Market Structure, Credit Expansion and Mortgage Default Risks

Transcription

1 Market Structure, Credit Expansion and Mortgage Default Risks Liu, Bo 1 Shilling, James D. 2 Sing, Tien Foo #3 Revised: May 12, Department of Real Estate; School of Design and Environment; National University of Singapore; 4 Architecture Drive, Singapore, Telephone: ; liubo@nus.edu.sg. 2 Michael J. Horne Chair in Real Estate Studies, Department of Finance, Kellstadt Graduate School of Business, DePaul 3 Department of Real Estate, National University of Singapore, 4 Architecture Drive, Singapore Telephone: (65) ; Facsimile: (65) ; rststf@nus.edu.sg. # Corresponding author. The authors would like to thank the Department of Real Estate, National University of Singapore for financial support. We would also like to thank Anthony B. Sanders, Geoffrey K. Turnbull, Nancy Wallace for valuable discussion and conversations.

2 Abstract A credit boom has been unprecedented in the periods In a perfect competitive market, banks increase credit origination without frictions to meet increases in demand for mortgages derived either directly by borrowers (households) in the housing markets or indirectly by investors in the securitization markets. However, credit market could be reduced when the entry barrier into a market becomes more stringent. Banks earn zero normal profits are unlikely to expand credit beyond the minimum point of the U-shaped cost curve. We use the Gini coefficient and the H-statistic of Panzar-Rosse (1982 and 1987) as the measures the concentration and competition in the banking markets, and test how these two indicators affect credit supply and credit quality using samples of non-agency mortgages in the US for the periods Our results show that concentration in banking markets increases credit supply, whereas contestability causes credit contraction in the markets. We also show that high concentration and high contestability in banking markets are associated with high mortgage default risks. The sub-market analyses also show significant between-group variations among banks operating in different the market structures. The results are consistent with the concentration-fragility view, which is against creating large and less competitive banks. Keywords: Market Structure, Concentration, Contestability, Credit Expansion, Mortgage Default Risks 1

3 Market Structure, Credit Expansion and Mortgage Default Risks 1. Introduction Prior to the subprime crisis in 2007, securitization has been lauded as one of the greatest financial innovations, which increases welfare of lenders and borrowers (Hakenes and Schnabel, 2009; Gerardi, Rosen and Willen, 2010). The securitization channel has also been efficiently used by mortgage originators to facilitate credit risk transfer (CRT). The periods of dramatic boom and bust in the subprime mortgage markets between 2001 and 2007, however, unveiled structural weaknesses in the securitization model. Frictions occurred in the securitization processes, such as lax screening (Keys, Mukherjee, Seru and Vig, 2010; Purnanandam, 2010), ease of liquidating high-risk mortgages by originate-to-distribute (OTD) lenders (Mian and Sufi, 2009), among others, which have caused large-scale defaults in the subprime mortgages. 1 The past studies have established that there was a significant rightward shift in the supply curve for subprime mortgages during the periods from 2001 to Mian and Sufi (2009) showed that distribution of these mortgages were more significantly concentrated in the subprime zip-code counties relative to other counties. The explosive growth of subprime mortgages and the deterioration of the credit quality of mortgages during the periods led to the subsequent collapse of the mortgage securitization markets (Gerardi, Lehnert, Sherlund and Willen, 2008; Demyanyk and Van Hemert, 2009; Mian and Sufi, 2009; Purnanandam, 2010; and others). Many researchers have linked the increases in sub-prime mortgages to an increasing rate of securitization especially in the subprime zip-code counties. The once successfully used financial engineering mechanism has been turned into a scapegoat for causing trillions of dollars to be wiped out of the US and the global financial markets. The securitization is not a new vehicle to the market; it has been instrumental in improving liquidity in the mortgage markets in the 1980s and 1990s (Gerardi, Rosen and Willen, 2010). What has gone so terribly wrong for the securitization technology in the 2000s? There were two major structural shifts in the mortgage markets during the periods that may have significant impact on the credit expansion. First, banks through embracing advance screening technology in the 1990s have become more productivity and efficient in loan origination (Pilloff, 1999; Bikker and Haaf, 2003; Claessens and Laeven, 2004; Yildirim and Mohanty, 2010 etc). Second, with the restrictions on inter-state and intra-state branching fully removed in 1994, the role of securitization that was designed to bridge liquidity mismatches between different geographical regions has been shifted to one that facilitates CRT by banks through capital market activities. The use of technology and the ease of CRT reduce the barrier of entry into mortgage markets, which as a result, attract more OTD lenders entering the markets during the periods. Purnanandam (2010) showed significant increases in OTD loans originated by capital-constrained banks relative to their less-constrained rivals in the post-deregulation periods. The sub-prime credit expansion is no longer an independent event. The interplays of three different but integrated markets, namely the capital market (securitization), the user market (housing financing) and the bank market (origination) drive variations in credit supply in 1 Ambrose, Lacour-Little and Sanders (2005) and Gan and Riddiough (2007), however, argue against the adverse selection motive of lenders in selling high-risk mortgages. Lenders sell low-risk mortgages and retain high-risk mortgages to signal reputation. Hakenes and Schnabel (2009) explain that adverse selection will only occur, if loan quality information is private. 2

4 zip-code counties. In this paper, we will focus specifically on the investigating the effects of different competing market structure on credit supply and risks. Two sets of hypotheses are defined in our study to explain the expansion of high-risk mortgages and the default risks in markets with varying levels of competition. We use two sets of indicators to represent the competitive market structure in our study. We expand the tests of the interactive effects of the concentration and the contestability measures on the credit supply and mortgage default risks, which we will further discuss in the subsequent section. Details of the two competitive market measures are described in Appendix 1.The concentration, used in a narrower sense 2, is measured by the number of banks and their respective market share in the markets. In a broader measure of market competitiveness, Baumol (1982) and Baumol, Panzer and Willig (1982) define the contestability measure as the entry barriers into the industry. In a perfectly contestable market, banks operate efficiently with the lowest costs of production and earn zero abnormal profits. Figure 1 shows the contestability of the banking system declines from 2001 to 2006, but the bank concentration increases over the same periods. The concentration and the competition have been found to be negatively correlated in some banking literature (Hanweck and Rhoades, 1984; Rhoades, 1995; Pilloff, 1999). There are three contributions in this study. First, we show that concentration increases credit supply in the markets, whereas the contestability causes credit supply to decrease in the markets. Second, we show that high concentration and high contestability in banking markets are associated with high mortgage default risks. High contestability in banking markets leads to contraction in credit supply. Banks are unwilling to expand loan supply at the expense of increasing the marginal costs. The expanded demand in users markets could be met if new credit supply could be generated by new entrants, given that zero profit condition is not violated in the markets. Third, the results of the between-group variations suggest that there are significant interaction effects between concentrated and contestable markets, either weak or strong, on credit supply and default risks. In particularly, we found that mortgage default risks are significantly higher in the strongly contestable markets compared to other sub-markets. Policy makers should not endorse the too-big-to-fail policy merely to expand credit supply; they should also evaluate possible default risks caused by the large and inefficient banks. The paper is organized into 6 sections: Section 1 gives the motivations of the study. Section gives an overview of the banking deregulation and concentration of the mortgage markets in the US. Section 3 reviews the literature in relevant subject areas. Section 3 discusses the theoretical framework and intuition of the market structure and the growth in mortgage supply and mortgage default. Section 4 presents the empirical methodologies that include data, empirical model specifications. Section 5 analyzes the empirical results and draws inferences. Section 6 concludes the paper. 2. Banking Market Deregulation and Mortgage Markets in the US The restrictions on inter- and intra-state branching of US banks have been fully lifted following the passage of the Riegel-Neal Interstate Banking and Branching Efficiency Act (IBBEA) in 1994 (Yildirim and Mohanty, 2010). The deregulation of banking industry in the 1990s has reshaped the competitive landscape of the banking industry and the OTD mortgage 2 A highly concentrated market may not necessary imply a high barrier of entry, whereas a perfectly competitive market may also lower costs and increase barrier to new entrants. In other words, we could expect perfectly contestable in a competitive market to have the same barrier to entry as the monopolistic contestable market (Pilloff, 1999; Yildirim and Mohanty, 2010). 3

