1. Is there an ouptut from each run of all the optimizations? Is it possible to see the top Nth performing systems for each run? I.e. the top 10 performing systems from each WFA run.
2. Is it possible to use the performance from all the systems to select the best system? Example, select the system with the highest custom criteria that maximizes distance in NET PROFIT from losing systems, i.e. pick the center cluster of high ground. It seems just glaring that nobody mentions this as a selection criteria.
3. Is it possible to graph the OOS WFA analysis equity curve in MC?
WFA Questions
-
- Posts: 108
- Joined: 05 Mar 2014
- Has thanked: 1 time
- Been thanked: 12 times
-
- Posts: 108
- Joined: 05 Mar 2014
- Has thanked: 1 time
- Been thanked: 12 times
- Vlada MultiCharts
- Posts: 293
- Joined: 22 Apr 2020
- Has thanked: 8 times
- Been thanked: 76 times
Re: WFA Questions
Hello tpredictor,
1. Only one result is saved for each run, so you'll see one best result on each. Technically it is not possible. to see the 10 best results.
2. Unfortunately there is no such pre-built functionality. All optimization reports are saved on your PC in Documents, so you can run optimization on different systems and review the reports and select the best one.
3. At the moment it is not possible. Our Engineering Team will consider adding this feature.
1. Only one result is saved for each run, so you'll see one best result on each. Technically it is not possible. to see the 10 best results.
2. Unfortunately there is no such pre-built functionality. All optimization reports are saved on your PC in Documents, so you can run optimization on different systems and review the reports and select the best one.
3. At the moment it is not possible. Our Engineering Team will consider adding this feature.
-
- Posts: 108
- Joined: 05 Mar 2014
- Has thanked: 1 time
- Been thanked: 12 times
Re: WFA Questions
Thanks. I also shared with you recently my blog post. I have came up with a new variation on the WFO which I called Weighted Anchored WFO. The idea is to weight newer results more heavily (decay older results). I think I could do this in a custom fitness function if I could access a bar number so that I can give a weight to more recent results.
Is there a way I can reference the bar number in the fitness function? I could use that a multiplier to decay the older results. Thanks. If not please consider this.
Is there a way I can reference the bar number in the fitness function? I could use that a multiplier to decay the older results. Thanks. If not please consider this.
- Vlada MultiCharts
- Posts: 293
- Joined: 22 Apr 2020
- Has thanked: 8 times
- Been thanked: 76 times
Re: WFA Questions
tpredictor,
Yes, you can reference the bar number in the fitness function.
The CFF values can be accessed from the script:
SetCustomFitnessNamedValue
SetCustomFitnessValue
It is possible to refer to the bar number as well.
Sorry, but we didn't find the link to the post you are referring to. Please describe in more detail what you are willing to achieve.
Yes, you can reference the bar number in the fitness function.
The CFF values can be accessed from the script:
SetCustomFitnessNamedValue
SetCustomFitnessValue
It is possible to refer to the bar number as well.
Sorry, but we didn't find the link to the post you are referring to. Please describe in more detail what you are willing to achieve.
-
- Posts: 108
- Joined: 05 Mar 2014
- Has thanked: 1 time
- Been thanked: 12 times
Re: WFA Questions
I suggest some methods here:
Weighted Anchored Walk Forward Analysis: You decay the profits/losses or fitness function over time .
Volatility/Variation Punishment: Idea is to penalize WFO systems/settings that vary greatly from one period to the next.
https://beyondbacktesting.com/2021/04/1 ... ns-in-wfo/
---
Also regarding bar number, is the fitness function called for every bar or just per trade? If it is per trade, is that the last bar of trade or first ? Can I access print functionality as well in the fitness function? Can I call the normal functions?
Weighted Anchored Walk Forward Analysis: You decay the profits/losses or fitness function over time .
Volatility/Variation Punishment: Idea is to penalize WFO systems/settings that vary greatly from one period to the next.
https://beyondbacktesting.com/2021/04/1 ... ns-in-wfo/
---
Also regarding bar number, is the fitness function called for every bar or just per trade? If it is per trade, is that the last bar of trade or first ? Can I access print functionality as well in the fitness function? Can I call the normal functions?
Re: WFA Questions
When I first started using WFA this was my first thought. The WFA report would be far more useful if each selected optimization had its own full time series report. The same way you can break apart portfolio trader into its individual reports. Not being able to see what's happening in the individual steps reduces the WFA's use. And as you mentioned being able to cross reference all the selected settings would be extremely useful for determining OOS fitness.1. Is there an ouptut from each run of all the optimizations? Is it possible to see the top Nth performing systems for each run? I.e. the top 10 performing systems from each WFA run.
2. Is it possible to use the performance from all the systems to select the best system? Example, select the system with the highest custom criteria that maximizes distance in NET PROFIT from losing systems, i.e. pick the center cluster of high ground. It seems just glaring that nobody mentions this as a selection criteria.
3. Is it possible to graph the OOS WFA analysis equity curve in MC?
-
- Posts: 108
- Joined: 05 Mar 2014
- Has thanked: 1 time
- Been thanked: 12 times
Re: WFA Questions
I posted the Weighted Anchored WFO code solution here:
https://beyondbacktesting.com/2021/04/2 ... hored-wfo/
@joebone Yes, being able to see individual trades would be useful to see if the performance was degrading over time.
https://beyondbacktesting.com/2021/04/2 ... hored-wfo/
@joebone Yes, being able to see individual trades would be useful to see if the performance was degrading over time.