BUG: factor_rank_autocorrelation infers turnover period in calendar...

luca-s commented 5 years ago

…space while periods could have different time space

quantile_turnover is affected by the same bug

closes Issue #323

twiecki commented 5 years ago

So this removes the ability to pass in anything more fine-grained than days? That's quite a hefty price-tag for fixing that edge case.

luca-s commented 5 years ago

No, don't worry. I didn't change the Alphalens behaviour. Due to the code change I noticed the are few test cases, which I created some time ago, that fail but also they don't make sense. Since it took some time to figure that out, I made sure the functions in utils reise now an exception when they receive a wrong parameter type.

luca-s commented 5 years ago

Just to be more detailed, the three tests were wrong for the following reason:

freq represents the factor trading calendar, which can be only Day, BusinessDay or CustomBusinessDay
the factor data itself can have any frequency, even hours, minutes, seconds. The period at which we analyse the factor can use these frequencies even though the trading calendar has daily resolution
the tests that I removed used the factor data frequency (10min, 1 hour and 3 hour) as value for freq, which is wrong So I deleted the wrong tests but we can still analyse a factor at any frequency we like.

quantopian / alphalens

BUG: factor_rank_autocorrelation infers turnover period in calendar... #324