When less is more (and when it’s not) why attribution is always going to be complex

March 14, 2019 in Attribution



When less is more (and when it’s not) why attribution is always going to be complex

Andrew Colin

Flametree Technologies

In this blog post, Andrew talks about the choices facing users of attribution software, cases where there’s no choice at all, and why the choices facing users should be embraced. Complexity is a fact of life - not just setting up the approach, but in handling individual securities.

What is the right attribution model? The first-time user of any attribution package might well be forgiven for feeling overwhelmed. The number of options available rival in complexity the controls in an airliner’s cockpit.

Is this complexity justified? In ’The Design of Everyday Things’ by Donald Norman, much play is given to the concept of ’affordance’ - making a system as simple and intuitive to use as possible.

What makes attribution inherently complex? I’d argue there are three reasons:

The number of investment strategies. The first is the sheer number of investment strategies available to a portfolio manager. Let’s consider a simple example.

Suppose your portfolio is hedged against movements in FX rates. This is a very standard tactic and ensures that your investments are protected against currency fluctuations. However, hedging comes at a price, which is typically the interest rate differential between the portfolio’s base currency and the currency to be hedged.

Should the return generated by these forward premiums be classed as interest holdings, or as carry return, or as curve return, or placed into a separate category entirely? There is no one correct answer; it depends on the investment process. Sometimes, hedging is run as part of a stand-alone portfolio; sometimes, it’s done over all non-domestic holdings, at a much later point in the investment process.

For this reason, it’s important that returns from hedges be allocated to the right part of the investment process.

Let’s look at another issue. Should we include returns due to convexity in an attribution report? Convexity just measures non-linearity of return against yield change. Everything else being equal, a bond with high convexity will outperform a bond with identical modified duration when yields change.

If the strategy is simply to be long (or short) against a benchmark, then probably not. Convexity has played no part in the investment process, and to report on this source of return in isolation is at best misleading.

On the other hand, if the investor has followed a strategy that relies on convexity (such as taking a barbell allocation), then it is proper and correct to report on convexity returns, so that the investor can verify that the return made by convexity outweighs any losses made elsewhere. It all depends on the investment strategy.

The number of security types. The second is the number of security types traded in the marketplace. Most FI portfolios are driven off changes in the yield curve, but even defining what a parallel change in the curve means is still an open question. Inflation-linked securities will be driven by the level of indexation to a reference index; MBS by the prepayment rate of the underlying security pool; and so on. There is no reason to expect that the pace and number of new security types will wane, so this requirement is unlikely to disappear.

In addition, one should be able to tailor the returns of a security to match its requirements.

Take the example of a future on a US treasury bond. This is one of the most liquid and heavily traded securities on the planet. Yet it’s not always clear what its returns should be without looking at its market price.

European bond futures have an embedded option on which bond to deliver (aka ‘quality’). To make matters more complex still, US bond futures have several embedded options, including quality, timing (when the bond is delivered), and end-of-month (switching deliverable bonds after the future’s last trading date).

In addition, there is the question of liquidity. A bond future that is two or three contracts away from being the ’nearest’ may be very lightly traded, in which case its quoted price may move away from the theoretical level.

It is difficult to unpack the effects of these options, and their presence can mean that the price of the future changes even when the price of the underlying remains unchanged. This translates into residuals.

So how important is this for attribution? Unless the portfolio manager is arbitraging futures positions, the answer is probably ’not very’. It’s probably not an overstatement to say that 95% of bond futures positions are taken to add, or subtract, interest rate risk from a portfolio as a whole. The great benefit of futures is that they let you do this in a cheap, liquid manner. In this case, residuals are not really of interest. Yet because of the large holdings of futures in some portfolios, they can appear enormous.

In this case, I’d argue that an attribution system should be capable of reallocating any residual return from futures to interest rate return. This requires the ability to map one or more sources of return for a given security type, or even on an individual security basis, to a different source of risk. How this is done is not important. What matters is that it should be possible.

The target audience. Paradoxically, the last reason to argue for complex, powerful attribution systems is the number of consumers of attribution reports. Given the insight that attribution provides into both a particular portfolio and an institution’s strategy, we should expect wide interest in the results. For results tell you about the fund; attribution tells you about the manager.

But not all readers are equal, nor should we expect them to be. While a retail consumer may be interested in seeing claims of consistent stock selection returns, the details of the same portfolio’s cash flow strategies may be too much detail. Ask yourself how much detail of your investment process should appear on an ad on the Tube, and you won’t go far wrong!

This means that not only should we provide a wide range of attribution analyses (as many as possible), but also as many ways of looking at the results as possible. In fact, I’d argue that any decent attribution system should offer (or at least support) a wide range of reporting options, including numerical reports, graphics, and interactivity.

Albert Einstein famously said ’Make it as simple as possible, but no simpler.’ Attribution systems are going to remain complex. And so they should, for they’ll be worthless otherwise. As with all other areas, complexity can be tamed. The best systems will combine simple, elegant design with deep and powerful functionality.

Leave a Reply

Your email address will not be published. Required fields are marked *