Advice on optimization, specially lookback period.
? Should I consider updating parameters every week
you need to study structures and think about the operating methods Hedge fund uses
To give you a basic idea, you may find many methods which works on short term buy and sell (scalping) but the main point here is about Exit. You can enter on a trend confirmation but you only exit when you find a signal of trend failure, If this trend is ongoing for 1 day, your period is 1 day, if trend is ongoing for 3 months, your period is 3 months, so a period can not be fixed or a number to give you which will perform best. It totally depends on how market reacts. but if you understand structure well then you will solve 70% of your problem yourself.
you need to study structures and think about the operating methods Hedge fund uses
To give you a basic idea, you may find many methods which works on short term buy and sell (scalping) but the main point here is about Exit. You can enter on a trend confirmation but you only exit when you find a signal of trend failure, If this trend is ongoing for 1 day, your period is 1 day, if trend is ongoing for 3 months, your period is 3 months, so a period can not be fixed or a number to give you which will perform best. It totally depends on how market reacts. but if you understand structure well then you will solve 70% of your problem yourself.
Hy, thanks for your reply. Do you mean to consider eliminating parameters and placing sl and tp at market structure? According to my data, it seems like short term periods don't work, especially I don't have enough data to justify a modification, which could be the reason why short periods don't work well. It seems I need to test more to find the right answer for my specific situation.
Hy guys!
I would need some advice and opinions how far to lookback when optimizing parameters and how often should I optimize? Should I consider updating parameters every week or even day, to include most current data
or is it enough to do it on a monthly basis? Logically I would assume, the more recent data it includes, the better the performance could be. I tested 1 months, 3 month and 6month lookback so far and 1month performed bad, 3-6 months has a similar good performance. Do you guys use the formula of Jezev and Shumsky to base the lookback period and parameter choice on your avg trade qty and the amount of parameters to tune?
If you are not able to analyse the market trends properly, you would be in danger. This is not an easy topic and it can take several years to develop an understanding of trend structure. It is also a subjective manner.
One thing you should do in optimization is out of sample tests. You can do forward tests and see the results. This will help you to have a consistent version.
As a rule of thumb, your optimization period should include
1- Mild trends,
2- Consolidation zones,
3- Breakouts or breakdowns combined with persistent trends.
If you optimize the algorithm within such periods, the algo will train almost all possible scenarios to go forward. Then with out of sample tests you would see the consistency.
This approach of course is very basic but powerful. Optimization is a difficult topic and it is like a gun. Basically, a blindly conducted optimization will create an illusion about reality and you will hit yourself with that gun.
Good luck.
Do you mean to consider eliminating parameters and placing sl and tp at market structure
If your strategy is able to find a perfect entry then you should already know SL but TP should not be fixed, you can use ATR, trendline breakout opposite of your direction, verify structure for trend reversal signal and consider exit when there is strong reversal. If you watch one period and code your strategy based on that, there may be rare chances of success as per my research. If market is in a mood to give you 1:50 risk:reward basrd on period of 200, why you would book 1:2 on period of 13 for example.
If you are not able to analyse the market trends properly, you would be in danger. This is not an easy topic and it can take several years to develop an understanding of trend structure. It is also a subjective manner.
One thing you should do in optimization is out of sample tests. You can do forward tests and see the results. This will help you to have a consistent version.
As a rule of thumb, your optimization period should include
1- Mild trends,
2- Consolidation zones,
3- Breakouts or breakdowns combined with persistent trends.
If you optimize the algorithm within such periods, the algo will train almost all possible scenarios to go forward. Then with out of sample tests you would see the consistency.
This approach of course is very basic but powerful. Optimization is a difficult topic and it is like a gun. Basically, a blindly conducted optimization will create an illusion about reality and you will hit yourself with that gun.
Good luck.
Thanks for your reply Evren! This absolutely makes sense. Would you include new data as fast as possible and roll the optimization window forward? (f.ex daily, weekly) or is monthly enough? I am getting good results when I do it monthly, however it somehow makes sense to do it weekly or even daily to include new data as fast as possible or would this be too responsive? I have to test this but also don't want to go down the rabbit hole. Would be great if you could share your experience.
- Free trading apps
- Over 8,000 signals for copying
- Economic news for exploring financial markets
You agree to website policy and terms of use
I would need some advice and opinions how far to lookback when optimizing parameters and how often should I optimize? Should I consider updating parameters every week or even day, to include most current data
or is it enough to do it on a monthly basis? Logically I would assume, the more recent data it includes, the better the performance could be. I tested 1 months, 3 month and 6month lookback so far and 1month performed bad, 3-6 months has a similar good performance. Do you guys use the formula of Jezev and Shumsky to base the lookback period and parameter choice on your avg trade qty and the amount of parameters to tune?