gbarber Veteran   Posts: 282 Joined: 12/30/2012 Location: Pearland, TX  User Profile | I forgot how to paste pictures into the text of these posts so just attached a word document with the pictures showing results and 'method' used. Jim Dean taught me how to do that a couple years ago but I have lost that info. Anyway, the point is that a little experimenting with ATMS could bring some pretty outstanding results without much effort. However, we all know that even with the most thoughtful and best testing using historical data, the results achieved in the real world on the right edge do not necessarily mimic the testing results. The market characteristics change making all that testing less and less relevant as time passes. That's why adaptive is the key word. ATM gives us that on the large scale with use of different strategies per market state. It also allows some adapting on a large time scale via optimization of strategies used and their parameters. Perhaps doing that periodically would help. Need to test that. But I think we will need something to adapt to short scale change of market characteristics to conquer the right edge on a continuing basis. Perhaps periodic retraining of the AI strategies would help. Perhaps use of filters like Lain Bogle is developing for ATS would help. Is the seminar coming up this week going to teach us how to divide the historical data into one chunk of back test and a set of smaller chunks of forward test so we could see how different means of adapting to market change would work?  Attached file : ATMS results.docx (290KB - 297 downloads) |
