Forecasting Daily Returns: A Comparison Of Neural Networks With Parametric Regression Analysis

Abstract

Since the seminal work of Fama (1965), many researchers have found that the actual distribution of stock returns, for the USA market, is significantly non-normal. Our study is focusing on the examining stock returns predictability for the Hellenic market given some macroeconomic variables. The objective is to use the given information set to reach an optimal way for forecasting. Hence, two basic models for forecasting are examined; a multivariable OLS regression approach and a non-parametric neural network approach and we compare them, based on the minimum forecasted error. Then, the approach that gives the minimum forecasting error is selected. The results indicated that better forecasting approach between the selected two ones is the neural network regression, since it has the smaller mean absolute percent error.