Riding the Yield Curve:

A Variety of Strategies
DAVID S. BIERI AND LUDWIG B. CHINCARINI

DAVID S. BIERI n its simplest form, the rational expecta- for the predictive power of forward rates in
is adviser to the general
manager at the Bank for
International Settlements
in Basel, Switzerland.
[email protected]
I tions hypothesis of the term structure of
interest rates (REHTS) posits that in a
world with risk-neutral investors, the
n-period long rate is a weighted average of the
future spot rates and thus any 1-period forward
terms of future sport rates, there is a small
strand of literature that examines the persis-
tence of excess returns from riding strategies
across different holding horizons with different
maturity instruments. In their seminal paper,

Draft
LUDWIG B. rate is an unbiased predictor of the corre- Dyl and Joehnk [1981] examine different
CHINCARINI sponding future 1-period spot rate. Conse- riding strategies for U.S. T-bill issues from 1970
is an adjunct professor and
financial consultant at
quently, the expectations hypothesis implies to 1975 and find that there are significant,
Georgetown University in that with the possible exception of a term pre- albeit small, excess HPRs to be made from
Washington, DC. mium, the holding period returns (HPRs) of riding the yield curve. They use a simple filter
[email protected] a class of fixed-income instruments are iden- rule based on break-even yield changes in
tical, independent of the instruments’ original order to quantify the ex ante riskiness of riding
maturity.1 Under this assumption, for example, the yield curve. Based on this filter, their results
the returns from purchasing a 3-month gov- indicate that the returns increase with both
ernment security and holding it until maturity the holding horizon and the maturity of the
and the returns from purchasing a 12-month instrument.
government security and holding it for 3 Grieves and Marcus [1992] are able to
months are identical. The strategy of pur- produce similar results by looking at a much
chasing a longer-dated security and selling it longer time series of monthly zero-coupon T-
before maturity is referred to as “riding the bill rates from 1949 to 1988. They apply the
yield curve.” same filter rule as Dyl and Joehnk to identify,
If the REHTS holds, then, for any given ex ante, under what type of yield curve envi-
holding period, riding strategies should not ronment excess returns from rolling can be
yield excess returns compared with holding a anticipated. Although their results confirm that
short-dated security until maturity. Any evi- longer-maturity rides outperform the simple
dence of persisting excess returns from such buy-and-hold strategy of the short-term
trading strategies would indicate the existence instrument, they conclude that, on a risk-
of risk premia associated with the term struc- adjusted basis, longer rides perform slightly
ture. The body of literature on different tests worse because of increased interest rate risk.
of the expectations hypothesis is very large, Overall, they find evidence against the pure
and overall the results remain inconclusive.2 form of the expectations hypothesis since it
Although the majority of tests of the appears that profitable trading strategies have
expectations hypothesis are hinged on testing gone unexploited. Using daily closing prices

6 RIDING THE YIELD CURVE: A VARIETY OF STRATEGIES SEPTEMBER 2005

for regular U.S. T-bill issues from 1987 to 1997, Grieves REHTS AND RIDING THE YIELD CURVE
et al. [1999] are able to confirm these earlier findings, and
they also find that their results are relatively stable over One implication of the REHTS is that with the
time. In contrast to Dyl and Joehnk, they conclude that exception of time-varying term premia, the return on a
conditioning the ride on the steepness of the yield curve longer-period bond is identical to the return from rolling
does not seem to improve the performance significantly. over a sequence of short-term bonds. As a consequence,
Most of the existing literature on excess returns from longer-term rates y nt are a weighted average of short-term
riding the yield curve is exclusively limited to examining rates y mt plus the term premia. This can be expressed as
the money market sector of the yield curve, i.e. maturi- follows:
ties below 12 months, and has thus far only studied the
k −1
U.S. Treasury market.3 1 (1)
In this article, we aim to add to this strand of liter-
ytn = ∑ E y m + s n, m
k h=0 t t +h
ature by looking at riding strategies for maturities beyond
one year, looking at different currencies (euro and ster-
ling), and also comparing rides between risk-free gov- where ymt+ h is the m-period zero-coupon yield at time t
ernment securities and instruments that contain some + h, Et is the conditional time expectations operator at
level of credit risk, namely, LIBOR-based deposits and time t, and σ n,m is the risk premium between n- and m-
swaps. In addition, we propose and test some forward- period zero-coupon bonds (with n > m). In Equation (1),
looking strategies based on either simple statistical mea- k = m–n is restricted to be an integer.
sures or economic models that incorporate the main In the absence of any risk premia, by taking expec-
drivers of the yield curve. The main purpose of such rules tations and subtracting ymt from both sides we can rewrite

