D o M o r t g a g e L o a n s R e s p o n d P e r v e r s e l y t o M o n e t a r y P o l i c y?

Transcription

1 D o M o r t g a g e L o a n s R e s p o n d P e r v e r s e l y t o M o n e t a r y P o l i c y? A u t h o r s Ali Termos and Mohsen Saad A b s t r a c t We investigate the response of loan growth to monetary policy shocks while controlling for loan securitization. Our major finding is that while commercial and industrial (C&I) loans and consumer loans respond to monetary policy asymmetrically according to theoretical predictions, mortgage loans show a reverse asymmetric response. In other words, while other loans are more responsive to contractionary than to expansionary shifts in monetary policy of the same magnitude, mortgage loans tend to respond inversely. Contrary to the bank lending channel predictions, expansionary monetary policy increases mortgage loan growth more than contractionary monetary policy of the same magnitude reduces this growth. We find that this reverse asymmetric response of mortgage loans is mainly driven by the securitization of these loans. Further, we show that this result is most pronounced for single-family home mortgages. We examine a well-known asymmetric effect of the transmission of monetary policy on bank lending related to loan securitization. Extensive empirical evidence shows that bank lending is more responsive to contractionary than to expansionary monetary policy (Gertler and Gilchrist, 1993; Bernanke and Gertler, 1995; Kashyap and Stein, 1995, 2000). This asymmetry is attributed to the bank lending channel of monetary policy transmission. The asymmetric effect of monetary policy on bank lending is more pronounced during recessions than during normal times as frictions generated by asymmetric information become more severe. Some researchers attribute this asymmetry in bank lending to moral hazard and agency problems that make banks more averse to issuing new loans or off-balance sheet commitments, especially during business cycle contractions (Stiglitz and Weiss, 1981; Holmstrom and Tirole, 1997; Diamond and Rajan, 2011; Caballero and Simsek, 2013). Others attribute the asymmetry to a drop in demand for bank loans during recessions (Bernanke and Blinder, 1992; Gertler and Gilchrist, 1993). The bank lending channel works effectively when the Federal Reserve Bank affects the supply of bank loans through open market operations. For example, a tightening of monetary policy contracts bank reserves and consequently leads banks to cut their lending (Bernanke and Gertler, 1995). However, these negative J R E R V o l. 3 8 N o

2 2 5 2 T e r m o s a n d S a a d shocks are not felt equally by different banks. Highly liquid banks are better able to mitigate such funding shocks than less liquid banks (Kashyap and Stein, 2000). Further, highly capitalized banks are less affected by monetary contraction than less capitalized banks (Kishan and Opiela, 2000). Moreover, small banks that are affiliated with a bank holding company (BHC) can more effectively insulate, through access to internal funds, their balance sheets from liquidity shocks than larger stand-alone banks (Peek and Rosengren, 1998; Campello, 2002). Loutskina (2011) examines the importance of bank loan securitization in shielding bank lending from monetary shocks. 1 She shows that securitization helps create a new source of liquidity that allows banks to transform illiquid loans into marketable securities. As such, securitization reduces the traditional dependence of banks on external sources of funds. She also finds that banks with more liquid loan portfolios (i.e., higher ratios of mortgages to total loans) experience a smaller drop in lending under conditions of monetary contraction than banks with less liquid loan portfolios. Loutskina s main conclusion is that the advent of securitization has not only altered the dynamics of bank liquidity management, but further weakened the bank lending channel. Securitizing banks are found to reduce their holdings of liquid securities while still able to increase their lending ability. We contribute to the literature on securitization by examining the impact of bank loan securitization on the asymmetric response of bank lending to shifts in monetary policy. We argue that liquidity, created through loan securitization, changes the contractionary as well as the expansionary monetary policy effects on bank lending in such a way that reverses the conventional asymmetry of the bank lending channel during the period of study. We postulate that securitizationinduced liquidity immunizes bank lending in response to a contractionary monetary policy and exacerbates lending in response to an expansionary policy. In effect, heavily securitized loans show a remarkable increase when monetary policy eases and decline only a little, if at all, when monetary policy tightens. Since this asymmetry occurs in the opposite direction of the conventional bank lending asymmetry, we refer to it as reverse asymmetry. We are the first to investigate the role of securitization in reversing the asymmetric response of bank lending to changes in monetary policy. We investigate the asymmetric effects of monetary shocks on each type of bank loan for the U.S. banks in continuous existence from 1985 to While these loan types vary in their degrees of securitizability, we emphasize mortgages as they lie at the center of the securitization market and are the most securitized of all loans (Exhibit 1). We test the hypothesis that mortgage loans show the most reverse asymmetry in response to monetary shocks. Consistent with our prediction, we find that in response to an expansionary monetary policy, mortgages do not just surpass other types of loans in growth, but at times continue to grow even in the face of a contractionary monetary policy. In other words, consumer, agricultural, and commercial loans show a conventional asymmetric response to monetary policy (i.e., they drop more due to a tightening than they rise due to an

3 J R E R V o l. 3 8 N o Exhibit 1 Ratio of Securitized to Outstanding Loans between 1985 and 2005 The aggregate data are obtained for the Financial Accounts of the U.S. as detailed in the Appendix. D o M o r t g a g e L o a n s R e s p o n d P e r v e r s e l y 2 5 3