5 markets. We use the H-statistic proposed by Panzar and Rosse (PR, 1987) (see Appendix 1 for details) to measure the level of efficiency and competition in the banking markets, where a high level of competition in the market is indicated by a value of 1 or higher. Figure 1 shows that there is an uptrend in the competitiveness of the mortgage market from 2000 (0.983) before the H-statistic reaches the peak in The bank efficiency and competitiveness in the markets drop since then before hitting the low in 2006 and The erosion of competitiveness in the markets could be due to entry of hit-and-run players into the markets. After the subprime crisis in 2007, some inefficient hit-and-run players may have exited the markets, and the market competitiveness increases in [Insert Figure 1] We compute zip-code level total loans originated in the respective years based on the pooled mortgage data of sample non-agency residential mortgage backed securities and superimpose the time trends in Figure 1. 3 The new loan originated of 45,595 in 1999 grew by an explosive rate to peak at 2,382,768 in Interestingly, the loan growth is inversely related with the degree of competition in the mortgage markets. The results are consistent with Gan s (2004) prediction that more franchise value is created when the market competition decreases. 4 The negative correlations support our Hypothesis (1a). Using the 2006 data, we plot the spatial distributions of the MSA-level bank concentration (Figure 2) and the zip-code level mortgage default rates (Figure 3) on the US maps. Figure 2 shows that MSAs with high bank concentration are found mainly in the East Coast states including New York, Ohio, Virginia, and North Carolina, the Mid-Western state of South Dakota and the South-Eastern state of Alabama. California has the highest Gini coefficient in the West Coast region. At the local zip-code level, high mortgage default rates areas are clustered around the Great Lake Region, Massachusetts, New York, the Florida in the South-East, and the San Francisco s Bay Area in California. The spatial pictures show some correlations between bank concentration and mortgage default probability at the zip-code level. [Figures 2 and 3] We plot the MSA-level Federal Housing Finance Agency (FHFA) housing price indices equally-weighted by states in the respective strong and weak contestability and concentration categories in Figure 4(a) and 4(c). There were significant upward trends in housing prices in all market categories since The price trends in the strongly concentrated market deviated significantly from the price trends in the weakly concentrated market over the periods. We also computed weighted average coupons (WAC) for sub-markets of strong and weak contestability and concentration, and plot the WAC time series over the periods in Figure 4(b) and 4(d). Sharp declines in the WAC, which is a proxy for securitization market risk, were clearly seen between 2000 and 2004, which were also the periods witnessing high securitization activities. We also observed stronger upturns in the WACs in both weakly concentrated and contestable markets during the periods [Insert Figure 4] 3 4 The strong growth trend was consistent with the evidence in other earlier studies, such as Mian and Sufi (2009), although different dataset was used in our study. Some studies, such as Pilloff (1999), Smirlock (1985), also show that the presence of big banks improves the profitability of other banks in the local markets. 4

6 The preliminary evidence above suggests that there are some correlations between market structure and credit supply and risk. More empirical analyses will be conducted in the subsequent sections to test the significance of causality from banking market structures to credit supply and mortgage default probability. 3. Literature Review The removal of inter- and intra-state branching restrictions resulted in more small banks being merged and acquired by large banks leading to a more concentrated banking industry (Yildirim and Mohanty, 2010). The banking deregulation has produced significant welfare gains in terms of accessibility to credit and lower costs of credit for disadvantaged families, and advancement in technology and product innovations for lenders. The income distributions of lower wage workers are also tightened in a more competitive banking environment (Beck, Levivne and Levkov, 2010). A deregulated banking market helps reduce borrower constraints and market imperfection (Gerardi, Rosen and Willen, 2010). There are two competing hypotheses that predict how the presence of big banks could shape the competitive landscape in the banking industry. First, the structure-conduct performance hypothesis argues that big banks will collude to charge monopolistic prices on mortgages, and thus increase profits of the banks at the expenses of borrowers (Gilbert, 1984; Berger, 1995; Wang, 2003; DeYong, Kiler, and McMillen, 2004; Hakenes and Schnabel, 2009; Dell Ariccia, Igan and Laeven, 2008; Dick and Lehnert, 2010). Second, the competing efficient market hypothesis predicts that if bank operate more efficiently, the profitability of banks increases when competition become keener in the market. The factor differentiating the two competing hypotheses lies with the question of whether market power (concentration) matters in determining the competition in the local banking markets (Smirlock 1985). On the negative aspect of the bank deregulation, Dick and Lehnert (2010) found that increased competition prompts large and more efficient banks to adopt sophisticated screening technologies, which result in credit to be extended to previously rationed borrowers. The increases in the supply of consumer credit cause a significant rise in bankruptcy rate. Competition in the banking market could also be detrimental to the banks effort in acquiring borrower-specific private information. Gan and Riddiough (2008) argue that incumbent banks in concentrated markets with full access to privilege information on prime borrower credit use limit pricing strategy to deter new entrant to prime markets. However, they price discriminately low-credit loans using public observed information. The two-tier pricing strategies allow them to trade-off their information rents for monopoly rents in the prime loan markets. The strategies resulted in higher market shares and credit spreads of prime loans relative to subprime loans in highly concentrated banking markets. Ogura (2010) further argues that inside banks offer interest rates on loans that are lower than the comparable rates offered by outside banks to acquire exclusive access to customer s private information. The banking market structure plays an important role in determining the mortgage risks and also mortgage supply in the market. The proponents of the concentration-fragility hypothesis argues that high concentration gives banks a too big to fail status, which leads large banks to originate risky mortgages is important (Boyd and De Nicolo, 2005). Large banks use market power to limit competition through various measures, such as credit rationing (Rothschild and Stiglitz, 1976; Stiglitz and Weiss, 1981), access to 5

7 borrower-specific private information (Gan and Riddiough, 2007; Ogura, 2010), strict underwriting process (Dell Ariccia, Igan and Laeven, 2008), and credit risk transfer (CRT) (Hakenes and Schnabel, 2009). Gan (2004) using the housing shocks in Texas during the periods as a natural experiment showed the fragility of highly risk-taking banks in the concentrated banking markets of Texas. The opposing view by Beck, Demirguc-kunt, and Levine (2006) supports the concentration-stability hypothesis, when they showed that concentrated banking systems reduces systematic risks using cross-country data for the periods from 1980 to Other than the mortgage suppliers concentration, the concentration of high-risk loans in at the zip-code and neighborhood levels was also found to have significant effects on mortgage default risks. Mian and Sufi (2009) found that mortgage increases in subprime zip-code counties were twice as much as the increases in prime zip-code mortgages during the periods from 2002 to They credit growth coincided with increases in securitization activities contributed to the sharp rises in defaults rate in the subprime zip-code counties in Agarwal, Ambrose, Shomsisengphet and Sanders (2009) when testing mortgage default in Phoenix, Arizona found high concentration of subprime mortgages in selected zip-codes, especially in the lower income neighborhoods and the Central Business District. Their results, however, showed no significant relationship between concentration of the subprime mortgages and default probability. They found, however, that credit supply led to rising housing price in the high subprime loan concentration neighborhoods. The concentration of banking structure, if leads to reduced competition in the market, will have some implications for mortgage market activities. In this study, we will test the relationships between the concentration and contestability levels of local banking markets and credit growth and default probabilities of mortgages. 4. Interconnectedness of Mortgage Markets 4.1. A 4-Quadrant Model for the Credit Market In our study, we assume that credit expansion is not an exogenous event. We modify and adapt the theoretical framework of 4-quadrant model for real estate asset and space markets by DiPasquale and Wheaton (1996) to illustrate the interconnectedness of the securitization (investors), origination (lenders) and mortgage financing (borrowers) markets (Figure 5). Economic activities in three distinct but inter-connected markets could influence credit supply (aggregate stock and flow of mortgages) in the market. Specific focus, however, is placed on the market structure of the originators (banks) and how it will influence the costs of capital (price) and the credit origination (supply) activities. [Insert Figure 5] The top right-hand (North-East) quadrant is the borrower (user) market, where the demand and supply of new mortgages is regulated by mortgage interest rates, i. If there were no income shocks, the supply growth is induced through lower interest rates (shift along the demand curve), i = D -1 (S), where denotes price elasticity, and S denotes the aggregate mortgage stocks. Moving in the counter-clockwise direction to the top left-hand (North-West) quadrant, the securitization market consists of investors and issuers of mortgage backed securities. The pricing elasticity of mortgage cash flows, which are represented by the weighted average coupon (WAC), reflects the risk premiums, r, placed on mortgage backed 6

