Working Paper No The Consequences of Mortgage Credit Expansion Evidence

Working Paper No. 15 The Consequences of Mortgage Credit Expansion: Evidence from the 2007 Mortgage Default Crisis Atif Mian University of Chicago Graduate School of Business and NBER Amir Sufi University of Chicago Graduate School of Business Initiative on Global Markets The University of Chicago, Graduate School of Business “Providing thought leadership on financial markets, international business and public policy” The Consequences of Mortgage Credit Expansion: Evidence from the 2007 Mortgage Default Crisis* Atif Mian University of Chicago Graduate School of Business and NBER Amir Sufi University of Chicago Graduate School of Business May 2008 Abstract We show that an expansion in the supply of mortgage credit to high latent demand zip codes led to a rapid increase in house prices from 2001 to 2005 and subsequent defaults from 2005 to 2007. From 2001 to 2005, high latent (unfulfilled) demand zip codes experienced relative declines in denial rates and interest rates and relative increases in mortgage credit and house prices, despite the fact that these zip codes experienced negative relative income and employment growth. The growth in securitization was significantly higher in high latent demand zip codes, suggesting a possible role of securitization in credit expansion. *We gratefully acknowledge financial support from the Initiative on Global Markets at Chicago GSB and the IBM Corporation. The data analysis was made possible by the generous help of Myra Hart, Jim Powers, Robert Shiller, Cameron Rogers, Greg Runk, and David Stiff. We thank Mitch Berlin, Stuart Gabriel, Jonathan Guryan, Bob Hunt, Erik Hurst, Doug Diamond, Mitchell Petersen, Raghu Rajan, Josh Rauh, Clemens Sialm, Nicholas Souleles and participants at the Chicago GSB finance seminar, Chicago GSB applied economics lunch, Emory University, Federal Reserve Banks of Philadelphia, New York, and San Francisco, IMF, the NBER Corporate Finance, Monetary Economics, and Risk of Financial Institutions conferences, the NYU-Moody’s Conference on Credit Risk, the Chicago Fed Bank Structure Conference, the University of Michigan, and Boston College for comments and feedback. We also thank Sim Wee, Rafi Nulman, and Smitha Nagaraja for excellent research assistance. Mian: (773) 834 8266, atif@chicagogsb.edu; Sufi: (773) 702 6148, amir.sufi@chicagogsb.edu Recent developments in the U.S. housing market are the focus of increased anxiety among policy-makers, investors, and financial markets. The rapid growth in mortgage credit and house prices from 2001 to 2005 has given way to grave concerns as mortgage defaults continue to increase. The market value of mortgage securities has fallen precipitously, with some tranches losing up to 70 to 80% of their value in less than a year. Many believe that weakness in the U.S. housing market poses a serious threat to financial markets and economic activity. Indeed, the April 2008 FOMC statement argues that “…tight credit conditions and the deepening housing contraction are likely to weigh on economic growth over the next few quarters.” This paper investigates the origins of the rapid growth in mortgage debt and house price growth from 2001 to 2005 and the subsequent mortgage default crisis of 2007. In particular, we explore whether recent trends are a result of supply or demand shifts in the mortgage market. The supply explanation argues that a greater willingness by lenders to assume risk led to a reduction in the risk premium and an expansion in the supply of mortgage credit. A demand explanation argues that increases in productivity or economic opportunities led to an expansion in the demand for mortgage credit due to a permanent income effect. It is impossible to separate the supply and demand hypotheses with aggregate time series data alone. As a result, researchers must rely on cross-sectional variation over time to empirically test whether supply or demand explains recent trends in the mortgage market. On this front, our unique advantage is a comprehensive zip code level data set from 1991 through 2007 that includes outstanding consumer debt, defaults, house prices, mortgage characteristics, income data, and demographic variables. This data set represents one of the most comprehensive and disaggregated data sets in the real estate and consumer credit literature. 1 In order to isolate the supply channel, we exploit variation across zip codes that differ in their initial latent or unfulfilled demand for mortgages, measured by the percentage of applicants in the zip code denied mortgage credit in 1996. Since a higher fraction of the population in these high latent demand zip codes is initially denied credit, a subsequent expansion in the supply of mortgage credit should disproportionately affect these zip codes. Our core results are strongly consistent with the supply expansion hypothesis. Zip codes with high latent demand for mortgages experience a sharp relative decrease in denial rates and a sharp relative increase in mortgage debt to income ratios from 2001 to 2005. In addition, high latent demand zip codes experience sharp relative growth in mortgage originations and house prices during this period. While high latent demand zip codes experience a strong relative increase in debt to income ratios, the price of mortgage credit risk – the spread between prime and subprime mortgages – declines to historical lows from 2001 to 2005. The primary counter-argument to our supply interpretation is that high latent demand zip codes experience relative mortgage origination and house price growth from 2001 to 2005 because of relative improvements in demand conditions such as credit quality or productivity. However, a number of facts dispute this concern. First, high latent demand zip codes experience negative relative income, wage, employment, and establishment growth from 2001 to 2005. Second relative growth in non-home debt (auto loans and credit cards) is negative for high latent demand zip codes, a fact that is inconsistent with the hypothesis that relative permanent income increases in these areas over this time period. Finally, an alternative concern is that high latent demand zip codes, which tend to be poorer with worse credit scores on average, are disproportionately affected by business cycle conditions. In particular, the concern is that these zip codes benefit more from declining interest 2 rates that characterize the post 2001 recession period. We mitigate this concern by replicating our methodology over the similar macroeconomic environment of 1990 to 1994, when interest rates declined rapidly in the aftermath of a recession. We find that despite similar business cycle conditions as 2001 to 2005, high latent demand zip codes experience disproportionately negative mortgage origination and house price growth from 1990 to 1994. We then demonstrate that the supply-driven expansion in credit to high latent demand zip codes is followed by a large increase in default rates. The magnitudes are striking: a one standard deviation increase in “supply-driven” mortgage debt from 2001 to 2005 leads to a one standard deviation increase in mortgage default rates from 2005 to 2007. Furthermore, a one standard deviation increase in supply-driven house price appreciation leads to a three standard deviation increase in mortgage default rates. To put these magnitudes in historical perspective, the relative increase in mortgage default rates from 2005 to 2007 for high latent demand zip codes is twice as large as the relative increase during the 2001 recession. What explains the increase in the supply of credit? One possible factor is that financial innovation through securitization allows loan origination risk to be distributed to non-traditional players in the mortgage market. Consistent with this explanation, we find that the relative growth in the securitization of mortgages is much stronger in high latent demand zip codes from 2001 to 2005. Similarly, credit growth from 2001 to 2005 and the growth in default rates from 2005 to 2007 are significantly higher for zip codes with larger increases in securitization. Interestingly, the positive correlation between the growth in credit/defaults and the growth in the sale of mortgages only holds for sales to financial firms that are not affiliated with the loan originator. This finding hints at moral hazard on behalf of originators as a factor contributing to the expansion in credit supply, although we believe that more research is needed on this issue. 