What have they been thinking? Homebuyer behavior in hot and cold markets.

Case, Karl E. ; Shiller, Robert J. ; Thompson, Anne K. 等

Comments and Discussion

COMMENT BY

DAVID LAIBSON It's very impressive to have called the housing bubble a few years before it popped. It's even more impressive to have conducted a perfectly timed series of housing surveys that anticipated the bubble. These authors are prophetic.

This paper by Karl Case, Robert Shiller, and Anne Thompson offers key insights into the thinking of participants in the housing market. The survey findings are required reading for anyone trying to explain the extraordinary up-and-down price movements that have whipsawed homeowners, homebuilders, banks, and by extension, most of the world economy.

There is much to praise in the remarkable survey studied in this paper, but I will focus on a survey design problem that leads me to reinterpret a few of its findings. Stated briefly, the 10-year forecasting data need far more trimming than the 10 percent trim (discarding the top and bottom 5 percent of the data) that the authors adopt.

In most survey data a 10 percent trim would be adequate--often more than adequate--to remove outliers. The survey data used here, however, suffer from an unusual bias that is not corrected with a 10 percent trim. Some fraction of the respondents--let's say a quarter for now--appear to be confusing the concepts of annualized returns and total returns. For the 1-year forecast data, these two concepts are the same, so that the bias that I am highlighting does not apply. But for the 10-year forecast data there is an enormous difference between an annualized return (what the survey question asks for) and a total return (what perhaps a quarter or more of the subjects are thinking. (1)

To see how this bias works, consider the following simplified example. Suppose that all respondents believe that housing will appreciate 3 percent per year for the next 10 years. Assume as well that three quarters of the subjects respond correctly to the question about 10-year returns, giving an answer of 3 percent, and one quarter give an answer of 30 percent (the total 10-year return, ignoring compounding to keep things simple). Then the mean response is an "annualized return" of 9.75 percent per year, more than triple the subjects' true belief. In this simple example, the researchers would need to trim 50 percent of the data to unbias the mean (since the bias is not symmetric).

Four related empirical facts about the survey results point to the existence of this bias:

--Some subjects give an answer for the 10-year annualized return forecast that is exactly 10 times the answer they give for the 1-year forecast (see the authors' appendix).

--A substantial fraction of the answers to the 10-year forecast question are so high that they are far more likely to be total returns than annualized returns. For example, my table 1 shows that at least 10 percent of the 2004 survey respondents say (if you take their answers literally) that housing prices will appreciate 50 percent per year for the next 10 years, implying a total price appreciation of 5,670 percent. Note that the authors' 10 percent trim removes only half of these respondents from the calculation of the mean.

--The 10-year annualized forecasts are far more right skewed than the 1-year annualized forecasts (see table 1 and the authors' appendix figures A. 1 and A.2).

--The mean 10-year forecast exceeds the mean 1-year forecast (both using the 10 percent trim) in all study years. In most years this gap is considerable (appendix tables A. 1 a and A. 1b).

It is hard to know what to do about this bias. For some of the reasons discussed above, I believe that the authors' 10 percent trim is not adequate. Indeed, I think that a quarter or more of the data may be corrupted by respondent confusion about annualized versus total returns.

Consider the following suggestive evidence. The 1-year forecast mean is unaffected by trimming. Whatever the trimming parameter used (10, 15, 20, 50, or 100 percent), the 1-year forecast "mean" barely budges (see appendix table A. l a). But for the 10-year forecast mean, each additional bit of trimming lowers the mean a little more. Only at a 50 percent trimming does the pattern of falling means abate, and in some years the trimming has to go to 100 percent before the mean stops falling (see appendix table A. lb).

It is still possible that 10 percent trimming is the right methodological answer. Nothing I have said explicitly rules that out. But the preponderance of evidence suggests that one should be wary of the 10-year forecasts. Researchers who wish to use then should explicitly take up the issue of confusion between annualized and total returns and offer a set of assumptions for modeling and measuring this bias.

It is tempting to think that the median is the conservative path to take. But even the median is not immune from these problems. Suppose that beliefs about annualized returns are uniformly distributed from zero to 6 percent, and again assume that one quarter of the respondents instead report the total return (again simplified to 10 times the annual return). If x is the median, then

0.75 x x/6 + 0.25 x x/60 = 0.5.

x [approximately equal to] 3.9.

Hence, the median is biased upward from the true value of 3 percent to a value of 3.9 percent.

In summary, I am awed by this survey that the authors fielded far in advance of the collapse of the housing market. But I urge my fellow economists who are thinking of using the 10-year forecasts in their own research to do so with caution. They cannot be interpreted literally, whether with a 10 percent or even a 100 percent trim.

(1.) The actual language in the question is, "On average over the next ten years how much do you expect the value of your property to change each year" The underlining appears in the actual survey instrument only in 2012.

Table 1. Distribution of Home Price Forecasts from the 2004
Case, Shiller, and Thompson Homebuyer Survey

 Reported 10-year
 growth forecast

 Reported Implied
Percentile of 1-year growth Percent total growth
respondents forecast (percent) per year (percent)

99 50 125 332,425
95 22 50 5,667
90 20 50 5,667
75 10 20 519
50 8 7 97
25 5 5 63
10 2 2 22
5 1 2 22
1 -4 -5 -40

Source: Author's calculations using data from the Case, Shiller,
and Thompson homebuyers survey.

