An Artificial Stock Market: Asset Pricing and Endogenous Expectations using Neural Nets

In this paper the relatively new technique of neural nets is integrated in a traditional model of portfolio choice. On the basis of Arrow’s State Preference Model the investment decision depends on the expectation building process which consists of two components. The individual information processing and the mutual influence upon one another. Therefore, each agent is represented by a single net but all individuals are connected with each other. On both levels the magnitude of impact for the final portfolio choice is reflected by the connection weights of the net. The aim of the heterogeneous agents is to learn the market structure in order to make forecasts of probable yield. By comparing the expected and the actual price the individuals adjust the weights according to the backpropagation algorithm. The simulation studies show, that the agents adapt to each other generating a decline in the total market error. Market entries can disturb this structure and induce erroneous forecasts of the remaining market participants. On the microeconomic level it can be seen that similar characters can profit from each other if some of them get a dominant market position.