The role of securitization and foreign funds in bank liquidity management

Transcription

1 The role of securitization and foreign funds in bank liquidity management Darius Martin * Mohsen Saad Ali Termos October 1, 2017 ABSTRACT Recent banking literature identifies two distinct sources of liquidity that banks can rely on in response to funding shocks: liquid funds on the balance sheet and securitizable loans. Using the Consolidated Reports of Income and Condition for all insured commercial banks in the U.S. between 1976 and 2007, we examine how domestic banks and global banks (banks with access to international capital) differ in tapping each source of liquidity. First, we find that global banks carry a less securitizable loan portfolio than their domestic counterparts. However, liquid securities and securitizable loans are more substitutable for global than for domestic banks. Keywords: Securitization; Monetary Policy Transmission; Bank Lending Channel; Liquidity JEL Classification: E51,l E52, E58, G21, G28. * Assistant Professor of Economics, Western Washington University (USA), Tel: , Darius.Martin@wwu.edu Associate Professor of Finance, American University of Sharjah (UAE), School of Business and Management, Tel: , msaad@aus.edu Assistant Professor of Finance, Qatar University, aatermos@gmail.com. 0

2 1. Introduction Recent changes in the banking industry have given banks new means to obtain liquidity to finance lending. Traditionally, bank liquidity was limited to liquid funds on the balance sheet. However, the last three decades have witnessed significant developments in securitization markets. Securitization allows banks to get liquidity by partially transforming their loan portfolio into liquid securities, thereby smoothing growth in lending. Another important channel is globalization, since global banks can transfer liquid funds from their international subsidiaries. The implications of these developments for bank liquidity management and monetary policy is the subject of a large literature. In this paper, we contribute to that literature by properly investigating the implications of both of these developments together. Securitization and globalization will both attenuate the efficacy of monetary policy provided the bank lending channel is an important monetary transmission mechanism. According to the bank lending view, expansionary monetary policy increases, and contractionary policy reduces banks supply of loanable funds. An open market sale will reduce a bank s total assets if the lost reserves cannot be replaced by uninsured liabilities such as certificates of deposit, internal transfers, or equity capital. Lending will fall in banks that lack a sufficiently high cushion of liquidity. Kashyap and Stein (2000) (henceforth K&S) show that the lending of banks with more balance sheet liquidity is less sensitive to shifts in monetary policy. They argue that this cross sectional difference in how banks respond to policy shocks is evidence in favor of a bank lending channel. However, their results are potentially subject to a negative bias if there is heterogeneity in risk aversion among bank managers. Banks with especially risk averse managers should both hold more liquidity, and lend to safer customers with more stable loan demand. In that case when there s a monetary contraction, lending doesn't fall at the more liquid banks not because they have more liquidity, but because the demand for their loans from their more stable, less risky clients doesn't change very much. 1

3 To rule out heterogeneous risk aversion, K&S do their estimations separately for large and small banks. Since large banks face fewer capital market frictions in raising uninsured liabilities, K&S argue that monetary policy should have a more limited effect on a large bank s loan supply schedule. Accordingly, the regression results for large banks should reveal the presence and direction of any bias. K&S indeed find that holdings of liquid funds on the balance sheet are not significantly related with the response of large bank lending to monetary shifts. But more recent research on bank liquidity management suggests two natural ways to enhance and verify K&S s results. First, Cetorelli and Goldberg (2012) argue that access to international capital, rather than bank size, is what insulates lending growth from monetary policy. Second, Loutskina (2011) argues that loan securitization has become a key source of bank liquidity. Suppose it is true that global, rather than large banks, are insensitive to monetary shocks. Then to rule out heterogeneous risk aversion, it is necessary to differentiate between global and domestic banks, rather than large and small banks, in examining the cross sectional differences in loan sensitivity to monetary policy. If lending contracts more at more illiquid global banks following a monetary shock, then Kashyap and Stein's results are more likely to be driven by differential fluctuations in loan demand and not loan supply. We do the test and indeed find that the level of balance sheet liquidity does not significantly affect how global banks respond to monetary policy, as expected. However, if banks can obtain liquidity from either liquid funds or securitizable loans, then the cross sectional tests should use cross-bank variation in both sources of liquidity. Accordingly, we also follow the regression framework of K&S using Loutskina s securitizability index as a measure of liquidity. In this case, we find that both domestic and global banks are more insulated from monetary policy. 2

