Hi there,

I'm new to TS and got a backtest issue to seek for help here.
When I apply different range of data, I got inconsistent returns and signals.
Here is the case for instance. I use 2 different range of data on the same symbol.
1. 2 years back
2. more than 2 years back, say from 1/1/2005
then I got inconsistent annual returns, say P/L in 2015 YTD is different in the 2 cases above.
Also, it's visibly clear on the chart that some (not all) signals are different.

My strategy is not a always-in-the-market one, so the 1st signal should not matter how it plays as time goes by.

Any one has similar experience with such situation?

Thanks!!

asobi,

this is possible, but there can be several reasons causing this.
Assuming you are using the same inputs in both cases I am guessing that
there is something in your code that causes the differences.
When you load a standard, built in strategy to a chart and check both
situations, do you see the same difference then?

Regards,
ABCTG

asobi 

Hi there, I'm new to TS and got a backtest issue to seek for help here. When I apply different range of data, I got inconsistent returns and signals. Here is the case for instance. I use 2 different range of data on the same symbol. 1. 2 years back 2. more than 2 years back, say from 1/1/2005 then I got inconsistent annual returns, say P/L in 2015 YTD is different in the 2 cases above. Also, it's visibly clear on the chart that some (not all) signals are different. My strategy is not a always-in-the-market one, so the 1st signal should not matter how it plays as time goes by. Any one has similar experience with such situation? Thanks!!

If you are using something like exponential averages, you will get different results because such calculations are starting point dependent.

If you are testing backadjusted continuous contracts, you could have differences if you are using percentage type calculations (close/close[1]).

Without more info, it is tough to say what is wrong...

Thanks @ABCTG and @kevinkdog for giving clues!!

Yes, I use EMAs, and that's exactly what the problem is!
This is tricky as the P/L drastically changes when different range of data being applied.
My initial idea to use shorter range of data is to save some memory usage when automation.
Now it seems not a good idea to do so.