- ago


Help, why am I not getting any results here? Walk forward efficiency says NaN and I get 0% apr for all out of sample intervals.

0
345
Solved
15 Replies

Reply

Bookmark

Sort
- ago
#1


I have no idea if this helps, but I copied this from the bottom of the screen.
0
- ago
#2
How can we reproduce this, what are the steps? e.g.

1. Does this apply to one particular, some or all strategies you've tested?
2. Maybe it appears that a specific DataSet or symbol is affected?
etc.
0
Cone8
 ( 6.71% )
- ago
#3
It looks as if there were no trades during any of the OOS period. That would result in NaNs for the WFE.

My bet is that you're using a 3rd party optimizer / scorecards. That's a good place to start looking.
0
- ago
#4
Eugene,

1. unfortunately I'm kind of new so this is my first real attempt at WFO,

2. The data set is Sp500 current and past from norgate
0
- ago
#5
Cone,

The strategy is a variation of Knife Juggler on the Sp500 so there should be hundreds of OOS trades

This was using the exhaustive WFO. I did optimize for MARratio. I don't understand why it would give in-sample results and nothing OOS

In-sample was 18 month intervals and OOS was 6 montsh.
0
Glitch8
 ( 11.74% )
- ago
#7
It’s really hard to say without having the strategy to try and reproduce this 🤷🏼‍♂️
0
Cone8
 ( 6.71% )
- ago
#8
Moptop13 is using the Knife Juggler (Post #5).

I'll give this a try.
Which S&P 500 data provider did you use?

Since the Knife Juggler in Sample Strategies is not set up to optimize, please export your Knife Juggler to C# Code and paste it here so we know how the optimization variables are set up.
0
Glitch8
 ( 11.74% )
- ago
#9
He’s not using Knife Juggler. He’s using a “variation” of Knife Juggler. The results show a parameter not contained in Knife Juggler.
2
- ago
#10
The original run was Norgate SP500 current and past. I ran it again last night using Norgate Nasdaq 100 current and past and the same thing happened.

Here is the whole strategy:




0
Glitch8
 ( 11.74% )
- ago
#11
You have too many windows. The OOS window is only 4 months long, that's not enough data to accommodate the one year (252 bar) period you defined as a LOWER limit.
0
Best Answer
- ago
#12
I'm purely a building blocks guy, I don't know how to code in C#, and I can try to upload the code if needed
0
- ago
#13
Thanks Glitch.

So the OOS sample period has to be longer than any of the periods of my indicators?

So If I used a 200 day sma in my strategy the, the OOS period would need to be longer than 6 months?

That leads me to another hypothetical question: If I used a 200 day sma and my OOS period was 250 days, would trades only be calculated on the last 50 days of that sample or would the whole period work?
0
Cone8
 ( 6.71% )
- ago
#14
It depends on where the OOS period starts relative to the beginning of the test data. If the first OOS period was more than 200 days from the start date, then it would be valid at the beginning of the OOS test... but this would mean that if the InSample period was 200 bars, it wouldn't be valid to create even one trade.

This is assuming blocks, because if you have more seed data, there is a way with a C# Strategy to make a long-period indicator "valid" from the first bar, but that's another topic.

0
- ago
#15
Thanks, I changed my longest parameter to 120 days and things seem to work now with a 6 month OOS period.
0

Reply

Bookmark

Sort