Forecasting inventory prices for purchasing decisions using multiple regression techniques

Abstract

The focus of this research was on developing a technique for forecasting prices to use in making inventory purchase decisions. The basic tool used to develop the forecasts was multiple regression analysis. The technique was tested on three commodities: copper, cotton, and wheat. The input (independent) variables used to develop the equations and forecasts were taken from government publications and the Wall Street Journal. Initially 20 variables were selected to forecast the price of each commodity used as a test in the study. The initial set of variables was reduced to those that had the most significant relationship to the price being forecast. The selection was accomplished by the stepwise regression technique. Forecast equations were developed using the set of selected variables. A forecast was made each month of the calendar year of the price for each delivery month for the test commodities. This eliminated the need to use dummy variables in the forecast equations. The accuracy of the forecast was evaluated by two methods. The first was the Theil U coefficient. The coefficient allows comparison of achieved forecast accuracy to the accuracy of a naive (no change) forecast. For all three test commodities forecast accuracy superior to naive forecasts was achieved. The second method of evaluation was the comparison of the inventory purchase costs assuming (1)all purchases in the current cash market, (2)all purchases in the futures market, and (3)all purchases based on the forecasts developed in the study. Again, for all three test commodities the method using the forecasts achieved results superior to the other two methods. i.e., lowest total purchase cost. The test period encompassed the years 1976-1979.