|
PureBytes Links
Trading Reference Links
|
Back in 1993 I developed what I felt was a reliable NN system for
trading stocks. I wrote an article that appeared in the June 93
issue of TAS&C which described my system. I also published a
newsletter that applied the system to trading stocks.
My work was cited in Barron's and another publication (a commodity
publication whose name escapes me) interviewed me regarding my
prediction of a market decline in 1994. They made a big deal that
my Neural Net was as good as a crystal ball.
For a variety of reasons I stopped publishing the newsletter and a
year later I stopped trading the system.
With that said I have the following observations:
I have some data that shows that for some stocks the NN accurately
predicts stock price gain 5 weeks into the future. The plot of
actual and predicted gain fall on top of each other. I believe that
this cannot happen by chance. With other stocks it a bit more
ragged and for others it plain doesn't work. As hard as I tried, I
couldn't figure out what was the discriminator in this. Another
problem I had was translating predicted price gain into trading
signals. Had I been using TradeStation at the time, I think I could
have solved that problem.
Someday when I have more time to devote to this problem I may dust
off these old programs and fire them up again. Since I have seen
some good results from fuzzy logic I have to agree with Pierre that
this could be a dynamite combination.
Marlowe
> Date: Mon, 15 Mar 1999 13:17:44 -0300
> From: Peter Hoon <phoon@xxxxxxx>
> To: "'omega-list@xxxxxxxxxx'" <omega-list@xxxxxxxxxx>
> Subject: Do neural nets work?
> Would appreciate comments, opinions, criticisms, personal stories on any one or all of the following:
>
> 1) For those who have used neural nets that are integrated into trade station: To what percent does the use of a neural network improve the net profit of a trading system above and beyond that whic>
> 2) Does the N_train software, or other neural net software, have problems which cause ts4 to crash when you run the net and its accompanying dll on real time data for inta-day day trading?
>
> 3) How do you get from a neural net to an empirically derived set of rules and a trading system that is useful in day trading?
>
> 4) What empirical data exist to show that neural net software added to
> ts4 actually produces improved bottom line day trading above and beyond what you can accomplish in EL code and existing ts4 (or ts 2000i) capabilities?
>
> 5) Are the hassles of data organization, pre-processing and massaging with neural nets worth it?
>
>
|