Designing forecasting model for a large retailer

Abstract

The study was to ascertain the relevance of ascertaining a forecasting model based on the past 3 years data for a brand. The details for the brand were not known for confidentiality reasons. The data was subjected to time series analysis to begin with. This was with trend analysis as each of the predicted values of Sales Data was plotted with respect to all the major predictors. For ensuring completeness, a predictor’s analysis was performed for checking the applicability of certain predictors to the model based on the standard deviations of some of these predictors All the above tests ensured that the number of predictors got narrowed down to a limited number. It is important to note that the competitor data was also used as predictor variables to allow for any ‘cross overs’. Since the aim was to build a forecasting model, the starting point was exponential smoothing followed by regression and finally a methodology to separate brands own initiatives vis-à-vis its competition. This framework was termed as “own effects and cross effects”. In order to compare the effectiveness of each of these models we contrasted the sigma/? or the coefficient of correlation which gave the following results. Additionally, it was found that own effect and cross effect and was also giving us additional info that was used in the second part.