What to attribute events to can be a challenge

At last week's North America Performance Measurement, Attribution & Risk (PMAR) conference, I provided an update on research I'm doing on holdings vs. transaction based attribution. Many investment professionals are aware that the holdings based approach typically results in residuals (i.e., unexplained differences between the sum of the attribution effects and the excess return). What isn't widely known is the mis-assignment of attribution effects, which can result in erroneous and misleading reports.

Yesterday we were hosting our immediate family for a picnic/pool party. As I walked outside to take bags of ice to our cooler, I tripped and fell. My wife responded by trying to come to my aid, only to stumble herself. And while I suffered no injury at all, she had a severe sprain. The cause of her injury? My falling, of course, for she wouldn't have fallen had I not done so myself.

Knowing where the blame or credit goes is important. What we don't need are reports that provide the wrong assessments.

Concerns with holdings-based attribution

On my "to do list" is the task to write an article an article for The Journal of Performance Measurement(R), detailing some of my findings from research I'm doing regarding the impact of trading on the accuracy of the holdings-based approach to performance attribution. Recall that firms can either (a) use a "buy and hold" approach, that only uses the starting position weights, ignoring any intraperiod activity (holdings-based) or (b) begin with the starting position weights, and then adjust them for trades, income, corporate actions that occur across the period (transaction-based).


We have known that the use of holdings-based models can result in a "residual." Let's briefly speak about this term. A "residual" is a non-zero difference between the sum of the attribution effects and the associated excess return. It can occur in two ways: across periods, as the result of linking arithmetic attribution subperiod effects (geometric attribution doesn't have these residuals) or within a period, by the use of a holdings-based model.


And so, we recognize that there's a flaw in using just the starting holdings and ignoring the activity. But, as with the "across periods" approach, there is often the assumption that the error is proportionate to the results, meaning that one could smooth the residual across the effects without encountering much of an error. My research has shown that this isn't the case. In fact, there's a second, more significant problem: the misassignment of effects. That is, we can have, for example, the allocation effect reflecting a totally incorrect value (e.g., negative when it should be positive).


My research will continue for the next several months, and I hope to have something that speaks to this in much greater depth later this year. In the mean time, I will provide an update on my most recent findings at this year's Spaulding Group PMAR conferences.

Do the numbers truly represent what they're supposed to?

Last week I had a post about holdings-based attribution, where I laid the groundwork for future commentary on analysis I've been doing on this subject, where I look at the results using both holdings and transaction-based methods. Well, it resulted in comments from my friend and colleague Andre Mirabelli, who questioned the fundamental model's ability to properly evaluate the attributes that produces the excess return. While not wishing to debate him on this topic, per se, it does bring up a broader question that is worthy of consideration.

When portfolio managers, prospects, and clients look at the numbers on a performance report, they draw various conclusions; and the folks who produced the reports no doubt hope that these conclusions are consistent with what they hope would be drawn. However, is everything working as it is intended? Just because a computer has run a particular model, which causes numbers to be produced, which then are assembled in a nice format on a piece of paper or a computer screen, does it mean that everything is correct?

The following graphic will be the basis for what we'll discuss today:

The center set of figures  represent the process that we typically employ: we gather data from a variety of sources; this data is fed into a model, or a series of formulas, and out comes the information, which is presented in reports or on computer screens, iPads, smartphones, etc.

The data issues are decades old: the acronym GIGO still lives on (garbage in, garbage out). One must take strides to ensure that the data is accurate and, of course, appropriate.

The real issue which Andre addressed is the model appropriateness. This is a fundamental issue which doesn't get enough attention. Although I've become less and less a fan of Warren Buffet, I will nevertheless quote him here: "beware of geeks bearing models." And yes, one must be cautious about what models they employ. Do we understand how they work? Do we understand what assumptions they make? What are the results intended to convey?

