Issue 42, Fall 2006
Fixed Income Performance Attribution
An Interview with James Hollis and Jim Hendrickson
With one major exception, the investment industry is in agreement on the appropriate methods and practices for all major performance-related areas. Thanks to GIPS, we agree on how to calculate and report on performance measurement and composites; thanks to Gary Brinson and many others, we generally agree on how to calculate equity performance attribution. The major exception to this industry-wide agreement is in the calculation of fixed income attribution - ask six people the best way to calculate fixed income attribution, and you are likely to get eight answers.
In this issue of the Cutter AdvantEdge, Jamie Hollis interviews Jim Hendrickson on the current state of fixed income performance attribution. Jamie is a managing director at CutterAssociates. He is also the former Chair of the Investment Performance Council, the organization within the CFA Institute (previously AIMR) chartered with administering GIPS, and he sits on the Board of the Institute. He has been involved with performance standards since the early days of AIMR-PPS.
Jim Hendrickson is a senior consultant at CutterAssociates and an expert on fixed income performance attribution. Jim built a leading-edge, proprietary fixed income performance attribution platform for a major fixed income investment firm, one of the first such systems created. Jim now consults to investment firms that are planning to buy or build such systems, helping them to verify fixed income performance attribution methodologies. He also assists clients in build vs. buy decisions and in selecting appropriate vendor products.
Jamie: Jim, what is the current status of vendor products for fixed income attribution?
Jim: There are a number of systems available that incorporate good methodologies, methodologies that are both logical and intuitive. One of the biggest changes in vendor products in the last several years is in providing choices. In the past, a vendor's system offered only that vendor's specific methodology. Now, many of these systems give the user a choice of methodologies. Some even are willing to incorporate a user's proprietary methodology within their systems. In short, many vendors provide the place to implement a solution, rather than defining that solution.
Jamie: What else has recently changed in this field?
Jim: The most helpful development has been greater access to constituent level index data and yield curve data on a daily basis. I should note, however, that few vendors of fixed income attribution systems actually provide this data; the user is responsible for finding a source of the data and populating the system's database each day. To me, fixed income attribution is a data and processing problem, not a conceptual or math problem. All that's being done is analyzing a portfolio's relative bets after the fact with all the returns given by the market, and the availability of the required data is critical.
Jamie: What are the key issues in producing useful fixed income attribution numbers?
Jim: The most important issue to me is getting the underlying risk-free return (typically that of Treasuries) out of each bond's returns. This is consistent with the market practice of quoting yields on these bonds as a spread over Treasuries with a like duration. It also separately identifies the risk-free return and the spread return. This separation is critical to obtaining logical results. The practical problem can be illustrated this way - in a 100 basis point rally, I cannot tell whether a 13-year duration bond outperformed a five-year duration bond until I remove the risk-free return in 13-year and five-year durations.
Jamie: Is there any agreement in the industry on a structure for fixed income attribution, as there is for equity attribution?
Jim: Within reason, yes. Preferences on yield curve analysis can vary somewhat, just as theories of yield curve movements and reshapings vary. That said, my experience is that most people are comfortable with the concepts of parallel and non-parallel shifts and see attribution of yield curve effects in that context. It is also somewhat common to add a butterfly level to the yield curve analysis. Obviously, the analysis and presentation of attribution for yield curve effects should be consistent with the risk presentation at the front end of the investment process; otherwise, it is impossible to tie excess performance back to the manager's decisions.
Attribution of the excess return effect (the incremental return over that of a comparable risk-free instruments) appears to be fairly well standardized at this point. Once the underlying risk-free return is factored out, the analysis of spread return is identical in all ways to equity attribution, using the same allocation/selection models. In my experience, industry practitioners are fully comfortable with this notion.
Jamie: Does fixed income attribution really differ that much from equity attribution?
Jim: Fixed income attribution differs from equity attribution solely in the issue of the underlying risk-free returns. Once the risk-free return is isolated and treated as a separate component, the balance of the attribution analysis - that of the spread return - is identical to equity attribution in both concept and application.
Jamie: Is it possible to have an integrated attribution solution?
Jim: There are a small number of vendors that currently provide both a standard equity attribution capability and robust fixed income attribution capabilities. I would expect this number to increase rapidly over the next few years.
Jamie: How involved are fixed income attribution models with derivative record keeping and performance measurement?
Jim: Performance attribution systems are not involved in record keeping or performance measurement. Record keeping is done by a portfolio accounting system, and performance measurement is done by a performance measurement system. Attribution systems expect to be fed security weights and returns from these other systems. Derivatives flow through attribution as just another security.
Jamie: Attribution systems, and especially fixed income attribution systems, have a voracious appetite for data. What problems are there in obtaining and maintaining this flood of data?
Jim: The major problem with data sourced within the firm is the difficulty of handling historical corrections in prices, returns, positions and transactions. A robust fixed income attribution system will provide tools to assist the user in making the required adjustments. The major problem with data sourced outside the firm, primarily relating to indices, is simply obtaining the data. All of the major suppliers of index data - firms such as Lehman, Citigroup, JPMorgan and IIC (iBoxx and iTraxx) - make constituent data available, but sometimes only to large customers or at a price that is prohibitive for smaller managers. I have sometimes run into a problem with global securities in getting daily returns in the accounting currency, but this can be overcome with spot and forward currency data.
Jamie: Fixed income indices typically have a very long list of constituents. Storing daily information on every constituent in every index, plus every holding in every portfolio, can be a challenge. How has the industry dealt with this?
Jim: There is no way around the fact that this is a great deal of data. Fixed income attribution is just very data intensive and requires a lot of disk space, fast processors and efficient code. On the positive side, once several years have past, the firm has a database with daily security-level returns on indices and portfolios, along with yield curve and currency data. This database will prove invaluable to the firm's quantitative researchers.
Jamie: What should a manager look for when reviewing vendor's fixed income attribution products?
Jim: First and foremost, the product should support a fixed income methodology consistent with the firm's investment process. If you present the portfolio's strategy in terms of parallel and non-parallel shifts, then you should not choose a model that provides attribution in the form of parallel, non-parallel and butterfly shifts. In the same vein, if you think in terms of the Brinson Hood Beebower model, then a system that only provides the Brinson Fachler model would be inappropriate. In other words, the vendor's product should provide information for you to be able to present your strategy to a client and clearly show how the key areas of that strategy affected the return on the client's portfolio relative to the index.
Simply put, that's the job - to tie the bets in the portfolio to the result. The math is simple, as is the concept. The details of implementing the concept, however, can be miserable, and I think it is this that makes fixed income attribution appear so complicated to many people.