CFA INSTITUTE RESEARCH FOUNDATION / LITERATURE REVIEW

PERFORMANCE ATTRIBUTION
HISTORY AND PROGRESS

CARL BACON, CIPM

Research
Foundation
Literature Review
PERFORMANCE ATTRIBUTION
History and Progress

Carl Bacon, CIPM

 Statement of Purpose

The CFA Institute Research Foundation is a not-

for-profit organization established to promote
the development and dissemination of relevant
research for investment practitioners worldwide.

Neither CFA Institute Research Foundation, CFA Institute, nor the

publication’s editorial staff is responsible for facts and opinions
presented in this publication. This publication reflects the views of
the author(s) and does not represent the official views of CFA Institute
Research Foundation.

CFA®, Chartered Financial Analyst®, and GIPS® are just a few of the trademarks owned
by CFA Institute. To view a list of CFA Institute trademarks and the Guide for the Use of
CFA Institute Marks, please visit our website at www.cfainstitute.org.
© 2019 CFA Institute Research Foundation. All rights reserved.
No part of this publication may be reproduced, stored in a retrieval system, or transmitted,
in any form or by any means, electronic, mechanical, photocopying, recording, or otherwise,
without the prior written permission of the copyright holder.
This publication is designed to provide accurate and authoritative information in regard
to the subject matter covered. It is sold with the understanding that the publisher is not
engaged in rendering legal, accounting, or other professional service. If legal advice or other
expert assistance is required, the services of a competent professional should be sought.
Cover photo credit: egon69 / E+ / Getty Images
ISBN 978-1-944960-89-6
 Contents

Introduction............................................................................................ 1
Early Development. . ................................................................................ 3
Foundations............................................................................................ 10
Multiperiod Analysis. . .............................................................................. 15
Multicurrency Attribution....................................................................... 20
Types of Attribution................................................................................ 23
Risk-Adjusted Attribution........................................................................ 26
Fixed-Income Attribution. . ...................................................................... 27
Other Attribution Issues.......................................................................... 30
Conclusion: The Evolution of Attribution Methodologies. . ..................... 33
Bibliography............................................................................................ 35

This publication qualifies for 2.5 CE credits under the guide-

lines of the CFA Institute Continuing Education Program.

© 2019 CFA Institute Research Foundation. All rights reserved.  iii

Performance Attribution
History and Progress

Carl Bacon, CIPM

Chief Adviser, StatPro

Introduction
The objective of performance attribution, as stated by Menchero (2000), is to
explain portfolio performance relative to a benchmark, identify the sources
of excess return, and relate them to active decisions by the portfolio man-
ager. Hensel, Ezra, and Ilkiw (1991) defined attribution as the “mathemati-
cal process of explaining an investment return by relating it to the different
risk-taking decisions implicit in the portfolio, and the extent to which each of
those risks was rewarded or penalized in the capital markets” (p. 66). Colin
(2014) said that the purpose of attribution is to disentangle the fund’s overall
return into the component returns generated by each risk.
In other words, attribution measures which of your investment decisions
about the portfolio’s underlying risks worked and which did not. This infor-
mation is critical business intelligence for anyone involved in selecting, man-
aging, or marketing investments. As the authors noted above, Fischer and
Wermers (2013) considered attribution analysis to be the “description and the
quantification of key factors on the returns of investment portfolios” (p. 395).
DiBartolomeo (2003) considered performance attribution to be the process
of disentangling component portions of the observed returns to draw conclu-
sions about the strengths and weaknesses of the investment process.
Note that these definitions are similar. The central question of an attribu-
tion analysis is thus the following: To which investment decisions can the active
return of a portfolio relative to its benchmark be traced over a given period?
Formally, Bacon (2008) stated that “performance attribution is a tech-
nique used to quantify the excess return of a portfolio against its benchmark
into the active decisions of the investment decision process” (p. 117). Murira
and Sierra (2006) considered performance attribution the ex post complement
to the ex ante risk decomposition provided by the risk measurement process.
“Performance attribution” is, therefore, perhaps a misleading term: As
these descriptions suggest, the excess, active, or relative return is examined,
not the whole of the return (or absolute return). “Relative return attribution”

© 2019 CFA Institute Research Foundation. All rights reserved.  1

Performance Attribution

or simply “attribution analysis” is perhaps better terminology. Absolute attri-

bution, or simply contribution analysis, would explain the sources of return of
a portfolio in isolation rather than in comparison with a benchmark.
As described in Bacon (2008), attribution analysis is a critical manage-
ment tool for several key stakeholders in the asset management process.
Above all, it is the main tool for performance analysts because it allows them
to ensure that the return is correct. It allows them to participate in the invest-
ment decision process by enabling them to demonstrably add value and thus
justify their salary and effort. Return attribution, together with risk analy-
sis, is the tool that allows the analyst to understand the sources of return in
the portfolio and to communicate that understanding to portfolio managers,
senior managers, and asset owners.
Investment management firms themselves, and the portfolio managers
they employ, are also major users of attribution analysis. Attribution analy-
sis provides a good starting point for a dialogue with asset owners on the
positive and negative aspects of recent performance. Attribution analysis can
be used aggressively to identify underperformance early and provide clients
with a thorough explanation of underperformance and, hopefully, a rationale
for improving performance. Attribution analysis is also crucial in gaining the
confidence of asset owners by demonstrating a good understanding of the
drivers of performance.
Portfolio managers inhabit the front office of asset management firms;
typically, performance analysts inhabit the middle office. The back office is
also able to use daily, security-level attribution, however, as a diagnostic tool
for identifying pricing and transaction errors.
Bacon (2008) also stated that “senior management take an active interest
in attribution analysis to provide them with a tool to monitor their portfolio
managers. They will be keen to identify performance outliers—good or bad—
and to ensure that value is added consistently across the firm” (p. 117).
Hence, attribution analysis has numerous uses across a wide range of
asset management stakeholders.
Lord (1997) suggested that the characteristics of a good attribution sys-
tem would include the following:
•• a consistent approach for analysing returns,
•• a solid theoretical framework for valuation,
•• the ability to analyse portfolios and indexes consistently,
•• attribution factors consistent with decision variables of the portfolio strat-
egy process,

2 © 2019 CFA Institute Research Foundation. All rights reserved.

 Performance Attribution

•• summary attribution analysis constructed from security details, which

themselves should be available for inspection,
•• transactions, and
•• robust analytic systems and high-quality data to provide meaningful
attribution results.
Murira and Sierra (2006) suggested that there is general agreement that
an attribution framework must meet two key requirements:
1. It should be consistent with the investment decision-making process of
the portfolio. Performance attribution is most useful if it mirrors the
investment management process because, in that case, it enables the iden-
tification and explanation of the sources of excess return.
2. It should be compatible with the organization’s performance measurement
and risk measurement systems. Managers would have difficulty making
effective use of the information provided by the performance attribution
process if the data being explained by this process were inconsistent with
the data originating from the internal risk and performance systems.
As in many literature reviews, here the approach to the task is historical—
beginning with the first attempts at performance measurement, about a half-
century ago, and progressing to the present.

Early Development
Arguably, comparative performance measurement began with Dietz’s (1966)
Pension Funds: Measuring Investment Performance and the Bank Administration
Institute (BAI 1968) report Measuring the Investment Performance of Pension
Funds for the Purpose of Inter-Fund Comparison. The purpose of the BAI study
was to develop methods of comparing the performance of pension funds to
differentiate between the abilities of their respective managers. The main
conclusions of their study are still relevant:
1. Performance measurement returns should be based on market value, not
cost.
2. Returns should be total returns.
3. Returns should be time weighted.
4. Performance should include risk as well as return.
5. Funds should be classified according to their investment objectives.

© 2019 CFA Institute Research Foundation. All rights reserved.  3

Performance Attribution

Although the report did not recommend the use of attribution analysis—
instead, emphasizing the importance of the total return of the portfolio—it
did suggest that comparison of various sector returns within the portfolio
might be useful. Interestingly, Eugene Fama was later invited to write and
wrote the chapter on risk for this report.
Fama Decomposition.  Fama was the first to fully delve into the sub-
ject of attribution analysis, which he did in “Components of Investment
Performance” (Fama 1972). In this seminal paper, Fama suggested breaking
down observed return into the part resulting from ability to pick the best
securities at a given level of risk (which he at the time called “selectivity”;
now, it is often referred to as “selection return”) and the part that is the result
of predictions of general market price movements—that is, the return from
systematic risk—as follows:

r − rF = r − β × (b − rF ) − rF + β × (b − rF ), (1)
   
Excess Selectivity Systematic risk
return

where:
r = average portfolio return
b = average market or benchmark return
β = portfolio systematic risk
rF = average risk-free rate
If a portfolio is fully diversified, it incurs no specific risk and the total
portfolio risk will equal the systematic risk. Portfolio managers will give up
diversification, however, to seek additional return. Therefore, the return from
selectivity can be broken down into (1) the additional return required to jus-
tify sacrificing a given amount of diversification and (2) the net return from
selectivity (the remainder of the manager’s value added). The return from
selectivity is equivalent to Jensen’s alpha, α.
The term “diversification” (used in a different context from the one we
would understand today) “is the return required to justify moving away from
the benchmark and taking on specific risk” (Bacon 2013, p. 88). To calcu-
late this diversification return, you first have to calculate the effective beta
required such that systematic risk alone is equivalent to the total portfolio
risk. This beta is the Fama beta, calculated as follows:

σ
βF = , (2)
σb

4 © 2019 CFA Institute Research Foundation. All rights reserved.

 Performance Attribution

where:
σ = portfolio risk
σ b = benchmark risk
Therefore, the return required to justify not being fully diversified is cal-
culated by using the difference between the Fama beta and the portfolio beta
as follows:

d = (β F − β) × (b − rF ). (3)

Note that the “Fama beta will always be greater than or equal to the port-
folio beta since total risk is greater than or equal to the systematic risk of the
portfolio” (Bacon 2013, p. 89).
The net return from selectivity (“net selectivity,” for short) is the remain-
ing return from selectivity after deducting the amount of return required to
justify not being fully diversified:

Net selectivity = Selectivity – Diversification.

Net selectivity = SNet = α − d .

If net selectivity is negative, the portfolio manager has failed to justify the
loss of diversification.
Timing is the return from the manager’s variations in systematic risk
(beta) around some policy or target amount:

β × (b − rF ) = (β − β I ) × (b − rF ) + β I × (b − rF ), (4)
      
Systematic Manager ’ s Systematic Investor ’ s Systematic
Risk Risk or Timing Risk

where β I = the investor’s systematic risk or target risk.

Figure 1 illustrates Fama’s decomposition for portfolio A. A ′ represents
the return from systematic risk (conventional beta) plus the risk-free rate, and
A ′′ represents the return from the Fama beta plus the risk-free rate.
Fama’s (1972) “Components of Investment Performance” is the first paper
to seriously analyse the sources of return of an investment portfolio and,
therefore, represents the birth of attribution analysis.
Notional Funds.  In response to the BAI (1968) paper, a working group
was set up towards the end of 1970 by the Society of Investment Analysts
(SIA) in the United Kingdom to make recommendations on the subject

© 2019 CFA Institute Research Foundation. All rights reserved.  5

Performance Attribution

Figure 1.  Fama Decomposition

A
Systematic Return from
Selectivity
r–
Net Selectivity
A” Diversification

A’
Return from

Risk

β βF
r–F

Systematic Risk (β)

Source: Redrawn from a diagram in Fama (1972).

of portfolio performance measurement for pension funds. Its approach,

eventually published in 1972 (SIA 1972), had two aims: (1) the determination
of a standard method of calculating a portfolio’s return, so that funds could
be fairly compared, and (2) the decomposition of the performance of the
fund into two components—selection of individual stocks and selection of
stock sectors. The first aim is achieved by calculating the time-weighted rate
of return using a simple approximation (which later became known as the
analyst’s test method and ultimately became redundant as more accurate
time-weighted returns became available). The second aim, the analysis of
performance, is produced by constructing “notional funds” based on market
indexes and comparing them with actual funds.
This early work dealt with equity-market sectors as the macro level of
selection and securities within the sectors as the micro level, but common
practice now is to regard asset classes (stocks, bonds, etc.) as the macro level
and securities within the asset class as the micro level. Thus, our discussion
will proceed from this latter starting point.
The working group established two notional funds as follows:
•• a fully restrained fund (index returns, benchmark weights) and
•• a partly restrained fund (index returns, actual weights).
Note that the actual fund consists of actual returns and actual weights.
The fully restrained fund (sometimes called the “fully notional fund”)
represents the return on the policy portfolio—that is, the strategic bench-
mark or policy weights set by the asset owner (represented by the pension

6 © 2019 CFA Institute Research Foundation. All rights reserved.

 Performance Attribution

fund trustee) multiplied by the market index returns for each asset class. In
mathematical notation,
i =n
b = ∑Wi × bi , (5)
i =1

where:
i =n
Wi = weight of the benchmark in the ith asset class (note ∑Wi = 1)
bi = return of the index in the ith asset class i =1

The partly restrained fund (sometimes called the “semi-notional fund”)

represents the intermediate return resulting from the actual weights deter-
mined by the manager multiplied by the market index returns. Again, in
mathematical notation,
i =n
bS = ∑ wi × bi , (6)
i =1

i =n
where wi is the weight of the portfolio in the ith asset class (note ∑ wi = 1).
The actual portfolio return in mathematical notation is i =1

i =n
r = ∑ wi × ri , (7)
i =1

where ri is the return of the portfolio assets in the ith asset class.
The performance of a fund—in this case, a pension fund—thus depends
on two factors: the selection of asset classes (typically, now described as asset
allocation) and the selection of securities within an asset class. (As noted ear-
lier, this analysis can also be applied to sector allocation and stock selection
within sectors in an exactly parallel way.) The contribution of each of these
components can be found by breaking down the transition from the actual
fund to the fully restrained fund as shown in Table 1.

