Summary: |
When a house is placed on the market, the seller must choose a “list” or “asking” price. While there is near universal agreement that a house seller faces a trade off in choosing this price, there is less agreement about how to measure this trade off. This paper offers a model based on the idea of managing risks and a trade off between the expected sale price and the expected time-till-sale. We show that this model produces an optimal decision rule as a function of the state of the market and the characteristics of the seller. We predict that a higher list price increases expected time-till-sale. The theory also predicts that, since house buyers must solve a signal extraction problem when deciding which houses to inspect, the effect of a higher list price is magnified for houses with a low predicted variance of the list price. Using a data set with 3874 observations from Texas, we test these hypotheses using a hazard model based on a Weibull model with gamma correction for unobserved heterogeneity. The proposed methodology differs from the most common method (two-stage least squares as used by, for example, Yavas and Yang, REE, 1995) and is closer to Glower, Haurin and Hendershott (REE, 1998). We estimate the effect of changes in list price on the time-till-sale (TTS) using what can be likened to a reduced form model. This paper does not attempt to predict the sale price because our model implies that this data set, like all previously used data sets, would be contaminated by omitted variables that have a nearly equal effect on the list price and the sale price. Our analysis shows that few of the characteristics of the house significantly affect the expected time on market while various indicators of market conditions do affect TTS. Our paper demonstrates the value of recognizing censoring in the data. Censoring reduces the information contained in duration data and its effect has only been considered by one other paper (Kalra and Chan, JRER, 1994). Approximately 40 percent of listings in our data set do not result in a house sale and our theory is sufficiently general that it can be used to analyse the behaviour of such sellers. We show that the list prices of unsold listings have a higher mean and higher variance than listings where the house sold. We demonstrate that the change in estimation methodology affects the magnitude and significance of the estimated coefficients. |