Some individuals who have read my materials argue that the aggregate method is, in fact, the superior method. They find that the case, for example, where all the composite's accounts have identical 4.00% returns, but where the composite has a different return, to be perfectly fine. In fact, that it is the correct return, and that the BMV+weighted cash flow approach to be the one in error (even though it matches the composite's).
I must confess that I found this all quite perplexing and, to some degree, shocking. However, while corresponding with some folks on this topic I had yet another epiphany. I believe there is something more foundational / fundamental going on here, which requires some thought. There is a point which is very important to consider: What is the composite return supposed to represent? I suggest that you take a moment to reflect on this before continuing to read this post.
If you want to check the definition that is in the GIPS® (Global Investment Performance Standards) glossary, you'll quickly discover something:
In reality, there are two different answers, depending upon which formula you use to calculate it:
- if you use the aggregate method, the composite return represents the return of the composite, as if it was an account itself.
Space doesn't permit me to go much further on this subject here, but I will comment at greater length later this month in our newsletter, so please stay tuned. I am of the firm believe that this is a fundamental issue with the standards, which has been overlooked by the GIPS Executive Committee, and its predecessor groups. I, too, hadn't really given it much thought until recently. But when someone firmly says, in response to my criticisms, that "no, the aggregate method is the most accurate method, and the BMV+WCF is actually inferior," I was forced to pause and reflect.
p.s., As you no doubt realize, the GIPS standards grew from the AIMR-PPS®. If you take a look at the first edition of the earlier standard you will only see one method to derive the composite return, and it is based solely on the beginning market value. The other two were added in the 1997 editions. I suspect that when this was done, no one on the AIMR-PPS Implementation Subcommittee realized that they were at that moment introducing a second definition or meaning for the composite's return. But they were.
.png)
I believe that the composite return is supposed to be an average of portfolio returns. I thought this was the common understanding, but perhaps this is not the case. We need to serve the various users of this composite return information without conflicting definitions or unnecessary calculation processes. For example, we can use the same set of portfolio returns to generate both asset weighted and equally weighted composite returns, as well as answering questions on the number of portfolios and the level of assets in the composite, along with information about dispersion of portfolio returns within the composite. You cannot do any of this with an aggregate method for calculating composite return. That being the case, why use an aggregate method at all? At best, it produces yet another answer to the asset weighted composite return that serves no unique purpose. Worse than being a waste of resources, this produces confusion as to what is the “correct” composite return. Seems that the aggregate method gives us more work, higher cost, some degree of confusion and the need for additional disclosures. All for the promise of greater “precision” in the composite return? No thanks. Haven't we got better things to do?
ReplyDeleteSteve, I agree: the composite return SHOULD reflect the average result of the underlying investors. After all, the manager is managing THEIR money; the composite isn't anything, per se, that is being managed. I believe that the crafters and champions of the aggregate method firmly believed that it produced more accurate results.
ReplyDeleteI recall speaking with a sales manager for an accounting system vendor around 1993, who told me of their approach (aggregate), and I said it wasn't permitted, since it wasn't in the standards. She informed me that it was, in fact, the BEST way to measure the results. I believe many people believed and still believe this.
But, it's measuring the wrong thing. "Precision" has nothing to do with it, if we're measuring the wrong thing. We can be accurate to 1/1,000,000 of a basis point, but who really cares if it's measuring the wrong thing.
Hopefully some others will take this matter as seriously as you and I.
I do hope that someone will present a thorough examination of the practical aspects of using an aggregate method for calculating the average return for a group of portfolios managed within a specific style (i.e. a "composite" return.) There seem to be several questions and problems that arise from the introduction of an aggregate method. These include:
ReplyDelete1. Do you have to duplicate your efforts by also calculating individual returns so that you can create an equally weighted composite return? That seems like a waste of time and effort.
2. Do you have to calculate individual portfolio returns to comply with the requirement to show the dispersion within the composite? (I think you do.) Does this also require an explanation of why the aggregate return is different? Will this erode confidence among clients because we have two answers to what they will see as the same question?
3. Has anyone thought of the additional burden on investment managers who now have to aggregate the cash flows from all of the portfolios within a composite? Right now, most systems are set up to produce individual portfolio returns, not the return from a net cash flow from a set of portfolios. This is not a trivial effort and cost. Shouldn't rule makers be required to assess the cost of new regulations, especially when there is no problem with accuracy of the existing methodology? Or is the simple claim of "greater accuracy" enough to add this burden onto the backs of those who wish to comply with GIPS?
4. Accounts entering and exiting composites is a relatively infrequent occurrence, and so it seems unlikely that there will be a material difference in composite returns using either the weighted average approach or the aggregate method. Has anyone assessed the degree to which changing membership in a composite is a frequent and significant enough occurrence to warrant the introduction of another calculation methodology? Or is another rule being mandated on the basis of nothing more than a few exaggerated hypothetical scenarios? It seems that this is a solution that's looking for a problem that doesn't exist.
Steve, you raise quite a number of interesting questions. As to whether anyone has done this analysis, my guess is probably not. And, based upon the response to the analysis I did, there is little likelihood that it would result in any serious consideration that what we do might, in fact, be wrong. I find this all a bit troubling. Perhaps there's a sense that "we have bigger fish to fry" or "we've always done it this way" makes any consideration of change unwelcome. It might also force someone to say "we were wrong," which is very difficult for most folks to do. I think I'll blog about this point, too.
ReplyDeleteYou may recall that in the pre-1993 days, AIMR published a method to link monthly standard deviations to arrive at an annual value. They actually then came out with an errata for it. When the '93 standards were published, I couldn't find this formula, which struck me as being quite odd. I pursued it and learned that "it had been dropped" from the standards, because it was later determined that standard deviations cannot be linked, and that the original idea was flawed. But there was no formal statement to this effect made.
This caused some confusion at the time, because several software developers had actually incorporated these formulas into their systems (I, for one, had developed a composite system for a client which had this method built in). What would the harm been to formally state "we were mistaken in thinking that blah blah blah"?
And so, to hear from the EC that the aggregate method produces a return which really has no place within GIPS, and the idea that the return should represent a composite if it was a single account doesn't actually make any sense most likely won't occur in my life time. But, given the encouragement I've received from various folks, I will continue to raise this issue.