Table 1.  Source of Profit/Loss

Selection of asset classes (asset allocation) Partly restrained fund minus fully restrained fund

Selection of securities Actual fund minus partly restrained fund

Total Actual fund minus fully restrained fund

© 2019 CFA Institute Research Foundation. All rights reserved.  7

Performance Attribution

The success of the manager’s asset allocation is measured by comparing

the performance of the partly restrained fund with that of the fully restrained
fund. The success of the manager’s security selection is measured by compar-
ing the performance of the actual fund with that of the partly restrained fund.
The SIA (1972) report is thus crucial because it introduces the concepts
of macro and micro levels of manager decision making and intermediate
notional portfolios to calculate these effects.
In “Investment Performance of Pension Funds” (Holbrook 1977), the
British actuary J.P. Holbrook, in effect, pulled together the BAI (1968)
report, Fama’s (1972) decomposition, and the SIA (1972) report. This virtu-
ally unknown paper provides the foundation for much of the development
in performance analysis that followed. Directing his advice to pension fund
trustees, Holbrook defined the decision-making process as follows:
The practical management of a pension fund investment portfolio involves
decisions at three levels.
i) Policy, i.e. the proportions of the assets which, as a long-term
aim, the trustees wish to hold in the different investment mar-
kets. The fundamental element of policy is the split between
equity-type and fixed-interest investments. Definition of pol-
icy may extend to a split of equity-type investments between
UK equities, overseas equities and properties, specific pro-
portions being set for each. If the trustees are not experts in
long-term investment, they will tend to formulate policy in
discussion with specialist advisers; it is gratifying to record
that actuaries who advise on the financing of pension schemes
are increasingly becoming involved in such discussions.
ii) Strategy, i.e. decisions to depart from the policy proportions
in the light of current market conditions, including decisions
to hold part of the fund on short-term deposit. The trustees
will usually seek advice from experts in the various markets
regarding the short-term outlook, up to the time of the next
investment meeting and over the next year or so. Except where
trustees delegate strategic decisions to managers and advisers,
investment meetings are usually held at short regular intervals,
and arrangements are made for special meetings whenever it
is thought that there has been a material change in market
prospects. Often trustees authorize the managers concerned to
determine the timing of purchases; thus, a manager is able to
retain part of the assets entrusted to him in the form of short-
term deposits.

8 © 2019 CFA Institute Research Foundation. All rights reserved.

 Performance Attribution

iii) Selection, i.e. the choice of the particular investments to hold,

buy or sell within the various markets. Again, practice varies;
many trustees give their managers carte blanche in this area—
some keep close control over day-to-day transactions. (p. 20)
Holbrook went on to describe the trustees’ needs. First, they need to
know whether the fund performed well, moderately, or badly. This knowledge
requires the calculation of return, comparisons with market returns, and con-
sideration of risk:
The performance of the fund as a whole, and within the different markets
and sectors, must be examined separately. Secondly, the trustees need to
be advised on the reasons for their fund’s performance; this may be inves-
tigated by analysing the results to show contributions to total performance
made by policy, strategy, and selection decisions. Finally, the trustees need
help in interpreting the results as a guide to future action. Identification of
points of strength and weakness in past performance may lead to reconsid-
eration of policy and practice, of the decision-making process, and possibly
to a change of manager. (p. 21)
To facilitate the breakdown of returns into policy, strategic, and selection
decisions, Holbrook suggested the return calculation of three notional funds:
rm = based on the model,
r p = based on the trustees’ policy proportions, and
rA = based on actual allocations.
Then, the geometric excess return, g, for the whole fund is given by

(1 + r ) (1 + r p ) (1 + rA ) (1 + r )
(1 + g ) = = × × . (8)
(1 + rm ) (1 + rm ) (1 + r p ) (1 + rA )

The first factor represents the contribution to total performance resulting

from the decision to adopt a policy that differs from the equity and fixed-
interest proportions in the model [the industry standard allocation between
equity and fixed income at the time]. The second, the contribution from
the strategic decision to depart from policy proportions. The third, the
contribution from selection decisions for the fixed-interest and equity
sectors taken together. (Holbrook 1977, p. 31)
This passage is the first articulation of successive notional portfolios each
of which isolates one step in the investment decision process. The fact that it
was written in 1977 shows how far ahead of his time Holbrook was.

© 2019 CFA Institute Research Foundation. All rights reserved.  9

Performance Attribution

Holbrook (1977) linked this successive notional portfolio approach to

Fama’s decomposition by equating the first factor to the investor’s return from
systematic risk, the second factor to the manager’s return from systematic
risk, and the third factor to selectivity.

Foundations
The Brinson Model.  The next major development in attribution analysis
stems from a series of papers written in the 1980s that collectively describe
Brinson attribution.
Brinson, Hood, and Beebower’s (1986) “Determinants of Portfolio
Performance” is perhaps the most well-known paper on attribution. It pro-
vided the name for the most common type of equity attribution model used
today: the Brinson model. Despite being a relatively short paper, it made two
contributions:
1. The general framework used to decompose total portfolio returns.
Brinson et al. (1986) suggested a model for breaking down the arithmetic
return in excess of the benchmark (r – b) under the assumption of a standard
top-down investment decision process in which the portfolio manager seeks
to add value through both timing and security selection. Figure 2 illustrates
their framework for analysing portfolio returns.
Quadrant I represents the policy benchmark return (b ) for a period—that
is, the return on a hypothetical portfolio wherein the asset-class weights are
those in long-term investment policy (Wi ) and the returns are those of the
asset-class benchmarks (bi ).
i =n
Quadrant I = b = ∑Wi × bi . (9)
i =1

Quadrant II represents the return effects of policy and timing.

In timing, the portfolio manager seeks to add value by holding asset-
class (or sector) weights in the portfolio that are different from the asset-class
benchmark weights. An asset-class weight in the portfolio that is greater than
the benchmark weight for that asset class would be described as overweight
and a lesser weight would be described as underweight. Clearly, the portfolio
manager will aim to overweight well-performing asset classes (or sectors) and
underweight poorly performing asset classes.
Quadrant II minus Quadrant I, therefore, describes the added or sub-
tracted value from timing only.

10 © 2019 CFA Institute Research Foundation. All rights reserved.

 Performance Attribution

Figure 2.  The Brinson Quadrants

Selection

Actual Passive

Quadrant IV Quadrant II

Portfolio Return Policy & Timing

Actual

i=n i=n
r= Σ wi × ri bS = Σ wi × bi
Allocation

i=1 i=1

Quadrant III Quadrant I

Passive

Policy & Selection Policy

i=n i=n
rS = Σ Wi × ri
i=1
b= Σ Wi × bi
i=1

Excess returns due to:

Timing II – I
Security Selection III – I
Other IV – III – II + I
Total IV – I

Source: Brinson et al. (1986).

Quadrant III represents returns resulting from policy and security selec-
tion combined. In security selection, the portfolio manager seeks to add value
by selecting individual securities within the asset class or sector.
Quadrant III minus Quadrant I, therefore, describes the added or sub-
tracted value from selection only.
Quadrant IV represents the actual return of the managed portfolio.
Brinson et al. (1986) called the remaining term “other” or the “cross-
product” term. It is now more commonly considered to be “interaction.”
Quadrant IV minus Quadrant III minus Quadrant II plus Quadrant I
represents the added or subtracted value from interaction.
2. In a more controversial practice, the framework is used on a sample of
pension funds to demonstrate that most of the difference in returns among
pension funds is the result of policy decisions, not timing or selection. This
use of the framework is responsible for much of its fame.

© 2019 CFA Institute Research Foundation. All rights reserved.  11

Performance Attribution

Attribution Formulas. Brinson et al. (1986) may be the better

known of the “Brinson papers,” but the more important paper in terms of
the development of attribution is “Measuring Non-US Equity Portfolio
Performance” by Brinson and Fachler (1985). Again, it is a short paper, but
it documented for the first time the formulas required to calculate individual
attribution effects:
nn Allocation (Brinson and Fachler [1985] used the term “market selec-
tion”).  The contribution to excess return from asset allocation in the ith sector is

Ai = (wi − Wi ) × (bi − b ). (10)

nn Selection.  The contribution to excess return from stock selection in

sector i is

Si = Wi × (ri − bi ). (11)

nn Interaction (Brinson and Fachler used the term “cross-product”).  The

contribution to excess return from interaction in sector i is

I i = (wi − Wi ) × (ri − bi ). (12)

With a few modifications, these formulas provide the basic foundations

for most of the attribution analysis used today. The terms “asset allocation,”
“security selection,” and “interaction” are now universal and familiar to all. At
the time, however, allocation might have been described as “timing,” “market
selection,” or “weighting.” To establish when the term “allocation” was first
used is difficult, but Cocks (1972) was among the first. He defined it as “all
those decisions by which the portfolio manager allocates his funds among the
selected securities (or broad classes of securities)” (p. 3). Asset allocation spe-
cifically locates the decision process in what asset classes to hold and in what
weights, and this terminology is associated with Ibbotson and Sinquefield
(1982), although no one knows when the term was first used.
Use of the term “interaction” (or “other” or “cross-product”) is difficult to
establish, but certainly, by the early 1990s, “asset allocation,” “security selec-
tion,” and “interaction” had become the most commonly used terms.
The Brinson et al. (1986) paper did not provide a formula for individual
attribution effects, but the following variation of the allocation formula is
associated with that paper:

12 © 2019 CFA Institute Research Foundation. All rights reserved.

 Performance Attribution

The contribution to excess return from asset allocation in the ith sector is

Ai = (wi − Wi ) × bi . (13)

It seems unlikely that Brinson et al. (1986) intended this interpretation,

and the Brinson and Fachler (1985) variant for asset allocation is far more
common and better aligned with most investment decision processes. If the
Brinson et al. definition is used, the contribution from asset allocation will be
positive for any overweight sector in which the sector return is positive, but it
will be positive when using the Brinson–Fachler definition only if the sector
return exceeds the overall benchmark return.
Graphically, the Brinson et al. (1986) method is easier to illustrate for
each sector i than the Brinson and Fachler (1985) method, as shown in
Figure 3. The contribution to portfolio return from sector i is the total area
ri × wi, and the contribution from the benchmark alone is area bi × Wi. In this
example the total contribution from the portfolio in sector i is greater than
the contribution from the benchmark.
The attribution to excess return in sector i is the sum of the areas rep-
resenting selection [Wi × (ri − bi )], allocation [(wi − Wi ) × bi ], and interaction
[(wi − Wi ) × (ri − bi )]. In this particular example, ri > bi, bi > 0 and wi > Wi;
therefore, all three contributions are positive.
The Brinson and Fachler (1985) method illustrated in Figure 4 is a little
less intuitive. As stated in Bacon (2008), “We observe no change to the areas
representing selection and interaction but allocation is now described by the
area (wi − Wi ) × (bi − b )” (p. 125).

Figure 3.  Brinson, Hood, and Beebower (1986) Attribution

ri
Selection Interaction

Wi × (ri – bi) (wi – Wi) × (ri – bi)

bi

Benchmark Contribution Allocation

Wi × bi (wi – Wi) × bi

Wi wi

Source: Bacon (2008).

© 2019 CFA Institute Research Foundation. All rights reserved.  13

Performance Attribution

Figure 4.  Brinson and Fachler Attribution

ri
Selection Interaction

(ri – bi) × Wi (ri – bi) × (wi – Wi)

bi

Allocation

Wi × (bi – b) (wi – Wi) × (bi – b)

b

Wi × b (wi – Wi) × b

Wi wi

Source: Bacon (2008).

The traditional formulas described in Brinson and Fachler (1985) were

clearly in use well before 1985. The same formulas as those suggested by
Brinson–Fachler and associated with Brinson et al. (1986) can be found in
a formal discussion response to a paper by Kingston (1973) presented to the
Faculty of Actuaries on 14 October 1974, more than 10 years before the
Brinson–Fachler paper was published.
The formulas may well have been developed prior to the publication of the
Brinson and Fachler (1985), Brinson, Hood, and Beebower (1986), Brinson,
Singer, and Beebower (1991), and Singer (1996) papers, but collectively these
papers established attribution analysis as an important aid in the critical
review of active strategies and decision-making processes for asset managers,
asset consultants, and asset owners.
Interaction.  Interaction is a controversial effect. Most authors—for
example, Laker (2001), Spaulding (2003/2004), Campisi (2004b), and Bacon
(2008)—would agree that the interaction effect is a directly calculated term,
not a residual, but what does it mean? .  .  . It is certainly not intuitive. As
noted, Brinson and Fachler (1985) and Brinson, Hood, and Beebower
(1986), arguably the first definers of the interaction effect, rather unhelpfully
described this effect as “other” or a “cross-product”; they did not use the
term “interaction.” By the early 1990s, the term had become fairly universal

14 © 2019 CFA Institute Research Foundation. All rights reserved.

 Performance Attribution

in the literature, however, probably popularised by the software providers of

the time. The interaction effect is clearly a cross-product, the combination of
allocation and selection decisions.
Laker (2000) suggested “that when calculated properly, the interaction
term plays a valuable role in explaining a portfolio’s active performance, and
indeed that sometimes it is clearly more explanatory than the other attributes”
(p. 43). Spaulding (2003/2004) was more nuanced in his comments, despite
pointing out potentially misleading results (from mispriced securities or
investing in sectors not in the benchmark). On balance, Spaulding suggested,
it is best to calculate interaction, take the time to do some analysis, and then
decide where to assign it, or better yet, show the interaction as a separate
item and prepare to explain it. Campisi (2004b) and Bacon (2008) were
less supportive of using the interaction effect; they suggested that it is not
an active part of the investment decision process and, therefore, need not be
calculated. Like many asset managers in practice, they suggested including
interaction in the selection effect.
The decision to include interaction with stock selection is not arbitrary.
If you assume that asset allocation decisions are taken first in a top-down
investment process, then the attribution from stock selection can be derived
from Quadrant IV minus Quadrant II in Figure 2 and interaction need not
be calculated at all. The decision to include interaction with stock selection is,
therefore, the direct result of the investment decision process.
The contribution to stock selection in the ith sector is now

Si = wi × (ri − bi ), (14)

and Figure 4 can now be redrawn as shown in Figure 5 with selection repre-
sented by the area wi × (ri − bi ).

Multiperiod Analysis
The attribution models described so far work well for single periods with static
data. No obvious way exists, however, to combine attribution effects over time.
Because of the multiplicative (geometric) nature of returns, arithmetic excess
returns fail to add up over multiple periods. Carino (1999, p. 5) eloquently
explained the problem this way:
Over multiple periods, the natural way to combine returns is to compound
them. The compounded return R over t periods is

R = (1 + ri ) × (1 + r2 ) ×  × (1 + rt ) − 1. [15]

© 2019 CFA Institute Research Foundation. All rights reserved.  15

Performance Attribution

Figure 5.  Selection including Interaction

ri
Selection

wi × (ri – bi)
bi

Allocation
Wi × (bi – b)
(wi – Wi) × (bi – b)
b

Wi × b (wi – Wi) × b

Wi wi

Source: Bacon (2008).

And for the benchmark,

B = (1 + bi ) × (1 + b2 ) ×  × (1 + bt ) − 1. [16]

It is clearly unsatisfactory to simply add effects over time, because the sum
of return differences does not equal the difference between compounded
returns.
Singer (1996, p. 54) made the same point:
While it is tempting to simply add the individual periods to obtain multipe-
riod attribution results, it is wrong:

R − B ≠ (ri − bi ) + (r2 − b2 ) +  + (rt − bt ). [17]

Nor is it satisfactory to compound the single period return differences because

that result does not equal the difference in compounded returns either.
Laker (2002) made the point through this equation:

R − B ≠ (1 + r1 − b1 ) × (1 + r2 − b2 ) ×  × (1 + rt − bt ) − 1. (18)

16 © 2019 CFA Institute Research Foundation. All rights reserved.