Draft
is to provide market practitioners with a simple tool set Equation (1) as
that not only allows them to identify potentially prof-
itable riding strategies but also enables an ex ante ranking k −1
1
of individual strategies. ytn − ytm =
k ∑y m
t+h
− ytm (2)
h =1

RIDING THE YIELD CURVE

Thus, under the REHTS, the future differentials on
Riding the yield curve refers to the purchase of a the short rate are related to the current yield spread
longer-dated security and selling it before maturity.4 The between the long-term and short-term zero-coupon rates.
purpose of riding the yield curve is to benefit from cer- Equation (2) forms the basis for most empirical tests of
tain interest rate environments. In particular, if a fixed- the REHTS, by running the regression
income manager has the choice between investing in a
k −1
1-month deposit or a 12-month money market instrument
and selling after 1 month, there are certain rules of thumb
1
k ∑y m
t +h (
− ytm = a + b ytn − ytm + et ) (3)
h =1
as to which strategy might yield a higher return. For
instance, when the yield curve is relatively steep and and testing whether β = 1. In practice, however, most
interest rates are relatively stable, the manager will ben- empirical studies report coefficients that are significantly
efit by riding the curve rather than buying and holding different from 1, which is almost exclusively taken as evi-
the short-maturity instrument. dence for the existence of (time-varying) risk premia.5
However, there are risks to riding the yield curve, Rather than postulating a linear relationship between
most obviously the greater interest rate risk associated with the future differentials on the short rate and the current
the riding strategy (as reflected by its higher duration). slope of the term structure as expressed in Equation (2),
Thus, if one is riding and yields rise substantially, the we calculate the ex post excess HPRs from riding the
investor will incur a capital loss on the riding position. Had yield curve. Thus, if the REHTS holds and there are no
the investor purchased the instrument that matched her risk premia, these returns should be zero.
investment horizon, she would have still ended up with Therefore, according to the REHTS, if all agents
a positive return. are risk neutral and concerned only with the expected
return, the expected one-period HPR on all bonds, inde-

SEPTEMBER 2005 THE JOURNAL OF FIXED INCOME 7

pendent of their maturity, should be identical and would • the yield differential between LIBOR rates and the
be equal to the return on a one-period asset: repo rate for General Collateral
• the slope of the term structure of risk-free
Et H tn+1 = ytm (4) interest rates
• the relative supply of government corporate debt.
n
where H t+1 denotes the HPR of an n-period instrument
between time t and t + 1. This result can now be used to There are also other non-default factors, such as li-
derive the zero excess holding period return (XHPR) quidity and yield spread volatility, that may play an impor-
condition of the REHTS by restating equation (4) as tant role in determining yield spreads.8
In line with the pioneering work by Dyl and Joehnk
XHtn+1 = Htn+1 − ytm = 0 (5) [1981], we also derive a formula for quantifying the risk
associated with a given riding strategy. This measure is
Hence, if the REHTS holds, we should not be able traditionally referred to as the “margin of safety” or
to find any evidence that fixed-income managers are able “cushion” and can be used as a conditioning moment or
to obtain any significant non-zero XHPRs by riding the filter for different rides. By calculating the cushion of a
yield curve. given riding strategy, the investor has an ex ante indica-
tion of how much, ceteris paribus, interest rates would have
Mathematics of Riding to have risen at the end of the holding period such that
any excess returns from riding would be eliminated. The
In this section, we derive the main mathematical cushion is therefore also referred to as the “break-even
formulae for riding the yield curve relative to a buy-and- yield change.” We will also derive an approximate for-

