Predicting Equity Returns from Securities Data

Our experiments with capital markets data suggest that the domain can be e ectively modeled by classi cation rules induced from available historical data for the purpose of making gainful predictions for equity investments. New classi cation techniques developed at IBM Research, including minimal rule generation (R-MINI) and contextual feature analysis, seem robust enough for consistently extracting useful information from noisy domains such as nancial markets. We will brie y introduce the rationale for our minimal rule generation technique, and the motivation for the use of contextual information in analyzing features. We will then describe our experience from several experiments with the S&P 500 data, illustrating the general methodology, and the results of correlations and simulated managed investment based on classi cation rules generated by R-MINI. We will sketch how the rules for classi cations can be e ectively used for numerical prediction, and eventually to an investment policy. Both the development of robust \minimal" classi cation rule generation, as well as its application to the nancial markets, are part of an