4 2 5 4 T e r m o s a n d S a a d easing of the same magnitude), but mortgage loans respond asymmetrically in the reverse direction. We show that for a 1 bp rise in the federal funds rate, commercial loans and consumer loans drop by 25 bps and 10 bps, while mortgage loans rise by 7 bps. On the other hand, for a 1 bp drop in the federal funds rate, commercial loans drop by 11 bps, consumer loans rise by 37 bps, and mortgage loans rise by 72 bps. Since mortgages are classified into four types, we separately test the reverse asymmetric response to monetary shifts of single-family, multifamily, commercial, and agricultural mortgages. The degree of the reverse asymmetry in response to monetary shifts is found to increase with the securitizability of each type of these mortgages. In particular, single-family mortgages, as the most securitized among all mortgage types, exhibit the most pronounced reverse asymmetry. To examine further the impact of securitization on this reverse asymmetry, we use the securitizability index of Loutskina (2011) ( Loutskina s index hereinafter), splitting banks into high and low securitization groups. For each group of banks, we investigate the response of mortgage lending to monetary policy. The reported evidence shows that the effect of the contractionary policy is mostly insignificant for both groups; however, the effect of the expansionary policy is significant (with the correct sign) only for the highly securitizing banks. In other words, the growth of mortgage lending in response to an expansionary monetary policy is higher for the highly securitizing banks than for the other banks. Finally, we conduct four different robustness tests. The first test relates to the use of Loutskina s index as a proxy for loan securitization. Loutskina s index is utilized because securitization data at the bank level are only available beginning in In her index she uses the ratio of all securitized loans to total loans outstanding at the economy-wide level and then calculates the securitizability of each bank s loan portfolio based on this national average. While Loutskina s index provides a proxy for measuring securitization by each bank for the entire sample period, it does not reflect the actual securitization of loans, but rather the securitizability of these loans. We address this issue in two ways. First, we use the Home Mortgage Disclosure Act (HMDA) dataset that details information on loans at the application level. Given that the HMDA data clearly specify whether the individual loans were later sold or securitized, we are able to aggregate the amount of mortgage securitization at the bank level. Based on this new HMDA measure of actual mortgage securitization, we split banks into either high or low securitization groups and test for the asymmetry of each of these two groups. Our results are very similar to those obtained using Loutskina s index. Second, we modify Loutskina s index to capture only the amount of mortgages that are actually purchased and securitized by government-sponsored enterprises (GSEs), rather than the total agency and GSE-backed securities plus asset-backed security issuers. Our measure is more conservative in that it is confined to commercial banks mortgage origination and the subsequent purchases of these mortgages by the GSEs, whereas Loutskina s index captures the ratio of economy-wide mortgage securitization to the aggregate mortgages outstanding, whether these

5 D o M o r t g a g e L o a n s R e s p o n d P e r v e r s e l y mortgages are issued by banks or non-bank financial institutions. Using this modified index, the results remain robust. In the second robustness test, we examine the reverse asymmetry of loan response to monetary policy for banks that are primarily engaged in mortgage lending. Banks are ranked based on the ratio of mortgages to total loans. Banks that rank above the median ratio are classified as real estate banks and those below the median as non-real estate banks. If loan securitization is the cause of the anomalous response of mortgages to funding shocks, real estate banks would be expected to exhibit a larger reverse asymmetric response to monetary policy than non-real estate banks. Our results show that the reverse asymmetric response of mortgage loans to monetary shocks is at least twice as large for real estate banks as for non-real estate banks. For instance, in response to a cut of 25 bps in the monetary policy rate, there will be growth of 325 bps in total mortgage lending for real estate banks versus growth of only 125 bps for non-real estate banks. However, in the event of monetary policy tightening of the same magnitude, total mortgage lending continues to increase by 275 bps at real estate banks, while no significant response is observed for non-real estate banks. As for single-family mortgages, the evidence shows a strong reverse asymmetry at real estate banks while none is observed at non-real estate banks. For our third robustness test, given that loan securitization drastically increased towards the end of our sample period (Exhibit 1), we carry out a subperiod analysis by splitting the sample period into two subperiods: and As predicted, the reverse asymmetry is found to be more pronounced in the 2000s than in the 1990s across different bank size groups. Interestingly, the magnitude of the asymmetry is amplified for the largest banks. The fourth robustness test addresses the potential selection bias problem that stems from confining our data to banks that existed throughout the entire period. Here the analysis is conducted using the entire universe of commercial banks. The main findings of our study remain by and large robust to these changes. The paper proceeds as follows. We begin with a brief literature review. We then discuss the data, the econometric strategy, and the results of the various asymmetric responses of bank loans to monetary shocks. Next, we investigate the differential degree of the reverse asymmetric response of mortgage securitization to monetary shifts. We also conduct a number of robustness tests. We close with concluding. R e l a t e d L i t e r a t u r e Our paper builds on three strands of literature: the asymmetric response of loans to monetary shocks, the bank lending channel, and loan securitization. Romer and Romer (1990) examine episodes of tight money obtained from the minutes of the Federal Open Market Committee (FOMC) meetings. They find that the growth rate of money supply (M1) falls within several months of these episodes, while J R E R V o l. 3 8 N o

