Maximum number of lines of code in a strategy?
- This topic has 17 replies, 4 voices, and was last updated 5 years ago by
.
-
I lowered my expectations of my ability to endure repetitive coding and have now coded a strategy of only 9623 lines! I wish we had arrays but then from discussions elsewhere on here I don’t think that the PRT engine could cope with the loops needed to fill the arrays with values anyway – so repetitive coding it is for now. I have now found out that find and replace is horribly slow when you try to replace thousands of items at a time!
So I now have a strategy that checks 400 combinations of variables x and y (10 to 200 in steps of 10) in a simulated strategy to see which combination of x and y has been the most profitable up to this moment in time. It then uses just the best set of variables in a very simple mean reversal strategy on the EURUSD Daily. First test results seem promising:
It appears that the values of 40 and 30 have been the most profitable since about mid 1995 which highlights to me that just looking at all time profitability might not be the best way to select variable values. So I am now working on a modified code with a third variable (p) in a loop that should allow me to base the decision on what look back period combined with what x and y values has been most profitable recently. Unfortunately this means modifying 400 different sections of code! I imagine the resulting strategy will be horrifically slow to compute as it will possibly be calculating 80000 possible combinations for every bar so I may have to lower my expectations again!
Yes – but I am starting with a strategy that is already pretty good and just trying to make it a little more robust and adaptive rather than starting with some conditions and seeing if machine learning can turn it into a profitable strategy. Really I’m just re-optimizing at each bar close. It may turn out to be more profitable or more robust or it might just leave me a bit out of synch with the market! I’m just over 50% through modifying my code at the moment – crashes allowing I hope to be finished and tested tomorrow.
1 user thanked author for this post.
I was able to finish the code today and have even managed one or two test runs. It is very slow to calculate a back test and the fact that it is Saturday and the bar behind our boat that I am ‘borrowing’ wifi from is very busy so the internet keeps dying on me and so the tests crash to a stop isn’t helping! Any how I have managed to see some results – I won’t post any images at the moment as I am still trying to get a feel for how this 10000 line ‘monster code’ works and to get an idea of what the results are actually telling me.
So far I have the datum/baseline of what seems to be a good strategy that is profitable over a broad range of two x,y variable values. The best set of optimized variables in this baseline strategy has so far beaten every other test with self auto optimizing strategies that I have carried out so far – and the equity curve of this base line strategy has also been the smoothest of them all so far.
I also have the results of my second test strategy which has two variables self optimizing between the values of 10 and 200 step 10 and only trading using the pair of values that has in all of history so far been the most profitable. This seemed to settle on x, y values in 1995 and then never change them after that. The third test was auto optimizing x, y and with an auto optimizing look back period of p to select x and y values in whatever p length look back period has achieved the best gain/per day with the most profitable x and y values for this look back period. The fourth test was to try various fixed look back periods of p in this last strategy to try to see if just one p value was better than a self optimizing p value.
Hopefully I will come to some conclusions soon. My first impressions are that the original strategy is too good! Auto optimizing multiple values seems to make it less profitable in the past although still a good strategy. Only forward testing will tell us if each new development is more robust than the baseline strategy. My first thoughts are that I am just automating the optimizing and it is a bit like doing a mini walk forward test and then using the last set of values that worked – but just because they are the last values that worked does not mean that they will work in the future any better than some other fixed value. Interestingly when I go from auto optimizing two variables to auto optimizing three the performance gets worse which sort of confirms that more complicated is probably more curve fitted whether that is manual curve fitting or auto curve fitting.
1 user thanked author for this post.
Find exclusive trading pro-tools on 
