Evolutionary inference of rule-based trading agents from real-world stock price histories and their use in forecasting

Abstract

We propose a representation of the stock-trading market as a group of rule-based trading agents, with the agents evolved using past prices. We encode each rule-based agent as a genome, and then describe how a steady-state genetic algorithm can evolve a group of these genomes (i.e. an inverted market) using past stock prices. This market is then used to generate forecasts of future stocks prices, which are compared to actual future stock prices. We show how our method outperforms standard financial time-series forecasting models, such as ARIMA and Lognormal, on actual stock price data taken from real-world archives. Track: Real World Applications (RWA).