Striking result! Andrea Unger, who is another fine educator, is also a great advocate of asset filtering. It's clearly worth exploring. The ideal strategy self-adapts to market conditions, but they are hard to find. I have a whole raft of rejected ideas I'll be revisiting to test with your filtering suggestions!
Hi Martyn! Very good video, thanks for sharing it. But, we can say that the results are not surprising since we are selecting in advance the currency pairs that we know have been noisy during the last years, for getting not bias result we would have to redone the pairs selection every month/year (analyzing only the data until that date), right? Thanks.
Remember that the assett stats were taken only from the last 2000 hrs. (approx 80 days). However the backtest was over 10 years. So I'm happy this is not the case.
Hi Martyn, I have a question: If the Asset ranking chart with AUDNZD being the most noisy is derived based on 20 bars period efficiency ratio over recent 4 months, and considering the noise characteristics is dynamic over time, how could you select the 8 most noisy pairs over just recent four months to test performance over 10 years? Shouldn’t we select 8 most noisy over each 4 months and test over next 4 months and have an approach similar to WFA? Thanks.
Have you done some research in this direction? If not, I will make a quaterly selection of the most noisy pairs, and see how it varied historically. For example, I doubt that AUDNZD stayed on top of the list since october 2022 when it became very trendy.
Really appreciate these videos Martyn. Can you please confirm how the asset ordering displayed at 3:59 was achieved using the ER with a look-back period of 20 [assumed from the (20) at the chart top] over the backtest period? Average of each asset's 20-period ERs over the entire backtest period? Thanks again.
20 period lookback (20 hours) but this is then played out across the last 2000 bars (hours) and an average produced. So basically it is an average analysis over the most recent 2000 hours or 83 days (approx 4 trading months)
There are 28 pairs (and for others 8 pairs) , how is it possible to have only one current_equity ?? If those pairs are merged , how did you do it ? Thanks
Helpful indeed but surprising as well. I always thought that usdcad, gbpjpy, and audnzd are of the same category cuz whenever I test a strategy and it works on one it will be as good on the other two. Does that mean a strategy that perform very well on one from each of the assets classes is probably overfit and adapt to noise? That assuming all precautions to not overoptimise and usage of out of sample walk forward were taken?
Not necessarily. There are of course many other factors in addition to noise than can impact similarity between assets. Volatility is of course one of these. And I will be talking a lot in the future about the relationship of volatility and noise and how they impact an asset's personality. Stay tuned...
Awesome study I can’t wait for the next episode!
Thanks for the feedback Eli
Striking result! Andrea Unger, who is another fine educator, is also a great advocate of asset filtering. It's clearly worth exploring.
The ideal strategy self-adapts to market conditions, but they are hard to find.
I have a whole raft of rejected ideas I'll be revisiting to test with your filtering suggestions!
Let me know how you get on with the testing of older strategies. Very interested to hear your conclusions.
Thank you Martyn!
Thanks for sharing this concept Martyn !
Super interesting.
Cheers Leon
Amazing and surprising results Martyn. Thank you very much for your videos!
Hi Martyn! Very good video, thanks for sharing it. But, we can say that the results are not surprising since we are selecting in advance the currency pairs that we know have been noisy during the last years, for getting not bias result we would have to redone the pairs selection every month/year (analyzing only the data until that date), right? Thanks.
Remember that the assett stats were taken only from the last 2000 hrs. (approx 80 days). However the backtest was over 10 years. So I'm happy this is not the case.
Why use a 20 period eff/ratio over the standard 10 period?
Hi Martyn, I have a question: If the Asset ranking chart with AUDNZD being the most noisy is derived based on 20 bars period efficiency ratio over recent 4 months, and considering the noise characteristics is dynamic over time, how could you select the 8 most noisy pairs over just recent four months to test performance over 10 years? Shouldn’t we select 8 most noisy over each 4 months and test over next 4 months and have an approach similar to WFA? Thanks.
Have you done some research in this direction?
If not, I will make a quaterly selection of the most noisy pairs, and see how it varied historically. For example, I doubt that AUDNZD stayed on top of the list since october 2022 when it became very trendy.
Really appreciate these videos Martyn. Can you please confirm how the asset ordering displayed at 3:59 was achieved using the ER with a look-back period of 20 [assumed from the (20) at the chart top] over the backtest period? Average of each asset's 20-period ERs over the entire backtest period? Thanks again.
20 period lookback (20 hours) but this is then played out across the last 2000 bars (hours) and an average produced. So basically it is an average analysis over the most recent 2000 hours or 83 days (approx 4 trading months)
Hi Martyn, thanks for this- very interesting. Would it be possible to post the code on github as I'd like to run it against the FTSE ?
Hi Eric. At the end of the series I am going to assess what can be shared on Github. Hold tight :)
There are 28 pairs (and for others 8 pairs) , how is it possible to have only one current_equity ?? If those pairs are merged , how did you do it ? Thanks
It is done using a multi-symbol EA. See here: ruclips.net/p/PLv-cA-4O3y95e9N3saUmDpQGKR11k-v1B Episodes 1.1, 1.2 and 1.3. Enjoy!
@@TradeLikeAMachine Thanks Martyn !
Can you make a video for meanreverting system
I'll consider for the future. Thansk for the suggestion.
Helpful indeed but surprising as well. I always thought that usdcad, gbpjpy, and audnzd are of the same category cuz whenever I test a strategy and it works on one it will be as good on the other two. Does that mean a strategy that perform very well on one from each of the assets classes is probably overfit and adapt to noise? That assuming all precautions to not overoptimise and usage of out of sample walk forward were taken?
Not necessarily. There are of course many other factors in addition to noise than can impact similarity between assets. Volatility is of course one of these. And I will be talking a lot in the future about the relationship of volatility and noise and how they impact an asset's personality. Stay tuned...
@@TradeLikeAMachine Awsome. Thanks again for helpful content you share for free with us.
I went to your website but my country of residence was not shown in your dropdown menu. How can I sign up otherwise?
Sir can u share strategies for intraday?