forecastingprinciples.com Reviews of Important Papers on Forecasting,
1985-1995 Reviews
Review of:

Terri A. Sexton (1987), "Foreasting property taxes; A comparison and evaluation of methods," National Tax Journal, 15, 47059.


This study compares the forecasting accuracy of six forecasting models. These include econometric and time series models. The models were calibrated using data on property taxes in 77 Minnesota counties during the period from 1974 through 1980. Forecasts of changes in the real market value of residential real estate were made for the years 1981 through 1983; forecast horizons of 1, 2, and 3 years were examined. The results were, in general, consistent with those from prior research. Interestingly, however, few of the relevant prior studies were cited. The study concluded that:

(1) Econometric methods yielded more accurate forecasts than did time series, especially as the forecast horizon increased.

(2) Improved fit to historical data had little relationship to forecast accuracy.

(3) A combined forecast based on econometric and time series forecasts was significantly more accurate than was either one alone.

(4) Econometric models within segments were no more accurate than an econometric model across the whole sample. This is the only surprising result, in light of previous research.

Now the bad news. All of the forecast comparisons were ex post. The author stated: "Following standard forecast evaluation practice, actual changes in the explanatory variables were used in predicting . . ." Clearly, ex post forecasts are of interest. However, recent practice has placed a heavier emphasis on ex ante (unconditional) forecasts as a way to more closely simulate the forecasting problem. The use of ex post forecasts and the small validation sample (three years) limits the confidence that we can place in the study. Nevertheless, with the exception of the result about econometric models by segment, the results add support for the current beliefs among forecasting researchers.