Theory-Regularized Deep Learning for Demand-Curve Estimation and Prediction

Advances in deep learning have significantly enhanced predictive accuracy in marketing decision-making, particularly in demand curve estimations and predictions. However, a key limitation of flexible deep learning methods is that their predictions become unreliable when applied to prices or environments different from those in the training and validation samples. To address this limitation, we propose incorporating economic theory of demand to inform and regularize the loss function. In our approach, the model’s training and validation on the pre-COVID-19 sample are regularized using the linear expenditure function, which describes individuals’ budget allocation within a product category. When tested on the post-COVID-19 sample, characterized by significantly different prices and environments, our method demonstrates a substantial improvement in predictive accuracy compared to the standard deep learning approach.