Draft
hold strategy. Because we evaluate different riding strate- mula that may appeal to the market practitioner because
gies for maturities beyond one year, we need to distinguish of its simplicity and intuitive form.
between riding a money-market instrument and riding a Riding the money market curve. For the analysis of
bond-market instrument. riding the money market curve, we assume that our rates
Furthermore, we are not only interested in evalu- are money-market or CD equivalent yields. We can pos-
ating riding returns for different maturities, but we also tulate that the price of an m-maturity money-market
consider the case where we use different instruments to instrument at time t is given by
ride the yield curve. In particular, we consider the case
of comparing a ride using a (risk-free) government bond 100 (6)
PmM, t =
with riding down the credit curve with a LIBOR/swap-  m
 1 + ym ,t · z 
based instrument. Because investors expect to be rewarded
for taking on non-diversifiable credit risk, two securities
that are identical except for the level of credit risk must where ym,t represents the current CD equivalent yield9 of
have different yields. Thus, comparing the returns from the instrument at time t, m is the number of days to the
two strategies that involve fixed-income instruments with instrument’s maturity, and z is the instrument and cur-
different credit risk would normally necessitate the spec- rency-specific day count basis.10 We can also denote the
ification of a framework that deals appropriately with price of this same maturity instrument after a holding
credit risk. period of h days as
However, drawing on results from the literature on
the determinants of swap spreads,6 we can assume that 100 (7)
PmM− h ,t + h =
the yield differential between government securities and  ( )
m-h 
swaps is not primarily a consequence of their idiosyn-  1 + ym − h ,t + h · 
 z 
cratic credit risk. This strand of literature argues that even
in the absence of any credit or default risk, swap spreads where ym-h,t+h represents the interest rate valid for the
would be non-zero,7 since they predominantly depend instrument, which has now m – h days left until final
on other factors such as redemption. Thus, the HPR of the ride of an m-matu-
rity instrument between time t and time t + h is given by

8 RIDING THE YIELD CURVE: A VARIETY OF STRATEGIES SEPTEMBER 2005

  m
PmM− h,t + h  1 = ym , t ⋅ z  Similarly, the excess holding returns from rolling
H[Mm , h ] = −1 = −1
PmM,t  (
m−h  ) (8) down the bond curve for h days are
 1 + ym − h, t + h · 
 z 

 (
1 + ym, t )
m/z 
  h
 1 + y ( m − h ) / z − 1 −  yh, t z  if h < 1 year
 ( m − h, t + h ) 
The XHPRs of this strategy of riding over the choice XH[Bm, h ] =  (13)
of holding an instrument with the maturity equal to the


(
1 + ym, t )
m/z 
(
− 1 −  1 + yh, t )
h/z
− 1 if h > 1 year
  1 + ym − h,t + h ( )
( )   
m−h / z

investment horizon h can be expressed as  

  m  It is important to reiterate at this point that Equa-

  1 + ym , t    h tions (10)–(13) are expressed in terms of zero-coupon
z
XH[Mm, h ] = − 1 −  yh, t  (9)
 1 + y m − h   z rates; hence there are no coupon payments to be consid-
  m − h, t + h
z 
  ered. This does not mean, however, that our simple frame-
work cannot be transposed to the (more realistic) world
It follows from Equation (9) that riding the yield of coupon-paying bonds. Using the approximation
curve is more profitable, ceteris paribus, 1) the steeper the [(Pt+h)/Pt] – 1 ≈ ym,th/z – ∆yt Dt+h, we can restate Equa-
yield curve at the beginning of the ride (i.e., large values tion (13) in a more applicable way:11
for ym,t – yh,t) and 2) the lower the expected rate at the
end of the holding period (i.e., ym-h,t+h is low).  h   h
  ym, t − Dyt Dm − h, t + h  −  yh, t  if h < 1 year
Riding the bond curve. In line with the assumptions  z   z
XH[Bm', h ] ≈ (14)

