While the role performance attribution has continued to grow within the investment industry, it is, in the words of "Brinson, Hood & Beebower," still evolving. And in its evolution, there remain areas that need to be re-explored, reconsidered, revisited, and rethought about (a lot of "re"s). Anyway, I thought it might be fun to identify some key points that firms should consider.
1. Contribution is a form of attribution. While there are some who will object to this, many of us believe that contribution is essentially "absolute" attribution, as it tell us how the various parts of a portfolio contributed to the total return. It is a commonly found statistic that often accompanies both client and prospect reports. Ideally, a "transaction" based approach should be used, to eliminate the presence of residuals and to ensure accuracy in the results. In addition, it's common to show the "top" and "bottom" five or ten; not typical to see all securities listed. In addition, it can be applied against sectors and virtually any other ways to carve up the portfolio, in order to provide valuable insights.
2. Get the model right -- it needs to align with the investment approach. Make sure the model you use ties into the investment approach, to ensure that the results can be aligned with the actual decisions being made.
3. But don't stop there, slice it up even further! To provide even more valuable insights, don't stop with the standard approach, but slice the portfolio up in additional ways, in order to gain more insights. For example, perhaps decisions aren't made based on style or market cap, but by running your attribution model against your portfolio broken up by large, mid, small cap within growth, core, and value, you can discover additional insights. We credit Ron Surz for this approach (in his words, "attribution with style"). There are many other ways to slice the portfolio up, too.
4. Speaking of which ... contribution of securities not held. An interesting twist to contribution is to run a form where you see what impact your decision NOT to hold certain benchmark securities had on your portfolio. Such information can be very helpful when communicating the reasons why you out- or underperformed the benchmark.
5. Have more information available when you underperform then when you outperform. If you can thoroughly and clearly identify the reason(s) why you underperformed, this will often satisfy your clients and prospects, as it shows that you have your "finger on the pulse," and can clearly articulate what occurred.
6. Don't ignore the currency effect. Too often we find firms that invest outside of their domestic market ignoring the currency effect; this is a huge problem, as it ignores a potential major contributor to performance. You can do well or poorly in the local market, and under or outperform the index because of how currency moved. It's the combination of what occurs in the local market to your securities, as well as the changes to exchange rates, that contribute to your performance. Don't ignore this effect.
7. Knowing when to use a sophisticated currency model. There is a fairly "naïve" currency model which gives just one figure. This approach is fine if you're not engaging in currency forwards, etc. I.e., if your portfolio is "naked" to the effect of currency movement. However, if you make currency bets or do any level of hedging, then you want a more sophisticated model (e.g., Karnosky-Singer) in order to bifurcate currency's effect into the change of the FX rates on your holdings as well as your use of derivatives.
8. If you're using a holdings-based model, seriously consider switching to a transaction-based one. Research I've done shows the ever presence of residuals when firms use holdings-based models. But worse, when using a holdings-based approach I found that frequently the signs of the effects will be opposite of what they should be (e.g., a negative selection effect when it should be positive). I plan to publish these details later this year, and believe it will add great credibility to abandoning the holdings-based methodology.
9. Treat the interaction effect with respect. While I realize that explaining what the interaction effect is can be a challenge, to simply place it with the selection effect is, in my view, improper, as it will occasionally give credit when credit belongs to allocation, or fault, when allocation is the cause. If you don't want to show interaction, then place it where it belongs. I wrote an article on this some time ago, and believe that such analysis can be easily incorporated into an attribution model.
10. If you're going to go below one level in your model, make sure you do your math correctly. Multi-level, multi-layer, nested attribution, if not implemented correctly, can either (a) result in incorrect results or (b) not have the sum of the parts equal the total. Steve Campisi, CFA developed a model that properly handles this type of analysis. He and I plan to do a follow up article, hopefully in the coming months, which will go into further detail on how to implement Steve's approach.