Multi-Day Holding Periods

I believe that I might have an idea on how to capture performance realistically for multi-day holding periods for a daily factor. For example, if you wanted to capture 21-day holding period performance on a factor of a factor that was refreshed each day, you would take your hypothetical portfolio and divide into 21 slices. You would rebalance one slice each day for 21 days. The return for tomorrow would be the average of tomorrow's returns for each of the 21 slices created over the period from 21 days ago though yesterday.

Has anyone tried this approach or have any ideas on how I might do this.

Regards, Eric

2 responses

Take a look at Alphalens. It might do what you're looking for. That tool was developed to analyze factor performance. Check it out https://www.quantopian.com/tutorials/alphalens#lesson1.

Attached is an example of how one would get the 21 (business) day forward returns of an arbitrary factor.

Maybe also look at Thomas Wiecki's post on some recent updates https://www.quantopian.com/posts/an-updated-method-to-analyze-alpha-factors.

2
The way I've sorted out for doing this is I keep a list of rolling portfolios context.rolling_portfolios = [] and use context.rolling_portfolios.pop(0) once the list exceeds my desired holding period, which gets rid of the oldest portfolio. . I use todays_weights = opt.calculate_optimal_portfolio(... and append the result to the list of rolling portfolios context.rolling_portfolios.append(todays_weights). Then I average them all together and feed the combined weights into calculate_optimal_portfolio().