Draft
for computing the returns for money market instruments,  y 
the zero-coupon prices for maturities beyond one year,   m, t
h
z  
(
− Dyt Dm − h, t + h  −  1 + yh, t )h / z − 1 if h > 1 year
where our rates used are zero coupon yields, can be cal-
culated. We can postulate that the price of an m-matu- where ∆yt = ym-h,t+h – ym,t and Dm-h,t+h is the modified
rity zero-coupon bond at time t is given by duration of the bond at the end of the holding horizon.
By virtue of this approximation, the subsequent parts of
100 (10) our analysis also apply to coupon-paying bonds.
PmB, t =
(1 + ym,t )(m / z) Break-even rates and the cushion. Given a certain
yield curve, the investor needs to decide whether to engage
where ym,t represents the current zero-coupon yield of in a riding strategy before making an informed decision
the instrument at time t, m is the instrument’s final matu- about selecting the appropriate instrument for the ride.
rity, and z is the appropriate day count basis. In line with The easiest way to make this decision is to use the cushion
Equation (7), we can denote the price of this same instru- or break-even rate change as an indication of how much
ment after holding it for h days as rates would have to have increased at the end of the
holding period h in order to make the riding returns equal
100 (11) to the returns from buying an h-maturity instrument and
PmB− h, t + h =
(1 + ym − h, t + h )(m − h) / z holding to maturity.
For example, if the yield curve is upward sloping,
longer-term bonds offer a yield pick-up over the one-
where ym-h,t+h represents the interest rate valid for the period short-term bonds. In order to equate the HPRs
zero-coupon bond, which is now an m – h maturity instru- across all bonds, the longer-maturity instruments would
ment that was purchased h days ago. Following Equation have to incur a capital loss to offset their initial yield advan-
(8), we can write the HPR from riding the zero coupon tage. Break-even rates show by exactly how much long-
bond curve as term rates have to increase over the holding period to
cause such capital losses. In other words, the break-even
rate is the implied end-horizon rate, y∗m–h,t, such that
H[Bm, h ] +
(1 + ym,t )m / z − 1 (12)
there are no excess returns from riding (i.e., XHt+h = 0).
(1 + ym − h,t + h )(m − h) / z m
By setting XH [m,h] = 0 and XH B[m,h] = 0 in Equations (9)

SEPTEMBER 2005 THE JOURNAL OF FIXED INCOME 9

EXHIBIT 1
Break-Even Rate and the Cushion

Yield

∗ C
ym−h,t
A
ym,t
Cushion

ym−h,t
B

Time to maturity
m−h m

Draft
and (13) respectively, we can derive the break-even rates C[ m , h ] = ym* − h, t − ym − h, t (17)
for both cases:
Money markets ride: Exhibit 1 provides a schematic illustration of a ride
on the yield curve from point A to point B. The cushion
 m h (15) is then defined as the vertical distance between points B
 ym , t z − yh, t z  z and C, that is, the amount by which interest rates have
ym* − h, t = ×
 1 + yh, t h  m − h to rise in order to offset any capital gains from riding the
 z  yield curve.
Thus, the concept of the cushion can now be used
Bond market ride: to define some simple filter rules for determining whether
to ride. For example, one such filter rule is based on the
 z / (m − h)
(16) assumption that interest rates display mean-reverting prop-
 
( )  erties and sends a positive riding signal whenever the cushion
m/z
  1 + ym ,t 
 − 1 ifÄ h Ä <Ä 1Ä year moves outside a prespecified standard deviation band

  1 + y h  
   around its historic moving average. The success rate of a
=    z  
h,t
ym* − h , t
 z / (m − h)
number of similar such rules is discussed in a later section.
 (
 1 + y m / z 
m ,t )  − 1 ifÄ h Ä >Ä 1Ä year
Selecting the best instrument for the ride. With such
 h/z  a simple decision-making strategy, the investor now needs
(
  1 + yh ,t ) 
to address the choice of the appropriate instrument for the
ride.12 In order to choose between two instruments, we
We can now see that under the REHTS without any need to compare the excess returns for a given riding
term premia, the break-even rate for a riding strategy strategy using either instrument. More formally, the excess
using an m-maturity instrument from time t to t + h is riding returns from using a government instead of a credit
equivalent to the m – h period forward rate implied by instrument are given by
the term structure at time t (i.e., y∗m–h,t = fm–h,m). The Money market ride:
cushion can now be written as

10 RIDING THE YIELD CURVE: A VARIETY OF STRATEGIES SEPTEMBER 2005

   m  riding with government bonds. Furthermore, the second
  1 + ym , t   factor also reveals that the slope differential gains in impor-
z
XH[M,ride = − 1
m, h ]

