The Introduction of Memory in Financial Choice David Ceballos Hornero, Mª Teresa Sorrosal Forradellas, Dídac Ramírez Sarrió IAFI, Research Group of the University of Barcelona Department of Economic, Financial and Actuarial Mathematics Av. Diagonal 690, E 08034, Barcelona (Spain) {ceballos, sorrosal, dramirezs}@ub.edu 1 Introduction Financial decision has been considered as a rational choice that can be axiomatized perfectly within the framework of Expected Utility Theory (EUT). Nevertheless, this scientific focus, in spite of its high level of applicability and dominant position, presents a number of paradoxes, and it has been increasingly noted that theoretical conclusions differ from the financial decisions and behavior observed. We propose an alternative, related to Behavioral Finance [1], that incorporates the effects of memory into the decisions of a financial agent. In our case, the agent is able to change the proportion of fixed and variable claims in an investment fund every month. This decision has no fiscal consequences according to current regulations. Bearing this in mind, we propose a modification of Gilboa and Schmeidler’s Case-based Decision Theory [3]. The latter defends financial choice in accordance with the similarity of the current problems to past experience and memories of previous problems and actors, and their consequences. The connexionist approach, Artificial Neural Networks (ANN) to be more precise will be used, to represent past knowledge and its influence on choice, traditionally estimated by means of an econometric approach. In particular, we work with Kohonen maps, which allow information to be organized according to the degree of similarity in the variables that describe each case or problem. Moreover, we introduce two substantial extensions to this analysis, namely (i) combining past cases/problems to generate new cases/outcomes within the general approach of the Theory of Evidence and Dempster’s Rule [4] (ii) an analysis of volitional uncertainty, both as regards the criteria for making a decision to be employed and the alternative chosen [2] [5]. In this way, we obtain a better approach to the representation of reality because we overcome standard paradoxes and the excessive simplification of reality that is produced by the optimization of expected utility as a criterion for rational choice. Moreover, we improve arbitrariness in the definition and estimation method for measuring similarity because we use an objective, neutral procedure to cluster past cases or memory. In addition, we explain the heterogeneity of the behavior or decisions observed not only on the basis of risk aversion but also by means of a different function of memory/similarity.