- Who are you?!!!
- I am your worst nightmare (overfitting). -
I know, I know: "Diego, refrain yourself from adding too many filters and variables" ==> I am attending to the psychoanalyst in order to solve this :D
What do you think of running back tests for periods entirely OOS? Let say if I optimized the strategy for the last 20 years, then I run a back test from 1993 to 2003 in order to be OOS. I am experimenting with this, but results are looking too good to be truth.
What am I missing?
- I am your worst nightmare (overfitting). -
I know, I know: "Diego, refrain yourself from adding too many filters and variables" ==> I am attending to the psychoanalyst in order to solve this :D
What do you think of running back tests for periods entirely OOS? Let say if I optimized the strategy for the last 20 years, then I run a back test from 1993 to 2003 in order to be OOS. I am experimenting with this, but results are looking too good to be truth.
What am I missing?
Rename
Seems like a valid approach. The only problem is that 1993 through 2002 was a tremendous bull market. And when I say "tremendous", I mean "bubble".
Maybe use a different market from 2003 - 2023. That's OOS too.
Maybe use a different market from 2003 - 2023. That's OOS too.
If you optimize for "new" data and check results with "old" data, this may be an interesting exercise, but it is not very helpful.
You should (always) do it the other way round:
Develop/change/edit your strategy for "old" data, say 1983 to 2003 (20 years)
If you are happy with the results, run it once for 2004 to 2023.
Use log Y-Axis.
Don't be sad, you'll find a lot of overfitting.
You should (always) do it the other way round:
Develop/change/edit your strategy for "old" data, say 1983 to 2003 (20 years)
If you are happy with the results, run it once for 2004 to 2023.
Use log Y-Axis.
Don't be sad, you'll find a lot of overfitting.
It is not entirely clear to me what the deeper reasons are for this, but:
Things that work in the future also tend to work in the past.
The converse is not true: Things that worked in the past tend to vanish in the future...
Things that work in the future also tend to work in the past.
The converse is not true: Things that worked in the past tend to vanish in the future...
Indeed, it is better the other way round! I feel more confident about the procedure. Let's fight the evil overfitting and win!!!
Your Response
Post
Edit Post
Login is required