 1+ y m − h   tance as the mismatch between the holding horizon and
  m − h, t + h
z 
 
the instrument’s maturity increases.
  m 
1 + yˆm, t 
  z 
− − 1 DATA AND METHODOLOGY

 1 + yˆ m − h  
  m − h, t + h
z 
 
Data
(18)
The data used in this study were obtained either via
Bond market ride: the Bank for International Settlements (BIS) or directly
from the relevant central bank. As such, the choice of
estimation methodology for the yield curves is deter-
 (
1 + ym , t )m / z − 1 mined by the BIS or the respective central bank.
m, h ] = 
XH[B,ride (19)
(
 1 + ym − h, t + h )(m − h) / z  We are estimating returns for different rolling strate-
 (1 + yˆ )m / z  gies using monthly U.S., U.K., and German interest rates
m, t
− − 1 for both government and corporate liabilities. In the case
 (1 + yˆm − h, t + h )
m−h/ z

of the government liabilities, these rates are zero-coupon
or spot interest rates estimated from the prices of coupon-
where the hats over the variables indicate the corre- paying government bonds. In the case of corporate lia-
sponding rates for the credit instrument at the respective bilities, the zero-coupon rates were estimated from
times. Defining ŷm-h,t+h = ym,t + ε, ŷm-h,t+h = ŷm,t – η and

Draft
LIBOR deposit and swap rates.
ym-h,t+h = ym,t – c, we can substitute these conditions into Government zero-coupon curves. The government
Equations (18) and (19) to derive an approximate, yet zero-coupon time series for the three countries begin on
very tractable, expression for the excess riding return from different dates, span different maturity intervals, and are
using the two instruments:13 estimated using different methodologies. The data for
Money market ride: Germany span a period of over 30 years from January
1973 to December 2003. The series for the United
Kingdom starts from January 1979 and the data for the
 
1  Unites States are available only from April 1982.
m, h ] ≈
XH[M,ride −
{ he + (c −
144244
h ) ( m − h
3
) (20)
z  initial spread Exhibit 2 plots the evolution of the 3-month, 2-
 slope effect 
year, and 10-year government zero-coupon rates, and
Bond market ride: Exhibit 3 shows how the slopes of different sectors of the
government yield curves have changed over the sample
period.
 
The zero-coupon rates for the three countries also
m, h ] ≈
XH[B,ride
1
−
{
z  initial spread 14442444
(
he + (c − h ) Dm − h, t + h z 
3
) (21)
vary with respect to the maturity spectrum for which
 slope effect 
they are available. Although the data are available for all
Equations (20) and (21) highlight the two main fac- countries at 3-month intervals for maturities from 1 to 10
tors that determine which instrument yields a higher profit years, reliable data for the money market sector (i.e. matu-
from riding. The first factor is the difference in rates or rities below 1 year) are available only for the United States.
yield pickup between the two instruments for any given This is mainly because, unlike its European counterparts,
maturity, whereas the second factor is a slope term.14 the U.S. Treasury through its regular auction schedule of
Therefore, the bigger the initial yield differential between Treasury and Cash Management bills has actively con-
the government bond and the credit instrument, the less tributed to making this part of the yield curve very liquid.
attractive is a riding strategy using the former. The second Since the yields on T-bills are de facto zero-coupon rates,
factor indicates that the steeper the slope of the govern- we use 3- and 6-month constant maturity rates published
ment yield curve compared with the slope of the credit by the Federal Reserve to extend the maturity spectrum
yield curve, the higher the relative excess returns from for the U.S. data.15

SEPTEMBER 2005 THE JOURNAL OF FIXED INCOME 11

EXHIBIT 2
Evolution of Zero-Coupon Yield Curves with Shaded Recessions
Government LIBOR/Swaps
15.0 12
3-month Bill 3-month LIBOR
2-year Bond 2-year Swap
10-year Bond 10-year Swap
12.5 10

10.0 8
Yield (%)

Yield (%)
7.5 6
US
5.0 4

2.5 2

0.0 0
1982 1984 1986 1988 1990 1992 1994 1996 1998 2000 2002 1987 1989 1991 1993 1995 1997 1999 2001 2003