8 securities in the securitized markets r =i/p, where P denotes the security price. The bottom left-hand (South-West) quadrant represents the loan origination markets. The supply of new loans, F, is a function of the capital costs of lenders, P, which equals the price of capital set in the securitization market. By keeping the risk premiums in securitization markets constant, the competition and contestability of the banking markets could affect the capital costs of lenders, which in turn influence the credit supply in the market, which is a function, F = C(P). In equilibrium, the supply in the origination market and the demand in the user market should match such that the aggregate stocks of mortgages as in the bottom right-hand (South-East) quadrant of Figure 5 can be determined. The mortgage stocks are dependent on new loan originated (inflow), F, and loan foreclosure or default, S, which is S = F - S. If default rate were higher than the new loan origination growth rate, the supply elasticity will be negative, [ S < 0], which implies a decline of outstanding mortgages in the market, and vice-versa. The above dynamics assumes no increases in welfare in the credit markets as the squared area in Figure 5 is unchanged. However, demand shifts (borrower s market) and reduction in risk premiums in the securitization market can induce changes that are associated with increase in credit market welfare. More complex dynamics could also arise as a result of interactions of activities in different market Testable Hypotheses The earlier literature on the structural-conduct performance hypothesis and the efficiency hypothesis invariably assumes that competitive banks could either charge monopolistic price, or increase profits via efficiency gains when market competition intensifies. Gan and Riddiough s (2008) model that allow limit pricing by large banks also makes the same assumption about the positive normal profits. The assumptions are inconsistent with Baumol s (1982) conditions for perfectly contestable equilibria, in which firms (banks) will always charge contestable price that is equal to marginal costs. In other words, banks in a perfectly contestable market should earn zero or negative normal profits. Therefore, market power does not exist for contestable banks (firms), whether large or small. Banks in a perfectly contestable market are not able to use cross-subsidy or predatory pricing tools as unfair market entry deterrent strategies. Before defining our testable hypotheses, we identify the different market effects for contestability and concentration of banking markets using the 4-quadrant mortgage and securitization markets framework. In a contestable market, banks do not change their production function in short-term as they are supposedly operate in the most efficient cost environment. The different levels of contestability (strong and weak) are represented by the shift along the credit supply curve in the South-West quadrant. In a strongly contestable market, banks operate efficiently and charge the lowest interest rates on new mortgages originated. They will not expand the credit supply beyond the optima (minimum) marginal costs. However, in a weakly contestable market, credit supply growth could be provided by relatively less efficient (high-cost) hit-and-run banks that will enter the market as long as they can earn positive normal profits. Therefore, the level of contestability in the mortgage market could be represented by the credit supply line, where strongly contestable market is denoted by a shift along the line towards the origin, and the weakly contestable market shows a shift but in an opposite direction away from the origin. The above contestability conditions also 7

9 imply that the mortgage market stock is reduced in a strongly contestable market (Figure 6a), whereas in a market with inefficient banks, the credit market is expanded (Figure 6b). [Insert Table 6] On the conditions on market concentration, we assume that banks could exploit market power to charge cut-throat pricing or to use predatory prices to deter new entrants. This is represented by a rightward shift of the supply line in the South-West quadrant for a strongly concentrated market with few big banks (Figure 6c). In a weakly concentrated market, where there are many small and competitive banks, the cost of production is higher because small banks could not exploit economies of scale production technology. The mortgage market shrinks and the supply function shift to the left (Figure 6d) Based on the four different market scenarios shown in Figure 6 on how changing lenders market structure could affect the credit supply and stocks, we define two sets of hypotheses to predict how market structure influences lenders behavior on credit supply and mortgage default. The first set of hypotheses relates contestability of market to credit supply and risks. In a strongly (perfectly) contestable market, banks operate efficiently, such that they originate mortgages at a rate (price) at least equal to the marginal cost. If competition increases to an extent that drives marginal costs down, hit-and-run banks that could not earn above zero normal profits may exist the market. The credit market shrinks, and some borrowers will be rationed-out of the market. Hypothesis (1a), if not rejected, implies that credit supply decreases when the market contestability increases (i.e. the market becomes more efficient), while keeping the demand shifts constant (no change to household income). 5 However, banks operating in a weakly contestable market increase credit supply by extending risky (subprime) loans to previously excluded (unqualified) borrowers. The alternative hypothesis thus predicts that mortgage supply increases in a less efficient market with a corresponding increase in the mortgage loan interest rate. Hit-and-run OTD lenders will enter the inefficient market and earn above zero normal profits. The effects of contestability on credit risks are not as clear cut as expected (Pilloff, 1999). 6 In a strongly contestable market where banks originate loan based mainly on the minimum cost consideration, a high proportion of high-risk mortgages is expected. Hypothesis (2a), if not rejected, thus implies that mortgage default risk is higher in a highly contestable banking market where banks are less discriminatory on credit quality. However, the alternative hypothesis argues that banks in a weakly contestable market are less willing to take additional risks at zero normal profits. They adopt advance loan screening technologies to support discriminatory pricing strategies, such that the proportion of under-priced high-risk loans will be small in the weakly contestable market. The second set of hypotheses involves testing market concentration structure on credit supply and risks. Gan and Riddiough (2008) were the first to examine how concentrated market structure affects pricing strategies and market shares of prime and subprime mortgages. They argue that incumbent banks choose pooled pricing strategy to preempt entry into the prime 5 6 The contestability hypothesis is not inconsistent with the negative relationships between credit growth and income growth in subprime zip-code counties, and also the flat price increases in the elastic sub-prime Metropolitan Statistical Areas (MSAs) during the credit expansion periods as in Mian and Sufi (2009). Dick and Lehnert (2010) when testing the US credit card markets found a negative relationship between banking market competition and bankruptcy rate. 8

10 loan market, and as a result, a high fraction of prime to subprime loan was observed in markets with highly concentrated banks. Based on the results, we state our Hypothesis (1b) that banks with larger market power (strong concentration) are more able to increase credit supply relative to markets with many small competitive banks. Large (monopoly) banks are also more likely to adopt cut-throat pricing strategies to expand their market shares. In contrary, the alternative hypothesis argues that smaller banks do not have economics of scale, the costs increases when the level of concentration decreases, which could then result in decreases in credit supply. If aggressive pricing strategies of monopoly banks either through pre-emptive (cut-throat) pricing, or cross-subsidizing sophisticated borrowers by exploiting naïve borrowers through confusing mortgage design that generate to (Gabaix and Laibson, 2006), which as a result increase default risk of mortgages. Hypothesis (2b), if not rejected, thus suggests that default risks of mortgages increase when bank concentration in the market increases. Hypothesis 1a: The level contestability in the banking market is negatively correlated with credit supply in the mortgage market. The Hypothesis (1a), if not rejected, implies that a strongly contestability market have relatively lower credit supply than weakly contestability market. Hypothesis 1b: The level of concentration in the banking market is positively correlated with credit supply in the mortgage market. The Hypothesis (1b), if not rejected, implies that credit supply in a strongly concentrated banking market is higher relative to a weakly concentrated banking market. Hypothesis 2a: The level of contestability of the banking market is positively correlated with the mortgage default risks Hypothesis 2b: The level of concentration of the banking market is positively correlated with the mortgage default risks. Both Hypotheses (2a) and (2b), if not rejected, predicts that mortgage default risks increase in markets where efficient banks earn zero profit, and/or where few large banks use cut-throat or predatory pricing strategies to expand credit and deter new entrant. Hypothesis (2b) supports the concentration-fragility argument that large banks are anti-competitive and they are too big to fail from the regulator s views. The market concentration may not necessary increase competitiveness and efficiency in the lenders markets. We could observe few large banks in weakly contestable markets, and similarly, we could also have many small banks but they operate highly efficiently (strong contestability). We thus include interactive variables, both in continuous and discrete terms, to tests the relationships between market structure and credit supply and risks. The four types of banking market structure can be represented below: 9

11 Market concentration (Gini Coefficient) Strong (top 10% deciles) Weak (lowest 10% deciles) Market Contestability (H-statistic) Strong (top 10% deciles) Weak (lowest 10% deciles) Monopolistic Monopolistic inefficient contestability (natural monopoly) Competitively Competitively inefficient contestability (perfectly competitive) 5. Empirical Tests and Methodology 5.1. Mortgage Data We collected mortgage-level data from 6000 non-agency residential mortgage backed securitization deals from Bloomberg. 7 The data were available for the periods from January 1991 to June The cut-off date for individual mortgage origination was December We also obtained loan-level information, which includes the origination mortgage balance, current mortgage balance, initial loan-to-value (LTV), mortgage interest rate, mortgage type, mortgage interest rate reset status, mortgage age, and FICO credit score of borrowers. After removing mortgage samples that do not have LTV data and other key information, a total of 7,108,724 sample mortgages was included in our subsequent empirical analysis Empirical Variables In additional to the contestability and concentration variables (see Appendix 1), there are two other sets of variables that will be used in our empirical tests. In the test of the first hypothesis on the market structure and credit supply relationships, we will use mainly the zip-code level variables as our control variables to explain variations in the zip-code level mortgage supply. The control variables for the second test on the market structure and mortgage default relationship are defined mainly from loan-level mortgage characteristics. Table 1 summarizes the list of the variables together with the descriptions Mortgage-level variables [Insert Table 1] Based on the mortgage contract, we sort our mortgage sample into an adjustable rate mortgage (ARM) or a fixed rate mortgage (FRM), which is represented by, where = 1, if it is an ARM, and otherwise, = 0, if it is a FRM. We convert the loan-to-value (LTV) and FICO score data of individual mortgage samples into categorical variables, which will be used in the proportional hazard risk models. Seven discrete categories based on the original LTV ratios of the mortgages are created, LTV i, where i = [1, 2, 3, 4, 5, 6, 7], in a descending order as follows: [LTV1 > 100%]; [100% LTV2 90%]; [90% LTV3 80%]; [80% LTV4 70%]; [70% LTV5 60%]; [60% LTV6 > 50%]; and [LTV7 50%] as the reference group. By borrowers FICO, four categorical variables are derived, FICO i, where [i = (1, 2, 3, 4)], such that [FICO1 634]; [634 < FICO2 686]; [686 < FICO3 736]; and [FICO4 > 736]. A dummy variable, NOFICO, is used to represent sample borrowers who 7 Compared to the database used in Keys, Mukherjee, Seru, and Vig (2010), which consists of a sample non-agency loans with outstanding balance amounting to 1.6 trillion, our sample non-agency loans with a cumulative outstanding balance of 1.77 trillion adding up to 2008 is reasonably representative. 10

