Topic: Number of combinations of strategies


You can set the EA Studio to search for strategies for a specific amount of time and the progress metric shows how much % of that time you have used up. But, is there a way to know how many actual combinations might be possible with your set up and how long it would take to work through them all?

Obviously, there is going to be a massive number of combinations and, depending on the speed of the machine, this is not going to be practical for most people to know. But how about if you had a very fast machine?

Also, Im assuming that the combinations that the software goes through are random and not incremental, i.e. the longer you leave it running, the more likely you are to get a good result. Thinking about this, is there an optimum amount of time that the software should be kept running to find a decent quality of strategies?

As an example, if it takes 2000 hours to go through ALL the combinations, you probably dont need to let it run for that whole time - you might cover a reliable % of that in 500 hours that would make the remaining 1500 hours more or less moot, especially when the portfolio approach takes up the slack of any poorly performing strategies.

Im interested to know, do most successful traders set it to 9999 minutes and leave it for a few days, e.g. on a VPS then just go with that?

Thank you,

Re: Number of combinations of strategies

Let me give you a quote:

If you have one MA with Period, Shift, Smoothing method and base price, the combinations are: 200*200*4*7 =  1120000

For one MACD indicator you have 1344000000 combinations.

If you have MA and a MACD, the combinations are: 1120000 * 1344000000 = 1505280000000000

It doesn't hurt to play with math, you have an example before you, do it for all your indicators. I guess it is not hard for you to estimate how fast your machine makes calculations, and from that point onwards you can arrive to a reasonable estimation, which might answer your question.

Re: Number of combinations of strategies

Thanks, yep - so that is more than huge so it doesnt seem worthwhile to know the info. Either that, or we have to contact NASA and use some of their machines.

So, is there a minimum timeframe that you use to find a strategy? How long do you let your machine run for (or are we not so concerned about timeframe, rather just passing the backtest/WF and MC tests)

Thank you.

Re: Number of combinations of strategies

The process is random, so on one hand it doesn't matter how long the generation process takes, but on the other hand the longer it runs, the more higher yield strats are found (the net balance threshold (or the chosen criteria) rises, so theoretically fewer strats make it to the collection). Whether it's a good thing or not depends on the workflow and methodology, in other words how well can bad apples be separated from the gold nuggets. And whether your working method relies on bigger number of strats or not.
Given the data mining nature, one needs MOAR strats most of the time smile Speaking from a personal experience - I don't let it run for multiple hours, 1-2h is fine to me, sometimes even less. In the end the question of generation length is trivial if you can't develop a workflow, which produces effective and profitable strat portfolios.

Re: Number of combinations of strategies

Hi matthew,

You mean a brute force optimization, just like MT4 optimization with genetic algorithm unchecked right?

do or do not there is no try

Re: Number of combinations of strategies

Thank you, Footon - that is exactly the answer I was looking for. My approach would be to get a minimum of say, 30 strategies, and while those are running (demo) then continue to look in the backround for better performing ones and replace them as necessary until we find something close to the risk/return profile that is acceptable overall. But its good to know about the 1-2 hours approach, that would suit me. Once I actually get the software and subscribe to it, I will be testing some different approaches in more depth. I hadnt hear of MOAR before (Im assuming you mean the Michael O'Higgins Absolute Return?) - fascinating and worth more research. I think the key is progressive elaboration and continual improvement - you will never (sic) test ALL the combinations so have to work with what you have.

Yonkuro - Yes, exactly. In other words, with a huge number of combinations, it seems not possible to test everything, so we need to dip in and out the possibilities with the means at our disposal. At least, a random approach will decrease the chances of similar strategies - I never had much luck with the genetic approach.

Re: Number of combinations of strategies

Hi Matthew,

Once I calculated how much time you need to find "all" combinations. It appeared that the universe will not last so long.

You may think that "they" throw us so many indicators just to not be able to find the best combination. However, we work hard to outsmart "them" smile

Re: Number of combinations of strategies

Popov wrote:

Hi Matthew,

It appeared that the universe will not last so long.

Thanks Popov - so many trades, so little time!