Predicting Mean and Variance in Inventory Order Decisions

Experimental newsvendor studies have shown that mean order quantities deviate from the normative predictions. In an effort to explain this phenomenon, a number of behavioral models have been proposed that can accurately account for this mean ordering behavior. However, an often overlooked result of newsvendor experiments is that order quantities exhibit considerable variability, despite the fact that price and cost parameters are constant. In this paper, we develop a simple behavioral model that predicts both mean order quantities and order variability. We begin by first describing our forecast anchoring model, which relies on a combination of simple decision heuristics, rather than an expected utility approach. We then fit the model by the generalized method of moments to a well-known newsvendor experimental data set. The estimated model parameters are used to generate predictions for ordering decisions in a separate external newsvendor data set, as an out-of-sample test. We find that our forecast anchoring model can predict both the mean and variability of order quantities well. Furthermore, as a robustness check, we conduct our own newsvendor experiment, which considers a unique setting with asymmetric two-point demand. We show that our model continues to fit ordering decisions well, whereas other models are not robust enough to provide an accurate fit in this setting. Across the different data sets, we also find a consistent pattern of subjects tending to anchor heavily on their demand forecast when the product profit margin is high. Overall, our study provides a new perspective for explaining and predicting order variability observed in behavioral inventory decisions. Our model predictions can help the upstream supply chain party better anticipate the order variability from the downstream buyer and thus improve profitability.