We recently completed our attribution survey, where we address a variety of issues on this important topic. It has amazed me how over the years, we've seen a shift from folks using the Brinson-Hood-Beebower model to the Brinson-Fachler model. BHB was published a year after BF and in fact is preferred by Gary Brinson. I believe we deserve much of the credit for identifying and communicating the huge difference between the models (through our training classes and various articles, not to mention our books). The late Damien Laker challenged me on this, writing articles which he posted on the Internet, that said that there was no difference; but there is a HUGE difference. If you're not already familiar with it, I'll briefly state that it's how the allocation effect is derived.

Well, folks for years used the BHB model with complete satisfaction; but did they really understand how it worked? Did they understand that there was an alternative, which they might prefer? In most cases the answers are "no." Years ago, when I was first composing my attribution book (which is long overdue for a rewrite), I did a fair amount of research and discovered that many folks simply said "we use the Brinson model." "THE" Brinson model. "The" means that there's only one, as in "THE" president of the United States. It should have been "A" Brinson model, as there are two. But, many developers weren't even aware of this. While the BHB model was published in the Financial Analysts Journal, which meant is was available to tens of thousands of individuals, the BF was published in The Journal of Portfolio Management, which has a much smaller subscriber base, and consequently, less opportunity to be read by the masses.

But even the employment of these models should call into question their appropriateness, given the basic rule that models should align with the investment process. Should a quantitative manager use a Brinson model, that only looks at allocation and selection effects (and, for the more enlightened, the interaction of these effects)? Most likely, no; instead, a multifactor model that looks at the factors they employ in their investment process would make more sense.

I (and many of my esteemed colleagues, such as Stefan Illmer and Steve Campisi) have been on our respective soap boxes for the past few years championing the merits of money-weighting; again, a hugely fundamental issue that is too rarely considered in model development and report production. I often challenge individuals who want me to review reports as to what they're trying to convey. What questions are they trying to answer? If you use the wrong formula, you're producing the wrong result, which can be misleading and not meet your reporting objectives.

This topic isn't a simple one, and covering it briefly in a blog post is impossible (as is obvious from today's attempt, which is only just scratching the surface). Perhaps I'll address it further in this month's newsletter.

By the way, the BF and BHB articles can both be found in Classics in Investment Performance Measurement.

What's wrong with holdings-based attribution?

The Spaulding Group recently completed its most recent survey, which deals with performance attribution. Jed Schneider, CIPM, FRM discussed many of the results at a luncheon held in NYC, where we learned that, as with the prior three editions of this research, most folks prefer transaction-based attribution, though roughly half use holdings-based. The probable reasons as to why the contradiction are interesting, but not part of this post.

For several years, I have attempted to encourage a proper and unbiased evaluation of the two methods, to determine the true differences, and whether there is a point when one cannot justify using the holdings-based approach; a point where the use of transaction-based attribution is a "must." But, no one chose to do the research, so I did. And I discussed some of my preliminary findings at last year's PMAR (Performance Measurement, Attribution & Risk) conferences, and will discuss further findings at this year's events.

Today, I merely want to discuss the three problems with the holdings-based approach. And, in "David Letterman" style, I will do it in reverse order of significance.

Number 3: we will usually have a residual. For many readers, this will seem odd to have been placed third, because surely the residual is the main problem with holdings-based, but I'd say it is not. A "residual" is a non-zero amount which reflects the inability to fully reconcile to the excess return. Recall that with relative attribution, our goal is to completely account for the excess return. However, with holdings-based models we often (actually, more like usually) can't do this, but have a "residual," meaning we don't fully account for the excess return. In reality, it's worse than that.

Number 2: getting the proportions wrong. Here I mean that the amount that is assigned to the different effects may, in realty, be incorrect. We may have too much or too little assigned to allocation, for example. Thus, it's contribution to the excess return is over or understated. This goes beyond merely not reconciling to the excess return; rather, the numbers that are produced can be allocated in a manner which doesn't properly align with reality; their proportions are incorrect.

