5

Fixed income risk attribution

Chithra Krishnamurthi
RiskMetrics Group
[email protected]

We compare the risk of the active portfolio with that of the benchmark and segment the difference
between the two to correspond to decisions made by the manager of the active portfolio. There
are three main decisions we consider: interest rate, which encompasses duration, allocation and
selection decisions; currency; and credit.

1 Introduction
A fixed income portfolio manager may generate high returns compared to the benchmark, at least
in the short term. But returns tell only half the story. For instance, it might turn out that the
returns are not high enough to justify the amount of risk taken. So measuring and reporting
portfolio risk is crucial for judging portfolio performance. Furthermore, understanding risk is
paramount to controlling it. So it is important to be able to measure the risk contribution of each
investment decision of the manager to the total relative risk of the portfolio with respect to the
benchmark.

We compare the risk of the active portfolio with respect to the benchmark and carve up the
difference between the two into components that correspond to decisions made by the manager of
the active portfolio. While there is a potential for greater returns when deviating from benchmark
positions by taking on more risk, there is also a concurrent potential for underperformance relative
to the benchmark. Therefore, it is essential to be able to measure the risk exposures of the
portfolio vis-à-vis the benchmark. We provide a methodology for risk attribution that is rigorous
from a mathematical standpoint and relevant from an investment standpoint.

Although there is now an extensive body of research, both in academia and industry, pertaining to
methodologies in fixed income attribution, there is no industry standard that may be used as a
guide. Much of the work has been focused on return attribution and paid little attention to risk
attribution for fixed income portfolios. Most of the fixed income return attribution methodologies
6 Fixed income risk attribution

Table 1
Components of return attribution

Interest Rate Currency Credit

Overall Duration
Sector Allocation
Security Selection

that are used in the industry may be classified into two groups: duration approach and principal
component analysis. We have drawn on some of the general principles of the duration approach to
come up with our own framework for risk attribution. In particular, we have mounted our risk
attribution methodology on a return attribution system that is similar to that in van Breukelen
(2000). This return attribution system is also close in spirit to the one described by Feser et al
(2002). For the risk attribution model, we have followed the methodology of Mina (2003).

The broad scope of any performance and risk attribution system is to compare the investment
outcomes of the active portfolio and the benchmark and to ascribe the difference between the two
to the decisions made by the manager of the active portfolio. This general principle applies equally
to equity and to fixed income attributions. However, within this broad outline, the investment
processes and the sources of risk for equity and fixed income portfolios differ significantly. Equity
returns are determined by market segments while fixed income returns are ruled primarily by
duration, yield curve and credit quality. This means that a fixed income attribution system cannot
be obtained by merely tweaking an equity attribution system. Nonetheless, we draw on the
similarities between the two systems to facilitate our understanding.

2 Return attribution
We start by establishing a return attribution system for fixed income portfolios and then use it to
come up with a risk attribution system. The difference in return between the benchmark and the
portfolio are due to different investment decisions made by the portfolio manager. Return attribution
is a way of measuring the contribution of each of these decisions to the difference in return
between the benchmark and the portfolio. Table 1 lists the return attribution components that we
analyze. In the simplest case, a fixed income portfolio consists of riskless government bonds which
may be denominated in different currencies. For example, Central Banks usually invest their
 Return attribution 7

international reserves in sovereign bonds denominated in different currencies.1 For such a portfolio
which has no credit risk, there are two components of return attribution that we will consider:
interest rate and currency. The interest rate component may further be subdivided into three
components: overall duration, market allocation and security selection.

One of the first decisions is the overall duration of the portfolio. If the manager has an overall
positive view of the bond markets, this will be reflected in a choice of a higher total duration for
the portfolio compared to the total duration of the benchmark.

The next decision concerns the manager’s view about markets or sectors. The markets or sectors
here could be countries, currency markets, maturity segments, duration buckets or credit classes. A
positive view in a particular market/sector is manifested as higher allocation in that market/sector
compared to the benchmark; this differential allocation could be achieved by choosing either a
longer duration than that of the benchmark or a larger weight compared to the benchmark or a
combination of both.

