genetic optimization for technical analysis

Discussion in 'Technical Analysis' started by vivek, Jun 22, 2012.

  1. vivek


    i need some advice on how to go about using matlab to perform genetic optimization for a combination of technical indicators.

  2. Mysteron


    Well I use matlab to assist in some trading scans. Regarding genetic algorthms, a few years ago I set up a program in good old fashioned fortran for combining various traditional indicators to generate buy and sell signals for swing trading UK stocks. The fortran code included a genetic algorithm to optimise trading profit over previous end of day prices, essentially backtesting, but also I tried to make the method robust to allow it to work well in forward testing on new price data for which the GA hadn't been applied.

    It was a fun excercise but the results didn't bear fruit and I soon realised that TA itself was the problem. Just too hit and miss, sometimes it would generate profit an other times a loss. So now I don't use TA at all and instead use my brain as its more reliable. In fact I now daytrade US stocks.
  3. Very good post. TA is the problem. Not matter how advanced the GP algorithm is, garbage in - garbage out. But I would also liem to add that GP is nothing more that a process for selecting the best fit. This works well for species in biology and for solving some math problems but it is a terrible choice for developing systems.
  4. gen opt = curve-fitting = you become loser

    Several snake oil salesman out there with gen opt software fitting endlesly indicators to data. I know guys who got burned bad. Next time you hear the word "genetic" run, run, run...

    Soon they will show up here defending their snake oil.