"degrees of Fredom" increases as sample size increases
To reproduce just increase your sample size and watch degrees of freedom increase without changing anything else. This is incorrect and unexpected. Here is all I can find from the documentation:
"It is in fact highly recommended for your strategy to have as little configurable parameters (degrees of freedom) as possible."
https://strategyquant.com/doc/optimization-profile-system-parameter-permutation-strategyquant/
Interestingly, this is already an area of confusion for some people:
"Building Winning Algorithmic Trading Systems: A Trader's Journey From Data Mining to Monte Carlo Simulation to Live Trading (Wiley Trading)" by Kevin Davey, for instance we believe in this book Degrees of freedom is incorrectly defined as discussed here:
https://roadmap.strategyquant.com/tasks/sq4_3414
However the fact remains all I did was add some more data and degrees of freedom went higher for the exact same strategies.
---EDIT---
I've been corrected in this thread, nothing was incorrectly defined:
https://strategyquant.com/forum/topic/degrees-of-freedomdof-bigger-is-better/
Rob Pardo is talking about degrees of freedom remaining in the data expressed as a percent. Kevin Davey is talking about degrees of freedom in the strategy.
Currently this stat is neither. Please select one of these two logical choices and stick with it. I suggest making it simply "degrees of freedom in the strategy" like it was before and a new column could be created for "degrees of freedom remaining in the data as a percent." (Hey look! I suggested this years ago! https://roadmap.strategyquant.com/tasks/sq4_3414 )
"It is in fact highly recommended for your strategy to have as little configurable parameters (degrees of freedom) as possible."
https://strategyquant.com/doc/optimization-profile-system-parameter-permutation-strategyquant/
Interestingly, this is already an area of confusion for some people:
"Building Winning Algorithmic Trading Systems: A Trader's Journey From Data Mining to Monte Carlo Simulation to Live Trading (Wiley Trading)" by Kevin Davey, for instance we believe in this book Degrees of freedom is incorrectly defined as discussed here:
https://roadmap.strategyquant.com/tasks/sq4_3414
However the fact remains all I did was add some more data and degrees of freedom went higher for the exact same strategies.
---EDIT---
I've been corrected in this thread, nothing was incorrectly defined:
https://strategyquant.com/forum/topic/degrees-of-freedomdof-bigger-is-better/
Rob Pardo is talking about degrees of freedom remaining in the data expressed as a percent. Kevin Davey is talking about degrees of freedom in the strategy.
Currently this stat is neither. Please select one of these two logical choices and stick with it. I suggest making it simply "degrees of freedom in the strategy" like it was before and a new column could be created for "degrees of freedom remaining in the data as a percent." (Hey look! I suggested this years ago! https://roadmap.strategyquant.com/tasks/sq4_3414 )
Attachments
No attachments
-
Votes +1
-
Project StrategyQuant X
-
Type Bug
-
Status Refused
-
Priority Normal
History
b
b
b
bentra
25.06.2019 00:31
Also interesting in the task I referenced the discussion ended as if everything was already correct but the task itself is marked "fixed" instead of "refused"
https://roadmap.strategyquant.com/tasks/sq4_3414
Edit: oops, the 0's were fixed obviously (the other part of that bug report)
https://roadmap.strategyquant.com/tasks/sq4_3414
Edit: oops, the 0's were fixed obviously (the other part of that bug report)
b
bentra
05.05.2020 18:21
It might be nice to first define it properly in the context of SQ so there is no confusion because sometimes the definition is different in trading books than it is in wiki:
https://en.wikipedia.org/wiki/Degrees_of_freedom_(statistics)
Pardo's book is cited here:
https://strategyquant.com/forum/topic/degrees-of-freedomdof-bigger-is-better/
https://en.wikipedia.org/wiki/Degrees_of_freedom_(statistics)
Pardo's book is cited here:
https://strategyquant.com/forum/topic/degrees-of-freedomdof-bigger-is-better/
PJ
gottogethelp
05.05.2020 18:48
Hi bentra. Do you agree with the Pardo definition? I believe SQ were going to implement this but I haven't checked in B127.
b
bentra
05.05.2020 18:59
Haha thanks for asking. Frankly, I don't really mind much either way. I slightly prefer wiki but I think most other traders prefer and expect Pardo's implementation. I'd just like to see it in the documents, so I know what to expect from this stat.
PJ
b
bentra
08.05.2020 19:25Description changed:
To reproduce just increase your sample size and watch degrees of freedom increase without changing anything else. This is incorrect and unexpected. Here is all I can find from the documentation:
"It is in fact highly recommended for your strategy to have as little configurable parameters (degrees of freedom) as possible."
https://strategyquant.com/doc/optimization-profile-system-parameter-permutation-strategyquant/
Interestingly, this is already an area of confusion for some people:
"Building Winning Algorithmic Trading Systems: A Trader's Journey From Data Mining to Monte Carlo Simulation to Live Trading (Wiley Trading)" by Kevin Davey, for instance we believe in this book Degrees of freedom is incorrectly defined as discussed here:
https://roadmap.strategyquant.com/tasks/sq4_3414
However the fact remains all I did was add some more data and degrees of freedom went higher for the exact same strategies.
---EDIT---
I've been corrected in this thread, nothing was incorrectly defined:
https://strategyquant.com/forum/topic/degrees-of-freedomdof-bigger-is-better/
Rob Pardo is talking about degrees of freedom remaining in the data expressed as a percent. Kevin Davey is talking about degrees of freedom in the strategy.
Currently this stat is neither. Please select one of these two logical choices and stick with it. I suggest making it simply "degrees of freedom in the strategy" like it was before and a new column could be created for "degrees of freedom remaining in the data as a percent." (Hey look! I suggested this years ago! https://roadmap.strategyquant.com/tasks/sq4_3414 )
"It is in fact highly recommended for your strategy to have as little configurable parameters (degrees of freedom) as possible."
https://strategyquant.com/doc/optimization-profile-system-parameter-permutation-strategyquant/
Interestingly, this is already an area of confusion for some people:
"Building Winning Algorithmic Trading Systems: A Trader's Journey From Data Mining to Monte Carlo Simulation to Live Trading (Wiley Trading)" by Kevin Davey, for instance we believe in this book Degrees of freedom is incorrectly defined as discussed here:
https://roadmap.strategyquant.com/tasks/sq4_3414
However the fact remains all I did was add some more data and degrees of freedom went higher for the exact same strategies.
---EDIT---
I've been corrected in this thread, nothing was incorrectly defined:
https://strategyquant.com/forum/topic/degrees-of-freedomdof-bigger-is-better/
Rob Pardo is talking about degrees of freedom remaining in the data expressed as a percent. Kevin Davey is talking about degrees of freedom in the strategy.
Currently this stat is neither. Please select one of these two logical choices and stick with it. I suggest making it simply "degrees of freedom in the strategy" like it was before and a new column could be created for "degrees of freedom remaining in the data as a percent." (Hey look! I suggested this years ago! https://roadmap.strategyquant.com/tasks/sq4_3414 )
MF
Mark Fric
12.06.2020 09:21Status changed from New to Refused
please see my comment o the forum: https://strategyquant.com/forum/topic/degrees-of-freedomdof-bigger-is-better/page/2/#post-258961
Votes: +1
© Copyright. All rights reserved. ProjectPanel.com
