wilmott.com

Hi All,I am working with rolling regression. lets say I am working with window size 10, 20 and so on. But is there any way by which we can say that particular window length (say 50) is optimal to use.RegardsGaurav

How about optimizing the R^2. (of course there's 100 other ways to do this - AIC, BIC etc).

- make sure you have enough samples given the number of independent vars you have in your regression - your window size should be bigger the more X's you put in.- i usually sweep various window sizes to find the optimal one - yes, it's a form of overfitting- lastly, make sure your sweep window looks backward. you don't want any forward looking bias in your betas.

Rolling regression attempts to "have it both ways", which is always dangerous.Your regression model assumes that the regression coefficients are static facts of nature just waiting for you to estimate them. But by computing rolling regressions you are assuming that the regression coefficients change over time. Which is it? Are the coefficients static or dynamic? If they are dynamic, then classical regression is a very inefficient tool for the job. Look into "dynamic linear models".

Have you read this document?http://rady.ucsd.edu/faculty/directory/ ... 10.pdfHope it helps!

Hi Everybody,Thanks a lot for your valuable inputs. Regards,Gaurav