I found Alexandre Voitenok and Rui Tang's Portfolio Management article (Why long/short attribution is difficult, Nov. 26) quite interesting. I agree with their suggestion that attribution for a long/short portfolio is more complex than against a long-only portfolio. However, there is hope to those wishing for a proper attribution analysis.
Jose Menchero wrote an article for The Journal of Performance Measurement (Performance Attribution with Short Positions, Winter 2002/2003) that addresses this topic in a very systematic, detailed, and well-thought-out manner. I cannot attempt to replicate Jose's model here, but will touch briefly upon a couple key points.
First, it's important that the longs and shorts be kept separate. Unfortunately, many portfolio accounting systems fail to do this. These positions are often netted, which can be problematic. In the event that a manager is short against the box, (i.e., going short against securities they already own, without selling those securities), for example, netting will dilute the ability to isolate the impact of the longs separately from the shorts. Netting can also cause the portfolio to be long a security at the start of the period, and short at the end: While deriving the return for the period isn't impossible, it's made more complicated.
Second, as for the benchmark dilemma, one could argue the same benchmark that is used for the long side should be used for the short. The attractiveness of shorts is that a manager that is bearish on a sector can, if he/she is correct, do better than simply being flat (i.e., have no investments in it), by going negative through shorting. As for a benchmark not having sectors that the portfolio is invested in (or vice versa), the Brinson models can handle these cases with some minor adjustments.
If the manager truly has no benchmark they're investing against, then traditional Brinson models won't work, meaning contribution analysis would be in order. Here, we would still want to segregate the longs from the shorts, so that we can fully understand how each contributed to the overall return. Hedge funds should generally use contribution (what some call absolute attribution) rather than relative attribution, since there aren't indexes that they actually manage against (a requirement for relative attribution).
This is a great topic that's worthy of more attention. Jose's article is an excellent starting point for developing a model that will serve the needs of most managers.
Publisher, The Journal of
President, The Spaulding