COMMENT BY

PAUL WILLEN My discussion of this paper by Karl Case, Robert Shiller, and Anne Thompson will attempt to place the paper in the context of our evolving understanding of the causes of the housing and financial crises that started in 2007. To me, the central research question about the crisis is why so many people made so many decisions that turned out so badly ex post. Why did borrowers take out loans they could not afford? Why did lenders lend them the money? Why did so many investors buy securities backed by loans that borrowers could not afford to repay? As Christopher Foote, Kristopher Gerardi, and Willen (2012) argue, two alternative narratives seek to answer this question.

The first, which Foote and coauthors refer to as the "insider/outsider" story, and which has dominated popular accounts and much academic research, is depicted in the top panel of my figure 1. According to this theory, the key to the crisis is the institutional fact that the securitization process placed intermediaries in between subprime borrowers and the ultimate investors. These intermediaries, the theory says, understood that the borrowers were going to lose their homes and that the investors were going to lose their money, but they deceived the borrowers and the investors into thinking that these transactions were going to end well for everyone. Foote and coauthors argue that the evidence, some of which I discuss in more detail below, is not kind to the insider/outsider theory. Among other things, intermediaries not only facilitated transactions but were major investors in them, and indeed, as my table 1 shows, the biggest losses were concentrated among the firms most closely associated with the securitization of subprime mortgages. In other words, the insiders appear to have been among the most deceived parties.

The alternative theory, depicted in the bottom panel of figure 1, is what Foote and coauthors call the "bubble theory." This theory holds that optimistic expectations of home price appreciation explain the bad decisions made by borrowers and investors [see follow-up e-mail of 2/6] that led to the crisis. According to this theory, both borrowers and investors were trying to cash in on the greatest real estate boom in U.S. history. If one believes that home prices are going to keep rising at a rate of 10 percent per year, one can easily rationalize ex ante decisions that appear absurd ex post. For example, taking out a mortgage at a monthly payment that exceeds one's monthly income does not normally make sense, but it does make sense if one expects to be able to sell the house in a year at a 10 percent profit.

The results in this paper fill an important gap in the debate over the bubble story. Before now, direct evidence already existed that optimistic beliefs were central to investors' decisions, but evidence on the role of beliefs on the borrower side was only circumstantial and anecdotal. The paper provides some of the first direct evidence on borrower beliefs.

[FIGURE 1 OMITTED]

On the investor side, the direct evidence is ample. Gerardi and others, in a 2008 Brookings Paper, provide extensive evidence along these lines. They point to a table (reproduced here as table 2) from a report titled "HEL Bond Profile across HPA Scenarios," prepared by analysts at Lehman Brothers and widely circulated in the fall of 2005, which illustrates the logic behind investing in subprime mortgage-backed securities. The purpose of the report was to measure the potential losses on subprime bonds across different home price scenarios. The table shows massive losses in the event of a fall in home prices. In the report's "meltdown" scenario, the Lehman analysts estimated that the deals would suffer losses of 17 percent, which implies, assuming that the average default results in a 50 percent loss to the lender, that about a third of the loans would end in foreclosure. Ex post, the Lehman forecast was remarkably accurate. Home prices fell twice as fast as in the meltdown scenario, at a rate of-10 percent rather than -5 percent, and actual investor losses were about 23 percent.

Table 2 provides strong evidence against the insider/outsider theory and in favor of the bubble theory. Contrary to the insider/outsider story, the table illustrates that investors were not deceived about the quality of the loans in which they were investing. Lenders typically foreclose on a fraction of a percent of mortgages rated prime; thus, the 5 percent losses in the "baseline" scenario imply that these loans were orders of magnitude riskier than the typical prime mortgage. Evidence in support of the bubble theory is the fact that, as table 2 also shows, the Lehman analysts were remarkably optimistic about home prices. Their "base" scenario (to which they assigned a probability of 50 percent) involves home price appreciation of more than twice expected inflation, and their "aggressive" scenario (with a probability of 15 percent) implies that despite having risen by almost 75 percent over the previous 5 years, home prices would rise another 35 percent over the next 3. Put another way, the Lehman analysts assigned three times as high a probability to the aggressive scenario as they did to the meltdown scenario (5 percent)--which was, in fact, a better outcome than what actually happened. Further, credit protection on subprime deals had to exceed at least 15 percent before any of the triple-A-rated securities lost money, so the implied probability of loss was less than 5 percent. In short, the Lehman analysis and similar research at other firms, as documented by Gerardi and others (2008), illustrates how the belief in continued rapidly rising prices led to the ex post bad decisions at the heart of the crisis.

On the borrower side, until now the evidence was not as complete. On the one hand, there was little evidence to support the idea that borrowers had been deceived. Proponents of the insider/outsider theory (for example, Eakes 2007) point to the use of adjustable-rate mortgages (ARMs) as evidence of deception. According to the theory, lenders used mortgages with deceptively low initial payments to lure borrowers into mortgages they ultimately could not afford. But my table 3 shows that ARMs could have played only a small role in causing the crisis, because the vast majority of borrowers defaulted while making payments that were the same or lower than the initial payment on the loan. Table 3 shows that whereas the performance of subprime ARMs was appalling, that of subprime fixed-rate mortgages was almost as bad.

