[R-SIG-Finance] Is there R code for measuring information coefficient?

Brian G. Peterson brian at braverock.com
Wed Mar 14 15:18:21 CET 2012

On Wed, 2012-03-14 at 09:01 -0500, Michael wrote:
> is an approximation...  

*Everything* that is not directly measured in an approximation.

>  So it's better measure IC and BR directly from raw trades data.   

Grinold isn't about measuring *trades*, it is about measuring skill in

> You said BR is trivial, could you please show me how to measure
> "independent bets" in practice? 
Breadth isn't about 'independent bets'. To quote[1], 'the value added by
the manager will be proportional to the information ratio squared'.
Breadth is 'the number of independent forecasts of exceptional returns
we make in a year'.  If you know the strategy or the manager or the
portfolio, all you do is *count* independent forecasts.

While the rest of chapter 6 in Grinold discusses other methods of
estimating Information Coefficient, these suggest using a degree of
forecasting power and process measurement that you have completely
failed to demonstrate.  Stick with the first approximation.

If that wasn't enough, I draw your attention to the end of the chapter
for another reason why the 'approximation' is probably superior to a
more nuanced mode in your case: "realized information ratios for optimal
portfolios based on {perfect} forecasts are statistically
indistinguishable from the forecasts of the fundamental law"[1 p.161].

Grinold and Kahn, Active Portfolio Management. Second Edition. 
Chapter 6, pp. 147 ff.

More information about the R-SIG-Finance mailing list