18 14
3-month Bill 3-month LIBOR
2-year Bond 2-year Swap
16 10-year Bond 10-year Swap
12

14

10
12
Yield (%)

Yield (%)
10 8
UK
8
6

4
4

2 2

Draft
1979 1981 1983 1985 1987 1989 1991 1993 1995 1997 1999 2001 2003 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003

14 11
12-month Bill 3-month LIBOR
2-year Bond 2-year Swap
10-year Bond 10 10-year Swap
12
9

10 8

7
Yield (%)

Yield (%)

8
Ger 6

6 5

4
4
3

2 2
1973 1976 1979 1982 1985 1988 1991 1994 1997 2000 2003 1988 1990 1992 1994 1996 1998 2000 2002

The majority of the central banks that report their The second column of Exhibit 2 shows the evolu-
zero-coupon yield estimates to the BIS MEDTS, including tion of selected LIBOR/swap rates, and the changes in
Germany’s Bundesbank, have adopted the so-called the slopes of different sectors of the yield curve are dis-
Nelson-Siegel approach [1987] or the Svensson [1994] played in Exhibit 3. Exhibit 4 plots the development of
extension thereof. Notable exceptions are the United the TED and swap spreads for the different currencies.
States and the United Kingdom, both of which are using The zero-coupon swap curves for each currency are esti-
spline-based methods to estimate zero-coupon rates.16 mated by the cubic B-splines method using LIBOR rates
LIBOR/swap zero-coupon curves. The commercial up to 1 year and swap rates from 2 to 10 years.
bank liability zero-coupon rates are estimated from
LIBOR deposit and swap rates. Unlike the government Methodology
data, the series are computed using the same method-
ology and span the same maturity spectrum, namely, 3 Zero-coupon curves are generally estimated from
months to 10 years at 3-monthly intervals. However, the observed bond prices in order to obtain an undistorted
starting dates of the series vary by country. The data for estimate of a specific term structure. The approaches com-
the United States are available from July 1987 to December monly used to fit the term structure can broadly be sep-
2003, from August 1988 for Germany, and from January arated into two categories. On the one hand, parametric
1990 for the United Kingdom.17 curves are derived from interest rate models such as the
12 RIDING THE YIELD CURVE: A VARIETY OF STRATEGIES SEPTEMBER 2005
EXHIBIT 3
Evolution of Zero-Coupon Yield Curve Slopes
Government Slopes LIBOR/Swaps Slopes
300 350

250 300

250
200

200
150
Basis Points

Basis Points
150
100
US 100
50
50

0
0

-50 -50
12-3 months 12-3 months
10-2 years 10-2 years
-100 -100
1982 1984 1986 1988 1990 1992 1994 1996 1998 2000 2002 1987 1989 1991 1993 1995 1997 1999 2001 2003

300 300

200 250

200
100

150
0
Basis Points

Basis Points
100
-100
UK 50
-200
0

-300
-50

-400 -100
12-3 months 12-3 months
10-2 years 10-2 years
-500 -150

Draft
1979 1981 1983 1985 1987 1989 1991 1993 1995 1997 1999 2001 2003 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003

300 300

250
200
200

100 150
Basis Points

Basis Points

100
0
Ger 50

-100 0

-50
-200
-100
12-3 months
10-2 years 10-2 years
-300 -150
1973 1976 1979 1982 1985 1988 1991 1994 1997 2000 2003 1988 1990 1992 1994 1996 1998 2000 2002

Vasicek term structure model; on the other hand, non- could significantly increase the risk-adjusted returns of
parametric curves are curve-fitting models such as spline- various riding strategies. The relative merits of these deci-
based and Nelson-Siegel-type models.18 The two types sion-making rules are evaluated by reporting the ex post
of non-parametric estimation techniques (Svensson and excess returns from riding down the yield curve, condi-
spline-based method) relevant for the data set used in this tional on the rule sending a positive signal. Risk-adjusted
article are described in more detail in Appendix 2. excess returns are expressed as Sharpe ratios in order to
compare and rank different riding strategies.
PRACTICAL IMPLEMENTATION Before describing the individual decision-making rules
in more detail, we present a brief overview of the literature
Most empirical studies on the term structure of describing the main factors that affect the yield curve.
interest rates find that the data generally offer little sup-
port for the REHTS. Our results are in line with these Determinants of the Term Structure
findings and suggest that market participants may be able of Interest Rates
to exploit violations of the REHTS. Although there is
some evidence that riding the yield curve per se may pro- For many years, researchers in both macroeconomics
duce excess returns compared with buying and holding, and finance have extensively studied the term structure
we suggest that using a variety of decision-making rules of interest rates. Yet despite this common interest, the
SEPTEMBER 2005 THE JOURNAL OF FIXED INCOME 13
EXHIBIT 4
Evolution of TED and Swap Spreads