But evidence against a theory is not necessarily evidence in favor of the alternative. If borrowers did not take out the loans because of unrealistically low initial mortgage payments, why did they take out loans they ultimately could not afford? The present paper provides an answer: like investors, they were incredibly optimistic about home prices. Figure 4 of the paper shows that in the key years of the boom--2004 and 2005--homebuyers expected home prices to rise by about 12 percent on an annualized basis.

Putting such beliefs into standard portfolio choice models yields exactly the sort of behavior that was observed. Equation 29 of Merton (1969), for example, shows that the share of wealth invested in the risky asset equals

(1) [omega]* 1/RRA x E([??])-[R.sub.f]/[sigma]([??]) (1)

where R is the return on the risky asset, [R.sub.f] is the risk-free rate, and RRA is the coefficient of relative risk aversion. Case and coauthors document that for housing, E([??]) rose dramatically, and in particular, came to be highly elevated relative to the relevant [R.sub.f], the mortgage rate.

To see how this would lead to an explosion in leverage, remember that [omega]* is the share of total wealth defined as the sum of financial and human wealth. In other words,

[omega]* = investment in risky assets = investment in risky assets total wealth financial wealth + human wealth.

As E([??]) - [R.sub.f] goes up, investment in the risky asset grows relative to financial wealth, and at some point the only way for the household to hit its target investment level is to start borrowing against its human wealth. In other words, changing beliefs about home price appreciation are all that is need to explain the rise in mortgage debt.

The bubble explanation, of course, leaves many questions unanswered. Why did such optimistic beliefs emerge? How do households form such beliefs? To understand the crisis, researchers must focus to some extent on the measurement and modeling of beliefs and on the typical work of economists, the measurement and modeling of behavior. To some degree the shift to research on beliefs has already happened: arguably, it has been going on ever since the rational expectations revolution in macroeconomics in the 1970s. One strand of the literature that has emerged in the wake of the crisis describes the phenomenon in the housing market in the boom years as "distorted beliefs. (1)

Let me conclude by suggesting some directions for future research. Case, Shiller, and Thompson's survey asks households about the first moment of their expectations, but the second moment is also important to most home purchase and default decisions. One can think of a mortgaged home as a call option: the borrower "sells" the house to the lender and gets an option to buy it back by repaying the outstanding balance of the loan. As any finance student can explain, the volatility of the underlying asset is the key to the valuation of this option. If one believed that home prices were equally like to rise by 20 percent or fall by 20 percent in 2005, then a mortgage allowed one to enjoy the upside and to default on the downside-which, indeed, millions of Americans did.

REFERENCES FOR THE WILLEN COMMENT

Cheng, Ing-Haw, Sahil Raina, and Wei Xiong. 2012. "Wall Street and the Housing Bubble: Bad Incentives, Bad Models, or Bad Luck?" Working paper. Princeton University.

Eakes, Martin. 2007. "Evolution of an Economic Crisis? The Subprime Lending Disaster and the Threat to the Broader Economy." Testimony before the Joint Economic Committee, September 19. www.responsiblelending.org/mortgagelending/policy-legislation/congress/senate- sept-07-final.pdf.

Foote, Christopher, Kristopher Gerardi, and Paul Willen. 2012. "Why Did So Many People Make So Many Ex Post Bad Decisions? The Causes of the Foreclosure Crisis." In Rethinking the Financial Crisis, edited by Alan S. Blinder, Andrew W. Lo, and Robert M. Solow. New York: Russell Sage and Century Foundations. Fuster, Andreas, David Laibson, and Brock Mendel. 2010. "Natural Expectations and Macroeconomic Fluctuations." Journal of Economic Perspectives 24, no. 4: 67-84.

Geanakoplos, John. 2009. "The Leverage Cycle." In NBER Macroeconomics Annual 2009, edited by Daron Acemoglu, Kenneth Rogoff, and Michael Woodford. University of Chicago Press.

Gerardi, Kristopher, Andreas Lehnert, Shane M. Sherlund, and Paul S. Willen. 2008. "MakingSense of the Subprime Crisis." BPEA, no. 2: 69-145.

Merton, Robert C. 1969. "Lifetime Portfolio Selection under Uncertainty: The Continuous-Time Case." Review of Economics and Statistics 51, no. 3: 247-57.

Simsek, Alp. 2012. "Belief Disagreements and Collateral Constraints." Working paper. Harvard University. www.economics.harvard.edu/faculty/simsek/files/ simsekBeliefDisagreementsCollateralConstraints7_ EMArevision.pdf.

(1). See, for example, Gennaioli and Shleifer (2010), S imsek (2012), Fuster, Laibson, and Mendel (2010), Geanakoplos (2009), and Cheng, Raina, and Xiong (2012).

Table 1. Mortgage-Related Losses to Financial Institutions in the
Subprime Crisis, as of June 18, 2008 (a)

Billions of dollars

Rank Institution Loss

1 Citigroup 42.9
2 UBS 38.2
3 Merrill Lynch 37.1
4 HSBC 19.5
5 IKB Deutsche 15.9
6 Royal Bank of Scotland 15.2
7 Bank of America 15.1
8 Morgan Stanley 14.1
9 JPMorgan Chase 9.8
10 Credit Suisse 9.6

Rank Institution Loss

11 Washington Mutual 9.1
12 Credit Agricole 8.3
13 Lehman Brothers 8.2
14 Deutsche Bank 7.6
15 Wachovia 7.0
16 HBOS 7.0
17 Bayerische Landesbank 6.7
18 Fortis 6.6
19 Canadian Imperial (CIBC) 6.5
20 Barclays 6.3

Source: Bloomberg, "Subprime Losses Top $396 Billion on Brokers'
Writedowns," June 18, 2008.
www.bloomberg.com/apps/news?pid=newsarchive&sid=aSGaivCMZu_M.

a. The date chosen precedes the Lehman bankrupcty to avoid
contamination from the wider financial crisis.