The final level of drill-down is the selection of a particular security in each individual market.

Currency attribution is dealt with separately by keeping interest rates constant and shocking the
currency rates to perform return and risk attributions. If we consider a portfolio which has
corporate bonds in it as well, then we need to incorporate return due to credit bets. And as with
the currency allocation, the decision to invest in corporate bonds, which may offer a higher rate of
return, has to be considered separately. In this case, we shock the credit spreads and perform
return and risk attribution for rating and/or sector allocation and selection.

An important underpinning of the fixed income investment process is the concept of duration.
Duration is also the feature of a fixed income portfolio that differentiates it from an equity
portfolio. Duration measures the sensitivity of the bond’s value to parallel changes in the yield
curve. We illustrate the role played by duration in a fixed income portfolio with the following
example. If the manager of an equity portfolio believes that a particular market is going to perform
well, he overweights that market relative to the benchmark. A fixed income portfolio manager with
the same belief about that market has two alternatives to express his view. He could choose to
overweight that particular market relative to the benchmark, or he could choose the same total

1 de Almeida da Silva Junior (2004) provides a specific example.

8 Fixed income risk attribution

weight in that market as the benchmark but invest in assets with a longer duration than the
benchmark. A valid return attribution system should consider this extra degree of freedom.

The best way to understand fixed income return attribution is by drawing a parallel with the more
familiar structure of equity return attribution. The example above shows that allocation decisions in
a fixed income portfolio are made along two dimensions: weights and duration. The best way to
understand fixed income return attribution is by drawing a parallel with the more familiar structure
of equity return attribution. The fixed income allocation decision may be implemented by changing
either the weight or the duration or a combination of both, that is, by changing the overall value
of weight times duration which is the term W D where W and D refer to the weight and duration
of a sector respectively. So in a fixed income portfolio, W D is analogous to weight in an equity
portfolio. Since we would like to make use of the product of weight and duration as the equivalent


of weight in an equity portfolio, we express the return the following way: W D Dr , where r is
the security return. This is a weighted average of the return where the weights are the product of
the true weights and durations. The term r/D denoted by R is analogous to return in an equity
portfolio.

Before we deal with each of the components of return attribution in greater detail, we set up a
general framework and establish the notation that will be followed in the remainder of this
document:

T = {1, 2, . . . , N} denotes the universe of securities. These are the securities present in the
portfolio and the benchmark.

WiB is the benchmark position in security i. If a security has a weight of zero, it is not a part of
the benchmark.

WiP is the portfolio position in security i. If a security has a weight of zero, it is not a part of
the portfolio.


WAB = i∈A WiB is the benchmark position in sector A.


WTB = i∈T WiB = 1.


WAP = i∈A WiP is the portfolio position in sector A.


WTP = i∈T WiP = 1.

ri is the return of an individual issue in the portfolio or benchmark in local currency.

 Return attribution 9

Di is the duration of security i.

Ri = ri
Di
is the local return normalized by duration.

B
i∈A Wi Di
DAB = WAB
is the average duration of benchmark issues in sector A.

WiB Di
DTB = i∈T
WTB
= i∈T WiB Di is the total benchmark duration.

P
i∈A Wi Di
DAP = WAP is the average duration of portfolio issues in sector A.

WiP Di
DTP = i∈T
WTP
= i∈T WiP Di is the total portfolio duration.

B
i∈A Wi ri
RAB = WAB DA
B is the normalized benchmark return for sector A.

P
i∈A Wi ri
RAP = P
WA DAP is the normalized portfolio return for sector A.

WiB ri
RTB = i∈T
DTB
is the normalized total benchmark return.

WiP ri
RTP = i∈T
DTP
is the normalized portfolio return for sector A.

Ec is the currency return for currency c.

Ei is the currency return for security i.