4 We argue that this anomalous result indicates the presence of an endogenous link between the cyclical sensitivity of loan demand and the securitizability index. Since commercial and industrial loans are by far the least securitizable loan category, banks that engage primarily in C&I lending will have lower values on the index. The endogeneity arises if C&I lending is more sensitive to the business cycle than other types of loans. Our result has important implications for research that uses the index in an analysis for bank liquidity management. For example, we provide a reinterpretation of Loutskina s findings on the substitutability of liquid funds and liquid loans. We then conclude the paper with a discussion of the challenges inherent in determining the consequences of asset securitization in both banking and the effectiveness of monetary policy. The rest of the paper proceeds as follows. Section 2 describes the related literature. Section 3 describes the data. Section 4 examines the liquidity dynamics for global versus domestic banks and estimates the substitutability of balance sheet liquidity and loan securitizability for the two bank classes. Section 5 tests the degree to which holdings of either type of liquidity affect the sensitivity of bank lending to monetary policy in global and domestic banks. Section 6 concludes. 2. Literature review Much empirical research on the bank lending channel has examined the cross-sectional sensitivity of bank lending to monetary policy. 1 While Kashyap and Stein (2000) focus on liquid securities, other authors look at variation in other sources of liquidity. Campello (2002), for example, studies internal capital market mechanisms, emphasizing internal capital flows between banks that are affiliates of the same bank holding company (BHC). He shows that small banks that are affiliated with a BHC are less sensitive to cost of funds shocks than stand-alone banks of the same size. Ashcraft (2008) similarly argues that affiliation with a 1 For example, see Bernanke and Blinder (1992), Bernanke and Gertler (1995), Kashyap and Stein (1994 and 1995), and Kishan and Opiela (2000). 3

5 multi-bank holding company (MBHC) better protects banks from monetary policy than being a stand-alone bank or an affiliate of a one-bank holding company (OBHC). Holod and Peek (2010) examine the incentives of MBHCs to shift capital to their subsidiaries when facing financial constraints. The authors report that MBHCs move capital from bank subsidiaries with weaker loan origination opportunities to those with a comparative advantage in originating loans. Cetorelli and Goldberg (2012), examine the role of internal capital transfers for banks that have access to international funds in foreign offices. They show that globally-oriented banks respond to a monetary tightening by transferring liquid funds from their international subsidiaries. C&G find that banks with access to international capital are less susceptible to monetary shocks than domestic banks. Moreover, they argue that size alone does not insulate banks from monetary shocks: large domestic banks are more responsive to these shocks than global banks. C&G conclude that access to international capital, more than bank size or affiliation, helps to smooth lending growth. Other papers that examine the importance of globalization in banking include Temesvary et al. (2015), and Morais et al. (2015). Temesvary et al. (2015) find evidence that U.S. monetary easing significantly increased cross-border flows of funds from U.S. banks in the pre-crisis period. They report that a 100 basis points decrease in the U.S. Fed Funds rate leads to 9.72 to percentage points increase in bilateral cross-border lending flows. Morais et al. (2015) mapped out borrowing by Mexican firms from international banks in Mexico. The authors find that U.S. banks in Mexico extend more loans to Mexican firms following an expansionary monetary policy in the U.S. They find similar results for European banks operating in Mexico; bank lending by UK and Eurozone banks increases to Mexican firms following expansionary monetary policy or quantitative easing in both UK and Eurozone respectively. The authors report that a one standard deviation decrease in the Fed Funds rate raises the average loan volume of U.S. banks in Mexico by 7.2%. 4

6 Since the key determinant of lending sensitivity to monetary policy is liquidity, many recent papers have focused attention on the appropriate measure of liquidity. Berger and Bouwman (2009) and Loutskina (2011) distinguish between balance sheet and loan portfolio liquidity. Bank holdings of securitizable loans, such as home mortgages, are a proxy for loan portfolio liquidity. Balance sheet liquidity consists of holdings of traditional liquid funds such as Treasury securities, Federal Funds, and other money market securities. Loutskina (2011) finds that banks with more securitizable loans hold fewer liquid funds. Several studies demonstrate that increased securitization has mitigated the effect of monetary shocks over time. 2 Loutskina (2011) argues that during the 1990s and 2000s, securitization allowed banks to immunize their lending growth: under a 100 basis point increase in the Federal Funds rate, a bank with high loan portfolio liquidity would increase commercial and industrial (C&I) lending from 5.15% to 5.25% more than a bank with low loan liquidity. She reports that the magnitude of the securitization effect on lending growth is between seven and ten times larger than the magnitude of the liquidity effect shown by Kashyap and Stein (2000). Altunbas et al. (2009) make a similar argument for Europe. 3. Data and variables 3.1. Sample selection and data filtering We collect bank-level data from the Reports of Condition and Income (call reports), submitted by all insured commercial banks in the U.S. to the Fed. Our data set begins in 1976:I and ends in 2007:IV. We end our sample in 2007 for two reasons. First, this allow us to use the federal funds rate as a measure of monetary policy. Second, by restraining our sample to the pre-financial crisis period, our results become comparable to Loutskina (2011). She, as well as Cetorelli and Goldberg (2012), use a similar sample. 2 See, for example, Duffee and Zhou (2001); Loutskina and Strahan (2009); Ashcraft and Schuermann (2008); Jiangli and Pritsker (2008); and Loutskina (2011). 5

