John Cotter, Stuart Gabriel, and Richard Roll
Can Housing Risk be Diversified? A Cautionary Tale from the Housing Boom and Bust
Review of Financial Studies | Volume 28, Issue 3 (Nov 2015), 913–936

Does geographic diversification reduce housing investment risk? To characterize diversification potential, we estimate spatial correlation and integration among 401 US metropolitan housing markets. The 2000s boom brought a marked uptrend in housing market integration. Numerous factors contributed to that trend, including eased residential lending standards and rapid growth in private mortgage securitization. As boom turned to bust, macro factors, including employment and income fundamentals, became important contributors to enhanced integration. Portfolio simulations reveal substantially lower diversification potential and higher risk in the wake of increased market integration. High levels of systemic risk and a reduced importance of local influences made geographic diversification less effective.

Geographic Diversification as a Method of Risk Mitigation

Geographic diversification long has been fundamental in risk mitigation among investors and insurers of housing, mortgages, and mortgage-related derivatives. The housing government- sponsored enterprises (GSEs), the Federal National Mortgage Association, (Fannie Mae) and the Federal Home Loan Mortgage Corporation, (Freddie Mac), now both in government conservatorship, diversified geographically in an effort to reduce the risk of investment in a single asset class (residential mortgages.) Similar logic was employed during the 2000s by prominent Wall Street firms, including Bear Stearns, Merrill Lynch, and Citigroup. More recently, geographic diversification has become central to the investment strategies of multi-family real estate investment trusts (REITs) and single-family housing investment funds.

However, during the late-2000s meltdown, anecdotal evidence suggests that geographic diversification of housing and mortgage investments was less effective. Indeed, diversification has limited power when returns are highly correlated. It appears that the substantial losses incurred by housing and mortgage investors during the downturn period depended in part to unforeseen and unprecedented contemporaneous price declines across geographically-distinct markets.

The effectiveness of geographic diversification as a method of portfolio risk mitigation is also of importance to private and government-backed insurers of residential mortgages. Substantial geographic correlation of credit losses, when coupled with sizable insurer guarantee liabilities and constrained access to credit markets, may render private mortgage insurance less viable. In such circumstances, policymakers may need alternative mechanisms, such as deeply-subordinated government-backed insurance on qualified mortgages, to assure the liquidity and stability of the housing finance system.

Metro Housing Returns were Highly Correlated during the Meltdown Period

Our study commences with an assessment of spatial correlation in housing returns. This includes an examination of contemporaneous and lagged return correlations among 401 metropolitan statistical areas (MSAs) over the 1985-2012 period. High levels of MSA return correlation raise concerns for mortgage or housing investors seeking to diversify risk associated with investment in this asset class. We find large numbers of MSA pairs with contemporaneous and lagged housing return correlations at high levels of statistical significance. Of the roughly 80,000 distinct MSA return pairs, over 53,000 pairs are statistically significant with a mean correlation level of 39 percent. Large MSA housing return correlations appear to be especially pronounced in California. In that state, 98 percent of MSA paired returns are significantly correlated, with a mean correlation level of about 77 percent.

Integration of MSA House Price Returns Trended Up during the Late-2000s

Given evidence of high levels of spatial correlation in returns, we turn to an assessment of the integration of housing markets. Our measure of integration is based on the proportion of a MSAs housing market returns that can be explained by an identical set of national factors (see Pukthuanthong-Le and Roll (JFE 2009)). The level of integration is indicated by the magnitude of R-square, with higher values representing higher levels of integration. Two MSAs are viewed as perfectly integrated if the same national factors fully explain housing market returns in both areas. In that case, the R-square would be 1.0, implying no diversification potential between the MSAs.

1: Housing Return Integration Trends for U.S. MSAs and California MSAs
cotter-gabriel-roll-1
The level of integration is measured by the average R-squares from the multi-factor housing returns model fitted over a sequence of 30-quarter moving windows, 1992:Q2 - 2012:Q4 for 401 US MSAs and for 29 California MSAs.