6 2 5 6 T e r m o s a n d S a a d bank loans do not drop until nearly six to nine months later. Bernanke and Blinder (1992) reach similar findings using innovations in the federal funds rate. They argue that a rise in the federal funds rate leads to an instant drop in bank deposits with GNP falling about two quarters later. 2 Gertler and Gilchrist (1993) examine an array of impulse response functions of bank loans to innovations in the funds rate between 1975 and They report an overall decline in mortgage and consumer loans that persists for about eight quarters after a tightening monetary shock. However, when the authors use the quarterly financial report (QFR) of manufacturing firms, the QFR bank loans appear to rise in response to a federal funds rate increase. The authors attribute this anomaly to an increase in demand for funds by manufacturing firms to finance inventories and other short-term obligations during times of liquidity crunch. Our study differs from Gertler and Gilchrist (1993) by investigating mortgage loans at the bank level rather than in aggregate, and by examining a different time period, that of We also control for mortgage securitization, which played a major role in shaping the banking industry during this period. In addition, we focus on bank loan supply rather than demand. This supply-side story of the bank lending channel also emphasizes the fact that the effect of monetary policy on bank lending is not transmitted evenly across banks. A large number of studies are dedicated to the investigation of this heterogeneous effect. Large, highly liquid, well-capitalized, or BHC-affiliate banks respond very sluggishly to tightening monetary shocks. 3 Cetorelli and Goldberg (2012) also find that international banks (i.e., those with foreign offices) are immune to monetary policy shifts as they depend on the internal capital flow from their foreign subsidiaries at times of financial distress. Considering all loans in aggregate, Kishan and Opiela (2006) suggest that the loan supply of capital-constrained banks responds more strongly to contractionary than to expansionary monetary policy. The authors also report that high-capital banks do not reduce lending in response to contractionary monetary policy, but they do increase lending in response to expansionary monetary policy. This asymmetry is especially pronounced for mortgage loans and arises due to capital constraints. Our study extends the work of Kishan and Opiela (2006) by examining the effect of securitizability, rather than capitalization, on the response of bank lending to monetary shocks. While we control for the level of capitalization in our models, we focus on mortgage lending vis-à-vis mortgage securitization. In particular, we find that mortgages, the most securitized loan type, are more responsive to expansionary than to contractionary monetary policy. In other words, mortgages exhibit a reverse asymmetric response to monetary shocks. Moreover, banks that engage in more mortgage securitization tend to exhibit a more pronounced reverse asymmetric response to monetary shifts than banks that engage in lower levels of mortgage securitization. Finally, banks with high levels of loan securitization are also found to exhibit resilience in the face of monetary policy tightening. Securitization allows banks to transform their illiquid assets into liquid assets at ease, thereby mitigating any

7 D o M o r t g a g e L o a n s R e s p o n d P e r v e r s e l y sort of liquidity drop (Estrella, 2002; Loutskina and Strahan, 2009; Loutskina, 2011). Loutskina (2011) presents a convincing argument that bank liquidity increased dramatically over the past three decades due to loan securitization. She finds that as the ability of banks to securitize loans increased over time, the holding of liquid assets on bank balance sheets has declined. Thus, securitization increased the supply of bank lending per dollar of capital and made loan growth less sensitive to the cost of funding shocks. While most studies in this line of literature consider the amount of the additional liquidity generated by securitization (e.g., Loutskina and Strahan, 2009; Loutskina, 2011; Berger and Bowman, 2012), we examine the role of this liquidity in reversing the asymmetric response of bank lending to monetary shocks. Our study is also related to the work of Aysun and Hepp (2011). They find that securitizing banks are not only more sensitive to the balance sheets of borrowers but are also more affected by monetary policy than non-securitizing banks. The impact of monetary policy is larger on securitizing banks: while loan growth is reported to be 4.5% higher than the average growth for the other subsidiaries of the same bank in different states, this relative growth is only 0.9% for nonsecuritizing banks. 4 Aysun and Hepp argue that the negative effect of securitization on the bank lending channel is counteracted by its effect on the balance sheet channel. In contrast to Aysun and Hepp, who examine the impact of securitization on the balance-sheet channel of the borrowers (the demand side), we emphasize the bank lending channel (the supply side). We also depart from them by focusing on the asymmetric response of various mortgage loans to monetary shocks given their respective degrees of securitizability. 5 We find that the effect of expansionary monetary policy increases with bank size, while the effect of contractionary monetary policy diminishes with bank size and vanishes in significance for larger banks. In the following sections, we build on these arguments to test the response of mortgage loans to monetary shocks as compared to the response of other loans. We then examine the growth of mortgage loans under both contractionary and expansionary monetary policy. D a t a A n a l y t i c s We use Call Report data on commercial banks obtained from the Federal Reserve Bank of Chicago for the period For consistency, and to assure proper lagging, only banks in existence throughout the sample period are included. Consequently, banks missing a quarter of data are dropped from the sample. This screening results in a final sample size of 5,777 banks. 7 In order to avoid bias, the period of study ends in 2005 when the subprime mortgage crisis in the U.S. was looming. Furthermore, during the sample period, the federal funds rate was the main monetary policy rate. We follow the literature in applying a standard procedure to filter the Call Report data (Kashyap and Stein, 2000; Campello, 2002; Kishan and Opiela, 2006; J R E R V o l. 3 8 N o

