Experiment With Multiple Regression Models for Sales Forecast and Location Analysis

Location is an important consideration for retailers as it is a huge upfront cost and has a large impact on sales. There are many methods that can be used for location analysis, and regression-based modeling is one that explicitly forecast sales performance and allows incorporating a large number of factors. Previous research in this area were limited due to the lack of real sales data for academic researchers. This paper uses data from a furniture retailer operating a large number of locations in Canada. Variables are chosen based on their impacts on furniture sales. Trade areas are delineated using the Thiessen Polygon approach and descriptive trade area method using real sales data. Two sets of sales forecasting models reconstructed, one using real furniture sales data and one without. A reasonably adequate sales projection model is developed with the best model having a range of projection errors of ±40%.