3 Research presented here is related to recent working papers examining the rise in default rates on subprime mortgages (Keys, Mukherjee, Seru and Vig (2008), Demyanyk and Van Hemert (2007), Doms, Furlong, and Krainer (2007), Gerardi, Shapiro, and Willen (2007), and Dell’Ariccia, Igan, and Laevin (2008)). Relative to these papers, we believe our analysis is unique in its strategy to isolate the causal effect of supply expansion on house price appreciation and defaults using within-county across-zip code variation in latent demand. Most closely related to our supply expansion results is work by Gabriel and Rosenthal (2007), who show that the expansion of a secondary mortgage market increased credit to high risk areas. Our work is also related to an earlier strand of literature that examines the relation between housing price changes and consumer borrowing (Poterba (1984), Case and Shiller (1989), Stein (1995), Genesove and Mayer (1997, 2001), Hurst and Stafford (2004), Glaeser and Gyourko (2005), Himmelberg, Mayer, and Sinai (2005), Brunnermeier and Julliard (2007)). I. Data, Summary Statistics, and Aggregate Trends A. Data Data on consumer debt outstanding and delinquency rates come from Equifax Predictive Services. Equifax keeps a credit history of most consumers in the U.S., and provided us with zip code level annual aggregate data for outstanding credit and defaults from 1991 to 2007, measured at the end of the year. The debt and default aggregates are broken down by the type of loans: mortgages, home equity lines, credit card debt, auto loans, student loans, and consumer loans. We classify mortgage and home equity loans as “home debt”, and other types of loans as “non-home debt”. The default data is aggregated by various degrees of delinquency. We use 30 days or more delinquent as our definition of default, but our results are materially unchanged using a stricter definition such as 60 days or more delinquent. 4 We collect data on the flow of new mortgage loans originated every year through the “Home Mortgage Disclosure Act” (HMDA) data set from 1990 through 2006. HMDA is available at the loan application level. It records each applicant’s final status (denied / approved / originated), purpose of borrowing (home purchase / refinancing / home improvement), loan amount, race, sex, income, home ownership status, and also (in the case of originated loans) whether the loan was sold to the secondary market within the year. We aggregate HMDA data up to the zip code level, and drop any zip codes with missing Equifax or HMDA data between 1996 and 2006, giving us a final sample of 18,419 zip codes.1 Our zip code level house price data from 1990 to the first quarter of 2007 come from Fiserv’s Case Shiller Weiss indices. FCSW use same house repeat sales data to construct zip level house price indices. One limitation of the data is that FCSW require a significant number of transactions in a given zip code to obtain reliable estimates of changes in house prices over time. As a result, FCSW has house prices for only 3,056 of the zip codes in Equifax-HMDA sample. While FCSW covers only 17% of the number of zip codes in the Equifax-HMDA sample, these zip codes tend to be larger and represent over 45% of aggregate home debt outstanding.2 We also add zip code level data on demographics, income, and business statistics through various sources: Demographic data on population, race, poverty, mobility, unemployment and education are from the decennial Census. Data on wages, employment, and business establishments in a given zip code come from the Census Business Statistics from 1996 through HMDA data contain census tract, but not zip code, information. We match census tracts to zip codes using a match provided by Geolytics. The match quality is high: 85% of the matched census tracts in our final sample have over 90% of their population living in the zip code to which they are matched. 2 The Appendix Table compares the sub-samples based on the house price index restriction, and shows that the primary difference is the fraction of households in urban areas. All of our results without house prices are materially unchanged if we use the full sample. In addition, all of our non-house price results hold if the analysis is done at the MSA level using the 199 MSAs for which we have data. We also collect zip code level price indices for 2,248 zip codes from Zillow.com, an online firm that provides house price data. House price changes for FCSW and Zillow have a correlation coefficient of 0.91, and all of our results are robust to the use of Zillow indices. 1 5 2004. Average adjusted gross income data at the zip code level for years 1991, 1998, 2001, 2002, 2004 and 2005 come from the IRS. The income variable from the IRS is important because it tracks the income of consumers living inside a given zip code, as opposed to Business Statistics which provide wage and employment statistics for individuals working, but not necessarily living, in a zip code. We also collect zip level statistics on total crime from 2000 to 2007 from CAP Index. B. Summary Statistics and Aggregate Trends Table 1 presents summary statistics for the final sample of 2,920 zip codes for which we have all data available in every year from 1996 to 2007. Mortgage debt represents 74% of total consumer debt in 1996. While mortgage and non-home debt increase at the same rate from 1996 to 2001, there is a rapid acceleration in mortgage debt from 2001 to 2005 relative to non-home debt (10.2% vs. 4.6%). The differential time-series pattern of mortgage debt relative to nonhome debt is also evident in Figure 1. In historical terms, the relative growth from 2001 to 2005 in mortgage debt is far larger than any other period since 1991. Table 1 demonstrates the strength of house price growth during our sample period, with house prices growing by an annualized rate of 7.3% from 1996 to 2001 and 11.3% from 2001 to 2005. There is also a dramatic increase of 26.3 percentage points in the fraction of originated mortgages sold to non-mortgage agency investors. 3 Figure 2 shows the time series of the fraction of mortgages sold. As the figure demonstrates, the sharp increase in this fraction begins in 2001. Figure 3A maps the median, and 75th and 90th percentiles of debt to income ratios of accepted mortgage applications from HMDA. There is a slight upward trend in the ratios from 1996 through 2000. However, the increase in mortgage debt to income ratios from 2001 to 2005 By “non-mortgage agency investors”, we mean investors other than Freddie Mac, Fannie Mae, Federal Farmers Home Administration, and Ginnie Mae. 3 6 is much larger, signifying the deterioration in observed credit quality as credit expands. The mortgage debt to income ratio of borrowers in the 90th percentile increases by 1 unit over this time period, which is a two standard deviation increase. In Figure 3B, we utilize an aggregate mortgage debt to income ratio as a measure of zip code credit quality. It represents mortgage debt originated for home purchase from HMDA scaled by the aggregate zip code income reported to the IRS. Figure 3B shows a very similar pattern to Figure 3A: debt to income ratios for the zip code increase sharply beginning in 2001. An alternative measure of credit quality is the debt to home value ratio. While we do not have mortgage level home value data, Demyanyk and Van Hemert (2007) show that debt to value ratios also increased from 2001 to 2005. Taken together, the evidence suggests an aggregate supply shift in mortgage credit, accompanied by a sharp increase in mortgage sales and deterioration in the credit quality of originated mortgages. Mortgage default rates increase sharply in the aftermath of the credit expansion. Table 1 demonstrates that mortgage default rates increase by 3.5 percentage points from 2005 to 2007. This increase represents more than doubling of the average default rate on mortgages since 1996. Figure 4 plots the historical default time series in order to place this increase in historical perspective. As it shows, the mortgage default rate is almost 100% higher in 2007 than in the recession of 2001, despite the fact that there is no recession in 2007. II. Empirical Methodology Our empirical methodology is designed to isolate the causal effect of the supply expansion on mortgage credit growth, house price growth, and subsequent defaults. Our approach attempts to separate the effect of an expansion in the supply of credit from potentially confounding effects of contemporaneous changes in the demand for mortgages. Consider customers living in zip code z in county c at time t. In every period customers of measure one are 7 interested in purchasing a new home that requires one unit of capital. For simplicity, we assume that a qualified customer takes the mortgage this period, and promises to completely pay off principal and interest next period. We define customers as “prime” if their income profile exceeds a certain threshold such that there is no possibility of default next period. As a result, all lenders are willing to lend to prime customers at the risk free rate normalized to 1. We denote the fraction of prime customers in a zip code by fzt (Izt), with the argument Izt reminding us that fzt depends on the overall income distribution within a zip code. We define customers with income profiles below the prime threshold as “subprime.” What distinguishes subprime customers is that they have a positive probability, p, of default if their realized income next period is sufficiently low. Subprime customers have different individual income profiles, and can therefore differ in their probability of default, p. We assume