ProbTrader - 24-7-2018 at 06:45 AM
Hi all
I thought I would start this thread to see if anybody has had success with systems on individual stocks (I have not). One thing that my intuition
isn't able to explain is why it seems so much harder to build great systems on single names vs indices. The answer, it seems, lies somewhere in the
difference in distributions on returns.
Is it so that the smoothing of outlier returns in indices make them more suitable to systems building? And why is that? I can make all sorts of "after
the fact constructions" that could explain this, but quite honestly, my intuition still says that it should be no harder to build systems tailored for
individual stocks given the price action they exhibit. And I'm obvisouly wrong.
If you take the beta effect out of the equation, alpha seems to be tough to acquire. It is "relatively easy" to build systems on names that trend, if
you go with the direction of the trend (long only for bull markets/short for bear). But then you're just collecting beta "after the fact" out of an
already known trend.
It would be interesting to your thoughts about this. One of my hypothesis for this is that it is key to derive indicators not only from the individual
name you are building systems on, but also from correlated names. Still playing around with that idea. Admittedly, so far, without much success.
Looking forward to hearing your thoughts.
rws - 24-7-2018 at 02:09 PM
When searching for succesfull ohlc patterns I also see it works much better OS when optimizing on an portfolio of all nasdaq stocks instead of the
nasdaq100 index
andrew.gibbs777 - 25-7-2018 at 12:44 AM
Swing trading systems work very well on equities with GSB. Long only systems are better as they take advantage of the natural upward bias in stocks.
Set day trading and market on close to False, hit play, GSB should come up with something. Perhaps also check the filters are not to stringent. I use
Pearson set at 0.9 and profit factor above 1.3 as the filter for systems, then you get a lot.
Try on an individual name initially and use other similar names to verify the system. I also like to build on say 10, 15, 20, 25 and 30 minute
timeframes to get a system. If that doesn't generate anything reduce the number to say 15, 20 and 25 minute and add 10 minute and 30 minute timeframes
to the verification list.
With trending names like AAPL, FB, AMZN etc GSB will build a system that fits all of them so you can use all of those datasets in the primary data
field..