7 We screen the call reports at the bank-quarter level following Kashyap and Stein (2000). We eliminate bank-quarters when: (i) the loan-to-asset ratio is more than 100% or less than 10%, (ii) there is missing information on deposits, liabilities, or any of our control variables (which are defined in Section 3.3 below), (iii) asset growth exceeds 50%, (iv) total loan growth exceeds 100%, and (v) credit card loans are greater than 50% of total loans. Moreover, we eliminate bank-quarters when a bank was involved in a merger 3, had negative equity capital, or had more capital than total assets. We also limit the analysis to U.S. commercial banks, and drop U.S. branches of foreign banks. 4 The sample contains 21,344 banks, and 1,440,981 bank-quarter observations. Table 1 reports the number of domestic and global banks in the first and last quarters of our sample period. Following earlier studies (Campello, 2002; and Cetorelli and Goldberg, 2012), we designate a bank large in a particular quarter if its gross total assets (GTA) are above the 95 th percentile of all bank-quarter observations, and small if its GTA are below the 75 th percentile. Note that according to our classification, a bank can be considered large in one quarter and not in another. Moreover, we specify a bank as global if it reports having assets in non-u.s. offices. Therefore, global banks are U.S. banks with an international presence, rather than non-u.s. banks with a presence in the U.S. The data reveal that the total number of banks has fallen by about one half, from 14,329 in 1976 to 7,308 in This drop in the total number of banks is associated with a decline in the number of small banks, from 11,814 to 4,265, and an increase in the number of large banks, from 508 to 639. We find that global activity is mostly exclusive to large banks; in 1976, only 14 small banks reported ownership of assets in foreign offices. By the end of 2007, all small banks were domestic. We note that though the number of large banks increased over our sample period, the number (percentage) of large global banks has dropped 3 Merger files from the Chicago Fed allow us to identify all bank mergers that have occurred since These entities are classified as international banking facilities (IBF), majority-owned edge, or agreement subsidiaries. 6

8 from 132 (26%) in 1976, 125 (20%) in 1990, 97 (18%) in 2000, to 56 (9%) in Finally, the percentage of the large domestic banks increased from 74% in 1976 to 91% in Measuring Monetary Policy Our monetary policy indicator is the quarterly change in the Federal funds rate. We use this as a proxy for the monetary stance follows Bernanke and Mihov (1998) who find it to be the best measure of monetary shifts in normal times. 5 An increase in the Federal Funds rate measures an increase in the cost of external funding that is exogenous to a bank s own decisions. Since our sample ends in 2007, we avoid the challenge of measuring the monetary policy in the post-crisis period. After 2008, the Fed followed an unconventional policy of large-scale asset purchases known as quantitative easing which rendered the fed funds rate less meaningful The securitizability index Following earlier studies 7 we measure the securitizability of a bank s loan portfolio using Loutskina s index, denoted S. The index is a weighted sum of the proportion of a bank s lending in each of six types of loans: home mortgages, multifamily, commercial, and farm mortgages, C&I loans, and consumer loans. The weights represent the securitizability of each loan type, and are defined as the ratio of economy-wide securitizations of loans in a category relative to total lending in the category. Data on aggregate lending and aggregate securitization in each category come from the Financial Accounts of the United States (known formerly as the Flow of Funds accounts). 8 Formally, the index is defined as follows: 5 Previous studies (e.g., Boschen and Mills, (1995); Strongin (1995)) use other measures of monetary policy such as the spread between the rates on six-month prime rated commercial paper and 180-day Treasury Bills or the change in non-borrowed reserves. Most studies find that the Federal Funds rate is the most sensible measure for monetary policy up until See Curdia and Woodford (2011), Bauer and Rudebush (2013); and Lombardi and Zhu (2014) 7 E.g., Berger and Bouwman (2012) 8 See the Data Appendix for more details. 7

9 6 economy-wide securitized type j loans at time t S it = [ economy-wide total type j loans outstanding at t ] [ share of type j loans in bank i's portfolio at t ] j=1 The index values may be interpreted as the fraction of a bank s loans that can be securitized. For example, according to the Financial Accounts, total home mortgages outstanding in 2007:IV amounted to $11.24 trillion while home mortgage securitizations totaled $6.58 trillion. Since 58.5% of all home mortgages were securitized, the index predicts that a bank with only home mortgages on its loan portfolio can securitize 58.5% of its loans. The weights used in calculating the securitizability index are presented in Table 2. Notice that home mortgages are by far the most securitizable loan category: in 2007, 57.8% of all home mortgages had been securitized. Also, farm mortgages and C&I loans are both substantially less liquid than any other types of loans. Again in 2007, only 4.2% of farm mortgages, and 5.9% of all C&I loans had been securitized. One major advantage of using the securitizability index is that it can be constructed for the entire sample period. Though the call reports provide data on mortgage securitization at the bank level, these data are only available starting in Moreover, these data do not identify the bulk of what one would properly consider to be securitization activity. For example, one call report item that specifically addresses home mortgage securitization is item RCFDb705, the outstanding principal balance of assets sold and securitized by the reporting bank with servicing retained or with recourse or other seller-provided credit enhancements. In the filing instructions, banks are advised not to include in this item the value of mortgages sold to Structured Investment Vehicles (SIVs) for conversion into mortgage-backed securities. In our analysis, this should be considered a securitization. In the call reports, however it would be considered a mortgage sale, and data on mortgage sales are only available beginning in