The return in the base currency of any given fixed income portfolio consisting of the securities
1, 2, . . . , N may be decomposed as


N
Return of the portfolio = Wi (ri + Ei ) (1)
i=1

N 
N
= W i ri + Wi Ei . (2)
i=1 i=1

Equation (1) states that the return of any single issue in the base currency of the portfolio is the
sum of the asset return in local currency and the currency return. Note that this is a linear
approximation.
10 Fixed income risk attribution

Table 2
Return attribution

Bench Reference port 1 Reference port 2 Active port

Overall duration B P P P
Sector allocation B B P P
Security selection B B B P

Equation (2) may be used to decompose the difference in return between the actively managed
portfolio and the benchmark in the following way:


N
Excess return = (WiP − WiB )Di Ri
i=1

N
+ (WiP − WiB )Ei . (3)
i=1

The first term on the right hand side of equation (3) explains the contribution of active fixed
income management, including choices of overall duration, market allocation and issue selection.
The last term in the equation explains the contribution of currency allocation to the difference in
returns. We will deal with each of these two aspects individually.

In order to undertake a detailed analysis of the fixed income attribution, we now construct two
portfolios: reference portfolio 1 and reference portfolio 2. Here we have used the terminology of
van Breukelen (2000). Reference portfolio 1 differs from the actual benchmark only in its overall
duration, which is that of the active portfolio. Reference portfolio 2 is constructed such that it has
the same overall duration and the same sector allocation as the actual portfolio. It differs from the
actual portfolio only in issue selection, that is R, at each level. It differs from reference portfolio 1
only in allocation across sectors or markets. Table 2 shows the differences and similarities of the
four portfolios. Specifically, if we consider a sector A, benchmark and portfolio allocation for A
are given by: WAB DAB and WAP DAP respectively. Similarly, by security selection for sector A, we
mean: RAB and RAP .

So we have the following: (a) the difference in return between reference portfolio 1 and the
benchmark explains the overall duration contribution, (b) the difference in return between reference
portfolio 2 and reference portfolio 1 gives the allocation contribution, and (c) the difference in
 Return contribution 11

return between reference portfolio 2 and the active portfolio lays out the selection contribution. We
now take a closer look at each of these three factors in turn.

3 Return contribution
3.1 Interest rate

Overall duration

We can construct the reference portfolio 1 by assigning the following weights to the securities in
T:

DTP
Wi(1) = WiB
DTB

where Wi(1) are the weights of the issues in reference portfolio 1. This will result in a portfolio
which assigns non-zero weights only to those securities which have non-zero weights in the
benchmark (this is equivalent to choosing the same subset of securities from T as the benchmark
does). Moreover, the non-zero weights assigned to the securities will be in proportion to the
benchmark weights.

The contribution of the overall duration to the differential performances of the portfolio and the
benchmark is given by the difference in returns between the benchmark and reference portfolio 1:


N 
N
DTP
(Wi(1) Di − WiB Di )Ri = (WiB Di − WiB Di )Ri .
i=1 i=1
DTB

This reduces to
 N
DTP
Overall duration = −1 WiB Di Ri . (4)
DTB i=1

Sector allocation

The sector allocation contribution is determined by the difference in return between reference
portfolio 1 and reference portfolio 2. Reference portfolio 2 has to: (a) have security weights
proportional to those of the benchmark, and (b) the same overall duration as the active portfolio.
These are criteria that are common with reference portfolio 1. Additionally, reference portfolio 2
12 Fixed income risk attribution

must have (c) the same sector allocation as that of the actual portfolio. By this we mean that for a
given sector A

WA(2) DAB = WAP DAP . (5)

Thus reference portfolio 2 is constructed by investing in the same securities as the benchmark, but
choosing the sector weights to satisfy

WAP DAP
WA(2) =
DAB

At the individual security level, we may choose the weights the same way as at the sector level.
For each sector A

WAP DAP
Wi(2) = WiB , ∀i ∈ A,
WAB DAB

so within a sector, security weights are proportional to benchmark weights.