8 2 5 8 T e r m o s a n d S a a d Exhibit 2 Distribution of Bank Assets ($ millions) Percentile % % % % , % 2, , , % 7, , , Notes: We report the size distribution of the bank assets during our sample period, , as well as during and subperiods. On average, around 75% of the banks have an asset size less than $125 million, 95% less than $700 million, and 98% less than $3 billion. Based on these size percentiles, each quarter, banks are split into four size groups: G1, G2, G3, and G4. We refer to banks that fall below the 75 th percentile of bank assets as G1, between 75 th and 95 th percentile as G2, between 95 th and 98 th percentile as G3, and greater than 98 th percentile as G4. Loutskina and Strahan, 2009; Loutskina, 2011). We eliminate bank-quarters when a bank has (1) zero assets; (2) zero deposits; (3) been engaged in a merger or acquisition; 8 (4) a loan observation more than five standard deviations from the mean; (5) a loan-to-asset ratio of less than 10%; or (6) less than 5% of C&I loans on its balance sheet. Exhibit 2 shows skewness of the size distribution towards large banks. That is, a small proportion of banks hold the majority of the total banking assets in the U.S. Subsequently, we control for bank size to have a robust assessment of the bank lending dynamics. Exhibit 1 reports the size distribution of bank assets for the entire sample period of , as well as for the and subperiods. Over the entire sample period, about 75% of the banks have less than $125 million in assets, 95% have less than $700 million, and 98% have less than $3 billion. Accordingly, we divide banks into four size groups each quarter: G1, G2, G3, and G4. G1 refers to banks that fall below the 75 th percentile of bank assets; G2 refers to banks that fall between the 75 th and 95 th percentiles; G3 between the 95 th and 98 th percentiles; and G4 for banks greater than the 98 th percentile. In addition to the main sample, this segmentation is also carried out for each quarter for the two subperiods, and , as illustrated in Exhibit 2. 9 To highlight the comparative development of banking activities over the study period, we report the descriptive statistics of balance sheet variables of interest in

9 D o M o r t g a g e L o a n s R e s p o n d P e r v e r s e l y Exhibit 3 for 1985:Q1 and 2005:Q4. In real terms, assets have grown remarkably for each size group between 1985 and Of the four size groups, the G4 bank size group witnessed the largest real asset growth. Panel C of Exhibit 3 shows that real asset growth rates for the four classes are %, %, %, and %, respectively. Over these two decades, the banking industry in the U.S. witnessed a wave of deregulation and reform that triggered significant banking consolidation through mergers and acquisitions. 10 That translates into a more skewed size distribution of the banking industry toward larger banks. During the sample period, real equity capital surged for all banks, as shown in Panel C of Exhibit 3. This increase in capital reflects certain major regulatory developments in the early 1990s that were aimed at enhancing banking efficiency and making improvements to the oversight of the equity capital of commercial banks. After the bank loan losses of the 1980s and early 1990s, the FDIC Improvement Act (FDICIA) introduced a more stringent formula for computing capital adequacy. 11 Moreover, as Panel C of Exhibit 3 demonstrates, the ratio of mortgage loans to total loans increased dramatically from 1985 to 2005 across the G1, G2, G3, and G4 size groups by 29.21%, 32.95%, 27.66%, and 19.50%, respectively. In 2005, the ratios of mortgage loans to total loans reached a staggering 61.17%, 72.56%, 61.47%, and 41.19% for the four bank size groups, respectively. Naturally, this shift of bank assets toward mortgage loans has come at the expense of other loans. For example, the C&I loan ratio fell in 2005 when compared to For small banks, the C&I loan ratio fell from 19.30% in 1985 to 14.88% in Similarly, the ratios of consumer loans and agricultural loans to total loans have gradually dropped over time across the four bank size groups. E x a m i n i n g t h e R e s p o n s e o f B a n k L e n d i n g t o M o n e t a r y P o l i c y Te s t i n g A s y m m e t r y f o r A l l L o a n s We start by examining whether all bank loans respond asymmetrically to monetary policy. We postulate the following hypothesis. Hypothesis 1: The decline of bank lending induced by monetary tightening is higher than the increase induced by monetary expansion (i.e., all bank loans respond asymmetrically to monetary policy). We investigate the asymmetric response of varying loans across the moving four bank-size groups. We say moving to emphasize that the bank size groups are recalculated each quarter throughout the period. We employ the following specification: J R E R V o l. 3 8 N o

10 Panel A: 1985:Q1 Exhibit 3 Descriptive Statistics Bank Size Group G1 G2 G3 G4 Assets (millions, 2005 $) 40, , ,435 8,434,902 Equity (millions, 2005 $) 3,859 12,919 47, ,355 Mortgage Loans/Total Loans (%) C&I Loans/Total Loans (%) Agricultural Loans/Total Loans (%) Consumer Loans/Total Loans (%) Agricultural Mortgages/Total Loans (%) Single-Family Mortgages/Total Loans (%) Multifamily Mortgages/Total Loans (%) Business Mortgages/Total Loans (%) T e r m o s a n d S a a d Panel B: 2005:Q4 Assets ($millions) 92, ,774 3,495,734 63,963,874 Equity ($millions) 9,755 57, ,669 7,635,615 Mortgage Loans/Total Loans (%) C&I Loans/Total Loans (%) Agricultural Loans/Total Loans (%) Consumer Loans/Total Loans (%) Agricultural Mortgages/Total Loans (%) Single-Family Mortgages/Total Loans (%)

