Subprime lending and the housing bubble Tail wags dog

ub erry A 926 ullerto ifornia Article history: Received 20 April 2008 Available online 20 September 2008 The cause of the ‘‘housing bubble” associated with the sharp rise and then drop in home itan areas. Figs. 1 and 2 depict the CS Index in levels and as quarterly returns over the period 1998–2007 in the aggregate and by three price-level tiers. The widespread availability of subprime loan products during this period, while arguably increasing consumption levels and homeownership rates, has been broadly blamed for this ‘‘bubble”. The share of subprime mortgage prod- ucts peaked at 23.5% of all mortgages originated during 1051-1377/$ - see front matter � 2008 Elsevier Inc. All rights reserved. * Corresponding author. Fax: +1 949 824 9308. E-mail addresses: mcoleman06@merage.uci.edu (M. Coleman), mlacour-little@fullerton.edu (M. LaCour-Little), kvandell@merage.uci.edu (K.D. Vandell). 1 Large lenders reporting large losses include Countrywide, Citigroup, Freddie Mac and Fannie Mae. High profile bankruptcies included New Century and American Home Mortgage. Journal of Housing Economics 17 (2008) 272–290 Contents lists available at ScienceDirect Journal of Housi journal homepage: www.e doi:10.1016/j.jhe.2008.09.001 1. Introduction Recent turmoil in the mortgage market—in particular a contraction in liquidity beginning in August 2007, sig- nificant increases in the rates of defaults and foreclo- sures, the failure of a number of mortgage firms, and large losses incurred by financial institutions and inves- tors in mortgage and mortgage-related assets1—has at- tracted considerable attention from the media, policymakers, and analysts. It is also now widely recog- nized that prices in the housing market, after a number of years of very rapid growth, have been declining at an increasingly rapid rate since sometime in early 2006. As of January 2008, based on the S&P/Case-Shiller repeat sales index (CS Index), house prices nationally had fallen 12.5% year-over-year, with declines over 20% in some metropol- JEL Classification: G21 G28 H81 R1 Keywords: Mortgage House prices Subprime Fannie Mae Freddie Mac Housing bubble prices over the period 1998–2008 has been the focus of significant policy and research attention. The dramatic increase in subprime lending during this period has been broadly blamed for these market dynamics. In this paper we empirically investigate the validity of this hypothesis vs. several other alternative explanations. A model of house price dynamics over the period 1998–2006 is specified and estimated using a cross-sectional time-series data base across 20 metropolitan areas over the period 1998–2006. Results suggest that prior to early 2004, economic fundamentals provide the primary explanation for house price dynamics. Subprime credit activity does not seem to have had much impact on sub- sequent house price returns at any time during the observation period, although there is strong evidence of a price-boosting effect by investor loans. However, we do find strong evidence that a credit regime shift took place in late 2003, as the GSE’s were displaced in the market by private issuers of new mortgage products. Market fundamentals became insignificant in affecting house price returns, and the price-momentum conditions charac- teristic of a ‘‘bubble” were created. Thus, rather than causing the run-up in house prices, the subprime market may well have been a joint product, along with house price increases, (i.e., the ‘‘tail”) of the changing institutional, political, and regulatory environment charac- teristic of the period after late 2003 (the ‘‘dog”). � 2008 Elsevier Inc. All rights reserved. a r t i c l e i n f o a b s t r a c t Subprime lending and the housing b Major Coleman IV a, Michael LaCour-Little b, K a The Paul Merage School of Business, University of California, Irvine, Irvine, C bCollege of Business and Economics, California State University at Fullerton, F cCenter for Real Estate, The Paul Merage School of Business, University of Cal ble: Tail wags dog? D. Vandell c,* 97, USA n, CA 92831, USA , Irvine, Irvine, CA 92697, USA ng Economics l sevier .com/locate / jhe M. Coleman IV et al. / Journal of Housing Economics 17 (2008) 272–290 273 15 0 20 0 25 0 H PI q ua rte rly re tu rn (% ) 2006, roughly coincident with the peak in the housing market (Inside Mortgage Finance, 2007). In this paper we empirically investigate the claim that increased credit availability in the subprime sector drove the housing boom against five alternative explanations for the current dynamic that have been, or could be, of- fered: (1) economic fundamentals (e.g., employment, in- come, population increases) were the primary drivers of price changes; (2) the problem was not subprime lending per se, but the Fed’s dramatic reductions, then increases in interest rates during the early- mid-2000’s; (3) the hous- ing ‘‘boom” was concentrated in those markets with