12 do not report the FICO scores in the data. As in the competing mortgage risk model of Deng, Quigley and Van Order (2000), we include a call option (PPAY) variable and a put option (PNEQ) variable. The probability of a mortgage having a negative equity for a sample mortgage k originated at time t, PNEQ k,t, is defined, as in Calhoun and Deng (2002), as a function of the current market value of a collateralized house, (CURMKV), the current outstanding mortgage balance, (CURAMT), and the house price volatility, p,t : PNEQ,,, P, (1) CURMKV, ORGAMT, LTV, P P O (2) where [.] is the standard normal cumulative distribution function. The current market value for k-th house at time t, CURMKV k,t, is not recorded in the data, and we compute it by assuming that the housing price, which is computed by dividing the original loan amount by the original LTV, that is ORGAMT k,t=0 /LTV k,t=0, drifts from at a rate measured by changes in Housing Price Index of the Federal Housing Finance Agency (FHFA), P t, the origination date t=0. We also compute the call option value, which is the difference between the contract interest rate of mortgage k at time t, R k,t, and the prevailing market interest rate at time t, R M, divided by the volatility of the 6-month LIBOR rate, R (source: Bloomberg): PPAY R, R M, R, (3) The non-linearity effects of the option values are indicated by (PNEQ k,t ) 2 for the put option value, and (PPAY k,t ) 2 for the call option value. In order to capture the cross-market variations in housing prices, we use an index-adjusted housing pricing variable that is free from collinearity problem in our default hazard model, which is defined as: = CURMKV, P ORGAMT, LTV, P O (4) where CURMKV k,t is the current value of housing underlying mortgage k at time t, and P t is the FHFA housing price index in quarter t Zip-Code Level Variables We derive five zip-code level mortgage variables weighted by the city-level or aggregate values at the zip-code level of the comparable mortgage variables. The zip-code level variables marked with a C as the initial of the variable identifier capture spatial variations in the local mortgage markets. The subscript i in the variable, [i = (zip, city)], indicates variables based on aggregate values estimated from mortgage samples identified by both the 11

13 zip-code, zip, and the city within which the zip-code is located, city. LAMTzip Weighted loan value (CL): CL (5a) LAMT Rzip Relative mortgage rates (CR): CR (5b) R LAMTARM Proportion of ARM by original loan value (CARM): CARM LAMT city city zip zip (5c) Proportion of low FICO score borrower (CLFICO), : CLFICO NHLTV Proportion of high LTV loans (CHLTV): CHLTV N zip zip NLFICO N zip zip (5d) (5e) where LAMT i is the aggregate loan origination amount; R i is the mean mortgage interest rate at origination; LAMTARM i the loan amount by mortgage type that is [LAMTARM i = LAMT i ], where = 1, if it is an adjustable rate mortgage (ARM), and otherwise, = 0, if it is of the fixed rate mortgage (FRM) type. The variables (d) and (e) are the proportion of mortgages of selected types, [NLFICO i and NHLTV i ], divided N i, that is the total number of mortgages at zip-code i, where NLFICOi counts the number of low FICO mortgages cut-off by a FICO score of less than or equal to 638 ( 634 FICO1 and :634<FICO2 638 subgroups); and NHLTV i counts the number of high LTV mortgages with a LTV cut-off of more than 90% ( LTV1>100% and 100% LTV2>90% subgroups) Descriptive Statistics Full Sample Table 2 summarizes the descriptive statistics for the key empirical variables defined in the early sections. The non-agency pooled of mortgages consists of 48.84% of ARM, and the remainders of 51.16% were made up of FRM. The average origination LTV for the full sample is 75.91%. The FICO score averages at 682 and the variance is 82.63, which imply that the credit quality of average pooled mortgage is higher than subprime mortgages, which is usually cut-off at 660 (or 620 in Keys, Mukherjee, Seru, and Vig, 2010). For the two option variables, the probability of negative equity is estimated at on average, and the prepayment probability is higher at on average. [Insert Table 2] We compute five zip-code level variables to describe the mortgage characteristics distributions in different zip-code counties. By loan amount at origination, the average per zip code loan origination relative to the city-level aggregate is about 1.69%. The average relative loan rate of 1.00 indicates no significant difference between zip-code and city-level average loan rates. The average CARM of 53.94%, which is the proportion of ARM originated by 12

14 local banks at the zip-code level over the total loan by banks in cities measured in terms of original loan amount, shows that the numbers of local (zip-code level) banks that originate more ARM than FRM are above the city-average. In terms of credit quality, the CLFICO of 42.74% and CHLTV of 13.50% indicate that the proportions of banks originating high FICO and low LTV loans are below the city-averages. The zip-code statistics imply significant neighborhood variations at the local zip-code level relative to the city average Sub-market Analysis For sub-market analyses of the relations between market structure and credit supply and mortgage characteristics, we sort our sample markets into two distinct submarkets based on the medium concentration (Gini coefficient) and the medium contestability (PR H-statistics) measures at the state-level. A weak market includes sample states ranked in the lower 50% percentile of the distributions; whereas a strong market includes the top 50% percentile states by the concentration and contestability statistics. The descriptive statistics are summarized alongside the full-sample statistics in Table 2. Whilst we observe no significant variations in terms of composition of ARM and FRM types in mortgage originations of banks in different markets; the results show, however, significant variations in terms of credit quality of the loans originated. By LTV and FICO criteria, loans are risky in markets where banks are weakly contestable and concentrated. The loan LTV increases from 75.21% to 77.04% and from 74.81% to 77.27% when bank market structures change from strong to weak in terms of contestability and concentration. We include tests of difference-in-difference between strong and weak sub-markets, and the t-test results are highly significant rejecting all hypotheses that the weak and strong submarkets based on the two treatment effects (concentration and contestability) are not different Zip-Code/ State Level Analysis The results of the two sub-market panels (b) and (c) in Table 2, one sorted by concentration (Gini) and another sorted by contestability (H-STAT), show significant variations between highly efficient (contestable) banks and large (concentration) banks. In the middle panel (b) where the samples are divided using the medium Gini coefficient into two sub-groups, the top 50% percentile of samples (strong concentration) has a mean Gini coefficient of compared to the mean Gini coefficient of for the lowest 50% percentile (weak concentration) sub-samples. Large banks (strong concentration) may possess market power, but they have the lowest contestability statistic of , compared to for sample banks in weakly concentrated markets. In panel (c), we sort the zip-code counties using the medium of the PR H-statistics into strongly contestable (mean H-statistic of ) and weakly contestable (mean H-Statistic of ) sub-samples. The concentration statistics for the strongly and weakly contestable sub-markets are estimated at and respectively. Next, we analyze variations in zip-code level loan characteristics in the concentration and contestability sub-markets. The loan origination amount (CL) by banks in strongly concentrated markets expressed as a ratio of city-level loans origination averages at 1.23%, which is half of the sample averages of the 2.44% of the loan amount originated by banks in weakly concentrated market. In the contestability sub-market the difference in loan amount between strong (1.50%) and weak (1.93%) submarket is relatively smaller. The zip-code average mortgage rates for loans originated by banks in the strong concentration and 13

