How many months of Historic Data are RELEVANT for back-testing and optimizing an EA that uses M15 timeframe?

New comment
 

Hi,

Given the current weakness of both EUR and USD, and the subsequent volatility of the market, it would seem to me that backtesting & optimizing an M15 timeframe-based EA on any more that the most recent 12 months of Historic Data would be a complete waste of time.

I would like to know what experienced, successful, EA-using traders have to say on this matter. All info and recommendations greatly appreciated.

Thank you,

~DV

 
It Depends. It Depends. Upon allot of different Variables. Your question is too broad. Try looking at the Statistical Behavior of the system.
 
dark_voyager:

Hi,

Given the current weakness of both EUR and USD, and the subsequent volatility of the market, it would seem to me that backtesting & optimizing an M15 timeframe-based EA on any more that the most recent 12 months of Historic Data would be a complete waste of time.

I would like to know what experienced, successful, EA-using traders have to say on this matter. All info and recommendations greatly appreciated.

Thank you,

~DV


I agree. For backtesting - I use since 2002 to at most 2007-8. Not even the past 2 years. It can be a far, far cry from what is happening at the moment.

I haven't much success with any particular trading on just a eurusd, but a lot of success doing hedges with eur's, usd's, chfs. Quick, fast ones. Cant say for certain in coming months. That I know now is the time for my balances to diversify, exclusion of holdings, etc, all in all, just play darn SAFE.

Last week, 5 eas manage to net in only 10+, over 30 frigging losses, luckily, just plain ol lady luck grabbed me 6 last minute winners each on a double barrel. Worse things may seem to happen. So, I rather wait longer than I thought.

Thank You.

 

Depends:

1. How many signals are you generating in a day/week/month?

2. Using an equal TP/SL for example 20/20, whats your win loss ratio?

3. using optimized TP/SL whats your perf/DD ?

First of all smaller TFs (those below H1) tend to flow in and out of phases. I doubt you will find a consistent pattern that lasts more than 3-9 months. On the other hand, if you can tune in your EA to the current 3-9 months, well then you may have something useful. Problem is dont expect it to last forever. Chances are it wont. Be flexible and constantly retune your parameters.

Some Suggestions:

1. use as much history as possible and look for patterns in the performance. Use these patterns to find time frames that are stable in recurring patterns.

2. find a recurring time period whether its 3 weeks, 3 months or whatever, where in a set of parameters is stable. this will help you find a time frame where you can tune your parms to.

3. Dont over optimize your parameter set. find a compromise between the stable time frame and a longer one such as a year or 18 months.

 
hkjonus:

Depends:

1. How many signals are you generating in a day/week/month?

2. Using an equal TP/SL for example 20/20, whats your win loss ratio?

3. using optimized TP/SL whats your perf/DD ?

First of all smaller TFs (those below H1) tend to flow in and out of phases. I doubt you will find a consistent pattern that lasts more than 3-9 months. On the other hand, if you can tune in your EA to the current 3-9 months, well then you may have something useful. Problem is dont expect it to last forever. Chances are it wont. Be flexible and constantly retune your parameters.

Some Suggestions:

1. use as much history as possible and look for patterns in the performance. Use these patterns to find time frames that are stable in recurring patterns.

2. find a recurring time period whether its 3 weeks, 3 months or whatever, where in a set of parameters is stable. this will help you find a time frame where you can tune your parms to.

3. Dont over optimize your parameter set. find a compromise between the stable time frame and a longer one such as a year or 18 months.

Hi Jonus,

Thanks to you (and others) for your comments. Your point about the difficulty of finding a consistent pattern that lasts more than 3-9 months is exactly in line with my thinking, and the reason I started this topic in the first place.

I have a couple of questions about your final suggestion:

1. I'm always reading advice about not over optimizing, but it seems that this is a very subjective thing.

My EA uses 8 parameters. On the first optimization run, I use the "Control Points" model, with the Genetic algorithm switched on, and the full range for each parameter, often giving over 300 million possibilities. I usually repeat this process for three runs, each time narrowing the range of each parameter, so that I end up with approx 600,000 to 1 million possibilities. At this stage, I continue with Control Points, but switch the Genetic algorithm off, running through an entire set of maybe 40,000 to 80,000 possibilities. I then analyse the results (in my own subjective way!) taking into account not only the final balance, but also the Profit Factor, the drawdown % and the Profit Trade %. Finally, having selected the "winning" parameter set, I run the Tester again in the "Every Tick" model with a range for the parameters close to the winning set from the Control Points phase. Is this about right?

2. I don't understand what you mean by: "find a compromise between the stable time frame and a longer one such as a year or 18 months".

Can you explain what this means and how to go about getting this compromise?

Thanks again for your help.

Cheers,

~DV

New comment
Reason: