I’ve been reviewing some points to make sure i got them right, and it seems that schweser pulls out some ‘concept’ out of nowhere. For instance “Optimization leads to lower tracking risk than stratified sampling”. I couldnt find confirmation of this in CFAI.
is that because optimization is often rebalancing?
Optimization I believe is more akin to say Monte Carlo and Stratified Sampling to MVO. I know they are not the same, just using them as comparisions for differences. Now, I believe Optimization uses a multifactor model to determine which securities will best track and represetn the index where as Stratified Sampling just breaks up the index into cells and then picks a stock or two out of each cell to represent that cell of the index…
Stratified sampling doesnt acount for correlation, so i would say it is not akin to MVO
whether it’s in CFAI or not, it’s right. stratified sampling is a “back of the envelope” type approach. you create a matrix and place your bonds / equities / whatever in it and pick those that appear representative of each cell, weight them appropriately in the portfolio, etc. optimization is a quasi-scientific approach that matches the factor risk exposures with a multifactor model and accounts for the correlation. how could it not have lower tracking risk? it specifically identifies and matches risks, hence optimization – more return per unit of risk.
CSK I was just using it to say Optimization is to Stratified Sampling as Monte Carlo is to MVO.
bigwilly Wrote: ------------------------------------------------------- > CSK I was just using it to say Optimization is to > Stratified Sampling as Monte Carlo is to MVO. reminds me of my SAT days. I scored like 400 on verbal! Good stuff good stuff
That’s what i was going for, glad you picked it up… yeah I sucked at verbal too, only around a 600 i think
I scored 440 on verbal. But anyway I think Schweser is right here. Because optimization is factor-based, it accounts for the covariances of the factors (ie it makes sure they are uncorrelated: industry, size, etc.) The sampling method only assumes this and therefore is subject to the correlation problem.