the mortgage market is competitive at the national level, and that lenders recover nothing in case of default. At each t, the interest rate offered to a subprime customer is given by: (1) ∞ In (1), θ reflects the “risk premium” that the market charges for bearing the probability of default, and is an interest rate ceiling above which no lender is willing to lend. We do not model explicitly the underlying friction that leads to an interest rate ceiling above which originators are unwilling to lend—borrower moral hazard (Diamond (1991), Holmstrom and Tirole (1997)) or adverse selection (Stiglitz and Weiss (1981)) are potential reasons.4 The net result of equation (1) is that only a fraction gzt of subprime customers in each period t obtain mortgages. The fraction gzt depends on the market risk premium (θt) and 4 Gabriel and Rosenthal (2007) explicitly model how a supply expansion affects borrowers with a Stiglitz and Weiss (1981) adverse selection problem. Their conclusions are similar to ours. 8 distribution of p among subprime customers, which in turn is a function of the overall income distribution Izt in the population.5 We can therefore write gzt as gzt (θt , Izt), with gθ < 0 and gI > 0. The preceding discussion gives us the equilibrium determination of mortgage originations in zip code z at time t (Lzt) as: 1 (2) We have suppressed arguments of f and g for notational simplicity. Allowing for other possible factors affecting Lzt, yields: 1 (3) αz reflects time-invariant determinants of loan origination for a given zip code, αct reflects timevarying county-level factors affecting loan originations, and εzt is an unobserved error term. The fundamental economic drivers of equilibrium loan originations in equation (3) are income factors, which are summarized by income distribution Izt, and credit supply factors, which are summarized by the mortgage risk premium θt. The challenge of our empirical methodology is to isolate the effect of changes in supply factors on loan originations while controlling for income factors. Since equation (3) includes county interacted with time fixed effects, any changes in income that are common across zip codes in the same county are nonparametrically removed. In order to clarify the identifying assumption we make to isolate the supply channel, we first make the assumption that all variation in income factors occurs at the county level. As we demonstrate below, we do not need this strong of an assumption but it is useful for illustrative purposes. Given that there is no residual time variation left in fzt (which does not depend on the risk premium), we can replace it by the initial fraction of prime customers in a zip code, fz0. Since 5 Solving explicitly, gzt is the subset of subprime customers with . 9 we are interested in shocks to loan originations, we first-difference equation (3) and suppress time subscripts for simplicity. Therefore, under the assumption that income factors only vary at the level of the county, first-differencing equation (3) gives us: ∆ 1 ∆ (4) where β =Δgz , which depends only on the credit supply shock θ. A negative θ reflects a reduction in market risk premium and hence a positive credit supply shock. A positive credit supply shock would lead to more subprime consumers obtaining mortgages and hence a positive β. In other words, identifies the impact of a credit supply shock on Lzt under the identifying assumption that all income shocks occur at the county level. We can relax our identifying assumption further. Income shocks may be zip code specific, but as long as they are orthogonal to the initial latent demand conditions (1-fz0), retains its interpretation. A natural corollary is that if zip code specific income shocks are negatively correlated with the initial fraction of subprime customers, then our interpretation of as a credit supply coefficient is still accurate, but the magnitude is an under-estimate of the true supply effect. As we show below, the fraction of subprime customers at the beginning of our sample is negatively correlated with observable measures of future income shocks. This negative correlation strengthens our identifying assumption that future income shocks are not positively correlated with the initial fraction of subprime customers, and further suggests that our estimates may understate the effect of credit expansion on outcomes. Equation (4) represents our primary regression specification. In order to estimate this equation, our data provides us with many possible measures of initial latent demand conditions, or equivalently “subprime” customers (1-fz0). We use the fraction of loan applications denied in 10 1996 and the fraction of borrowers with a credit score under 660 as our main measures of high latent demand zip codes. 6 Table 2 presents the correlation of our main measure of latent demand in a zip code, the fraction of mortgage applications denied in 1996, with other variables. This measure is strongly correlated with alternative measures of high latent demand/high credit risk, such as the fraction of subprime borrowers or the fraction of loans backed by FHA. It is also strongly correlated with poverty and unemployment, and negatively correlated with household income measures. The bottom panel of Table 2 demonstrates that measures of future growth in economic opportunities are negatively correlated with the fraction of 1996 mortgage applications denied. As mentioned above, the critical identifying assumption of our empirical methodology is that areas with high initial latent demand do not experience subsequent increases in income, credit quality, or economic opportunity. The correlations in Table 2 strongly support the identifying assumption, given that observable measures of future growth in economic opportunity are negatively correlated with our primary measure of high 1996 latent demand. This also suggests that our estimate is an under-estimate of the true supply shock coefficient β. III. Results: Credit Expansion A. Credit Expansion to High 1996 Latent Demand Zip Codes We begin our empirical analysis by demonstrating that high 1996 latent demand zip codes experience a relative increase in credit supply to riskier borrowers from 2001 to 2005. In Figures 5 through 7, we plot coefficient estimates from a year-by-year set of county fixed effects regressions of the following general specification: , 6 β , , (5) We obtain similar results using alternative measures of latent demand such as the fraction of population with a credit score as of 2000, fraction of mortgages backed by federal housing administration as of 1996, and number of bank branches per capita as of 2000. 11 1997, 1998, … , 2007 In other words, for each year t from 1997 to 2007, we estimate a first-difference county fixed effects specification relating the change in outcome y for zip code z in county c from year 1996 to year t to our primary measure of high 1996 latent demand, which is the fraction of 1996 mortgage applications denied in the zip code. We plot the coefficient estimates of β for each year t, along with the corresponding 95% confidence interval. The plotted coefficient estimates represent the differential effect on the change in outcome y from 1996 to t for high latent demand zip codes, after controlling for county fixed effects (αc). The county fixed effects control for any shock at the county level. Figure 5 examines the differential pattern of denial rates, debt to income ratios, and loan sales to non-mortgage agency investors for high 1996 latent demand zip codes. Figure 5A demonstrates a dramatic differential decrease in denial rates for high 1996 latent demand zip codes beginning in 2001 and lasting through 2006. The coefficient estimate for 2004 implies that a one standard deviation increase in 1996 latent demand (0.08) leads to a reduction in the denial rate of 2 percentage points from 1996 to 2004, which is a one-third standard deviation of the left hand side variable. Figure 5B shows a corresponding increase in the average debt to income ratio of high 1996 latent demand zip codes that begins after 2001. The coefficient estimate for 2005 implies that a one standard deviation increase in 1996 latent demand leads to a one-third standard deviation increase in mortgage debt to income ratios from 1998 to 2005. The relative reduction in denial rates and relative increase in debt to income ratios suggest a supply expansion to high latent demand areas. Figure 5C shows a source of this expansion. Beginning in 2001, there is a sharp relative rise in the fraction of mortgages sold to non-mortgage agency investors for high latent demand zip codes. The estimate for 2006 implies 12 that a one standard deviation increase in latent demand leads to a 2.4 percentage point increase in the fraction of mortgages sold from 1996 to 2006, which is more than a one-third standard deviation of the left hand side variable. Interest rates to subprime borrowers also decline during this period. While we do not have mortgage level data on interest rates, Chomsisengphet and Pennington-Cross (2006) show that the subprime-prime mortgage spread for 30-year