Forecastability of Returns with Neural Networks: An Application to Spot and Futures Italian Bond Markets

We discuss the possibility of applying neural networks for the analysis of financial markets. We consider simple forecasting and more complicated devising of trading rules. The former can be performed with a single neural network or with a combination of neural networks, while the latter can be performed with an unsupervised learning methodology that applies genetic algorithms to neural networks. We apply these techniques to the Italian bonds spot and futures markets. We find few signs of predictability out-of-sample, with a high (more than 30%) degree of explanation in-sample.