Number 1: having the wrong signs. To me, this is the major problem. What do I mean? Well, with holdings based attribution, we may be showing a positive selection effect, but in reality, it's negative! And so, instead of saying "great job!" we should be saying "you've got to do better!" And the problem is, you won't know this. You'll see a positive selection effect and conclude that those decisions were good ones that contributed to the excess return, when in reality the decisions hurt your performance!

This third problem (#1, actually) means that the results will be misleading,  spurious, invalid. What's the risk of this happening? I think you'll be surprised.

An article will be forthcoming which will provide additional details on my research. Suffice it to say, the results are fairly startling, and should encourage most users of holdings-based models to seriously consider switching.

p.s., for more details on the attribution survey, contact Patrick Fowler.

p.p.s., our next survey deals with the GIPS(R) Standards (Global Investment Performance Standards). Please join in! It will begin this summer.

Doing transaction-based attribution? Don't forget about the benchmark!

I am working on a research project, where I am creating a variety of portfolios to compare the implementation of transaction-based attribution versus holdings-based. I began this study more than a year ago with my first test portfolio, and am now putting several more together. I chose the S&P 500 index for this project.

To avoid complicating the matter, I began by selecting months where the S&P 500's constituents saw no changes, as I didn't want to be bothered with worrying about capturing those transactions (one or more securities leaving during the month, and being replaced by other securities, possibly altering the weights of the various sectors). The problem is that it's often difficult to find such a month.

And so I decided to discard this requirement, and select any month, even if changes had taken place. The problem was that when I made this decision, I forgot about the need to capture the transactions on the index side.

Almost immediately residuals appeared when no portfolio transactions had occurred: they weren't terribly large (usually only one or two basis points), but when there is supposed to be no residual whatsoever, this posed a dilemma. At first I thought the fault lied with my spreadsheet somehow: that perhaps in loading the data, I had altered a cell's formula. But after some reflection it hit me: the benchmark's composition had changed, and I hadn't accounted for it (ironic, because my initial intention was to avoid this situation, but somehow forgot about it when I decided to forgo the original constraint). In other words, the benchmark sector weights used in the formulas were based solely on what they looked like at the start of the period, without accounting for the changes that occurred during the month. And so, the formula:

fails. And yet, the sector returns are the result of changes in their weights, which aren't captured here. The result: we have a residual.

The solution? Well, in "real life," one would want to make sure you captured the changes in the sector weights that resulted from securities leaving or entering the sector during the month. In my case, because I am evaluating changes to the portfolio, not the benchmark, I will simply alter the benchmark's overall return so that the formula is satisfied: something I have the luxury of doing in my analysis, and arguably a "must" to properly assess what the study is all about. Hopefully, you'll read more on this in an upcoming article.

My point? Make sure that your implementation of a transaction-based model properly captures the benchmark's changes, too; i.e., not just the ones to your portfolios!

Dealing with zero weights

I got an email from a client recently, asking how one should treat the case where a manager has zero exposure to one of the sectors in the benchmark.

If you are using one of the Brinson models and simply use the formulas as written, you'll find that the interaction effect gets most of the credit (or blame) for the decision. But does this make any sense? No!

And so, you may want to shift the interaction results to one of the other effects: selection or allocation.

Now, which do you think it should be? [pause, to allow you to reflect a moment] I think allocation, since the manager made the decision to go void in the sector. Thus, I would move the entire interaction effect there.

I'll address this in greater detail in this month's newsletter.

How to implement transaction-based attribution

I got an email from someone today asking how to calculate transaction-based attribution. I addressed this during our recent Attribution Week, but will touch on it briefly here, and in greater detail in this month's newsletter.

Recall that attribution relies on returns and weights.

The Weights

With a holdings-based appraoch we use the weight at the start of the period. For example, if the portfolio's initial value is 100,000, and Technology has 10,000 as its starting value, then its weight is 10% (10,000 / 100,000).