15 contestability submarkets are lower at and compare to the rates of and in the weakly concentration and contestability submarkets. The compositions of ARM mortgages originated by banks in the strong sub-markets are relatively higher with the CARM estimated at 55.12% (concentration) and 55.40% (contestability). Banks in weakly concentrated and contestable markets have the highest proportion of high risk loans, which include 44.77% and 43.86% (CLFICO values) of loans with FICO below 686; and of 14.87% and 16.29% (CHLTV values) of loans with LTV above 90%. In comparison, the lowest CHLTV values of 12.65% and 11.25% in strong concentration and strong contestability counties imply that large and highly efficient banks are more risk-averse and stricter in processing and originating high LTV loans Time-series Analysis of Credit Supply and Risks Mian and Sufi (2009) and Demyanyk and Van Hemert (2009) and other researchers argued that sudden surge in subprime mortgages in periods has contributed to the subprime outbreak in Did loan characteristics at the zip-code level change drastically during the periods? We plot four zip-code level loan variables on a yearly basic across the sample periods from 1999 to We present the results for the submarket analysis based on the contestability criteria in Figure 7. 8 [Insert Figure 7] The results show that the mortgage interest rates decline across time from 1999 to 2003 in all three contestable markets. The interest rates for the periods from 2003 to 2006 increase, but interest rates in the three submarkets show more divergence during the periods. The FICO scores of borrowers in the three submarkets generally trend upward over the periods The zip-code LTV trends mirror the interest rate trends, which saw LTV declines from 1999 and hits the lowest in The LTV then increases again in 2004 and the submarket variations widen during the periods On the share of ARM loans, increasing number of ARM loans were originated especially in the periods after The deviations of CARM across the three submarkets increase during periods. 5. Empirical Methodology and Testing As shown by the preliminary statistics in the earlier section, significant variations are found in loan characteristics across states/zip-code counties with different banking market structure. The market power (concentration) and efficiency in banking industry (contestability) have different impact on mortgage origination activities in lenders markets. In the following sections, we will describe our empirical methodologies designed to test the two hypotheses relating to the market structure Hypothesis 1: Banking Market Structure and Mortgage Supply In Mian and Sufi (2009), they found that mortgage supply increased sharply in the subprime zip-code countries, which is based on the lowest zip-code quartile in the distributions of subprime loans (FICO below 660) in their sample, at a rate that was twice as high as the credit growth rate in other prime countries during the They also found that income in these subprime zip-code counties decreased during the same periods. The rapid 8 As the results of time series trends of the four zip-code level variables in concentration sub-markets are not significantly different, we will include the comparable time series results where mortgages are sorted by the concentration criteria. 14

16 expansion seemed to be coincided with the increasing securitization activities in the markets. In our first hypothesis, we attempt to test the causal-relations between zip-code level mortgage supply and local market factors. Instead of assuming that the zip-code level credit supply and securitization are exogenous, we allow our zip-code variations in local markets to be endogenously driven by different market structure and changes in local market (supply origination (supply) activities. We define our supply as flow activities based on the amount of new loan origination each year in each zip-code. A total of 234,079 zip-code-year observations is generated for the zip-code level mortgage supply computed in logarithm term and is used to represent the dependent mortgage supply variable, F it, in the pooled regression below: F it 0 1Cit 2Zit 3Yt it (6) where the zip-code-year observations are indexed by the subscript zip-code i and year t in the pooled regression. The regression model includes two vectors of regressors, C it and Z it, and a year vector, Y t. C it represents the state-level market structure through a vector of variables, which include the PR H-Statistics and Gini Coefficient, their respective square terms and also an interactive term, [HSTAT Gini]. We also replace the continuous variables by discrete measures of contestability and concentration, where strong and weak indicators are represented by dummy variables indicating the top 10% deciles and the lowest 10% deciles of the respective indicators. The interactive terms are included to capture the joint effects of contestability and concentration of markets. Z it is a vector of zip-code level variables, which include two zip-code to city-level ratio variables that are the relative loan number, CL it, relative weighted mortgage interest, CR it, and three zip-code level share variables in terms of compositions of ARM, CARM it, low FICO mortgage, CLFICO it, and high LTV mortgages, CHLTV it. We first run the regression using the full sample mortgages, and then we repeat the regressions using submarket samples (divided by medium samples). Two sets of sub-market analyses are included for testing of two supplementary Hypotheses 1a and 1b. The first set of tests sorts the sample into strong, medium and weak markets using the state-level Gini coefficients, and the second set of tests sorts them into the same three submarkets but based on the PR H-statistics of contestability. The empirical results are summarized in Table 3. The regression models have strong goodness-of-fit with R 2 ranging between (column 3) and (column 2) for the full-sample models. A large majority of the estimated coefficients in the models are significant, where some of them have p-values of less than [Insert Table 3] We first examine the results for various market structure variables as shown in Panel (A) of the Table. The full-sample results (column 1) show significant but opposing effects of market power (Gini Coefficients) and efficiency in the markets (PR H-statistics) on the mortgage supply at the zip-code level. The coefficient on the contestability variable is negative, but weakly significant at a 10% level. The coefficient on the non-linear contestability term is positive and significant, which shows a downward convex relationship of the market contestability on mortgage supply. The coefficients on the concentration variable and the squared term are both significant, but the signs are different. There is positive correlation between credit supply and the level of concentration in banking market. The mortgage supply 15

17 is a positive concave function of the concentration of banking market, i.e. the credit growth increases at a decreasing rate when banking concentration increases. On the interactive term, the coefficient is significantly negative, which implies that in counties with few large banks and entry barrier is high, credit supply is likely to be small compared to other less contestable markets with weakly concentrated banks. The results support Hypothesis (1a) that mortgage supply decreases in local markets with highly concentrated banks; and Hypothesis (1b) that credit expansion is more likely to occur in markets with few large banks. We further examine the hypotheses using the top and lowest deciles of H-Statistics and Gini coefficients to create four different dummies to represent strong and weak markets in terms of contestability and concentration. The coefficients on the two contestability dummies are positive highly significant, but the credit supply is significant higher in the weakly contestable markets (H-STAT 10%) (0.3358) than the strongly contestable markets (H-STAT 90%) (0.0661). Hypothesis (1a) is not rejected as the results did suggest a negative correlation between credit supply and the degree of contestability in the markets. On the coefficients for the two concentration dummies (Gini 90% and Gini 10%), both are negative and significant. The lower value of for the weakly concentrated market relative to the value of for the strongly concentrated market again implies a positive slope in the credit supply curve in relation to the level of market concentration. Thus, the results support Hypothesis (1b), which implies that there is a positive relationship between credit supply and market concentration. In column 3, we add four interactive terms to further separate the market structure effects, and each interactive term represent one market structure type: monopolistic contestability (H-STAT 90% * Gini 90%), monopolistic inefficient (H-STAT 10% * Gini 90%), competitive contestable (H-STAT 90% * Gini 10%) and competitively inefficient (H-STAT 10% * Gini 10%) (see Section 4.2). The results show that the interactive term representing monopolistic contestable market has a negative but significant coefficient. All other three terms have positive coefficients. The results indicate that in markets where entry barriers are high, smaller mortgage market is expected when banks acquire more market power. A contestable market with only few large banks, mortgage rationing policy is more likely to be adopted that may lead to a reduction in smaller mortgage stocks. However, when there are many banks that possess limited market power, credit expansion is expected in the market through more aggressive loan origination strategies of banks. In our sub-sample analyses, we use the medium H-Statistic and Gini coefficient to sort the sample into a strong sub-market and a weak sub-market in each of the market structure. The results (columns 4 and 5) show that the coefficients on the Gini term and the squared Gini term are consistent with the full-sample estimates. However, the contestability measure has stronger impact on banks in highly concentrated markets ( ) compared to banks in weakly concentrated markets ( ). The entry deterrent strategy will have stronger adverse impact on the credit supply in markets with few big banks. The negative interactive term on the strongly concentrated sub-market affirms the effects. On the contestability sub-markets analyses (columns 6 and 7), we found that market concentration increases credit supply in both strong and weak concentration sub-markets. However, the coefficient on the Gini for the strongly contestable market (6.2630) is smaller than the coefficient (8.1936) for the weakly contestable market. The results suggest that market power has positive and significant effects on credit supply in a market with low entry barriers. In the models, we also control for other zip-code and city-level mortgage effects on mortgage 16

