Can Urban Indicators Predict Home Price Appreciation? Implications for Redlining Research
Li, Ying, Rosenblatt, Eric, Real Estate Economics
A 20% difference in home ownership rates by race (Wachter and Megbolugbe 1992) has long motivated allegations that lenders discriminate against minority neighborhoods, a process known as "redlining" (Dreier 1991). Research into these allegations has been characterized by inconsistent results and a host of methodological problems, such as simultaneous equation bias, sample selection bias and omitted variables (see Perle, Lynch and Horner 1993; and Rachlis and Yezer 1993 for examples). However, there seems to be widespread agreement that redlining models should control for "economically justifiable and appropriate" neighborhood risk indicators, collected by institutions such as the Bureau of Census and Department of Housing and Urban Development (Holmes and Horvitz 1994). The levels of such indicators are presumed to affect the probability of default, and thus the profitability of lending, by altering the distribution of future collateral values. Underwriters and lending officers, who are skilled at interpreting such risk factors and can predict which neighborhoods are likely to prosper and which to fail, are therefore in possession of valid business reasons for preferring one neighborhood over another.
Redlining researchers have seen little reason to be conservative in the number of risk indicators inserted into such models, nor have they concerned themselves with testing their significance in regressions. For instance, Schill and Wachter (1993) include median household income, median house value, median age of houses, percent of vacant housing, percent of occupied units and the percent of households receiving public assistance, though their regressions seldom reveal the variables to be significant. This raises a serious irrelevant variables issue. Most of these variables are highly correlated with the race variables which are the subject of these models, and the specifications can lead to misleading regression results due to multicollinearity. In other words, if race-correlated indicators simultaneously enter the equation, it becomes virtually impossible to support the hypothesis of redlining.
In a study of three California primary metropolitan statistical areas (PMSAs), this paper undertakes to determine if the use of neighborhood indicators as proxies of default risk is justified. The mechanism by which these factors alter default risk is assumed to be exclusively by means of their effect on future collateral values. Future collateral values depend on present collateral values and future home price appreciation. However, future home price appreciation, and therefore future collateral values, can only be estimated. Neighborhood indicators may provide additional information beyond current prices in predicting future home price appreciation. This can be true, for instance, if the housing market is not informationally efficient with respect to neighborhood qualities. Inefficiency in the form of autocorrelation in real house prices is found in Case and Shiller (1989) and numerous later papers. See Cho (1996) for a survey of the literature.
Following the suggestion of Rachlis and Yezer (1993), the contribution of neighborhood risk factors to local home price appreciation and its variation is estimated. This is an effort with implications not only for redlining research but for pricing mortgage credit risk and for real estate portfolio management, as well. A California sample of the Fannie Mae/Freddie Mac joint repeat transactions database is used to construct annual census tract level home price appreciation rates in the Los Angeles-Long Beach, Oakland and Anaheim-Santa Ana PMSAs. The neighborhood indicators in these three PMSAs are constructed from the 1990 Census.
As there is little theoretical work on the expected relationship of indicator levels and home price appreciation, the main hypothesis is the simplistic theory that levels associated with contemporaneously higher home prices are associated with higher future home price appreciation and lower home price appreciation variance. …