Table 2. Lehman Brothers Conditional Forecasts of Losses on
Subprime Investments, August 2005

Scenario Assigned Cumulative
name Description (a) probability loss

"Aggressive" 11 percent annual HPA over 15% 1.4%
 the life of the pool

 8 percent annual HPA over 15% 3.2%
 the life of the pool

"Base" HPA slows to 5 percent 50% 5.6%
 annual by end of 2005

"Pessimistic" Zero HPA for next 3 years, 15% 11.1%
 5 percent annual thereafter

"Meltdown" -5 percent annual for next 5% 17.1%
 3 years, 5 percent annual
 thereafter

Source: "HEL Bond Profile across HPA Scenarios," from Lehman
Brothers, "U.S. ABS Weekly Outlook," August 15, 2005.

a. HPA=home price appreciation.

Table 3. Payment Changes and Default, 2007-10

 2007 2008 2009 2010 All

FRM share (percent) 38 48 62 74 59
Payment changes (percent of loans
 eventually foreclosed upon),
 Interest rate reset 18 20 18 11 17
 Payment increase 12 17 11 9 12
 Payment reduction 0 0 4 8 4
 No change since origination 88 82 85 83 84
Privatelabel (percent of loans) 68 54 37 23 41
No. of observations (thousands) 374 641 874 756 2,646

Source: Foote and others (2012).

a. Only those changes that occurred before delinquency spell that
led to foreclosure are counted.

Table 4. Relative Performance of Subprime Adjustable-Rate and
Fixed-Rate Mortgages, 2005-07

 All subprime mortgages Subprime fired-rate mortgages

 Percent of
 Originations Percent Originations all subprime
Year (thousands) defaulting (thousands) mortgages

2005 529 41.9 198 37.3
2006 504 55.9 258 51.2
2007 246 55.9 208 84.5
Total 1,278 50.1 663 51.9

 Subprime Subprime adjustable-rate (2/28)
 fired-rate mortgages (a)
 mortgages
 Percent of
 Percent Originations all subprime Percent
Year defaulting (thousands) mortgages defaulting

2005 37.1 332 62.7 44.8
2006 50.7 246 48.8 61.4
2007 53.8 38 15.5 66.8
Total 47.6 615 48.1 52.8

Source: Foote and others (2012), using data from Lender
Processing Services, Inc.

a. In a 2/28 mortgage, the interest rate remains constant for the
first 2 years and then adjusts periodically over the remaining
28 years of the mortgage.

GENERAL DISCUSSION Karen Pence observed that in the typical residential real estate transaction, many people--the realtor, the mortgage lender, the title insurer, and others--have a financial incentive in seeing the transaction go through, but no one profits directly if it does not. As a result, potential buyers may hear more about the benefits of homeownership than the risks. It therefore seemed to Pence unsurprising that unrealistic beliefs about future home price appreciation could persist.

To Frederic Mishkin the authors' own charts suggested that the problem of respondents anchoring on the expected 1-year price appreciation rate when answering the question about the 10-year rate was even greater than David Laibson feared. He himself had seen signs of such anchoring in responses about long-term inflation expectations in the University of Michigan survey and therefore tended to ignore those data. Mishkin acknowledged that the authors' expectations data, especially the short-term data, could still be of some value to the extent that people acted on the basis of their irrational expectations, but he felt the measurement problem needed further study. Justin Wolfers pointed out that in the latest Michigan survey, many of the answers even about short-term inflation made no sense either: 10 percent of respondents thought that inflation would exceed 10 percent in the next year, and some believed it would be 40 percent or more. That indicated either that many respondents are profoundly economically illiterate, or that they were not giving serious answers.

Bradford DeLong wished that the authors' survey had focused on the expectations of the marginal homebuyer--the household right on the edge between buying and not buying--rather than on those of the average buyer. One thing he took away from the study, however, was that even today, in the wake of a major housing bust, many people remain overoptimistic about home prices, which suggested to him that there will always be homebuyers who are overoptimistic about home prices. That, together with Pence's point about the bias of third parties' incentives toward completing the transaction, argued in DeLong's view for greater attention to credit standards as an essential governing mechanism.

Ricardo Reis proposed that the authors address the problem of irrational price expectations either by trimming the outliers further or by focusing on the median rather than on the trimmed mean responses. Whatever measure of central tendency the authors chose, he thought they should discuss the reasons for that choice in the paper. He himself had looked at a number of surveys of expectations of various kinds--of stock prices, for example, as well as inflation--and found that in most cases the median response of the general public is in line with the median response of professional analysts. The difference is in the dispersion: whereas the analysts' answers are usually tightly distributed, typically at least 25 to 30 percent of the lay respondents give answers that are off the chart. Therefore, simply trimming 10 percent from the tails of the distribution and calculating the trimmed mean will still give biased results.