The sector allocation contribution is the difference in return between reference portfolio 2 and
reference portfolio 1. Mathematically, this may be expressed as
 
Sector allocation = WA(2) DAB RAB − WA(1) DAB RAB
A A
 (1) B
= (WAP DAP − WA DA )(RAB − RTB ), (6)
A

where

B
i∈A Wi ri
RAB =
WAB DAB

RAB is the normalized return for sector A. The term RTB is defined similar to RAB with the sum
taken over T instead of over A. We also define RAP and RTP analogously.

Security selection

The active portfolio and reference portfolio 2 have the same total duration and the same allocation
across sectors and differ only in the security bets. Thus the security selection contribution is
 Risk attribution 13

determined by the difference between reference portfolio 2 and the active portfolio:

Security selection = portfolio return − reference portfolio 2 return (7)

 
= WAP DAP RAP − WAP DAP RAB (8)
A A

= WAP DAP (RAP − RAB ). (9)
A

3.2 Currency contribution

Currency allocation decisions, which are dealt with separately, can be written as
 
Currency performance = WcP Ec − WcB Ec (10)
c c

= (WcP − WcB )(Ec − E BT ), (11)
c

where Wc is the weight of the portfolio or the benchmark aggregated over a particular currency c


and E BT = c WcB Ec is the benchmark weighted average of the currency component.

4 Risk attribution
4.1 Tracking error

With a return attribution system defined, we proceed to define risk attribution in the same way as
in Mina (2003). As in Mina (2003), we begin with the relationship between the portfolio excess
return and the forward looking portfolio tracking error. The excess return is the sum over all
securities of the portfolio bet on each security times the security’s realized return. The forward
looking tracking error is defined similarly, but with the security returns treated as yet-to-be-realized
random variables. Using a model for future volatilities and correlations of the security returns, we
may calculate the standard deviation of the future portfolio excess return. This is defined as the
portfolio tracking error. Note that at the portfolio level, we are faced with a decision of whether to
use the actual security returns in the base currency or the linearized returns as in Equation (2).
Using the actual returns has the obvious attraction of making our tracking error correspond to the
true excess return, while using the linearized returns allow us to further decompose the risk
according to the return attribution system. Since our goal is to attribute the risk along the same
dimensions as the return, we proceed with the linearized security returns.
14 Fixed income risk attribution

For each level of decomposition, the return attribution specifies the amount of portfolio excess
return that was due to a particular part of the investment process. Analogously, our risk attribution
describes the amount of risk, on a forward-looking basis, inherent in the existing investment
choices. Thus, just as at the portfolio level, we define risk attribution at any level of drilldown by
using the expression for return attribution, but treating the returns as random variables. Again,
using a model for return volatilities and correlations, we define the risk attribution as the standard
deviation of the return expression. Note that the specific model for volatilities and correlations is
not relevant here; we are concerned only with defining the quantities for which we will compute
standard deviation. We now treat the various stages of the risk attribution in detail.

As explained earlier, fixed income return attribution consists of two components: interest rate and
currency taken together and credit or spread. The interest rate and currency components of return
may be subdivided into returns due to interest rate, namely, overall duration, sector or maturity
allocation, security selection and currency allocation. For the interest rate and currency component,
we may define the overall tracking error of the interest rate and currency component, denoted
by
, as the standard deviation of excess returns due to the overall interest rate and currency
components:
 N 
 
n 
N 
N

= std WiP Di Ri − WiB Di Ri + WiP Ec − WiB Ec . (12)
i=1 i=1 i=1 i=1

At the next level of drill-down, we may split the interest rate risk and currency risk to deal with
them individually. Currency risk attribution may be dealt with separately just as in the case of
currency return attribution. From equation (11), we get the total currency risk
 

std (WcP − WcB )(Ec − E BT ) . (13)
c

From this we may infer the next level of drilldown, stand-alone currency risk for currency c:

std[(WcP − WcB )(Ec − E BT )]. (14)

