FIXED INCOME ATTRIBUTION:

ANALYZING SOURCES OF RETURN

by

Alexandre Melnikov
Bachelor of Business Administration, Simon Fraser University, 2009

and

Sanja Simic
Bachelor of Science, La Sierra University, 2006

PROJECT SUBMITTED IN PARTIAL FULFILLMENT OF

THE REQUIREMENTS FOR THE DEGREE OF

MASTER OF FINANCIAL RISK MANAGEMENT

In the
Faculty
of
Business Administration

© Alexandre Melnikov and Sanja Simic 2010

SIMON FRASER UNIVERSITY
Summer 2010

All rights reserved. However, in accordance with the Copyright Act of Canada, this
work may be reproduced, without authorization, under the conditions for Fair Dealing.
Therefore, limited reproduction of this work for the purposes of private study, research,
criticism, review and news reporting is likely to be in accordance with the law,
particularly if cited appropriately.
APPROVAL

Name: Alexandre Melnikov and Sanja Simic

Degree: Maser of Financial Risk Management

Title of Thesis: Fixed Income Attribution: Analyzing Sources

of Return

Supervisory Committee:

_____________________________________________

Dr. Anton Theunissen

Senior Supervisor

_____________________________________________

Dr. Andrey Pavlov

Supervisor

Date Defended/Approved: ______________________________

ii
ABSTRACT

This paper investigates several methods of analyzing performance of bond

portfolios and presents an empirical framework for conducting fixed income

attribution calibrated to a particular portfolio. First, we discuss characteristic of fixed

income portfolio management and explain some of the challenges for attribution

reporting. Our primary focus is on depicting deficiencies in methodologies when

measuring shift, twist, butterfly movements, and credit spread changes in a non-

smooth yield curve environment. In our empirical example, we present a systematic

approach to fixed income performance measurement. We also show that attribution

results are consistent with manager’s strategy and changes in the interest rate

environment.

Keywords: fixed income; performance measurement; return attribution

iii
ACKNOWLEDGEMENTS

We would like to thank Dr. Anton Theunissen and Dr. Andrey Pavlov for

providing valuable insights and challenging us to consider ideas from all vantage

points.

iv
TABLE OF CONTENTS

APPROVAL ....................................................................................................................................ii 
Abstract...........................................................................................................................................iii 
Acknowledgements .........................................................................................................................iv 
Table of Contents ............................................................................................................................v 
List of Figures .................................................................................................................................vi 
List of Tables .................................................................................................................................vii 

1: Introduction ................................................................................................................................. 1

2: Fixed Income Performance Measurement .................................................................................3

2.1  Overview of Fixed Income Management Process .................................................................3 
2.2  Fixed Income Performance Attribution Models....................................................................4 

3: Data and Methodology................................................................................................................8

3.1  Data......................................................................................................................................... 8 
3.2  Methodology ......................................................................................................................... 11 
4: Results........................................................................................................................................ 14

5: Discussion And Conclusion ....................................................................................................... 17

5.1  Discussion.............................................................................................................................. 17 
5.2  Conclusion ............................................................................................................................. 18 
Appendices ..................................................................................................................................... 20
Appendix 1: Total Return Calculation ......................................................................................... 21 
Appendix 2: Treasury Return decomposed into Shift and Twist ............................................... 22 
Appendix 3: Comprehensive list of formulas used in attribution calculations............................ 23 
Appendix 4: Excel model inputs ................................................................................................... 24 
Appendix 5: Duration Matched Treasury calculations — Part1 .................................................. 25 
Appendix 6: Duration Matched Treasury calculations — Part2 .................................................. 26 
Appendix 7: Excel model attribution calculations — Part 1 ........................................................ 27 
Appendix 8: Excel model attribution calculations — Part 2 ........................................................ 28 
Appendix 9: Excel model graphical output .................................................................................. 29 
Glossary .......................................................................................................................................... 30 

Reference List ................................................................................................................................ 31

v
LIST OF FIGURES

Figure 1: Benchmark Sector Allocation ................................................................................ 9

