Posted By: boatman
Considering weekend gaps - 07/25/15 10:10
This article raises some interesting questions: Time series analysis and data gaps
The main point is that weekend returns (and the first few intraday Monday bars) can have vastly different statistical properties than weekday returns, and that treating them the same in our development and testing can lead to reduced performance, inaccurate optimization and skewed results.
My conclusion from this is that it is worth testing our strategies by constraining them to trade only from say a few hours after the open on Sunday night (UTC) to some time prior to the close on Friday, including closing out all positions prior to the weekend.
For inter-day strategies, this would add to the transaction costs, however based on the article there is evidence that it could lead to more robust models and better performance.
Since this would be simple to implement in Zorro, it seems a useful tool to add to the arsenal, at least as an exercise in comparisons.
I would really like to hear anyone's thoughts and opinions.
Cheers
The main point is that weekend returns (and the first few intraday Monday bars) can have vastly different statistical properties than weekday returns, and that treating them the same in our development and testing can lead to reduced performance, inaccurate optimization and skewed results.
My conclusion from this is that it is worth testing our strategies by constraining them to trade only from say a few hours after the open on Sunday night (UTC) to some time prior to the close on Friday, including closing out all positions prior to the weekend.
For inter-day strategies, this would add to the transaction costs, however based on the article there is evidence that it could lead to more robust models and better performance.
Since this would be simple to implement in Zorro, it seems a useful tool to add to the arsenal, at least as an exercise in comparisons.
I would really like to hear anyone's thoughts and opinions.
Cheers