st: RE: Methodology question

["Lachenbruch, Peter" <Peter.Lachenbruch@oregonstate.edu>]

This procedure is essentially a stepwise regression and has all the headaches associated with this procedure.  The book by Frank Harrell in Springer series (about 2002) discusses this.  You might also investigate the lars command, but you need to read up on variable selection methods a bit.

Tony

Peter A. Lachenbruch
Department of Public Health
Oregon State University
Corvallis, OR 97330
Phone: 541-737-3832
FAX: 541-737-4001


-----Original Message-----
From: owner-statalist@hsphsun2.harvard.edu [mailto:owner-statalist@hsphsun2.harvard.edu] On Behalf Of Prasad Ramani
Sent: Wednesday, June 16, 2010 10:10 AM
To: statalist@hsphsun2.harvard.edu
Subject: st: Methodology question

Dear Statalist,

I am new here and my question is quite different from what is normally
asked here. I have a few questions more from an application point of
view.

The Project
===========
I am analyzing a multi asset class portfolio whose composition has
changed over the years from mainly equities to a mix of equities,
fixed income, hedge funds & private equity. The objective of the
analysis is to find which risk factors the portfolio is exposed to and
how to hedge them. The data is a monthly series of returns of this
portfolio for the past 7-8 years.

My Proposed Methodology
====================
1. Get monthly returns for a list of indices that represent the major
asset classes: For equities: SP500, MSCI World etc., for Fixed Income:
BarCap US Aggregate Bond fund, JP Morgan Emerging Market Bond index,
for Commodities: Gold, Oil, for Interest rates: 3 month LIBOR etc. I
end up with about 15 such factors...Factor 1 to Factor 15.

2. I come up with a correlation structure for these 15 factors based
on weekly/monthly returns going back to about 3 years.

3. I regress the returns of my portfolio against these 15
factors...and based on the t-stats of the factors and the overall adj
r-squared, I eliminate those factors that are insignificant at 5%
level.

4. I expect the ones with low t-stats to be highly correlated to some
other factors...and this can be verified from the above Var-Covar
matrix (point 2)
Finally I end up with those factors that have significant t-stats,
F-stat and adj r-squared.

I would really appreciate if you can give me your views on this.

Many thanks,
Raman
*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/

*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/