A forecasting model addressing the issue of correlation with application to automobile sales

Forecasting is a method of predicting future events from historical outcomes. Forecasts are not free from error. However, error can be minimized through the careful analysis of data and selection of relevant variables for inclusion in forecasting models. Multicollinearity has presented problems with respect to correlation among the independent variables. Many times the analyst assumes independence among the predictor variables and allows errors due to multicollinear relationships to be accumulated with other sources of random error. Factor analysis is a statistical technique used in the social sciences to examine correlative relationships in variables. The outcome of this analysis procedure is a minimum set of independent factors that explain the multicollinear relationships in the variables. By utilizing these factors in a regression model a less costly forecast can be developed which accounts for correlation among the variables in terms of the independent factors. The decision for usage of a forecast is primarily based on the benefits derived in increased profits versus the cost of developing a more realistically modeled forecast. It is assessed in this thesis that overstated forecasts versus understated forecasts are deemed more favorable in cases where inventory holding costs are lower that stock-out costs. The reverse is true when stock-out costs are lower than inventory costs. A case study was presented for analysis regarding the issue of automobile sales. The problem identified from review of market activity was an unrecognized change in demand for full-size cars relevant to the auto industry. Before implementing a change in production to increase sub-compact units, the characteristics and amount of demand for sub-compact cars should be forecast. By developing an accurate forecast of demand for subcompact vehicles, American automobile manufacturers will be able to minimize additional costs incurred by changed production processes for the modified product mixture (increase production of sub-compacts, decrease production of full-sized cars). An analysis was conducted on data available from the Department of Commerce. Results of the analysis indicate that a factor analysis would be appropriate for development of the forecasting model for sub-compact car demand. The risk of increasing holding costs due to an overstated forecast were considered worthwhile in consideration of the profits that would be lost by adhering to an understated forecast. The example used in this thesis demonstrates the use of factor analysis and supports the implementation of this technique for model development of sub-compact car demand. In consideration of the successful utilization of this procedure, recommendations were made to pursue research toward the practical application of factor analysis in forecasting methods.