To have a transaction-based approach, we need to adjust the weight for any buys/sells/income that occur during the month. For example, if we bought another 2,000 at the middle of the month, then we weight the transaction the same way we do with Modified Dietz [(CD-D+1)/CD] and multiply this by the value (and so, 0.5 x 2,000 = 1,000). We add this to the starting value and get a revised weight [(10,000 + 1,000)/100,000 = 11%]
We use this weight in our formula.

The Returns

With holdings-based, we can simply account for the starting values to derive our returns. With transaction-based, I suggest you use Modified Dietz, so that you capture the activity that occurs. And so for our example above, if we ended with Technology being valued at 14,000, we'd have R = (VE-VB-CF)/(VB+w*CF)=(14,000-10,000-2,000)/(10,000+0.5*2,000).
[I'll let you do the math]

That's it! Again, I'll spend a bit more time on this in the newsletter.

Shall we interact?

I had a conversation with a client last week, where we addressed a variety of aspects regarding performance attribution. They asked how common it is for firms to report their interaction effect. While I don't believe we've addressed this question in any of our surveys, my gut tells me that most firms avoid doing so, in order to avoid having to explain it.

What is interaction? All one need to is to look at the formula to answer this: it's the difference in weights (portfolio minus benchmark) times the difference in returns (again, portfolio minus benchmark); and these components represent allocation and selection, respectively. And so, it's the interaction of these two decisions.

Why is it needed or why does it appear? Because the individual effects (allocation and selection) are isolated from one another, so as to not allow the influence of the other to impact its results (e.g., the selection effect is the difference in returns times the BENCHMARK weight, not the portfolio, since portfolio weight would reflect allocation). But this answer fails to mean much to most folks, thus the desire to avoid it.

The topic of whether it should be included or not is often hotly debated, as there are passionate voices on both sides of this issue. I've written an article on the topic, as have others, including my friend, Steve Campisi. In the end it's up to the firm to decide how they wish to handle it.

We will include a question on this topic in our upcoming attribution survey; more details to follow. We hope you'll join in, as we want as many participants as possible, all of whom will receive a complimentary copy of the results.

Goldman Sachs & Currency Attribution

In yesterday's Wall Street Journal, the editorial writer Holman W. Jenkins, Jr. wrote a piece about Goldman Sachs, explaining how they, and their chairman, Lloyd Blankfein, are being vindicated, as the SEC's attempt to shift attention from their (the SEC's) failure to catch Bernie Madoff, to a highly successful firm and individual, is faltering. This is sadly reminiscent of the government in Ayn Rand's Atlas Shrugged, where this populist strategy, to assign blame and negative attention to the successful firms and individuals, is an easy one to employ, as the public often seems willing to damn these targets, which can be akin to killing the proverbial goose that is laying our golden eggs.

And so, what does this have to do with currency attribution? The proper assignment of blame (and credit).

I was speaking with a client recently who explained that because they don't manage currency, they do not do currency attribution; rather, they use the basic Brinson Fachler (BF) model that only shows allocation and selection (they avoid interaction) effects, based on the base (US$) returns in their global and international portfolios. I suspect that many other firms hold to this same approach. And so, is there anything wrong with it?

Well, consider what would happen if, for example, we have a stock in Euroland, that had a zero percent annual return (i.e., its price was unchanged for the year) as measured in local (Euro) terms,during a year where the Euro's return relative to the US dollar was up 15 percent. By using the security's base return (15%, which arises solely from the currency change), we are unable to identify the true source for the security's return, and attribute it entirely to its selection.

To properly evaluate a portfolio's performance, when multiple factors are at work (the local dynamics of the security, coupled with the exogenous impact of currency movements over time), we must separate their contributions in our attribution analysis, by using the local (i.e., Euro in this case; not base, i.e., US$) returns in the BF evaluation of market effects, and a currency attribution approach to complete the job, in order to fully reconcile to our excess return, by addressing the impact of currency movements (and the contributions from any hedging we may have employed).

