Stalking Information: Bayesian Inventory Management with Unobserved Lost Sales

摘要: Retailers are frequently uncertain about the underlying demand distribution of a new product. When taking empirical Bayesian approach Scarf 1959, they simultaneously stock product over time and learn distribution. Assuming that unmet is lost unobserved, this learning must be based on observing sales rather than demand, which differs from in event stockout. Using framework results Braden Freimer 1991, cumulative captured by two parameters, scale parameter reflects predicted size market, shape indicates both market precision with known. An important simplification result 1960 Azoury 1985, allows to removed optimization, shown extend setting. We present examples reveal interesting phenomena: 1 A retailer may hope that, compared stocking out, realized will strictly less level, even though out would signal stochastically larger distribution, 2 it can optimal drop after period successful sales. also specific conditions under following hold: Investment excess stocks enhance occur every dynamic problem, never dropped poor The model extended multiple independent markets whose distributions depend proportionately single unknown parameter. argue smaller should given better service as an effective means acquiring information.