Figure 2: Portfolio Sector Allocation .................................................................................... 9
Figure 3: Treasury Yield Curve Movement......................................................................... 11
Figure 4: Treasury Yield Curve Change ............................................................................. 11
Figure 5: Total Return Decomposition ............................................................................... 12
Figure 6: Price Change Effect Decomposition ..................................................................... 12
Figure 7: Contribution to Return for Bond Portfolio and Benchmark .................................. 14

vi
LIST OF TABLES

Table 1: Benchmark Data ................................................................................................... 9

Table 2: Portfolio Data........................................................................................................ 9
Table 3: Five-Year Key Rate Durations ............................................................................. 10
Table 4: Treasury Yield Curve Change .............................................................................. 11

vii
1: INTRODUCTION

Performance attribution helps investors and managers understand sources of

portfolio absolute and relative returns. It answers questions such as if the portfolio

beat the market, whether the risks a manager took paid off, or did manager add value

and was it due to skill or chance. Performance attribution is an important tool for

both investors and fund managers. Investors use attribution to evaluate fund

managers, their concern is largely the return on initial investment and income

received. Investment managers use performance attribution to assess strategies and

determine if those need to be reinforced or rethought.

Unlike for the equity attribution based on Brinson-Fachler (1986) model there is

no one-size-fits-all, standardized approach to fixed income attribution. The literature

on this subject has been diverse. General understanding is that bonds are unlike

stocks, consequently sources of risk and decision-making processes differ, therefore the

traditional equity-style attribution introduced by Brinson-Fachler (1986) is not

suitable for fixed income portfolios. The appropriate method should be representative,

reflecting the decision making process within the fixed income portfolio.

Major determinants of fixed income performance are income and changes in the

treasury (default-free) yield curve such as shift, twist, and credit spreads.

Furthermore, the spreads between the benchmark and portfolio returns are minute

relative to equities and require additional precision when calculating. Models that use

a reference curve, may interpolate using different methods, introducing a dose of bias.

Additionally, as risks and models become more complex, the process becomes more

tedious. Lastly, complexities introduced by different models may have intricate data

1
feed requirements which can introduce significant expenditures, but offer little

improvement in accuracy.

Our approach takes a real world bond portfolio, and presents a step-by-step

example of performance attribution based primarily on Campisi’s framework (2000).

We show that this particular method is consistent with management’s decision-

making process and therefore appropriate for reporting attribution.

Following is the organization of this paper. Section 2 discusses some aspects of

fixed income asset management. We compare and contrast that to equity asset

management and show why it is that traditional equity frameworks are not

appropriate for fixed income attribution. We also list some of the challenges in

designing and implementing a fixed income attribution tool. Finally, we describe some

of the characteristics of a good fixed income attribution framework. Section 2 also

provides a literature review of the existing frameworks and their analysis. Section 3

outlines data and details of Stephen Campisi’s framework that we have adopted.

Section 4 presents and summarizes results. Section 5 concludes with a discussion of

concerns that a fixed income manager may have, and presents several suggestions for

improving the model in the future.

2
2: FIXED INCOME PERFORMANCE
MEASUREMENT

2.1 Overview of Fixed Income Management Process

Factors that drive the performance of bonds are fundamentally different from

those of stocks. As financial instruments, bonds and stocks differ in structure, pricing,

potential upside returns and market in which they trade. Investment managers

consider this when valuing bonds and making investment decisions, thus it is

important for performance measurement to account for the same difference.

Yield curves are essential to fixed income management as changes in the curve

have immediate impact on prices of fixed income securities. At any point in time, a

yield curve shows market consensus of where the interest rates are expected to be in

the future across different maturities (Colin, 2005). Moreover, yield curves carry an

embedded view on future inflation, economic growth, exchange rates, perceived default

probability of the issuer, and much more. Fixed income managers form investment

strategies with respect to their expectation of the movement in the yield curve (up,

down, steepen, flatter, etc). This is different from the equity approach where managers

assess the growth potential of a particular stock or sector and implement selection and

allocation strategies accordingly. Because of the large variety of fixed income

securities, different sources of risk, and wide range of scenarios in terms of yield

structure movements, fixed income attribution should go beyond simple selection and

allocation approach.