18 supply in the respective zip-code counties. The results are consistent in the full-sample and the sub-sample analyses. The relative loan number, CL it, and relative mortgage rate, CR it, are negatively related with the mortgage supply in the zip-code counties. Mortgage supply is significantly higher in zip-code counties with high fraction of ARM loans, CARM it. The proportion of high-risk loans (low FICO), CLFICO it, and high LTV loans, CHLTV it, will have negative effects on mortgage supply. The zip-code and city-level coefficients are significant and stable after controlling for the levels of contestability and concentration in the banking market in the respective zip-code counties. In terms of year effects, we observe significant and positive increases in credit supply during the periods 2002 to 2007, when 2008 is controlled in the models as the reference year. The periods showed slower growth in credit supply compared to the reference year The supply trends were consistent with the evidence of credit expansion during the periods found in Mian and Sufi (2009) and other researchers Hypothesis 2: Banking Market Structure and Mortgage Default To test Hypothesis 2 that banking market structure affecting mortgage default risks, we use the proportional hazard model proposed by Deng (1997) and Deng, Quigley and Van Order (2000) with an expanded set of determinants. We include, in additional to the standard variables such as mortgage characteristics, competing option values and housing value, state levels banking market structure and zip-code level credit supply variables in the mortgage default hazard function. We eliminate possible endogeneity in the mortgage supply and market structure variables by using a two-stage least squares (2SLS) approach, where the predictive values of mortgage supply estimated from the earlier Equation (6) are included as the explanatory variable in the second stage default hazard function. The 2SLS model is defined as follows: ˆ F it 0 1Cit 2Z it 3 Y (6 ) t λ t λ t; X λ t exp X β Y β C β F β (7) where 0 (t) is the baseline hazard that is related with the year of amortization (age) of mortgage I; X i is a vector of exogenous variables that include two categorical variables: LTV and FICO score, a dummy on ARM type,, a continuous relative housing price variable,, and year dummies from 1999 to 2008; Y i include default and prepayment option values as in Equations (1) and (5) respectively; and C i are the market concentration and contestability variables (both the continuous and discrete measures) including the non-linear and interactive terms; and Fˆ is the predicted mortgage supply variable derived from Equation (6 ). ' i In Equation (7), the proportional hazard, i (t), is a function of t, which is defined as the duration from the date of mortgage origination to the first date of occurrence of mortgage foreclosure, or the right-censoring date of our sample data that is June 2009, whichever is the earliest. The records of month-by-month payment and foreclosure for our sample mortgages are available for a 2-year window before the June 2009 cut-off date. For sample mortgages 9 We regressed the mortgage growth on year-on-year changes in the market structure and zip-code / city-level loan characteristics. The results show that annual rates of change in the levels of contestability and concentration in the banking markets and the interactive term have significant positive effects on annual growth in credit supply. We the marginal change in market contestability has negative impact on the supply in the markets when banks moderately concentrated. The results remain consistent, but they are not shown in the paper due to space constraints. 17

19 have matured and/or foreclosed more than 24 months from the cut-off date, the first date of the 2-year window is used to define the foreclosure. 10 Based on the predicted mortgage supply in Equation (6 ), we estimate the vector coefficients in the proportional hazard model (10), j where j = [1, 2, 3, 4], using the partial likelihood method of Cox (1972), with the full-sample and the sub-sample data. The results are summarized in Table 4. [Insert Table 4] We first analyze the results in Panel (C) of Table 4. All the estimated regression coefficients on predicted mortgage supply and market structure variables are significant, but the signs vary across different sub-markets, which suggest different borrowers default behaviors in sub-markets with different banking concentration and competition. In contrary to the supply-based hypothesis of Mian and Sufi (2009), our results show that the predicted credit expansion have negative effects on the mortgage default hazard in all markets, except for the weakly concentrated market, where banks are perfectly competitive. The supply-driven default story is thus only supported in sub-markets served by a large number of banks that possess no market power. On the two variables for concentration and contestability in the banking markets, we found positive and significant effects of the two variables on mortgage default probability. The effects are, however, non-linear. The negative and significant interactive term suggests that there are positive concave relationships between mortgage default risks and the levels of market concentration and contestability. The results do not reject the Hypotheses 2(a) and 2(b), which suggest that banking market structure has significant positive impact on mortgage default risks, after controlling for the credit supply in the model. The interactive term is significant, but the coefficient has a negative sign. The results may suggest that mortgage default rate is lower in markets with few large efficient banks operating at the lowest marginal costs. We further investigate different market structure effects using the discrete dummies that represent the top and the bottom deciles of the samples based on the H-Statistics and Gini measures. In columns (2) and (3), we found that the coefficient on the strong contestability market dummy was significant and positive. The results imply that in markets where barriers of entry are high and banks earn zero profits, the probability of mortgage default was relatively higher compared to other submarkets. The coefficients on the other three market structure dummies are though significant and negative. The risk of mortgage default was the lowest in the markets with many small but profit maximizing (competitive) banks. We observe a clear increasing relationship in values of the coefficients from the weak to strong sub-markets. The results are thus not inconsistent with Hypotheses 2(a) and 2(b). When we include the interactive terms in column (3), the results show that the natural (inefficient) monopoly submarket with few large banks with low entry barriers was the worst performer in term of mortgage default risks. The competitively contestable submarket consisting of many banks with no controlling market power, but originate mortgages efficiently at the lowest marginal interest rates, is likely to have lower mortgage default risks. 10 The dead mortgage samples are treated as cases with delayed foreclosure, which may bias downward (underestimate) the probability of default. It will, however, not have significant effects on the overall results. We conducted robustness tests by using only the most recent sample mortgages originated not more than 10 years ago, and the results, which were not included in the paper due to space constraints, are in general consistent with the main findings in this paper. 18

20 By subdividing the mortgage samples at the medium points of the H-Statistics and Gini coefficients, we test between-group variations in the mortgage default risks in markets with strong and weak concentration and contestability structures. We found significant but contrasting coefficients between strong and weak submarkets in both the concentration (Columns 4 and 5) and contestability (Columns 6 and 7) structures. In a highly concentrated sub-market (Gini>50%), the contestability in the banking market reduces the mortgage default risks. In contrast, high contestability could result in high default risks for mortgages originated in a weakly concentrated market. The squared H-Statistic terms also show opposite signs for the strong and weak contestability market. The results suggest a clear divergence of the mortgage default probability in the two sub-markets when the level of contestability increases. The interaction of bank concentration and contestability are positively significant in explaining default risks in the two submarkets. When we examine the strong and weak contestable submarkets (columns 6 and 7), the results show that the market concentration has significant but positive effects on the mortgage default risks in the strong contestable submarket. The opposite effects of concentration were found in the weakly contestable submarkets. Again, the non-linear terms also show opposite signs, which imply divergence in the mortgage default risks when concentration increases in both sub-markets. The interactive terms in contestable submarkets are both negative but significant. The coefficients for the mortgage and borrower variables in the competing risk models are stable across different sub-samples. Using [LTV7 50%] as the reference group, the coefficients on the LTV2 to LTV6, except for the [LTV1>100%] variable, are positive and significant indicating higher risks when LTV increases. On the borrowers credit records, the borrowers with no FICO records and those with FICO of 736 and below have relatively high mortgage default risks, when the [FICO4 736] group is used as the reference group. ARM mortgages are more risky relative to mortgages of FRM type. The coefficients on the two option variables and the squared terms are also significant, and the signs are consistent across different sub-market samples. The probability of negative equity, PNEQ, and the prepayment probability, PPAY, both increase the default risks of the sample mortgages. The effects are, however, non-linear as shown by the negative signs of the squared terms that are significant. The year dummies show that significant rates of increases of default probabilities over the years. Using 2008 as the reference year, the coefficients on year dummies 1999 to 2003 were negative in most of the models, but the magnitude of coefficient decline over the years. The coefficients on the year dummies were positive and significant after 2004 culminating to the peak in 2007, which highlighted clearly the rising trends in mortgage default risks in the samples Robustness Tests Effects of Legislation Risks Ambrose and Pennington-Cross (2000) found that the judicial foreclosure laws and deficiency judgments influence the market shares of the conventional loans and the Federal Housing Administration (FHA) insured loans originated by the government sponsored enterprises (GSEs). We use Ambrose and Pennington-Cross s (2000) definitions to sort US states into four broad groups with different foreclosure and deficiency judgments, that are Q1 includes states with non-judicial foreclosure but allow deficiency judgments; Q2 includes states with non-judicial foreclosure and no deficiency judgment; Q3 includes states with 19

21 judicial foreclosure and allow for deficiency judgments; and Q4 includes states with judicial foreclosure but no deficiency judgments. We then repeat the estimation for mortgage supply (Equation 9 ) and mortgage default hazard models (Equation 10) in a 2SLS process. The results are summarized in Tables 5 and 6. We found significant variations in how contestability and concentration in banking market structure affect the supply (origination) of non-fha in our models in different jurisdictions. In states with judicial foreclosure rules [Q3 and Q4], more efficient and competitive banks can cause decreases in credit supply. However, banks acquiring market power could increase credit supply, and the effects are the strongest in states with judicial foreclosure, but do not deficiency judgments, [Q4]. When we include the predicted credit supply into the step-2 proportional hazard models, the results in Table 5 show that higher default risks are correlated with credit expansion in markets that allow deficiency judgments, [Q1 and Q3].The competition and concentration of banks have significant positive effects in all states, except for Q1 states that have non-judicial foreclosure laws but allow for deficiency judgments. Our market structure results could not be separated from the legislative risks in different states in the US, but our results supporting hypotheses that concentration and competition in banking market do influence credit supply and risks remain unchanged Credit Expansion Phases [Insert Tables 5 and 6] Mian and Sufi (2009) have argued that the sharp increases in credit during the periods significantly caused sharp increases in mortgage defaults in We test whether the rapid credit expansion was associated with changing market structure in the banking markets. To test the robustness of the results, we split our sample periods into three phases, which include the slow growth phase from 1998 to 2001, the exponential growth phase from 2002 to 2006, and the post financial phase from 2007 to We repeat the 2-step estimation procedures, and the results are summarized in Tables 5 and 6. Interestingly, we observe that large bank and more efficient banks caused mortgage supply to decrease in the early slow growth phase We do not see strong competition in banking to cause sharp increases in mortgage supply in the high growth phase. The Gini coefficient is positive and significant, which show that the credit expansion was caused by the present of large banks. However, the the negative interactive term implies that large banks, if operate more efficiently, cause credit supply to decrease. We thus cannot rule out the hypothesis the bulk of the credit expansion in the periods were caused by the origination activities of inefficient banks. In the mortgage hazard model (Table 6), our results were not consistent with the supply story of Mian and Sufi (2009), which hypothesizes that the large credit supply in the periods was associated the subsequent increases in mortgage default. The coefficients on the predictive mortgage supply was negative in the rapid growth phase, but was positive in the post-crisis phase. Our results, however, show that the presence of either large banks or highly efficient banks leads to high default in the markets during the two periods and If large banks operate efficiently, the default rates in the same periods could be reduced significantly. Our sub-period results are not inconsistent with the two hypotheses on market structure effects on credit supply and risks. 6. Conclusion 20