Replying to Paul Willen on the role of incorrect beliefs in forming bubbles, Reis observed that an important lesson from attempts to model bubble formation is that buyers with optimistic beliefs need to be matched with sellers with more pessimistic beliefs; the model needs to capture not so much those beliefs that are extreme as the full dispersion of beliefs. He therefore urged the authors to include in the paper an analysis of some measure of disagreement among respondents.

William Brainard thought it important to distinguish between new and existing homes when discussing home prices and price expectations, both because the factors that affect the prices of the two types of homes differ, and because the two types differ in the way changes in their prices affect output. New home prices primarily reflect construction costs and the price of undeveloped land; changes in new home prices, in turn, affect housing construction, a direct contributor to GDP and source of jobs, but have minor effects on household wealth. Changes in existing home prices, in contrast, have major effects on household wealth, and therefore on consumption, but in cities like Boston and San Francisco these price changes primarily reflect changes in the shadow price of land rather than construction costs. They have only indirect effects on incentives to build new homes on undeveloped land at the periphery. Since the factors that would be expected to affect home prices differ, Brainard thought it would be interesting to report separately the results for the expectational behavior of buyers of the two types of homes. He also thought it would be interesting to compare the expectations of buyers and sellers in the resale market with those of buyers and speculative builders in the new home market.

Robert Hall was impressed with the case Paul Willen had made that the finance industry during the housing boom was not recklessly throwing money away, as is often alleged. It seemed that the underwriting of mortgage-backed securities during that period had been careful and rational after all, using probability distributions just as prescribed by state-ofthe-art decision theory. The problem was that when the analysts wrote down the distribution, they assigned only a tiny probability to a price decline, and then the model's undoing was simply an unbelievably unlucky draw from the that distribution: the sudden collapse in home prices across the United States. Under no principle of economics, Hall concluded, were the underwriters following the wrong decisionmaking procedure. From a longer historical perspective, it seemed likely that they understated the probability of a large price decline.

On the question of trimmed means versus medians, Hall suggested instead using quantile regressions so as to capture the characteristics of the whole distribution, as he had done in a forthcoming paper in the American Economic Review. He thought the most interesting approach here would be to analyze the joint distribution of the 1-year and 10-year price expectations.

Steven Davis saw the paper as an important effort at the frontier of developing a more extensive collection of data about the beliefs of economic agents. The evolution of beliefs during the housing boom and subsequent crisis was central to the theory of the crisis that Willen had sketched out, and Reis had noted that a number of interesting theories stress the dispersion of beliefs in explaining what happened. Yet, Davis observed, the federal statistical agencies devote almost no resources to eliciting information about beliefs. Data from nongovernmental sources on business and consumer sentiment are available but often difficult to interpret. The profession, in Davis's view, was a long way from having data sets equal to the task of evaluating the theories that Reis had described, or of understanding the beliefs of DeLong's marginal buyer. A particular need was for data matching the beliefs of individuals with their demographic and other objective characteristics. The federal government or the Federal Reserve, Davis thought, was surely in by far the best position to take on these tasks.

Nellie Liang saw the paper as supporting the idea that the formation of expectations matters for understanding economic behavior, and she suggested that this was true not just of households buying homes but also of sophisticated participants in equity markets. During the dot-com bubble, for example, many equity analysts held views about the future long-term growth of technology firms that seemed to mirror the expectations reported in the authors' survey. That indicated to her that it might be worth investigating whether there was some sort of commonality in how these expectations are formed. Unfortunately, information about these expectations is difficult to obtain in real time. Liang also was interested in knowing to what extent expectations correlated with lending standards, because the most costly bubbles tend to be those whose aftermath involves the widespread unwinding of leverage.

Christopher Carroll agreed that there is a scarcity of data of the kind that would yield insight into how beliefs are formed, and he surmised that macroeconomists have tended to shy away from developing models of belief formation in part because of this lack of data. He pointed out, however, that some data useful for that purpose were available, in the Michigan survey and elsewhere, yet were not being fully exploited. He thought it would be worth investigating what these data could say about expectations and their formation, so that economists doing empirical research are not limited to the unsatisfying choice of simply accepting or rejecting the standard rational expectations model. Indeed, Carroll saw the study of belief formation as a potential growth area for macroeconomics.

Laibson's point about long-term reversion to the mean in asset price series reminded Carroll of a 1996 paper by Shiller and John Campbell about mean reversion in stock price-earnings ratios--that was the paper that seemed to have inspired Alan Greenspan's speech invoking "irrational exuberance" a few weeks later. Carroll had recently returned to Shiller and Campbell's model to see what level of stock prices it would have predicted for October 2008, shortly after the Lehman Brothers meltdown: he found that the 1996 model predicted prices on that date almost perfectly, after being wrong for the whole intervening period. That, to Carroll, suggested that this was the kind of model that one ought to be investigating for this set of issues.

Robert Pozen offered further evidence that a large part of the general public lacks an understanding of basic financial concepts: surveys by the investment community have found, he said, that half of all individual investors think that bond prices rise when interest rates rise. Given the importance of mortgage interest rates in home buyers' decisions, Pozen suggested that the authors add some questions to their survey aimed at determining the respondents' understanding of the relationship between prices and interest rates.

David Laibson seconded Hall's call for studying whole distributions rather than some single measure of central tendency, but with a caveat in this case. The authors' data on long-term expectations consisted not of a single outcome--the answer to a particular question--with some noise. Rather, two-thirds of the data were answers to one question, and one-third the answer to a different question, and any model used to analyze such a data set needed to take that into account.