10 Table 3 reports the distribution of the securitizability index (S) across time and across global and domestic banks. As in Table 1, we report the statistics for the years 1976, 1990, 2000, and Notice that the distribution of S for global banks is to the left of the distribution for domestic banks. Global banks consistently have a less securitizable loan portfolio than their domestic counterparts. For instance, in 1990, a large domestic bank in the middle of the size distribution could securitize 13.27% of its loan portfolio, versus 7.46% for a large global bank. Note that while both domestic and global banks loan portfolios have become more securitizable over time, the difference between loan portfolio securitizability for the two bank classes has diminished. However, difference-in-means tests show that the differences in the securitizability index between global and domestic banks are highly significant. These descriptive statistics show that banks with access to international capital hold fewer securitizable loans than domestic banks. In section 4, we investigate this observation with a more rigorous empirical analysis. 3.4 Control variables In our regressions, we control for bank size (SIZE), equity capital (CAPITAL), profitability (PROFITABILITY), outstanding letters of credit (LC), which, according to Loutskina (2011), is a proxy for a bank s reputation, nonperforming loans (NON_PERFORMING), and two measures of financing sources: the share of deposit financing (DEPOSITS_SHARE), and the cost of deposits (DEPOSITS_COST). We calculate the value of these controls using the call report data. SIZE is the natural logarithm of the bank s GTA. CAPITAL, PROFITABILITY, and LC are the ratio of equity capital, net income, and letters of credit relative to GTA, respectively. NON_PERFORMING is the value of all loans 90 days late relative to GTA. DEPOSITS_SHARE is the ratio of deposits to total liabilities, and DEPOSITS_COST is interest paid on deposits divided by total deposits. We also include dummy variables indicating whether the bank is a member of a Bank Holding Company (BHC) or has been recently engaged in a merger or acquisition. We use 9

11 the BHC Reports from the Chicago Fed to find the number of banking institutions controlled by each holding company. We match a bank with its holding company using the variable RSSD9348 which identifies holding companies in both the call report and the holding company data. We define MBHC equal to 1 if the bank is member of a multibank holding company and OBHC equal to 1 if the bank is a member of one-bank holding company. Finally, we define M&A equal to 1 if the bank was engaged in a merger and acquisition in the previous 12 quarters. The data appendix includes the data sources and the exact derivation of each variable. Table 4 presents descriptive statistics for the securitizability index (S) as well as the control variables. We report the statistics for banks within three size categories: those with real GTA less than $500 million, between $500 million and $1 billion, and above $1 billion. Within each size group, we report the statistics for domestic and global banks. Across all size categories, global banks are much larger than domestic banks. Focusing on banks with at least $1 billion of real GTA, global banks are around 6 times larger than their domestic counterparts, with an average size of $14.66 billion versus $2.46 billion. Global banks have fewer loans as a proportion of total GTA than domestic banks; 58.69% versus 60.21%. Across all size categories, global banks loan portfolios are less securitizable: S is smaller. Global banks are also less liquid, issue substantially more letters of credit, are less profitable, and have a lower share of deposit financing (78.36% versus 83.82% in the largest size class). 4 Liquidity dynamics of global and domestic banks We first check whether, controlling for other sources of variation, global banks do indeed issue proportionately fewer securitizable loans than domestic banks. We estimate the coefficients in the following equation: S i,t = α + β 1 GLOBAL i,t + β 2 Q i,t + β 3 (GLOBAL Q i,t ) + β 4 Controls i,t + ε i,t (1) 10

12 The subscripts i and t refer to the bank identity and the time period, respectively. Q represents balance sheet liquidity, and is the ratio of all liquid assets owned by the bank, such as Federal Funds and Treasury securities, relative to GTA. 9 S is the securitizability index. Controls includes all the variables defined in subsection 3.4. We estimate four different specifications of equation (1) for all banks in the sample. Regression results are reported in Table 5. To capture any nonlinearity in the relationship between GLOBAL and SIZE, we include the interaction of these variables in specifications (2) and (4). Specifications (3) and (4) include the indicators for holding company status (MBHC and OBHC) and recent merger activity (M&A). All specifications include time-specific fixed effects to control for an unobserved time effect on S. The dummy variable, GLOBAL, is our main variable of interest. Across all specifications, the coefficient β 1 is negative and significant at the 1% confidence level, confirming that global banks hold fewer securitizable assets than their domestic counterparts. Controlling for other sources of variation, specification (1) implies that a global bank holds on average percentage points fewer securitizable loans than a domestic bank. The results also show that SIZE is positively correlated with S, suggesting that the loan portfolios of larger firms are more liquid. The interaction term SIZE GLOBAL is positively related to S, indicating that the impact of size on loan portfolio liquidity is higher for global banks than domestic banks. Note that the coefficient β 1 is larger when we include the interaction term (specifications 2 and 4). Why do global banks have a less securitizable loan portfolio? One possible interpretation is that global banks have less of a need to hold such loans because, with access to international funds, global banks are more impervious to monetary shifts. This is consistent 9 See the Data Appendix for more details. 11