22 A credit boom has been unprecedented in the periods In a perfect competitive market framework, we invariably assume that credit supply and growth are exogenously. Banks could increase credit origination without frictionl to meet increases in demand for mortgages derived either directly by borrowers (households) in the housing markets or indirectly by investors in the securitization markets. However, in a market where banks earn zero normal profits, banks are unlikely to expand credit beyond the minimum point of the U-shaped cost curve. Therefore, in a highly contestable banking market, credit market could be reduced when the entry barrier becomes more stringent. In contrary, in a market dominated by few big banks, banks could operate inefficiently by using cut-throat (limit) pricing on prime loans to deter new entry Gan and Riddiough (2008). In this market, mortgage demand could be induced by uncompetitive large banks that possess significant market power. It will thus be wrong to assume that large banks and banks in strongly contestable markets will always lead to welfare gains for borrowers. Therefore, it is important to test how market structure changes could affect credit expansion and default risks in the markets. The empirical results could help explain some of the questions such as whether the rapid credit expansion was caused by large entry of hit-and-run lenders, or by cut-throat banks that operate inefficiently at below the optimal costs. We constructed H-statistics and Gini coefficients using MSA-level banking data, and tested the causal-effects of banking market structure on credit supply and mortgage default risks that are measured using mortgage data of the pools of non-agency residential backed securities. The concentration measure widely used in studying banking market structure (Dick and Lehnert, 2010; Gan, 2004; and others) is shown to have significantly different economic implications compared to the contestability definition proposed by Baumol (1982) and Baumol, Panzer and Willig (1982). We found that high contestability in banking markets could lead banks to reduce credit supply. Banks are unlikely to expand loan supply that will increase the marginal costs. The expanded demand in users markets could be met if new credit supply could be generated by new entrants, given that zero profit condition is not violated in the markets. In the strongly contestable markets, mortgage default risks are found to be higher compared to other sub-markets. In contrast to the strongly contestable markets, an opposite credit growth path is observed in highly concentrated banking markets. Large banks are more likely to increase credit supply, and still earn above zero normal profits through their possession of market powers. The monopoly power of the banks will increase mortgage default risks. We observed significant deterioration in credit quality in highly concentrated markets when banks are run inefficiently (inefficient monopolists). The results were consistent with the concentration-fragility view of some studies in the banking literature (Beck, Demirguc-Kunt and Levine, 2007; Boyd and De Nicolo, 2005; and others), which shows that banking concentration reduce competition and make banks vulnerable to default risks. When we extend our analyses to the sub-markets, between-group variations were observed, but the general results remain consistent that are contestability increases mortgage risks in highly concentrated banking market; and that large banks in a highly contestable market (a low marginal cost environment) are also exposed to higher mortgage default risks. The above findings have important implications on the too big to fail policy that aimed to protect inefficiently run banks with strong market power. These large banks tend to expand credit supply at the expenses of increasing default risks in the credit markets. Similarly, policies that promoting operating efficiency (contestability) in banking markets could results 21

23 in credit contraction in the markets, as banks find it infeasible to increase supply beyond the lowest marginal cost point. The strongly contestable banks are likely to be less discriminatory in pricing risks, and as a result, the mortgage default risks increase when banks become more competitive and entry barriers increase. On the economic intuition, our findings help to shed some light on the differences between contestability and concentration in banking markets. The banking literature, which invariably focuses the debate on concentration-fragility and concentration-stability, should not overlook the contestability factor that may drive profit down to zero. In a zero-profit environment, banks are that run efficient could cause welfare reduction in the housing markets, as many borrowers will be rationed out of the markets. The strongly contestable banks are also more vulnerable to originating mortgages that have high default risks. The default risk increases, if the contestable banking markets are made up of only few large banks. 22

24 Appendix 1: Contestability and Concentration o Banking Markets The removal of the inter-state and intra-state branching restrictions has been found to have significant impact on the competitiveness in the banking industry (Pilloff, 1999; Bikker and Haaf, 2003; Claessens and Laeven, 2004; Yildirim and Mohanty, 2010 etc). 11 The banking literature has still not addressed the issues on the measures of competition. Some researchers argue for the use structural measures of competition using the marginal approach proposed by Brensnahan (1982) and Lau (1982) and the input factor elasticity approach by Panzar and Rosse (PR, 1982, 1987). In this study, we adopt the three measures with the inputs of banking data to proxy market power, competition and contestability (efficiency) in the banking markets. We rely on Wharton s banking database of the University of Pennsylvania as our inputs for the computations of the market structure variables. The data on income statement and bank location are available for the sample periods from 1999 to 2008 on a quarterly basis for a total of more than 9000 banks, which give a pooled of 347,356 observations. a. Bank Number The deregulation of the banking industry and increasing use of technologies including telecommunication, internet banking and automated teller machine (ATM) in the 1990s have resulted in declines in the number of banks over the periods (Figure A1), except for some states such as Arizona, Alaska, California, Delaware, Florida, Georgia, Idaho, and Louisiana, where the number of bank increases through new banks and/or new branches. The numbers of banks and branches at the local level represent the market reach and coverage in the selected neighborhoods. [Insert Figure A1] b. Gini Coefficient Measure of Bank Concentration Gini coefficients 12 measures inequality of banks, which is defined as the area between equality line and Lorenz curve divided by total area under the equality line: N G 1 [( X X ) ( Y Y )] i 0 i i 1 i i 1 (A1) where Y i is the cumulative proportion (discrete density function) of the measured variable y i sorted in an ascending sequence; X i is the cumulative proportion of the number counts x i. G has its value in a close interval of [0 1], where a value 0 means equality, and a value 1 means inequality. As a measure of market concentration, a Gini coefficient of 1 indicates a high level of concentration in the market. 11 The passage of the Interstate Banking and Branching Efficiency Act (IBBEA) of 1994 has mandatorily removed geographical branching restrictions in all MSAs in the US. The staggered timing of deregulating banking branching restrictions in different MSAs leading to the passage of IBBEA in 1994 has had different impact on the contestability and market power of banks in different states (Dick and Lehnert, 2010; Beck, Levine and Levkov, 2010; Pilloff, 1999 and others). 12 We also use Herfindahl-Hirschman Index of concentration (HHI), which at city level is calculated using the sum of the squared market shares. The smaller HHI indicates higher concentration level in the city scope. HHIL is firstly used, which is based on loan market share measured by percentage of total loans on total loan of banks in city level. Other general HHIA is based on the market share measured by percentage of total assets on total banks asset (Equity and debts) in city level, which is assumed to measure the bank industry concentration. The regression results of their sign are not changed by using proxy of HHIA and HHIL. 23