One of the major challenges of fixed income attribution is as mentioned earlier

the lack of uniformed approach. One of the reasons for this is that market pricing and

3
risk factors are not as straight forward as for equity securities. Another reason involves

data flow issues, mathematical background and computing knowledge. Lastly,

sophisticated fixed income models may be costly and out of reach for many

institutions or managers.

2.2 Fixed Income Performance Attribution Models

We have established earlier that in order for attribution to prove meaningful it

needs to account for the decision making process in portfolio management. According

to Campisi (2000), following are some of the characteristics of a comprehensive fixed

income attribution framework.

- Representative: consistent with the investment and decision-making process,

demonstrating attribution of return for taking on systematic risk;

- Rigorous: tells a story of what happened during the holding period; accurately

explains sources of over and under performance;

- Reasonable: offers a balance between rigor and cost;

- Responsive: provides ability to meet client needs by customizing benchmark to

match manager’s strategy;

When applied to fixed income portfolios, many equity-focused attribution

frameworks fail to address systematic risk drivers of bond returns, and ignore some or

all drivers of manager’s decision making process. For example, a manager may feel

optimistic about an economic outlook and hence overweighs the portfolio towards

corporate bond sector. A “Sector Allocation” Brinson-Fachler (1985) attribution model

may show a positive excess return due to the sector allocation decision. Unfortunately,

we cannot be exactly sure that excess returns were derived solely from the sector

overweighting. Such approach would ignore manager’s decision about the maturity of

4
the bonds. Perhaps the manager invested in long term corporate bonds, which

delivered a positive return because of a downward shift in long term interest rates. In

this case, a “Sector Allocation” model would provide misleading results.

The goal of fixed income attribution models is to show a link between changes in

the yield curve environment and portfolio performance. The appropriate models

should explain how the return was generated and distinguish skill from luck. Our

research shows that Campisi’s (2000) attribution methodology fits best with fixed

income portfolios. Our implementation of the framework, which we will further

present in a practical example, has been executed with enough rigor to tell a story of

how the value added was created.

Brinson and Fachler (1985) and Brinson et al (1986) commonly known as the

Brinson model has set a foundation for performance attribution. This approach is

widely used and generally expected in equity-style attribution. Often times Brinson

model is used for fixed income, however as discussed earlier this may not be the most

suitable technique.

Wagner and Tito (1977) use a duration approach to fixed income attribution

based on Fama (1972) where the duration was used as a measure of systematic risk as

opposed to beta in the original Fama framework. Duration alone, however, is not

sufficient to explain non parallel yield curve movements.

In explaining how actual portfolio returns were achieved Fong, Gifford, Pearson

and Vasicek (1983) framework decomposes return first on a macro level, and drills

down to more of a micro analysis. In simplest terms total return on the fixed income

portfolio can be contributed to the external changes in the interest rate environment

and management contribution. The change in the interest rate environment is that

one that management has no control over and can be partitioned into an expected

5
return on the Treasury portfolio and an unexpected return. The management

contribution can be further decomposed into three categories: return from maturity

management, return from spread/quality management and return from selection of

selected securities.

Kahn (1991) introduced a multi-factor single and multi period fixed income

attribution models. His multi-factor analysis is developed in great detail, while multi-

period performance offers a useful tool to distinguish skill from luck. The framework

identifies six different sources of fixed income return: portfolio moving closer to

maturity, default-free term structure has changes (sovereign curve moves), sector and

quality spreads have changes, unexpected cash flows, unexpected changes in quality

ratings, bond specific price changes.

Van Breukelen (2000) combines Wagner and Tito duration-based approach with

Brinson equity-style model. This “weighted duration approach” first calculates

duration contribution to the total return and then computes allocation and selection

components.

Campisi’s (2000) framework on macro level decomposes total returns into

income return and price change. The price change can be further partitioned into

duration and yield change. Where yield change is composed of treasury change and

spread change. Campisi model is easy to implement and requires minimum inputs,

while at the same time considers the management process and provides meaningful

decomposition of the total return.

Silva Jr. et al (2009) uses a simple combination of duration-based attribution

with asset selection. First, a sovereign yield curve is fitted using the Nelson-Segal

(1987) approach. Second, three hypothetical portfolios are created so that the returns

may be classified according to appropriate factors.