fixed rate mortgages drops sharply from 2001 to 2004. Demyanyk and Van Hemert (2007) reach a similar conclusion using a different data set. The simultaneous decrease in denial rates and interest rates for subprime borrowers suggests a shift in the supply of mortgage credit to subprime households. B. The Effect of Credit Expansion on Mortgage Debt and Housing Prices Figure 6A shows a sharp relative increase from 2002 to 2006 in the volume of home purchase loan originations for high 1996 latent demand zip codes. The coefficient estimate for 2006 implies that a one standard deviation change in 1996 latent demand leads to a relative increase in the growth rate of originated mortgage amounts for home purchase of 28%, which is one-half standard deviation of the left hand side variable. There is a slight increase from 1998 to 2000, but this increase is less than half the increase from 2002 to 2005. Figure 6B examines the relative growth in mortgage debt outstanding of high 1996 latent demand zip codes. The figure demonstrates that the sensitivity of mortgage debt growth in a zip code to high 1996 latent demand increases from 1999 through 2007. The coefficient estimate for 2007 implies that a one standard deviation increase in 1996 latent demand leads to a relative increase in the growth rate of mortgage debt outstanding from 1996 through 2007 of 5 percentage points, which is one-eighth of a standard deviation of the left hand side variable.7 The estimates in Figure 6B are relatively imprecise and smaller in magnitude compared to other estimates of mortgage growth because the Equifax measure of mortgage debt used in the figure does not differentiate mortgage 7 13 Figure 7 demonstrates the effect of increased supply on house price growth. High 1996 latent demand zip codes do not experience higher growth in house prices from 1996 to 1998. However, as credit supply starts to expand disproportionately in high latent demand zip codes in 1999, they start to experience a relative increase in house price appreciation. The relative growth in house price appreciation accelerates from 2001 onward. The coefficient estimate for 2000 implies that a one standard deviation increase in 1996 latent demand leads to a relative increase in house price appreciation from 1996 to 2000 of 0.8%, which is less than a one-fifteenth standard deviation in house price appreciation. The coefficient estimate for 2006 implies that a one standard deviation increase in 1996 latent demand leads to a relative increase in house price appreciation from 1996 to 2006 of almost 6%, which is one-third of a standard deviation. It is important to emphasize that the relative increase in housing prices for high latent demand zip codes occurs despite relatively negative income and employment growth for these zip codes during this period. In fact, as we demonstrate below, the period from 2001 to 2005 is the only period in recent U.S. history where house prices rise in zip codes with relatively negative income growth. These findings suggest that house price growth from 2001 to 2005 is closely linked to the mortgage credit expansion, and they caution against treating house prices as exogenous to credit conditions. Table 3 presents the equivalent regression coefficients for the results seen in Figures 6 and 7, where the specifications control for possible changes in economic and social conditions at the zip code level. The estimated coefficients come from the following first difference county fixed effects specification: debt for new home purchase versus mortgage debt obtained through refinancing. This is important because high 1996 latent demand zip codes do not refinance as aggressively in response to declining interest rates as low 1996 latent demand zip codes (something we confirm in the HMDA data that separates originations for refinancing versus home purchase). 14 , (6) where X represents a matrix of control variables. We choose the period 2001 to 2005 for the regressions given the evidence from Figures 5 through 7 that this is the main period over which supply expansion occurs. Minor variations of this time frame do not affect the results. The results in Panel A confirm the findings from Figures 6 and 7. Panel B examines the differential effect of credit expansion on mortgage debt and house prices using the fraction of subprime borrowers in a zip code in 1996 as an alternative measure of latent demand. The results are similar. IV. Could Results be Due to Changes in Demand? The results in Section III demonstrate a relative decline in denial rates, credit quality, and interest rates for high latent demand zip codes, in conjunction with a relative increase in the fraction of mortgages sold by originators to non-agency investors. These relative changes correspond to a relative increase in mortgage origination and house price growth for high latent demand zip codes. Taken together, these results strongly suggest a shift in the supply of mortgage credit. In this section, we explore the concern that differential changes in demand are responsible for the growth in house prices and originations in high 1996 latent demand zip codes. A. Improvements in Income and Business Opportunities One concern is that high 1996 latent demand zip codes subsequently experience relatively stronger income, wage, or productivity growth that justifies lower denial rates, lower interest rates, more lending, and higher house prices. However, as Table 2 demonstrates, from 2001 to 2005, high 1996 latent demand zip codes experience relatively negative income, establishment, and employment growth. In other words, mortgage credit is originated at a faster pace in relatively declining areas. These correlations contradict the argument that the relative increase in 15 credit and house prices in high latent demand zip codes is due to relative improvements in economic conditions in these areas. Furthermore, Table 4 demonstrates that the correlation between income growth and loan origination growth and between income growth and house price growth is negative between 2001 and 2005. This negative correlation is unique in recent U.S. history. Since 1990, in all other periods, income growth is positively correlated with credit growth and house price growth. If cross-sectional variation in mortgage and house price growth across zip codes were driven primarily by demand side factors from 2001 to 2005, then we would witness a positive correlation between income growth and credit / house price growth, just as we witness in all other period of recent U.S. history. We see the exact opposite in the data. While contemporaneous income and business opportunities decline in relative terms in high latent demand zip codes from 2001 to 2005, an alternative concern is that expected income in these areas increases. We note initially that it is hard to construct examples where expected relative income increases in a zip code that continues to experience relative decreases in realized income. Nonetheless, our data set allows us to directly test this alternative expected income hypothesis. If consumers in high latent demand zip codes expect increases in future income, then they would increase borrowing on all margins. However, results in Table 5 show exactly the opposite. When we examine the pattern in non-home debt outstanding, which consists mainly of automobile and credit card debt, we find relative declines for high latent demand zip codes from 2001 to 2005. High latent demand zip codes experience an increase in mortgage debt outstanding from 2001 to 2005, despite experiencing a decline in non-home debt over the same time period. B. Business Cycle Effects 16 We control for any level effect of county business cycle trends by including county fixed effects in the first-differenced specifications above. However, an alternative concern is asymmetric effects of the business cycle on lower credit quality zip codes. For example, one worry is that marginal neighborhoods with a higher concentration of subprime borrowers may demand relatively more credit as the economy emerges from the 2001 recession. Alternatively, a mortgage-specific business cycle concern is the impact of declining interest rates from 2001 to 2005 on subprime borrowers. The concern is that subprime borrowers increase their demand for housing relatively more in response to a lower nominal risk free rate than prime borrowers. There are two facts that mitigate this concern. First, as mentioned above, subprime borrowers experience a relative decline in non-home debt balances from 2001 to 2005, which contradicts the argument that the emergence from a recession in 2001 coupled with low nominal risk free rates mechanically increases borrowing by lower credit quality households. Any demand-based business cycle concern must explain why high latent demand zip codes experience a simultaneous increase in mortgage debt and decrease in non-home debt. Second, if the differential effect of the business cycle explains our results, then we would expect to find similar results during the 1990 to 1994 period in which the U.S. economy experiences a similar