The interest rate tracking error may correspondingly be defined as

 N 
 
n
std Wi Di Ri −
P B
Wi Di Ri . (15)
i=1 i=1

The next drilldown level of the fixed income management component consists in deriving the
individual tracking errors of the overall duration, sector allocation and security selection.
 Risk attribution 15

From the overall duration return in equation (4), we define total duration attribution risk to be
 
 DP
std (WAB TB DAB − WAB DAB )RAB . (16)
A∈T
DT

From equation(6), the sector allocation risk for sector A is

std[(WAP DAP − WA(1) DAB )(RAB − RTB )]. (17)

Finally, from equation (9), the security selection risk for security i in sector A is given by

WAP DAP
std[ri (WiP − WiB )]. (18)
WAB DAB

4.2 Incremental VaR

We now consider the calculation of the relative incremental VaR. In the case of an equity portfolio,
relative IVaR was introduced in order to overcome the obstacle presented by the tracking error (or
relative VaR) not being additive.2 We use the same technique here for that very same reason, and
such that it is in keeping with the spirit of the risk attribution methods for an equity portfolio.

At the topmost level of the risk attribution report we have the Overall Fixed Income Tracking Error
which, when it is based on the true returns, will not be decomposed into relative IVaRs. At the
next report level, the overall interest rate and currency tracking errors taken together, i.e.,
, and
overall credit tracking error may be decomposed into relative IVaRs. Suppose the excess return due
to interest rate and currency is denoted by ICR, then we have

ICR = rI R + rCC , (19)

where rI R and rCC refer to the interest rate and the currency management risk respectively:


N 
n
rI R = WiP DiP RiP − WiB DiB RiB (20)
i=1 i=1

N 
N
rCC = WiP Ec − WiB Ec . (21)
i=1 i=1

2 Mina (2003)
16 Fixed income risk attribution

Table 3
Weights and durations

Sectors Benchmark Wts Portfolio Wts Active Wts Benchmark Dur Portfolio Dur
1 - 3 0.299 0.262 −0.038 0.596 0.419
3 - 5 0.234 0.452 0.218 0.922 1.718
5 - 10 0.297 0.194 −0.103 1.994 1.357
10+ 0.171 0.093 −0.078 2.251 1.160
Total 1.0 1.0 0.0 5.763 4.654

The overall interest rate and currency risk attribution may be made additive in the following way:

cov(ICR, rI R ) cov(ICR, rCC )

= + . (22)

The overall interest rate risk may be decomposed in a similar manner. Suppose the returns due to
overall duration, sector allocation and security selection (equations (4), (6), (9)) are denoted by
rDD , rAA and rSS respectively, and the overall interest rate tracking error is denoted by ITE. Then
we have

cov(rI R , rDD ) cov(rI R , rAA ) cov(rI R , rSS )

ITE = + + . (23)
ITE ITE ITE

5 An example
We present an example of a risk attribution report for a fixed income portfolio and explain how to
interpret the statistics. The benchmark consists of a total of 233 sovereign bonds in seven different
currencies with U.S. dollars as the base currency. The durations range from one to seventeen years.
The active portfolio is invested in a subset of 25 of the benchmark bonds.

In order to calculate the interest rate risk we choose the sectors to be the following duration
buckets: 1-3 years, 3-5 years, 5-10 years and 10+ years. Table 3 shows weights, bets and durations
of each sector. The portfolio is significantly overweight, that is 21.8%, in the 3-5 year duration
sector and the largest difference in durations between the portfolio and the benchmark is in the
10+ sector where there is a difference of −1.09 years.
 An example 17

Table 4
Fixed income weights

Sectors Benchmark Reference portfolio 1 Reference portfolio 2 Active Portfolio

1 - 3 0.596 0.482 0.419 0.419
3 - 5 0.922 0.745 1.718 1.718
5 - 10 1.990 1.610 1.357 1.357
10+ 2.251 1.817 1.160 1.160
Total 5.763 4.654 4.654 4.654

