortgages. Suppose in 2001:I both banks decide to increase their home mortgage lending by 10% (to 90% for Bank A and 30% for Bank B) in response to increasing liquidy of home mortgages due to expanding securization market. Then in 1976:I Bank A would have had S of 4.53% and Bank B S of 1.13%, whereas in 2000:I, they would have had S of 51.18% (0.9*56.37% 0.1*4.51%) and 20.06% (0.3*56.37% 0.7*4.51%) respectively. In constructing the instrumental variable I assume that the loan portfolio structure for both banks does not change from 1976:I to 2001:I. Hence, the computed instrumental variable for 2001:I is equal to 46% (0.8*56.37%0.2*4.51%) for Bank A and 14.9% (0.2*56.37% 0.8*4.51%) for Bank B. Since I control for the bank specific fixed effects in the instrumental variable regressions, the coefficient of S is driven by the whin bank variation of the instrumental variable (4.53% to 46% for Bank A and 1.13% to 14.9% for Bank B). 14

17 Panel A of Table 4 presents the results of the ordinary regressions and Panel B of Table 4 presents the results of the instrumental variable regressions. The results are consistent across Panel A and Panel B. I find that the level of on-balance-sheet liquidy is indeed negatively correlated wh the widening of the securization market. The coefficients of S have higher magnude in Panel B than in Panel A due to the potential posive bias in the relationship between B and S that is not accounted for in ordinary regressions of Panel A. The evidence suggests that as the securizabily of a bank loan portfolio increases by 1%, the level of on-balance-sheet liquidy maintained by a bank decreases on average by around 14 basis points. As I have shown in Table 2 the average securizabily of a bank loan portfolio increased from around 1.5% in the 1970s to around 25% in period. This corresponds to around 3.29% decrease in the total assets held as liquid securies, which is in turn equivalent to roughly a 36.5% decrease in the amount of liquid funds per dollar of equy held by banks. Moreover, one can look at these results from a different perspective. Since the liquid funds and loans are two dominant components of the asset side of a bank balance sheet, a 1.65% decrease in the share of liquid funds in total assets is likely to lead to an increase in the share of the loan portfolio of a similar magnude. The results of the regression analysis also indicate that, first, the amount of liquid assets tend to decrease wh an increase in bank size or reputation. Second, the availabily of access to the securization market affects the level of liquidy for large, more reputable banks to a smaller extent. Third, banks tend to hold more liquid funds when they face higher loan wre-offs. Finally, more capal constrained banks and banks wh higher share of deposs in liabilies tend to have less liquid funds on their balance sheets. The empirical evidence presented in this sub-section provides statistically and economically significant evidence that confirms the presence of the substutabily between the liquid funds and securizable loans on banks balance sheets. The evidence also shows that small banks can benef from securization as a liquidy substute on banks balance sheets to a greater extent relative to large banks. 15

18 6. The Effect of Securization on the Banks Lending Under Funding Shocks The results presented in the previous sub-section provide statistically and economically significant evidence that securization acts as a substute for liquidy on a bank s balance sheets. If true, the increasing liquidy of bank loans ought to change the link from bank funding availabily (e.g., deposs) to their willingness to supply cred. In this section I empirically evaluate this question exploing the Federal Reserve s abily to affect banks costs of funds via open market operations known in the lerature as the bank lending channel Testable Predictions The main postulation behind the lending channel of monetary policy is that by selling bonds in the open market the Federal Reserve drains the reserves of the deposary instutions, thus, causing a reduction in the availabily of insured deposs the cheapest source of the loanable funds for banks. It is not optimal for banks to completely offset this decline in deposs by borrowing directly from economic agents using the uninsured financing instruments. 19 Consequently, in the past, a bank facing tightened monetary policy would reduce lending in response to an increase in the marginal costs of raising deposs. Today, however, securization offers an addional mechanism to finance loans in the face of restricted availabily of external financing. Apart from cutting back on lending and draining down liquid funds, a bank can securize existing loans, thus, obtaining funds for new lending opportunies; can also finance new loans by securizing them. Following this intuion, I argue that under a restricted availabily of external sources of funds, a bank wh higher securizabily of s loan portfolio will experience a smaller contraction in s lending activy than a bank wh lower securizabily of s loan portfolio. 20 Furthermore, since large banks have 19 The uninsured financing instruments are not free from the tradional principal-agent problem and, hence, require the addional risk premium. CDs in excess of $ , for example, are not protected by the depos insurance and, therefore, carry more risk as well as the necessy for monoring by lenders. For detailed discussions of the sufficient condions for the existence of the bank lending channel of the monetary policy see Bernanke and Blinder (1992), Bernanke and Gertler (1995), and Kashyap and Stein (1994). 20 The existing lerature documents that the availabily of addional internal and external sources of funds partially alleviates the effect of funding shocks on the supply of loans. Kashyap and Stein (2000) document that more liquid banks are less susceptible to monetary authory moves than less liquid ones. Campello (2002) shows that internal capal markets also help shield banks 16