signif- icant supply-side restrictions, which tend to be more price- volatile; (4) the problem was not in the excess supply of 10 0 1998q3 2000q3 20 Year Fig. 1. Case–Shiller housing price indices by Price Tier. Single-family home purch tiers are standardized to 100 at the start of 2000. Source: S&P Case–Shiller Hom -2 0 2 4 6 H PI q ua rte rly re tu rn (% ) 1998q3 2000q3 20 Year, Fig. 2. Quarterly returns in the Case–Shiller housing price indices by Price Tier. Si of the home. Source: S&P Case–Shiller Home Price Index. Low Tier Index Mid Tier Index credit in aggregate, or the increase in subprime per se, but rather in the increased or reduced presence of certain other mortgage products; and (5) fundamental changes in the legal/political/regulatory environment resulted in strong incentives for a surge in lending and securitization by private issuers under looser underwriting standards which were not sustainable.2 To preview our results: we confirm the dominance of macroeconomic fundamentals 02q3 2004q3 2006q3 ,Quarter High Tier Index ase transactions are classified into terciles by price of the home. All three e Price Index. 02q3 2004q3 2006q3 Quarter Low Return Mid Return High Return ngle-family home purchase transactions are classified into terciles by price 2 Note that this does not exhaust all possible alternative explanations for this dynamic. Others offered include the arguments that the problem with prices was primarily in the supply of new housing, not with the availability and cost of mortgage credit, and that the problem was primarily one of fraud on the part of aggressive mortgage brokers or borrowers. prior to late 2003, find moderate support for the influence of supply constraints (especially in high-end markets), but find virtually no support for the proposition that subprime lend- ing per se had much to do with subsequent price rises (although investor loans and to a degree alt-A loans clearly did). The most surprising and important result, however, was the great impact that the displacement of the GSE’s by private ABS issuers in late 2003 to the end of 2006 had in disconnecting market prices from their fundamentals, boosting loan and unit-production volumes, and accelerat- ing house price returns. tals, in particular, the growth of income and the decline in interest rates. Himmelberg et al. (2005) also focus on the ability of economic fundamentals to explain recent house price pat- 274 M. Coleman IV et al. / Journal of Housing Economics 17 (2008) 272–290 2. Literature review In the interest of brevity, we will not provide an exten- sive literature review here, preferring to focus on a few key papers. With respect to the measurement of house price movements, there has been substantial effort expended over the years in the creation and evaluation of alternative house price indices. Most such indices are based upon median sales prices or hedonic price estimates (e.g., FHA, NAR, DataQuick). Case and Shiller (1987, 1989) first apply the repeat-sales estimation technique, now generally re- garded as the best available method for assessing house price movements over time because it is the only tech- nique that is able econometrically to come close to elimi- nating the potential biases associated with uncontrolled- for variations in neighborhood or structural amenities across units.3 We apply the CS approach in this study. With respect to the literature on housing bubbles, there has been much recent activity owing to recent increases in the volatility of price cycles in the housing market. Case and Shiller (2003) argue that the term ‘‘bubble” refers to a situation in which widespread expectations of future price increases cause prices to be temporarily elevated. In turn, the expectation of large price increases may have a strong impact on demand if households believe that home prices are very unlikely to fall, and certainly not likely to fall for long, so that there is little risk associated with a home purchase. They note, too, that the mere presence of rapid price increases is not in itself conclusive evidence of a bubble, since economic fundamentals may explain much of the observed increase. They find that that income growth alone explains the pattern of recent home price in- creases in most states and falling interest rates explain much of the recent run-up nationally. Likewise, McCarthy and Peach (2004) argue that the recent upturn in homes prices is largely attributable to strong market fundamen- 3 Case and Shiller acknowledge that their index built upon earlier work by Bailey et al. (1963) who first introduced the repeat-sales estimation methodology to the housing market. CS includes all recorded sales representing a transfer of property ownership. They exclude refinancings and REO sales by lenders on defaulted loans and take into account the effects of interim property improvements on observed prices. Their primary results are available for 20 large metropolitan regions, as well as larger