Predicting mean and variance in inventory order decisions

We develop a simple forecast-anchoring model to explain and predict the mean and variance of observed inventory order decisions in a newsvendor problem. The model assumes that people employ a two-step decision heuristic. In the first step, a behavioral bias may gravitate the decision maker's point forecast toward a random forecast versus a constant unbiased forecast. In the second step, a behavioral bias of the same magnitude may cause the decision maker to treat the point forecast as if it is the mean of potential demand, and then make an upward or downward adjustment depending on the underage and overage costs. We evaluate the performance of this descriptive forecast-anchoring model across five experimental newsvendor data sets. First, we fit the model to a setting with uniform demand. We then use the corresponding estimates to generate predictions in a secondary data set with uniform demand, as an out-of-sample test. We proceed to fit the model to three additional newsvendor data sets, two with normal demand and one with asymmetric two-point demand. In all cases, the model predicts the mean and variance of inventory order decisions well. We further investigate the profit implications under the forecast-anchoring model and find that the predictions match well with the experimental data. Through improved predictions, the model can help upstream supply chain parties anticipate inventory order decisions from buyers and improve profitability.