Forecasting models were developed for annual and monthly average price of Maine potatoes, using Crop Reporting Board data. Several approaches to the forecasting problem were explored, as were a number of statistical aids to model development and evaluation. Attention was focused on providing timely forecasts as data became available.

Specifically, three models were developed. The first provided forecasts of the season average price. Attention here was placed on the selection of a subset of influential variables from a set of potentially important ones, as well as the detection of outliers and colinearities. These problems were especially significant given the small number of observations in the annual price series. Production data for five regions were deemed most influential.

The other two models were developed to provide forecasts of the monthly average price from November through May. A modified ARIMA (0,1,0) (2,1,0) model was estimated and used to provide forecasts based on the price data series alone. An alternate model, using production levels from four regions, current stocks levels and a monthly trend variable (and, implicitly, lagged price), was also developed.

Both monthly models provided current forecasts for each month remaining in the marketing season which were updated as new information became available. The second model, which incorporated more information, was deemed to provide somewhat more accurate forecasts, at all forecast distances, than the simpler ARIMA model, when compared over the 1977–1981 period.