12-Month TED Spreads

160
USD Rates
EUR Rates
140
GBP Rates

120

100
Basis Points

80

60

40

20

-20
1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003

10-Year Swap Spreads

250
USD Rates
EUR Rates
GBP Rates
200

Draft
150
Basis Points

100

50

-50
1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003

two disciplines remain remarkably far removed in their determinants of the latent factors identified by empirical
analysis of what makes the yield curve move. The building studies. In their pioneering work, Ang and Piazzesi [2003]
blocks of the dynamic asset-pricing approach in finance develop a no-arbitrage model of the term structure that
are affine models of latent (unobservable) factors with a incorporates measures of inflation and macroeconomic
no-arbitrage restriction. These models are purely statis- activity in addition to the traditional latent factors—level,
tical and provide very little in the way of explaining the slope, and curvature. They find that including the two
nature and determination of these latent factors.19 The macroeconomic factors improves the model’s ability to
factors are commonly referred to as “level,’’ “slope,’’ and forecast the dynamics of the yield curve. Compared with
“curvature’’ (Litterman and Scheinkman [1991]) and a traditional latent factor models, the level factor remains
wide range of empirical studies agree that almost all move- almost unchanged when macro factors are incorporated,
ments in the term structure of default-free interest rates but a significant proportion of the slope and curvature fac-
are captured by these three factors. In contrast, the macro- tors are attributed to the macro factors. However, the
economic literature still relies on the expectations hypoth- effects are limited as the macro factors primarily explain
esis of the term structure, despite overwhelming evidence movements at the short end of the curve (in particular
of variable term premia. inflation), whereas the latent factors continue to account
A handful of recent studies have started to connect for most of the movement for medium to long maturities.20
these two approaches by exploring the macroeconomic Evans and Marshall [2002] analyze the same problem
14 RIDING THE YIELD CURVE: A VARIETY OF STRATEGIES SEPTEMBER 2005
using a different, VAR-based approach. They formulate implement and, with increasing access to real-time data,
several VARs and examine the impulses of the latent fac- are often already implemented in many standard software
tors to a broad range of macroeconomic shocks. Although packages. We consider the following three simple rules.
they confirm Ang and Piazzesi’s results that most of the Positive slope. In the simplest of all cases, assuming
variability of short- and medium-term yields is driven by relatively stable interest rates over the holding horizon,
macro factors, they also find that such observable factors a positive slope is a sufficient condition for riding the
explain much of the movement in long-term yields and yield curve. We define the slope of the term structure as
that they have a substantial and persistent impact on the the yield differential between 10-year and 2-year rates
level of the term structure. and implement a riding strategy whenever this slope is
Wu [2001, 2003] examines the empirical relation- non-zero.
ship between the slope factor of the term structure and Positive cushion. The cushion, or break-even rate
exogenous monetary policy shocks in the United States change, is a slightly more sophisticated measure of the rel-
after 1982 in a VAR setting. He finds that there is a strong ative riskiness of a given riding strategy. As discussed above,
correlation between the slope factor and monetary policy the cushion indicates by how much interest rates have to
shocks. In particular, his results indicate that such shocks change over the holding horizon before the riding trade
explain 80-90% of the variability of the slope factor. begins to be unprofitable. A positive cushion indicates that
Although the influence is short lived, this provides strong interest rates have scope to increase without the trade incur-
evidence in support of the conjecture by Knez et al. [1994] ring a negative excess return. With this filter rule, we imple-
on the relation between the slope factor and Federal ment a riding strategy whenever the cushion is positive.
Reserve policy.21 75%ile cushion. In most instances, the absolute basis-
Most recently, Rudebusch and Wu [2003] have point size of the cushion will have an influence on the

