How do you determine how many different indices you use when performing a returns based analysis? Saw a few questions on this but still hazy on it.
5 criteria: - covers most liquid investments - covers most investable assets - homogenous - uncorrelated - no overlap
Yeah I get that part… That’s not what I"m referring to. THere are a few questions where they use: - Large Cap Universe - Value - Small Cap Universe - Value - Mid Cap Universe - Value and then the same 3 again but Growth and they use all 6 indices to determine if there has been style drift.
If you use LCV, LCG, MCV, MCG, SCV, and SCG you are going to do a pretty good job at covering your bases. You would then look at the results of the RBSA and determine how much % of portfolios return is from LCV --> SCG .
Same, answer, with more explanation. BW is right, those six will do it for you most times, unless you start getting into finer breakdowns and geographical categories. Remember what you’re doing here, it’s multiple linear regression, and you’re trying to figure out how many independent variables. If you go back to L2 Quant, there’s an explanation there of how you go from not having enough variables to having too many. In the former case, R^2 is too low, and in the latter, you get intervariable correlation and other “bad” stuff that screws up your regression. Don’t remember all the details but it still boils down to the five things I said in my first post.
Remember you are fitting the indices to Numbers. Of course it doesnt know what you are holding so it may come out that 5% of the return is due to Cash, but you might not have any cash and wonder why…well if you have high dividend paying holdings this will make it appear that you were holding cash. So its fitting numbers into numbers. They have tried to do this with Hedge funds, to some success but the problem is you really need to know what they are holdings first before you can do a RBSA.
make sure that you don’t choose the similar index(duplication). it should be exhaustive enough but not duplicative…