ABSTRACT
The importance of the latter is particularly evident for applications of an evolutionary / artificial neural network (E/ANN) under supervised learning, where the process of network training is based on a chosen statistical criterion, but when economic performance is generally sought. Addressing the issue of ANN topology dependency, simulations reveal optimal for financial applications network settings. The experiment establishes that training ANN with the performance surface optimised with genetic algorithm for directional accuracy, discounting least values or minimizing number of large errors generally improves strategies’ profitability. Only the degree of improvement over ‘efficient prediction’ shows some robust links with returns’ measures. The research demonstrates that the performance surface set-up is a crucial factor in search of a profitable prediction with an agent-based model. To model the turmoil in an economic system with frequent shocks, short memory horizons are considered optimal, as older data is not necessarily informative for the current/future state modelling/forecasting.