11 Exhibit 3 (continued) Descriptive Statistics Bank Size Group J R E R V o l. 3 8 N o Panel B: 2005:Q4 (continued) G1 G2 G3 G4 Multifamily Mortgages/Total Loans (%) Business Mortgages/Total Loans (%) Panel C: Change between 1985 and 2005 Percentage Change in Assets Percentage Change in Equity , Difference in Mortgage Loans/Total Loans (%) Difference in C&I Loans/Total Loans (%) Difference in Agricultural Loans/Total Loans (%) Difference in Consumer Loans/Total Loans (%) Difference in Agricultural Mortgages/Total Loans (%) Difference in Single-Family Mortgages/Total Loans (%) Difference in Multifamily Mortgages/Total Loans (%) Difference in Business Mortgages/Total Loans (%) Notes: Panels A and B report the descriptive statistics of the banks balance sheet variables of interest for the two quarters 1985:Q1 and 2005:Q4, respectively. All reported figures (apart from assets and equity) are ratios of total assets. Panel C presents the real percentage changes from 1985:Q1 for assets and equity and the difference in percentages for the loan ratios. For example, the percentage increase in assets for small banks between 1985:Q1 and 2005:Q4 is %; while, the difference in the percentage of mortgage loans to total assets is 29.21%. The dollar figures for assets and equity in Panel A are real values expressed in 2005 dollars. D o M o r t g a g e L o a n s R e s p o n d P e r v e r s e l y 2 6 1

12 2 6 2 T e r m o s a n d S a a d 4 4 L L F (F D) it j1 1 j itj j1 2 j tj 3 t 4 4[F(1 D)] t 5Xit1 6Cit1 7 jytj j1 3 j it j1 q. (1) L it stands for a real bank loan at bank i at time t. We run four separate regressions with different definitions of L. L represents variously total loans, mortgage, C&I, consumer, and agricultural loans, all in real 2005 U.S. dollars using the implied GDP deflator as a measure of inflation. The right-hand side of equation (1) includes four lags of the dependent variable, and four lags of the change in the federal funds rate (F) as a proxy for monetary policy. 13 D is a dummy variable for the monetary policy change. It is equal to one for a monetary policy tightening and zero otherwise. X is a control variable for the bank s liquidity measured by the ratio of cash and Treasury securities to total assets. 14 C is the bank s equity capital ratio. Y stands for real GDP growth to control for loan demand variation over time. q is a quarter dummy to control for seasonal effects, and is a difference operator. All financial variables are in logs. We focus on the coefficients 3 and 4. In the bank lending channel theory, both coefficients are negative and 3 is greater than 4, reflecting the conventional asymmetric response to monetary shocks. In what follows, we conduct an array of tests to detect whether the response of each individual type of loan to monetary policy conforms to this conventional view in light of the growing mortgage securitization market during the two decades of the study. Exhibit 4 reports the estimated coefficients for the baseline model described in equation (1) using a pooled GLS panel regression while controlling for bank fixed effects. All errors are corrected for heteroscedasticity and autocorrelation. Unlike mortgage loans, for G1 banks, C&I and consumer loans show negative and significant coefficients on the monetary tightening variable, meaning that as the Federal Reserve raises its monetary policy rate, the growth of these loans drops. 15 For G2 banks, C&I loans also show such a response. However, for mortgage loans, the positive sign and statistically significant coefficient mean that for small banks, these loans increase when the interest rate rises. No significant response is observed for larger size groups. The effect of the expansionary monetary policy on mortgage loans is not only significant but larger than the effect of the contractionary monetary policy, in clear contrast to the response of other loans to such shocks. In particular, this reverse response is more pronounced for larger banks. The coefficients on expansionary monetary policy for mortgage loans are significant and remain negative for all

13 D o M o r t g a g e L o a n s R e s p o n d P e r v e r s e l y Exhibit 4 Asymmetry in Loan Responses to Monetary Shocks across Bank Size Groups Bank Size Group Type of Loan G1 G2 G3 G4 Total loans F t D ** * (0.046) (0.596) (0.062) (0.657) F t (1 D) ** (0.112) (0.230) (0.156) (0.043) Mortgage loans F t D *** (0.000) (0.111) (0.521) (0.981) F t (1 D) *** *** ** * (0.000) (0.001) (0.022) (0.100) C&I loans F t D *** ** (0.001) (0.048) (0.487) (0.552) F t (1 D) *** (0.001) (0.704) (0.977) (0.875) Consumer loans F t D *** (0.000) (0.315) (0.370) (0.544) F t (1 D) ** ** (0.044) (0.720) (0.016) (0.152) Agricultural loans F t D (0.433) (0.753) (0.189) (0.676) F t (1 D) (0.254) (0.225) (0.358) (0.190) Notes: The figures reported in this table are the result of equation (1). The change in L it is regressed on four of its own lags, four lags of the federal funds rate, F t D represents contractionary monetary policy, F t (1 D) represents expansionary monetary policy, four lags of change in GDP, three quarter dummies, while controlling for bank fixed effects. L it stands for bank loan at bank i at time t. Besides total loans, at this stage L takes on four types of loans: mortgage, commercial and industrial (C&I), consumer, and agricultural. D is a dummy variable for monetary policy shift. It is one when the change in the policy rate is positive (tightening) and zero otherwise. All financial variables are in log and in real terms. P-values are in parentheses. G1 G4 indicates bank size groups, where G4 is the largest. Only the coefficients of monetary policy shifts are reported for brevity. The coefficient estimates of the other regressors are available from the authors. *Significance at the 10% level. **Significance at the 5% level. ***Significant at the 1% level. J R E R V o l. 3 8 N o