6
 From what we found, Campisi’s (2000) framework is the only one to account

explicitly for portfolio’s income component. Many investors choose bond portfolios

because they provide a predictable stream of cash flows, therefore, we feel that it is

important to make sure that attribution results account for income return. In North

America, it is the market convention to quote bond prices in terms of “clean price”,

which is the price that is most often used in attribution calculations. If attribution

professionals take the extra step to incorporate accrued interest in price calculations

(“dirty price”), then other models may provide results that account for income returns.

7
3: DATA AND METHODOLOGY

3.1 Data

Our study focuses on building an instrument suitable for the use at SIAS fixed

income portfolio. SIAS fixed income portfolio is benchmarked against DEX Universe

Bond Index. Accordingly, SIAS portfolio data is obtained from BNY Mellon

Workbench platform and DEX data is collected from the PC Bond application.

Selected data covers a period between March 31st, 2010 and June 30, 2010. Following

are inputs that went into our model:

- Total return - calculated as a percentage price change over the holding period,

plus an income component.

- Weight - this model takes in the beginning/ending weight and assumes that

the weights were held constant over the period

- Coupon - annualized coupon rate

- Price — price at the beginning/end of the period

- Duration - modified duration extracted out of PC bond

- Key rate duration (KRD) — sourced from PC Bond

- Treasury yield curve at the beginning and end of the period.

Below we present data for the benchmark and portfolio respectively.

8
 Table 1: Benchmark Data
Sector Total Prices
Sectors Coupon Duration
Weights Return at t-1
Federal 47.27% 0.0193 3.84 5.04 $ 105.75
Prov 24.46% 0.0165 5.53 8.23 $ 110.48
Muni 1.42% 0.0128 5.32 6.18 $ 106.16
Corp 26.85% 0.0152 5.67 5.37 $ 107.34
Total: 100% 1.0637 4.77 5.92 $ 107.33

Table 2: Portfolio Data

Sector Total Prices
Sectors Coupon Duration
Weights Return at t-1
Federal 17.73% 0.0111 4.55 2.14 $ 105.66
Prov 46.41% 0.0224 5.60 8.05 $ 110.51
Muni 3.55% 0.0132 5.54 5.06 $ 109.90
Corp 32.31% 0.0153 5.57 4.70 $ 106.46
Total: 100% 1.0621 5.40 5.81 $ 108.32

Below, sector weight graphs show that portfolio is overweight credit risk in

provincial, corporate and municipal sectors. We may infer that portfolio manager’s

allocation strategy revolves around spreads narrowing. In a flight-to-quality scenario,

however, we would expect spreads to widen and portfolio to underperform as a result.

Figure 1: Benchmark Sector Allocation Figure 2: Portfolio Sector Allocation

9
 In terms of measuring parallel yield curve shift, and twist effect, we require

portfolio and benchmark sensitivity to changes in 5-year rate. We pick 5-year point as

the key rate because both portfolio’s and benchmark’s durations are near the 5 year

mark. This part of the attribution process offers flexibility. For portfolios that are

heavily invested in long maturity bonds, a 10-year point may show a more meaningful

result. Below we present KRD (key rate duration) data for both benchmark and

portfolio.

Table 3: Five-Year Key Rate Durations

Benchmark Portfolio
Beginning Ending ΔKRD Beginning Ending ΔKRD
Federal 4.98 5.08 0.1 2.14 1.92 -0.22
Provincial 8.07 8.26 0.19 7.91 7.88 -0.03
Municipal 6.11 6.79 0.68 5.03 4.85 -0.18
Corporate 5.29 5.38 0.09 4.63 4.55 -0.08
Total 5.84 5.97 0.13 5.72 5.66 -0.06

Key rate durations measure sensitivity of the portfolio and the benchmark to

changes in five-year yields, holding all other maturities constant.

Next, we present the interest rate environment at the beginning and at the end

of the attribution period. Figure 4 shows a scenario that includes an upward shift in

short term interest rates and downward move in long end of the curve. As we will see

further, this move will be decomposed into a shift and twist components. What

follows is a detailed description methodology for fixed income attribution analysis.

