Comparing Statistical and Machine Learning Methods for Sales Forecasting During the Post-promotional Period

Abstract

Retail sales promotions lead to complexity in the retail supply chain. Generally, sales promotions consist of three types of periods: a normal period, a promotional period, and a post-promotional period. However, research on forecasting with sales promotions has focused specifically on promotion effects and the subsequent sales uplift due to the promotion. Often, they are not incorporating post-promotional periods with the sales forecasts. Thus, we model sales forecasts at the stock keeping unit level using popular univariate and machine learning models to investigate their performance in the presence of the post-promotional period. Our study shows that machine learning models can identify the post-promotion effect in the presence of the right features and perform better than univariate methods. We believe our findings will help retailers plan future sales promotions effectively.