regional aggregations and 3 price tiers. The S&P/Case Shiller Indices are currently calculated by Fiserv, Inc., which also offers indices covering a number of zip codes (about 15% of the total) and metro areas using the Case–Shiller methodology. Since November 2006, a futures product based upon the Case–Shiller Index has been traded on the Chicago Mercantile Exchange (see http://www.cme.com/trading/prd/re/housing.html). terns, constructing measures of the annual cost of single- family housing for 46 metropolitan areas in the United States over the period 1995–2004 and comparing those costs to the cost of renting. They argue that metrics such as the growth rate of house prices, the price-to-rent ratio, and the price-to-income ratio fail to account both for the time series pattern of real long-term interest rates and pre- dictable differences in the long-run growth rates of house prices across local markets. They find that from the trough of 1995–2004, the cost of owning rose somewhat relative to the cost of renting, but not, in most cities, to levels implying that houses were overvalued. Pavlov and Wachter, in a series of papers (2004, 2006a, 2006b), develop and test models that examine the implica- tions of aggressive non-recourse asset-based lending that under-price default risk. They demonstrate expectations of greater asset price volatility and deeper asset price ‘‘crashes” following negative demand shocks. The causes are relaxed income constraints (on the up side) freeing up latent demand for home ownership and (on the down side) the decline in the availability of aggressive lending activities following the demand shock. Empirical tests make use of international data and data from Los Angeles to provide evidence of under-pricing of default risk on the upside, coupled with over-valuation of assets, along with more extreme declines afterward. Another strand of the literature focuses on supply con- straints. Glaeser et al. (2005) focus on regulatory constraints affecting the elasticity of housing supply. They argue that a declining supply elasticity resulting from increased local development regulations in certain cities has caused prices to rise excessively in recent years. These arguments are con- sistent with Malpezzi (1999a, 1999b) and Malpezzi and Maclennan (2001), that cross-sectional variation in regula- tory constraints helps explain variation in house price dynamics through its effect on supply elasticity. The above papers focus on factors unique to U.S. eco- nomic market conditions as the cause of the bubble. Shiller (2007), however, notes that the recent run-up in house prices has occurred, not just in the U.S., but also in Australia, Canada, China, France, India, Ireland, Italy, Korea, Russia, Spain, and the United Kingdom. The coincidence of housing booms across countries would seem to cast doubt on the argument that purely local phenomena, such as supply constraints, could be responsible for house price growth patterns.4 Moreover, Shiller argues, the boom in the U.S. may be best understood as a series of regional booms, starting at different times. Shiller characterizes the boom in home prices as a classic speculative bubble, driven by extravagant expectations for future price increases, and argues that survey research measuring consumer expecta- tions confirms this description. 4 We note that the same argument could be made cross-sectionally across MSAs within the U.S., where there exist widely varying price trends and development regulations of widely varying degrees of restrictiveness. More recently, Mian and Sufi (2008) use data from approximately 3000 zip codes to examine the effect of freeing latent demand through an increase in the supply of mortgage credit driven by securitization. They measure house price movements using zip-code level Case–Shiller indices and find a positive effect of high latent demand the increased percentage of lending on properties that are not owner-occupied, and the increased use of simulta- neous-close second liens (also called ‘‘piggybacks”). The GAO has reported to Congress on the growth of non-tradi- tional mortgage products (GAO, 2006), and regulatory bodies have set forth guidance on risks and best practices for financial institutions engaged in such lending FFIEC (2006, 2007).6 LaCour-Little and Yang (2007) trace the his- tory of recent mortgage contract innovations such as inter- est-only and pay-option ARMs showing that such products are rationally preferred by households with lower risk aver- M. Coleman IV et al. / Journal of Housing Economics 17 (2008) 272–290 275 on house price growth during 2001–2005, after controlling for income and employment growth, and other fundamen- tals. They measure latent demand by the percentage of home purchase loan applications denied in 1996. While Mian and Sufi argue that an increase in the supply of mort- gage credit had a discernable positive effect on subsequent house price growth rates, their estimate of its