The decision to use a multi-currency attribution model has nothing to do with the management of currencies, but simply the exposure to two or more currencies, and thus the associated foreign exchange movements which occur, independent of what happens to the securities we invest in.

Attribution analysis ... when the numbers don't add up!

I was sent an attribution problem by a client to research for them. The details appear here:

I've highlighted the issues they were concerned with. First, in yellow we find that the technology sector's allocation effect is positive (0.222%), even though (a) it's overweighted (75% vs. 67%) and (b) the sector's index return (7.37%) is below the overall index return (7.51%).  How can this be? Note that the Brinson-Fachler model was used, meaning that this situation should result in a negative allocation effect. Why? Because the manager overweighted a sector that underperformed the overall index return (meaning there were other sectors that performed better).

Second, in blue we find that the finance sector's selection effect is positive (0.212%) although the portfolio underperformed the index (7.47% vs. 7.79%). The model expects the effect to be negative in this case. And why? Because the manager underperformed the index, so the selection effect should be negative to reflect poor selection decisions.

And so, what's the problem?

Well, if you duplicate these numbers you'll see that the sector effects were arrived at by adding the effects of the underlying sub-sectors. This isn't the way it's done. We need to apply the attribution effect formulas to the sector levels. And when we do that, we get:

  

By calculating the sector effects using the same formulas as we use at the subsector level, our numbers make sense. The technology sector's allocation effect (-0.011%), as well as the finance sector's selection effect (-0.105%), are now both negative, as we'd expect. 

Conclusion: sometimes the numbers don't add up ... because they're not supposed to!

An animated debate on geometric vs. arithmetic attribution

Here's a new way to communicate ideas, issues, and controversial topics. Hope you like it!

Private equtity attribution

I'll be teaching a class for a client later this month who asked for some specific topics to be included, one of which is private equity attribution. The challenges with private equity are:

1. They typically don't have indexes, so we can't use relative attribution
2. They use IRR, not TWRR, since the manager controls the cash flows.

And so, what is a private equity manager to do if they want attribution? The answer is simple: absolute attribution (aka contribution)! We can utilize IRR at all levels and reconcile to the overall return. We calculate our sector returns, for example, using IRR; calculate their weights, taking cash flows into consideration; and simply multiply the weights by the individual returns. Their sum should tie out to the overall return. For example:

We can see how each sector contributed to the overall return. And just like hedge fund attribution, we can slice up our private equity portfolio in countless ways (e.g., by industry, subsector, country). Does this not make sense? Thoughts?

Attribution question...answered!

A colleague sent me an interesting question, which I will summarize here:

An account has a single cash flow during the month, so there are two sub periods for performance to generate a TWRR. Although we can link the returns to get the full month return, it isn't as easy to derive a single weight that would maintain the integrity of a weighted average calculation. So, the preference is to calculate attribution for each sub period and link. However, we still face a scenario where many indices aren't available daily, so how do we get index return periodicity to match the account returns? This poses a number of questions:

• Can we assume linear performance for the benchmark?
• Do the cons of making assumptions around indexes outweigh the benefits? Should we be driven by the least common denominator? In other words if index data is only available monthly then attribution should only be produced monthly. This then means you need to explain the difference between the published return and attribution analysis.
• Can we manipulate portfolio weights so that we can use monthly analysis and linked returns?

In my view this essentially speaks to the holdings- versus transaction-based issue with attribution. There is a misperception many hold that daily transaction-based attribution is a superior methodology; the reality is that monthly is sufficient for transaction-based attribution; holdings-based, on the other hand, warrants a consideration for daily in order to improve accuracy. And there's the rub: once we move to daily we need to be able to capture the index details for the corresponding periods. And as noted in the question, even breaking a month into two pieces would require this. And, as correctly noted, not all indices provide daily information.

