Jay,
You asked how my nets were doing. Well, I started out with really simple nets that just had the target stock with long, short, and volume and an industry index that is specific to the stock that had long and short terms. I constructed a dozen nets using this approach (mostly semiequip stocks) and have been training each one for 1 1/2 hours per day using the batch mode. The typical one has 9 or 10 hours on it now. I am pretty pleased with the way they look in the verification period, being pretty much solid long during the big runup in oct/nov and often catching dips in dec correctly. I am a bit puzzled as to why some of these stocks were out in the face of this big run-up last week in semi stocks though. I view being out as being the same as a short signal since the net is predicting that the price is going to go down.
One of my biggest problems right now is a lack of a systematic way to evaluate and compare nets made with Neruostock. I mean, we're given this confidence factor number. What does that mean exactly? It changes as the new data is added so it is a statement about the net's confidence of the current prediction. It doesn't really allow you to compare nets made with different stocks, and we really don't have a way to see how influencial a given input is on the net. Furthermore, we need to be able to export the net prediction as a function of time. We can see the prediction on the chart, but we need to be able to quantify the correlation between predicted and actual. I think I am pretty close to being ready to make some more complicated nets with more relateds to see if I can improve the net performance, (if I can figure out how to measure it).
I think I've decided to spring for BioComp Profit tomorrow. It gives much more detailed information about the net and allows one to see the relative influences of the inputs. It may take me a lot of work to get nets that perform as well as the Neurostock nets but it should be a good learning experience. Optim and others on this thread and the other thread have provided a lot of good input about Profit and Neuroshell Trader vs Neurostock.
You might not be able to tell from this posting of complaints, but I'm really pleased with my first toe wetting experiences with NN market prediction. The experience is probably the most valuable addition to my toolbox that I could get into. |
