Noisy Bayesian Learning in an Oligopolistic Newsvendor Market with Demand Inertia

Newsvendors, such as newspaper companies, face the challenge that they need to decide how many units to produce before facing the demand by customers, and that their products are perishable. Experimental studies have documented that individuals do not find the optimal solution to that problem when they are asked to solve it under controlled conditions in the laboratory. This paper sheds light on the cognitive processes that explain individuals' inability to act optimally, using the data by Nagel and Vriend (1999). These authors implement the problem as a game, in which subjects need to make two decisions: the number of units they produce, and the amount of resources spent on advertisement. While the advertisement decisions converge to the Nash equilibrium, average production is inconsistent with the equilibrium even after many rounds of play. Our findings are interesting. First, advertisement decisions are well-described by a hill-climbing algorithm, linking the adjustment steps to the posterior uncertainty. Second, the bias in production is consistent with Bayesian learning, predicting slow adjustment. Third, the volatility patterns in production are well-described by Thompson Sampling, a Bayesian heuristic from the Operations Research literature