How to Predict in An Unpredictable World? Forecasting Automobile Consumer Demand during and after a Pandemic

COVID-19 has caused unprecedented disruptions in global supply chains, changing consumers’ behavior and purchase patterns at an unanticipated scale and speed. As a result, traditional demand forecasting models trained primarily on a company’s historical data may have acute and wide-spread biases. Therefore, firms are urgently searching for newer and smarter ways to forecast demand, especially in manufacturing industries. In this study, we aim to solve this problem by providing a new machine learning approach that can accurately predict consumer demand for the automobile manufacturing industry. We collaborate with two automobile manufacturers to obtain a unique proprietary operational dataset and show that the industry’s traditional demand forecasting models are no longer valid during and after the pandemic, even when pandemic-related variables are incorporated. We propose a new modeling technique, i.e., the boosting method, to predict consumer demand and show that this method could perform quite well by integrating social-media related information. We conduct extensive data mining analysis to identify how consumer behavior changes during and after the pandemic. Our analysis and results provide useful tools for firms to quickly anticipate how consumer demand can shift and adjust their production planning accordingly. We also outline how policymakers can implement policies to improve automobile supply chain continuity in the context of a pandemic such as COVID-19