Abstract:This paper considers a fírm facing an uncertain demand curve. The firm can experimentally adjust its output in order to gain information that will increase expected future profits. We examine two basic questions. Unde= what conditions is it worthwhile for the fírm to experiment? How does the firm adjust its output away from the myopic optimum to exploit its ability to experiment? Two necessary conditions are established for experimentation to occur, involving requirements that experimentation be informative an… Show more
“…Here the construction is more subtle because after some histories bidders may find themselves in a situation where they are uncertain about whom they bid against. Then they face a dynamic programming problem similar to a monopolist who experiments with prices in order to learn about an uncertain demand function, as for example in Rothschild [1974], McLennan [1984], Aghion, Bolton, Harris and Jullien [1991], and Mirman, Samuelson and Urbano [1993].…”
We study collusion in repeated first-price auctions under the condition of minimal information release by the auctioneer. In each auction a bidder only learns whether or not he won the object. Bidders do not observe other bidders' bids, who participates or who wins in case they are not the winner. We show that for large enough discount factors collusion can nevertheless be supported in the infinitely repeated game. While there is a unique Nash equilibrium in public strategies, in which bidders bid competitively in every period, there are simple Nash equilibria in private strategies that support bid rotation. Equilibria that either improve on bid rotation or satisfy the requirement of Bayesian perfection, but not both, are only slightly more complex. Our main result is the construction of perfect Bayesian equilibria that improve on bid rotation. These equilibria require complicated inferences off the equilibrium path. A deviator may not know who has observed his deviation and consequently may have an incentive to use strategic experimentation to learn about the bidding behavior of his rivals.
“…Here the construction is more subtle because after some histories bidders may find themselves in a situation where they are uncertain about whom they bid against. Then they face a dynamic programming problem similar to a monopolist who experiments with prices in order to learn about an uncertain demand function, as for example in Rothschild [1974], McLennan [1984], Aghion, Bolton, Harris and Jullien [1991], and Mirman, Samuelson and Urbano [1993].…”
We study collusion in repeated first-price auctions under the condition of minimal information release by the auctioneer. In each auction a bidder only learns whether or not he won the object. Bidders do not observe other bidders' bids, who participates or who wins in case they are not the winner. We show that for large enough discount factors collusion can nevertheless be supported in the infinitely repeated game. While there is a unique Nash equilibrium in public strategies, in which bidders bid competitively in every period, there are simple Nash equilibria in private strategies that support bid rotation. Equilibria that either improve on bid rotation or satisfy the requirement of Bayesian perfection, but not both, are only slightly more complex. Our main result is the construction of perfect Bayesian equilibria that improve on bid rotation. These equilibria require complicated inferences off the equilibrium path. A deviator may not know who has observed his deviation and consequently may have an incentive to use strategic experimentation to learn about the bidding behavior of his rivals.
“…Balvers and Cosimano [3] modeled demands as a linear function of prices with unknown slopes and intercepts, which motivated to learn by estimating parameters in the linear model. [19] further examined the incentives of demand 3 learning, and established two necessary conditions for a firm to learn uncertain demand curve from experiments. Later, Petruzzi and Dada [20] considered a demand model with both additive and multiplicative stochastic components, whose distributions are updated over time using Bayes' rule.…”
In this paper, we propose a revenue optimization framework integrating demand learning and dynamic pricing for firms in monopoly or oligopoly markets. We introduce a state-space model for this revenue management problem, which incorporates game-theoretic demand dynamics and nonparametric techniques for estimating the evolution of underlying state variables. Under this framework, stringent model assumptions are removed. We develop a new demand learning algorithm using Markov chain Monte Carlo methods to estimate model parameters, unobserved state variables, and functional coefficients in the nonparametric part.Based on these estimates, future price sensitivities can be predicted, and the optimal pricing policy for the next planning period is obtained. To test the performance of demand learning strategies, we solve a monopoly firm's revenue maximizing problem in simulation studies. We then extend this paradigm to dynamic competition, where the problem is formulated as a differential variational inequality. Numerical examples show that our demand learning algorithm is efficient and robust.
“…Firms often know more about their costs and/or demand than the contracting party and thus, the third party may be able to learn about private information of the …rm through contracting variables and publicly observable information. While there is a vast literature on experimentation by …rms under di¤erent market structures and using di¤erent choice variables (see Aghion, Espinosa and Jullien (1993), Mirman, Samuelson and Urbano (1993) and Belle ‡amme and Bloch, (2001) for example), the issue of learning by a principal has only been studied recently. , in a pioneering paper study learning by a principal when the agent has private information and the outcome is noisy (see also, Jeitschko, Mirman and Salgueiro (2002)).…”
This is the accepted version of the paper.This version of the publication may differ from the final published version.
Permanent
AbstractIn this paper, we examine the e¤ect of potential entry on learning by a lender when the demand shock has a general distribution. We show that under this type of noise, entry does not lead to any changes in the equilibrium expected signals and therefore, there is no e¤ect on learning by the lender, unlike the case when noise is uniformly distributed. The result holds even when contracts are not observable.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.