Assessing store performance models

One of the major problems faced by the management at supermarket chains is the determination of a fair and equitable assessment of individual store performance keeping in mind the variation in store features, competitive environment, and socio-demographic characteristics of the consumers facing each location. Mainstream literature has thus far tried to define "trading areas" with finite borders around supermarkets to identify their customer base. However in a densely populated highly developed urban environment, such trading areas have very high overlaps (90-95%) between stores, rendering the use of finite border models not only ineffective, but incorrect to use. In this paper, we formulate and estimate four different models based on various assumptions such that in all models, market force of a store is conceptualized as a function of internal and external features of the store. These models are estimated using a multistart Newton-gradient method and Simple Genetic Algorithm with Elitist Strategy. We tested our models using scanner data from a multistore supermarket chain in the Northeast region of the USA, facing 1075 major competitors in 9 chains and covering approximately 6000 customer buying districts as designated by the US Census Bureau. We find that out of the four models estimated, the sales based model gives the best fit. We achieve 77% prediction accuracy of sales. We examine the residuals, then, as mainly due to fluctuations in store management quality. The resulting approach enables management to estimate expected sales for each supermarket in each buying district and compare it with their real performance. This way both supermarkets and buying districts can be evaluated. Moreover it is also possible to estimate the intensity of competition in a given location and predict sales of a planned store with given features, thus providing input for a location planning model.