Benjamin Kay suggested that the noise in the data might be reduced by adding questions about the opinions of the respondents' friends and family members, as Wolfers had done in some recent work. In the face of a declining response rate, such a strategy might boost the information that could be elicited for a given sample size.

Gerald Cohen wondered whether part of the problem in analyzing people's beliefs about the housing market was that a home is both an investment and a consumption good. Changes over time in the degree to which individual respondents purchase homes primarily for investment or primarily for consumption, he surmised, could result in substantial shifts in observed expectations about price appreciation. Cohen noted that the various official inflation measures themselves differed on this score, with the PCE deflator treating housing more as consumption whereas the CPI, with its concept of owner's equivalent rent, treated it more as investment.

Responding to the discussion, Karl Case began by noting that although the responses to the question on expected 10-year home price appreciation seemed problematic, they were consistent across the four cities surveyed. Moreover, the expectations of the California respondents of a 10-year average appreciation exceeding 10 percent a year were not completely unreasonable given that state's history of home prices: prices there had risen at an annual average rate of 10.6 percent from 1996 to 2006, and at a similar rate in the earlier boom from 1975 to 1985. Prices had risen by more than 10 percent a year in several other cities around the nation as well. Of course, real returns in the earlier period were much smaller because of the high inflation of the time, but Case stressed that whereas economists focus on real prices, participants in the real estate market tend to focus on nominal values. This, too, was not unreasonable, because mortgages are denominated in nominal terms.

Case went on to note that people often forget just how hot the housing market was between 2002 and 2006. In Orange County in 2003, for example, homebuyers expected prices to rise by a (trimmed) average of 11.5 percent annually. And in 2000 and 2001 prices had actually rose by 11 percent and 10 percent, respectively. But then things changed: actual price increases for Orange County were 17 percent in 2002, 20 percent in 2003, 28 percent in 2004, and 21 percent in 2005. Long-term expectations as measured in the survey peaked in 2004 at 17 percent, which seemed unreasonable to Case and his colleagues, but the market at the time was witnessing even faster growth.

Finally, Case remarked that little research has been done into the interaction of the home ownership and home rental markets on the supply side. It seemed to him that prices are determined in very different ways in the two markets: whereas rental housing is priced on the basis of capitalization rates and revenue flows, home purchase prices are largely driven by buyers trying to figure out how large a mortgage they can obtain or afford.

Robert Shiller pointed out that whatever problems lay in the 10-year price expectations data, the changes observed in those expectations over time should still be instructive, because the questions remained the same every year. In fact, the pattern for the 10-year expectations differed over time from that for the 1-year expectations, the former being significantly more sluggish. Shiller cited his book Animal Spirits, co-written with George Akerl of, as providing evidence that although people's expectations about the distant future matter enormously, those expectations are fuzzy and are not captured well by any existing forecasting model. He acknowledged that the survey questions might have been better worded in retrospect, and additional questions such as about the opinions of friends and family, as suggested by Kay, might be worth adding, but in principle the attempt to elicit 10-year price expectations was worth undertaking. He and his coauthors had considered changing the wording of some questions before the most recent survey but felt that doing so would cloud the interpretation of the results.

Replying to DeLong's suggestion to focus on the marginal home buyer, Shiller thought that by surveying only households who had recently purchased a home, he and his coauthors were coming as close to the marginal buyer as one reasonably could.

Finally, Shiller was intrigued by Hall's proposal that the authors use quantile regressions in analyzing their data set, but the balance of the discussion led him to think that their first approach in revising the paper should be to further trim the tails of the distributions. He pointed out that Reis's suggestion to use the median instead of the mean was equivalent to trimming 50 percent of the distribution from each tail. The problem with using the median in this survey, however, was that because many people give answers in round numbers, the median tends to stick at one such number from year to year and any movement in the central tendency gets lost.

Anne Thompson provided some detail on how she had trimmed the data. In one version she simply trimmed the top and bottom 10 percent of responses, which resulted in the mean response for the 1-year expectation declining 0.2 percentage point and the annualized 10-year expectation falling by about 1.0 percentage point, but the overall pattern was otherwise the same. Alternatively, she looked at the responses one by one to identify extreme outliers and obvious misreadings of the question. Specifically, she discarded any response that exceeded 40 percent, and where the 10-year expectation was a 10-fold multiple of the 1-year, she changed the former to equal the latter. This procedure lowered the mean response by about 0.9 percentage point but again did not alter the pattern.

KARL E. CASE

Wellesley College

ROBERT J. SHILLER

Yale University

ANNE K. THOMPSON

McGraw-Hill Construction

Table 1. Response Rates in the Homebuyers Survey, 1988-2012

 Surveys Response rate
Year returned (percent)

1988 886 43.6
2003 705 35.3
2004 456 22.8
2005 441 22.1
2006 271 13.6
2007 300 15.0
2008 545 27.3
2009 370 18.5
2010 375 18.8
2011 319 16.0
2012 328 16.4
All years 4,996 22.7

Source: Authors' calculations from homebuyers survey data.

Table 2. Correlations between Actual and Perceived Home Price Trends,
by Survey Location, 2003-12a
Correlation coefficients

 Actual price trend

Perceived Alameda Middlesex Milwaukee Orange
price trend County County County County All

ng rapidly 0.67 0.86 0.89 0.81 0.76

Falling -0.88 -0.65 -0.80 -0.71 -0.76
rapidly

Source: Authors' calculations from homebuyers survey data.