 Performance Attribution

Linking and Smoothing.  Obviously, if multiperiod arithmetic excess

returns do not add up, then clearly arithmetic attribution that adds up for sin-
gle periods will not add up for multiple periods. Value added through active
management decisions compounds over time, a fact that should be accounted
for in multiperiod frameworks. Naturally, the end users of attribution analy-
ses expect their analyses to add up without residuals.
Solutions can be categorized into two types, smoothing algorithms, such as
those described by Carino (1999) and Menchero (2000), and linking algorithms
(sometimes described as “dollar attribution”), such as those in GRAP (1997),
Frongello (2002a, 2002b), and Bonafede, Foresti, and Matheos (2002).1
The authors of each paper suggested that their method was the most
appropriate to implement. Smoothing algorithms, in effect, take the result-
ing residual and methodically redistribute it across other factors. Carino
(1999) used logarithms, and Menchero (2000) redistributed the residual in
an optimised way that minimised the change in any single factor. GRAP
(1997), Frongello (2002a, 2002b), and Bonafede et al. (2002) compounded
attribution effects through time and achieved identical results, but cru-
cially, their methods are order dependent: If the time periods are reversed,
the attribution result will change. Frongello acknowledged the order
dependence but regarded it as a strength; others, including Laker (2002),
would describe it as a weakness.
Davies and Laker (2001) suggested an “exact” method for using the
Brinson model to calculate attributes over multiple periods at a portfolio level,
but not sector level. Because of the so-called exact nature of their model,
they suggested it would be helpful to evaluate the accuracy of other pro-
posed methods by comparison with their method at the total portfolio level.
Many articles have been written critiquing and defending existing methods
and offering refinements and new methods, including Banchik (2004/2005),
Carino (2002), David (2012/2013), Frongello (2002a, 2002b), Jiang and
Saenz (2014/2015), Kirievsky and Kirievsky (2000, 2004), Laker (2002,
2005), Menchero (2003), Mirabelli (2000/2001), Reztsov (2011/2012),
Spaulding (2002), and Wong (2003/2004, 2007). By and large, however, the
end user has proven to be unconcerned about these black box approaches,
provided that the multiperiod arithmetic attribution adds up without residu-
als. Frongello (2006) concluded that “although the mathematics can differ
greatly from one method to another, the story told by the resulting attribution
is the same regardless of the method” (p. 366).

1
Software providers have typically found solutions but often keep their methods more or less
proprietary.

© 2019 CFA Institute Research Foundation. All rights reserved.  17

Performance Attribution

Geometric Attribution.  Some would suggest that the problem with

multiperiod attribution analysis stems from an incorrect definition of excess
return itself. Bacon (2002) defined the arithmetic excess return (r − b ) as the
profit in excess of the notional fund expressed as a percentage of the ini-
tial value of the portfolio. That work defined the geometric excess return,
[(1 + r ) / (1 + b )] − 1, as the profit in excess of the notional fund expressed as
a percentage of the final value of the notional fund. In other words, the geo-
metric excess return measures the added value in terms of a percentage of the
value of the notional fund assets at the end of the period, not the start, which
is surely the main concern of any asset owner.
Bacon (2002) suggested that geometric excess returns are preferable
because they are compoundable (geometric excess returns compound over
multiple periods without residuals), they are convertible among currencies,
and they are proportionate. The preferred definition of excess return should
determine the choice of attribution methodology. Bain (1996), Burnie,
Knowles, and Teder (1998), and Bacon (2002) each presented a geometric
attribution model. Although developed independently and presented in dif-
ferent ways, the models are, in fact, identical, which is an indication of the
robustness of the approach. Geometric models built on the idea of successive
notional portfolios can be found in Holbrook (1977), Hymans and Mulligan
(1980), and Allen (1991). As shown in Figure 6, each successive portfolio

Figure 6.  Successive Notional Portfolios

Step III
Portfolio Return
i=n
r = Σ wi × ri
i=1

Step II
Semi-Notional
i=n
bS = Σ wi × bi
i=1

Step I
Benchmark Return
i=n
b = Σ Wi × bi
i=1

Source: Bacon (2008).

18 © 2019 CFA Institute Research Foundation. All rights reserved.

 Performance Attribution

represents an isolated decision in a top-down investment decision process.

Step I represents the strategic, policy, or benchmark return. The difference
between Step II and Step I represents the value added by active asset alloca-
tion, and the difference between Step III and Step II represents the value
added by security selection. More steps can be introduced for more complex,
multistep investment decision processes.
In the geometric attribution model, the difference between each step is
calculated geometrically:

(1 + r ) (1 + bS ) (1 + r )
−1= × − 1. (19)
(1 + b )  +
(1
 b) (1 + bs )

Allocation Selection

Because geometric excess returns compound through time and because

geometric allocation effects and geometric selection effects compound
together within each period, geometric allocation and geometric selection
must also compound through time. Bain (1996), Burnie et al. (1998), and
Bacon (2002) provided formulas for individual attribution effects as follows:
nn Allocation.  The contribution to geometric excess return from asset
allocation in the ith sector ( AiG ) is

 1 + bi 
AiG = (wi − Wi ) ×  − 1 . (20)
 1+ b 

nn Selection.  The contribution to geometric excess return from stock

selection in sector i (SiG ) is

(ri − bi )
SiG = wi × (21)
(1 + bS )

or

 1 + ri  (1 + bi )
SiG = wi ×  − 1 × . (22)
 1 + bi  (1 + bS )

In geometric attribution, interaction is typically included in the selection

effect, although a geometric interaction effect can be derived if desired (see
Weber 2018).

© 2019 CFA Institute Research Foundation. All rights reserved.  19

Performance Attribution

Burnie et al. (1998) described perhaps the most well-known geometric

methodology. Interestingly, they made the case for geometric attribution in
a single-period context, not necessarily a multiperiod one. They also drew a
distinction between top-down and bottom-up investment decision processes.
A bottom-up investment decision process, they noted, assumes that security
selection takes precedence over asset allocation or is performed first.
Weber (2018) would consider the Bain (1996), Burnie et al. (1998),
and Bacon (2002) models to be semi-geometric models. The reason is that
although total selection and allocation effects compound to the geometric
excess return, the effects within selection and allocation are still arithmeti-
cally calculated. Menchero (2001/2002) offered a fully geometric model in
which individual selection and allocation effects compound through time.

Multicurrency Attribution
In this section, we consider three works specifically directed to performance
attribution in currency management.
Allen (1991).  In an article entitled “Performance Attribution for Global
Equity Portfolios,” Allen took the Holbrook (1977) concept of successive
intermediate portfolios and applied it in a multicurrency environment. Allen
stated that the trick to quantifying the impact of each type of active manage-
ment decision involved calculating the return on a theoretical portfolio where
one type of decision had been neutralised. Comparison of the return on the
manager’s portfolio (where no decisions had been neutralised) with the return
on the portfolio where a class of management decisions had been neutralised
allowed the impact of those neutralised decisions to be measured.
When carried out correctly, this process is quite neat. No unexplained
residuals remain. Allen’s article provides a framework for quantifying the
marginal impact that both policy decisions and active management decisions
have on the performance of a global equity portfolio. At the heart of this
framework is the assumption that the total return on an actively managed
global equity portfolio is explained by seven types of decisions:
1. the choice, at the policy level, of the unhedged dollar-denominated index
against which the active equity manager is to be measured
2. security selection decisions by the active manager within each country
3. overweighting or underweighting of a country relative to the index based
on the manager’s expectation of the country’s equity returns
4. overweighting or underweighting of a country relative to the index based
on the manager’s expectation of the country’s currency returns

20 © 2019 CFA Institute Research Foundation. All rights reserved.

 Performance Attribution

5. the timing of purchases and sales of securities (which implicitly affect the
currency appreciation of the portfolio)
6. the choice, at the policy level, of the benchmark percentage of the portfo-
lio that is to be hedged passively against currency fluctuation
7. active decisions in the currency forward markets that cause the portfolio’s
return to deviate from that of the passively hedged benchmark
Allen’s article is noteworthy in that despite being based in the United
States where arithmetic attribution prevails, he favored a geometric approach
because the attribution factors are computed on a multiplicative basis, so they
can be “chain-linked” across any time period to generate a cumulative factor.
This method allows for evaluation of the long-term contribution of a selected
investment strategy to total return.
Ankrim and Hensel (1992, 1994).  In 1992, Ankrim and Hensel pub-
lished a Russell Research Commentary setting out a framework for multicur-
rency performance attribution. This article, building on work by Brinson and
Fachler (1985), separated out the impact of currency decisions—in particular,
it recognised the significance of differential interest rates in currency decision
making. Ankrim and Hensel proposed
a method of performance attribution that retains the simplicity and intui-
tive appeal of the Brinson and Fachler approach but breaks the returns from
currency into two components: one that recognizes the opportunity cost of
returns achievable in forward-currency markets and a second that measures
the return attributable to the currency being less than fully hedged, both
in the portfolio and in the benchmark with which the portfolio is com-
pared. (p. 2)
In the 1992 paper, Ankrim and Hensel identified three reasons that
including currency exposure in performance attribution causes problems:
First, and most obvious, it alters the returns ultimately received by the cli-
ent. In Brinson and Fachler [1985], all returns are denominated in the home
currency, but this convention makes it impossible to separate the contribu-
tion to returns made by the manager’s country and security selection deci-
sions from the impact of currency translation gains and losses.
Second, in practice, many portfolios have some portion (but not necessarily
all) of their currency exposure hedged away. This requires an approach to
performance attribution that allows the proportion of currency exposure
hedged to be variable.
Third, even if a portfolio is completely hedged, the client may choose to
adopt a benchmark that is less than fully hedged. (p. 2)

© 2019 CFA Institute Research Foundation. All rights reserved.  21

Performance Attribution

The authors defined two components of currency return: currency surprise

and the forward premium. Currency surprise is uncertain, but the forward
premium element is not because it depends on interest rate differentials at the
time of purchase:
Any approach to multicurrency attribution that lumps the forward premium
and currency surprise together will credit positive performance to manag-
ers who have large exposures in countries with positive forward premiums.
Such an approach would penalize managers who hold large positions in
countries with negative forward premiums. (p. 12)

Karnosky and Singer (1994). Arguably the single most important

paper published on attribution in the 1990s is Karnosky and Singer. This
paper not only demonstrates that managing multicurrency portfolios is
sub-optimal if currency is not managed independently but also provides
a framework for calculating attribution effects, while critically, taking into
account interest rate differentials. Karnosky and Singer resolved the issue
of market and currency compounding by using continuously compounded
returns in their model, admittedly this is only referred to in the small print,
and solved the forward premium concern by thinking in terms of a “return
premium” above local interest rates. Karnosky and Singer explained that the
term “return premium” is used, rather than the more familiar “risk premium,”
to make a subtle distinction in the underlying cash return. “Risk premium”
refers to the return of a risky asset in excess of the return of a riskless asset,
but in the authors’ analysis, the premiums depend on forward contracts with
long maturities, where risk is probably not the dominant factor; hence, the
term “return premium” is used.
Karnosky and Singer’s framework demonstrates that separate currency
and market strategies can be implemented within a global portfolio in which a
specific currency weight is different from the weight of the associated country
market. In effect, the authors applied the standard Brinson methodology
twice—once to global portfolios by using local-return premiums (rather
than simply local returns) to measure allocation and selection effects and,
separately, to cash returns in the base currency of the asset owner in order to
calculate independent currency effects, as shown in Figure 7.

22 © 2019 CFA Institute Research Foundation. All rights reserved.

 Performance Attribution

Figure 7.  Karnosky and Singer (1994) Multicurrency Attribution

Excess Return

Local-Return Premium Attribution Currency Attribution

Quadrant IV Quadrant II Quadrant IV Quadrant II

Quadrant III Quadrant I Quadrant III Quadrant I

Source: Bacon (2008).

Types of Attribution
Three types of attribution have been identified, and they are characterized by
the information used to calculate attribution effects, as follows:
1. Holdings-based attribution.
Holdings-based or position-based attribution is calculated by analysing
the underlying beginning-period holdings of the portfolio only. Typically,
holdings-based attribution is calculated from monthly, weekly, or daily
data, with the shorter time periods leading to greater accuracy.
2. Transaction-based attribution.
Transaction-based attribution is calculated by using both the holdings
of the portfolio and the transactions (purchases and sales) that occurred
during the evaluation period.
3. Returns-based attribution.
Returns-based (or factor) attribution uses the historical returns of the
total portfolio and/or individual securities to decompose the total return
into multiple factors. The multifactor model is a regression equation that
describes the way a particular security or portfolio reacts to industry sec-
tor, fundamental, macroeconomic, and other market factors.

© 2019 CFA Institute Research Foundation. All rights reserved.  23

Performance Attribution

Asset-Grouping Attribution.  The type of attribution favoured by the

asset manager or asset owner will vary according to the investment decision
process, investment objective, quality and availability of data, cost and time
considerations, complexity, and reporting objective. Holdings-based and
transaction-based attributions are mutually exclusive, and the choice is based
on (for holdings based) cost, quality of data, ease of use, and rapid implemen-
tation against (for transaction based) the advantages of reconciliation to the
client-reported return, completeness, and accuracy. A number of authors—
Giguere (2003), Bonafede and McCarthy (2003), Spaulding (2003a),
Menchero and Hu (2003), and Bacon (2008)—argued for one approach or
another. Spaulding, without an eye to existing software (and thus genuinely
independently), concluded that the question “which is better?” does not have
a simple answer. Too many variables are at work; in the end, the choice comes
down to user preference. In later research, Spaulding (2018) showed that the
residual implicit in holdings-based attribution can be extensive and is not nec-
essarily correlated with turnover. Both holdings-based and transaction-based
attribution are also described as “asset-grouping” or “variance type” attribu-
tion, reflecting their requirements for asset allocation data and variance from
benchmark data.
Attribution Based on Returns.  Returns-based attribution, in contrast
to the asset-grouping approach, might complement either holdings- or trans-
action-based attribution in that it might reveal unexpected biases and make
communication easier for asset managers using a factor-focused investment
decision process. Returns-based attribution is a direct descendant of Fama
(1972) decomposition, which is a single-factor model, via Sharpe (1988) and
Fama and French (1993)—both three-factor models that include a general
market factor together with a size and value factor—and Carhart (1997), a
four-factor model that introduced a momentum factor.
Amenc, Sfeir, and Martellini (2003) described four types of factor models:
1. Implicit factor model. In this model, some factor analysis (e.g., a principal
components analysis) is performed to statistically extract the factors from
the return’s time series. These factors may not be easily interpretable.
2. Explicit macro factor model. In this approach, macroeconomic variables
are used as factors.
3. Explicit micro factor model. In this approach, microeconomic attributes
are used as factors.
4. Explicit index factor model. In this approach, stock market indexes
(including style subindexes of the market) are used as factors.