13 with much literature (e.g. Peek and Rosengren, 2000; Acharya and Schnabl, 2010; Cetorelli and Goldberg, 2012). We can interpret all of the estimation results in this light. For example, we also find that the balance sheet liquidity Q is negatively associated with S. All else constant, a one percentage point drop in balance sheet liquidity is associated with a 1.20 basis point increase in holdings of loans that can that can be securitized, such as residential mortgages. Presumably, more liquid banks have less of a need to carry securitizable loans. Note also the negative and significant coefficient on the interaction term GLOBAL Q across all specifications. Again, access to foreign funds makes global banks less exposed to monetary policy shocks. Global banks with a large buffer of liquid funds are therefore highly insulated from such shocks and have a correspondingly smaller need to make securitizable loans. Additionally, the reported results in specifications (3) and (4) show that members of bank holding companies carry less securitizable loan portfolios; OBHC and MBHC are negatively related to S. This is consistent with the results of Campello (2002) and Ashcraft (2008). Banks that are members of BHCs have access to internal capital from the head office and as such have less incentive to make securitizable loans. Note also that S decreases in equity capital, the proportion of deposit financing, the use of letters of credit, profitability, and non-performing loans. Among all control variables, S increases only in the cost of deposit financing. Finally, recent M&A activities have a negative, though insignificant, effect on S. But we could also explain this finding, that global banks can securitize less of their loans, in a different way. Recall, first, that we identify global banks as those with an office outside of the U.S., and second, that C&I loans are the least securitizable loan category. We would then expect to find a negative sign on β 1 simply if banks with foreign offices are more likely to specialize in C&I lending. This is more likely to be the case if U.S. banks do not for the 12

14 most part open foreign offices to provide retail services, but to provide commercial services and trade financing: issuing letters of credit and acceptances and so forth. Although this is consistent with our reading of Cetorelli and Goldberg (2012), we have not found much relevant research on the incentives of U.S. banks to open foreign branches. So in the following sections, we conduct more tests to determine which of these interpretations is more reasonable. 5 Foreign Funds and Securitization in Monetary Policy Again, our basic goal is to understand monetary transmission in an environment where banks can obtain liquidity from either foreign affiliates or from asset securitization. Kashyap and Stein argue that the lending of banks with more liquid assets is less sensitive to monetary policy, and they estimate their model separately for large and small banks. Since K&S assume that large banks are completely insensitive to monetary shocks, they argue that the signs of the estimated coefficients for large banks indicate the direction of any bias in their results. But recall that Cetorelli and Goldberg (2012) and others argue that global orientation, rather than size, protects banks from monetary policy. So it is natural to run regressions similar to those in K&S, except differentiating between global and domestic banks, rather than between large and small banks. We accordingly estimate the following the regression equation separately for global and domestic banks: 4 4 log L it = α + j=1 β j log L i,t j + j=0 μ j M t j + γq i,t 1 + j=0 λ j ( M t j Q i,t 1 ) 4 + δs i,t 1 + j=0 η j ( M t j S i,t 1 ) + φcontrols i,t + ρt + ε i,t 4 (3) L it is lending of bank i at time t. M is the quarterly change in the Federal funds rate, our monetary policy indicator. Note that an increase in M indicates a monetary contraction. ρ is the coefficient on a linear time trend. The remaining notation and controls are the same as before. 13

15 We are interested in the sign and significance of the interaction between changes in the monetary policy indicator with the balance sheet liquidity ( M Q) and with loan portfolio liquidity ( M S). The sum of the estimated coefficients on ( M Q ) and ( M S) measures the impact of balance sheet and loan portfolio liquidity on the sensitivity of bank lending growth to shifts in monetary policy. A positive value of the sum of lags on either term implies that the lending of banks with more of either type of liquidity is less sensitive to monetary policy. Assuming that global banks are insensitive to monetary policy to begin with, we thus expect that the sum of the lags should be positive and significant for domestic banks, and insignificant for global banks. We estimate the coefficients in equation (3) using the change in both total bank lending, and C&I lending as the dependent variable. We use C&I lending growth because K&S (2000) argue that the change in C&I loans better reflect the effect monetary policy on bank lending activity than other loans such as home mortgages. This is for two reasons: first, C&I loans are the least securitizable loan category. Second, C&I loans have a shorter maturity than mortgages and can therefore be more readily adjusted in response to changing economic conditions. Table 7 reports the results of three different specifications of equation (3): the first specification includes the interaction of the monetary policy indicator M with S only, the second includes the interaction of M with Q only, and the final specification includes both interaction terms. Note first that in all cases, the coefficients are substantially larger when the dependent variable is the change in C&I lending. This implies that in the face of funding shocks, liquidity is more important for smoothing growth in illiquid, relatively nonsecuritizable C&I loans. Confining attention first to the results for specifications (2) and (5), notice that, as expected, the sum of the lagged coefficients on M Q is significantly positive for domestic banks only. The effect of monetary policy shocks on the lending of global banks, however, 14