10
 Table 4: Treasury Yield Curve Change Figure 3: Treasury Yield Curve Movement

Yield Yield
Years (t-1) (t) Δ Yield
0.08 0.21 0.31 0.1
0.16 0.24 0.4 0.16
0.25 0.29 0.51 0.22
0.5 0.47 0.75 0.28
1 0.94 1.04 0.1
2 1.73 1.44 -0.29
3 2.27 1.89 -0.38
4 2.8 2.3 -0.5
Figure 4: Treasury Yield Curve Change
5 2.91 2.36 -0.55
7 3.11 2.78 -0.33
10 3.57 3.1 -0.47
15 3.82 3.36 -0.46
20 4.08 3.61 -0.47
25 4.11 3.67 -0.44
30 4.07 3.65 -0.42
40 4.07 3.65 -0.42
41 4.07 3.65 -0.42

3.2 Methodology

In our glossary section we offer details on some of the terms and formulas used

in the framework. Detailed formulas are outlined in appendices. As we stated earlier

our methodology closely follows that outlined by Campisi (2000). After collecting the

necessary data and importing it into our model, we define total return as the price

change effect and income effect over the attribution period. Appendix 1 provides

detailed formulas for calculating total return.

11
 Price Change
Total Return Effect

Income Effect

Figure 5: Total Return Decomposition

We first calculate the Income effect by dividing the coupon rate by the ending

price. This is equivalent to Current Yield, not to be confused with, Coupon Yield or

Yield To Maturity (YTM). Unlike YTM, Current yield does not reflect reinvestment

risk or total return over the life of the bond. Moreover, current yield fluctuates with

changes in bond prices, and doesn’t assume a constant reinvestment rate. Another

component of total return - Price change effect is calculated as time weighted return

for the period. We take the change in bond’s clean price over the time period and

divide it by the price at the end of the period.

We can further decompose Price Change Effect it into four categories: shift,

twist, spread and selection effects.

Shift Effect

Twist Effect
Price Change
Effect
Spread Effect

Selection Effect

Figure 6: Price Change Effect Decomposition

12
 Shift and twist effects are derived from change in the reference curve (usually a

risk-free treasury curve) and portfolio’s sensitivity to curve movements. This can be

further broken down into sector level analysis. For detailed calculations refer to

Appendix 2.

We then multiply the change in treasury curve with negative modified duration

to get the total treasury return. We further decompose the treasury effect into shift

and twist. To help us clarify shift and twist effects we first calculate the change in key

rate durations for both portfolio and a benchmark. This is accomplished by taking the

difference between ending and beginning key rate duration values. Shift effect is a

product of the change in key rate duration and negative modified duration. Twist

effect is obtained by multiplying the difference in changes in the yield curve and key

rate duration with negative modified duration. Spread effect is calculated by

subtracting income and treasury effects from the total return. Selection effect is the

amount remaining once income, treasury, and spread effects are subtracted from the

total return. Consequently for the benchmark there is no selection effect, however for

the portfolio there will be a selection effect relative to the benchmark. Moreover,

selection effect may incorporate the difference in convexities. This will be addressed

further in the conclusion.

13
4: RESULTS

In our empirical example, the fixed income portfolio has underperformed relative

to DEX Universe benchmark by 16 bps. We attempt to explain where this different-

from-the-benchmark performance came from. To do so we look at both benchmark

and portfolio performance during the attribution period. We can clearly see that the

portfolio has outperformed the benchmark on income, treasury and spread elements.

However, underperformance was due to the negative selection effect.

Figure 7: Contribution to Return for Portfolio and Benchmark

First, we look at the benchmark in more detail. We explain the total return by

decomposing it into income, treasury and spread. There is no selection component to

14
the benchmark return because the assumption is that the benchmark includes the

entire universe of securities. Selection effect is relevant for portfolio management as it

demonstrates the skill of actively managing the portfolio. Income return for the

benchmark represents the income earned during the attribution period. Treasury

return is further decomposed into parallel effect (shift) and non-parallel effect (twist).

Spread return shows how much credit exposure the portfolio had and how much

spread return was generated as a result of the spread changes.

