Abstract

To model market dynamics is a challenge that has attracted the interest of practitioners and researchers alike. This problem has been addressed from the perspective of Game Theory, in models that explicitly include profit-maximization schemes for the companies, and also from the point of view of Data Mining, with models that consider multivariate functions to model customer demands and related phenomena. In this work we present a two-stage model that unifies both approaches. A hybrid neural network-support vector machines model estimates multiclass demand at a customer level, which then serves as input for a game-theoretic model that considers the strategic relationships between costs and demands in pricing schemes for Bertrand equilibria. The model was applied to a database in a loan-granting institution with good results. New knowledge discovered includes insights about cost structures and the institutions' competitive behavior, providing new business opportunities. © 2010 IEEE.