0
$\begingroup$

My company's multi-asset fund has been using risk metrics methodology to calculate ex-ante VaR and tracking error for years. Due to hardware limitation, the calculation only reflects active risk from asset allocation (e.g. O/W S&P 500 by 2ppt), assuming our portfolio managers largely replicate the benchmarks.

However, my boss is concerned about the active risk from security selection. He tasked me to explore whether it is possible to combine the ex-ante active risk from asset allocation with the ex-post active risk from security selection to estimate the total risk and tracking error. My question is whether this addition makes sesnse? What is the assumed correlation between the two active risk measures(i.e. 1 to be prudent or 0 to be realistic)?

Kindly advise if you have any ideas or come across any literature on this. Thanks

$\endgroup$
  • $\begingroup$ I think what you are interested in is generally discussed as "Risk-based performance attribution" or "ex-ante performance attribution". Try googling some of these terms. $\endgroup$ – noob2 Aug 31 '18 at 2:06
0
$\begingroup$

If you don’t find any better codified solution in literature, I think you could try to run a factor analysis of the securities in your portfolio, aggregate the factor sensitivities into the portfolio sensitivities to factors, and see how you are exposed to factors and what is the exposure of the reference benchmark. Then you can try to see how that works on a validation sample of returns (different from the sample used) to check whether it is reliable enough in explaining ex post tracking error. If so, you have a closed-form expression to map the risk factors sesntivities, so that you build a bridge between ex ante portfolio composition vs benchmark and ex post differential returns, provided that your model is good enough.

This is a suggestion in case you do not find anything you like more in past literature, so search literature before.

$\endgroup$
  • $\begingroup$ What you said is interesting but when you say "factor analysis" are you talking about a vendor type factor model such as Barra or actually doing your own multivariate statistical factor analysis ? I ask because I find that the term "factor model" is often used without description and I always wonder what is being referred to. Even if one restricts oneself to the financial type so stays in the finance factor model world, there are probably so many factor models just in there. Or maybe it doesn't matter for the OP's question ? Thanks for your comments and interesting suggestion. $\endgroup$ – mark leeds Dec 29 '18 at 18:17

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.