Daily Sales Forecasting for Variable-Priced Items in Retail Business

Modern retail business manages products from various sources to serve consumers. To be able to respond to customers’ needs, accurate sales forecasting is essential to prepare appropriate levels of stocks. This research aims to find methods and features to forecast the daily sales for a case study retail store chain having many categories of products with varied promotion prices. Three main models are considered: (1) Time series forecasting, i.e., TBATS model, (2) explanatory forecasting method, i.e., multiple linear regression, and (3) machine learning model, i.e., XGBoost. Different types of dependent and independent variables transformation in Stepwise regression are also considered in order to find the most accurate results. Since there were records of the number of Covid-19 cases in Thailand from 2020 and there was also government's welfare money policy during Covid-19 crisis, this paper attempts to find how these variables affects retail sales. Weighted absolute percentage error (WAPE) is used to compare the accuracy among different models.