Tuesday, June 5, 2012

Why security level attribution?

I am working on my doctoral dissertation (hoping to have it completed by the end of the year), and wanted to cite comments I've made regarding my questioning of the value of having security level attribution, but can't find anything in writing. This doesn't mean I haven't written on it; just that I can't find it. And so, I decided to briefly pen my thoughts as a start; more details will be presented in this month's newsletter.

Sector, industry, portfolio attribution I get: we want to capture our allocation and selection decisions (for the purpose of this post, I'm limiting my discussion to equity attribution, though similar arguments can also be made in other asset classes). The manager looks at the index and decides to overweight some sectors and underweight others; and so, we need to capture the impact of these decisions.

But how many managers actually allocate at the security level? Perhaps overweight within the portfolio, relative to other securities they hold, but relative to the index?

Let's say that the manager is using the S&P 500 where we find that the consumer discretionary sector has roughly 80 securities. Chances are the portfolio will hold much fewer securities in this sector. One of them is Best Buy, which is in the index, and it makes up 2.5% of the portfolio. It's about 0.15% of the index. Does this mean the manager overweighted Best Buy? No, it means that he has much fewer consumer discretionary stocks and stocks in total, and is therefore able to invest more in them than the index, which has 500 stocks. But if we run attribution at this level as we do at the sector, we'll get a contributor for the overweighting; does this make any sense?

Because the portfolio holds much fewer stocks than the index, it won't hold most of what the index has; does this mean the allocation is zero percent? Well, technically this may be, but isn't it really about selection?


  1. Stephen Campisi, CFAJune 5, 2012 at 9:42 AM

    I am pleased to see your comments on so-called "security level attribution." This is an area of performance analysis that really needs better definition. We often hear claims of "sector level attribution" without a clear definition of what this actually means, much less what value such alleged information provides. It's sad that performance analysts, writers and vendors perform detailed work in this area without having first established so much as a clear definition of the term.

    I believe it would be most helpful for everyone to remember to relate their attribution analysis to the investment decision making process. After all, this is what attribution is supposed to evaluate. When investing, there are only two real decisions that are made - the first decision is about the allocation of capital and the second decision is about the implementation of that allocation decision. The "allocation" decision addresses capital committed to asset classes, sectors, sub-sectors and also to both risk, style or other structural factors. The challenge is in preserving the unique ("orthogonal") nature of these characteristics, since these often overlap. It's also difficult to distinguish a deliberate decision from an unintended consequence of a different decision - that's why this is challenging! Nonetheless, we clearly see a decision or a set of decisions around the allocation of capital.

    The second decision relates to how to turn this plan into an actual portfolio, and that is the job of execution. This is what most performance analysts are referring to when they use that loosely-defined term "selection." Obviously, the portfolio manager must select specific investments which represent the allocation strategy. In doing so, we deliver the structural characteristics of the strategy (what some call the "beta" return.) We also see the introduction of an idiosyncratic component of return, also called an "alpha" return (although the better term is probably "excess" return, since this does not imply that the additional return has been adjusted for differences in risk.)

    So, how does this actual investment process "get attributed" at the security level? I believe that it doesn't - this is simply definitional because each attribution effect is unique. But perhaps we can stretch the definition and say that each issue makes its contribution to the allocation of capital, although I don't think this conveys any meaningful information about the selection process. This is what happens when you naively calculate things without a clear understanding of the content - everything gets muddled.

    In the end, performance analysis answers two questions: how effectively did we allocate capital, and how effective was our execution? This makes sense at the macro level; it makes no sense at the issue level. Issue level attribution only makes sense in the context of contribution to return. Frankly, "That's all she wrote..."

  2. Steve, thanks for your commentary. I am confident that a great deal more can be written on this subject; perhaps we can collaborate on something!

  3. The way that I look at attribution at the security level is arguably meaningless. Consider each security to be it's own "group". So, within each group, "selection" will be zero for every group, and "allocation" across groups will sum up to the active return (portfolio minus benchmark).

    Where this does become meaningful is within an education framework, where I've taught how the number of groups in an attribution will have a bearing on the amount of allocation versus selection. I use two extremes -- first, the case above, where every security is its own group, and thus allocation picks up all of the basis points. Second, where there is only one group -- say "equity" for an equity portfolio -- where Allocation to equity in both the portfolio and benchmark are 100%, so Selection picks up all of the basis points. The lesson is that if the number of groups is very close to 1, then it's mostly Selection, while if the number of groups is close to the total number of stocks in the benchmark plus the portfolio, then it's mostly Allocation.

    I should pause for a second and point out that the best Attribution framework for a portfolio is the one that reflects the investment process. The lesson above is simply academic, and my portfolio managers had no interest in either extreme. (Incidentally, they were equity managers that were heavily bottom-up but with a top-down feedback loop -- so at best we could force a Brinson-style approach, but it was a force.)

    That all said, we did use three other attribution-like approaches, which we called "Absolute Contribution", "Active Contribution", and "Stock Picking". The Absolute Contribution measured the individual stock contribution to the portfolio's absolute return (independent of benchmark). The Active Contribution viewed every stock in the portfolio and benchmark as a relative weight (overweight, underweight, or rare same weight), and multiplied by the relative weight by the stock's benchmark-relative return -- so unheld stocks that were large in the benchmark and underperformed the benchmark would positively contribute to the Active Contribution. Stock Picking falls between the two Contributions -- Stock Picking measures the absolute weight in the portfolio times the stock's benchmark-relative performance -- stocks in the benchmark but not in the portfolio will have no contribution. Stock Picking takes the perspective that the PM is trying to outperform the benchmark with each and every stock (rather than outperforming the benchmark sector return) -- and that any sector allocation is a second-order reflection of risk control and the "bunchiness" of attractive stocks across sectors.

    Hope this is useful -- I enjoy reading your blog.

  4. Dorian, thanks for your note; sorry it didn't get posted sooner (problem w/blog). Interesting points, and no doubt more to be offered. Thanks!


Note: Only a member of this blog may post a comment.