| Summary: | Retailers need accurate forecasts for inventory planning. Poor forecasts lead to either out-of-stock or over-stock conditions. If products are regularly out of stock, customers may get frustrated and eventually switch their loyalty to other supermarkets. Retailers also do not want to overstock items because it is costly. To promote the various products in their stores, retailers employ marketing mix activities such as price reductions and promotions. There is an established marketing literature that focuses on identifying and estimating the effects of the marketing mix activities. However, little research effort has been devoted to incorporating the information of the marketing mix activities in forecasting retailer sales at the Universal Product Code (UPC) level. The forecasting models in previous studies only take into account the effects of the marketing mix activities for the focal product. This thesis proposes econometric forecasting methods that also take into account the effects of competitive marketing mix activities. The selection of competitive marketing mix variables becomes important because it is not obvious which UCPs compete against each other. The relationship between the marketing mix activities and the product sales can change permanently. For example, consumer taste towards a particular product may change or we can consider the availability of a new close substitute product. However, traditional econometric models with fixed parameters assume that the relationship is time invariant. As a result, the model may be subject to structural breaks and thus produce biased forecasts. This thesis implements various recently developed techniques to adjust the fixed parameter model with respect to the forecast bias caused by structural breaks, which may potentially improve the forecasting accuracy. The empirical analysis suggests that the inclusion of competitive marketing mix variables offers worthwhile benefits and the adjusted econometric models which take into account structural changes can produce more accurate forecasts than traditional econometric models.
|
|---|
