Demand Forecasting using Machine Learning Methods: An Empirical Study on Walmart Retail Sales Forecast

Abstract: Economists put considerable effort and research into demand estimation, modeling consumer be- havior uncovering causal relationships between product attributes, consumer preferences, prices, and demand. However, demand forecasting is less explored in the economics literature. This paper provides insights into machine learning methods and econometric methods that applied economists can use to forecast demand. In particular, Long Short-Term Memory (LSTM), and eXtreme Gradient Boosting (XGBoost) are compared with the more traditional econometric model, ARIMA in forecasting product demands from a multinational retail corporation, Wal- mart. Machine learning models performed better in terms of prediction accuracy compared to ARIMA. LSTM demonstrated the highest performance in efficiently capturing non-linear com- ponents in sales data.