@tornellFRM okay thank you. We just discourage "bumping" (re-posting) for what I hope are obvious reasons. I didn't immediately reply because I couldn't instantly decide if the answer is wrong (probably) or just meaningless (possibly). Whenever I disagree with GARP's materials, which is often, I try to be very careful because if my disagreement is incorrect, then I've done a real disservice to our members. Consequently, a proper reply to your question is time-consuming because I should go to the original data and make sure i've got the whole context. I will just say that

(1) I get a bit conceptually stuck on the idea of "annualizing a t-stat;" what exactly does it mean? I could make an argument it is a meaningless idea as a t-stat is a standardized (of the raw distance between the observation and a zero null) standard deviation of an observed coefficient. It's not the same thing as annualizing a monthly standard deviation. The answer given (i.e., same) actually reinforces an argument that time-scaling a t-stat is meaningless in the first pace.

(2) Per Amenc, the information ratio can be viewed as (or, actually really is) an annualized t-stat, where the relationship is given by IR = t-stat / SQRT(T) or t-stat =

**IR*SQRT(T)**; e.g.,. see

https://www.bionicturtle.com/forum/threads/question-on-t-statistic.2588/post-8122
So if I were forced to annualize a t-stat, and it meant something different than re-running the regression with a new periodicity (?), then like you, I would scale the SE with a SQRT(T). Put another way, I'd scale the t-stat like I'd scale the IR, by multiplying by SQRT(T) which is just to multiply it by

**[T / SQRT(T)]** = [SQRT(T)*SQRT(T)] / SQRT(T) = SQRT(T); i.e., i'd be linear in the numerator but SQRT(T) in the denominator. But, again, I'm not certain this is even meaningful .... Thanks,

## Stay connected