24 © 2019 CFA Institute Research Foundation. All rights reserved.

 Performance Attribution

Within this framework, Amenc and Le Sourd (2003) described a wide

range of multifactor models in detail.
William F. Sharpe, who is associated with style attribution starting with
his 1988 paper (see Sharpe 1988, 1992), used a 12-factor model to determine
the passive return of a portfolio with the same style as the active asset man-
ager. The difference between the actual portfolio return and the passive return
with the same style would be a result of the manager’s selection ability. This
type of style analysis provides a natural method for constructing appropriate
style benchmarks. Sharpe (1988) stated that the usefulness of the asset-class
model depends on the asset classes chosen for implementation. It is desirable
that asset classes
•• be mutually exclusive,
•• be exhaustive, and
•• have returns that differ.
Each asset class should be proxied by a market capitalization–weighted
index of the securities in it, and no security should be included in more than
one asset class. Also, the asset classes should have low correlations with each
other (or different standard deviations if correlations are high, as is the case in
many fixed-income styles or asset subclasses).
Surz (1999/2000) suggested a holdings-based or transaction-based form
of style attribution based on the exclusive style characteristics of the hold-
ings in the portfolio and benchmark. Clarke, de Silva, and Thorley (2005)
proposed a regression-based attribution system linked to the parameters from
the “Fundamental Law of Active Management” (Grinold and Kahn 1995,
pp. 117–35). In a slight oversimplification, the fundamental law says that
alpha is a function of manager skill as well as breadth (the number of inde-
pendent bets made by the manager). The investigation by Clarke et al. (2005)
indicated that the “law can be used to estimate and interpret the regression-
based results without substantial error while showing the investor how much
of the active return is related to forecasting accuracy and how much is noise
related to constraints in constructing the portfolio” (p. 81).
Returns-based or regression-based attribution has the following
characteristics:
•• It does not necessarily require the composition of the portfolio.
•• It complements traditional holdings- or transaction-based attribution.
•• It is relatively easy and cost-effective to implement.

© 2019 CFA Institute Research Foundation. All rights reserved.  25

Performance Attribution

•• It is potentially most consistent with the investment decision process of

the asset manager.
•• It is easy to interpret.
However, this method does have limitations:
•• It is less accurate than other methods, particularly for concentrated strategies.
•• It relies on the appropriate definition of factors.

Risk-Adjusted Attribution
Campisi (2000) listed the characteristics of a good performance attribution
system as follows:
•• It is consistent with the investment process and the manager’s decision-
making process.
•• It uses a benchmark that reflects the manager’s strategic (long-term) asset
allocation.
•• It measures the effect of the manager’s tactical (short-term) allocation shifts.
•• It adjusts attribution of return for systematic risk(s).
Campisi (2000) suggested that many attribution systems fail at least one
of these criteria and some fail all of them. Often, this failure occurs because
returns are not adjusted for systematic risk and style (high vs. low beta, large
vs. small size, value vs. growth orientation, etc.). Simply applying the alloca-
tion/selection model often results in attribution results that are not neces-
sarily meaningful because they do not reflect the investment process or the
risks that drove return over the evaluation period. Risk-adjusted attribution
is the real Cinderella of attribution analysis: In practice, risk-adjusted attri-
bution is rarely performed. Ankrim (1992) proposed a form of risk-adjusted
attribution using beta, as did Bacon (2008) and Obeid (2005). Kophamel
(2003), thinking in terms of the volume of “performance cubes” rather than
the familiar “performance areas” of the Brinson model, proposed attribution
analysis in three dimensions. Spaulding (2016) suggested adjusting for total
risk rather than systematic risk—in effect, attributing M2 returns (described
in Modigliani 1997).2 Rather satisfactorily, Fisher and D’Alessandro (2019)

2
Modigliani risk-adjusted performance (also known as M 2, M2, the Modigliani–Modigliani
measure, or RAP) is a measure of the risk-adjusted returns of some investment portfolio. It
measures the returns of the portfolio adjusted for the risk of the portfolio relative to that of
some benchmark (e.g., the market).

26 © 2019 CFA Institute Research Foundation. All rights reserved.

 Performance Attribution

took attribution full circle and tied risk-adjusted attribution back to the origi-
nal Fama decomposition.
Van Breukelen (2000) suggested a form of risk-adjusted attribution for
fixed-income portfolios that takes into account the weighted duration “bets”
of portfolio managers. Van Breukelen described a top-down investment deci-
sion process in which overall duration is the first decision followed by a mar-
ket allocation decision—a combination of country weight and duration—and
finally, an issue selection decision within country. Currency effects are mea-
sured separately using a Karnosky and Singer (1994) type of approach. The
value added by each step of the decision process is measured by using a series
of reference or successive notional portfolios.

Fixed-Income Attribution
The simple Brinson model is widely regarded as being insufficient for all
but the simplest of fixed-income investment strategies. McLaren (2002)
explained that classical attribution models (e.g., Brinson’s model) are inap-
propriate for fixed-income portfolio analysis because the allocation decision
in equity models does not explicitly account for yield-curve positioning (dura-
tion) set by fixed-income managers. This omission is significant because the
duration decision, which has no direct equity counterpart, plays an impor-
tant role in the fixed-income decision-making process. Campisi (2000) also
explained why fixed income needs its own attribution model and pointed out
five critical differences between stocks and bonds:
1. Bonds are temporary lending agreements with a stated maturity, whereas
stocks are permanent investments.
2. Bonds promise a fixed return, with a strictly defined upside if held to
maturity; stocks promise uncertain returns and the unlimited upside (and
downside) associated with ownership.
3. Bonds are typically purchased by institutions, which often hold them
until maturity. This pattern results in a limited secondary market for
bonds. Stock investing takes place in an active secondary market. As a
result, bonds are illiquid compared with stocks.
4. Bond performance is driven by promised income and by changes in market
yields: When market yields rise above a bond’s stated yield, the price of that
bond falls. As a result, the relevant risk for bonds is sensitivity to changes in
yields. Stock performance is driven by the market’s economic sectors, and
the principal risks for stocks are (a) sensitivity to the overall market, or beta,
and (b) the individual fortunes of the company invested in.

© 2019 CFA Institute Research Foundation. All rights reserved.  27

Performance Attribution

5. Bonds are simply the promise of a stream of cash flows; thus, they are
relatively homogeneous in their pricing. For example, bonds with the
same maturity and default risk will generally sell at the same price.
Therefore, bonds reflect little selection effect. Stock prices, in contrast,
respond dramatically to company-specific conditions and thus reflect a
large selection effect.
Distinct approaches for fixed-income attribution emerged quite quickly
in the early development of attribution. The method described in Wagner and
Tito (1977) is derived directly from Fama decomposition, with beta replaced
by a systematic risk measure for bonds—namely, duration. In this approach,
fixed-income portfolio performance is driven by promised income and by the
effect that changes in yields have on bond prices.
No single, standardised way of defining fixed-income attribution exists.
The asset class has more securities than in equity investing, more types of
instruments, more variety in investment process, and more quantitative mod-
elling requirements. It also has various ways of describing change in the yield
curve and a broad array of investment mandates. Murira and Sierra (2006)
explained that each of the commercially available systems uses its own partic-
ular brand of return attribution. In fact, because of differences in the defini-
tion and computation of the factors, different systems often provide different
values for the same return factor.
Other authors have made various contributions to developments in
fixed-income attribution—developments often linked to specific invest-
ment decision processes or specific software capabilities. These authors
include Ramaswamy (2001), Giguere (2005), Colin (2005, 2014), Gillet and
Hommolie (2006), Colin, Cubilie, and Bardoux (2006), Silva, de Carvalho,
and Ornelas (2010), Simmons and Karadakov (2013/2014), and Dias (2017).
Figure 8 illustrates the attribution effects in a typical fixed-income attri-
bution model. Here, carry represents the return resulting from the passage of
time: coupon payments, accrued interest, and rolling down the yield curve (the
pull-to-par effect as the instrument approaches maturity).
In Figure 8, the yield-curve factor represents the changing shape of the
yield curve, parallel shift (the impact of a parallel move of the yield curve),
twist (a change in slope of the yield curve), and curvature (change in the
curvature of the yield curve). Often, twist and curvature are combined and
described as nonparallel changes in the yield curve or curve reshape.
Spread is the return that comes from the widening or narrowing of credit
spreads.

28 © 2019 CFA Institute Research Foundation. All rights reserved.

 Performance Attribution

Figure 8.  Fixed-Income Attribution Effects

Excess Return

Carry Yield Curve Spread Selection Other

Coupon Parallel Shift Convexity Price Effect

Roll Down Twist Optionality Trading Effect

Curve Reshape

Curvature Residual Currency

Source: Redrawn from a diagram in Dai and Elliot (2014/2015).

Selection effects include convexity, any optionality in the payment of future

coupons (important with mortgage-backed securities because mortgages can
be prepaid), and a residual. Normally quite small, residuals are classified by
Dias (2017) as caused by the model or caused by data errors and systematic or
random effects.
In the other category, the price effect captures the difference in price
sources between the bonds in the portfolio and bonds in the benchmark. This
is a nonmanagement effect that is a particular problem for fixed income. The
trading effect measures the impact of intraday trading. Currency captures the
impact of independent currency management, typically by using a Karnosky
and Singer (1994) type of approach.
Fixed-income attribution models can be characterised as either top-down
successive portfolio methods (e.g., Van Breukelen 2000) or bottom-up (yield-
curve decomposition) methods. Bottom-up methods calculate the impact of
yield-curve movements and credit spreads on individual bonds in both the
portfolio and the benchmark and aggregate results. Top-down models are
perhaps easier to implement and more effective in terms of communicating to
asset owners and other stakeholders.

© 2019 CFA Institute Research Foundation. All rights reserved.  29

Performance Attribution

Gillet and Hommolie (2006) described the bottom-up approach as a suc-

cessive spread methodology that can be split into two phases. The first phase
consists of decomposing the bond prices into its various characteristics (pas-
sage of time, evolution of yield curves, variation of spreads). This decomposi-
tion is applied simultaneously to the constituents of the portfolio and to the
constituents of the benchmark. The second phase consists only of regrouping
the separated elements according to the investment decision process (p. 65).
Bottom-up models are more difficult and expensive to implement but more
appropriate to the detailed investment decision processes of the front office.
McLaren (2002) offered a geometric fixed-income attribution model,
which is not common. Most fixed-income attribution models are arithme-
tic in nature, although for the most part, they can be adapted for geometric
analysis. The reason may be that excess returns are typically smaller than in
equity investing and the underlying benchmark returns less volatile, mak-
ing the difference between geometric and arithmetic excess return smaller in
magnitude.

Other Attribution Issues

Other issues in attribution pertain to derivatives, transaction costs, frequency
of the analysis, and application to money-weighted returns.
Derivatives.  The Brinson model is an incredibly robust methodology.
Stannard (1997), Menchero (2002/2003), Bacon (2008), Fischer and Wermers
(2013), and Bacon, Thompson, and van der Westhuizen (2018)—all described,
using appropriate notional assets and notional returns, how the standard
Brinson model can be adapted to handle derivative instruments, including
index futures, options (of various types), and swaps and short positions.
Transaction Costs.  Laker (2001) pointed out that transaction costs—
explicit costs (such as brokerage and taxes) and implicit costs (such as the
bid–offer spread, market impact, and opportunity cost)—are not dealt with
explicitly in most performance attribution methods but are normally included
within the stock selection effect. In a bottom-up, security-level attribution
approach, Laker suggested using the standard Brinson selection formulas
at the security level to calculate the impact of transaction costs (explicit and
implicit) and intraday timing effects. He suggested that any return difference
between an individual security’s daily return and its return in the index must
be the result of transaction costs and timing within the day.
Frequency of Analysis.  DiBartolomeo (2003) challenged the assump-
tion that aggregated daily attribution analysis is more appropriate than

30 © 2019 CFA Institute Research Foundation. All rights reserved.

 Performance Attribution

monthly aggregated analysis and suggested that applying typical attribution

methods to daily data draws on assumptions that are highly biased and unre-
liable. Beyond simply observing the attributed returns, he believed, you must
be able to judge their statistical significance. He suggested that recommend-
ing daily observation rests on four fallacies:
1. Daily performance is economically meaningful. Not necessarily—for
example, information is lost by adjusting for the timing of cash flows in
time-weighted returns.
2. Daily returns are themselves accurate. Not necessarily—many securities
are traded on multiple exchanges or electronic markets at different times,
so exactly which closing price is used to compute the daily returns can
have a sizable impact. Also, coordinating the timing of corporate actions
such as dividends, spin-offs, and mergers with index providers is difficult.
3. Observations are assumed to be independent. Investment portfolios
do not start as a fresh set of management decisions every trading day.
Thus the rationale of an independent “start fresh” assumption is weak.
Substantial path dependencies are present in portfolio management in
practice, which reduces any justification for the assumption of indepen-
dent periodic observations of manager skill.
4. Parametric distributional assumptions are accurate. Most of the aca-
demic literature illustrates that the imperfections of our assumptions with
respect to quarterly or monthly return data are small, whereas for daily
data, we must consistently reject the correctness of the assumptions. In
general, daily security returns have fat-tailed distributions, with large
return events occurring far more often than would be consistent with
a normal distribution. This issue is less of a problem with quarterly or
monthly returns.
DiBartolomeo (2003) concluded that daily performance attribution com-
pounds to a more accurate measure of the time-weighted return than attribu-
tion using less frequent data—but at a tremendous cost in our ability to judge
the statistical significance of the results. Most of the imperfections of daily
attributions result in upwardly biased confidence levels in our estimates of
manager skill, leading to persistent Type 1 errors (believing an estimate is
statistically significant when it is not).
Darling and MacDougall (2002/2003) certainly questioned the ben-
efits of more frequent and more detailed performance measurement. They
suggested that the cost of daily stock-level attribution is difficult to justify
(although given recent developments in technology their views may have

© 2019 CFA Institute Research Foundation. All rights reserved.  31

Performance Attribution

changed). They suggested that monthly modified Dietz calculations might

actually be near the optimum balance and that a simple alternative might be
running actual profit and loss for each position.3
Frongello (2006), in contrast, made the case for daily attribution analysis.
He explained that in the real world, risk factor mismatches may occur unin-
tentionally—caused by neglect, accidents, or simply changes in the risk pro-
file of securities. Performance attribution should seek to explain performance
resulting from all risk factor mismatches. Daily attribution is the best method
for monitoring the status and consequence of risk factor mismatches (p. 367).
Zangari and Bayraktar (2005/2006) suggested that daily attribution, if
done methodically, should produce results that are superior to monthly attri-
bution. They pointed out that regardless of the methodology used, what is
ultimately critical is the process governing attribution. The five core compo-
nents of what they considered to be a good return attribution process are as
follows:
1. Data management and quality control. This task is the most important.
Positions, prices, corporate actions, and so on, must be accurately cap-
tured, processed, and incorporated.
2. Infrastructure. Without proper infrastructure (efficient application of
hardware and databases), the chance of errors and delays increases.
3. Analytics. Algorithms perform the attribution calculations.
4. Methods. For example; daily buy-and-hold, monthly buy-and-hold, and
transaction-based attribution.
5. Reports. The way in which information is presented is important, so
timely, accurate, and easy-to-understand attribution reports are necessary.
Zangari and Bayraktar (2005/2006) concluded that process plays a criti-
cal part in determining the accuracy of attribution results. Return attribution
should be based on a process that monitors key input data and evaluates attri-
bution results on a daily basis.
Attribution of Money-Weighted Return.  For purposes of comparative
analysis, most attribution analysis is naturally undertaken on time-weighted
rates of return (TWR). Attribution analysis of money-weighted rates of
return (MWR) is less common. Yet, attribution analysis of MWR is still
undertaken in the form of the modified Dietz return calculation method.
3
The modified Dietz method is a mathematical technique to evaluate a portfolio’s return
based on a weighted calculation of its cash flow. The method takes into account the timing of
cash flows and assumes a constant rate of return over a specified period of time.

