What's your opinion on optimization? I need some advice. - page 3
You are missing trading opportunities:
- Free trading apps
- Over 8,000 signals for copying
- Economic news for exploring financial markets
Registration
Log in
You agree to website policy and terms of use
If you do not have an account, please register
Hi Rafael !
If you wishes to show its stability over the last x years to sell it as a ready to use for example, then you could do 3/4 + forward 1/4.
But in truth you won't need more than 6 months to match last market conditions. To repeat quaterly or as soon as you notice it slow down.
Maybe someone will answer you better than me.
Most articles I read about Walk Forward Analysis uses as example 10/1 - 10 months of optimization to 1 month of backtesting, but I think that's too much.
I'll probably do something closer to what you described. I just wanted to know how people chooses these periods.
I've also read about Walk Forward Matrix - a software runs several Walk Forward Analysis with different periods and then we choose the best one ... but to me it seems like this would be considered overfiting.
I also need to decide how many parameters I should optimize.
Most articles I read about Walk Forward Analysis uses as example 10/1 - 10 months of optimization to 1 month of backtesting, but I think that's too much.
I'll probably do something closer to what you described. I just wanted to know how people chooses these periods.
I've also read about Walk Forward Matrix - a software runs several Walk Forward Analysis with different periods and then we choose the best one ... but to me it seems like this would be considered overfiting.
I also need to decide how many parameters I should optimize.
Search for Strategy Quant, QuantAnalyzer is an advanced optimization tool, you should have a look at the free demo version, it's yet advanced.
Hey guys.
I've added several new features to my EA: It now calculates the distance from the last opened trade and calculates the lot size it should open. Also, because I use renko charts, it now considers the bar size to make some adjustments (each pair has an optimal bar size which is based on ATR).
Next week I'll create a demo account and create a free signal on it, if anyone is interested in seeing how it's performing you'll be able to follow it.
The purpose of this post is actually ask a few more advices about Walk Forward Analysis:
- I've read that OOS (ou of sample) data should not exceed 25% of IS (in sample) data: I'm thinking about using 12 months of IS data to optimize and run 3 months of OOS data (most examples online use 12/1).
- I've not come to a conclusion of how many parameters should be optimized. Some of you said the minimum the better, but I've talked to a guy that created a WFA software that said that it doesn't matter the number of parameters, all parameters that can have a big influence on the EA resutls should be optimized.
- After running the WFA, I'll get a number of results. How do you choose which one to use? The one that have bigger progit? Best Profit Factor? Select the 10 best results and make an average of the parameter values?
I really appreciate any advice you can give me.
Search for Strategy Quant, QuantAnalyzer is an advanced optimization tool, you should have a look at the free demo version, it's yet advanced.
About StrategyQuant, I've seen it, but for now I want to know the basics to run some tests manually. Maybe in the future I'll consider buying it.
About StrategyQuant, I've seen it, but for now I want to know the basics to run some tests manually. Maybe in the future I'll consider buying it.
:)
Then do whatever Rafael.
But in 10 years of backtest there's high & low, you won't be able to appreciate by seeing the 10 yrs curve, the risk is to be desesperated during few week once you start the EA. Why wouldn't you try it on a demo account first, just enough to gain confidence in this new system ?
Hi.
I've been thinking... it doesn't matter how many parameters I choose to optimize on each phase of WFA, as long as I keep using the same model, the results should be consistent on every phase.
But I am getting an interesting result: the results are almost the same independent of the values of the parameters. This is a 1 year optimization on EURUSD - the first step of WFA (it only about 33% complete):
I am running a 1 year optimization because my plan is to optimize 12 months and run for 3 months (12/3).
But it doesn't seem to matter what combination of parameters is being used, the results are 97% on the expected range, 1% above expected and 2% below (with a few iterations that would mean account being blown without a DD% limit).
Of course when I say that it doesn't matter the values of the parameters, I have defined min. and max. values that makes sense (for example, bollinger bands period can be from 20 to 100, MAs can be from 5 to 30, etc).
From my tests, it seems the EA can run on any pair. The results are almost the same on any random period and any pair I run it.
I just created a DEMO account and started to run this EA. The EA is running on all majors and most crosses - currently 18 pairs. It will monitor the spread to filter (not open trades) if it's too high.
I am still making some final code adjustments, like defining how much DD% is acceptable and what to do if it reaches such loss (for example: it can close all trades or it can close the trade with the biggest loss and try hedging/averaging again - this would mean a smaller loss at the cost of risking suffering some consecutive losses).
One month, or two should be sufficient for training purposes.
Any sample size above that does not.... improve overall performance.