14 2 6 4 T e r m o s a n d S a a d size groups. For a 1 bp drop in the federal funds rate, mortgage lending grows by 6, 12, 22, and 32 bps at the size groups G1, G2, G3, and G4 respectively. These results clearly suggest an inverse asymmetric response of mortgage loans in the face of monetary policy shifts in a way that is contrary to the response of other loans to such shifts. For a 1 bp point rise in the federal funds rate, commercial loans and consumer loans drop by 25 and 10 bps respectively, while mortgage loans rise by 7 bps. On the other hand, for a 1 bp drop in the federal funds rate, commercial loans drop by 11 bps, consumer loans rise by 37 bps, and mortgage loans rise by 72 bps. 16 To demonstrate the economic significance of these results, consider the following figures that are based on the aggregate data from the Financial Accounts of the U.S. in 2005:Q3. 17 Total mortgage loans were $2,839.1 billion, C&I loans $1,133.7 billion, and consumer credit $708.2 billion. Consider only small banks (G1). The results presented in Exhibit 4 imply that an increase of 25 bps in the federal funds rate would lead to a $49.68 billion rise in mortgage loans at these banks, while C&I and consumer loans would drop by $34.01 billion and $17.70 billion, respectively. 18 Thus, the rise in mortgage growth is almost offset by the drop in the other loans, negating the effect of a monetary policy shift on bank lending, as illustrated in the empirical evidence of the bank lending channel (Kashyap and Stein, 2000). On the other hand, a drop of 25 bps in the federal funds rate results in a $42.58 billion rise in mortgage loans, a $31.17 billion drop in C&I loans, and a $7.08 billion rise in consumer loans. The drop in the demand for C&I loans may outweigh the drop in supply during recessions. This point is made clear by Gertler and Gilchrist (1993), who also find that following tight money, bank loans to businesses rise slightly. E x a m i n i n g t h e I n v e r s e A s y m m e t r y f o r M o r t g a g e L o a n s To understand the reasons behind the reverse response of mortgage loans to monetary policy, we focus on the fact that mortgage loans are the most securitized of all loans. Loan securitization in the U.S. is much more prevalent than in other industrialized economies, thanks to the presence of the GSEs. These institutions were intended to provide mortgage credit and to promote liquidity and stability in the secondary market for mortgages. The steady absorption of residential mortgages by the GSEs and the creation of an active and sophisticated secondary market for them encouraged commercial banks and other financial institutions to securitize residential mortgage loans. Hence, securitization became a dominant source of liquidity creation for mortgage originators, allowing banks to increasingly transform their illiquid assets into liquid assets. Consequently, the GSEs market share grew dramatically, reaching approximately 60% of the nonjumbo mortgage market and 43% of the overall mortgage market in 2003, for a total issuance of $3.6 trillion (Green and Wachter, 2005). Coupled with financial innovation, we postulate that the dramatic surge in liquidity in the mortgage market ultimately explains the reverse asymmetry of mortgage loans to monetary

15 D o M o r t g a g e L o a n s R e s p o n d P e r v e r s e l y policy. In other words, mortgage lending is far more responsive to monetary easing than to tightening. Therefore, securitization allows commercial banks to absorb liquidity shocks induced by a monetary contraction and to immunize mortgage lending to such shocks. In this section, we test for the reverse asymmetric response of mortgage loans to monetary policy. Hypothesis 2: Different mortgage loans show various degrees of reverse asymmetry in response to shifts in monetary policy. In the Call Reports, mortgage loans are decomposed into four types: single-family home mortgages (or alternatively, home mortgages), multifamily residential mortgages, business mortgages, and agricultural or farm mortgages. These different mortgage loans vary by levels of securitization. To the extent that securitization explains the reverse asymmetry, we should expect that the mortgages that are most securitized will exhibit higher reverse asymmetry in response to monetary shifts. As singlefamily mortgages were central to the activities of the GSEs, it is not surprising that these loans were the most securitized (Exhibit 1). We employ the same model as in the previous section to test the response of these four types of mortgages to monetary policy changes. The results are reported in Exhibit 5. Exhibit 5 shows that the reverse asymmetry for single-family mortgages is most pronounced for the groups G2 and G3. Despite the statistical significance of the contractionary monetary policy that the G1 group exhibits for single-family mortgages, the sign of the coefficient is positive. This indicates that single-family mortgage lending growth increases following a contractionary monetary policy shift. As for the G2 and G3 groups, the results show that single-family mortgage loans are stimulated by an expansionary monetary policy and are unresponsive to a contractionary policy. For instance, for a 1 bp drop in the federal funds rate, single-family mortgages grow by 13 bps for the G2 group and 38 bps for the G3 group. However, a 1 bp increase in the federal funds rate triggers an anomalous, positive lending growth of 8 bps at small banks and has little effect on mortgage lending at larger banks. A similar reverse asymmetric response is shown for multifamily and business mortgages for the G3 group only. Other groups show a stagnant or insignificant reaction to liquidity shocks. Finally, agricultural mortgages are not significantly affected by monetary policy shifts. The diminishing effect of monetary policy, as proxied by the federal funds rate, is consistent with the findings discussed in the literature review section. More recent evidence, however, is provided by Miles (2014). Interestingly, Miles uses time series data to determine the relative ability of the federal funds rate and the 30-year mortgage rate to predict housing prices in the U.S. The author concludes that while the 30-year mortgage rate appears to have a significant impact on home prices over the last three decades, the federal funds rate exhibits virtually no significance. These results, along with ours, support the literature on the increasingly weaker control of the Federal Reserve Bank over the housing market during the period. 19 J R E R V o l. 3 8 N o

16 2 6 6 T e r m o s a n d S a a d Exhibit 5 Reverse Asymmetry of Mortgage Loans across Bank Size Groups Bank Size Group Type of Mortgage Loan G1 G2 G3 G4 Single-family mortgages F t D *** (0.005) (0.301) (0.979) (0.213) F t (1 D) *** *** (0.126) (0.003) (0.002) (0.139) Multifamily mortgages F t D ** (0.636) (0.032) (0.475) (0.686) F t (1 D) * (0.335) (0.829) (0.096) (0.504) Business mortgages F t D ** (0.023) (0.649) (0.722) (0.157) F t (1 D) *** (0.508) (0.999) (0.004) (0.753) Agricultural mortgages F t D (0.126) (0.998) (0.998) (0.154) F t (1 D) (0.635) (0.671) (0.866) (0.852) Notes: The figures reported in this table are the result of equation (1). The change in L it is regressed on four of its own lags, four lags of the federal funds rate, F t D represents contractionary monetary policy F t (1 D) represents expansionary monetary policy, four lags of change in GDP, three quarter dummies, while controlling for fixed effects. L it stands for mortgage loans at bank i at time t. L takes on four types of mortgages: Single-Family, Multifamily, Business, and Agricultural. D is a dummy variable for monetary policy shift. It is one when the change in the monetary policy rate is positive (tightening) and zero otherwise. All value variables are in log and in real terms. P-values are in parentheses. G1 G4 indicates bank size groups, where G4 is the largest. Only the coefficients of monetary policy shifts are reported here for brevity. The coefficient estimates of the other regressors are available from the authors. *Significance at the 10% level. **Significance at the 5% level. ***Significant at the 1% level. S e c u r i t i z a t i o n a n d t h e R e v e r s e R e s p o n s e o f M o r t g a g e L o a n s t o M o n e t a r y P o l i c y In this section, we test for the correlation between securitization and the reverse response of mortgage loans (i.e., the degree to which the reverse asymmetric response of mortgage loans to monetary policy shifts is a function of loan securitization). Hence, we examine the following hypothesis. Hypothesis 3: The

17 D o M o r t g a g e L o a n s R e s p o n d P e r v e r s e l y degree of the reverse response of mortgage loans is most pronounced for highly securitizing banks. Since data on mortgage securitization at the bank level is only available beginning in 2001, we follow Loutskina (2011) by employing a securitizability index that is based on estimating the average securitization of loans at a given bank as proxied by a national average. 20 For example, in 2005:Q3, 46.37% of mortgage loans were securitized. If a bank holds $100 million of mortgage loans on its balance sheet, then the index calculates the amount of mortgage loan securitizability for this bank as $46.37 million. 21 Within each of the four size groups, we split banks into high and low classes on the securitizability index. The aggregate data on securitization from the Financial Accounts of the U.S. is merged with the bank-level data on loans from the Call Reports. A bank is ranked low in any quarter where it scores below the 50 percentile on the securitizability index, while a bank that falls above the 50th percentile is identified as high in securitizability. This classification is carried out across the four size groups each quarter. Using Loutskina s index to detect the impact of the degree of securitizability on the reverse response of mortgage loans to monetary policy, we estimate the same model as in equation (1) for mortgage loans within each size group for the two securitization classes, high and low. The results are reported in Exhibit 6. Generally, highly securitizing banks show a more pronounced response to expansionary than to contractionary monetary shifts. Most coefficients on the contractionary side are either statistically insignificant or positive, as in the cases of G1 and G2. The coefficients on the expansionary side are all significant for highly securitizing banks and have the expected negative sign. Examining the response to expansionary monetary policy, the growth of mortgage lending is higher at highly securitizing banks than at low securitizing ones within the same size group. This result conforms to the finding that securitizing banks are more sensitive to monetary policy than non-securitizing banks (Aysun and Hepp, 2011). In addition, the response of larger banks is more pronounced than that of smallor medium-size banks. For a 1 bp drop in the monetary policy rate, lending growth rises by 8, 14, 26, and 34 bps for the highly securitizing banks in the four size groups, respectively. To illustrate the magnitude of this reverse asymmetry of highly securitizing banks, we compute the asymmetry gap : the difference between the two coefficients on the two opposite monetary policy stances. Coefficients that are not statistically significant are counted as zeroes. This gap is highest for the largest bank group (40 bps), as shown in the last column of Exhibit 6. Therefore, the effect of an expansionary monetary policy increases with bank size, while the effect of a contractionary monetary policy diminishes with bank size and vanishes in significance for larger banks. These results lend support to our hypothesis that loan securitization explains the reverse asymmetric response of mortgage loans to monetary policy shifts, and consequently to the diminishing effect of the bank lending channel between 1985 and J R E R V o l. 3 8 N o

