Strategic bidding under uncertainty using genetic algorithms

Abstract

This paper presents a new approach based on Genetic Algorithms (GA) to find the optimal bidding strategies of a market participant. Uncertainty about competitors behavior is included in our model, which incorporates risk management. Two uncertainty levels have been considered. The first one is used to weight nominal competitors behaviors (e.g. aggressive, conservative, etc). The second one represents small deviations around these nominal behaviors. A Probabilistic Residual Demand Lineal Function is defined and used to generate an arbitrary number of competitor behavior scenarios for each nominal strategy. Once scenarios are built, a GA is used to find the optimal bids. Each individual in the GA population is a bid curve. Its fitness is computed taken into account the profit and the risk the participant is willing to assume.