“The one thing that I don’t see now compared with 2005-2007 (when the housing market turned the last time) is a lot of talk about speculative excess. The financial crisis of 10 years ago was a record setter, and we don’t expect things like that to repeat.” Robert Shiller, Bloomberg Interview, April 5, 2019.
“Credit-financed housing price bubbles have emerged as a particularly dangerous phenomenon.”
Òscar Jordà, Moritz Schularick and Alan M. Taylor, Leveraged Bubbles, 2015.
Following the boom and bust of the 2000s, there is widespread agreement that residential real estate is a key source of vulnerability in advanced and emerging economies alike. Housing accounts for a significant fraction of wealth, especially for people in the middle of the income distribution, who are much less likely to own risky financial assets (see our earlier post).
Furthermore, housing is highly leveraged, creating risks to both homeowners and their lenders. Indeed, while the aggregate declines of U.S. equity wealth in the early 2000s and of housing wealth in the late 2000s were of similar scale in nominal dollars, the former led to the mildest recession in postwar experience while the latter precipitated the deepest and broadest financial crisis since the Great Depression. As experience during the 2007-09 crisis highlights, leveraged losses—like those resulting from real estate exposure—that undermine the capital in the financial system can trigger a panic in short-term funding, threatening the supply of credit to healthy borrowers. Only dramatic policy action can arrest the resulting adverse and destabilizing feedback between the financial system and the economy.
In the United States, real housing prices have rebounded by nearly 40 percent from their 2012 trough. Today, they are only about 10 percent shy of their 2006 peak (see chart). As such, it is natural to ask whether we are once again facing a heightened risk of a crash. Enter “House Prices at Risk” (HaR)—a new worst-case metric created by the IMF to assess the likely scale of a housing price bust conditional on a bad state of the world. Consistent with the IMF’s previous work on “GDP at Risk” (see our earlier post), we view HaR as a valuable addition to the arsenal of risk indicators that allow market professionals and policymakers to monitor financial vulnerability.
U.S. Real House Price Index (1890=100), 1890-Jan 2019
Source: Website of Robert Shiller; based on Figure 3.1 of Irrational Exuberance.
Before getting to the details of Housing at Risk, it is worth taking a brief look at more conventional measures of risk in the residential real estate market. Compared to the Shiller real house price index shown above, other indicators of aggregate U.S. housing price conditions give notably less cause for concern. For example, the OECD’s U.S. housing price-to-rent ratio is currently about one standard deviation above the post-1970 average―far below its level in March 2006, when it stood three standard deviations above the norm. Similarly, the price-to-income ratio—a measure of housing affordability—is currently less than one standard deviation above its post-1970 trend, well below the peak of two standard deviations reached in 2005. Finally, the ratio of housing wealth at market prices to the service flows from that housing—an indicator analogous to a price-earnings ratio—is largely in line with its long-term norm (see following chart).
Ratio of U.S. Housing Wealth at Market Prices to Housing Service Flows, 1959-2018
Note: The horizontal blue line is the mean ratio (11.0), while the shaded area depicts one standard deviation (plus or minus 1.5) on each side of the mean. Sources: Federal Reserve (Financial Accounts of the United States, LM155035015.Q) and FRED.
Housing-related credit indicators also appear relatively benign (see, for example, the 2018 Financial Stability Report of the Federal Reserve). First, as a result of stronger underwriting standards, the expansion of mortgage debt in recent years reflects borrowing by households with prime credit scores. Indeed, the bottom decile of mortgage-eligible FICO scores is now about 40 points higher than in the early 2000s (see page 15 in the monthly chartbook of the Urban Institute’s Housing Finance Policy Center). Second, using home values estimated based on fundamentals, housing leverage appears back in line with norms prior to the pre-crisis lending boom (see Figure 2-12 here). Third, the share of mortgaged homes with negative equity dropped to 4.2 percent at end-2018, down from 25 percent as recently as 2011. Fourth, while cash-out refinancing has generally been rising (particularly as a share of mortgage originations), the volume remains low compared to the boom (see page 10 here). Finally, amid the continued economic expansion, delinquencies and foreclosures have dropped to pre-crisis lows (see page 24 here).
Against these relatively sanguine readings, what is the IMF April 2019 GFSR’s new contribution to assessing the vulnerability of residential real estate valuations? Like Value at Risk (VaR), House Prices at Risk (HaR) is an estimate of the worst possible decline of house prices over a specific horizon. Specifically, HaR measures the 5th percentile of the distribution of house prices one- or three-years ahead.
