Which parameters to tune when developing automated strategies?

• 03/09/2023 at 10:28 AM #211183
  Lavallette

  Dear Fellow Traders,

  I would like to invite your expert opinion on assessing the tradability of a system. I understand that each trader is unique and may have varying capital and risk tolerances, making a single answer impossible. However, I am curious about the criteria you use to backtest your system.

  Personally, I develop my system using a limited history. If it performs well, I extend it to 100K bars and optimize it if necessary. Then, I test it on longer histories.

  Initially, I check the % gain, % winning trades, and total gain. Subsequently, I examine the regularity of profits made over weeks and months.

  However, I still have some questions that I hope you can help me with. For instance, how do you balance the criteria when evaluating a trading system? Should one prioritize a higher win rate or a better W/L ratio? What are the trade oof ? etc…

  Although I researched a bit, it was challenging to obtain insights into the criteria and methodologies that bot developers use. Would you be willing to share yours?

  Thank you for your time and expertise.

  

  Lavallette

  Participant

  Topics: 8

  Replies: 23

  Been thanked: 3 times
  03/11/2023 at 11:07 AM #211356
  PeterSt

  Hi there,

  Lavallette wrote:

  how do you balance the criteria when evaluating a trading system? Should one prioritize a higher win rate or a better W/L ratio? What are the trade oof ? etc…

   

  Over time I varied these matters. From my experience :

  A highest gain is not the thing to go for. This most often ends up in infinitely long running trades, ending in the SL after all.

  Similarly, a highest win rate is not the thing to go for.  Notice that I can show you win rates of over 98% long term. This looks nice, but ends up with the lowest gain. This is because with this a a target, the SL will be relatively high and it is virtually the same “issue” as the previous topic.

  A system with too few trades, is not the thing to go for, because you will have too few backtesting experience. And No, using 1 million bars on anything that is a 1 minute TimeFrame or even more, is useless because too many things change underway. I use 6 months of backtesting data, though that is a bit arbitrary (because it is 200K of 1 minute TF). Still I like that.
  Notice that “too few trades” means a few per day, to me. I work with 20-30 trades a day at least. This is an explicit target.

  A ratio of 1 between SL and TP appears the “nicest” thing to have. This is not a target because it won’t bring sufficient gains (one could attest that this will be zero in that case).
  In practice I now end up in the lower 60% of winning trades. Not a target, but that comes from the systems I am satisfied with and which win for the general flow. Say 63%. However, all kinds of aftermath tweaks may lift that to around 68%. They comprise of technical means, like an again better trailing (the Exit).

  Implied 63-68% does not tell a thing about the gain. So throughout time a formed a target for that, and this is more explicit. This starts with what one can achieve (but with the wrong basics, see above) but what’s “In it” because you have the experience of it. This is somewhat harder to express because the (backtesting-) figure is always based upon an entry amount of 200K in my case, but from my other post from early this morning I see the figure still on screen, you see below (2nd attachment – the 1st failed). That says 14,73% with 200K as a base – which is over 6 months. The real figure is the profit of 29454, with the knowledge that the contract amount is 50000 here. Now, I always use a base of a known amount, which in my case is 20000 (the 50000 used here is just on screen). Thus, 50000/20000 = 2.5 and 29454 / 2.5 = 11782.
  The 11782 is for me relative to what I always strive for, also knowing that this very system delivered ~ 7000 one month ago. But it is also my base. Hence, it was this a year ago just the same. And two years ago. If this is not in there, the base of the strategy is wrong and I won’t continue with it. Thus what I am saying also is : with that reference base, a system is apparently able to grow from 7000 of gain to 11782 by means of backdoor tricks. Each week new ideas.

  In the end this obviously means that with 20000 as investment and a revenue of 2x 11782 = 23564, this means more than 100% revenue in one year. This is without re-investment of gains.
  The way to get there is in the first part of this post. 🙂 Would those “rules” be disobeyed, then for example underway I would lose almost all because of a too large Drawdown somewhere. Obviously this is all about how the future can be dealt with (if we’d only know that – hehe).

  Last thing is : because I have the base of it all working to satisfaction, I test things for reality on Live (not Demo, not forward backtest). Only from that you will learn the best, which includes letting a system stay in when it is losing. Of course we do this with the lower amounts (different for everybody indeed). This also means that it requires some money of which you will know you will lose it. It is part of the job to get there. Just calculate it in.

  Now you know (about) all. 🙂 🙂
  Peter

  1 user thanked author for this post.

  Lavallette

  

  PeterSt

  Participant

  Topics: 73

  Replies: 2501

  Been thanked: 805 times
  03/23/2023 at 8:31 AM #211965
  Lavallette

  Hello Peter,

  I’m just back from vacations and found your very detailed answer. Thank you very much for taking the time to answer and very happy to see that some are making good money.

  Cheers.

  

  Lavallette

  Participant

  Topics: 8

  Replies: 23

  Been thanked: 3 times
• Author
  Posts