Day 2: 15:30-16:30. Fixed Income Return Attribution Analysis. Presented by Frank J. Fabozzi, Professor of Finance, EDHEC Business School

Transcription

1

2 Day 2: 15:30-16:30 Fixed Income Return Attribution Analysis Presented by Frank J. Fabozzi, Professor of Finance, EDHEC Business School 2

3 A performance measure does not answer two questions: 1. How did the portfolio manager perform after adjusting for risk associated with the portfolio strategy employed? 2. How did the portfolio manager achieve the reported return? 3

4 Answers critical for both the client and the investment management team For client: Used in assessing how well or how poorly manager performed relative to some benchmark and why For management team: Used in assessing how team members contributed to performance. 4

5 Look at methodologies for adjusting returns for risk so as to 1. Determine whether a superior return was realized 2. Analyze actual return of a portfolio to uncover reasons why a return was realized. Begin by looking at various benchmarks that can be used to evaluate the performance of an asset manager. 5

6 Benchmark Portfolios To evaluate performance of a portfolio manager, client must specify a benchmark against which the portfolio manager will be measured. Issue: Are standard fixed income benchmarks appropriate? Theme of many papers written by ERI researchers is that standard benchmarks used in fixed income are inappropriate Benchmarks used for clients who are pursing liabilitydriven strategies (e.g., defined benefit plans) are inappropriate. (Ben Cotton to elaborate) 6

7 In the beginning performance was based solely on return relative to a benchmark 7

8 It was not until the 1960s that there was recognition that risk must be considered along with return 8

9 This lead to the introduction of several measures, the most notable being the Sharpe measure or Sharpe ratio. 9

10 Sharpe ratio Measure of the reward-risk ratio. Introduced by William Sharpe in 1966 (Portfolio return Risk-free rate) Standard deviation of the portfolio s return 10

11 Other Single-Index Performance Evaluation Measures include 1. Treynor measure 2. Information ratio 3. Jensen measure 11

12 The problem with these measures which were commonly used to evaluate mutual fund performance they tell us only if a fund manager outperformed but not why. The way may be even more important 12

13 Today two approaches are used in evaluating performance Return attribution analysis Peer analysis 13

14 Return attribution analysis Basic idea Given an index (benchmark) determine the risks associated with the benchmark That is, view the index as a package of risks Quantify those risks The portfolio return of a client can be attributed to those risks. 14

15 One should not think of return attribution analysis as a separate exercise from portfolio construction. In portfolio construction, one determines the risk factors that drive return and builds a risk-controlled optimal portfolio accordingly In performance evaluation, one uses the same risk factors to assess how the risks taken when constructing the portfolio paid off. 15

16 Basically, return attribution analysis is used to determine why a manager outperformed or underperformed the benchmark. No longer acceptable to be satisfied with outperformance because it could have been accomplished by taking on substantial risks unacceptable to boards/clients. 16

17 The more the portfolio s risks match in quantity that of the benchmark, the closer the portfolio s performance will be to that of the benchmark (i.e., closet indexing). The greater the departure of the portfolio s risks relative to the benchmark, the greater the degree of active management. 17

18 What does return attribution seek to explain? Difference between the return on the portfolio and the return on the benchmark. This difference is referred to as the active return Return attribution analysis therefore tries to determine how the active return was generated. The active return is attributed to the different risks that have been quantified for the benchmark 18

19 In broad terms, return performance of a portfolio can be explained by three actions followed by a portfolio manager: 1. Actively managing a portfolio to capitalize on factors that are expected to perform better than other factors. 2. Actively managing a portfolio to take advantage of anticipated movements in the market. 3. Actively managing the portfolio by buying securities that are believed to be undervalued, and selling (or shorting, usually in fixed income portfolio via credit default swap) securities that are believed to be overvalued. Ben Cotton has further insightful points he will discuss. 19

20 Since there is an individual that is part of the portfolio team that is responsible for each risk taken (i.e., bet relative to the benchmark), return attribution analysis allows the assessment of individual team member performance. 20

21 The challenges in implementing a return attribution model are 1. Selecting a suitable benchmark (either a currently available index or a customized index) 2. Identifying and quantifying the risks embedded in the index selected (done using statistical /econometric techniques) Ben Cotton has some further insights 21

22 Types of Performance Attribution Models In the early 1970s, Eugene Fama [1972] suggested a framework for the partitioning of risk-adjusted portfolio performance for equity managers. Extensions of the Fama framework were then proposed for mutual funds. In 1980, Dietz, Fogler, and Hardy [1980] appear to have been the first to apply performance attribution to bonds. A few years later, Fong, Pearson, and Vasicek [1983] provided a more elaborate bond performance attribution model. 22

23 Today, performance attribution models can be classified into three types: 1. Sector-based attribution models 2. Factor-based attribution models 3. Hybrid sector-based/factor-based attribution models 23

24 Sector-Based Attribution Models The simplest model (also referred to as the Brinson model developed in Brinson and Fachler [1985]) In this model, the portfolio return relative to the benchmark is represented by two decisions: 1. allocation of funds among the different sectors (asset allocation decision) 2. selection of the specific securities within each sector (security selection decision) There is also an interaction decision that results when two factors contribute to the performance of the return but vendors will fold that term into one of the other two. 24

