Back testing and curve fitting

New comment
 

Hi, I've been back-testing my EA over longer periods (1-4 years) in order to find a 'good' set of parameters.

I'm trying to avoid curve fitting by doing so, with the assumption that curve fitting can be minimized over longer time frames (I'm using a 15 min chart).

Curve fitting has been a problem in my back-testing and walk-forward analysis over shorter time frames.

I'm currently trying to develop an EA that is profitable whatever the market conditions/month/currency pair etc. used.

I hoping that someone reading this has tried longer term back-testing and used the optimized parameters live and can give me some advice.

Also what are your opinions/experiences using walk-forward analysis (four month back-test - 1 month forward)?

Thanks

Dave.




 

You must not forget the markets changes and therefore the optim. parameters!

2007/2008 was the USA bank crash, 2011 (I as far as I remember) was ths EUR-trouble when even smaller majors could move the EUR-price for some hours, now the EUR is very week and USD very strong => stronger trend then before.If you optimize under such conditions you might get into troubles if e.g. EUR turn into a sideways range.

 
Yes curve fitting and then by the time is done, the market movement changes.
 

There is another way to optimize parameters. Not curve fitted.

From the book "Trading Systems that Work" - Stridsman 2001 (pdf available in torrent 651DBAE1C71CC37B6274F6EADF5D95862BEBB595) [the book has no useful systems IMO]

Create list of trades: profit and all parameters (including symbol/TF) at the end of each optimization pass. See OnTester. Optimize on multiple symbols.

At the end, load into a spread sheet, and for each parameter: sort by parameter value, use value with highest ave(profit)/stdev(profit).

The result is each parameter is set independently of the others (not curve fitted) to generate the highest, consistent profits.

I'm going to try this approach next. FWIW

New comment