I have frequently encountered things that have been implemented, that just seemed like SUCH good ideas at one time. However, often we later discover that what may have, indeed, seemed like a good idea isn’t, in reality.
This post will focus on a few of cases where concepts seemed like a good idea in the world of performance measurement, but we’ll start with items that are TOTALLY non-performance, just to set the mood.
A good idea for the bathroom…maybe
We’ll start from the bottom and work our way up. And what better “bottom” than the bathroom? Specifically, public men’s rooms.
In the United States, it is quite common to find urinals at two different heights: one rather low and others a bit higher. And so, the question is: why are there lower urinals in men’s rooms?
Well, the obvious answer is for children, right?
But, that’s not the reason.
It’s for people with disabilities.
But, seriously: have you (I’m addressing men; sorry) ever seen a man with a physical disability (e.g., wheel chair bound) use one? Probably not. Oh, well. But someone thought it was a good idea and so they’re required. Lovely. Well, at least young boys can use them.
That center brake light you have near your car’s rear window … what’s it for?
For a number of years the U.S. Federal Government has required cars to have a brake light centered, in the vicinity of your automobile’s back window. Why?
It’s so that if the car in front of the car that is in front of you puts their brakes on, then you will be able to see it through the windows of the car in front of you, alerting you that you’ll soon have to stop, thus avoiding or reducing rear end collisions.
Well, that’s the reason, and it’s based on a theory, which turns out not to hold in most cases. Just try to see if you can, in fact, observe the brake lights of the car that’s in front of the one you’re immediately behind. Given the various sizes of cars, it just doesn’t work.
But, manufacturers must have them, meaning higher costs to the consumer.
Now, let’s talk about cases in performance where what seemed like a good idea actually wasn’t!
Why not link monthly standard deviations?
In 1992, as the AIMR-PPS® (Association for Investment Management & Research (the predecessor to the CFA Institute)-Performance Presentation Standards (the predecessor to GIPS®)) introduced a method to link monthly standard deviations. What a great idea? Since the goal is to report measures of dispersion on an annual basis, if we link the composite’s monthly values, we don’t have to worry about accounts being present the full year: excellent!
The original formula was adjusted in a subsequent version.
But, a funny thing happened in 1993, when the first edition of the AIMR-PPS appeared:
the linking formula was missing!
Well, someone figured out that you can’t actually link monthly standard deviations. I actually took this up myself, not too long ago, as I found someone who WAS trying to link. At first, I thought that perhaps they had found a way to do it successfully, but en route to reviewing this, I came upon a reason why it’s impossible.
A (not so) great way to link returns
The Bank Administration Institute (BAI) came up with the very first standard for performance measurement (1968). They identified three ways to calculate returns:
- the Exact method (where you revalue for all cash flows)
- the Linked Internal Rate of Return (known today as “Modified BAI”)
- through regression (no one, NO ONE does this; it’s too complicated).
Okay, so two out of three isn’t bad, right?
And so, we now have returns for individual periods. In the case of the Linked IRR, we have monthly returns; and, for the Exact method, returns between cash flows.
Now the question was, how to link them?
Today, we take it for granted that we would geometrically link, but apparently this idea didn’t exist in 1968, so a method was needed.
I can imagine the conversation:
- How can we link the subperiod returns?
- Well, each subperiod can be of different lengths, right?
- Yes, so what?
- Well, shouldn’t we take the length of time into consideration?
- I’m not sure: why should we?
- Well, shouldn’t the return for a longer period count for more than one for a shorter period? For example, if there’s a return for three months, shouldn’t that count for more than one for just three days?
- You know what, you’re on to something … great idea!
To me, this makes absolute intuitive sense: why wouldn’t we want to assign more weight to periods that were longer?
In fact, when we discuss geometric linking in our Fundamentals of Performance class, students often wonder why we don’t. For example, consider the linking of these period returns:
- December 2015
- July 1 – 10, 2017
We have returns for, respectively:
- One month
- One year
- One quarter
- One quarter
- 10 days.
Some students are amazed that we link these returns without taking the period length into consideration. But we don’t!
But the BAI thought it made sense to do this, and so, they came up with a method by which the lengths were taken into consideration; they called it “time-weighting.” And, in fact, this is what the term means.
Sadly, sometimes intuition fails us. Their method doesn’t work properly: geometric linking is the way to link subperiod rates of return.
And, the fact that NO ONE links this way doesn’t matter: the term, “time-weighting,” remains. (please checkout an earlier post to review this further).
Composite returns: why asset-weight?
The Global Investment Performance Standards (GIPS(R)) require compliant firms to asset weight the underlying account returns to derive their composite returns: why?
We need to back up to the AIMR-PPS, which is mentioned above. Actually, more correctly, the Financial Analyst Federation’s (FAF) draft standards that were first issued in 1987. Both required asset-weighting of returns. This was so that the return would supposedly look like that of a single portfolio.
But why would we want this? Aren’t averages normally equal-weighted (putting aside for a moment the variety of ways to derive index measures). Imagine if we wanted to measure the average height of boys in a classroom, but asset-weighted heights by their individual weights! What would that result represent?
If you are thinking of investing with a manager, wouldn’t you want to know how the firm did, on average, for their clients?
When the AIMR-PPS was being finalized, there were two groups that vehemently opposed asset-weighting: the Investment Managers Consultants Association (IMCA) and the Investment Council Association of America (ICAA). Both felt that by asset-weighting, larger accounts might be favored, since their results would count more.
As a GIPS verifier, I have, on occasion, found composites with several small accounts and one quite large one. Often, the composite’s return matches that of the large account, even though the smaller might have returns that are different.
Was this idea from the mid-80s really a good one? Should we re-examine it to see if it really was a good idea? I think so, but have great confidence that this won’t happen: asset-weighting will remain.
Speaking of “asset-weight,” why standard deviation?
It was in the second edition of the AIMR-PPS (1997) that the requirement to include a measure of dispersion within the compliant presentation was introduced (talk about overuse of the passive voice!).
This was a great idea, since a return doesn’t really convey much if we don’t know how consistent the manager was in executing the strategy.
And while a variety of methods were offered (equal-weighted standard deviation*, high/low, range, quartiles), the measure the AIMR-PPS Implementation Committee favored was asset-weighted standard deviation.
I can imagine what the discussion might have been like:
- we require asset-weighted composite returns, right?
- well, doesn’t it make sense that the dispersion measure be asset-weighted, too?
- perhaps; sounds intriguing
- what if we find a way to “asset-weight” standard deviation? Wouldn’t that be better than the traditional way to calculate it?
- I think you’re right!
And so, they found a method; well, the initial form was found to be unsound, so they revised it. And, asset-weighted standard deviation was born, with the support and encouragement of the AIMR-PPS Implementation Committee. The result was that most compliant firms adopted it.
Now, fast forward 20 years … we find that GIPS does NOT recommend it. Granted, it’s permitted, but the enthusiasm for it has waned a bit. And why might this be?
Perhaps because the result is not interpretable. Unlike the “equal-weighted” form, where, for example, plus/minus one standard deviation encompasses roughly two-thirds of the distribution, there is no way to assess what the asset-weighted version result means.
The result means nothing; well, at least nothing that we can articulate. When speaking with someone about your results, what’s there to say?
Another case of “it sounded like a good idea, but …” I always recommend to our GIPS verification clients that they abandon this method.
Have any more?
There are more, I’m sure, but I’ll stop here and invite you to suggest others and/or comment on these. Thanks!