|
Abstract:
|
Data-driven managerial decision-making is gaining increasing popularity recently. It usually consists of two steps: 1. first learning the uncertain/random demand from historical data using statistical techniques; 2. then deriving the optimal decision strategy based on some cost-benefit analysis. The two steps are typically separate in the sense that the first step of statistical learning is merely gauged by classical statistical measures (goodness-of-fit or out-sample prediction), which may not be optimal. The need to integrate statistical learning with decision optimization has been noted in the operations management/operations research literature. In this paper, we investigate how to tailor the high dimensional statistical learning of the random input for the subsequent decision making objective in a big data environment. We study model selection and aggregation problems in operations management, including the feature selection and covariance learning problems in inventory management, revenue mangement and financial portforlio management. Our proposed managerial decision-oriented statistical learning methods outperform the traditional methods in several management problems.
|
