PDF | Free | 71 pages
Download PDF

Predicting House Price Bubbles and Busts with Econometric Models

What We’ve Learned, What We Still Don’t Know

James R. Follain and Seth H. Giertz

September 2012, English

This paper builds upon our previous study (Follain and Giertz, 2011b) of the house price bubble and bust that has affected many US metropolitan housing markets in the last several years. Here, data are compiled for a balanced panel of 342 MSAs from 1990–2010. A vector error correction model (with as many as five second stage equations) is employed to estimate the relationship between house prices and an array of economic variables. Out-of-sample house price paths are projected for several different three-year periods, with special focus on projections from 2006:Q3 to 2009:Q2 and from 2008:Q1 to 2010:Q4. The model is also estimated for several additional time periods (such as through 2000:Q4 and 2010:Q4). For each specification, Monte Carlo simulations are conducted to project an array of house price paths. The cumulative three-year price changes from the Monte Carlo simulation are ordered, so as to examine both median projections, as well as low probability events—such as the price path at first or fifth percentile of the distribution. Attention is focused upon the ability of the model to predict actual house price changes from 2008:Q1–2010:Q4. Follain and Giertz (2011b) adequately captured the ranking of the MSAs that were hardest hit during this period, but substantially underestimated the magnitude of the price declines. Here, the predictions from the various models generate much larger declines than the models from our previous work and, in some instances, exceed the actual declines that occurred. Excluding from the in-sample data all periods after 2006:Q2 has a major impact on the three-year projections (and in the forecast error). With only data through 2006:Q2, the cumulative three year projections for house prices generally under predicted what occurred and by a much greater degree than the model with just 18 more months of data. We find some evidence of a growing importance of price momentum in the models estimated using data ending at different points during the first half of the 2000s. This increased momentum appears to reflect the enhanced sensitivity of the housing market to negative shocks in other variables and, especially, to declines in the growth rates of house prices. In effect, an exogenous shock may have started the drop in house prices, but the greater importance of momentum may have led to a snowball effect. The model is also estimated using data through 2010:Q4 and predictions for 2011–2013 are provided. These initial projections suggest the road to recovery is a long way off.