Artificial Neural Network forecasting models: An application to the Italian stock market

In literature there is a growing interest about the predictability of the financial stock markets. The classical hypothesis that variations of stock prices are independently log-normally distributed has not been settled in an increasing number of empirical works. In this paper we analyze the daily returns of the Italian stock market index COMIT by using some particular non-linear and non-parametric models coming from the soft Artificial Intelligence approach, the so-called (feedforward) Multi-layer Perceptron Artificial Neural Netwoks, finding evidence of predictability. In particular, we obtain interesting results in forecasting the trend and the signum of the analyzed financial time series.