Dupont Forecasting Options

Introduction

DuPont manufactures products that cater to a variety of industries such as agriculture, healthcare and construction. As anticipated, these industries are susceptible to seasonal fluctuations. Consequently, these fluctuations impact DuPont’s revenue stream as apparent in Exhibit 1 which shows DuPont’s quarterly revenues from Q1-1999 to Q2 2008. In addition to the evident seasonal component, DuPont’s revenue data also exhibits a slight trend. The goal of this assignment is to select the best forecasting model that takes into account DuPont’s trend and seasonal fluctuations when projecting the company’s quarterly revenues from Q3 2008-Q2 2009.

A plethora of techniques were implemented on DuPont’s training set (Q1 1999-Q2 2008) to arrive at the optimum model that exhibited minimum forecasting error and highest accuracy. Exhibit 2 shows the range of models tried and a summary of the criteria on which each of the models were evaluated. Eventually, the MAD and RMSE levels were analyzed to select the best model for forecasting DuPont’s quarterly sales.

Initial Computations for Each Model

• Seasonal Decomposition: This model was applied on the revenues variable to take the seasonality component out of the data and to arrive at seasonal indices (Additive and Multiplicative). Eventually, equations containing only the trend component were arrived at and the seasonal indices were added/ multiplied to estimate the forecast for Q3 2008-Q2 2009. Based on the forecasting error measures, the Additive method proved to be the better of the two seasonal decomposition models.

• Multiple Regression: Dummy variables for Q2, Q3 & Q4 were created. Based on the correlation matrix and a stepwise analysis of the Multiple Regression model, the following model was picked (out of 6 potential models) with an adjusted R2 of 0.752, a Durbin Watson of 1.121, statistically significant variables and a homoscedastic residual plot....