32 © 2019 CFA Institute Research Foundation. All rights reserved.

 Performance Attribution

It was, in fact, recommended by Menchero and Hu (2003) in the context of

transaction-based attribution and by Darling and MacDougall (2002/2003).
O’Shea and Jeet (2018) suggested single-period attribution of modified Dietz
returns for private equity. Spaulding and Campisi (2007) made the case for
money-weighted attribution. They suggested that because cash flow move-
ments are a factor that defines the investment process, these cash flows must
be part of the performance measurement and evaluation system instead of
being neutralised (as occurs in the time-weighted calculation).
The modified Dietz method is a first-order approximation of the internal
rate of return (IRR) and, according to Bacon (2007a), an intermediate step
in the evolution from calculating IRR (which some might describe as a true
MWR) to the true TWR. The IRR is a standard return measure for illiquid
asset classes—such as private equity, real estate, and infrastructure—and is
common at the total-asset owner level, where comparative analysis is less rel-
evant. The modified Dietz method is easy to disaggregate and more straight-
forward to attribute than the IRR. Nevertheless, Illmer and Marty (2003)
and Illmer (2009) offered a solution to the attribution of the true MWR in
terms of both return and profit and loss.
D’Alessandro (2016) suggested analysts should end the debate about
which method of return to use, TWR or MWR, and require the presentation
of both with an attribution between the two.

Conclusion: The Evolution of Attribution Methodologies

The evolution of attribution analysis is clear, easy to trace, and illustrated
in Figure 9, which maps, not necessarily in chronological order, the major
developments in attribution analysis discussed in this review.
Comparative performance measurement starts with the BAI (1968)
study, but attribution analysis itself starts with Fama (1972). From that point,
a divergence occurs between asset-grouping and variance-type attribution,
with the UK SIA (1972) introducing the concept of selection, allocation, and
notional portfolios and return-based attribution developed a little later by
Sharpe (1992), Fama and French (1993), Carhart (1997), and many others.
Fixed-income attribution naturally veers in a different direction via Wagner
and Tito (1977), who replaced beta with duration in the Fama decomposition,
and Lord (1997), who responded to the investment decision processes of bond
managers, which are so different from those of equity managers.
Very early, a split occurred between, on the one hand, the dominant
arithmetic Brinson and Fachler (1985) and Brinson et al. model (1986) and,
on the other hand, the total level geometric approach of Holbrook (1977).
Smoothing and linking algorithms were soon needed in order for multiperiod

© 2019 CFA Institute Research Foundation. All rights reserved.  33

Performance Attribution

Figure 9.  The Evolution of Attribution Methodologies

Fama
1972

Fixed Income Asset Grouping Return Based

Wagner and Tito UK Society of Investment Sharpe

1977 Analysts 1972 1992

Arithmetic Geometric

Lord Brinson, Hood, and Holbrook Fama–French

1997 Beebower 1986 1977 1993

Multicurrency

Brinson and Burnie, Knowles, Allen Carhart

Fachler 1985 and Teder 1998 1991 1997

Multicurrency

Van Breukelen Smoothing and Ankrim and Factor

2000 Linking Hensel 1992 Attribution

Karnosky and
Singer 1994

Yield-Curve Multicurrency
Decomposition Geometric

Source: Author.

arithmetic models to add up, which were, for the most part, designed and
implemented by software firms to meet client requirements. Algorithms
started to be published several years after their initial use in Carino (1999),
soon followed by GRAP (1997), Menchero (2000), Frongello (2002a, 2002b),
and Bonafede et al. (2002). Allen (1991) is a direct link from Holbrook but
addressed multicurrency issues. Burnie et al. (1998), together with Bain
(1996) and Bacon (2002), developed geometric selection and allocation calcu-
lations that were missing from Holbrook (1977) and from Allen. Ankrim and

34 © 2019 CFA Institute Research Foundation. All rights reserved.

 Performance Attribution

Hensel (1992) and, with more success, Karnosky and Singer (1994) adapted
the basic Brinson model for multicurrency attribution, including, importantly,
taking into account interest rate differentials. Van Breukelen (2000) adapted
the Brinson model for successive portfolio fixed-income attribution.
Figure 9 firmly cements Brinson and Fachler (1985) at the centre of
developments in attribution analysis. In more recent years, we have seen
many variants of yield-curve decomposition models, which reflects the diver-
sity of fixed-income investment decision processes. Because of low uptake,
risk-adjusted attribution is not illustrated here but could be represented by a
link between return-based methods and the Brinson model.

Bibliography
Allen, Gregory C. 1991. “Performance Attribution for Global Equity
Portfolios.” Journal of Portfolio Management 18 (1): 59–65.
This article provides a framework for quantifying the marginal impact that
both policy decisions and active management decisions have on the per-
formance of a global equity portfolio. The article is noteworthy because,
influenced by currency effects, it represents an early US-based example
of geometric or multiplicative attribution; it emphasises the value of “no
unexplained residuals” and attribution factors that “chain-link.”
Amenc, Noel, and Veronique Le Sourd. 2003. Portfolio Theory and Performance
Analysis. John Wiley & Sons.
This wide-ranging academic work evaluating the quality of the portfolio
management process connects each of the major categories of techniques
and practices to the unifying and seminal conceptual developments of
modern portfolio theory. Of particular relevance to attribution analysis
is Chapter 6 on multifactor models and their application to performance
measurement, Chapter 7 on evaluating the investment management pro-
cess and decomposing performance, and Chapter 8 on fixed-income secu-
rity investment.
Amenc, Noel, Daphne Sfeir, and Lionel Martellini. 2003. “An Integrated
Framework for Style Analysis and Performance Measurement.” Journal of
Performance Measurement 7 (4): 35–41.
This article explains that risk-adjusted performance evaluation consists of
using a factor model to compare the overall performance of a manager with
the “normal” performance the manager would get from being exposed to
rewarded sources of risk. The article suggests the use of two important

© 2019 CFA Institute Research Foundation. All rights reserved.  35

Performance Attribution

tools: (1) return-based style analysis, which allows the formation of peer
groups of comparable managers, and (2) a customised multifactor model
to estimate superior risk-adjusted performance (alpha) within each group.
Ankrim, Ernest M. 1992. “Risk-Adjusted Performance Attribution.”
Financial Analysts Journal 48 (2): 75–82.
The author suggests that classical attribution analysis does not consider
risk and posits that in general, such methods will tend to overreward risky
managers and underreward conservative ones. The author proposes an
attribution approach that adjusts for the systematic risk of each sector.
Ankrim, Ernest M., and Chris R. Hensel. 1992. “Multi-Currency
Performance Attribution.” Russell Research Commentary (November).
Ankrim, Ernest M., and Chris R. Hensel. 1994. “Multicurrency Performance
Attribution.” Financial Analysts Journal 50 (2): 29–35.
These articles, both originally published in a Russell Research
Commentary, expand the work of Brinson and Fachler (1985), separate out
the impact of currency decisions, and, in particular, recognise the signifi-
cance of differential interest rates in currency decision making.
Bacon, Carl R. 2002. “Excess Returns—Arithmetic or Geometric?” Journal
of Performance Measurement 6 (3): 23–31.
This article sets out to define arithmetic and geometric excess returns. The
author explains that the choice of an arithmetic or geometric attribution
method is really a choice between using arithmetic or geometric excess
return, and he expresses a preference for geometric excess returns based
on compoundability, convertibility, and proportionality. Finally, the author
presents a geometric attribution model that, although developed indepen-
dently, is actually identical to that of Burnie, Knowles, and Teder (1998).
Bacon, Carl R. 2007a. “Money-Weighted versus Time-Weighted
Attribution.” Journal of Performance Measurement 11 (Performance Attribution
Supplement): 21–27.
The author illustrates and explains the evolution of return calculations from
internal rates of return to true time-weighted rates of return. In summary,
the author suggests that time-weighted forms of attribution will remain
dominant because of the following:
•• The time-weighted rate of return is most appropriate for comparisons
of managers.

36 © 2019 CFA Institute Research Foundation. All rights reserved.

 Performance Attribution

•• Internal rates of return assume a constant rate of return difficult to

attribute.
•• Investment decisions are independent of the total amount invested.
•• Time-weighted attribution can measure the manager’s timing
decisions.
•• Money-weighted attribution misallocates estimates of manager skill
between stock selection and asset allocation.
Bacon, Carl R. 2007b. “Multi-Currency Attribution—Part 1: The Real
Nature of Multi-Currency.” Journal of Performance Measurement 12 (1): 22–29.
The author lists the issues of genuinely multicurrency investment decision
processes:
•• compounding effects of market and currency returns
•• interest rate differentials
•• unrealized gains/losses in forward currency contracts
•• securities denominated in currencies other than the currency of eco-
nomic exposure
Building on the work of Allen (1991), the author moves on to describe
a naive form of currency attribution (naive in that it ignores currency
differentials).
Bacon, Carl R. 2007/2008. “Multi-Currency Attribution—Part 2: Factoring
in Interest Rate Differentials.” Journal of Performance Measurement 12 (2): 8–19.
The author introduces the concepts of Karnosky and Singer (1994) into a
geometric framework that takes full account of interest rate differentials.
Bacon, Carl R. 2008. Practical Portfolio Performance Measurement and
Attribution. John Wiley & Sons.
This general book on performance analysis includes specific chapters on
performance attribution, multicurrency attribution, fixed-income attri-
bution, multiperiod attribution, and performance measurement for
derivatives.
Bacon, Carl R. 2013. Practical Risk‐Adjusted Performance Measurement. John
Wiley & Sons.
This book documents with extensive references a large number of discrete
ex-post risk measures and explains how they are related.

© 2019 CFA Institute Research Foundation. All rights reserved.  37

Performance Attribution

Bacon, Carl R., Ian Thompson, and Pierre van der Westhuizen. 2018.
“Abnormal Returns.” Journal of Performance Measurement 20 (3): 43–49.
The authors demonstrate the robustness of the Brinson–Fachler (1985)
model for short positions.
BAI. 1968. Measuring the Investment Performance of Pension Funds for the
Purpose of Inter-Fund Comparison. Bank Administration Institute.
The purpose of the BAI study, the starting point for performance analysis,
was to develop methods of comparing the performance of pension funds in
order to differentiate between the abilities of their respective managers. The
report recommended, among other things, the use of market value, not book
value; total return; time-weighted rates of return; and taking risk into account.
Bain, William G. 1996. Investment Performance Measurement. Woodhead
Publishing.
This book is, for the most part, concerned with the measurement and anal-
ysis of pension fund investment portfolios. Most notable is its description
of the WM Company’s multiplicative (or geometric) attribution methodol-
ogy (pp. 65–74).
Banchik, Sean. 2004/2005. “Pure and Inter-Period Interaction Effects in
Multi-Period Attribution.” Journal of Performance Measurement 9 (2): 53–63.
The author concludes that there is no single correct way to link single-
period arithmetic effects to produce multiperiod effects. The appropriate
methodology is generally one of personal preference. The most relevant
question for the reader is not which method is correct but whether the reader
understands where these effects are included in the chosen methodology.
Bonafede, Julia K., Steven J. Foresti, and Peter Matheos. 2002. “A Multi-
Period Linking Algorithm That Has Stood the Test of Time.” Journal of
Performance Measurement 7 (1): 15–26.
The authors introduce a linking algorithm for multiperiod arithmetic attri-
bution. The method was in use from the early 1980s, but this description
was not published until 2002.
Bonafede, Julia K., and Mary Cait McCarthy. 2003. “Transaction-Based
vs. Holdings-Based Attribution: The Devil Is in the Definitions.” Journal of
Performance Measurement 8 (1): 42–51.
The authors suggest that because full benchmark and portfolio valuation is
not realistically available at each transaction time, true transaction-based

38 © 2019 CFA Institute Research Foundation. All rights reserved.

