A bridge between statistical learning and agent based modelling in stock market predictions

Abstract

The research related to agent based modelling and statistical learning has increased in the recent years, as alternatives to main stream econometrics. This project combines the three for constructing and evaluating a forecasting experiment of capital markets. The model is applied on the stock indexes S\&P500 and FTSE100 in a time period of 15 years. The performance of the model is measured in economic profits, where transaction costs are taken into account. The model produces a daily Sharpe ratio that outperforms the buy-and-hold strategy on the FTSE100 index to above 4 BPS in transaction costs. On the S\&P500 index, the model underperforms the buy-and-hold. The model is capable of capturing gains in different market regimes on both indexes.