«Elena Loutskina University of Virginia, Darden School of Business & Philip E. Strahan Boston College, Wharton Financial Institutions Center & NBER ...»
We exploit the idea that the impact of this increased subsidy varies across local housing markets. For example, in a market where all home prices fall below the jumbo-loan cutoff in t-1, home buyers there would receive no incremental benefit from an increase in the cutoff in year t;
all potential homebuyers would already be subsidized. In contrast, in markets with substantial demand near the jumbo-loan threshold, potential homebuyers would benefit greatly when the cutoff rises.
We use two strategies to measure differences across markets in the impact of changes in the jumbo-loan cutoff on housing demand. Detailed data for all mortgage applications to lenders above $50 million in assets are collected annually under the Home Mortgage Disclosure Act (HMDA). The HMDA data include loan size, whether or not a loan was accepted, some information on borrower credit characteristics, and the location of the property down to the Zip code level. Using these data, we estimate the fraction of loan applications in CBSA i and year tthat are above the jumbo cutoff then, but would fall below that cutoff in the subsequent year (year t) as a consequence of the increase in the cutoff between the two years. This ratio captures the percentage of borrowers that would benefit from the change in the cut-off though getting access to more readily available and/or cheaper credit.
the area just below the cut-off (recall Figure 2A). A large fraction of home buyers reduce their borrowing to fall below the cut-off in year t-1, but many would also benefit from an increase in the jumbo-loan cutoff. For example, often home buyers will increase their equity investment in a property to be able to finance their borrowed funds in the subsidized, non-jumbo segment.
Others will split their borrowing into a senior, non-jumbo mortgage (to gain the subsidy), and finance the remainder with a second-lien mortgage from a portfolio lender (i.e. a lender who holds the mortgage) plus equity. Thus, many mortgage applicants below – but not too far below – the jumbo-loan cutoff would also benefit from its increase. To capture this portion of demand, we build an instrument equal to the total fraction of applications within 5% of the jumbo-loan cutoff (on either side) in year t-1, multiplied by the percentage change in the cutoff between years t-1 and t.
elasticity built for 263 CBSAs based on physical impediments to expansion in the housing stock, such as waterways, mountains, and so on.13 Saiz (2010) shows that cities with high supply elasticity have both slower increases in housing prices over time and faster population growth, compared to low-elasticity cities. These results make sense because low barriers to the expansion of housing implies that increased demand from population growth can be accommodated without increasing the cost of housing (e.g. land is not scarce in these areas). In our setting, we expect prices to respond more to the demand shocks associated with changes in the jumbo-loan cutoff in markets with low housing-supply elasticity than in markets with high elasticity.
13 We use the elasticity estimates available online at: http://real.wharton.upenn.edu/~saiz/ and then convert them to the new definitions of CBSA using the zip-code overlap.
market where most of the demand for housing is already subsidized by the GSEs and with very high supply elasticity (e.g. Wichita, where supply elasticity equals 5.5 and only 0.5% of total mortgage applications lie within 5 percentage points of the jumbo-loan cutoff), versus a market with a large mass of demand near the jumbo-loan cutoff and with low supply elasticity (e.g. Los Angeles, where supply elasticity equals 0.63 and about 5.4% of total mortgage applications lie within 5 percentage points of the jumbo-loan cutoff). An increase in the GSE jumbo-loan cutoff shifts housing demand only slightly in Wichita but substantially in Los Angeles. Because supply responds elastically in Wichita, prices barely rise. In LA, however, prices rise sharply, both because demand shifts further from the increased subsidy and because supply responds very little. Thus we trace a shock in a supply of funding to the housing price changes accounting for both CBSA-specific demand shifts and the CBSA-specific supply conditions.
The first-stage model then takes the following form:
where Share-New-NJi,t-1 equals the fraction of applications in CBSA i and year t-1 that will fall below the jumbo-loan cutoff next year (year t); Share-Near-NJi,t-1 equals the share of applications within +/- 5% of the cutoff in year t-1 times the percentage change in the cutoff between t and tWe expect housing prices to grow fastest in markets with a large mass of demand that would benefit from an increase in the jumbo cutoff; thus, we expect: β1HP 0, and β3HP 0. Since house prices should react less if supply is elastic, we expect the interaction terms to offset, meaning β2HP 0, β4HP 0. We estimate Equation (5) with year and CBSA fixed effects, and we
elasticity measure, which is constant over time, is absorbed by the CBSA fixed effects.) Results Table 4 reports summary statistics for our instruments, for housing price growth and for personal income, employment and GDP growth during the 1994-2006 period. We obtain the CBSA-year level data on employment (and employment by segment) from the Bureau of Labor Statistics; the personal income data from the Bureau of Economic Analysis; and the local geography GDP from Moody’s Analytics.14 The analysis begins in 1994 because the financial integration data, based on deposits, become available starting in 1994, and because HMDA data become available only in 1992. We end the analysis in 2006 for two reasons. First, we do not want our estimates to be driven by the Financial Crisis and the ensuing Great Recession. Second, our identification strategy relies on the consistent and mechanical increase in the jumbo-loan cutoff. This cutoff was raised aggressively and in response to political pressure during the Financial Crisis, and has subsequently remained fixed despite falling housing prices in an effort to bolster prices and sustain mortgage credit. The instrumental variables thus lose power after 2006, as they only generate an expansion in housing demand when the cutoff increases.
Table 5 reports the first-stage equation (Eq. (5)) linking the instruments to house-price appreciation, along with the time and CBSA fixed effects, industry share and banking sector control variables. We report the models first for each instrument separately (columns 1 and 3),
14 The CBSA-year level GDP estimates are also available from Bureau of Economic Analysis (BEA) but only starting in 2001. We cross-reference the Moody’s Analytics data with BEA and find the correlation of 98.7% between two data series.
