This preview shows page 1. Sign up to view the full content.
Unformatted text preview: wn volumes, bond rate etc. 1 Prediction using past stock values only
This first category is based on the theory that all the information available regarding any
economic indicators is already contained in the time history of the index and future index
values are dependent on data upto the present moment. Some applications of this category
are discussed below.
Dual module neural network
One application  uses a dual module neural network, with one module learning the long
term trend of the market and the other module trained to learn the short term rate. The
data used is a 4-tuple of index values, namely the high, low and closing values and trading
volume for each day. By taking moving averages and variances, this set of 4-tuples for all
days are combined into a 16-tuple which is level insensitive and time-invariant. These
extracted features are then used to train the two networks. The long term net uses a
training window spanning the last trading quarter while a 12-day window is used fo...
View Full Document
- Spring '12