25 Example: return relative to the same benchmark (i.e., the active return) shown in basis points Decision Portfolio A Portfolio B Portfolio C Asset allocation Security selection All three managers outperformed the benchmark by 200 basis points. However, they did it in different ways. Manager of Portfolio A made the right calls both in the allocation of funds among the different sectors and the selection of securities within each sector. Portfolio B s outperformance is attributable to the portfolio manager making a good asset allocation decision while at the same time doing a poor job of security selection (i.e., manager picked the right sectors of the benchmark to allocate funds but selected the poor performing securities within those sectors) Portfolio C s manager outperformed due to the selection of the better performing securities within a sector despite not selecting the better performing sectors. 25

26 A more detailed analysis would break down the active return from each sector and the securities within each sector. Performance attributable to security selection would probably differ by sector Example: consider Portfolio C and assume 245 basis points due to security selection was such that there was a zero active return in terms of security selection for all sectors in the benchmark except for the CMBS and agency MBS sectors. Assume further that the respective active returns were 100 and 145 basis points respectively. This would indicate that the analysts responsible for the selection of the securities in these two sectors did a superb job in making security selection decisions. 26

27 Factor-Based Attribution Models Uses framework for portfolio construction based factor-based models. Factor models (i.e. risk-factor models) determine the key common drivers of returns and their risks for the benchmark and can then be used to construct a portfolio relative to a benchmark so as to control the tracking error. These common factors are referred to as the systematic risk factors. 27

28 REVIEW OF RISK FACTORS Tracking Error A factor model attempts to identify the specific risks that contribute to the forward-looking tracking error. All of the risks are quantified in terms of forward-looking tracking. 28

29 Can decompose the tracking error in order to understand where the risks are. Analysis begins with a decomposition of the risks into two general categories: 1. Systematic risk and 2. Nonsystematic risk (also referred to as residual risk). 29

30 DECOMPOSITION OF SYSTEMATIC RISK Can be decomposed into two risks: 1. term structure factor risk and 2. non-term structure factor risk. A portfolio s exposure to changes in the general level of interest rates is measured in terms of exposure to 1. a parallel shift in the yield curve and 2. a nonparallel shift in the yield curve. Taken together, this risk exposure is referred to as term structure risk. 30

31 Non-Term Structure Risk The other systematic risk that is not due to exposure to changes in the term structure These risk factors include sector risk, quality risk, optionality risk, and MBS risks (sector, prepayment, and convexity risks) 31

32 Nonsystematic Risk Divided into 1. Issuer specific and 2. components that are issue specific. This risk is due to the fact that the portfolio has greater exposure to specific issues and issuers than the benchmark index. 32

33 Consider the following three hypothetical portfolios all with a 110 basis point active return: Risk Factor Portfolio D Portfolio E Portfolio F Yield curve risk Swap spread risk Volatility risk Government-related spread risk Corporate spread risk Securitized spread risk Portfolio D s manager made one right bet: an interest rate bet. All other bets had an adverse impact on performance. It seems Portfolio E s manager took no view on interest rates. All the non-interest rate bets paid off, with the major one being due to bets on the risks associated with corporate credit and securitized products. Portfolio F s manager also made the right bets on non-interest rate risk and, as did Portfolio E s manager, the majority of the outperformance was due to the risks associated with corporate credit and securitized products. However, Portfolio F s manager made the wrong call on the interest rate bet, causing the portfolio s performance to just match that of Portfolio E s manager. 33

34 Factor-based attribution models actually allow a decomposition of the yield curve risk into level risk and changes in the shape of the yield curve. For example, for the three hypothetical portfolios, D, E, and F suppose that the attribution due to yield curve risk is determined to be as on the next slide: 34

35 Portfolio D Portfolio E Portfolio F Yield curve risk Level risk Shape risk Once yield curve risk is decomposed as shown above, it turns out that Portfolio E s did indeed make interest rate bets. It turns out that the two bets almost offset each other so that net there was only a one-basis-point return attributable to the interest rate bet. Portfolio D s manager basically made a major duration bet but virtually no bet on changes in the shape of the yield curve. The interest rate bet by Portfolio F manager was on changes in the shape of the yield curve but otherwise was basically duration neutral. 35

36 Hybrid Sector-Based/Factor-Based Attribution Models As the name suggests, combines the previous two attribution models. This model allows for much more detail regarding not only the bets on the primary systematic risk factors driving returns but also the impact of decisions with respect to sector and security selection. The level of detail in such models depends on what might be sought by the client or the portfolio manager. 36

37 Some concluding comments: 1. Search for models that can incorporate the use of derivatives 2. Need to integrate performance attribution models into the entire portfolio management process 3. Tie in to best execution performance Ben Cotton will identify others in his wrap up of this session 37

38 Peer Analysis: Used for Mutual Funds Public relations versus performance Return to the olden days where only consideration is given to return and not to risk relative to a benchmark Moves us from to 38

39 Peer Analysis Issues Selecting the appropriate peer group Since risk is not considered, will always have inferior performance than the most aggressive fund manager whose bet(s) pay off 39

40 REFERENCES CITED IN PRESENTATION Eugene F. Fama, Components of Investment Performance, Journal of Finance 27, 3 (1972), pp Peter O. Dietz, H. Russell Fogler, and Donald J. Hardy, The Challenge of Analyzing Bond Portfolio Returns, Journal of Portfolio Management 6, 3 (1980), pp H. Gifford Fong, Charles Pearson, and Oldrich A. Vasicek, Bond Performance: Analyzing Sources of Return, Journal of Portfolio Management 9, 2 (1983), pp Gary Brinson and Nimrod Fachler, Measuring Non-U.S. Equity Portfolio Performance, Journal of Portfolio Management 11, 3 (1985), pp