Neural Networks for High Accuracy Forecasts

In USING NEURAL NETWORKS: AN ANALYSIS. OF METHODS AND ACCURACY, Jason E. Kutsurelis accomplishes some really amazing empirical runs, predicting 10-day returns of the S&P500 index. It seems incredible that he would make this awesome literature freely available to the world, and yet very few have yet to embrace the concepts successfully simply because it takes some brain labor.

Key points from his Neural Network model 10-day forecast tests:

Condition Probability of predicting a market rise: could reach 93%
Condition Probability of predicting a market decline: could reach 88%

When compared to traditional multiple regression, the NN outperformed it in several ways. Lower error bounds and higher accuracy.

Predictive daily variables used for forecasting 10-day future SPX closing price:
1) SPX Day-High Price
2) SPX Day-Low Price
3) SPX Closing Price
4) DWT, DOW Transportation Index Closing Price
5) DWU, DOW Utilities Index Closing Price
6) XOI, AMEX Oil Index Closing Price
7) CRB, Commodity Research Bureau Closing Value
8) XAU, Gold and Silver Mining Index Closing Value

A lot of mathematical or statistical software today have capability to do neural network work, SPSS and Matlab have pretty user-friendly interfaces. Lots of stuff to play with!

*Quick note, a lot of theoretical stuff goes behind the above, and some studies have found Support Vector Regression to outperform Backpropogation Neural Networks in financial forecasting. This area of research has still got a long way to go.