As shown in Figure 1, there was a pronounced uptrend in US housing market integration over the period of boom and bust. Prior to the 2000s boom, average integration for the 401 MSAs held roughly steady at around 45 percent over the 1992-2004 period. However, starting in late 2004, average integration levels turned up and peaked at 67 percent in 2010. During 2011 and 2012, as the crisis abated, housing return integration trended back down to about 55 percent. Among California MSAs, integration is generally higher but the recent movements are similar, rising from about 63 percent in 2004 to around 90 percent late in the decade. By 2012, integration within California housing returns had declined to just over 80 percent. These recent movements in housing integration are robust to variations in MSA cohorts and estimation methods.

Drivers of Integration Trends

We are able to identify factors associated with the increased integration during the latter half of the 2000s. To do so, we compute the contribution to integration R-square associated with each factor. As shown in Figure 2, innovations in mortgage finance, notably including securitization of non-conforming mortgages and ease of mortgage underwriting, were strongly associated with higher integration during the 2004-2007 boom period. The economic significance associated with those factors also moved up substantially during the boom period. These results coincide with arguments in the literature (see, for example, Favilukis, Ludvigson, and Van Nieuwerburgh (WP 2013), Garriga, Manuelli, and Peralta-Alva (WP 2012), Duca, Muellbauer, and Murphy (WP 2012) and Mian and Sufi (QJE 2009)) that the boom in house prices was fueled in no small measure by widespread easing in mortgage qualification and in the provision of non-conforming secondary market liquidity.

2: Factor Contributions to MSA Housing Return Integration for U.S. MSAs 1992:Q2 through 2012:Q4
cotter-gabriel-roll-2
The level of integration is measured by the R-square from a multi-factor housing returns model estimated for sequential 30-quarter moving windows. The R-square contribution of each factor to the level of integration is plotted.

As boom turned to bust and the influence of mortgage liberalization waned, our results show that other macro factors, including employment and income fundamentals, contributed importantly to the ongoing trending up in housing return integration. Indeed, those factors were responsible for the majority of the increment in U.S. housing market integration during the post-boom period. Similarly, the economic significance of those macro fundamentals was heightened during the crisis period. More recently, and in the wake of the attenuation of the crisis, those same macro factors accounted for much of the downward adjustment in metro return integration.

Does Geographic Diversification Reduce Housing Investment Risk?

Finally, to capture diversification potential for housing investors, we examine an equal-weighted portfolio of our U.S. metropolitan housing cohorts (1992:Q2-2012:Q4). This portfolio exhibits sharply rising levels of housing risk over the 2000s boom and bust. We then measure diversification potential by the difference between average individual MSA volatility and portfolio volatility.

3: Housing Return Integration, Portfolio Risk and Diversification for U.S. MSAs
cotter-gabriel-roll-3
Integration is measured by the R-squares from the multi-factor housing returns model fit for a sequence of 30-quarter moving windows. Risk is measured by the standard deviation (volatility) of housing returns. Diversification is the average volatility of individual MSAs within an equal-weighted portfolio less the portfolio's volatility. The 30-quarter windows end on 1992:Q2 through 2012:Q4 for 401 US MSAs.

Changes in U.S. portfolio risk correlate strongly with the level of housing market integration. During the 2000s housing boom and bust, the simple correlation between the integration R-square and the standard deviation of portfolio returns is 0.96! As shown in Figure 3, during the crisis period, housing portfolio diversification provided only limited benefits in risk diversification. Indeed, the negative correlation between portfolio integration and diversification benefits averaged over –0.82 during the period of housing boom and bust. While integration slowed in 2011 and 2012, our findings still suggest substantial limitations to geographic diversification as a strategy for portfolio risk mitigation. Results suggest that investors and insurers of housing credit risk were rather exposed to the market downturn. Taken together, our findings offer a cautionary tale about geographic diversification as a mechanism to mitigate housing risk.