25 We compute Gini coefficients based on real estate loan value and total asset (equity and debt) from our sample of 72,570 bank-year observations over the 10-year periods. Figure A2 shows that by loan values, the Gini coefficients reflect a relative high level of concentration, where the values fall within a narrow band of to In comparison, the Gini coefficients by total asset value of between 0.72 and reflect a weaker concentration relative the loan value Gini coefficients. The two Gini coefficient series move in an opposite direction during the pre-crisis periods from 2002 to By loan size, the market shares of banks were diluted over the periods; but more big banks with larger total asset value emerged during the same periods. At the same time, more hit-and-run small OTD banks that entered the markets have also caused rapid increases in credit supply in the pre-crisis periods. As these hit-and-run banks actively sold their mortgages in the securitization markets, the total asset size of the banks did not grow as fast as the loan growth. c. H-statistics for Bank Competiveness [Insert Figure A2] The use of H-statistic proposed by Rosse and Panzar (1977) and Panzar and Rosse (PR, 1982, 1987) as a measure of contestability and competitiveness in the banking market has been widely adopted in the banking literature (Bikker and Haaf, 2002; Claessens and Laeven, 2004; Bikker, Spierdijk, Finnie, 2007; Yildirim and Mohanty, 2010). With the state-level bank data in the US, we estimate log-version of the H-statistics using the PR approach as: 13 ln ln F ln PE ln PCE (A2) Unlike Bikker, Spierdijk, Finnie (2007) that use the first differences for the variables, the absolute (level) values for the total interest income, ; annual expense on funds, F; personal expense, PE; and physical capital expense that include furniture, fixture, equipment and auto; PCE, are used in our models. From Equation (A2), the H-statistics at the state-level can be computed as the sum of the three input factor elasticity measures with respects to the bank s total revenue: H (A3) In a high H-statistic state, banks operate more efficiently in producing the same level of services at a lower marginal cost. We run Equation using yearly data for a sample of 50 US states and 6 other territories / islands (that are American Samoa (AS), Federated States of Micronesia (FM), Guam (GU), Puerto Rico (PR), Rhode Island (RI) and Virgin Islands (VI)), and repeat the estimates on yearly basis for the periods The results as in Figure 1 show that the state-level bank efficiency decreases for the pre-subprime crisis periods from 2001 to 2006 admit increases in the number of loans originated. 13 The small bank sample at zip-code level restricts our estimation of the H-statistics at the zip-code level, and the MSA-level H-statistics are also less susceptible to estimation errors. 24

26 Reference: Ambrose, B.W., Lacour-Little, M. and Sanders, A.B. (2005). Does Regulatory Capital Arbitrage, Reputation, or Asymmetric Information Drive Securitization? Journal of Financial Services Research, 28(1/2/3), Ambrose, Brent W. and Pennington-Cross, Anthony, (2000). Local economic risk factors and the primary and secondary mortgage markets. Regional Science and Urban Economics, 30, Agarwal, Sumit, Ambrose, Brent W., Chomsisengphet, Souphala, and Sanders, Anthony B. (2009). Thy Neighbor s Mortgage: Does living in a Subprime Neighorhood Impact your probability of default? Working paper, October 7, 2009 edition. Baumol, W. J. (1982). Contestable Markets: An Uprising in the Theory of Industry Structure. The American Economic Review, 72(1):1-15. Baumol, W. J., Panzer, J., and Willig, R. (1982). Contestable Markets and the theory of industry structure. Harcourt Brace Jovanovich (New York). Berger, A. N. (1995). The Profit-Structure Relationship in Banking--Tests of Market-Power and Efficient-Structure Hypotheses. Journal of Money, Credit and Banking, 27(2): Beck, T., A. Demirgüç-Kunt, et al. (2006). Bank Concentration, competition and crises: First results. Journal of Banking & Finance, 30: Beck, Thorsten, Levine, Ross and Levkov, Alexey (2010). Big Bad Banks? The Winners and Losers from Bank Deregulation in the United States. The Journal of Finance, 65(5), Bikker, Jacob A. and Haaf, Katharina (2002). Competition, concentration and their relationship: An empirical analysis of the banking industry, Journal of Banking & Finance, 26, Boyd, J.H., and De Nicolo, G. (2005). The theory of bank risk taking and competition revisited. The Journal of Finance, 60, Brensnahan, Timorthy F. (1989). Studies of industries with market power. In Handbook of Industrial Orgainization, Vol II, edited by Richard Schmalensee and Robert D. Willig, Amsterdam: Elsevier Science Publishers. Claessents, Stijn and Laeven, Luc (2003). What drives bank competition? Some international evidence. Journal of Money, Credit and Banking, 36(3), Dell'Ariccia, G., D. Igan, et al. (2008). Credit Booms and Lending Standards: Evidence from the subprime Mortgage Market. Working Paper WP/08/106, IMF. Demyanyk, Yuliya and Hemert, Otto Van (2009). Understanding the Subprime Mortgage Crisis. Review of Financial Studies, DeYong, Kiler, and McMillen. (2004). The changing Geography of the U.S. Banking Industry. The Industrial Geographer, 2(1): DiPasquale, D. and Wheaton, W.C. (1996). Urban Economic and Real Estate Markets, Prentice Hall. Dick, Astrid A. and Lehnert, Andreas (2010). Personal Bankruptcy and Credit Market Competition. The Journal of Finance, 65(2),

27 Gabaix, X. and Laibson, D. (2006). Shrouded Attributes, Consumer Myopia, and Information Suppression in Competitive Markets, Quarterly Journal of Economics, 121, Gan, J. (2004). Banking market structure and financial stability: Evidence from the Texas real estate crisis in the 1980s. Journal of Financial Economics, 73, Gan, J. and T. J. Riddiough (2008). Monopoly and Information Advantage in residential mortgage market. Review of Financial Studies 21(6): Gerardi, Kristopher S., Rosen, Harvey S. and Willen, Paul S. (2010). The Impact of Deregulation and Financial Innovation on Consumers: The Case of the Mortgage Market. The Journal of Finance. 65(1), Gerardi, Krostopher, Lehnert, Andreas, Sherlund, Shane and Willen, Paul (2008), Making Sense of the Subprime Crisis. Brookings Papers on Economic Activity. Gilbert, R. A. (1984). Bank Market Structure and Competition: A Survey. Journal of Money, Credit and Banking 16(4/2): Hakenes, H. and I. Schnabel (2009). Credit Risk Transfer and Bank Competition. Preprints of the Max Planck Institute for Research on Collective Goods. Bonn, Max Planck Society. Hanweck, G.A., and Rhoades, S.A. (1984). Dominant firms. Journal of Economics and Business, 36, Keys, B.J., Mukherjee, T., Seru, A., and Vig, V. (2010). Did Securitization lead to lax screening? Evidence from subprime loans. The Quarterly Journal of Economics, 125, Lau, Lawrence (1982). On indentifying the degree of competitiveness from industry price and output data. Economics Letters, 10, Markusen, J. R. and D. T. Scheffman (1978). Ownership Concentration and Market Power in Urban Land Markets. The Review of Economic Studies, 45(3): Mian, Atiff and Sufi, Amir (2009). The Consequences of Mortgage Credit Expansion: Evidence from the U.S. Mortgage Default Crisis. The Quarterly Journal of Economics, 124, Panzar, J. C. and J. N. Rosse (1982). Structure, conduct and comparative statistics. Bell Laboratories Economic Discussion Paper. Panzar, J. and J. N. Rosse (1987). Testing for mono equilibrium. Journal of Industrial Economics. 35, Pilloff, Steven J. (1999). Does the Presence of Big Banks Influence Competition in Local Markets? Journal of Financial Services Research. 15(3), Purnanandam, Amiyatosh (2010). Originate-to-distribute Model and the Subprime Mortgage Crisis. Review of Financial Studies, Ogura, Yoshiaki (2010). Interbank competition and information production: Evidence from the interest rate difference. Journal of Financial Intermediation, 19, Rhoades, S.A. (1995). Market share inequality, the HHI, and other measures of the firm-composition of a market. Review of Industrial Organization, 10,

28 Rothschild, M. and Stiglitz, J. (1976). Equilibrium in competitive insurance markets: An essay on the economics of imperfect information. The Quarterly Journal of Economics, 90, Smirlock, M. (1985). Evidence on the (non) relationship between concentration and profitability in banking. Journal of Money, Credit and Banking, 17, Stiglitz, J., and Weiss, A. (1981). Credit rationing in markets with imperfect information. The American Economic Review, 71, Wang, J. C. (2003). Productivity and Economies of Scale in the Production of Bank Service Value Added. FRB of Boston Working Paper No Boston, FRB. Yildirim, H.Semih and Mohanty, Sunil K. (2010). Geographical Deregulation and Competition in the U.S. Banking Industry. Financial Markets, Institutions & Instruments, 19(2),

29 Figure 1: Mean of State-level H-statistic and the number of loans through year Note: The figure shows the State-level H-statistic through year. Higher H-statistic means high bank s efficiency in its service operation in a contestable state level submarket. The right side vertical line is the total number of loans that originated at the specific years. 28

30 Figure 2: Spatial distribution of Market concentration of U.S. bank in year 2006 Note: It shows the market concentration of U.S. bank in loan service market for the year 2006 (Mapped from our data by Gini Coefficient). The market concentration is measured by Gini coefficient based on individual banks total loan amount serviced from their report data.. The pattern of the market concentration groups is dispersed geographically, which is different with the spatially 12 Federal Reserve Districts (1996) by U.S. Federal Reserve Board ( Gini coefficiency in the figures is separated into 10 groups by equal quantile interval of 10%. 29

31 Figure 3: Residential Mortgage default rate) at zip-code country level (Originated at year 2006 Note: This figure is based individual mortgage performance date with origination date that originated by non-agency institutions. It is calculated by number of default/number of total loan at each zip-code level. The location is geo-coding based on zip code level, with 82% matching. 30