Table 5: Analysis of Benchmark Return

Treasury Spread Return
Bench Income Return Shift Twist Return Selection (Yield)
Total 0.011 2.656 -0.770 3.426 -1.603 0.000 1.064
Federal 0.009 2.750 -0.504 2.029 -2.740 0.000 0.019
Provincial 0.013 3.188 -1.564 4.683 -3.184 0.000 0.016
Municipal 0.013 2.597 -4.202 7.561 -2.597 0.000 0.013
Corporate 0.013 2.735 -0.483 3.356 -2.733 0.000 0.015

Portfolio strategy was to generate more income by underweighting federal bonds

and overweighting corporate and provincials bonds, which delivered higher income

return. The portfolio income return was positive, as was the treasury return. Spread

returns for benchmark and portfolio were negative as a result of widening in spreads,

however the spread excess return was positive. Selection for the quarter was negative

mainly driven by municipal sector.

Table 6: Analysis of Portfolio Return

Treasury Spread Return
Port Income Return Shift Twist Return Selection (Yield)
Total 0.012 2.677 0.349 2.328 -1.573 -0.054 1.062
Federal 0.011 0.648 0.471 0.177 -1.163 0.516 0.011
Provincial 0.013 3.051 0.242 2.809 -3.115 0.073 0.022
Municipal 0.013 2.750 0.911 1.839 -2.126 -0.623 0.013
Corporate 0.013 2.515 0.376 2.139 -2.392 -0.120 0.015

15
 As we can see from attribution results, portfolio underperformed from the twist

effect due to being underweight federal long-maturity bonds. During the attribution

period, the yield curve has twisted resulting in a decline in long-term yields at the

same time driving the price of long-term bonds up. This performance was direct

consequents of management’s decision to underweight long-term bonds by remaining

short duration.

Table 7: Detailed Excess Portfolio Attribution Analysis by Sector

   Income Shift Twist Spread Selection Total

Federal 0.0017 0.9748 -1.8523 1.5763 0.5160 -0.0082
Provincial 0.0002 1.8052 -1.8734 0.0696 0.0735 0.0059
Municipal 0.0001 5.1132 -5.7218 0.4706 -0.6230 0.0005
Corporate -0.0001 0.8593 -1.2178 0.3410 -0.1203 0.0002
Total 0.0014 1.1190 -1.0977 0.0301 -0.0545 -0.0016

In summary, the portfolio has done better then the benchmark in three out of

the five categories. Non-parallel changes in the yield curve have contributed to

underperformance, as did poor selection. However, the underperformance was not

significant, and it can very well be described by the management’s strategy. Most

fixed income portfolios are managed for long term, thus small deviations in the short

run are not uncommon.

1.1190

0.0014 0.0301

-0.0545 -0.0016

-1.0977
Income Shift Twist Spread Selection Total

Figure 8: Summary of Excess Returns

16
5: DISCUSSION AND CONCLUSION

5.1 Discussion

Our model uses a buy-and-hold approach to attribution as oppose to

transaction-based approach. The buy-and-hold method assumes that portfolio

weighting is constant over the attribution period. It captures a “snapshot” of the

portfolio weights and holdings. The buy-and-hold approach assumes that there are no

transaction costs and that all transactions happen at the end of the holding period.

Consequently, the shortcomings of this approach are that it ignores transaction costs

and change in weights of individual holdings. (Spaulding 2003). The buy-and-hold

approach is quite common in fixed income analysis and given the infrequent activity

in the portfolio used in our empirical section, we believe this approach to be relevant

for our analysis.

Additionally, it is important to note that our model does not include convexity.

Bond price change is approximated by duration and spread or yield change. When we

add convexity, we move away from a linear model to a quadratic one. While linear

model allow for straightforward calculations of various return effects, quadratic model

offers no mathematical equivalence for (Duration + Convexity) * Treasury Change,

and (Duration + Convexity) * Spread Change formulas. The two equations would not

be mathematically equivalent. We also know that convexity tends to have a very

small impact on excess basis between benchmark and portfolio. However, some

managers do take active convexity bets, for example through asset/mortgage backed

securities. In that case, convexity effect can be calculated at the total yield level. We

would calculate portfolio's and benchmark's total yield change by using respective

durations and convexity. Further we could infer from both results the return

17
component generated by an active bet in convexity. Although we don't explicitly

break out convexity effect in the existing model, it is aggregated in the Selection effect

with other factors.