16 does not vary with their balance sheet liquidity. This is consistent with Kashyap and Stein, and suggests that their results are not driven by heterogeneity in risk aversion. In the remaining specifications, our results are anomalous for global banks. First, note that the sum of the lags on M S is significant and positive for both domestic and global banks. For domestic banks, this is reasonable: both balance sheet and loan portfolio liquidity should immunize lending from funding shocks. A larger buffer of liquidity in either form (liquid funds or liquid loans) helps banks shield their total lending growth against monetary shocks, thereby reducing the efficacy of monetary policy. For global banks, the results are more difficult to understand. At face value, the sensitivity of the lending of global banks to monetary policy is strongly attenuated by holdings of liquid loans, but not liquid funds. One possible interpretation is that global banks are sensitive to monetary shocks, and that their insulation against an unexpected draw on liabilities is managed through holdings of liquid loans, rather than traditional liquid funds. But it seems more reasonable to us that our results indicate the presence of an endogenous link between securitizability and the cyclical sensitivity of loan demand. Our results imply that when money is tight, lending grows more quickly among global banks with a more securitizable loan portfolio, and more slowly when money is loose. If this is due to the effects of loan demand rather than supply, then global banks with a more securitizable loan portfolio have less cyclically sensitive customers. In other words, when there's a monetary contraction, lending doesn't fall as much at the more securitizable global banks. But we argue that this is not because of their loan portfolio liquidity, but because they have more stable, less risky clients, and so their demand for loans is less sensitive to interest rates. To understand this, recall again that the securitizability index is a weighted sum of the proportion of bank lending within six different categories of loans. The weights for each loan category are the percentage of loans across the entire economy within that category that are 15

17 securitized. Of these six categories, C&I loans and farm mortgages are far less securitizable than the others. For example, Table 2 shows that in 2007, 57.8% of home mortgages and 26.3% of commercial mortgages had been securitized, but only 5.9% of C&I loans were. The index is therefore tightly correlated with a bank s involvement in commercial and industrial lending. In other words, the securitizability index may just identify banks that specialize in commercial and industrial lending. If so, then our regression results in Table 7 would imply that the demand for C&I loans are more cyclically sensitive. 6 Liquid Funds vs. Securitizable Loans In the previous section, we found that the lending of global banks with more securitizable loans grows more quickly in periods of monetary tightening. Since global banks should be relatively insensitive to monetary policy, this indicates that banks with more securitizable loans lend to less cyclically sensitive customers. Now recall that we took the securitizability index from Loutskina (2011), who used it to argue that the deepening of the securitization market reduced the amount of liquid funds that banks optimally hold. In other words, Loutskina argues that banks with more securitizable loans hold fewer liquid securities because the two types of assets are substitutable. Our results in the previous section suggest an alternative interpretation. Banks with loan portfolios that are high on the securitizability index may simply hold less balance sheet liquidity simply because they specialize in lending to customers whose loan demand is more stable and predictable. To shed further light on this matter, we replicate Loutskina s regressions, but run them separately for global and domestic banks. The regression equation is: Q i,t = α + β 1 S i,t 1 + β 2 SIZE i,t 1 + β 3 Controls i,t 1 + ε i,t (2) Controls is the same as before, containing bank affiliation, capitalization, and the availability of external funds. In estimating equation (2), we replicate Loutskina s regression framework as closely as possible using the methodology described in her paper. Following 16

18 Loutskina, we include time-specific fixed effects rather than a linear time trend, and lag the regressors to mitigate endogeneity issues. The main variable of interest is β 1, which should be negative. If the relationship between Q and S were driven by the substitutability of the two forms of liquidity, we would not expect the coefficient to be large in magnitude or significant for global banks. With their access to foreign funds, global banks can afford to be less liquid. Moreover, since global banks tend to be larger, they can more easily raise uninsured funds such as CDs to finance both lending opportunities and unanticipated withdrawals. Larger banks also have a more diversified base of depositors, and therefore a more stable demand for withdrawals. 10 Since global banks have less of a need for liquidity, there should not necessarily be any systematic relationship between the liquidity of their securities and lending portfolios. One concern is potential endogeneity in the relationship between S and Q that may result from managerial discretion. We would expect managers who prefer to maintain higher levels of liquidity to invest proportionately more in both liquid funds, and liquid loans. To address this concern, we also run an instrumental variables regression using Loutskina s instrument. This is the securitizability of a loan portfolio with a fixed proportion of lending across the six loan types. The fixed weights correspond to the bank s average portfolio structure in the first four quarters of the sample. The estimated coefficients in equation (2) are presented in Table 6. Column (1) and (4) report results of a pooled regression that includes all banks. The results are comparable, and strikingly close to those found in Loutskina (2011) 11. As expected, β 1 < 0, so S is strongly negatively correlated with Q. A one percentage point increase in S reduces Q by 5.4 basis points. (Note that Loutskina finds that the same change in S reduces Q by 5.24 basis points.) 10 See Loutskina (2011), or Jayarante and Morgan (2000). 11 Columns (1) and (4) in Table 6 may be compared with columns (1) and (6) in Table 4 on page 372 of Loutskina (2011). 17

