Anybody done a trial of running 2 (or more) duplicate Systems where one is left running and the other (same code) is optimised every week / month / whatever period?
So it would be a comparison type trial?
I’ve intended to for ages but never got organised enough, but I feel like I might be getting real close now! 🙂
My thoughts / logic is that so many members say that Systems work good around the time they optimised and then gradually performance reduces.
So regular optimisation may maintain performance levels around that which was seen during original optimisation. Worth a trial anyway?
It goes without saying (just in case some naysayer would have said it anyway! 🙂 ) that if the System is – for example – a trend follower and the trend came to an end just after optimisation then the regular optimisation approach is not going to bring good results.
Anyway I’ve got a scalper I’m going to optimise each week on a Sunday … I’ll let you know how it goes.
Hey GraHal. Ive been talking to former trading cup winner Kevin Davey about the same topic, and im paraphrasing here but this is what he said:
“I have had systems that have been working for 10+ years now without any optimization, and i have systems that require re-optimizing every 3-6 months. The only way to truly find out if your system “needs” re-optimizing is by doing a walk forward test with different settings.”
In other words if your wondering how often strategy A should be re-optimized, if it all, just do it in backtest. Take a 10 year data-set and re-optimize walking forward with 1 optimization per year, or 2, or 3. or 1 for every 2 year of data.
If the results are improving from re-optimizing then that is probably what you should do in real life as well.
Now here is my 2 cents: The less re-optimizing => more robust/simple code. The more variables to optimize => Less robust and might need optimizing more often. Logically if you’re trading at a pattern you believe is working for you, then that pattern shouldn’t change too much before becoming a different pattern than what you where originally trading. In other words when re-optimizing (if needed) you definitly should keep your variables within very very short ranges. if you used a rsi 14 originally, you dont want to be looking at a reoptimized RSI of 4 because that is just way too different to make a “small change”. Thats just my thoughts.
I also need to go over all my algos and see if i can find some sort of magic pattern to reoptimizing them.
Thank you @jebus89 for sharing Kevin Davey experience and your thoughts and views … all excellent advice … which I agree with.
Very Good Idea Grahal. I thought about this as well.
I will give it a shot but it requires a lot of effort to maintain the statistics
requires a lot of effort to maintain the statistics
Yeah I agree, easiest would be to leave the original version running and then the weekly / monthly optimised version can be run alongside and easily be compared on an ongoing basis?
Sod’s law the System I had in mind to re-optimise this weekend has not traded all week since I set it going last weekend!!!
EricParticipant
Master
Whats the difference between optimize,curvefit and machine learning?
LeoParticipant
Veteran
With machine learning you are always optimizing by curve-Fitting
😛
Whats the difference between optimize,curvefit and machine learning?
Hahah very astute question … what an erudite chap you are Eric! 🙂
And thank you for the answer @Leo, it kinda clears the air a bit, levels the black art of machine learning ( 🙂 ) and sets a baseline!
I can’t wait to optimise every week and come back with glowing results!! 🙂
I can’t wait to optimise every week and come back with glowing results!!
….or at least nice and steady losses.
