Serving the Quantitative Finance Community

 
User avatar
tw
Topic Author
Posts: 592
Joined: May 10th, 2002, 3:30 pm

multi-regression question

August 10th, 2006, 10:05 am

Hi I have a question regarding multiple regression, would be grateful for any pointers.The situations it that I have a couple of potential sets of variables that I think are significant for a particular independent variable:visually one is much more significant than the other.I can run regressions with one variable, and then with two (and see that the coefficient of determination increases with the addition of the second variable), but how can I test that the addition of the second variable is significant? I imagine this situation must occur in numerous situations, is there a named test I can look up?Many thanks.
 
User avatar
gardener3
Posts: 8
Joined: April 5th, 2004, 3:25 pm

multi-regression question

August 10th, 2006, 3:22 pm

Look up AIC (Akaike Information Criterion) or BIC (BAyesian Information Criterion)
 
User avatar
Maelo
Posts: 0
Joined: July 28th, 2002, 3:17 am

multi-regression question

August 11th, 2006, 3:42 am

Check "Linear Regression Analysis" by Douglas C. Montgomery...can't remember the exact cahpter, but I think this is well-covered there. A F-tets between two models is performed: a reduced model vs a full model (with all variables) than a forwrad or backawrd methodology (mostly depending on your preferences) is used toa dd or elimnate varoables.Hope this help.M
 
User avatar
ntruwant
Posts: 0
Joined: August 3rd, 2004, 9:50 am

multi-regression question

August 11th, 2006, 8:01 am

The most easy rule is retaining variables that have a t statistic of at least 2 . But you need a book indeed to go further than this (check for multicollinearith,...)
 
User avatar
csa
Posts: 0
Joined: February 21st, 2003, 3:16 am

multi-regression question

August 11th, 2006, 1:00 pm

QuoteOriginally posted by: ntruwantThe most easy rule is retaining variables that have a t statistic of at least 2 . But you need a book indeed to go further than this (check for multicollinearith,...)Eliminating parameters with t-stat less than 2 is not always a good thing. If there is an economic rationale as to why that variable should be there, then it should be there regardless of what the t-stat is.