Read an Excerpt
Housing and the Financial Crisis
By Edward L. Glaeser, Todd Sinai
THE UNIVERSITY OF CHICAGO PRESSCopyright © 2013 National Bureau of Economic Research
All rights reserved.
House Price Moments in Boom-Bust Cycles
The United States experienced a remarkable boom and bust in house prices in the 2000s. According to the Fiserv Case-Shiller ten-city index, house prices grew by 125 percent in real terms from their trough in 1996 to their peak in 2006 and subsequently fell by 38 percent over the next five years. The impacts of this house price cycle have been wide-ranging and severe.
Explaining the causes of this episode of house price growth and decline and its effects on the rest of the banking sector and the real economy is the subject of much current research, some of which is collected in this volume. Potential explanations of the boom and bust in house prices include changing interest rates, subprime lending, irrational exuberance on the part of home buyers, a shift to speculative investment in housing, contagion and fads, and international capital flows.
The goal of this chapter is to describe a set of patterns in house prices among housing markets in the United States and to compile a set of empirical facts that potential explanations of the housing boom and bust should seek to explain. While some of the empirical relationships detailed here have been discussed to varying degrees in prior research, this paper seeks to assemble a broad collection of empirical facts. A unified theory of housing booms and busts would presumably be able to explain the entire set of facts. Of course, it is possible that there is no single mechanism that generated all the economic fluctuations that were experienced, and that a combination of causes needs to be explored.
For this chapter, I consider only house price dynamics. Evaluating the many potential determinants of these housing market dynamics is generally outside the scope of this chapter. However, the role of demand fundamentals, such as rents, income, and employment, is lightly addressed. Other potentially important contributors to housing market dynamics are the purview of other authors in this volume, such as Haughwought et al.'s chapter on the supply side of housing markets; Glaeser, Gottlieb, and Gyourko's chapter on interest rates; and Keys et al.'s chapter on housing finance. In addition, I broadly define housing markets as metropolitan statistical areas (MSAs), intended to correspond to labor market areas that workers are willing to commute amongst. During the boom-bust, there were also important within-MSA house price dynamics. These dynamics are addressed by Genesove and Han in this volume as well as Ferreira and Gyourko (2011) and Bayer, Geissler, and Roberts (2011).
I highlight six stylized facts. First, despite the sizable boom-bust pattern in house prices at the national level, individual housing markets in the United States experienced considerable heterogeneity in the amplitudes of their cycles. The seventy-fifth percentile MSA experienced 111 percent trough-to-peak growth in real house prices in the 1990s and 2000s (using Federal Housing Finance Agency [FHFA] data), whereas the twenty-fifth percentile MSA had only 32 percent trough-to-peak real house price growth.
Second, the boom-bust of the 2000s bears remarkable similarities—as well as some differences—to the boom-bust of the 1980s. We observe two types of MSAs in the data. One set experienced house price cycles in both the 1980s and the 2000s, whereas the other set experienced a boom-bust only in the 2000s. The MSA-level correlation in trough-to-peak real house price growth in the late 1980s and the early 2000s for all MSAs is quite high at 0.45 and would be higher still if those MSAs that did not experience a 1980s cycle at all were excluded.
Third, housing markets also experienced differences in the timing of their cycles. Most MSAs in the 1990s saw their house prices bottom either between 1990 and 1993 or between 1996 and 1997, and house prices generally peaked in 2006 and 2007. In the 1980s, house prices peaked between 1986 and 1990. Potential explanations for housing booms, therefore, need to generate both differences in the amplitude and timing of price changes across MSAs.
Fourth, the largest booms and busts, and their timing, seem to be clustered geographically. The largest amplitude cycles in the boom/bust of the 1990s and 2000s occurred in coastal MSAs and in Florida. These geographic concentrations also had price peaks and troughs that started at similar times, but distinct from the rest of the country.
Fifth, other interesting patterns emerge when one considers annual house price growth, rather than house price changes from trough to peak and back again. In particular, the cross-sectional variance of annual house price changes increases in booms and decreases in busts.
Lastly, these five patterns remain even when house prices are purged of demand fundamentals such as rents, incomes, or employment. Although changes in fundamentals are correlated with changes in house prices, cycles in these fundamentals do not have the same amplitude as price cycles and the time pattern of the growth in fundamentals does not match the timing of the growth in house prices. In fact, controlling for demand fundamentals makes the remaining boom-bust patterns in house prices even starker.
