With the current move from equity to fixed interest sectors, the ability to perform comprehensive risk management on portfolios of fixed interest instruments is now more important than ever before. A major component of this is the ability to perform fixed interest attribution, or to isolate the effects of various sources of risk on the overall performance of a portfolio.
Here, we look at what we think is possible in FI attribution.

Attribution standards: It’s unlikely there will ever be any attribution standards, useful though this would be for comparing investment returns. The reason is that every portfolio manager has their own investment process, and this affects the order and the nature of the calculations. Unlike performance calculation standards – where it makes sense to insist on industry-wide measurement standards - don’t expect to see attribution reporting standards anytime soon.
An obvious question is why bother? If we have accurate, GIPS-compliant returns for a set of managed portfolios, why go to the considerable trouble of breaking down these returns? The reasons are several-fold:
For the manager, attribution gives vital insight into the strengths and weaknesses of the investment process, and (in particular) their risk management systems. A manager that claims to make returns on the basis of duration bets, but that continually loses because of other types of yield curve movements, clearly has to improve internal risk control and ensure that they are hedged against unwanted risk – which they may not be qualified to assume.
For the asset consultant and the investor, attribution acts as a quick, verifiable guide that the investment process, the technology and the expertise of the manager are all working together. This allows the manager to demonstrate that quoted returns are not just due to a few lucky bets over the previous years, but are due to a rigorous, reproducible set of investment criteria. How often have you seen investment advertising the stock-picking expertise of a fund manager? How often have you seen attribution returns that back up these claims? In summary:
o Returns tell us about the fund
o Attribution tells us about the manager.

Sources of risk: What are the main sources of risk in a fixed interest portfolio?
Fixed interest investment risks are qualitatively different from equity investment risks, where the decisions made refer to asset allocation and stock selection. While a fixed interest portfolio can be analysed in this way, this decomposition does not follow the way that fixed interest portfolios are actually managed, and therefore gives limited information about the success of the investment strategy. A more realistic list is as follows:
o Coupon
o Yield curve
o Credit shifts
o Trading effects
o Liquidity
o Cheap/dear effects
o Tax effects
Let’s examine each of these in turn.

Coupon: This is just the return due to pre-determined, regular payments from an instrument.

Yield curve: Shifts due to a reference yield curve, usually calculated from a set of reference AAA bonds.
To mirror investment process, we prefer to break these shifts down into parallel component, in which the entire curve moves up or down as a whole; twist component, in which we measure by how much the curve flattens or steepens; curve (or butterfly) component, in which we measure whether the curve becomes more or less convex.
This decomposition reflects the way that many fund managers manage their portfolios. To measure these components is necessarily harder than performing sector attribution over duration buckets, and requires the development of some specialized algorithms.

Trading risk: Trading risk is the value that a dealer adds, or removes, from a portfolio by dealing at non-end of day rates.

Credit risk: A bond that has lower credit rating than those that lie on the reference yield curve will trade at a premium to the curve. This difference is called the credit spread. For instance, the BBB yield curve is the AAA yield curve plus the BBB yield curve spread. A bond priced off the CCC yield curve therefore has its return determined by movements in the AAA yield curve, the BBB yield spread and the CCC yield spread, while a bond priced off the BBB spread will be unaffected by changes in the CCC yield spread.
An important part of credit attribution is the determination of these different credit curves, and measurement of how changes in their shape have affected the performance of an instrument. This requires robust algorithms to model the shape of the various credit curves. Currently, the Nelson-Siegel curve algorithm is widely used for curve fitting to yield and credit curves.
Credit risk is only modeled at the broadest level by most commercially available attribution systems. This does not reflect the growing needs of many fund managers, who are moving away from interest rate risk to take greater exposures in credit risk.

Liquidity risk: Consider a corporate bond that is rated AAA but that has a market yield 50 bp above a government-issued instrument with a similar maturity. Since everything else is equal, we regard this spread as being due to the market’s perception of the bond’s lower liquidity, for which a higher yield is demanded.

Yield curve noise, cheap/dear effects, tax and other effects: Although the above list is fairly comprehensive, it cannot account for all of the causes of a fixed instrument’s price changing. Other effects include:
o Shape of the yield curve being more complex than modeled
o Cheap/dear effects – such as a particular bond issue being bid in the repo market or for tax reasons

Market noise: In practice, these will be included in the liquidity risk category, as it is very difficult to discriminate between these minor effects. However, this should have little effect on the overall analysis.

Multiple yield curves: An infrequently realised need for attribution systems is that of having several yield curves within the system.
For instance, consider a portfolio that contains instruments that are priced off both short and medium term areas of the term structure. While descriptions of how the bond curve changes will apply to instruments with duration greater than – say – a year, the parallel shift will not apply to instruments with very short duration of a year or less, where the shape of the curve is closely tied to the cash rate. We therefore see it as important that an attribution system be able to model multiple yield curves.
Reporting technology and
frequency: There is no reason why daily reporting is not possible. This has the twin benefits of higher accuracy and reduction of residuals.
On reporting, we expect to see a move from paper-based reporting to more interactive, slice-and-dice type displays of results – particularly for use by the front office. This On-line Analytical Processing (OLAP) technology is new and we expect to see more take-up over the next few years as its advantages become apparent.

The hurdles: Fixed interest attribution is vastly different from equity attribution. For equities, a simple statement of the rate of return for each security or sector is enough to get started. For fixed interest, an attribution system must be able to reprice instruments several times to measure the effects of yield and credit curve shifts. It is therefore crucial to have internal pricing algorithms for each instrument represented within a portfolio, whether that be a simple bank bill or a complex swap with embedded options. In our view, this represents a major hurdle for most fixed interest attribution vendors.
This requirement also impacts on the use of benchmarks. For detailed attribution, it is vital to be able to represent a benchmark at the stock level and to be able to perform the same performance decomposition that is carried out on the portfolio.
Depending on the benchmark currently in use, this can be a major undertaking. The data maintenance requirements to model a benchmark – which may have many hundreds of stocks – can be large.
It’s not all bad, however. Unlike equity attribution – which requires sector or stock level asset allocations and returns for both portfolio and benchmark – fixed interest attribution can be carried out on a portfolio in isolation, so that the overall return of the portfolio can be decomposed into yield curve, sector and other returns. This at least allows the manager to back up claims of being a consistent manager in their chosen risk areas.

Multi-currency: Everything we’ve talked about applies to single-currency portfolios. For multi-currency portfolios, there are the extra risks of exchange rate variations and interest rate differentials.
Matters become even more complex when derivatives are used. A sophisticated investor with substantial exposure to options may wish to examine how changes in the volatility curve affected overall returns.
The technology and techniques are already available to perform fixed interest attribution to levels of detail that will substantially assist fund managers in improving their returns. It’s now up to vendors to supply these capabilities to the marketplace.
Andrew Colin is director of fixed income research at StatPro in London