19 a competive advantage in their abily to access the securization market (see discussion in Section 5.1), the effect of securization on bank lending activy should be more pronounced for large banks than for small banks. The substutabily between liquid funds and securizable loans on banks balance sheets has another implication. I argue that a bank wh a wide access to the securization market can drain s liquid funds more aggressively in maintaining s loan portfolio under a funding shock, than a bank wh a limed access to the securization market, since a manager of former bank knows that he can easily replenish lost liquid funds later through securization. My strategy of analyzing the relationship between the amount of liquid funds maintained by banks, their lending behavior, and their degree of access to the securization market in the framework of bank lending responses to the external funding shock have number of advantages. First, at allows me to adust for potential endogeney. Since the Federal Reserve induced funding shocks are exogenous to financial intermediaries decisions, is possible to isolate the cross-sectional differences in the relationship between financial instutions investments (loans) and their abily to securize them (sell them) using the monetary policy as a state variable. This approach also gives me an identification strategy for the simultaney between the securizabily of bank loan portfolios and on-balance-sheet liquidy that might have not been correctly accounted for in the regression analysis presented in Section 5.3. Second, this approach allows me to test whether securization offers an addional mechanism to finance loans in the face of central bank tightening, thereby potentially weakening the link from monetary policy to loan supply Econometric Specifications In choosing the regression specification to evaluate the empirical predictions above, I start from so called univariate one-step regression specification similar to Kashyap and Stein (2000) and Ehrmann et.al. (2001). I regress the log real loan growth ( log( L ) ) against: (i) four lags of self; (ii) five lags of from the impact of increase in the costs of funds. Since securization provides banks wh an addional source of financing, my argument are in tune wh this lerature. 17

20 changes in a monetary policy indicator ( M t ); (iii) linear time trend; (iv) quarter dummies; (v) bankspecific fixed effect; (vi) liquidy of a bank s balance sheet ( B ) as well as cross effect of liquidy and 1 changes in the monetary policy indicator. 21 i 4 = ) µ M t Θ0Timet Θ1Quartert B 1( b β M t ) = 0 = 0 log( L ) = λ α log( L ε (2) Where M t is a monetary policy indicator and an increase in the level of M t corresponds to a monetary tightening. Following Ehrmann et.al. (2001) I adopt the bank-specific fixed effects since the variation in the level of liquidy as well as the composion of a banks balance sheet, growth of the loan portfolio, etc. might be affected by the bank internal factors such as a clientele base, a management team, a mainstream and availabily of the lending opportunies (e.g., individual home mortgages versus business loans). 22 The tradional theory of the lending channel of monetary policy transmission argues that a contractionary monetary impulse drains banks insured deposs and, hence, causes decrease in a bank s lending volumes. Thus, the sum of µ s should be negative. Following Kashyap and Stein (2000), the availabily of liquid funds reduces bank s loan growth sensivy towards a posive cost of funds shock. Thus, the sum of β s should be posive. To capture the securization market influence on a bank s loan growth I augment the set of the independent variables in the basic univariate regression specification (2) wh the bank-specific index of 21 In the dynamic models for panel data which contain the individual specific fixed effects lagged dependent variables become non-exogeneous if the sample has small time dimension (T). Arellano and Bond (1991) and Andersen and Hsiao (1982) propose the solution for this problem by using GMM estimation procedures. However, the lerature considers this problem to be present only in samples wh time-series number of periods below 15. Since my sample contains bank-quarters from 112 periods and I restrict each depos instution to have at least 20 quarters of data present (84% of the depos instutions in my sample are present for more than 40 quarters) I do not use the GMM procedures proposed by Arellano and Bond (1991) and Andersen and Hsiao (1982). 22 Consider an example, let us think about two banks: one wh average level of on-balance-sheet liquidy of 15% over the years (less conservative), another one wh average level of on-balance-sheet liquidy of 25% (more conservative). If the on-balancesheet liquidy of both banks spikes up to 30% in year 1, then one should treat this 30% differently for two banks. For the less conservative bank 30% implies significant increase in on-balance-sheet liquidy above preferable (or sustainable) level of 15%, whereas for the more conservative bank this increase in liquid funds is less dramatic (only 5%). Consequently, the first bank is more likely to drain s liquid funds back to 15% of total assets in year 2 by converting the liquid funds into loans, whereas the second bank might consider 15% level of on-balance-sheet liquidy below sustainable and thus issue less loans than the first one. The bank-specific fixed effects accommodate these differences. 18

21 19 securizabily of a loan portfolio S proposed in this paper, cross-effects B S, five lags of M t S 1, and five lags of t M S B 1 1 ( 4 = 0, ). t t t t t t i M d S B M c S M b B Quarter Time M L L ε ρ ξ β µ α λ Θ Θ = = = = = = ) ( ) ( ) ( ) log( ) log( (3) I argue that a bank wh higher securizabily of s loan portfolio should have smaller contraction in lending activy under a monetary tightening than a bank wh lower securizabily of s loan portfolio. Hence, I expect the sum of ξ s to be posive. In addion, if securization acts as a substute for liquidy on banks balance sheets then an increase in the securizabily of a bank loan portfolio would allow banks to dig deeper into their liquid funds to protect s loan portfolio under a contractionary monetary impulse, and the sum of ρ s should be posive. While I focus on the question of how a bank s decisions on the asset side of the balance sheet affect s loan growth, there are other mechanisms that can generate a similar effect on bank lending. Specifically, a similar effect might be generated by a bank s inabily to fulfill capal adequacy requirements. 23 To disentangle this inadequate capalization alternative, I implement so called bivariate regression analysis where I add the log real GDP growth to the set of the independent variables. 24 Adusting for the GDP growth also allows me to control for loan demand differences across quarters. t t t t t t t i S B S B GDP M d S B M c S M b B Quarter Time M L L ε θ θ θ θ ρ ξ β µ α λ Θ Θ = = = = = = = ) ( log ) ( ) ( ) ( ) log( ) log( (4) 23 A number of papers (see e.g., Diamond and Raan (2000), Hubbard, Kuttner and Palia (2002), and Sharpe (1995)) argue that insufficient bank s equy capal can be one of the restricting forces behind bank s lending activy. According to this story, monetary tightening simply raises rates and suppresses economic activy, thus, causing banks to experience loan losses and, hence, reduction in capal. This, in turn, forces weaker, more capal constraint banks to cut back on new lending. 24 See Kashyap and Stein (2000) for the discussion of the bivariate regression approach.