Collectively, these facts limit the possible explanations of the housing boom and bust. The fact that changing demand fundamentals cannot match the boom/ bust pattern of house prices indicates that the house price cycles were due to changes in the price of owning a home rather than changes in the underlying demand for a place to live. However, the factors that are commonly believed to determine asset prices—the cost of credit, changing growth expectations, or time-varying risk premia—often vary over time only at the national level and thus cannot account for the different magnitudes of the booms and busts across metropolitan areas. Some have postulated that a national factor, such as the availability of subprime mortgages, might interact with local characteristics, such as the elasticity of housing supply, to create different-sized booms across MSAs. However, those explanations do not address why the booms would start at different times in different MSAs. Another set of explanations postulate that idiosyncratic MSA-specific conditions, such as differential availability of subprime mortgages, an influx of speculators, or excessive optimism, led to booms in a subset of MSAs. However, the remarkable similarity between the location and size of the booms of the 1980s and 2000s implies that, for this to be explanation, those same conditions must have reoccurred in the same MSAs.
The remainder of this chapter proceeds as follows. I first describe the data used in the chapter and the algorithm for identifying peaks and troughs in each MSA's time series. In section 1.2, I describe the aggregate national patterns in house price dynamics. Section 1.3 makes the point that MSAs can vary considerably from the national average and documents heterogeneity in the amplitude of the housing boom/ bust of the 1990s and 2000s. It also shows that house price booms were typically followed by house price busts. The similarities and differences between the housing boom of the 1980s and the boom of the 2000s is discussed in section 1.4. The next section documents the fact that MSA-level house prices hit their troughs and peaks in different years. Section 1.6 shows that MSAs with similar amplitudes and timing in their housing booms and busts are clustered near each other geographically. Section 1.7 moves from the trough-to-peak house price growth concept to consider annual growth in house prices and the distribution of that growth rate across MSAs. I briefly discuss housing demand fundamentals in section 1.8, finding that house price cycles remain even conditional on cycles in housing fundamentals. Finally, section 1.9 briefly concludes.
The primary source of data is the FHFA's quarterly house price index from data on repeat sales of homes. By comparing repeat transactions on houses that sell multiple times, the index controls for the size or quality of the house to the extent that the house is not renovated. The most significant benefit of the data is that it is available over a very long time (it is reliable as early as 1980) and for a large number of MSAs. However, to be included in the index's sample, the houses need to actually transact—and multiple times at that—and have conforming mortgages securitized by Fannie Mae or Freddie Mac. This sample of houses may not be representative of the overall housing stock and may not reflect the full volatility of the underlying housing market. In addition, the FHFA indexes are normalized within each MSA and thus cannot be used for cross-MSA house price comparisons. The FHFA data contains 344 MSAs with data between 1990 and 2010 and 163 MSAs with data covering 1980 through 2010. I annualize the data by averaging the index over the four quarters in a calendar year and convert the price indexes from nominal to real terms by deflating using the Consumer Price Index (CPI) (all urban consumers).
The FHFA repeat sales index is augmented with rent data from Reis Inc. Reis surveys "class A" apartment buildings, which are typically among the nicest in a given market, and adjusts the rents for concessions, such as months of free rent, to calculate a measure of effective rent. This is the rent concept we use as a proxy for rental values of owner-occupied houses. Because the Reis and FHFA indexes measure two different quantities of housing—the housing stock comprised by the apartment buildings in the Reis data can be quite different than the housing stock in the single-family detached houses in the FHFA data—I will not try to interpret the differences in house price levels versus rents in a given MSA. Instead, in section 1.8, I will compare the growth in FHFA prices to the growth in Reis rents, which merely requires that the growth in rents for apartments tracks the growth in (unobserved) rental value of houses.
As an alternative to using apartment rents as a proxy for housing demand, one could use demand fundamentals. For that reason, I collect data on median per capita income by MSA and employment by MSA from the Bureau of Labor Statistics (BLS). Income is converted into real dollars using the CPI.
Much of the analysis in this chapter concerns the peaks and troughs of housing cycles. The algorithm to determine those troughs and peaks starts by determining the peak of house prices in the 2000s. For each MSA, that peak is defined by first finding all the local maxima—years where the average annual real house price exceeds that of the adjacent years—in the 1999 through 2010 period and then choosing the local maximum with the highest real house price. After that, the algorithm works backward in time: it finds the local minimum—a year where house prices are lower than in the adjacent years—in the period prior to the 2000s peak year that is closest in time to the 2000s peak year and calls it the 1990s trough. The next preceding local maximum is labeled the 1990s peak, and the local minimum that precedes it is called the 1980s trough. Some MSAs do not have cyclical enough house prices for there to be local maxima and minima for all possible peaks and troughs. In those cases, the algorithm defines only those peaks and troughs it can identify. In addition, the trough in house prices in the 2000s is defined as the lowest real house price subsequent to the 2000s peak. However, that so-called trough often occurs in 2010, the last year of the data, and thus may not reflect the actual bottom in house prices. The peak/ trough algorithm is repeated for all the MSA-level economic variables in the data set—house prices, apartment rents, median incomes, and MSA employment—as well as the ratios of house prices to rents, incomes, and employment.
