Date: Mon, 12 Jan 2009 22:45:40 -0500
Reply-To: Sigurd Hermansen <HERMANS1@WESTAT.COM>
Sender: "SAS(r) Discussion" <SAS-L@LISTSERV.UGA.EDU>
From: Sigurd Hermansen <HERMANS1@WESTAT.COM>
Subject: Re: Bootstrapping Question
Content-Type: text/plain; charset="us-ascii"
Bootstrapping does not improve estimates of parameters of a model derived from a sample. Bootstrapping may, though not always, yield a better estimate of confidence intervals around parameter estimates.
As a rule, parameter estimates derived from any one sample tend to be too narrow. Using a bootstrap to improve estimates of confidence intervals reduces "sample fit" bias. Reducing the number of observations in an observed sample degrades slightly the accuracy of parameter estimates but may reduce confidence intervals and improve the accuracy of the model.
From: SAS(r) Discussion [mailto:SAS-L@LISTSERV.UGA.EDU] On Behalf Of Abhay Kaushik
Sent: Monday, January 12, 2009 4:48 PM
Subject: Bootstrapping Question
I have 320 funds and I am running bootstrap regressions with 1000 resampling for each fund (replacement of response variable with residual). When I ran my codes I do get the parametric p value, bootstrapped p value, parametric t values as well as coefficient of my main variable "Intercept" (alpha).
I am little confused because When I ran a simple OLS, I still get the same coefficient for intercept (alpha) for each fund as I obtained by running bootstrap method. Is it correct? I was reading chapters on bootsrapping and it seems that bootstraping improves the confidance intervals and I also feel that we should get a different value for coefficient.
Any suggestion is appreciated.