macroeconomic environment. Figure 8 shows that the evolution of 3-month Treasury bill interest rates from 1990 through 1994 (Figure 8A) is analogous to 2001 through 2005 (Figure 8B). The macroeconomic environment is also similar, as the U.S. emerges from a recession during both of these time periods. In Figure 9, we examine the differential pattern in origination growth and house price appreciation from 1990 to 1994 for high denial zip codes, measured as of 1990. For comparison purposes, we also plot the coefficients for the 2001 to 2005 period, where the denial rate is measured as of 2001. Figure 9 shows that we do not see the 17 relative increases in originations or house prices for high denial rate areas from 1990 to 1994 period, despite the similar macroeconomic environment. In fact, the evidence suggests that origination growth and house price growth is relatively negative from 1990 to 1994 for high 1990 denial rate zip codes. V. Results: Default Rates Figure 10 demonstrates the dramatic relative rise in default rates for high latent demand areas from 2005 to 2007. In terms of magnitudes, the point estimate for 2007 implies that a one standard deviation increase in latent demand as of 1996 leads to a one-half standard deviation increase in default rates in 2007. To put this in historical perspective, the point estimate for the recession year 2001 implies that a one standard deviation increase in latent demand as of 1996 results in less than one-fifth a standard deviation increase in default rates. In other words, high latent demand zip codes experience an increase in default rates more than twice as large in 2007 than in 2001, despite the fact that there is no recession in 2007. In columns 1 and 2 of Table 6, we examine a reduced form specification relating the change in default rates from 2005 to 2007 to the denial rate and the subprime share as of 1996. The two right hand side variables measure zip codes that experience a relative increase in the supply of credit from 2001 to 2005; therefore, the coefficient estimates represent the reduced form effect of credit expansion on default rates. The estimate in column 1 implies that a one standard deviation increase in 1996 latent demand leads to a 1.4 percentage point increase in default rates from 2005 to 2007, which represents more than one-third standard deviation increase in the left hand side variable. In columns 3 through 6 of Table 6, we examine the effect of supply-driven mortgage growth and house price growth from 2001 to 2005 on the change in default rates from 2005 to 18 2007. We define supply driven mortgage growth and house price growth as the predicted values from first stage regressions relating mortgage growth and house price growth from 2001 to 2005 to either the fraction of 1996 mortgage applications denied or 1996 fraction of subprime borrowers in the zip code. The first stage estimates used to predict supply driven mortgage growth and house price growth are in columns 1 and 3 of Table 3. The estimates in columns 3 through 6 of Table 6 demonstrate that supply driven mortgage growth and house price growth from 2001 to 2005 have a strong effect on the increase in default rates from 2005 to 2007. The estimate in column 3 implies that a one standard deviation increase in supply driven mortgage growth from 2001 to 2005 leads to a 4 percentage point increase in default rates, which is a one standard deviation increase in the left hand side variable. A one standard deviation increase in supply driven house price appreciation leads to a 12 percentage point increase in default rates, which is a three standard deviation change in the left hand side variable. The magnitudes are only slightly smaller when we use the fraction of subprime borrowers as of 1996 instead of the 1996 denial rate. VI. Securitization and Moral Hazard The results above imply large losses for mortgage investors from credit expansion into high latent demand zip codes. In order to estimate the marginal losses, we replicate the specification in column 3 of Table 6 with the growth in the default amount as the left hand side variable instead of the change in the default rate. The coefficient estimate from this unreported specification is 0.32, which represents the elasticity of default amount increases from 2005 to 2007 with respect to the increase in lending to high 1996 latent demand areas from 2001 to 2005. In other words, a 10% increase in lending to high denial rate zip codes leads to a 3.2% increase in default amounts. Given that foreclosure recovery rates are typically between 40 and 70% on 19 defaulted mortgages (Pence (2006)), this elasticity implies enormous losses for mortgage investors. In addition, the subprime-prime mortgage spread fell to historical lows during this period, which suggests that investors were not compensated for the additional ex post risk. These losses beg the question: Why did originators make these mortgages? Figure 2 above shows that the sharp rise in mortgage growth coincides with an increase in originators selling loans to non-mortgage agency investors. Moreover, the sale of loans by originators is significantly stronger in zip codes with high 1996 latent demand for loans (Figure 5C). In Table 7, we present further evidence that the process of selling loans is correlated with the default patterns we observe in high latent demand zip codes. Column 1 in Panel A reaffirms the result shown earlier in Figure 5C: High initial latent demand zip codes experience a larger increase in the fraction of loans sold to investors within the year. Columns 2 through 6 of Panel A disaggregate the fraction sold by the identity of the party buying the mortgage from the originating institution. The correlation in column 1 is driven by mortgages sold in private securitizations to unaffiliated investors, and to non-bank financial firms. The largest non-bank financial firms are mortgage banks that are primary arrangers of securitization pools. In other words, the increase in mortgage sales in high latent demand zip codes is driven by mortgages sold for the purpose of securitization. Column 1 of Panel B demonstrates that zip codes experiencing a relative increase in securitization also experience increases in mortgage debt to income ratios, which is consistent with originators shedding credit risk during the 2001 to 2005 expansion. Columns 2 through 6 show the correlation of the fraction of loans sold in a zip code from 2001 to 2005 by the type of investor buying the mortgage with subsequent default rates from 2005 to 2007. The estimates demonstrate that zip codes in which a larger fraction of mortgages are sold in private 20 securitizations and to non-commercial bank financial firms for the purpose of securitization experience relatively larger increases in default rates from 2005 to 2007. In contrast, column 2 of Panel B shows that zip codes in which originators sell more mortgages to affiliated investors do not experience an increase in default rates. Under the assumption that originators’ incentives are more closely aligned with affiliated versus nonaffiliated investors, these results suggest that undetected moral hazard is a potential cause for the higher default rates on mortgages sold to non-affiliated investors. In addition, column 5 of Panel B demonstrates that zip codes in which originators sell more mortgages to other commercial banks do not experience an increase in default rates. Given that commercial banks have specialized screening, these results suggest that originators only sold bad loans to unaffiliated investors lacking the skills to judge loan quality. Together with the findings in Panel A, these findings are consistent with the hypothesis that moral hazard on behalf of originators is a main culprit for the rise in default rates. As a caveat it is important to emphasize that we view our evidence on moral hazard as suggestive. It is difficult to assert that undetected moral hazard on behalf of originators caused the spike in mortgage defaults for two reasons. First, there is a lack of exogenous within-county variation across zip codes in the ability of originators to sell mortgages. Without such variation, it is difficult to rule out alternative explanations. Second, we do not have loan-level interest rate data, which makes it difficult to examine whether moral hazard is priced. VII. Concluding Remarks: What are the Macroeconomic Magnitudes of the Supply Shift? The process of mortgage originators selling and securitizing loans led to a sharp shift in the supply of mortgage credit from 2001 to 2005. The expansion in supply affected subprime customers who were traditionally marginal borrowers unable to access the mortgage market. The 21 shift in mortgage supply consequently led to a rapid rise in the risk profile of borrowers, and a surge in supply-induced house price and mortgage credit growth. These changes caused a subsequent spike in default rates, which have in turn depressed the housing market and caused financial market turmoil. The main contribution of our work is to empirically isolate a mechanism, the magnitude, and the consequences of