Since we have been selling and using the NeuroShell Trader, we have learned quite a bit about financial forecasting. This is knowledge we would not have found out in years with our more general purpose software; the Trader is so much easier and faster that experience comes quicker. I am going to share with you, our NeuroShell Trader users, what we have learned. Please keep what I am about to tell you confidential.
Basically, we have learned that financial forecasting in not as hard as we used to think it was before the NeuroShell Trader. Following a few simple guidelines, we have built many successful models. Here are the guidelines that I personally have used (there are certainly other successful techniques that others have followed. I am by no means as good as many of our users):
1. Pick a volatile instrument to predict; one that repeatedly rises and falls is the best way to phrase it. Don’t pick something that has only gone up in recent memory. This is the most important thing you can do; I can’t emphasize it enough.
2. Pick an instrument that is rising and falling because of trading, not fundamental factors, if you want to try to predict with technical indicators.
3. Predict the percent change rather than the change or the actual price. It is more normalized.
4. Use only about three to six indicators as input. This is the best way to prevent over-fitting problems.
5. Make sure that your inputs are all as different as possible when graphed. Don’t pick an indicator and 4 lags of it; they will be too similar.
6. Which of our 700 indicators should you pick from? I’ve had good luck with the ones in the Price Momentum category, and to a lesser extent the Volume category. I rarely go outside these categories, except some occasional experiments with acceleration, velocity, and the Regression category. My view is, if you pick a predictable issue, you can be successful with lots of different indicators.
7. Then do your experimenting to find the answer to:
a. How far ahead to predict? (I like 1 to 3 bars).
b. How long should the training set be? (I have had the most success with 2 to 5 years if daily bars).
c. How long should the evaluation set be? ( I like 4 to 6 months).
8. Don’t look for perfection; have realistic expectations and don’t wait for the biggest wave to surf. If you are too analytic about it or too much a perfectionist, you will never learn to surf. Get some reasonably good nets and just get going! Start out with stocks (long at first) and only 100 shares; most of you won’t be risking your life’s savings. All surfers crash sometimes and nets will fail, but I’ve lost much more listening to my own intuition or advice of brokers. I am still burned about the time in early 1990 that I wanted to invest $20,000 in Microsoft and I let my broker talk me out of it. Figure out what that one cost me!
9. Make your trades on the internet to keep costs down. If you don’t pay large commissions, you can take advantage of even small moves.
10. Once you start trading, don’t be afraid to do more research or go beyond our examples. Look into better things to predict, better trading strategies, and multiple nets.
Let me know how you are doing.