However, if we employ a transaction-based approach, then we don't have to segment the month nor do we have to worry about index details for the sub periods. By using the beginning sector weights, plus weighted flows (just as we do with Modified Dietz) we can accomplish our objective. The returns, too, have to take into consideration the intra-month activity to ensure accuracy.

Why don't attribution effects link?

I was teaching an attribution class recently, and a student asked for a simple explanation as to why arithmetic attribution effects don't link (geometric do, which is one of its advantages over arithmetic). The person recognized that the effects don't link, as do their clients, but he was still wanting to have a pithy response to the question when posed by clients. Simply saying "because they don't" didn't seem to work.

Well let's consider two other items before we answer this question: returns and excess returns. Returns link, right? And why is this? Because returns compound. That is, returns build upon the performance of prior periods. If I start with $10,000 and have a 10% return in January, $1,000 is added to my value and the portfolio ends the month at $11,000. Then if I have another 10% return in February, I don't add another $1,000, but another $1,100, which is 10% on the initial $10,000 plus 10% on the added value ($1,000) in January; i.e., $1,000 + $100. Returns compound and therefore we link from month to month: arithmetic linking (i.e., simply adding returns together) don't take the compounding effect into consideration; thus we use geometric linking.

Excess returns (i.e., portfolio return minus benchmark return) don't link. Why not? Because excess returns themselves don't compound. And while the portfolio and benchmark returns compound, they may compound in different fashions depending on their individual results. But excess returns don't compound.

Attribution effects reconcile to excess returns, right? And if excess returns don't compound how can attribution effects? But we want to be able to reconcile to the linked period excess return, which is based on taking the difference between the linked period returns (what a mouthful!). We accomplish this through a smoothing technique, such as the ones developed by David Cariño, Jose Menchero, and Andrew Frongello (and no, you don't have to have an "o" at the end of your name to develop such a linking method (but it can't hurt!) The French group, GRAP, also developed a method to link attribution effects).

To summarize, attribution reconciles to excess returns. Unlike the returns themselves, arithmetically derived excess returns don't compound. Therefore arithmetic attribution effects don't compound. In the case of geometric attribution, their excess returns do compound, so the attribution effects compound, too.

Hopefully this makes sense, though I'm open to your thoughts.

Karnosky-Singer attributes

Perhaps the best known model to reflect the attribution effects due to currency is the one developed by Denis Karnosky and Brian Singer, both formally of Brinson Partners. The fact that they worked with Gary Brinson is probably part of the reason for its tie in with the Brinson equity models. The K/S model is actually quite flexible and can be adapted to any attribution model, including fixed income models, though little has been written about this. This is "on my plate" to do in the near future.

Two important features of the K/S model should be emphasized. First, on the market side (where we look at the traditional effects one might find from a Brinson (e.g., Brinson-Fachler) model, for example) we use local return premiums rather than merely local returns. That is, we back out the risk-free rate from the returns, as this rate ends up on the currency side.

Second, we show two attribution effects for the currency side: the contribution from the underlying assets (that is, the impact simply due to changes in the FX rates on the assets in the portfolio over the period) and from currency forwards (that arise from our employment of various forward strategies). While one might want to lump these effects together, separating them provides the reader with a much better insight into what has occurred in the portfolio.

If, for example, we don't employ any hedging, then all of the effects would presumably come from the underlying assets. If, however, some hedging is taking place, then we would see how each is contributing to our excess return.

Attribution standards

Someone e-mailed me this week asking about the status of attribution standards. He cited my article, "A Case for Attribution Standards" (Journal of Performance Measurement, Winter 2002/2003) and wondered if any progress has been made. Sadly, no.

When the Performance Measurement Forum developed our draft standards close to a decade ago we hoped they would spur some interest and an endorsement, but this hasn't occurred.

Since the release of the article and the draft, a French group (GRAP) released guidelines for fixed income attribution and the European Investment Performance Council also provided guidelines.