19 The results for domestic and global banks are reported respectively in columns 2 and 3, estimated using OLS, and in columns 5 and 6 using IV regressions. Note that β 1 significant, and substantially larger, for global banks. At face value, this again implies that loan portfolio and balance sheet liquidity is substantially more substitutable for global banks. For a one percentage point increase in loan portfolio liquidity S, the level of balance sheet liquidity Q drops by 44.8 basis points for global banks compared to 37.2 basis points for domestic banks. We consider this to be good evidence that the negative relationship between Q and S is in fact driven by a negative relationship between S and the cyclical sensitivity of loan demand. Because the regression includes time effects, the results are driven by variation in S within a quarter. Note that the instrument does not capture this particular source of endogeneity. Since C&I loans have consistently been among the least securitizable, the instrument is negatively correlated with the degree to which a bank specializes in C&I lending. It is the case, however, that over time, banks holdings of balance sheet and loan portfolio liquidity have moved in opposite directions. For example, between 1976 and 2007, holdings of liquid securities in the median global bank dropped 3.42 percentage points (from 25.31% to 21.89%). For the median domestic bank, the figure is percentage points (from 37.85% to 24.13%). According to Table 3, a global bank could securitize on average 0.8% of its loans in 1976, and 23% in For domestic banks, the statistics are 1.1% and 27%. Now suppose that the drop in Q is indeed due to the rise in S. Our IV regression results reported in Table 6 imply that the increase in securitizability should correspond to an approximately 9.63 percentage point decrease in liquid funds holdings for domestic banks 12 and 9.95 percentage points for global banks. Thus, securitization would explain about 70% of the decline in liquid fund holdings for domestic banks, where 0.7 = 9.63 / But for global banks, holdings of liquid funds did not fall nearly as much as we would expect, given their increased holdings of securitizable assets. Although this may be due to an offsetting 12 To understand this, note that 9.63 = (27 1.1) = β 1 ΔS 18

20 change in the other regressors, we interpret it as evidence against a causal link between the increase in the securitizability index and the drop in balance sheet liquidity. 7 Conclusion In this paper, we investigated liquidity management in two classes of banks, domestic and global, that may invest in two different sources of liquidity, liquid funds and securitizable loans. We found that global banks on average can securitize a smaller percentage of their loans. However, among global banks, those with a more securitizable loan portfolio carry a substantially less liquid securities portfolio, and have a much smaller drop in their lending in the quarters following contractionary monetary policy. Since global banks are known to be less sensitive to monetary policy, we argue that this implies that securitizability is inversely related to this cyclical sensitivity of loan demand. This is reasonable because banks that are more specialized in C&I lending will by construction have a lower score on the securitizability index. Altogether, our results illustrate the challenges inherent in determining the implications of increased loan securitizability for monetary policy and bank liquidity management. In particular, it is difficult to draw conclusions about securitization using only cross sectional variation in bank lending across different loan categories. It may be possible to shed more light on the issue by directly measuring individual bank involvement in securitization. This is possible using call report data beginning in Further research may profitably assess whether securitization has indeed weakened the bank lending channel by studying banks actual activity in the securitization markets and how that activity responds to changes in monetary policy. 19

21 References Acharya, V., Schnabl, P. (2010). Do Global Banks Spread Global Imbalances? The Case of Asset-Backed Commercial Paper during the Financial Crisis of IMF Economic Review, 58, Altunbas, Y., Gambacorta, L., Marques-Ibanez D. (2009). Securitization and the bank lending channel. European Economic Review, 53, Ashcraft, A. (2008). Are Bank Holding Companies a Source of Strength to Their Banking Subsidiaries? Journal of Money, Credit, and Banking, 40(2-3). Ashcraft, A., Schuermann, T. (2008). Understanding the securitization of subprime mortgage credit. Foundations and Trends in Finance, Bauer, M., Rudebusch, G. (2013). Monetary Policy Expectations at the Zero Lower Bound. Federal Reserve Bank of San Francisco, Working Paper, Berger, A.N., Bouwman, C. H. S. (2009). Bank Liquidity Creation. The Review of Financial Studies, 22 (9), Bernanke, B. S., Blinder, A. S. (1992). The Federal Funds Rate and the Channels of Monetary Transmission. American Economic Review, Bernanke, B. S., Gertler, M. (1995). Inside the Black Box: The Credit Channel of Monetary Policy Transmission. Journal of Economic Perspectives, Bernanke, B., Mihov, I. (1998). Measuring Monetary Policy. Quarterly Journal of Economics, 113(3), Boschen, J., and Mills, L. (1995). The Relation Between Narrative and Money Market Indicators of Monetary Policy. Economic Inquiry, 33, Campello, M. (2002). Internal Capital Markets in Financial Conglomerates: Evidence from Small Bank Responses to Monetary Policy. The Journal of Finance, LVII(6), Cetorelli, N., Goldberg, L. S. (2012). Banking Globalization and Monetary Transmission. The Journal of Finance, Vol LXVII, NO. 5. Curdia, V., Woodford, M. (2011). The Central-Bank Balance Sheet as an Instrument of Monetary Policy. Journal of Monetary Economics, 58, Duffee, G., Zhou, C. (2001). Credit derivatives in banking: useful tools for managing risk? Journal of Monetary Economics, 48, Holod, D., Peek, J. (2010). Capital Constraints, Asymmetric Information, and Internal Capital Markets in Banking: New Evidence. Journal of Money Credit and Banking, 42(5), Jayaratne, J., Morgan, D.P. (2000). Capital Market Frictions and Deposit Constraints at Banks. Journal of Money, Credit, and Banking 32, Jiangli, W., Pritsker, M. (2008). The Impacts of Securitization on U.S. Bank Holding Companies. (Working paper). Kashyap, A., Stein, J. (1994). Monetary Policy and Bank Lending. (N.G.Mankiw, Ed.) Kashyap, A., Stein, J. (1995). The Impact of Monetary Policy on Bank Balance Sheets. Carnegie-Rochester Conference Series on Public Policy, 42(1), Kashyap, A., Stein, J. (2000). What do a Million Observations on Banks Say about the Transmission of Monetary Policy. American Economic Review, 90(3), Kishan, R., Opiela, T. (2000). Bank Size, Bank Capital, and the Bank Lending Channel. Journal of Money, Credit, and Banking, 32(1), Lombardi, M., Zhu, F. (2014). A Shadow Policy Rate to Calibrate U.S. Monetary Policy at the Zero Lower Bound. BIS Working Paper No

