Estimating Land Values Using Residential Sales Data

In this research, we attempt to develop a statistically rigorous, but straightforward and tractable method by which vacant land sales on the periphery of a metropolitan area can be used to estimate underlying implicit land values throughout the city, including areas in which there are few or no vacant lot sales. Our methodology involves a two-step process. In the first step, we use single-family residential sales data to estimate a constant-quality price surface using a locally-weighted regression (LWR) model. This standardized property value surface is obtained by pricing an identical housing bundle at every location, where this bundle includes both the typical dwelling characteristics and the typical lot characteristics.

The second step of our analysis involves using vacant land sales in neighborhoods on the periphery to “pin down” the absolute level of the standardized value surface. We do this by imputing the value of the standard dwelling at the locations of the vacant lot sales, and subtract the observed selling prices of the lots to come up with the “value” of the standard dwelling. Subtracting this imputed dwelling value from our standardized price surface provides us with estimates of absolute land values at each location throughout the city.

The land surface we estimate is consistent with our understanding of the local market from which are data are obtained (Wichita, Kansas) and is decidedly non-monotonic, with neighborhoods of significant premium in land prices over others. We also find a number of regularities at odds with the basic urban model in a setting we had anticipated to behave much like the proto-typical monocentric city: polycentric land pricing and land at the periphery varying widely rather than being price according to some nominal “agricultural” use. Though our discussion focuses primarily on the land price recovery exercise, these are potentially interesting findings that speak directly to validity of the basic urban model in its standard form.