The adaptive selection of financial and economic variables for use with artificial neural networks

摘要

It has been widely accepted that predicting stock returns is not a simple task since many market factors are involved and their structural relationships are not perfectly linear. Recently, a promising data mining technique in machine learning has been proposed to uncover the predictive relationships of numerous financial and economic variables. Inspired by the fact that the determinant between these variables and their interrelationships over stock returns changes over time, we explore this issue further by using data mining to uncover the recent relevant variables with the greatest predictive ability. The objective is to examine whether using the recent relevant variables leads to additional improvements in stock return forecasting. Given evidence of non-linearity in the financial market, the resulting variables are then provided to neural networks, including probabilistic and feed-forward neural networks, for predicting the directions of future excess stock return. The results show that redeveloped neural network models that use the recent relevant variables generate higher profits with lower risks than the buy-and-hold strategy, conventional linear regression, and the random walk model, as well as the neural network models that use constant relevant variables.