Factor Rank Autocorrelation for each day/period about alphalens HOT 3 CLOSED

This idea makes complete sense to me. The way we are currently doing it is with a pandas timerule, I'm kind of debating whether we want to eschew pandas timerules all together in the package, and just focus on a number of periods analysis. Right now there is kind of a mix, not sure how i feel about that

from alphalens.

The current rule used by the tear sheet is time_rule='D'. That means it works fine If we pass daily factors/data to the tear sheet (because it won't do any resampling at all) but it will result in surprising graph if we pass minutely/weekly (anything that is not daily) data.

It seems to me it would make sense to make the resampling an option (used only by aware users) and let the code that follows the resampling decides the autocorrelation period:

In factor_rank_autocorrelation function:

autocorr = asset_factor_rank.corrwith(asset_factor_rank.shift(1), axis=1)

shift(1) should be changed to shift(period) where period becomes a function argument. In this way the feature is available when calling the function directly (at least). Later it can be decided how to enable this option via tear sheet (probably via tear sheet function argument or simply calling factor_rank_autocorrelation for each day/period)

from alphalens.

That sounds good I'd go ahead and submit a PR.

from alphalens.