>I'm still confused on the influence period though. It almost
>seems like it might be a 'weighting' parameter.
In a sense I think it is. I think of it as just the number of inputs you give to the model. For example, if you say a 20 day short-term influence, it would use yesterday's data, two days ago, etc., until it gets to the 20 day mark. The more a model can see, hopefully the more accurate it gets. This isn't always the case though. Neural nets are part science, part art, and part luck in my opinion.
>I've only been running NS since the weekend, but so far it's give
>two timely sell signals and one bad hold (ie. should've sold).
You have to remember that NS will never work 100%. In the long run however, it should pay off. You may also want to 'paper trade' a system for a few months first, to make sure it works. And by creating multiple nets on the same stock and then using a consensus view of the recommendations, you better your odds. Try using different periods, influence settings, related stocks, etc. The more diverse the inputs, the better the chances of a good outcome.
Optim |
