Data mining for financial prediction and trading: application to single and multiple markets

An alluring aspect of financial investment lies in the opportunity for respectable returns even in the absence of prediction. For instance, a portfolio tied to the S and P500 would have yielded a compound annual return in the teens over the last half century. Over the same period, a portfolio tracking the fast-growth economies of the Far East would have provided even higher returns. Previous researches in learning methods has focused on predictability based on comparative evaluation even these techniques may be employed to forecast financial markets as a prelude to intelligent trading systems. This paper explores the effect of a number of possible scenarios in this context. The alternative combinations of parameters include the selection of a learning method, whether a neural net or case based reasoning; the choice of markets, whether in one country or two; and the deployment of a passive or active trading strategy. When coupled with a forecasting system, however, a trading strategy offers the possibility for returns in excess of a passive buy-and-hold approach. In this study, we investigated the implications for portfolio management using an implicit learning technique (neural nets) and an explicit approach (CBR)