magnitude is relatively small: about 10% of the aggregate house price appreciation during 2001–2005 is due to a that supply shift (Mian and Sufi, 2008,, pp. 32). Moreover, their measure of increase in the supply of mortgage credit is represented by the increase in the volume of all loans originated, not just subprime loans. This increase is found to be highly cor- related with the intensity of non-agency securitization.5 In the paper arguablymost closely related to ours,Whea- ton and Nechayev (forthcoming) (henceforthW–N) investi- gate whether the growth in housing prices between 1998 and 2005 can explained by increases in demand fundamen- tals such as population, income growth and the decline in interest rates. Wheaton and Nechayev estimate time series models for multiple markets using data from 1975 to 1998 and use those models to predict house price growth occur- ring during 1998–2005, finding that in all markets actual house price growth outstripped that which would be pre- dicted by economic fundamentals by a considerablemargin. They use an AR(1) model of log changes in house prices as measured by the OFHEO repeat sales indices for 59 MSA markets, controlling for total employment, total personal in- come divided by employments, and the 30-year fixedmort- gage rate. Wheaton and Nechayev hypothesize that house price growth in excess of that implied by economic funda- mentals is related to the emergence of risk-priced subprime mortgage lending and theunusual growth in thedemand for second homes and/or investment properties over the time period studied. To test these hypotheses they examine cross-sectional forecast errors produced by using the eco- nomic fundamentals model to predict house price changes. Results establish a statistical association betweenmeasures of credit availability and the volume of second home pur- chases and the cross-sectional forecast error in house price changes, but W–N caution that inferring causality from this relationship is difficult. Later, we will compare the assump- tions and results from our current effort with those ofW–N. Mortgage market trends and the changing mix of credit instruments being made available and demanded by bor- rowers have also received considerable attention. Using Home Mortgage Disclosure Act (HMDA) data, Avery et al. (2007) document the rapid growth of non-prime lending, 5 It should be noted that Mian and Sufi’s working paper is still in draft form and has been subject to a number of criticisms, including possible omitted variable bias, sample selection bias, the reduced form nature of their model, constraints imposed by their model specification, and conclu- sions that extend beyond their specification and results (see for example Gabriel, 2008). sion and in markets with greater expected house price appreciation. Gramlich (2007) details the rise of subprime lending, its role in increasing home ownership rates among traditionally under-served households, and associated risks. 3. Methodology and model specification We are interested in a simple model for home prices that explicitly allows for changes in loan type intensities to be a leading indicator of future home prices. Starting with a structural model with both supply and demand relationships: QDt ¼ at þ b0;tPt þ b1;t�sLt�s þ b2;tMt þ b3;tKt þ eDt Q St ¼ at þ B0;tPt þ B1;tRt þ B2;tCt þ eSt ð1Þ where: QDt = Quantity of housing demanded in period t Pt = Housing prices at time t Lt�s = Vector of loan type intensity lagged s periods. Mt = Vector of macroeconomic, demographic, and finan- cial controls Kt = Cost of capital QSt = Quantity of housing supplied in period t Rt = Housing market supply regulation Ct = Cost to supply housing bn, Bn = Structural coefficients a, a = Intercepts eDt, eSt = Classical error terms we impose the equilibrium condition QDt = QSt, which implicitly requires market imbalances to be corrected over time by price adjustments. The result is a reduced form equation with prices as our main endogenous variable: Pt ¼ p0 þ p1;t�sLt�s þ p2;tMt þ p3;tKt þ p4;tRt þ p5;tCt þ et ð2Þ where pn are reduced-form impact multipliers.7 6 Though such regulatory guidance was absent earlier, a situation that has been the subject of recent debate between Alan Greenspan, who has been accused by some for primary blame for the situation, and his critics (see Greenspan, Financial Times, April 6, 2008 (http://blogs.ft.com/wolffo- rum/2008/04/alan-greenspan-a-response-to-my-critics/). 7 It is acknowledged that an estimate for user cost should be included in model specification. While we have not done so explicitly, virtually all the elements composing the user cost relationship are included as explanatory variables. We note that appreciation expectations are often the most vexing of user cost elements to proxy for. In our model these are considered explicitly as future house price returns. Our priors are that the predominant effect of increased density of a particular alternative loan type intended to