There was much confusion when the subject of standards was first broached, with some interpreting the idea as defining what type of model, for example, one must use. But rather the idea is about disclosure: that is, for firms that employ attribution to disclose how they do it (e.g., arithmetic or geometric; holdings- or transaction-based; which model).

Perhaps this inquiry and some additional discussion might inject some renewed interest in this concept.

Linking fixed income attribution

As you're probably aware, arithmetic attribution results in subperiod effects which are linking challenged; that is, if we attempt to link these subperiod (e.g., monthly) results to obtain longer (e.g., annual) period results, using, for example, simple geometric linking (as we use with returns), the results won't reconcile to the longer period excess return: a residual typically results [geez! what a sentence!]. As a result, several individuals (e.g., Jose Menchero, David Cariño, and Andrew Frongello) developed models which eliminate the residuals. These models are typically demonstrated with one of the Brinson (i.e., Brinson-Hood-Beebower and Brinson-Fachler) models.

But what if you're doing fixed income attribution and using, for example, the Campisi model: how do we link these subperiod results?

Answer: the same as we would with equity attribution; that is, use one of the linking models. Special models aren't required to make this work.

Residuals when you don't expect to see them

Most folks in our industry are aware that if you employ a holdings-based model for attribution, you're subject to residuals creeping in. Residuals are the unexplained differences between the excess return (portfolio return minus benchmark return) and the sum of the attribution effects. Because holdings-based models don't capture intraperiod cash flows, they're subject to residuals. And so, our research confirms that most asset managers prefer to use a transaction-based model to be more accurate.

However, if you're using a monthly return formula (e.g., Modified Dietz), then you may still end up with residuals, because the transaction-based approach is capturing an exact return, while the Modified Dietz arrives at an approximation to the true, TWRR. Ideally, you should be using a daily return method so that you will eliminate the residual.

Note: you can still have what might be called a longitudinal or intertemporal (occurring across time) residual, unless you employ an appropriate linking method.

p.s.,  I believe I'm the first to use the term "intertemporal" to describe this form of a residual. I believe it's a handy way to distinguish between single-period residuals which can arise from using a holdings-based model, and residuals across time, when using an arithmetic approach.

What happens when your classification is different than the index's?

I'm at the Spring meeting of the North American chapter of the Performance Measurement Forum in Boston. One of the issues that was raised deals with the case when a firm has a different classification structure than the index and the resulting impact this has on performance attribution. For example, if a firm assigns a stock that's in the index to a different industry than what the index has.

If we ignore this difference then the attribution results are questionable. If we change how the benchmark has the security assigned then we've manipulated the benchmark and arguably falsified the comparison. If we change our classification so that it aligns with the benchmark than we might feel that this doesn't truly represent how we manage. No good solutions.

Probably the best answer is not to have a different classification. But this may not always be practical. Perhaps in those cases where the firm feels that such differences are necessary we should have a new attribution effect: "classification effect." Just as we can have "pricing effect" to identify pricing differences between the benchmark and portfolio, we can have a "classification effect" to identify the impact of differences in how we classify our securities. To my knowledge no one has done this but it can't be that difficult to accomplish.

Thoughts?

What does "daily attribution" mean?

We're working with a client who needs to be able to respond to RFPs that ask "do you have daily attribution?" But what exactly does this mean? Does it mean "do you calculate attribution daily" or "can you report attribution from any day to any day"?

As I mentioned in March's newsletter, when it comes to transaction-based attribution there is no reason to calculate the effects daily: monthly works fine. But what if someone wants to report attribution ending (and/or beginning) for a day other than month-end? Not a problem: you just value the portfolio as of that date and calculate the effects between that point and the prior (or ending, if it's the starting point) month end, taking  into consideration the trading that occurs in the interim.

Why should a prospect care whether or not you actually calculate the effects daily? They should be concerned with the ability to report from any date to any date accurately, and this can be achieved without calculating daily effects.

And, as I mentioned previously, this doesn't apply to holdings-based, where more frequent processing is required to ensure a better degree of accuracy.