 Performance Attribution

performance attribution is not possible. Only buy-and-hold returns are

consistent with benchmarking needs; therefore, performance attribution
for portfolio management should use holdings-based return analysis.
The authors propose an additional term called “trading” to explain the
difference.
Brinson, Gary P., and Nimrod Fachler. 1985. “Measuring Non-US Equity
Portfolio Performance.” Journal of Portfolio Management 11 (3): 73–76.
This hugely important paper documented, for the first time, selection and
allocation formulas at the sector level. The paper helped lend “Brinson” to
the name of the most commonly used attribution model.
Brinson, Gary P., L. Randolph Hood, Jr., and Gilbert L. Beebower. 1986.
“Determinants of Portfolio Performance.” Financial Analysts Journal 42 (4):
39–44.
This extremely well-known paper is important for two reasons. First, it
provided the general framework for the decomposition of portfolio return
into selection, allocation, and interaction effects (although this last piece
is described as “other” in the paper). Second, and more controversially, the
paper made the observation that policy decisions make by far the biggest
contribution to the differences in returns among pension funds.
Brinson, Gary P., Brian D. Singer, and Gilbert L. Beebower. 1991.
“Determinants of Portfolio Performance II: An Update.” Financial Analysts
Journal 47 (3): 40–48.
This paper updated and confirmed the results first published in Brinson,
Hood, and Beebower (1986).
Burnie, J. Stephen, James A. Knowles, and Toomas J. Teder. 1998.
“Arithmetic and Geometric Attribution.” Journal of Performance Measurement
3 (1): 59–68.
Most attribution models are arithmetically formulated and measure the
manager’s “value added”—or the difference between a fund’s return and a
passive benchmark return. The authors of this paper contended that such
models are prone to error and may produce inconclusive results. This paper
illustrates the benefits of using geometric attribution, which defines the
“management effect” as the ratio of the fund return factor to the bench-
mark return factor.
This paper is of fundamental importance in addressing the relative advan-
tage of geometric over arithmetic attribution, multiperiod attribution, and

© 2019 CFA Institute Research Foundation. All rights reserved.  39

Performance Attribution

the ambiguity of the interaction or cross-product effects and defining top-

down and bottom-up attribution.
Campisi, Stephen. 2000. “Primer on Fixed Income Performance Attribution.”
Journal of Performance Measurement 4 (4): 14–25.
The author explains that the drivers of fixed-income performance are fun-
damentally different from those of equity portfolios. Bond portfolio per-
formance is driven by promised income and by the effect that changes in
yields have on bond prices. The author proposes a simple attribution model
that provides a straightforward method for identifying the systematic risk
and stylistic factors of the asset manager’s strategy.
Campisi, Stephen. 2004a. “The Case for Money-Weighted Performance
Attribution.” Journal of Performance Measurement 8 (3): 31–41.
The proposal in this paper is that although the case is irrefutable for accept-
ing the time-weighted return as the preferred method for measuring the
return of an asset manager, an equally compelling case can be made for
accepting the money-weighted return as the appropriate method for evalu-
ating the source of that active return; that is, the money-weighted return is
the correct method for performance attribution.
Campisi, Stephen. 2004b. “Debunking the Interaction Myth.” Journal of
Performance Measurement 8 (4): 63–70.
The author refutes the idea of “interaction” and demonstrates that interac-
tion arises from either a poor attribution model or a poor implementation
of a sound attribution model. When a sound attribution model is imple-
mented correctly, the interaction effect is eliminated. Despite the appearance
of mathematical rigor in the calculation of interaction, investors and clients
understand that interaction is not a legitimate part of the investment process.
Campisi, Stephen, and David Spaulding. 2007. Readings in Fixed Income
Performance Attribution. TSG Publishing.
This book is a useful compendium of many of the articles included in this
review.
Canty, Mike. 2017/2018. “Evolving Performance Attribution to Support
Exploratory Excess Return Decomposition.” Journal of Performance
Measurement 22 (2): 27–38.
The author proposes three ways to improve arithmetic performance attri-
bution. First, the total weight associated with the stock selection effect

40 © 2019 CFA Institute Research Foundation. All rights reserved.

 Performance Attribution

should exclude the asset allocation effect. Second, the assessment of an

asset allocation decision should consider the actual returns connected
to that decision. Third, excess return decomposition should exclude the
impact of overlapping issue-level positions.
Carhart, Mark M. 1997. “On Persistence in Mutual Fund Performance.”
Journal of Finance 52 (1): 57–82.
In this highly influential paper, the author proposes an extension to the
Fama–French (1993) three-factor model by adding a momentum factor.
Carino, David R. 1999. “Combining Attribution Effects over Time.” Journal
of Performance Measurement 3 (4): 5–14.
This paper discloses Frank Russell Company’s method of combining
attribution effects over time and displaying results in an additive man-
ner. The paper is important in that it is the first time a software firm pub-
lished its method for linking multiple-period attribution results without
residuals, thus encouraging other firms to publish their in-house linking
methodologies.
Carino, David R. 2002. “Refinements in Multi-Period Attribution.” Journal
of Performance Measurement 7 (1): 45–53.
The author offers a derivation of the dollar family of linking algorithms
(see Frongello 2002a, 2002b; GRAP 1997; and Bonafede, Foresti, and
Matheos 2002) and discusses further refinements whilst cautioning the
performance analyst to balance complexity against practicality.
Clarke, Roger, Harindra de Silva, and Steven Thorley. 2005. “Performance
Attribution and the Fundamental Law.” Financial Analysts Journal 61 (5): 70–83.
The authors propose a regression-based attribution system linked to the
parameters from the “Fundamental Law of Active Management” (Grinold
and Kahn 1995). Their investigation indicates that the law can be used
to estimate and interpret the regression-based results without substantial
error while showing the investor how much of the active return is related
to forecasting accuracy and how much is noise related to constraints in
constructing the portfolio.
Cocks, Graham. 1972. “An Objective Approach to the Analysis of Portfolio
Performance.” Investment Analyst 34: 3–7.
The author argues that the primary purpose of portfolio analysis should be
to provide the portfolio manager with information to help improve future

© 2019 CFA Institute Research Foundation. All rights reserved.  41

Performance Attribution

performance. The objective of performance analysis should be to measure

the various elements of performance that reflect the decisions involved in
the investment process. The author then suggests that investment decisions
may be classified in two ways: selection and allocation. This paper is one of
the first to use the term “allocation.”
Colin, Andrew M. 2005. Fixed Income Attribution. John Wiley & Sons.
This book, one of the few devoted to fixed-income attribution, explains the
theory and practice of fixed-income attribution in detail. The author also
shows how to reveal the effects of multiple investment decisions in fixed-
income portfolios, including yield return, term structure effects, credit and
liquidity effects, and others. The book contains both theoretical and practi-
cal information about fixed-income attribution, including the mathematics
of attribution, yield-curve modelling, practical limitations, benchmarks,
and presentation tools.
Colin, Andrew M. 2014. Attribution in Finance. Flametree Technologies.
This comprehensive book devoted to the subject of performance attribution
is very much focused on all aspects of fixed-income attribution but includes
equity attribution, currency attribution, and smoothing algorithms.
Colin, Andrew M., Mathieu Cubilie, and Frederic Bardoux. 2006. “A New
Approach to the Decomposition of Yield Curve Movements for Fixed Income
Attribution.” Journal of Performance Measurement 10 (4): 18–28.
The authors discuss the most common ways to describe yield-curve
movements: shift, twist, and curvature. Shift is a parallel movement of
the curve across all maturities. Twist is a steepening or flattening of the
curve, and curvature occurs when a curve becomes more or less humped.
The authors suggest an approach to decomposing curve changes that cir-
cumvents the difficulties inherent in principal components analysis and
model fitting.
Dai, Tianci, and Mark Elliot. 2014/2015. “Fixed Income Attribution with
Carry Effect.” Journal of Performance Measurement 19 (2): 7–18.
These authors propose a way to capture the true carry return. Typical fixed-
income return decompositions attempt to include the carry return, but the
return is usually (inadequately) approximated by yield. The authors argue
that using an accurate calculation better isolates the effects of yield-curve
shifts, shape, and position; spread changes; and other factors.

42 © 2019 CFA Institute Research Foundation. All rights reserved.

 Performance Attribution

D’Alessandro, Joe. 2016. “Using Brinson Attribution to Explain Differences

between Time-Weighted (TWR) and Money-Weighted (IRR) Returns.”
Journal of Performance Measurement 20 (4): 11–21.
The author recommends ending the debate about which method of return
to use—time-weighted or money-weighted returns—and requiring the
presentation of both with a comparison of, or attribution between, the two.
Darling, Robert, and Alastair MacDougall. 2002/2003. “Using Performance
Statistics: Have Measurers Lost the Plot?” Journal of Performance Measurement
7 (2): 22–32.
The authors question the benefits of carrying out performance measure-
ment that is more frequent than monthly or in more detail. They suggest
that the cost of daily stock-level attribution is difficult to justify and argue
that monthly modified Dietz calculations may actually be near the optimal
balance. A simple alternative, they note, might be to run actual profit-and-
loss data for each position.
David, Mark R. 2012/2013. “A Case for Arithmetic Attribution.” Journal of
Performance Measurement 7 (2): 26–38.
The author questions the advantages of geometric attribution, propor-
tionality, convertibility, and compoundability and suggests an arithmetic
money-based compounded notional approach based on Davies and Laker
(2001).
Davies, Owen, and Damien Laker. 2001. “Multiple-Period Performance
Attribution Using the Brinson Model.” Journal of Performance Measurement
6 (1): 12–22.
The authors agree that multiperiod arithmetic attribution is difficult but
suggest an exact method for calculating attributes over multiple periods in
the Brinson model at the portfolio, not the sector, level. They further sug-
gest that because of the exact nature of their model, evaluating the accu-
racy of other proposed methods would be helpful.
Davis, Ben, and Jose Menchero. 2010/2011. “Beyond Brinson: Establishing
the Link between Sector and Factor Models.” Journal of Performance
Measurement 15 (2): 8–20.
The authors show that the classic Brinson model is a special kind of factor
(return-based) attribution model. They also demonstrate how this factor
model can be augmented with other factors, such as styles, to gain insight
that would not be possible using the classic Brinson framework.

© 2019 CFA Institute Research Foundation. All rights reserved.  43

Performance Attribution

Dias, Joao S. 2017. “Residuals on Duration-Based Fixed Income Attribution.”

Journal of Performance Measurement 21 (4): 8–20.
The author usefully classifies residuals in fixed-income attribution that are
the result of the model or data errors (including various types of data error)
and considers whether these errors are systematic or random. The author
also asks the important question, What is an acceptable residual?
DiBartolomeo, Dan. 2003. “Just Because We Can Doesn’t Mean We Should.”
Journal of Performance Measurement 7 (3): 30–36.
This paper presents four fallacies about daily observation and concludes
that daily performance attribution compounds to a more accurate measure
of the time-weighted return but at a tremendous cost in the analyst’s ability
to judge the statistical significance of results. Most of the imperfections of
daily attributions result in upwardly biased confidence levels in estimates
of manager skill, leading to persistent Type 1 errors.
Dietz, Peter O. 1966. Pension Funds: Measuring Investment Performance.
Graduate School of Business of Columbia University and Free Press.
This book fired the starting gun for comparative performance measure-
ment and established the dominance of time-weighted rate of return as the
measure. Performance analysis is referenced but considered to be beyond
the scope of the book.
Eadie, Dugald. 1973. “A Practical Approach to the Measurement and
Analysis of Investment Performance.” Investment Analyst 37: 12–18.
The author offers an early practical demonstration of attribution analysis
using the notional fund concept.
Fama, Eugene F. 1972. “Components of Investment Performance.” Journal of
Finance 28 (3): 551–67.
This truly seminal paper was the first to suggest decomposing the returns
of a portfolio in the manner the author proposes. Fama suggests finer
breakdowns of performance than given by the Sharpe, Treynor, or Jensen.
For example, methods are presented for distinguishing, on the one hand,
the part of an observed return that results from ability to pick the best
securities of a given level of risk (“selectivity”) from, on the other hand,
the part that is a result of predictions of general market price movements
(“timing”). The author also suggests methods for measuring the effects of
forgone diversification when an investment manager decides to concentrate
the portfolio’s holdings in what the manager thinks are a few “winners.”

44 © 2019 CFA Institute Research Foundation. All rights reserved.

 Performance Attribution

Fama, Eugene F., and Kenneth R. French. 1993. “Common Risk Factors in
Stock and Bond Returns.” Journal of Financial Economics 33: 3–56.
The authors propose that three common factors can empirically explain the
cross-relationships between stock returns: an overall market factor and fac-
tors related to size and value. This paper sets forth the widely used Fama–
French three-factor model.
Fama, Eugene F., and Kenneth R. French. 2015. “A Five-Factor Asset Pricing
Model.” Journal of Financial Economics 116 (1): 1–22.
The authors propose that a five-factor model is better than the three-factor
model.
Feibel, B.J. 2003. Investment Performance Measurement. John Wiley & Sons.
This wide-ranging book on investment performance measurement is valu-
able for its in-depth spreadsheet examples of attribution analysis. Part IV
of five parts describes performance attribution.
Fischer, Bernd, and Russell Wermers. 2013. Performance Evaluation and
Attribution of Security Portfolios. Elsevier.
A substantial work on performance evaluation, this book includes detailed
chapters on equity attribution, fixed-income attribution, analysis of multi-
asset-class portfolios and hedge funds, and attribution analysis with
derivatives.
Fisher, Jeffery D., and Joseph D’Alessandro. 2019. “Risk-Adjusted
Performance Analysis of Real Estate Portfolios.” Working paper. https://
w w w.ncreif.org/globalassets/public-site/research/ncreif-insights/risk-
adjusted-attribution-analysis-of-real-estate-portfolios-7-28-19-final---copy-
for-ncreif-website.pdf.
This article proposes a risk-adjusted performance attribution analysis that
integrates risk measures with the Brinson models of attribution. The ben-
efit is that it allows analysts to decompose the excess portfolio return into
components of risk allocation, selection, and net selectivity that are addi-
tive and consistent with financial theory.
Frongello, Andrew S.B. 2002a. “Attribution Linking: Proofed and Clarified.”
Journal of Performance Measurement 7 (1): 54–67.
The author provides the mathematical proofs for attribution linking while
illustrating the importance of order dependence. He also critiques the
multiperiod Brinson methodology suggested by Laker (2002).

© 2019 CFA Institute Research Foundation. All rights reserved.  45

Performance Attribution

Frongello, Andrew S.B. 2002b. “Linking Single Period Attribution Results.”

Journal of Performance Measurement 6 (3): 10–22.
This article proposes a linking algorithm that allows the summing of
single-period attribution to arrive at cumulative results that explain the
exact difference in cumulative returns. In addition to the judgement criteria
introduced by Carino (1999), the author adds the criteria of sincerity,
intuitive, and order dependence.
Frongello, Andrew S.B. 2006. “Linking of Attribution Results.” In Portfolio
Analysis, edited by Timothy P. Ryan, 343–68. Risk Books.
The author reviews popular linking methodologies and provides descrip-
tions and examples of small differences. His conclusion is that although
the mathematics may differ greatly from one method to another, the story
told by the resulting attribution is the same regardless of the method.
Giguere, Claude. 2003. “Transaction-Based vs Holdings-Based Attribution:
A Perspective.” Journal of Performance Measurement 8 (1): 24–26.
The author defines transaction-based and holdings-based attribution and
expresses a strong preference for the transaction-based approach.
Giguere, Claude. 2005. “Thinking through Fixed Income Attribution—
Reflections from a Group of French Practitioners.” Journal of Performance
Measurement 9 (4): 46–65.
The author provides reflections on a group of practitioners whose objectives
were to review the theoretical and methodological aspects of fixed-income
attribution and to define best practices. In the context of fixed income, this
group agreed to the following:
•• Excess return is generally low; therefore, methodologies must be accu-
rate with small residuals.
•• The attribution results must match the investment decision process.
•• The price differential between the portfolio and the benchmark is not
an effect attributable to the manager.
•• The calculation frequency should be as high as possible to avoid distortion.
The group described two types of attribution:
1. Yield-curve decomposition, which breaks down performance into
passage of time, yield-curve movements (however described), and spread
variations. This analysis is mostly intended for portfolio managers.

