Maybe not so nonsensical after all
Submitted by Econbrowser
How far will house prices fall? Implications from the latest WSJ survey.
About 5 months ago, I posted a figure depicting the impending resets on subprime mortgages, option adjustable rate and Alt-A ARMs. Resonding to one comment, another reader made the following observation:
…a fall of 50% in [valuations on residential housing].
Nonsense.
…
I have to say that a 50% drop struck me as implausible as well. But I thought I would re-examine this view in light of recent developments, and what the economists surveyed by the WSJ say.
First, consider the OFHEO House Price Index (HPI). In Figure 1, I plot the historical (purchases-only) index (blue line), the average forecast as implied by the March WSJ survey (red line). I also include the high estimate and low estimates.
Figure 1: OFHEO seasonally adjusted HPI (purchases only) (blue line), mean forecast from WSJ March survey (red line), and “hi” and and “lo” foreasts (teal lines). NBER-defined recession highlighted gray. Sources: OFHEO, WSJ March survey, NBER, and author’s calculations.The two year high estimate is from James F. Smith of Western Carolina University and Parsec Financial Management, while the corresonding low estimate is from Maria Fiorini Ramirez/Joshua Shapiro of MFR, Inc. So while the average forecast is for continued price decline, there is a wide dispersion of views. Figures 2 and 3 depict a histogram of percentage changes for 07Q4/08Q4 and 08Q4/09Q4, respectively.
Figure 2: Histogram of forecast 07Q4/08Q4 price changes, in percent. Source: WSJ March survey.
Figure 3: Histogram of forecast 08Q4/09Q4 price changes, in percent. Source: WSJ March survey.The histograms indicate not only a lot of dispersion, but considerable skewness in the 08Q4/09Q4 forecasts. These latter forecasts are not distributed Normally, according to the Jarque-Bera test statistic, using a 10% MSL.
In figure 4, I plot a detail of the OFHEO index, in log terms, normalized to the peak equal to zero in 07Q2. This graph implies that house prices in 09Q4 will be 8.4% lower (in log terms) according to the average forecast. The worst case scenario depicts a 25% decline. For some perspective, the forecast of UCLA’s Ed Leamer — who predicted accurately the 2001 recession — is also plotted (green line). Despite predicting a less than 45% chance of a recession, his forecasts imply a 17.3% decline in the OFHEO HPI. (Jan Hatzius, who has been prominent in the debate over the impact of bank losses and deleveraging (discussed here), did not provide a forecast.)
Figure 4: Log OFHEO seasonally adjusted HPI (purchases only) (blue line), mean forecast from WSJ March survey (red line), “lo” forecast (teal line), and Ed Leamer/UCLA Anderson School forecast (green line), all normalized to zero in 07Q2. Source: WSJ March survey, and author’s calculations.Now, even the worst-case price decline does not seem that disasterous. But as has been discussed on a number of occasions, the OFHEO HPI (in this case the purchase-only index) has some characteristics which makes interpretation difficult. Most importantly, it measures the prices of houses that are financed with conforming loans.
What would be of interest is to know what the forecasted decline in the OFHEO index implies for the Case-Shiller index of house prices, which could be construed as being more representative (for debate, see [1], [2], [3]). I’m going to take a particularly naive and atheoretical approach, and use the linear relationship observed between log differenced prices, viz.:
dcsxr t = -0.018 + 1.754(dhpi t) + 0.610(dhpi t-1) + 0.302(dhpi t-2) + e t
Adj. R 2 = 0.72, SER = 0.0108, DW = 1.515, smpl = 1991Q4-07Q4.
bold face coefficients indicate significance at the 10% level, using Newey-West HAC standard errors (lag truncation = 3). csxr is the log Case-Shiller 10-city index, hpi is the OFHEO log HPI (purchases only), and “d” prefix denotes a first differenced term.
The regression is not a perfect one, but CUSUM and CUSUMSQ tests fail to reject the null of no structural breaks. After the first year of residuals, the 1-step ahead and n-step ahead recursive residual tests fail to reject at the 1% MSL.
Using this relationship to project out the implied decline in the Case-Shiller index for the 2008-09 period yields the purple line in Figure 5.
Figure 5: Log OFHEO seasonally adjusted HPI (purchases only) (blue line), mean forecast from WSJ March survey (red line), log Case-Shiller house price index (green line) and implied Case-Shiller 10-city index based upon estimated relationship with OFHEO HPI. Case-Shiller index monthly data converted to quarterly by arithmetic averaging on unlogged data. NBER-defined recession highlighted gray. Sources: OFHEO, S&P, WSJ March survey, NBER, and author’s calculations (see text).The combination of the WSJ mean forecast, and the observed correlation over the 1991Q4-07Q4 period leads to an implied 40% log decline in the Case-Shiller price index (34.7% in percentage terms). (If one uses a static-no lag specification, then the implied drop is 34.8% in log terms.)
(By the way, the above results pertain to the 10-city index. The 20-city index is only available from January 2000 onward. Estimating an analogous relationship (in log first differences, using 3 lags), yields an Adj. R2 of 0.78. The implied decline from the 2006Q3 peak is 38.3% in log terms. Hence, while the 20-city index runup is less pronounced than the 10-city runup, the decline relative to peak is essentially the same.)
I certainly don’t want to assert that house prices will decline by 40% in log terms. What is true is if (1) the mean forecast for the OFHEO HPI is realized, and (2) the historical correlation between the OFHEO and Case-Shiller indices continue to hold, then a 40% decline in 10-city prices is implied.
One corollary of this result is that only a slightly more pessimistic than average forecast implies a 50% decline in house prices as measured by Case-Shiller, relative to peak (the “Lo” estimate implies a 84% log decline).
In the same post I referenced above, I was admonished by one reader, to this effect:
…You must understand that when Menzie serves up one of his Dishes of Doom, the Doomsters dig in, adding their own special spices to the meal.
Never let it be said that I do not take into account readers views. Hence, in order to present the upside scenario, let me observe that if the “Hi” case (which in 07Q4/08Q4 is 2.2 standard deviations above the mean prediction) is realized, then the Case-Shiller 10-city index will be only 9.6% below peak (in log terms), and less than half a percent lower than what it was at 07Q4.
Visit 1800blogger to see all of our industry leading blogs.
Are you an investor? Have something you want to say about the economy? Register on Econoimist Blog now and get published within minutes. Before posting, it is recommended that you review our posting guidelines.
