Unformatted text preview: nd averages are all normalized using eq. (1), with 10 different values for A (A = 0.1 for the index and the averages and is = 0.02 for the
absolute value of the 3 first order differences).
The interest rates fluctuate about a certain value and there is no cumulative or time
integral effect on their behavior, hence, the interest rates are not detrended and these are
normalized using simple linear normalization. The same applies for the absolute difference
of the interest rates. All the trends have a value of either 0.8, 0.2 or 0.0 and as such don’t
need further normalization. This completes the data normalization for the neural network.
The same techniques and constants are used for any other range of training and testing
data. 4. Test results
The neural network was trained using standard backpropagation for 2 different test cases,
which are of critical interest from the point of view of desired performance. Fig. 1 shows
the variation of the S&P 500 index and is reproduced below for ready reference. It can be
seen that there is a big market crash around the 730th data point (October, 10th, 1986) and
there is a strong bull market beginning around...
View Full Document
- Spring '12
- Neural Networks, Artificial neural network, Stock market index, neural network