Why long/short attribution is difficult

112612 port man
Alexandre Voitenok (left) is director, long/short strategies, and Rui Tang is a vice president, associate portfolio manager, at Acadian Asset Management LLC, Boston.

Every client invested in a market-neutral portfolio wants to know how much value is added by the short positions. Answering this question is much harder than it appears. Indeed, there is no commercially available tool that yields a thorough, unbiased and widely accepted analysis of the contributions from the short side.

There are three main issues that complicate the attribution of returns to shorts.

The first is related to the fact that, philosophically speaking, short positions are not independent from long ones. It is often difficult to spell out why some shorts exist in a market-neutral portfolio. It could be that short positions are taken so the overall portfolio is country, sector and size neutral. In other words, the shorts serve to balance the longs in the risk dimension. It could also be that the shorts have negative return forecasts. Often a stock is shorted for a combination of negative forecast and risk reduction purposes, given all the desirable longs. If the shorts perform no worse than the benchmark, they might seem to add no value, but it is precisely those shorts that allow the overall portfolio to own the promising long positions.

Even if investors ignore this issue for now and treat the short side as a stand-alone portfolio, they still run into the same problem that arises with performance attribution for long-only portfolios. This relates to potentially substantial distortions in size, sector and country distributions on the short side of the portfolio. These distortions are often more pronounced on the short side than on the long and are caused by a variety of factors, ranging from intentional biases in allocating capital to certain segments (such as market cap or sector) based on the manager's stock-picking ability, to stock loan availability.

It is even harder to find an appropriate benchmark with which to compare these shorts. An inappropriate benchmark would give an incorrect attribution. For example, if the short side has less exposure to financials than the benchmark, but financials underperform, this will make the short side look bad. However, this ignores the fact that the long side of the portfolio could have an even smaller exposure to financials, creating an overall net short bias, which would benefit the overall portfolio. Finally, the portfolio might have no exposure to a particular segment at all, because of the decision to allocate to other segments where stock-picking skill is believed to be higher.

The third issue is the difficulty in coming up with an approach for choosing a benchmark that resembles the portfolio in question. In the long-only case, if the portfolio consists of small caps, then the investor would use a small-cap index such as the Russell 2000 as the benchmark. But what if the portfolio has no financials? What benchmark should be used in that case? Even if the investor finds a benchmark with the same industry and country exposures as the long-only portfolio, what if they have different size exposures? The investor needs to match all the characteristics of the benchmark and the portfolio, whether these relate to country, sector, industry or size. As the number of characteristics increases, the benchmark will gradually overlap with the portfolio.

To summarize, long/short portfolios have a structure that makes benchmark selection challenging for investors seeking to produce, say, Brinson-type attribution for separate long and short positions. The challenge arises from conceptual differences between long/short and long-only portfolios, mainly in the allocation space. In other words, in the absence of an explicitly specified benchmark, long/short portfolios tend to allocate positions to segments (i.e. countries) based on perceived opportunity/risk/cost trade-off. While net long/short exposures might look very reasonable from the overall net exposure viewpoint, and be similar to active exposures of a long-only fund run against the appropriate benchmark, individual long and short portfolios probably will look nothing like the benchmark. The problem is further exacerbated by leverage.

We believe there are a number of ways in which this problem can be tackled. The investor could choose a benchmark that approximates the long/short portfolio based on a few key characteristics, such as country and size. This benchmark could then be scaled to match the portfolio's leverage, producing more meaningful allocation/selection results. While still not necessarily very intuitive (and highly sensitive to benchmark selection), these results become more meaningful when netted for the long and short sides.

A less obvious approach that tends to produce more intuitive results separately for the longs and the shorts is based on creating an “all-opportunities” benchmark. Generally speaking, such a benchmark could be constructed from the entire opportunity set available in the portfolio construction process. This set includes all stocks theoretically eligible for inclusion, after they have been screened for liquidity and other eligibility criteria. It is then weighted according to the expected liquidity profile of the portfolio, often along the lines of natural logarithm or square root of market cap (to account for greater liquidity in large-cap stocks, while avoiding an excessive overweight in megacap stocks). While still not fully addressing the issue of short positions existing only to offset some longs in the risk space (or vice versa), this creates a benchmark that one would reasonably expect to underperform the portfolio's longs but beat its short positions.

Such an approach also produces a reasonably meaningful allocation/selection picture, at least as it pertains to the process of correctly choosing positions from the available opportunity set, on both sides of the long/short alpha dimension. A further refinement of such an approach is particularly applicable in market-neutral framework, where the portfolio manager is not expected to be taking sizeable net exposures to any market segment. In such framework, by applying certain transformations to realized stock returns, it is possible to produce a definitive analysis for the stock selection component even in the absence of a well defined benchmark, thus answering a key question: “do our shorts add value?”

The non-trivial task of disentangling the performance of longs and shorts demonstrates just one of the distinctive challenges facing long/short managers. In our view the challenge is still not widely recognized but answers are certainly available and indeed they are essential to a successful long/short operation.

This article originally appeared in the November 26, 2012 print issue as, "Why long/short attribution is difficult".