1.2 National Patterns
The history of national average house prices in the United States is by now well-known. According to data from the FHFA, real house prices rose more than 55 percent between the mid-1990s and the end of 2006 and had declined by almost 17 percent by late 2010. A national boom-bust in house prices also was experienced in the 1980s, with real prices rising nearly 15 percent between the mid-1980s and late 1989 and subsequently falling by 8 percent. This data is plotted in figure 1.1. The dashed line, labeled "HPI" (house price index) corresponds to the FHFA national series, deflated by the CPI, and is normalized so that it equals one in 1990.
Another index, Fiserv Case-Shiller, uses a similar repeat-sales methodology but, unlike the FHFA index, is not limited to housing transactions with conforming mortgages and does not exclude sales of foreclosed homes. The real Fiserv Case-Shiller index is plotted in figure 1.1, in the dotted line, for comparison with the FHFA index. The Fiserv Case-Shiller index demonstrates substantially more volatility than the national FHFA index, more than doubling between 1997 and 2007 and subsequently falling by about one-third. However, the Fiserv Case-Shiller index in this figure is a composite of just ten cities, and it turns out that the differences in volatility between the FHFA index and the Case-Shiller index is more a function of the composition of cities that make up the index than of the composition of housing transactions within a city. To show this, we plot (with a dash-dot line) a composite real FHFA index for the same ten cities that are in the Fiserv Case-Shiller ten-city composite index. The two ten-city indexes are quite similar, with the Fiserv Case-Shiller index exhibiting slightly more volatility. This chapter uses the FHFA series because the data covers a longer time span—for many MSAs, it starts (reliably) in 1980 rather than Fiserv Case-Shiller's 1987—and because it covers more metropolitan areas. However, based on figure 1.1, the results should be similar if other house price indexes are used.
1.3 Heterogeneity in Amplitude
The national pattern of house price dynamics masks considerable heterogeneity within the United States. One way to see this cross-MSA variation is to consider the amplitude of the trough-to-peak and peak-to-trough cycles in house prices experienced across various housing markets. Appendix Table A reports these statistics for each MSA in the data. In the 1990s and 2000s, most MSAs did not experience nearly the price growth reflected in the national average. However, a number of MSAs experienced considerably more growth, skewing the distribution of house price growth. The dispersion in the cumulative real price growth from each MSA's trough to peak is graphed in figure 1.2. For instance, the solid line plots a kernel estimate of the distribution of total real price growth for the entire set of MSAs available in the FHFA data, weighting each MSA equally. Most MSAs, 57 percent, experienced real house price growth below the national average growth of 55 percent. Indeed, the mode of the distribution is below 50 percent. However, the right tail of the price growth distribution is skewed, with a number of MSAs experiencing a doubling or more in their real house prices over the period.
The skewness across MSAs in trough-to-peak real house price growth is accentuated when the MSA observations are weighted by their 1990 population of households. This result can be seen in the dashed line in figure 1.2. The peak of the distribution is reduced, with the mass redistributed to the right. This change implies that the highest trough-to-peak house price growth in the 1990s and 2000s was experienced by larger cities, so weighting by the number of households reduces the emphasis at the low-growth portion of the distribution and shifts out the right tail of the distribution.
Most of the skewness in house price growth in the 1990s and 2000s arises from exceeding the prior house price peak of the 1980s and 1990s rather than from recovering to the prior high-price-level after the trough of the 1990s. Evidence can be found in the dashed-dotted line, which is a kernel density estimate of the real house price growth between the real house price peak in the 2000s and the prior peak in house prices, which typically occurred in the late 1980s. That distribution, which is unweighted, looks very similar to the unweighted distribution of trough-to-peak real house price growth, but shifted a bit to the left.
Excerpted from Housing and the Financial Crisis by Edward L. Glaeser, Todd Sinai. Copyright © 2013 National Bureau of Economic Research. Excerpted by permission of THE UNIVERSITY OF CHICAGO PRESS.
All rights reserved. No part of this excerpt may be reproduced or reprinted without permission in writing from the publisher.
Excerpts are provided by Dial-A-Book Inc. solely for the personal use of visitors to this web site.