22 Loutskina, E. (2011). The Role of Securitization in Bank Liquidity and Funding Management. Journal of Financial Economics, 100, Loutskina, E., Strahan, P. E. (2009). Securitization and the Declining Impact of Bank Finance on Loan Supply: Evidence from Mortgage Originations. Journal of Finance, 64, Morais, Bernardo; Jose-Luis Peydro;, and Claudia Ruiz (2015), The International Bank Lending Channel of Monetary Policy Rates and QE: Credit Supply, Reach-for-Yield, and Real Effects, International Finance Discussion Papers 1137 Peek, J., Rosengren, E. S. (2000). Collateral Damage: Effects of the Japanese Bank Crisis on Real Activity in the United States. The American Economic Review, Vol. 90, No. 1, Strongin, S. (1995). The Identification of Monetary Policy Disturbances: Explaining the Liquidity Puzzle. Journal of Monetary Economics, 35, Temesvary, Judit; Steven Ongena; and Ann L. Owen (2015), A Global Lending Channel Unplugged? Does U.S. Monetary Policy Affect Cross-border and Affiliate Lending by Global U.S. Banks? Center for Financial Studies, Goethe University, Working Paper Series No

23 Table 1 Number of banks, by size and global orientation The table reports the numbers of large and small banks in the first and last periods of our sample, and in the second quarters of 1990 and Below each number we report the respective percentage within parentheses. The number of banks has decreased from 14,329 in 1976:I to 7,308 in 2007:IV. In 1976:I, out of 508 large banks, 376, or 74.02% are domestic and 132 or 25.98% are global. Note by 2007 all small banks are domestic. To calculate these numbers, we sum the number of banks that filed a call report in the indicated period. A bank is considered large (small) in any quarter when its size is greater than (less than) the 95 th percentile (75 th percentile) of the size distribution across all bank-quarter observations. Bank size is the real value of a bank s gross total assets (real GTA). 1976:I 1990:II 2000:II 2007:IV Domestic Global Total Domestic Global Total Domestic Global Total Domestic Global Total Number of Large Banks (74.02%) (25.98%) (100%) (79.74%) (20.26%) (100%) (81.87%) (18.13%) (100%) (91.24%) (8.76%) (100%) Number of Small Banks 11,800 (99.88%) 14 (0.12%) 11,814 (100%) 9,242 (99.90%) 9 (0.10%) 9,251 (100%) 7,593 (99.95%) 4 (0.05%) 7,597 (100%) 4,265 (100%) 0 (0%) 4,265 (100%) Total No. of Banks 14,329 12,191 10,270 7,308 22

24 Table 2 Securitizability Index Weights This table presents the weights ω j used in calculating the securitizability index (S) between 2000 and 2009, where 6 share of type j loans S it = ω j [ in bank i's portfolio at t ] j=1 Where the weight ω j is the fraction of all loans in the category j that are securitized across the whole economy. We calculated the weights using aggregate data from the Financial Accounts of the U.S. (The Flow of Funds). The six loan categories are home, multi-family, commercial, and farm mortgages, C&I loans, and consumer loans. For example, note that in 2009, 61.4% of all home mortgages had been securitized. Year Home Multi-family Commercial Farm Consumer C&I Loans Mortgages Mortgages Mortgages Mortgages Loans

25 Table 3 Distribution of the securitizability index This table presents the distribution of the securitizability index (S) in 1976, 1990, 2000, and 2007 for large domestic and global banks and then for all domestic and global banks. S is the value of the Securitizability index at the midpoint of the indicated interval. Below each distribution in a specific year, we report the Difference-in-means test of the Securitizability Index between global and domestic banks. t-statistics are in parentheses. *** indicates significance at the 1% level. Panel A Percentile Large Banks All Banks Domestic Global Domestic Global # of # of # of S S S Banks Banks Banks S # of Banks 1976:I Difference-in-means test 0.262*** (4.780) 0.458*** (10.352) 1990:II Difference-in-means test 6.511*** (11.438) 5.484*** (11.636) 2000:II Difference-in-means test 7.701*** (6.052) 7.097*** (6.334) 2007:IV Difference-in-means test 4.510*** (3.246) 3.808*** (3.234) 24