[jaupe]

I cannot fault your enthusiam but I can fault your method :)

Read Tsay's Financial Time Series Analysis for a better idea how to forecast financial data.

The biggest problem you will face is stationarity: the statistical properties of the data is not constant over time. For example, the mean and std dev is not constant over time. Using returns instead of raw prices helps to make better financial forecasts.

Two methods to explore:

1. You are better off predicting stock prices by predicting future returns and then forecasting is the current price plus predicted future return.

2. You could use your neural model to predict absolute size of returns using realized volatility.