Decomposition of Multiple Sales Promotion Effects on Infrequently Purchased Category

摘要

A Bayesian inference is an important technique for data analysis in large scale as discovery science. Kondo and Kitagawa introduced a Bayesian method in order to decompose price promotion effect of store level scanner sales into brand switching and category expansion, together with long-term component of baseline (trend) sales and cyclical day-of-week effect. This paper presents another analytical example of an infrequently purchased category, Chinese Tea, by adding one more promotional variable of display as explanatory variable. The analysis on this category shows that there exists day-of-week effect within display effect that causes very sharp spikes of incremental sales.