(columns 2 and 4), and last for all instruments together (column 5, which is the first stage regression used subsequently). All of the sets of instruments are powerful – with significant effects individually and collectively – although Share-Near-NJ is clearly stronger than ShareNew-NJ (compare columns 1 and 3). Moreover, the signs and magnitudes of the coefficients are economically sensible individually. For example, a standard deviation increase in Share-NearNJ leads to an increase in housing price growth of 2.7% (a little more than one-half of a standard deviation – see Table 4). Each instrument is also more positive in markets with low supply elasticity (columns 2 and 4). Sign patterns are difficult to interpret in the final regression, with both instruments and interaction terms, because the instruments are highly correlated (ρ=0.92).
The F-test on all four instruments is 45.29 and passes the test for weak instruments with flying colors (Stock and Yogo, 2005).
Table 6 reports a baseline set of IV estimates linking the exogenous component of housing price appreciation to economic outcomes (Equation 4a). We estimate all models with time and CBSA fixed effects and with time-varying industry share variables, and time-varying measures of banking system characteristics. Table 6 reports a total of eight specifications - with and without the lagged dependent variable, times four different measures of output: personal income growth (columns 1 & 2), total employment growth (columns 3 & 4), the growth of total employment excluding employment in financial firms and construction (columns 5 & 6), and GDP growth (columns 7 & 8). Employment without construction and finance allows us to test whether any effects that we observe spillover beyond segments not directly tied to housing finance.
specifications, ranging from 0.14 to 0.26. An exogenous 1% increase in housing prices (stemming from a credit supply increase) thus causes the local economy to expand by 0.14 to
0.26 percentage points faster than otherwise. The coefficients on total employment growth are smaller than GDP growth, which makes sense because GDP includes all sources of production from local sources (i.e. it includes returns to capital as well as labor).15 Moreover, the coefficient on employment growth without segments directly tied to housing suggests that spillovers from higher collateral values raise output beyond the housing sector. Coefficients on personal income growth tend to be somewhat smaller because some of the variation depends on sources of income not tied specifically to the local area.16 Table 7 reports our last test, where we introduce an interaction between housing price growth and financial integration (Equation 4b). For this model, we add the branching restrictions index and its interaction with the other instruments and model housing price growth, financial integration and their interaction as endogenous variables.17 Identification for the direct effect of financial integration is weak due to the inclusion of the CBSA fixed effects, but we are able to get strong identification for the interaction between housing prices and integration (since the interaction has both cross sectional and time series variation).
15 We have also estimated these models separately for the early (1994-2000) and late (2001-2006) portions of our sample. We find that housing is positively and statistically significantly related to economic outcomes in both samples, with somewhat larger magnitudes in the first half of the sample.
16 We have explored several alternative ways to build instruments to check for model robustness. In one set of models, we only use the interaction between the Saiz elasticity measure with the share near non-jumbo * change in cutoff; these results are close to those reported in Table 6, both statistically and quantitatively. We have also estimated models in which we eliminated the time-variation in the share near non-jumbo by using its average value at the beginning of our sample. These results lead to somewhat larger coefficients on the house-price growth variable that have a higher level of statistical significance than those reported in Table 6.
17 The branching index will also help identify housing growth, as Favara and Imbs (2010) show that housing prices grew faster in states more open to interstate banking due to greater availability of credit.
in financially integrated markets. Across all four specifications, housing price growth and financial integration are jointly significant at better than 1%. Moreover, the interaction term suggests that better integration has an economically important effect on the size of the causal impact of housing prices on economic output. For example, at the mean of the In-CBSA ratio (0.81), a 1% increase in housing prices would generate an increase in GDP growth of 0.15% (0.15 = -0.70+0.81*1.044); in markets one-standard deviation above the mean level of integration (0.81+0.15), the same 1% housing-price shock would lead to an increase of 0.30% (0.30 = -0.70+0.96*1.044). The interaction effect of integration on housing is statistically significant across all four models, with a magnitude that varies from 1.0 to 1.4. Because credit supply can respond more elastically to increases in collateral values when local markets are better integrated, an increase in housing prices generates a larger positive spillover in integrated markets. In these areas, the higher demand for credit can draw financial resources in from other sectors.
The Financial Crisis and subsequent Great Recession of 2007-2011 have emphasized for everyone the importance of a strong housing market to the economy. Housing markets not only increased sharply during the 2000s, but they also became more volatile across local markets. We show that this volatility increase is explained in part by better financial integration. We then demonstrate a causal link from housing to the overall economy, using variation in the impact of credit-supply subsidies from the GSEs to construct an instrument for housing price changes that is unrelated to economic conditions in the local economy. Our estimates suggest that a 1% rise in housing prices increase growth by about 0.25%. This effect is larger in localities that are
Ashcraft, A. B. and T. Schuermann, 2008, “Understanding the securitization of sub-prime mortgage credit,” http://papers.ssrn.com/sol3/papers.cfm?abstract_id=1071189&rec=1&srcabs=1093137.
Adelino, M, A. Schoar and F. Severino, 2011, “Credit Supply and House Price: Evidence from Mortgage Market Segmentation,” mimeo.
Brunnermeier, M. K., 2008, “Deciphering the 2007-08 liquidity and credit crisis,” Journal of Economic Perspectives.
Demyanyk, Y. and C. Ostergaard and B. Sorenson, 2007, “US Banking Deregulation, Small Business and Interstate Insurance of Personal Income, Journal of Finance.
Demyanyk, Y. and O. Van Hemert, 2010, “Understanding the subprime mortgage crisis,” Review of Financial Studies.