The introduction of fixed income attribution model has provided useful insight

into the nature of the portfolio returns. The goal was to assist fund management in

forming strategies and help client better understand sources of return. We have

adopted Campisi framework to fixed income attribution and calibrated it according to

the needs of this portfolio. In our review, we found Campisi’s method to be most

compatible with the management process.

5.2 Conclusion

We have showed a number of different approaches to performance attribution.

We explained how unique fixed income environment is and why it requires a special

approach to attribution. Furthermore, we implemented an empirical example using

the Campisi (2000) method. Empirical calculations used a Canadian fixed income

portfolio. Our performance attribution model is parsimonious yet it provides useful

insight into the sources of return. More importantly, this model is unique because it

includes income return in addition to price return. Generally, fixed income models

focus on price return only and ignore income return. This is incorrect, because bonds

are primarily income instruments, and over time price returns tend to revert to zero so

that most of the total long-term effect is generated through income. Our model also

reflects the management’s decision-making process.

Additional improvements can be done to our model. First, we could introduce a

transaction-based approach to accounting for returns. Another improvement is to use

a more sophisticated method such as Nelson-Segal to interpolate the curve when

determining changes in Duration Matched Treasuries (DMT). Furthermore, our

18
approach is limited to sector level attribution; possible enhancement would introduce

attribution down to the security level.

The attribution model that we have presented in the paper decomposes total

return into components related to portfolio income, and yield curve movements,

however it can also be applied to portfolio volatility. To better understand portfolio

volatility and greatest sources of tracking error when compared to benchmark, total

portfolio volatility can also be decomposed using this model.

19
APPENDICES

20
Appendix 1: Total Return Calculation
V   I   V
  (1)
V  

Income Component is dependent on selected attribution period. For annual

attribution period use (2) but for quarterly use (3).

(2)

  (3)

Example:

Benchmark
Coupon
Price Price Price Return
Sector Mar-31 Jun-30 Return Coupon (quarterly) Total
Federal 105.75 108.04 0.01024 0.0384 0.0091 0.01932
Provincial 110.48 112.26 0.00394 0.0553 0.0125 0.01646
Municipal 106.16 107.88 0.00023 0.0532 0.0125 0.01276
Corporate 107.33 108.13 0.00200 0.0567 0.0132 0.01520
Total 107.33 109.11 0.01658 0.0477 0.00011 0.01637

109.11  0.0444  107.33 1

  0.0167 or 1.67%
107.33 4

Sector Prices were obtained by taking a weighted average of individual security

prices within each sector.

21
Appendix 2: Treasury Return decomposed into Shift and Twist

Changes in Key Rate Durations, as well as changes in Duration Matched

Treasury (“DMT”) are needed for calculating shift and twist effect. Key Rate

Durations can be manually calculated or obtained from the PC Bond application.

DMT however requires its own calculation using the method that we have adopted.

ΔDMT stands for change in treasury rate corresponding to each sector duration. In

the data that is available at sector level, we are given duration for the total portfolio

and each sector individually.

Example:
Portfolio DMT DMT
Sectors Durations at t-1 at t ΔDMT
Federal 2.14 1.806 1.503 -0.303
Provincial 8.05 3.271 2.892 -0.379
Municipal 5.06 2.916 2.373 -0.543
Corporate 4.70 2.877 2.342 -0.535
Total 5.81 2.99 2.53 -0.460

DMT (at t-1) is a treasury rate on a 2.14 year treasury bill at the beginning of

the attribution period. DMT (at t) is a treasury rate on a 2.14 year treasury bill at

the end of the attribution period. It is unlikely that we are going to find 2.14 year

treasury bill trading in the market at any given point in time. As such, we will be

required to interpolate it’s yield from a standard treasury yield curve. There are

several choices available for interpolation, with the simplest one being linear

interpolation. Models that are more complex may apply quadratic, cubic interpolation,

or Nelson-Siegel (1987) approach. As long as interpolation approach is consistent for

both benchmark and portfolio, the bias is kept to minimum.