the historic shift in mortgage supply. To help understand the macroeconomic implications of our findings, we conduct two analyses in this section. First, Figure 11 provides a geographic representation of our findings when the specifications from Tables 3 and 6 are estimated separately for each state in our sample. More specifically, Figure 11 shades each state according to the point estimate that relates mortgage origination growth (Panel A), house price growth (Panel B), and the increase in mortgage default rates (Panel C) for high 1996 latent demand zip codes state-by-state. As the figure demonstrates, our results are robust in most states in our sample.8 The supply shift documented in our results is not unique to one or two states; it is a nation-wide effect. Second, given that we have identified the expansion in credit and increase in house price due to the shift in the supply of credit, we can use our microeconomic estimates to answer an important macroeconomic counter-factual: How would mortgage lending and house prices have evolved if the shift in supply in the mortgage industry had not occurred? To answer this question, we sort zip codes by 1996 denial rates and categorize them into 20 equal bins with 5% of zip codes in each bin. Let i index each bin, and denote by di the median denial rate inside a 5% bin. Given a coefficient of 2.11 (Table 3, column 1) for the marginal effect of initial denial rate on mortgage growth from 2001 to 2005, the incremental supply-induced loan origination in bin i is We expand our sample to all zip codes in the US when using mortgage growth and growth in default rate as dependent variables to provide a more comprehensive picture across the US. The home price growth map (Panel B) is limited to states which have zip code level home price data. 8 22 equal to 2.11*Li,2001*(di – d1), where Li,2001 is aggregate loan origination in bin i in 2001. The total supply-induced loan origination in 2005 is thus equal to: 2.11 L , d –d . (7) A similar calculation can be done for house prices using the estimate of 0.34 from column 3 of Table 3. Solving the above expressions in our data gives us $83 billion of additional mortgage originations in 2005 due to supply shift, or 15% of total mortgage originations in 2005. Similarly, we find a 4.3% increase in house prices between 2001 and 2005 due to supply shift, or almost 10% of aggregate house price appreciation in the US between 2001 and 2005. It is important to emphasize that the calculations described above are an underestimate of the true impact of supply shift for two reasons. First, changes in borrower credit quality are negatively correlated with initial latent demand for mortgages which biases downward our regression estimates (as described in Section II). Second, since our empirical methodology is based on a difference-in-differences estimator, we can only estimate the relative impact of the shift in supply. In other words, we estimate the differential effect of the supply shift on high 1996 latent demand zip codes relative to low 1996 latent demand zip codes. Consequently, our calculation above disregards any level impact of the supply shift which impact all zip codes. There is evidence to suggest that this may be a significant omission. For example, even zip codes with low denial rates in 1996 took advantage of the lower lending rates by taking out home equity loans and refinancing in large amounts. The effect of the supply shift on the intensive margin of higher credit quality homeowners is material for future research. 23 References Brunnermeier, Markus and Christian Julliard, 2007. “Money Illusion and Housing Frenzies,” Review of Financial Studies, forthcoming. Case, Karl, and Robert Shiller, 1989. “The Efficiency of the Market for Single-Family Homes,” American Economic Review 79: 125-137. Chomsisengphet, Souphala and Anthony Pennington-Cross, 2006. “The Evolution of the Subprime Mortgage Market,” Federal Reserve Bank of St. Louis Review 88: 31-56. Dell’Ariccia, Giovanni, Deniz Igan, and Luc Laeven, 2008, “Credit Booms and Lending Standard: Evidence from the Subprime Mortgage Market,” Working Paper, IMF, February. Demyanyk and Van Hemert, 2007, “Understanding the Subprime Mortgage Crisis”, Working Paper, New York University. Diamond, D., 1991, “Monitoring and reputation: The choice between bank loans and privately placed debt,” Journal of Political Economy, 99, 689-721. Doms, Mark, Fred Furlong, and John Krainer, 2007. “Subprime Mortgage Delinquency Rates,” Federal Reserve Bank of San Francisco Working Paper. Gabriel, Stuart and Stuart Rosenthal, 2007, “Secondary Markets, Risk, and Access to Credit: Evidence from the Mortgage Market,” Working Paper, Syracuse University. Genesove, David, and Christopher Mayer, 1997. “Equity and Time to Sale in the Real Estate Market,” American Economic Review, 87: 255-269. Gerardi, Kristopher, Harvey Rosen, and Paul Willen, 2007, “Subprime Outcomes: Risky Mortgages, Homeownership Experiences, and Foreclosures,” Working Paper, Federal Reserve Bank of Boston, July. --, 2001, “Loss Aversion and Seller Behavior: Evidence from the Housing Market,” Quarterly Journal of Economics, 1233-1260. Glaeser, Edward and Joseph Gyourko, 2005, “Urban Decline and Durable Housing,” Journal of Political Economy 113: 345-375. Himmelberg, Charles, Christopher Mayer, and Todd Sinai, 2005, “Assessing High House Prices: Bubbles, Fundamentals, and Misperceptions,” Journal of Economic Perspectives 19: 67-92. Holmstrom, B. and J. Tirole, 1997, “Financial intermediation, loanable funds, and the real sector,” Quarterly Journal of Economics, 112, 663-691. 24 Hurst, Erik and Frank Stafford, 2004, “Home is Where the Equity Is: Mortgage Refinancing and Household Consumption,” Journal of Money, Credit, and Banking 36: 985-1014. Keys, Benjamin, Tanmoy Mukherjee, Amit Seru and Vikrant Vig 2008, “Securitization and Screening: Evidence From Subprime Mortgage Backed Securities”, working paper. Pence, Karen, 2006, “Foreclosing on Opportunity: State Laws and Mortgage Credit,” Review of Economics and Statistics, vol. 88 (February 2006), pp. 177-182 Poterba, James, 1984. “Tax Subsidies to Owner-Occupied Housing: An Asset-Market Approach,” Quarterly Journal of Economics 99: 729-752. Stein, Jeremy, 1995. “Prices and Trading Volume in the Housing Market: A Model with DownPayment Effects,” Quarterly Journal of Economics 110: 379-406. Stiglitz, Joseph and Andrew Weiss, 1981. “Credit Rationing in Markets with Imperfect Information,” American Economic Review 71: 393-410. 25 Figure 1 Mortgage and non-Mortgage Debt Outstanding, Indexed to 1996 This figure presents total mortgage and non-mortgage consumer debt outstanding for the U.S. from 1992 to 2007, indexed to 1996. Total non-mortgage consumer debt includes student loans, auto loans, consumer loans, and outstanding credit card balances. Data are from Equifax Predictive Services. 3.5 3 2.5 2 1.5 1 0.5 0 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 Mortgage debt outstanding non-Mortgage debt outstanding Figure 2 Fraction of Mortgages Sold to Non-Mortgage Agency Institutions This figure presents the fraction of originated mortgages that are sold to non-mortgage agency institutions within one year of origination. Non-mortgage agency institutions include all third parties except for Fannie Mae, Freddie Mac, Ginnie Mae, and Farmer Mac. Data are from HMDA. 0.6 0.55 0.5 0.45 0.4 0.35 0.3 0.25 0.2 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 Figure 3A Debt to Income Ratios for Accepted Mortgage Applications, Relative to 1996 This figure presents the mortgage debt to income ratios of accepted mortgage applications at the median, 75th, and 90th percentiles from 1996 to 2006. The 1996 level is substracted from each series. Data are from HMDA. 1.4 1.2 1 0.8 0.6 0.4 0.2 0 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 Debt to Income of Accepted Applications--Median D bt t I fA t d A li ti M di Debt to Income of Accepted Applications--90th Percentile Debt to Income of Accepted Applications--75th Percentile D bt t I fA t d A li ti 75th P til Figure 3B Originated Mortgage Debt for Home Purchase to Income Ratios, Relative to 1996 This figure presents the average originated mortgage debt to aggregate income ratio across zip codes from 1998 to 2005. The 1998 level is substracted from the series. Originated mortgage debt is from HMDA and aggregate income is from the IRS. * indicates data missing for the year in question. 0.16 0.14 0.12 0.1 0.08 0.06 0.04 0.02 0 1998 1999* 2000* 2001 2002 2003* 2004 2005 Figure 4 Default Rates for Mortgage and non-Mortgage Debt , Indexed to 1996 This figure presents the default rate for consumer debt outstanding for the U.S. from 1992 to 2007, indexed to 1996. The total non-mortgage default rate is calculated using non-mortgage debt which includes student loans, auto loans, consumer loans, and outstanding credit card balances. Data are from Equifax Predictive Services. 