46 © 2019 CFA Institute Research Foundation. All rights reserved.

 Performance Attribution

2. Exposure decomposition, a top-down approach based on intermediate

reference portfolios. It can be better understood by noninvestment
professionals than yield-curve decomposition and is useful when
explaining the source of added value.
Gillet, Philippe, and Bernard Hommolie. 2006. “Fixed Income Attribution:
A Combined Methodology.” Journal of Performance Measurement 10 (4): 64–78.
The authors describe two existing forms of fixed-income performance
attribution—“successive portfolios” and “successive spreads”—and suggest
a combined model.
GRAP. 1997. “Synthèse des modèles d’attribution de performance.” Paris:
Groupe de Recherche en Attribution de Performance (March).
This early publication of a multiperiod arithmetic linking algorithm was
created by a French working group of performance measurement experts.
Grinold, Richard C., and Ronald N. Kahn. 1995. Active Portfolio Management.
Irwin Professional Publishing.
This encyclopaedic work on quantitative equity investing sets forth, among
many other concepts, the “Fundamental Law of Active Management” the
observation that the information ratio of any trading strategy is propor-
tional to the square root of the number of independent bets made per year.
Heatter, Craig, Charles Gabriel, and Yi Wang. 2004. “A Four-Factor
Performance Attribution Model for Equity Portfolios.” Journal of Performance
Measurement 9 (1): 51–59.
The authors discuss Carhart’s (1997) four-factor model in the context of
performance attribution. They are able to identify and accurately quantify
a portfolio’s risk exposures, sources of excess return, and sources of added
value.
Hensel, Chris R., D. Don Ezra, and John H. Ilkiw. 1991. “The Importance of
the Asset Allocation Decision.” Financial Analysts Journal 47 (4): 65–72.
Building on the work of Brinson, Hood, and Beebower (1986), the authors
suggest that if US T-bills are the appropriate naive alternative for US
investors, then asset allocation away from T-bills is, as commonly thought,
the single decision with the greatest impact on an investor’s return. But if
a diversified mix is the alternative, then the impact of departing from this
alternative may be no greater than the impact of other decisions, including
security selection.

© 2019 CFA Institute Research Foundation. All rights reserved.  47

Performance Attribution

Holbrook, J.P. 1977. “Investment Performance of Pension Funds.” Journal of

the Institute of Actuaries 104 (1): 15–91.
This important paper pulls together the work of the BAI (1968), Fama (1972),
and the Society of Investment Analysts (SIA 1972). The article establishes
the importance of investment objectives, the nature of liabilities, the need for
investment policies, and the responsibilities of trustees. Crucially, the author
decomposes the decision-making process into policy decisions, strategy deci-
sions, and selection. Holbrook suggests the use of notional funds for each
step of the decision process and establishes a geometric relationship for the
value added (or subtracted) at each step of the process.
Hsu, Jason C., Vitali Kalesnik, and Brett W. Myers. 2010. “Performance
Attribution: Measuring Dynamic Allocation Skill.” Financial Analysts Journal
66 (6): 17–26.
The authors propose an extension of the Brinson model that subdivides
asset allocation into static and dynamic components.
Hymans, Clifford, and John Mulligan. 1980. The Measurement of Portfolio
Performance. Kluwer Publishing.
This extremely useful early book on performance measurement includes, in
particular, a useful history of the subject to 1980, together with appropriate
references. Chapter 18 on performance analysis brings Holbrook’s work to
a wider audience beyond UK actuaries, and Appendix 6 provides a worked
example of attribution.
Ibbotson, Roger G., and Rex A. Sinquefield. 1982. Stocks, Bonds, Bills, and
Inflation: The Past and the Future. Charlottesville, VA: CFA Institute Research
Foundation.
The authors document, from a starting date of 1926, the monthly and
annual returns on US stocks, bonds, T-bills, and what they call “inflation” (a
hypothetical asset returning the rate of change in the US CPI). They calculate
historical (achieved) risk premiums based on differences between the average
returns of these assets and make probabilistic projections of future returns.
Illmer, Stefan J. 2009. “Decomposing the Money-Weighted Rate of Return—
an Update.” Journal of Performance Measurement 14 (1): 22–29.
This paper illustrates two methodologies that help analysts understand
where historical return (expressed as an internal rate of return) and
historical profit and loss are coming from. The author suggests that money-
weighted attribution has a lot of advantages because its results are in line

48 © 2019 CFA Institute Research Foundation. All rights reserved.

 Performance Attribution

with profit-and-loss attribution and it correctly separates the timing effect

between the asset allocation and stock-picking effects.
Illmer, Stefan J., and Wolfgang Marty. 2003. “Decomposing the Money-
Weighted Rate of Return.” Journal of Performance Measurement 7 (4): 42–50.
This article addresses return attribution from a client’s point of view. It
illustrates the decomposition of the money-weighted rate of return and
shows its relationship to the time-weighted rate of return.
Jiang, Yindeng, and Joseph Saenz. 2014/2015. “The Associative Property of
Attribution Linking.” Journal of Performance Measurement 19 (2): 19–24.
The authors suggest that the “associative” property is necessary for attri-
bution linking. Among the popular smoothing/linking methods, they
show that Carino (1999) and Frongello (2002a) are associative, whereas
Menchero (2000) and various ad hoc methods are not. They also suggest
that Frongello and Carino produce similar results.
Karnosky, Denis S., and Brian D. Singer. 1994. Global Asset Management and
Performance Attribution. Charlottesville, VA: Research Foundation of the
Institute of Chartered Financial Analysts.
The most important work written on performance attribution in the 1990s,
this book develops an analytical framework for evaluating global asset
markets and uses that framework to construct a performance attribution
system that isolates the effects of market allocation, currency management,
and security selection on global portfolios.
This book is important for demonstrating that a multinational portfolio
is suboptimally managed if currency is not managed independently and
for demonstrating the importance of interest rate differentials between
currencies.
Kingston, T.D. 1973. “Measuring Investment Performance [with discussion].”
In Transactions of the Faculty of Actuaries 34: 241–98. Cambridge University
Press.
This work is another early paper on the whole subject of measuring invest-
ment performance. Section 3 deals particularly with performance attri-
bution. Methods are suggested for comparing actual rates of return with
those of indexes, notional portfolios, and other funds. Of most interest is
the formal record of the discussion after the paper’s presentation. Note the
response from Mr Stewart (pp. 268–74), which documents the now familiar
Brinson and Fachler (1985) method (Method 2) and Brinson et al. (1986)

© 2019 CFA Institute Research Foundation. All rights reserved.  49

Performance Attribution

method (Method 1) formulas at least 10 years before the Brinson papers

were published.
Kirievsky, Leonid, and Anatoly Kirievsky. 2000. “Attribution Analysis:
Combining Attribution Effects over Time Made Easy.” Journal of Performance
Measurement 4 (4): 49–59.
The authors propose a smoothing algorithm that consists of “separating
pure effects from changes in each factor and the remainder (cross-product)”.
Kirievsky, Leonid, and Anatoly Kirievsky. 2004. “Attribution Analysis: Old
and New.” Journal of Performance Measurement 9 (1): 60–74.
In this article, the authors describe in detail the problem of attribution
analysis and provide a precise method of solving it. They analyse the short-
comings of other multiperiod linking and smoothing algorithms and the
geometric model and critique the original Brinson et al. (1986) paper.
Based on an example of historical Australian funds data, they suggest that
security selection is the main determinant of return.
Kophamel, Andrew. 2003. “Risk-Adjusted Performance Attribution: A New
Paradigm for Performance Analysis.” Journal of Performance Measurement
7 (4): 51–62.
This article proposes a three-dimensional approach to link risk and return
via the intuitive construct of volume and the “performance cube” rather
than using the familiar two-dimensional areas of the Brinson model.
Laker, Damian. 2000. “What Is This Thing Called ‘Interaction’?” Journal of
Performance Measurement 5 (1): 43–57.
In this paper, the author suggests that the interaction term has a bad
reputation. He states that portfolio managers do not distinguish between
interaction and error terms but arbitrarily add them to other terms. As a
result, they find interaction difficult to explain to clients.
The author argues that the interaction term is sometimes more explanatory
than the other attributes and that attempts to avoid confusion by “getting
rid of ” the interaction term in a Brinson model are, in fact, likely to create
more confusion.
Laker, Damian. 2001. “Incorporating Transaction Cost Measurement into
Performance Attribution.” Journal of Performance Measurement 5 (4): 13–24.
This paper points out that transaction costs (explicit costs, such as brokerage
and taxes, and implicit costs, such as the bid–offer spread, market impact,

50 © 2019 CFA Institute Research Foundation. All rights reserved.

 Performance Attribution

and opportunity cost) are not dealt with explicitly in most performance
attribution methods and are normally included in the stock selection effect.
The author suggests a stock-level, bottom-up style of attribution approach
that identifies and separates the impact of intraday timing and transaction
costs.
Laker, Damian. 2002. “A View from Down-Under.” Journal of Performance
Measurement 6 (4): 5–13.
The author suggests an exact method for multiperiod arithmetic attribu-
tion at a total level that can be used to evaluate the accuracy of various
linking and smoothing methods, including naive compounding.
Laker, Damian. 2003. “Perspectives on Transaction-Based Attribution.”
Journal of Performance Measurement 8 (1): 10–23.
The author defines transaction-based attribution, holdings-based attribu-
tion, and transaction-based attribution that identifies transaction costs,
and he provides worked examples of each. The paper explains that the
simplicity, reduced number of calculations, and reduced data burden of
holdings-based attribution come at the cost of accuracy.
Laker, Damian. 2005. “Toward Consensus on Multiple-Period Arithmetic
Attribution.” Journal of Performance Measurement 9 (3): 26–37.
The author defends criticism of his “exact” method for calculating multipe-
riod attributes at the total-fund level.
Lord, Timothy J. 1997. “The Attribution of Portfolio and Index Returns in
Fixed Income.” Journal of Performance Measurement 2 (1): 45–57.
This paper focuses on performance attribution as it applies to fixed-income
portfolios and indexes. More clearly described as bottom-up contribution
analysis of the portfolio compared with its benchmark, this paper
emphasizes the requirement for the attribution to be consistent with the
investment strategy of the portfolio manager. Income and price returns are
calculated separately. Price return is then attributed according to duration,
yield-curve distribution, sector allocation, and issue selection. Shift and
twist are considered components of duration return; and sector and specific
issuer effects are considered components of spread return. Included also
are a calendar return (the price return resulting solely from the passage of
time) and a residual return (the difference between the actual return and
the model’s estimate of return).

© 2019 CFA Institute Research Foundation. All rights reserved.  51

Performance Attribution

McLaren, Andrew. 2001. “A Geometric Methodology for Performance

Attribution.” Journal of Performance Measurement 5 (4): 45–57.
The author proposes a framework for absolute and relative multicurrency
geometric attribution.
McLaren, Andrew. 2002. “A Framework for Multiple Currency Fixed
Income Attribution.” Journal of Performance Measurement 6 (4): 59–79.
This article presents a framework for analysing the investment decisions
of a multicurrency fixed-income portfolio. The significant difference from
the classical model is the addition of a reference yield-curve portfolio to
measure the return generated from interest rate exposure decisions.
Menchero, Jose G. 2000. “An Optimized Approach to Linking Attribution
Effects over Time.” Journal of Performance Measurement 5 (1): 36–42.
In this paper, the author presents an intuitive and robust algorithm for
linking single-period attribution effects over time. The method consists of
first deducing the correct scaling between the single-period excess returns
and the linked excess return and then optimally distributing the residual
among the different time periods.
Menchero, Jose G. 2000/2001. “A Fully Geometric Approach to Performance
Attribution.” Journal of Performance Measurement 5 (2): 22–30.
In this paper, the author examines the close relationship between geo-
metric and arithmetic calculations and their effects and uses the analogy
to arrive at an intuitive mathematical form for attribution analysis. Also,
while addressing the issues involved in aggregating the attribution effects
across sectors, the author shows how they can be linked across time.
Menchero, Jose G. 2002/2003. “Performance Attribution with Short
Positions.” Journal of Performance Measurement 7 (2): 39–50.
In this paper, the author shows how classical attribution analysis can be
adapted to include short positions. The keys to the methodology are
•• explicit inclusion of the cash positions and
•• independent treatment of the long and short sides of each sector.
Menchero, Jose G. 2003. “Linking Differences Do Matter.” Journal of
Performance Measurement 7 (3): 47–50.
In this short article, the author refutes the claim that differences between
smoothing and linking algorithms are immaterial.

52 © 2019 CFA Institute Research Foundation. All rights reserved.

