This is outstanding. Aside from the obvious quality of the content, there's also a "suspense" component: "How will it all end?", which is keeping me so interested and watching these videos.
Some of the most useful and genuine information for algos. I love these videos. The content is top notch and has helped refine my algo trading immensely...so thank you Martyn. I also made the best pot of chili I've ever made while watching this channel and I like to think Martyn helped with that as well.
Martyn, your videos are absolutely amazing! You are helping us a lot! The information you provide is clear and easy to understand! However, for the current video I don't understand how you compare the optimization result with the walk-forward phase. A walk-forward is just a staged optimization so the result is the same as with the optimization for the analyzed period of time. Please let us know in the next video the way that you obtain the result data for this specific video. The confusion for me is coming from the correlation you make for the parameter value (you connect them with an optimization and walk-forward). If you make correlation for an optimization of say 8 years period and walk-forward which is 2 years / 1 year and then compare the parameters obtained for the CAR/MDD (for example) at the end of the 8 year period it would be OK but I left with the impression that you compare the exactly the same input parameters from Optimization and Walk-forward phase. Thank you :)
Hi Martyn. I wondering if the number of samples shouldn't be the number of entry signals regarding statistical significance. For example, let's say I have an indicator that generates 10,000 entry signals in 10 year and this gives me a 1.1 profit factor. If I add to this trading strategy a filter, let's say for example ATR, and the number of total trades drops to 3,000 with a profit factor of 1.5. Intuitively I think both backtests have the same level of confidence because they were all 10,000 trading signals. I didn't think much about it. Maybe I'm wrong. But what do you think?
This is outstanding. Aside from the obvious quality of the content, there's also a "suspense" component: "How will it all end?", which is keeping me so interested and watching these videos.
Some of the most useful and genuine information for algos. I love these videos. The content is top notch and has helped refine my algo trading immensely...so thank you Martyn. I also made the best pot of chili I've ever made while watching this channel and I like to think Martyn helped with that as well.
gold! ; how about using criteria such as Sortino or Ulcer indices?
Thanks Martyn for these educational videos. Can't wait for next 2 parts! (DKM)
Thanks for the feedback. Should be released this week.
It's a cliffhanger! Looking forward to the next episode.
Almost as exciting as an episode of Breaking Bad eh? Maybe not :)
Yes@@TradeLikeAMachine
Martyn, your videos are absolutely amazing! You are helping us a lot! The information you provide is clear and easy to understand! However, for the current video I don't understand how you compare the optimization result with the walk-forward phase. A walk-forward is just a staged optimization so the result is the same as with the optimization for the analyzed period of time. Please let us know in the next video the way that you obtain the result data for this specific video. The confusion for me is coming from the correlation you make for the parameter value (you connect them with an optimization and walk-forward). If you make correlation for an optimization of say 8 years period and walk-forward which is 2 years / 1 year and then compare the parameters obtained for the CAR/MDD (for example) at the end of the 8 year period it would be OK but I left with the impression that you compare the exactly the same input parameters from Optimization and Walk-forward phase. Thank you :)
Hi Martyn. I wondering if the number of samples shouldn't be the number of entry signals regarding statistical significance. For example, let's say I have an indicator that generates 10,000 entry signals in 10 year and this gives me a 1.1 profit factor. If I add to this trading strategy a filter, let's say for example ATR, and the number of total trades drops to 3,000 with a profit factor of 1.5. Intuitively I think both backtests have the same level of confidence because they were all 10,000 trading signals. I didn't think much about it. Maybe I'm wrong. But what do you think?
Good point on overfitting. What is you optimisation phase doing Martin? Is it running 1000s of trades through permutations of hyperparameters?