2.2 2 1.8 1.6 1.4 1.2 1 0.8 0.6 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 Mortgage debt default rates non-Mortgage debt default rates Figure 5A Mortgage Denial Rates For High 1996 Denial Zip Codes This figure plots the estimated coefficients of β and 95% confidence intervals for each year for the following first difference county fixed effects specifications: Denied zct − Denied zc ,1996 = α c + β t * HighLatentDemand zc ,1996 + ε zct for t = 1997,1998,...,2007 0.1 0.05 0 1996 -0.05 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 -0.1 -0.15 -0.2 -0.25 -0.3 -0.35 Figure 5B Originated Mortgage Debt to Income Ratio For High 1996 Denial Zip Codes This figure plots the estimated coefficients of β and 95% confidence intervals for each year for the following first difference county fixed effects specifications: D 2 I zct − D 2 I zc ,1996 = α c + β t * HighLatentDemand zc,1996 + ε zct for t = 1997,1998,..., 2007 * indicates data missing for the year in question. 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0 1998 1999* 2000* 2001 2002 2003* 2004 2005 Figure 5C Disintermediation For High 1996 Denial Zip Codes This figure plots the estimated coefficients of β and 95% confidence intervals for each year for the following first difference county fixed effects specifications: Sold zct − Sold zc ,1996 = α c + β t * HighLatentDemand zc ,1996 + ε zct for t = 1997,1998,..., 2007 Disintermediated loans are loans sold to any third party except for Fannie Mae, Freddie Mac, Ginnie Mae, and Farmer Mac within 1 year of origination. 0.35 0.3 0.25 0.2 0.15 0.1 0.05 0 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 -0.05 Figure 6A Amount of Originated Mortgages for Home Purchase For High 1996 Denial Zip Codes This figure plots the estimated coefficients of β and 95% confidence intervals for each year for the following first difference county fixed effects specifications: Ln( Amt ) zct − Ln( Amt ) zc,1996 = α c + β t * HighLatentDemand zc ,1996 + ε zct for t = 1997,1998,..., 2007 4 3.5 3 2.5 2 1.5 1 0.5 0 1996 -0.5 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 Figure 6B Outstanding Mortgage Debt For High 1996 Denial Zip Codes This figure plots the estimated coefficients of β and 95% confidence intervals for each year for the following first difference county fixed effects specifications: Ln(MortgageDebt ) zct − Ln( MortgageDebt ) zc ,1996 = α c + βt * HighLatentDemand zc ,1996 + ε zct for t = 1997,1998,...,2007 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0 1996 -0.1 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 -0.2 Figure 7 Relative House Price Appreciation For High 1996 Denial Zip Codes This figure plots the estimated coefficients of β and 95% confidence intervals for each year for the following first difference county fixed effects specifications: Ln( HP) zct − Ln( HP ) zc ,1996 = α c + βt * HighLatentDemand zc ,1996 + ε zct for t = 1997,1998,..., 2007 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0 1996 -0.1 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007Q1 1995-01-01 1994-11-01 1994-09-01 1994-07-01 1994-05-01 1994-03-01 1994-01-01 1993-11-01 1993-09-01 1993-07-01 1993-05-01 2006-01-01 2005-11-01 2005-09-01 2005-07-01 2005-05-01 2005-03-01 2005-01-01 2004-11-01 2004-09-01 2004-07-01 2004-05-01 Figure 8A Yield on 3 Month Treasury Bill, 1990 - 1994 1993-03-01 1993-01-01 1992-11-01 1992-09-01 1992-07-01 1992-05-01 1992-03-01 1992-01-01 1991-11-01 1991-09-01 1991-07-01 1991-05-01 1991-03-01 1991-01-01 1990-11-01 1990-09-01 1990-07-01 1990-05-01 1990-03-01 1990-01-01 8.00 7.00 6.00 5.00 4.00 3.00 2.00 Figure 8B Yield on 3 Month Treasury Bill, 2001 - 2005 2004-03-01 2004-01-01 2003-11-01 2003-09-01 2003-07-01 2003-05-01 2003-03-01 2003-01-01 2002-11-01 2002-09-01 2002-07-01 2002-05-01 2002-03-01 2002-01-01 2001-11-01 2001-09-01 2001-07-01 2001-05-01 2001-03-01 2001-01-01 6.00 5.00 4.00 3.00 2.00 1.00 0.00 Figure 9A Growth in Originated Mortgages for High Denial Zip Codes During Falling Interest Rate Environment This figure plots the estimated coefficients of β and 95% confidence intervals for each year for the following first difference county fixed effects specifications: 2.5 2 1.5 2001 - 2005 1 0.5 0 -0.5 1990 - 1994 -1 -1.5 Year 0 Year 1 Year 2 Year 3 Year 4 Figure 9B Relative House Price Appreciation for High Denial Zip Codes During Falling Interest Rate Environment This figure plots the estimated coefficients of β and 95% confidence intervals for each year for the following first difference county fixed effects specifications: 0.4 0.35 0.3 2001 - 2005 0.25 0.2 0.15 0.1 0.05 0 1990 - 1994 -0.05 -0.1 Year 0 Year 1 Year 2 Year 3 Year 4 Figure 10 Mortgage Default Rates for High 1996 Denial Zip Codes This figure plots the estimated coefficients of β and 95% confidence intervals for each year for the following first difference county fixed effects specifications: DefRatezct − DefRatezc ,1996 = α c + βt * HighLatentDemand zc ,1996 + ε zct for t = 1997,1998,...,2007 0.25 0.2 0.15 0.1 0.05 0 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 -0.05 Figure 11 Panel A The Effect of Mortgage Credit Expansion on Mortgage Amounts, By State The map displays coefficient estimates of the effect of high 1996 latent demand in a zip code on growth in originated mortgage amount for home purchase from 2001 to 2005 by state. All specifications include county fixed effects.       Figure 11 Panel B The Effect of Mortgage Credit Expansion on Home Price Growth, By State The map displays coefficient estimates of the effect of high 1996 latent demand in a zip code on growth in home prices from 2001 to 2005 by state. All specifications include county fixed effects. (States in pure white color do not have home price data and hence do not have an estimate)         Figure 11 Panel C The Effect of Mortgage Credit Expansion on Growth in Default Rate, By State The map displays coefficient estimates of the effect of high 1996 latent demand in a zip code on growth in default rate from 2005 to 2007 by state. All specifications include county fixed effects.     Table 1 Summary Statistics This table presents summary statistics for the 2,920 zip codes in our sample. Mean Equifax Data Mortgage debt as fraction of total debt, 1996 Home equity debt as fraction of total debt, 1996 Non-home debt as fraction of total debt, 1996 Mortgage debt annualized growth, 1996 to 2001 Non-home debt annualized growth, 1996 to 2001 Mortgage debt annualized growth, 2001 to 2005 Non-home debt annualized growth, 2001 to 2005 Mortgage default rate, 1996 Non-home default rate, 1996 Mortgage default rate change, 1996 to 2005 Non-home default rate change, 1996 to 2005 Mortgage default rate change, 2005 to 2007 Non-home default rate change, 2005 to 2007 Sub-prime consumer fraction (under 659) Fiserve Case Shiller Weiss Data Home price annualized growth, 1996 to 2001 Home price annualized growth, 2001 to 2005 HMDA Data Mortgages for home purchase annualized growth, 1996 to 2001 Mortgages for home purchase annualized growth, 2001 to 2005 Fraction of applications denied, 1996 Fraction of mortgages sold to non-agency investors, 2001 Change in fraction sold to non-agency investors, 2001 to 2005 Decennial Census, IRS, and Census Statistics of U.S. Business Median household income, thousands, 2000 Fraction non-white, 2000 Fraction with education less than high school, 2000 Fraction unemployed, 2000 Fraction of housing units rented, 2000 Income growth (from IRS), 2001 to 2005 Wage growth (from Business Conditions), 2001 to 2004 Employment growth, 2001 to 2004 Establishment growth, 2001 to 2004 Crime index growth, 2001 to 2005 0.741 0.044 0.215 0.069 0.061 0.102 0.046 0.030 0.069 -0.003 -0.008 0.035 0.018 0.288 SD 0.106 0.028 0.085 0.042 0.042 0.046 0.046 0.024 0.035 0.028 0.030 0.038 0.026 0.112 10th 0.600 0.018 0.121 0.023 0.019 0.054 -0.007 0.005 0.029 -0.032 -0.041 -0.003 -0.010 0.156 50th 0.761 0.037 0.201 0.068 0.061 0.103 0.047 0.024 0.063 -0.004 -0.007 0.027 0.016 0.270 90th 0.854 0.075 0.330 0.115 0.103 0.149 0.098 0.060 0.114 0.025 0.024 0.088 0.049 0.447 0.073 0.113 0.023 0.044 0.041 0.039 0.075 0.117 0.102 0.166 0.095 0.134 0.219 0.277 0.263 0.046 0.078 0.079 0.074 0.057 0.039 0.048 0.122 0.184 0.191 0.097 0.136 0.211 0.275 0.262 0.152 0.220 0.329 0.374 0.334 55.7 0.202 0.166 0.051 0.300 0.127 0.072 0.025 0.057 0.024 20.0 0.197 0.117 0.032 0.160 0.094 0.121 0.192 0.100 0.050 33.4 0.028 0.051 0.023 0.105 0.033 -0.050 -0.172 -0.039 -0.027 52.5 0.132 0.135 0.042 0.277 0.114 0.072 0.017 0.046 0.016 81.8 0.496 0.320 0.091 0.529 0.241 0.196 0.217 0.170 0.087     Table 2 Correlation of Fraction of 1996 Mortgage Applications Denied with Other Variables This table presents correlations of our main measure of latent demand for mortgages in a zip code, the fraction of 1996 mortgage applications denied, with other variables, after controlling for county fixed effects. The sample includes 2,920 zip codes. Correlation with fraction of 1996 applications denied Alternative measures of latent demand Fraction of subprime borrowers in 1996 (under 619) Fraction of subprime borrowers in 1996 (under 659) Bank branches per capita in 2000 Fraction of population with credit score in 2000 Fraction of loans backed by FHA in 1996 Fraction of housing units rented in 2000 Demographic variables from Census 2000 