 Performance Attribution

Menchero, Jose G., and Ben Davis. 2009. “Multi-Currency Performance

Attribution.” Journal of Performance Measurement 14 (1): 45–55.
The authors present a comprehensive model for attributing the performance
of multicurrency portfolios based on the Karnosky and Singer (1994)
framework but with certain refinements. In particular, they explicitly
account for the cross-product effect derived from repatriating local profits
back to the base currency of the investor.
Menchero, Jose G., and Junmin Hu. 2003. “Errors in Transaction-Based
Performance Attribution.” Journal of Performance Measurement 8 (1): 57–69.
The authors identify three types of transaction-based attribution: true
time-weighted methods that use a buy-and-hold assumption and two types
of transaction-based attribution based on modified Dietz return calcula-
tions. The two latter types are as follows: Type 1, which accounts for exter-
nal and internal cash flows, and Type 2, which accounts for internal cash
flows only. The authors acknowledge that the holdings-based approach will
lead to differences between the actual return and the computed return, but
they suggest that the magnitude should be small and typically depend on
volatility and portfolio turnover. They suggest that this difference should
be attributed to trading effects. The authors show that both Type 1 and
Type 2 approaches will lead to attribution errors, so they ultimately prefer
the holdings-based approach.
Mirabelli, Andre. 2000/2001. “The Structure and Visualization of
Performance Attribution.” Journal of Performance Measurement 5 (2): 55–80.
This paper proposes an alternative to multiperiod arithmetic performance
attribution methods.
Modigliani, Leah. 1997. “Risk-Adjusted Performance, Part 1: The Time
for Risk Measurement Is Now.” Morgan Stanley’s Investment Perspectives
(February).
The author introduces the risk-adjusted return measure M2. The measure was
developed by Franco Modigliani and his granddaughter Leah Modigliani.
Murira, Bernard, and Hector Sierra. 2006. “Fixed Income Attribution: A
Unified Framework—Part 1.” Journal of Performance Measurement 11 (1): 23–35.
This article provides a framework for understanding fixed-income
performance attribution. The article discusses two complementary types
of attribution model: (1) a factor model that captures the impact on
the portfolio return of changes in economic variables, such as the term

© 2019 CFA Institute Research Foundation. All rights reserved.  53

Performance Attribution

structure of interest rates, volatility, and prepayment speed, and (2) an

allocation model that assesses the returns in excess of the benchmark
returns of the allocation strategy in various sectors.
Murira, Bernard, and Hector Sierra. 2006/2007. “Fixed Income Attribution: A
Unified Framework—Part 2.” Journal of Performance Measurement 11 (2): 8–21.
A continuation of Part 1 (Murira and Sierra 2006/2007), this part dis-
cusses some practical issues related to the attribution process—in particu-
lar, the fact that duration is not a perfect measure. Other issues addressed
are key-rate duration versus partial duration, selecting the appropriate
point to use on the yield curve, dealing with transactions, linking multipe-
riod returns, and the accuracy of security prices.
Obeid, Alexander. 2005. “Reformulating Ankrim’s Risk-Adjusted
Performance Attribution.” Journal of Performance Measurement 9 (3): 8–25.
This paper builds on the work of Ankrim (1992) but presents a more ade-
quate risk adjustment for evaluating market timing.
O’Shea, Luis, and Vishv Jeet. 2018. “Single-Period Brinson-Style Performance
Attribution for Private Capital.” Working paper, Burgiss Applied Research
(June).
In this working paper, the authors develop a methodology for Brinson-
style attribution of private capital returns.
Ramaswamy, Srichander. 2001. “Fixed Income Portfolio Management: Risk
Modeling, Portfolio Construction and Performance Attribution.” Journal of
Performance Measurement 5 (4): 58–70.
The author outlines a methodology for risk modelling, portfolio
construction, and performance attribution of fixed-income portfolios.
Reztsov, Andrei. 2011/2012. “Geometric and Arithmetic Approaches to
Attribution Linking Are Equivalent.” Journal of Performance Measurement
16 (2): 46–56.
The author offers an algorithm that is not order dependent and has a linear
nonsmoothing construction.
Sharpe, William F. 1988. “Determining a Fund’s Effective Asset Mix.”
Investment Management Review (December): 59–69.
This landmark paper was the first to present return-based style analysis
(attribution identifying exposure to growth, value, large-cap, and small-cap

54 © 2019 CFA Institute Research Foundation. All rights reserved.

 Performance Attribution

factors as well as the market factor). Sharpe describes the returns as leaving
“tracks in the sand” that can be interpreted, by using regression analysis, as
style exposures.
Sharpe, William F. 1992. “Asset Allocation: Management Style and
Performance Measurement.” Journal of Portfolio Management 18 (2): 7–19.
The author, building on Sharpe (1988), suggests a 12-asset-class model
that determines the passive return of a notional portfolio having the same
style as the asset manager. The manager’s selection return is the difference
between the actual return and the passive return in the same style.
SIA. 1972. The Measurement of Portfolio Performance for Pension Funds.
Society of Investment Analysts.
Toward the end of 1970, a working group was set up by the UK Society of
Investment Analysts to make recommendations on the subject of portfolio
performance measurement for pension funds. The resulting report spe-
cifically puts forward methods of calculating portfolio performance, thus
making possible a general comparison of funds.
The approach recommended has two aims: (1) the calculation of a rate of
return for comparing funds and (2) the decomposition of the performance
of a fund into two components—selection of stocks and selection of sec-
tors. This report is, therefore, really the starting point of performance attri-
bution because it both defines stock selection and sector allocation (called
“sector selection” in the report) and provides a worked example.
Silva, Antonio F.A., Pablo J.C. de Carvalho, and Jose R.H. Ornelas. 2010.
“A Performance Attribution Methodology for Fixed Income Portfolios.”
Portfolio and Risk Management for Central Banks and Sovereign Wealth
Funds (February).
This paper presents a simple fixed-income attribution model suitable for
the investment process of central banks. The approach does not require a
specific methodology for yield-curve fitting.
Simmons, Peter, and Anton Karadakov. 2013/2014. “A Simplified Fixed
Income Performance Attribution Model.” Journal of Performance Measurement
18 (2): 44–55.
The authors create an attribution model that follows a top-down approach
but is built from the bottom up. The attribution groups the output into five
sections: duration management, spread management, issue selection, yield
management, and currency management.

© 2019 CFA Institute Research Foundation. All rights reserved.  55

Performance Attribution

Singer, Brian D. 1996. “Evaluation of Portfolio Performance: Attribution

Analysis.” Journal of Performance Measurement 1 (2): 45–55.
The author explains that attribution analysis is an important aid in the
critical review of active strategies and in the decision-making process for
a portfolio. This article provides a basic framework for gleaning accurate
information regarding active portfolio management decisions.
Spaulding, David. 2002. “Is Linking Attribution Effects as Hard as It
Looks?” Journal of Performance Measurement 6 (3): 32–39.
The author compares a number of smoothing and linking algorithms.
Spaulding, David. 2003a. “Holdings- vs. Transaction-Based Attribution: An
Overview.” Journal of Performance Measurement 8 (1): 52–56.
The author concludes that the question “which is better?” has no simple
answer. Too many variables are at work. Some firms’ trading volumes are
such that a simple buy-and-hold model will not provide even the remotest
degree of precision. But how much more must a firm spend? And for many
firms, any incremental benefit can be justified by the added cost.
Spaulding, David. 2003b. Investment Performance Attribution. McGraw-Hill.
The book is the first devoted solely to the subject of performance attribution.
Spaulding, David. 2003/2004. “Demystifying the Interaction Effect.” Journal
of Performance Measurement 8 (2): 49–54.
The author explains that the interaction effect is the product of two deci-
sions: selection and allocation. Despite pointing out potentially misleading
results caused by calculating interaction (for example, mispriced securities
and/or investing in sectors not in the benchmark), the author determines
that on balance, it is better to calculate interaction, take the time to do
some analysis, and then decide where to assign it—better yet would be to
show the interaction as a separate item and prepare to explain it.
Spaulding, David. 2016. “Performance Attribution Risk-Adjusted: Why It
Makes Sense and How to Do It.” Journal of Performance Measurement 20 (4):
42–56.
This paper briefly reviews various risk-adjusted attribution methodologies
and suggests a simple methodology based on M 2 (the Modigliani and
Modigliani risk-adjusted performance measure).

56 © 2019 CFA Institute Research Foundation. All rights reserved.

 Performance Attribution

Spaulding, David. 2018. “Transaction- vs. Holdings-Based Attribution: The

Differences Are Not So Clear but Quite Important.” Journal of Performance
Measurement 22 (3): 49–76.
In this detailed study, the author finds that the residuals caused by
holdings-based analysis can be extensive and are not necessarily correlated
with turnover, as might be expected.
Spaulding, David, and Stephen Campisi. 2007. “A Case for Money-Weighted
Attribution.” Journal of Performance Measurement 11 (Performance Attribution
Supplement: 8–20.
The authors make the case for money-weighted attribution. Since cash
flow movements are a factor that defines the investment process, cash flows
must be part of the performance measurement and evaluation process and
should not be neutralised, as they are in a time-weighted calculation. The
authors point out that transaction-based attribution methodologies that
use modified Dietz calculations are, in fact, money-weighted attributions
over a single period.
Stannard, John C. 1997. “Measuring Investment Returns of Portfolios
Containing Derivatives: Part II—Performance Attribution.” Journal of
Performance Measurement 1 (3): 5–9.
This paper demonstrates that a derivatives position can be represented as a
series of (normally, two or more) notional asset exposures representing physi-
cal securities and any implied borrowing used to finance the security position.
Standard performance attribution techniques can be used on notional assets
to represent the derivatives in terms of the underlying investment decisions.
Surz, Ronald J. 1999/2000. “Attribution with Style.” Journal of Performance
Measurement 4 (2): 50–64.
This paper makes the case that manager style is a powerful determinant
of performance. Style-based attribution produces much more reliable
indicators than sector-based methods because most investment managers
manage to a style rather than to some mix of economic sectors. Using sector-
based attribution analysis can produce significantly misleading inferences.
Van Breukelen, Gerard. 2000. “Fixed Income Attribution.” Journal of
Performance Measurement 4 (4): 61–68.
This article discusses performance attribution for fixed-income portfolios.
The starting point is a top-down decision structure, which means that the
portfolio’s overall duration is decided first, followed by the distribution

© 2019 CFA Institute Research Foundation. All rights reserved.  57

Performance Attribution

over markets. The risk factor for markets is a combination of weight

and duration. After the market allocation, individual issue selection is
measured. Currency allocation is independent and measured separately.
Wagner, Wayne H., and Dennis A. Tito. 1977. “Definitive New Measures of
Bond Performance and Risk.” Pension World (May): 17–26.
This article is an early attempt to analyse fixed-income portfolios in a man-
ner similar to Fama (1972) decomposition but replacing beta with duration.
Weber, Arno. 2018. “Geometric Attribution and the Interaction Effect.”
Journal of Performance Measurement 22 (4): 6–19.
The author presents a variant of the Burnie, Knowles, and Teder (1998)
model that includes interaction effects.
Wong, Celia. 2003/2004. “Attribution—Arithmetic or Geometric? The Best
of Both Worlds.” Journal of Performance Measurement 8 (2): 10–18.
The author lists the pros and cons of the arithmetic and geometric methods
and concludes the geometric method is clearly superior in almost every way.
The author states that the arithmetic method produces a distortion when
applied to multiple-period attribution because returns naturally compound
geometrically, not arithmetically. Nothing can be done to improve the
arithmetic method to overcome this basic fact. What is not so well known
is that the arithmetic method produces a distortion even for a single period
when multiple currencies are involved. In conclusion, the author suggests a
method that is computed geometrically and presented arithmetically.
Wong, Celia. 2007. “Comparing Arithmetic and Geometric Attribution.” In
Advanced Portfolio Attribution Analysis, edited by Carl Bacon, 135–59. Risk
Books.
The author provides a new arithmetic formula that eliminates the
drawbacks of multiperiod arithmetic attribution with no need for linking
algorithms. She also presents an easy way of converting between arithmetic
and geometric attribution so that expressing attribution results is just a
matter of preference.
Zangari, Peter J., and Mehmet Bayraktar. 2005/2006. “Which Is Better: Daily
or Monthly Attribution?” Journal of Performance Measurement 10 (2): 8–19.
This article explores important issues in the debate about the merits of daily
versus monthly attribution and concludes that daily attribution, if done
methodically, should produce results that are superior to monthly attribution.

58 © 2019 CFA Institute Research Foundation. All rights reserved.

 The CFA Institute
Research Foundation
Board of Trustees
2019–2020
Chair JT Grier, CFA* Aaron Low, CFA
Ted Aronson, CFA Virginia Retirement LUMIQ
AJO System
Mauro Miranda, CFA
Heather Brilliant, CFA Joanne Hill Panda Investimentos
Diamond Hill CBOE Vest Financial AAI Ltda.

Margaret Franklin, CFA Roger Ibbotson, CFA* Lotta Moberg, PhD, CFA
CFA Institute Yale School of Management
William Blair
Bill Fung, PhD Joachim Klement, CFA
Sophie Palmer, CFA
Aventura, FL Independent
Jarislowsky Fraser
Daniel Gamba, CFA Vikram Kuriyan, PhD, CFA
Dave Uduanu, CFA
BlackRock GWA and Indian School
Sigma Pensions Ltd
of Business

*Emeritus

Officers and Directors

Executive Director Secretary
Bud Haslett, CFA Jessica Lawson
CFA Institute CFA Institute
Gary P. Brinson Director of Research Treasurer
Laurence B. Siegel Kim Maynard
Blue Moon Communications CFA Institute
Associate Research Director
Luis Garcia-Feijóo, CFA, CIPM
Coral Gables, Florida

Research Foundation Review Board

William J. Bernstein Paul D. Kaplan, CFA Krishna Ramaswamy
Efficient Frontier Morningstar, Inc. University of
Advisors Pennsylvania
Robert E. Kiernan III
Elroy Dimson Advanced Portfolio Andrew Rudd
London Business School Management Advisor Software, Inc.

Stephen Figlewski Andrew W. Lo Stephen Sexauer

New York University Massachusetts Institute Allianz Global Investors
of Technology Solutions
William N. Goetzmann
Yale School of Alan Marcus Lee R. Thomas
Management Boston College Pacific Investment
Management Company
Elizabeth R. Hilpman Paul O’Connell
Barlow Partners, Inc. FDO Partners
 Named Endowments
The CFA Institute Research Foundation acknowledges with sincere gratitude the
generous contributions of the Named Endowment participants listed below.
Gifts of at least US$100,000 qualify donors for membership in the Named Endowment
category, which recognizes in perpetuity the commitment toward unbiased,
practitioner-oriented, relevant research that these firms and individuals have
expressed through their generous support of the CFA Institute Research Foundation.

Ameritech Miller Anderson & Sherrerd, LLP

Anonymous Nikko Securities Co., Ltd.
Robert D. Arnott Nippon Life Insurance Company of
Theodore R. Aronson, CFA Japan
Asahi Mutual Life Insurance Company Nomura Securities Co., Ltd.
Batterymarch Financial Payden & Rygel
Management Provident National Bank
Boston Company Frank K. Reilly, CFA
Boston Partners Asset Management, Salomon Brothers
L.P. Sassoon Holdings Pte. Ltd.
Gary P. Brinson, CFA Scudder Stevens & Clark
Brinson Partners, Inc. Security Analysts Association
Capital Group International, Inc. of Japan
Concord Capital Management Shaw Data Securities, Inc.
Dai-Ichi Life Insurance Company Sit Investment Associates, Inc.
Daiwa Securities Standish, Ayer & Wood, Inc.
Mr. and Mrs. Jeffrey Diermeier State Farm Insurance Company
Gifford Fong Associates Sumitomo Life America, Inc.
Investment Counsel Association T. Rowe Price Associates, Inc.
of America, Inc. Templeton Investment Counsel Inc.
Jacobs Levy Equity Management Frank Trainer, CFA
John A. Gunn, CFA Travelers Insurance Co.
John B. Neff, CFA USF&G Companies
Jon L. Hagler Foundation Yamaichi Securities Co., Ltd.
Long-Term Credit Bank of Japan, Ltd.
Lynch, Jones & Ryan, LLC
Meiji Mutual Life Insurance
Company
Senior Research Fellows
Financial Services Analyst Association
For more on upcoming Research Foundation
publications and webcasts, please visit
www.cfainstitute.org/en/research/foundation.
 ISBN 978-1-944960-89-6

Available online at www.cfainstitute.org 9 781944 960896