22
Appendix 3: Comprehensive list of formulas used in attribution
calculations

       

1
     

1
Changes in 5 year or 10 year key rate duration

   

     

   
   

     

23
Appendix 4: Excel model inputs

24
Appendix 5: Duration Matched Treasury calculations — Part1
In order for the excel MATCH() function to find the upper and lower bounds

for the duration matched treasury to interpolate from, Yield curve rates need to be

sorted in both Ascending and Descending order.

Yield Curve Inputs

25
Appendix 6: Duration Matched Treasury calculations — Part2

Match() and VLOOKUP() functions are used

extensively to sort data for easy yield curve rate

interpolation. The interpolation of the Duration

Matched Treasury is calculated in columns AC and

AD. Change in DMT which is then used in the model

is calculated in column AE.

26
Appendix 7: Excel model attribution calculations — Part 1

27
Appendix 8: Excel model attribution calculations — Part 2
Formula view of Appendix 7 shows how formulas in Appendix 3 are used to perform
calculations.

28
Appendix 9: Excel model graphical output

29
Glossary
Duration (modified): a linear measure of the sensitivity of the bond's price to

interest rate changes.

Key Rate Duration (KRD): measures the sensitivity of a security or the value

of a portfolio to a 1% change in yield for a given maturity, holding all other maturities

constant.

Duration Matched Treasury (DMT): a point on the treasury yield curve that

corresponds to a specific duration number. i.e. 2.14 duration would correspond to the

yield on a 2.14 year treasury bond.

Convexity: a measure of the curvature of how the price of a bond changes as

the interest rate changes. Second derivative, that measures how the duration of a

bond changes as the interest rate changes.

Current Yield: coupon rate divided by the price of the security. It represents

the return an investor would expect if they purchased the bond and held it for a

year/quarter/month/day.

Yield to Maturity (YTM): return anticipated on a bond if it is held until the

maturity date. Assumes that coupons can be reinvested at YTM rate.

Spread (credit): difference in yield between securities with different credit

quality i.e. difference in yields on a Corporate and a Treasury bond.

30
REFERENCE LIST

[1] Brinson, G., and N. Fachler. 1985. “Measuring Non-U.S. Equity Portfolio
Performance”, Journal of Portfolio Management, vol. 11, no. 3 (Spring), pp. 73-76.

[2] Brinson, G., Hood, R. and Beebower, G. 1986. “Determinants of portfolio

performance”, Financial Analyst Journal, (July/August), pp. 39—44.

[3] Campisi, S. 2000. “Primer on fixed income performance attribution, Journal of

Performance Measurement”, vol.4, n° 4, (Summer), pp 14-25.

[4] Charles R. Nelson and Andrew F. Siegel. 1987. “Parsimonious Modeling of Yield
Curves”, The Journal of Business, Vol. 60, No. 4. (October), pp. 473-489.

[5] COLIN, Andrew. Fixed Income Attribution. John Wiley & Sons, Ltd. 2005.

[6] Fama, E. 1972.”Components of investment performance”, Journal of Finance 27

(2), pp. 551-67.

[7] Fong, Gifford, Charles Pearson and Oldrich Vasicek, 1983. “Bond performance:
Analyzing sources of return”, Journal of Portfolio Management 9, pp. 46-50.

[8] Kahn, Ronald N. 1991. “Bond performance analysis: A multifactor approach”,

Journal of Portfolio Management 18, pp. 40-47.

[9] Silva, De Carvalho, Ornelas. 2009. “A Simple Performance Attribution

Methodology for Fixed Income Portfolios in a Typical Central Bank Investment
Process”.

[10] Spaulding, D. 2003. “Holdings vs. transaction-based attribution — an overview”,

Journal of Performance Measurement, Fall, 52—56

[11] Tito, D., and W. Wagner. 1997. "Definitive New Measures of Bond Performance
and Risk", Wilshire Associates.

[12] Van Breukelen,G. 2000. "Fixed Income Attribution."Journal of Performance

Measurement, vol. 4, no. 4 (Summer), pp. 61-68.

31