New metric that shows correlation between long-only and short-only equity curve

I´ve noticed that my most stable strategies have a very similar (straight) equity curve separately(!) for the long and for the short trades only. These strategies do better than other strategies that also have a combined(!) very straight equity curve (so long and short combined) but have a very uneven long only and short only equity curve if seen separately (basically showing that it´s just luck and good timing (which will never happen like this again) that the combined equity curve looks good, but the strategy is weak actually). So would it be possible to make a new correlation metric that measures the correlation (something like R-squared maybe) between the long-only equity curve and the short-only equity curve (also as an acceptance criteria of course)? This way we could see how close the long only and short only equity curve correlate with each oher and the higher the correlation, the more stable the strategy is (at least from my experience).


Thank you :-)

Attachments
No attachments
  • Votes +6
  • Project StrategyQuant X
  • Type Feature
  • Status Archived
  • Priority Normal
  • Assignee None
  • Milestone Archived (To be done later)

History

g
#1

geektrader

20.01.2019 11:53

Task created

g
#2

geektrader

20.01.2019 11:56
Voted for this task.
IH
#3

clonex / Ivan Hudec

20.01.2019 13:47
Voted for this task.
KL
#4

kainc301

20.01.2019 16:24
Voted for this task.
m
#5

mikeyc

13.02.2019 22:01
Voted for this task.
KW
#6

Sean

24.02.2019 11:41
Voted for this task.
KL
#7

kainc301

08.08.2019 19:15
While I definitely think this is a good idea and should be explored for research purposes, I have been thinking about what this measurement would actually tell us. If the correlation was based on when trades are taking place and the trades were being executed in opposite directions, how useful would this information be? I don't understand why it would be good if these values were highly correlated as you suggest.


If anything, I look at correlation that exists on only one side to see if all my short strategies are trading around the same times and vice versa for long strategies. Curious to get more thoughts on this though.

b
#8

bentra

09.08.2019 18:16
Voted for this task.
b
#9

bentra

09.08.2019 18:19
When you find them uncorrelated, it may be because of the symmetry issue:
https://strategyquant.com/forum/topic/what-should-happen-when-short-and-long-signals-fire-at-the-same-time/
KL
#10

kainc301

11.08.2019 00:50
@bentra These are not for symmetrical strategies. These are for long-only vs short-only strategies which will have different rules and thus not be symmetrical. The point of long only/short only strategies is to be asymmetrical as there isn't a short or long counterpart to the side that is trained.
b
#11

bentra

13.08.2019 15:41
@keinc: I'd be willing to bet geektrader is referring to an origin strategy that is symmetrical and having separate longs and shorts equity lines from it analyzed and compared.

If this were implemented like a cross check, we could use all the standard tools to find that the two lines were both straight.
KL
#12

kainc301

13.08.2019 20:01
He said long-only vs short only which are asymmetrical by design. I am confused about what you are referring to. Standard correlation already compares symmetrical strategies.
b
#13

bentra

13.08.2019 21:06
As in analyzing a single symmetrical strategy by forming it's long only and short only versions of itself and then comparing the equity curves of these to each other....
KL
#14

kainc301

14.08.2019 23:18
Hmm. I understand what you are trying to say now. But I don't know how there would be any correlation between the two sides if the orders should never be activated at the same time if they are opposite entry conditions. Similar to the conversation on the other thread about a simultaneous entry. 
b
#15

bentra

14.08.2019 23:33
I think he means the two equity lines shapes are correlated as in both equity lines are the same shape - both straight. They'd be correlated on a larger (weekly / monthly) level right?


g
#16

geektrader

18.08.2019 15:08

Attachment Untitled2.jpg added

Attachment Untitled3.jpg added

@bentra: yes, that is exactly what I meant. Example (from another platform that already has this feature) is attached. If both, the separate long and short equity curve have pretty much equal shape like in this example, I´ve found the strategies to be more stable in live trading. I´ve also attached a second example that I would not trade, the long side is much weaker than the short side in this case.
h
#17

hankeys

18.08.2019 20:16
isnt ranking for LONG or SHORT enough? because you can set whatever you want also only on 1 side


you can set filter for SYMMETRY for example

b
#18

bentra

19.08.2019 00:06
But to compare the two lines and filter based on the contrast is not possible yet. Nor (I don't think) is comparing short stats (RT/DD for instance) to long stats via filter, we can filter via one side or the other by a static number (short RT/DD > 3 and long RT/DD > 3) but not via a comparison of the two to each other (short RT/DD = long RT/DD)

The symmetry stat is either still bugged or doesn't seem to care about anything but finished PL. 99% symmetry is reported for wildly asymmetrical strategies where it produced 100 short trades and 1000 long trades just because they had the same profit. Completely different equity lines. This stat is currently not the best way for trying to identify symmetry.
h
#19

hankeys

19.08.2019 08:06
but you could also set the ranking filter as a comparation of short RDD vs long RDD


even with percentage, for example


short RDD > 80% long RDD

short RDD < 120% long RDD

b
#20

bentra

19.08.2019 21:52
Oh yeah %! Thanks for the reminder.
MF
#21

Mark Fric

16.02.2020 13:52

Status changed from New to Archived


Votes: +6

Drop files to upload

or

choose files

Max size: 5MB

Not allowed: exe, msi, application, reg, php, js, htaccess, htpasswd, gitignore

...
Wait please