Residential land prices prior to development

Journal of Real Estate Research, The, 1997 by Karl L Guntermann

A total of seventy-nine transactions are used in this study, extending from areas that are substantially developed to the largely vacant urban fringe.7 The data are from Mesa and Gilbert, Arizona, which are located in the southeastern portion of the Phoenix MSA in what is known as the east Valley. Portions of those cities underwent substantial development during the 1980s and they would be representative of what occurred throughout the Phoenix area during the 1983-91 period. Both substantial appreciation and depreciation in land prices is reflected in the transactions data. Approximately onehalf of the transactions used here resulted in residential development, usually within one to two years of sale, while the other parcels had not been developed by the end of 1993. The parcels were zoned for either an agricultural or residential use of varying densities and ranged in size from 3 to 483 acres. Descriptive statistics for the variables used in the regressions presented here are in Exhibit 2.

Results

Ideally, an estimated model would include any variables that could be used to test a particular hypothesis, such as population and employment forecast variables to test one of the development expectations hypothesis. However, multicollinearity problems typically preclude such a direct approach. A correlation matrix for the development expectations and local growth variables is presented in Exhibit 3. The variables for the number of active subdivisions within one, two or three miles of each parcel are significantly and highly correlated with each other, as might be expected. As a consequence, separate models have been estimated including only one of the subdivision variables. The same situation exists in the other test of development expectations using the population and forecast variables and with the tests involving the growth capitalization variables. Hence, alternative models test only one variable of each type.

The results for specifications of the model with the employment forecast and three mile subdivision variables and alternative growth variables are presented in Exhibit 4 for the semi-log models and in Exhibit 5 for the log-linear models. The log-linear models typically had greater explanatory power with better significance levels for many variables. The SIZE variable generally is significant and the coefficient is consistent with the results reported in earlier studies. The availability of utilities and infrastructure to each parcel (OFFSITES) is highly significant and has a substantial effect on value, as might be expected.8 The accessibility variables, DISTFWAY and DISTDEV are generally not significant, especially DISTDEV, and the zoning variables are also insignificant in all cases.

The employment forecast variable generally is significant, providing evidence that development expectations and, hence, land prices are sensitive to variations in those forecasts across small planning areas. Use of the population forecast variable instead of employment produced generally similar results in the semi-log specifications but somewhat weaker results in the log-linear models (Appendix 1). Based upon the correlations between POPFORE and the dependent variable, real price per acre (RPRIACRE) and EMPFORE and RPRIACRE (Exhibit 3), these results are to be expected. Irrespective of which forecast is the better proxy for development expectations, the conclusion that can be reached from these findings is that the market is responsive to credible near-term forecasts (one to five years) of growth and development in an area and they are capitalized into land prices.