Table 5
Interest rate tracking error

Sectors Interest Rate Duration Allocation Selection

1 - 3 13.017 7.754 1.174 5.736
3 - 5 56.874 13.077 16.596 9.568
5 - 10 43.509 24.866 1.533 10.277
10+ 58.930 23.316 9.417 11.531
Total 60.145 67.916 23.967 14.45

Table 4 shows the effective weights, that is W D, of the two reference portfolios. The row with the
Total W D numbers gives the overall durations of the four portfolios and as we can see, both
reference portfolio 1 and 2 have the same overall duration as the active portfolio. Notice that for
reference portfolio 1, the allocation across sectors is proportional to the benchmark allocation across
sectors. For example, for sectors 1-3 and 3-5, the sector allocation proportions for the benchmark
and the reference portfolio 1 are 0.596/0.922 = 0.646 = 0.482/0.745.

In Table 5, we have the total interest rate tracking error along with the duration, allocation and
selection tracking errors for each of the four duration buckets. We see that duration risk is highest
in the 5-10 and 10+ duration buckets. As mentioned earlier, the difference in duration between the
portfolio and the benchmark is highest in the 10+ sector where it is −1.09 years. However, the
portfolio is not significantly underweight in the 10+ sector compared to the underweight in the
5-10 sector where the duration difference is −0.64 years. This has the effect of magnifying the
18 Fixed income risk attribution

Table 6
Interest rate IVaR

Interest Rate Duration Allocation Selection

60.145 58.772 −8.239 9.285

duration risk in the 5-10 sector. As pointed out earlier, the active weight is highest in the 3-5
sector which explains the fact that the 3-5 is the sector with the maximum allocation risk.

With risk attribution, we not only get stand-alone numbers but we can also analyze correlation
effects across sectors and securities. While stand-alone numbers are useful, the largest stand-alone
tracking errors do not always contribute the maximum amount to the total tracking error.

In Table 6, we have the interest rate IVaR at the total portfolio level. The duration, allocation and
selection risk contributions add up to the total interest rate tracking error of 60.15 bp. We see that
duration risk is larger than the allocation and selection risks in almost all the sectors. Hence it is
not surprising that duration risk contribution to the total interest rate tracking error is much higher
than the risk contributions of either allocation or selection. The incremental risk of allocation and
selection are smaller than the tracking error for allocation and selection. This is explained by
correlation effects which come into play across sectors. In particular, the incremental allocation risk
being negative implies that one way to decrease the interest rate tracking error would be to
increase the allocation bets in those sectors which have negative risk contributions.

6 Conclusion
We developed a risk attribution model for fixed income portfolios that measured the impact of
investment decisions on the difference in risk between an actively managed portfolio and a given
benchmark. We did this by establishing a return attribution system and defining a risk statistic. We
split up the investment process into a series of decisions: interest rate, encompassing duration,
allocation and selection decisions, and currency. We then measured the contribution of each of
these components to the difference in risk between the benchmark and the portfolio.
 Conclusion 19

References

de Almeida da Silva Junior, F. A. (2004). Performance attribution for fixed income portfolios in
Central Bank of Brazil international reserves management, Risk Management for Central Bank
Foreign Reserves, European Central Bank.
Feser, M., Rimaud, C., Wilson, M., and Fisher, S. (2002). Investment process and the sources of
active excess return, Investment Insight: International Fixed Income, J.P. Morgan Investment
Management, Quarter 1.
Mina, J. (2003). Risk Attribution for Asset Managers, RiskMetrics Journal, 3(2): 33–57.
Ramaswamy, S. (2001). Fixed Income Portfolio Management: Risk Modelling, Portfolio
Construction and Performance Attribution, BIS Banking Papers, Issue 6, April 2001.
van Breukelen, G. (2000). Fixed-Income Attribution, Journal of Performance Measurement,
4(4): 61–68.
20 Fixed income risk attribution