(a.) Results are simple correlations between the percentage of
respondents in the indicated location who gave the indicated response
and the actual annual percentage change in the S&P/Case-Shiner Home
Price Index for that metropolitan area (measured from the second
quarter in the year before the survey to the second quarter of the
survey year; see figure 1 for the wording of the survey question).
Data for each location are pooled across all 10 survey years.

Table 3. Short-and Long-Term Home Price Expectations, by Survey
Location and Year, 2003-12

Mean response (percent) (a)

 Survey location

Survey Alameda Middlesex Milwaukee Orange
year County County County County

 "How much of a change do you expect there to
 be in the value of your home over the
 next 12 months?" (b)

2003 7.6 4.4 5.5 9.4
2004 9.3 7.6 6.4 13.1
2005 9.6 6.3 6.6 8.7
2006 7.4 1.9 5.9 6.0
2007 4.9 2.9 6.1 -0.1
2008 -1.6 -0.7 2.4 -2.6
2009 2.4 2.0 1.5 0.7
2010 4.4 2.2 3.7 3.8
2011 2.3 2.3 1.7 0.3
2012 4.4 2.3 2.3 3.6

 "On average over the next ten years how much
 do you expect the value of your property to
 change each year?" (c)

2003 12.3 8.9 7.1 11.5
2004 14.1 10.6 10.4 17.4
2005 11.5 8.3 11.9 15.2
2006 9.4 7.5 9.9 9.5
2007 10.7 5.3 8.1 12.2
2008 7.9 6.4 7.2 9.4
2009 8.5 6.2 8.2 6.9
2010 9.8 5.0 7.3 5.7
2011 7.6 4.1 4.7 7.1
2012 5.4 3.1 3.1 5.0

Source: Authors' surveys.
(a.) Means are 10 percent trimmed means; that is, we dropped
the highest and lowest 5 percent of responses before
calculating the mean.
(b.) Survey question 6.
(c.) Survey question 7; in the 2012 survey only, the words
"On average" and "each year" were underlined.

Table 4. Expected versus Actual Short-all d Long-Term Home Price
Expectations in Orange and Middlesex Counties

 Expected annual
 price increase
 (percent) Actual
 1-year price
Survey Next 10 increase Implied value of
location Next years (percent) a home worth
and year year Orange County $100,000 in 1996

 Orange Country

1996 n.a. (a) n.a. $100,000
1997 n.a. n.a. 2.4 102,440
1998 n.a. n.a. 12.8 115,594
1999 n.a. n.a. 11.5 128,902
2000 n.a. n.a. 10.2 142,074
2001 n.a. n.a. 9.8 155,986
2002 n.a. n.a. 11.8 174,318
2003 9.4 11.5 18.2 206,043
2004 13.1 17.4 31.1 270,205
2005 8.7 15.2 18.5 320,167
2006 6.0 9.5 14.9 367,883
2007 -0.1 12.2 -3.3 355,662
2008 -2.6 9.4 -24.3 269,082
2009 0.7 6.9 -19.6 216,212
2010 3.8 5.7 8.9 235,450
2011 0.3 7.1 -2.9 228,595
2012 3.6 5.0 -2.1 223,901

 Middlesex County

1996 n.a. n.a. $100,000
1997 n.a. n.a. 6.0 105,962
1998 n.a. n.a. 8.8 115,298
1999 n.a. n.a. 12.3 129,497
2000 n.a. n.a. 14.1 147,810
2001 n.a. n.a. 16.4 172,090
2002 n.a. n.a. 10.8 190,655
2003 4.4 8.9 11.3 212,161
2004 7.6 10.6 9.6 232,443
2005 6.3 8.3 8.4 252,031
2006 1.9 7.5 -1.4 248,583
2007 2.9 5.3 -4.2 238,218
2008 -0.7 6.4 -6.0 224,001
2009 2.0 6.2 -6.9 208,446
2010 2.2 5.0 4.3 217,499
2011 2.3 4.1 -3.2 210,551
2012 2.3 3.1 0.0 210,476

Sources: S&P1Case-Shiner, Fiserv, Inc., and authors'
calculations from homebuyers survey data.
a. n.a. = not available.

Table 5. Regressions Testing for Rational Expectations of the
One-Year Change in Home Prices (a)

 Survey location

 Alameda Middlesex Milwaukee
 County County County

 Using S&P/Case-Shiller Indexes, 2003-12

Constant -12.79 -4.75 -5.67
 (8.84) (2.85) (4.52)

Own-city expected 2.57 1.50 1.43
 12-month price change (b) (1.42) (0.71) (0.94)

No. of observations 9 9 9
[R.sup.2] 0.32 0.39 0.25

 Using FHFA home price data

Constant -8.60 -4.82 -6.96
 (4.12) (2.50) (3.45)

Own-city expected 2.03 1.73 1.86
 12-month price change (b) (0.66) (0.62) (0.72)

No. of observations 9 9 9

[R.sup.2] 0.57 0.52 0.49

 Survey location

 Orange
 County All

Constant -9.48 -9.13
 (5.16) (2.52)

Own-city expected 2.71 2.34
 12-month price change (b) (0.78) (0.46)

No. of observations 9 36
[R.sup.2] 0.63 0.43

Constant -8.75 -8.11
 (2.88) (1.48)

Own-city expected 2.81 2.32
 12-month price change (b) (0.44) (0.27)

No. of observations 9 36

[R.sup.2] 0.86 0.69

Sources: Authors' regressions using data from S&P/Case-Shiller,
Fiserv, Inc., the Federal Housing Finance Agency, and the homebuyers
survey.

