Hi Matthew and Samuel,
I recently read through this post and decided to try the strategy you outlined in your 17/11/22 00:42 post. I have been retrospectively testing it, that is to say I run the reactors and validations as if todays date were Saturday 6th August 2022. I then pick the best 10 strategies as per your post and using the validator i see how they perform over the following 2 weeks, ie between Monday 8th August to Friday 19th August. Then I repeat the process as if todays date were Saturday 20th August, and so on and so forth. That way I can build a picture over time as to how this strategy works over a period of a few months. I have had some partial success with this, the results have definitely been more encouraging than another system that I had been using til recently. I had some extra questions and wondered if you could help Matthew, or Samuel (as I know you had been providing advice to Matthew).
Do you only run one reactor with the 20 – 200 TP / SL settings?
If you run more than one reactor what other settings do you use?
How long do you leave the reactor(s) to run for?
What collection size do you set?
Your instructions state 30 months is the initial data horizon, then 2 months for validation 1 and an additional 1 month for validation 2. @Samuel I noticed in one of your replies you suggested to Matthew that validation 1 should be 3 months minimum, @Matthew have you done that and if so can you comment on whether 1 month or 3 months work better for validation 1?
Also re the initial 30 month data horizon, my broker (IC Markets) data only goes back to 1/8/2020 for 1HR, which means the earliest 30 month period I can input to generate strategies is 1/8/2020 to 31/1/2023. 1 months for Val 1 is 31/1/23 to 28/2/23 and add 1 more month for val 2 is 1/3/23 to 31/3/23. Which means I cant really try this strategy using my own broker data until the end of this month. I have experimented with a shorter initial data horizon of 15 and 18 months (and lowering the minimum trade count accordingly in acceptance criteria) but the results have been mixed. So do you both feel that the long initial data horizon is fairly critical to the success of this system?
Also Matthew in your post you mention you use the Premium data set to generate the strategies over the initial 30 month period, but then you must load the 10 successful strategies onto an MT4 or MT5 account which will use your broker data? I assume you find this still works fine? My instincts usually say I should only load strategies generated using IC Markets data on an IC markets account, because its the same data so you should be able to have confidence that strategies that perform on a backtest of IC Markets data should perform going forward on IC Markets data? But then I realised you generate strategies using the Premium data and must load them on an account using a different broker data set and you are still getting results so presumably you don’t see this as a problem? The Premium data set starts in 2007 so certainly I could do much longer range historical testing with this data set.
I should also check – do you do val 1 and val 2 using premium data or using your own broker’s data.
Thanks for your time and sorry about all the questions, but it really sounds like you are onto something so I am keen to exploring and running my own tests with this system.
Thanks,
Steve