<Great minds think alike.> Ok, I think I'm on to something. Think influence settings!! When I first started training this model I started with 30 days for the s/t settings. The net trained slow, amd there was this damn "buy" in the verify that should have definitely been a "sell". To make a short story long... I forgot and retrained 3 more times lowering the settings (for the target AND the relateds) to 21, then 13, then 8 (love those Fibro #'s). Each time the net got better until finally at 8 days it has given me a greater profit and virtually eliminated most of the "holds" I had gotten both in the training and verify sections. The verify looks great. Jay, remember the accuracy of OMPT with those wierd settings? Could it be that by shortening the s/t day settings, the net is actually becoming more responsive to more current data?
I know that the documentation states that,in general, the higher the s/t influence setting the better, but that was written, who knows how many years ago, in a different market. Today's market is much more volatile and apt to turn on a dime, so maybe the s/t settings are inversely proportional to the volatility of the targets and relateds. The greater the beta the lower the s/t settings. Kinda sounds logical doesn't it? Things are getting "beta" all the time! :))
"use one of those index share stocks"
I'll have to give it a try... thanks! |