(a) Each column in each panel reports results of a single regression.
The dependent variable is the actual percentage home price change in
the indicated city from the second quarter of the year to the second
quarter of the following year. Standard errors are in parentheses.

(b) Trimmed mean of responses to question 6 of the homebuyers survey.

Table 6. Regression Testing for Rational Expectations of the
One-Year Change in Home Prices with Additional
Information Variablesa

 Regression
 coefficient

Constant -12.91
 (3.82)

Own-city expected 12-month price 3.28
change (percent) (b) (0.93)

Lagged own-city actual 12-month -0.25
price change (percent) (0.29)

Lagged national (10-city) actual -0.03
12-month price change (percent) (0.26)

No. of observations 36

[R.sup.2] 0.48

Sources: Authors' regression using data from S&P/Case-Shiller,
Fiserv, Inc., and the homebuyers survey.

(a) The dependent variable is the percentage change in actual
home prices in the respondent's metro area from the second quarter
of the survey year to the second quarter of the next year.
Data are pooled across all locations and survey years.
Standard errors are in parentheses.

(b) Trimmed mean of responses to question 6 of the homebuyers survey.

Table 7. Regressions of the Expected One-Year Change in
Home Prices on lagged Actual Price Changes (a)

 Survey location

 Alameda Middlesex Milwaukee
 County County County

Constant 4.87 2.79 3.76
 (0.64) (0.49) (0.30)

Lagged own-city 0.18 0.29 0.30
 actual 12-month price (0.04) (0.07) (0.05)
 change (percent)

No. of observations 10 10 10
[R.sup.2] 0.70 0.64 0.82

 Survey location

 Orange
 County All (b)

Constant 3.25 3.72
 (0.61) (0.28)

Lagged own-city 0.26 0.23
 actual 12-month price (0.04) (0.02)
 change (percent)

No. of observations 10 40
[R.sup.2] 0.87 0.73

Sources: Authors' regressions using data from S&P/Case-Shiller,
Fiserv, Inc., and the homebuyers survey.

(a.) Each column reports results of a single regression.
The dependent variable is the trimmed mean of the expected 1-year
change in home prices in the indicated location. Standard errors
are in parentheses.
(b.) Data are pooled across all locations and survey years.

Figure 1. S&P/Case-Shiller Home Price Indexes for the Four Survey
Locations, 1987-2012 (a)

Alameda County, Calif. (San Francisco metro area)

Index, Jan. 2000 = 100

 1988 2004 2006 2008 2011 2012

Perceptions: Which best describes the area home price trend?

Rising rapidly 84% 69% 21% 1% 0% 5%
Rising slowly 13 31% 42 4 15 50
Not changing 3 0 17 4 28 24
Falling slowly 1 0 21 57 47 19
Falling rapidly 0 0 0 33 10 2

Expectations: It's a good time to buy because prices likely to
increase.

Agree 95% 87% 81% 80% 87% 85%

Middlesex County, Mass. (Boston metro area)

Index, Jan. 2000 = 100

 1988 2004 2006 2008 2011 2012

Perceptions: Which best describes the area home price trend?

Rising rapidly 3% 33% 7% 1% 0% 5%
Rising slowly 34 51 22 2 30 47
Not changing 37 12 18 8 22 32
Falling slowly 2 4 47 71 46 16
Falling rapidly 3 0 5 17 1 0

Expectations: It's a good time to buy because prices are likely
to increase.

Agree 78% 83% 67% 82% 82% 81%

Figure 1. S&P/Case-Shiller Home Price Indexes for the Four
Survey Locations, 1987-2012 (a)

Milwaukee County, Wise.

Index, Jan. 2000 = 100

 1988 2004 2006 2008 2011 2012

Perceptions: Which best describes the area home price trend?

Rising rapidly 9% 39% 28% 5% 0% 0%
Rising slowly 53 57 51 17 14 22
Not changing 24 3 9 17 23 24
Falling slowly 12 1 11 55 57 35
Falling rapidly 3 0 2 6 5 18

Expectations: It's a good time to buy because prices likely to
increase.

Agree 85% 83% 78% 85% 88% 89%

Orange Country, Calif. (Los Angeles metro area)

Index, Jan. 2000 = 100

 1988 2004 2006 2008 2011 2012

Perceptions: Which best describes the area home price trend?

Rising rapidly 91% 87% 23% 0% 2% 2%
Rising slowly 9 13 51 3 25 48
Not changing 0 0 15 6 38 28
Falling slowly 0 0 11 63 36 23
Falling rapidly 0 0 0 27 0 0

Expectations: It's a good time to buy because prices are likely
to increase.

Agree 93% 72% 75% 79% 92% 86%

Sources: S&P/Case-Shiller and Fiserv, Inc.

a. Vertical lines indicate quarters in which the homebuyers survey
was conducted. The questions in each table are from survey questions
14 and 25: the full survey questionnaire is available on the
Brooking Papers website at www.brooking.edu/about/projects/bpea/,
under "Past Editions."

Note: Table made from bar graph.