Price Discimination at Slashdot Speeds (Research Seminar, March 21st, 2002) Paul Kantor Distinguished Professor of Information Science Rutgers University Abstract It is well known that the demand curve for a given product changes with time. In the case of fashion items, the characteristic time parameter will be of the order of months. For seasonal goods, the changes occur over a year or several years. In the work of electronic communication, and of electronic commerce, the characteristic times may be reduced to days, hours, and even to minutes. This raises the challenge of promptly detecting variations in the demand curve, and adapting to them with prices that optimize revenue. We consider changes in magnitude, in location, and in shape. We exhibit several approaches to the problem of optimizing revenue, including methods based on Bernoulli trial models, and on Bayesian analysis. In all cases, it seems possible to revise an offered price fairly promptly in response to changes in location and in shape. However, the underlying problem is very ill-posed, as exhibited by an occasional unstable revision. We consider ways to select modeling methods, based on controlling the risk of errors of a given magnitude. Problems of embedding the whole decision process in a total cost model are considered, as are problems of maintaining positive customer relationships while rapidly adjusting prices.