Fraction non-white Median household income Per capital income Poverty rate Fraction with less than high school education Fraction unemployed Measures of future growth in economic opportunity Income growth, 2001 to 2005 Wage growth, 2001 to 2004 Employment growth, 2001 to 2004 Establishment growth, 2001 to 2004 **,* Correlation statistically distinct from 0 at the 1% and 5% levels, respectively 0.795** 0.789** -0.282** -0.609** 0.516** 0.399** 0.680** -0.585** -0.568** 0.643** 0.662** 0.567** -0.209** -0.028 -0.078** -0.122**       Table 3 The Effect of Mortgage Credit Expansion on Mortgage Amounts and House Prices Columns 1 and 2 of Panel A present estimates of the effect of high 1996 latent demand in a zip code on growth in originated mortgage amount for home purchase and growth in mortgage debt outstanding from 2001 to 2005, respectively. Column 3 presents estimates of the effect of high 1996 latent demand in a zip code on growth in house prices from 2001 to 2005. Panel B examines how outcomes vary with an alternative measure of high latent demand zip codes: the fraction of borrowers with a credit score under 659. All specifications include county fixed effects. Panel A: Fraction of 1996 applications denied (1) (2) (3) Originated mortgage Mortgage debt House price growth amount for home outstanding growth 2001 to 2005 purchase growth 2001 to 2005 2001 to 2005 Fraction of 1996 applications denied 2.109** (0.115) 0.103 (0.083) -0.036 (0.057) 0.008 (0.042) 0.011 (0.090) -0.050 (0.162) 2897 0.41 0.444** (0.064) 0.500** (0.046) 0.044 (0.031) -0.015 (0.023) 0.669** (0.050) 0.408** (0.090) 2897 0.30 0.339** (0.016) -0.031** (0.011) -0.004 (0.008) 0.009 (0.006) -0.032** (0.012) -0.079** (0.022) 2897 0.96 Income growth, 2001 to 2005 Wage growth, 2001 to 2004 Employment growth, 2001 to 2004 Establishment growth, 2001 to 2004 Crime growth, 2001 to 2005 N R2 Panel B: Fraction of subprime borrowers in 1996 (1) (2) (3) Originated mortgage Mortgage debt House price growth amount for home outstanding growth 2001 to 2005 purchase growth 2001 to 2005 2001 to 2005 Fraction subprime borrowers, 1996 1.331** (0.077) 0.278** (0.087) -0.025 (0.057) 0.013 (0.042) -0.029 (0.091) -0.126 (0.163) 0.269** (0.043) 0.533** (0.048) 0.046 (0.031) -0.015 (0.023) 0.660** (0.050) 0.387** (0.090) 0.266** (0.010) 0.018 (0.011) -0.001 (0.007) 0.010 (0.005) -0.036** (0.012) -0.069** (0.021) Income growth, 2001 to 2005 Wage growth, 2001 to 2004 Employment growth, 2001 to 2004 Establishment growth, 2001 to 2004 Crime growth, 2001 to 2005 N R2 **,* Coefficient estimate statistically distinct from 0 at the 1% and 5% levels, respectively     Table 4 Income, Mortgage Origination Growth and House Price Appreciation Panels A and B examine the correlation between income growth and origination and house price growth for given time periods. Panels C and D examine the correlation between initial income levels and origination and house price growth for given time periods. All specifications include county fixed effects. Panel A: Origination growth and income growth (1) (2) (3) Origination growth Origination growth Origination growth 1991 to 1998 1998 to 2001 2001 to 2005 Income growth, 1991 to 1998 0.447** (0.071) Income growth, 1998 to 2001 0.461** (0.083) Income growth, 2001 to 2005 -0.241** (0.086) Panel B: House price growth and income growth (1) (2) (3) House price growth House price growth House price growth 1991 to 1998 1998 to 2001 2001 to 2005 0.140** (0.008) 0.061** (0.013) -0.080** (0.012) Income growth, 1991 to 1998 Income growth, 1998 to 2001 Income growth, 2001 to 2005 **,* Coefficient estimate statistically distinct from 0 at the 1% and 5% levels, respectively       Table 5 Non-home Debt Growth This table presents estimates of specifications that replicate specifications reported in Table 3 but with non-home as the dependent variable. All specifications include county fixed effects. Non-home debt growth and defaults (1) (2) Non-home debt growth, 2001 to 2005 Fraction of 1996 applications denied -0.117* (0.059) -0.146** (0.039) 0.400** (0.043) 0.013 (0.029) -0.012 (0.021) 0.585** (0.046) 0.386** (0.083) 0.362** (0.044) 0.010 (0.029) -0.013 (0.021) 0.584** (0.046) 0.360** (0.083) Fraction of subprime borrowers in 1996 Income growth, 2001 to 2005 Wage growth, 2001 to 2004 Employment growth, 2001 to 2004 Establishment growth, 2001 to 2004 Crime growth, 2001 to 2005 N 2897 R2 0.24 **,* Coefficient estimate statistically distinct from 0 at the 1% and 5% levels, respectively 2897 0.24           Table 6 The Effect of Mortgage Credit Expansion on Mortgage Default Rates This table presents estimates of how the shift in mortgage supply from 2001 to 2005 affects default rates from 2005 to 2007. Columns 1 and 2 report estimates from the reduced form relating the increase in defaults from 2005 to 2007 to measures of zip codes experiencing relative supply increases from 2001 to 2005. Columns 3 through 6 relate the increase in default rates from 2005 to 2007 to measures of supply driven house price growth and mortgage origination growth, which are obtained from a first stage that relates these variables to two different supply shifters: the fraction of denied applications in 1996 and the fraction of subprime borrowers in 1996. The first stage estimates are in Table 3. All specifications include county fixed effects. (1) (2) (3) (4) (5) (6) Change in mortgage default rates from 2005 to 2007 Fraction of 1996 applications denied Fraction of subprime borrowers in 1996 Supply driven originated mortgage amount for home purchase growth, 2001 to 2005 Supply driven house price growth, 2001 to 2005 0.169** (0.010) 0.136** (0.006) 0.080** (0.006) 0.498** (0.033) -0.078** (0.007) -0.001 (0.005) 0.005 (0.004) 0.012 (0.008) 0.040** (0.014) Reduced form -0.052** (0.007) 0.001 (0.005) 0.006 (0.003) 0.010 (0.008) 0.046** (0.014) Reduced form -0.086** (0.009) 0.002 (0.006) 0.005 (0.005) 0.011 (0.010) 0.044* (0.018) 1996 Denied applications 2897 0.01 -0.062** (0.008) 0.001 (0.005) 0.001 (0.004) 0.028** (0.009) 0.079** (0.016) 1996 Denied applications 2897 0.23 -0.080** (0.010) 0.003 (0.007) 0.005 (0.005) 0.013 (0.011) 0.059** (0.020) 1996 Subprime fraction 2897 0.01 0.102** (0.007) 0.513** (0.028) -0.061** (0.008) 0.001 (0.005) 0.001 (0.004) 0.028** (0.009) 0.081** (0.016) 1996 Subprime fraction 2897 0.21 Income growth, 2001 to 2005 Wage growth, 2001 to 2004 Employment growth, 2001 to 2004 Establishment growth, 2001 to 2004 Crime growth, 2001 to 2005 Supply shifter? N 2897 2897 R2 0.39 0.42 **,* Coefficient estimate statistically distinct from 0 at the 1% and 5% levels, respectively       Table 7 Evidence of a Securitization Channel Panel A presents coefficient estimates relating the change in the fraction of originated mortgages sold in a zip code to latent demand as of 1996. Column 1 of Panel B relates the change in median debt to income ratios of accepted applications from 2001 to 2005 to the change in the fraction of loans sold to investors. Columns 2 through 5 of Panel B presents estimates relating default rates from 2005 to 2007 to the fraction of loans sold by originators to investors from 2001 to 2005. All specifications include county fixed effects and control variables for income, wage, employment, establishment, and crime growth. Panel A: High latent demand zip codes and mortgage sales (1) (2) (3) (4) Change in fraction Change in fraction Change in fraction Change in fraction sold to all sold to affiliates, sold in private sold to banks, 2001 investors, 2001 to 2001 to 2005 securitizations, to 2005 2005 2001 to 2005 Fraction of 1996 applications denied N R2 0.161** (0.014) 2897 0.47 -0.072** (0.008) 2897 0.54 0.177** (0.006) 2897 0.63 0.004 (0.004) 2897 0.45 (5) Change in fraction sold to non-bank financial firms, 2001 to 2005 0.125** (0.006) 2897 0.58 Panel B: Mortgage sales and changes in default rates (1) (2) (3) (4) (5) Change in Debt Change in mortgage default rates from 2005 to 2007 to Income Ratio 2001 to 2005 Change in fraction sold to all investors, 2001 to 2005 Change in fraction sold to affiliates, 2001 to 2005 Change in fraction sold in private securitizations, 2001 to 2005 Change in fraction sold to banks , 2001 to 2005 Change in fraction sold to non-bank financial firms , 2001 to 2005 N 2919 2897 R2 0.77 0.33 **,* Coefficient estimate statistically distinct from 0 at the 1% and 5% levels, respectively 0.314** (0.056) 0.081** (0.014) -0.220** (0.025) 0.332** (0.030) -0.008 (0.046) (6) 2897 0.34 2897 0.35 2897 0.32 0.405** (0.029) 2897 0.36   Appendix Table Characteristics of Zip Codes With and Without House Price Data This table compares the 3,056 zip codes for which FCSW collects house price data to the 16,312 zip codes for which house price data are unavailable. House price data available Fraction urban, 2000 Population, 2000 Median household income (thousands), 2000 Poverty rate, 2000 Fraction of housing units built in last 5 years, 2000 Fraction of households that have moved in last 5 years, 2000 Mortgage default rate, 1996 0.919 6,630 55.7 0.092 0.084 0.352 0.030 House price data not available 0.464 4,763 39.8 0.129 0.116 0.337 0.028