Newsvendor 'Pull-to-Center' Effect: Adaptive Learning in a Laboratory Experiment

摘要

In the newsvendor game, the expected-profit-maximizing order quantity is higher in the demand interval when the per-unit profit margin is high and lower in the demand interval when the per-unit profit margin is low. However, laboratory experiments show a "pull-to-center" effect: average order quantities are too low when they should be high and vice versa. We replicate this pull-to-center effect in laboratory experiments and construct an adaptive learning model that incorporates memory, reinforcement, and probabilistic choice to explain individual decisions. The intuition underlying the model's prediction is that the most recent demand observation is more likely to have been greater than the optimal order quantity if the optimal order quantity is low, in which case a recency bias tends to pull the order quantity upward. A countervailing downward pull exists if the optimal order quantity is high. The recency effect may be augmented by a reinforcement bias, which causes subjects to focus more on the profitability of decisions they actually make and less on counterfactual payoffs that would have resulted from other order quantities. The predictions of this model track the observed data patterns across treatments. A pull-to-center pattern is also observed in designs involving doubled payoffs and reduced order frequency.