18 2 6 8 T e r m o s a n d S a a d Exhibit 6 The Effect of Loan Securitization on the Reverse Asymmetric Response using Loutskina s Index Bank Size Group Securitization Level F t D F t (1 D) Asymmetry Gap G1 High *** *** (0.000) (0.000) Low * (0.631) (0.054) G2 High ** *** (0.047) (0.000) Low (0.538) (0.493) G3 High ** (0.540) (0.018) Low (0.715) (0.484) G4 High * (0.853) (0.075) Low (0.659) (0.171) Notes: Each bank size group is split into two subgroups: banks with high securitization level and banks with low securitization. For each subgroup, we estimate the model described in equation (1). We regress the change in mortgage loans on four of its own lags, four lags of the federal funds rate, F t D represents contractionary monetary policy, F t (1 D) represents expansionary monetary policy, four lags of change in GDP, three quarter dummies, while controlling for bank fixed effects. D is a dummy variable for monetary policy shift. It is one when the change in the monetary policy rate is positive (tightening) and zero otherwise. All financial variables are in log and in real terms. The calculation of the asymmetry gaps is explained in the text. A negative gap reflects a conventional asymmetry while a positive one reflects a reverse asymmetry. P-values are in parentheses. G1 G4 indicates bank size groups, where G4 is the largest. Only the coefficients of monetary policy shifts are reported here for brevity. The coefficient estimates of the other regressors are available from the authors. *Significance at the 10% level. **Significance at the 5% level. ***Significant at the 1% level. Thus far we have demonstrated the three hypotheses outlined above: first, that mortgage loans exhibit an inverse asymmetric response to shifts in monetary policy, whereas other loans present a conventional asymmetric response. Second, among the four components of mortgage loans, that the reverse asymmetry is most pronounced for single-family mortgages. Finally, that this reverse asymmetry of loans is higher for banks with highly securitizable loan portfolios. The asymmetry gap is also found to increase with bank size.

19 D o M o r t g a g e L o a n s R e s p o n d P e r v e r s e l y R o b u s t n e s s Te s t s We conduct robustness tests in this section. We revisit Loutskina s index as an indicator for securitizability in order to examine loans that are actually securitized as opposed to those that are merely eligible for securitization. Next, we further investigate the role of loan securitization as it relates to the response of mortgage loans to funding shocks. Accordingly, we split banks into two groups: real estate banks and non-real estate banks. We conjecture that real estate banks exhibit a larger reverse asymmetric response to monetary policy than non-real estate banks. We then split the sample period into two subperiods, and , and conduct a subsample analysis. Lastly, we test for selection bias. U s i n g t h e H o m e M o r t g a g e D i s c l o s u r e A c t D a t a f o r S e c u r i t i z e d M o r t g a g e s Loutskina s index for securitizability does not necessarily reflect the actual amount of loans a given bank securitizes. While the same national average is applied to each individual bank in a given quarter, a given bank may have securitized more or less than the national average. In order to address this issue, we re-estimate the regression model using the Home Mortgage Disclosure Act (HMDA) data, which give detailed information on individual loans at the bank level. 22 We merge the HMDA data with the Call Report data. Using the newly merged dataset, we calculate the percentage of loans that were actually sold and securitized by banks during the study period. Within each of the four size groups, we split banks into either high or low securitization classes for each quarter. Banks that score below the median based on the percentage of securitized loans are ranked low, while banks that fall above the median are identified as high. The results of the estimation are reported in Exhibit 7. Confirming our initial finding, banks ranked high on the securitization index show a significant increase in their lending activity in response to expansionary monetary policy, while their lending activities are not affected by monetary contraction. Second, we modify Loutskina s index by calculating the purchase of home mortgages directly by the GSEs as an indicator of mortgages that are actually sold to the GSEs for securitization. We collect the aggregate data on home mortgage purchases by the GSEs from the Financial Accounts of the U.S. and calculate the ratio of mortgage-backed securities issued by the GSEs to the aggregate home mortgage pools issued by commercial banks. Then we multiply this ratio by the individual bank s share of home mortgage loans. Therefore, parallel to Loutskina s index, we construct the following index: J R E R V o l. 3 8 N o

20 2 7 0 T e r m o s a n d S a a d Exhibit 7 The Effect of Loan Securitization on the Reverse Asymmetric Response using HMDA Data Bank Size Group Securitization Level F t D F t (1 D) G1 High *** (0.330) (0.006) Low (0.378) (0.321) G2 High ** (0.610) (0.015) Low (0.769) (0.404) G3 High *** (0.302) (0.001) Low * (0.274) (0.332) G4 High ** (0.281) (0.050) Low (0.226) (0.122) Notes: Each bank size group is split into two subgroups: banks with high securitization level and banks with low securitization. For each subgroup, we estimate the model described in equation (1). We regress the change in mortgage loans on four of its own lags, four lags of the federal funds rate, F t D represent contractionary monetary policy, F t (1 D) represents expansionary monetary policy, and four lags of change in GDP, while controlling for fixed effects. D is a dummy variable for monetary policy shift. It is one when the change in the monetary policy rate is positive (tightening) and zero otherwise. All financial variables are in log and in real terms. P-values are in parentheses. G1 G4 indicates bank size groups, where G4 is the largest. Only the coefficients of monetary policy shifts are reported here for brevity. The coefficient estimates of the other regressors are available from the authors. *Significance at the 10% level. **Significance at the 5% level. ***Significant at the 1% level. Hit GSEs Home Mortgage Purchases at time t Total Home Mortgages Originated by all Banks at time t Share of Home Mortgages, in Bank i s loan Portfolio at time t where H it is home mortgage securitization by bank i at time t. 23 This index is more conservative than Loutskina s in that it only captures mortgages that are