Focusing on low-probability tail events like HaR is precisely what a risk manager aims to do. Careful analysis of these tail risks can reveal nonlinear relationships that may not affect (and may be concealed by) the rest of the distribution. For example, following a sufficiently large house price decline, collateral quality will deteriorate, making it more difficult for potential borrowers to qualify for a loan. And, as a real estate boom turns to a bust, loan losses can undermine the capitalization of leveraged intermediaries, reducing the supply of credit. In other words, when HaR is elevated, there is a risk that lending markets will collapse.
So, what are the properties of the new HaR measure? The following figure displays the estimated densities for the advanced economies over a one-year-ahead horizon for the 95th (dashed gray line), 50th (dotted gray line), and 5th percentiles (black line) of the housing price distribution. The last of these is the HaR measure for the advanced economies. The latest reading (for the fourth quarter of 2017) of -6.0 percent is actually a tad closer to zero (meaning less negative and less indicative of vulnerability) than the median HaR (-6.1 percent) over the full period. Presumably reflecting the negative impact of increased risk on house prices, the 50th percentile projection declines with increased dispersion between the 95th and 5th percentiles. Most of that changing dispersion reflects the fluctuations in HaR, which is notably more volatile: its standard deviation is 81 percent larger than the 95th percentile projection. That is, similar to the case for growth, a heightened HaR is associated with an increase in the negative skewness of the distribution.
One-year Ahead Density of Advanced Economies House Prices, June 1990-December 2017
Source: IMF GFSR, April 2019. The IMF authors separately supplied the 95th and 50th percentile estimates for the advanced economies.
What leads to a rise in HaR at the one-year horizon in the advanced economies? The GFSR identifies four key factors: Housing price overvaluation (relative to fundamentals), tighter financial conditions, slower economic growth, and faster credit growth. Consistent with the presence of nonlinearities, the impact of these four factors on one-year projections generally is greater for the 5th percentile of the housing price distribution than for the 50th percentile of the distribution.
What about the U.S. HaR? Shown as the red line in the chart above, it also ended 2017 reasonably close to its median since 1990 (-7.8 percent vs. -6.6 percent). That level of vulnerability seems roughly consistent with the modest risk premium in the second chart above (showing the ratio of housing wealth to housing service flow). Thus, in contrast to the run-up to the financial crisis, the latest U.S. HaR does not signal a major threat either domestically or (through a leading relationship with the advanced economy measure) abroad. Interestingly, while it was the dominant driver of HaR in the years before the crisis, overvaluation relative to fundamentals contributes only modestly to the recent U.S. one-year HaR (see Figure 2.8.1 in the GFSR).
More generally, what can the HaR tell us about the future? First, consistent with the view that housing price busts generate wide economic distress, HaR helps forecast GDP at Risk. (For an early discussion of this relationship, see here.) For the advanced economies, a one-percentage-point improvement in HaR improves GDP at Risk by 0.3 percentage point over a 1- to 2-year horizon. Similarly, adding the HaR to standard models for predicting financial crises improves their forecasting accuracy. For the advanced economies, a one-year-ahead HaR of -12 percent (the largest projected decline of -13.5 percent occurred in late 2008) raises the estimate of the probability of a financial crisis by 31 percent at a two-year horizon (see Figure 2.11.3 in the GFSR).
So, both monetary and regulatory policymakers should care about the HaR. And, according to the GFSR, they have tools that can influence it. For example, tightening debt-service-to-income and loan-to-value ratios lowers the HaR in both advanced and emerging economies. In contrast, the risk-diminishing impact of easier monetary policy works only for one to two quarters, and only in advanced economies. These results underscore the stabilization challenge in the United States, which lacks a consistent framework across banks and nonbanks for systemic regulation (see our earlier post).
The bottom line: no one knows when or where the next crisis will come. We will never know when a shock will translate vulnerabilities in the residential real estate market into a price plunge that triggers a financial collapse and a deep economic downturn. All we can do is try to manage risks so that they remain at a socially acceptable level. For this, HaR and GDP at Risk are valuable diagnostic tools.
We commend the authors of the IMF GFSR for their continuing work to improve our risk monitoring toolbox.
Acknowledgments: We are grateful to Tobias Adrian and Nico Valckx of the IMF for providing the data shown in the third chart.