Draft
extended this research into the macroeconomic deter- profitability of the riding strategy, since for a given level
minants of the yield curve by incorporating a latent factor of interest rate volatility, a small positive cushion may not
affine term structure model into an estimated structural offer sufficient protection compared with a large one.
New Keynesian model of inflation, the output gap, and Assuming the cushion itself is normally distributed around
the federal funds rate. They find that the level factor is a zero mean, we compute the realized distribution of the
highly correlated with long-run inflation expectations cushion over a two-year interval prior to the date on
and the slope factor is closely associated with changes of which a riding trade is put on. A riding strategy is imple-
the federal funds rate. mented whenever the cushion lies outside its two-year
Changes in the yield curve ultimately determine the moving 75%ile.
relative success of riding the yield curve vis-à-vis buying
and holding. Any filter rule that aims to improve the per- Macro-Based Rules: Monetary
formance of riding strategies must therefore somehow be Policy and Riding
conditioned on various (ex ante) measures of changes of
the term structure of interest rates. In this context, we are In order to translate the link between the steepness
examining the performance of two broad categories of deci- of the yield curve and monetary policy into potentially
sion-making rules, namely, statistical and macro-based rules. profitable riding strategies, we need to formulate a tractable
A given rule is said to send a positive signal if a certain trigger model of the interest rate policy followed by the central
point has been reached by the observable variable(s), the bank, such as the Taylor rule.
behavior of which is modeled by the rule. The approach of a simple model of the Federal
Reserve’s behavior was first suggested by Mankiw and
Statistical Filter Rules Miron [1986], who found that the REHTS was more
consistent with data prior to the founding of the Federal
Statistical filters are a well-established relative value Reserve in 1913. This strand of literature argues that there
tool among market practitioners. The main motivation is a link between the Federal Reserve’s use of a fund rate
for using this type of rule is the belief that many financial target instrument and the apparent failure of the REHTS.22
variables have mean-reverting properties, at least in the Rather than developing an elaborate model of term premia
short to medium term. In addition, such rules owe much coupled with Federal Reserve behavior, our approach
of their current popularity to the fact that they are easy to takes the well-established Taylor rule (J. Taylor [1993]) as

SEPTEMBER 2005 THE JOURNAL OF FIXED INCOME 15

EXHIBIT 5 conditioning the riding strategies
The Fed Funds Rate and the Slope of the Yield Curve on lagged “real-time” recessions
rather than “look-ahead” reces-
The impact of a change in the Fed funds rate on the slope of the term structure is assessed by regressing the
changes in the 10–2-year yield differential ( Slope) on the changes in the Fed funds target rate ( FFTR). sions.24
2
Estimates are multiplied by a factor of 10 for ease of interpretation. Standard errors appear below the coefficient
estimates in parentheses and are corrected for serial correlation and heteroskedasticity using Newey et al. [1994].
The slope of the yield curve
Asterisks *,** indicate significance at the 90% and 95% level (two-sided test). All variables are stationary and the Taylor rule. In this section,
according to augmented Dickey-Fuller unit root tests.
we examine how we can effectively
employ a simple Taylor rule to pre-
Slopet = 0 + 1 FFTR t + t dict future changes in the term
Constant –0.21 structure of interest rates from
0
(0.11) changes in the federal funds rate.
FFTRt –25.35** As a first step, we verify that there
1
(0.43)
is a significant link between changes
Sample period 1982:02–2003:12
No. obs. 263 in the slope of the yield curve (i.e.,
Adjusted R
2 0.1715 the degree by which the yield
Durbin-Watson 2.09
curve changes its slope over time)
and changes in the short-term
a model for central bank behavior and tests for its predictive interest rates, as suggested earlier.
power for excess returns by indicating changes in the slope A first visual inspection of slope changes and target
of the yield curve. In a second approach, we do not model rate changes displayed in panel 3 of Exhibit 8 appears to
the Federal Reserve’s behavior explicitly; instead, we support such a linkage. By regressing changes in the fed-

Draft
extract the market’s expectations of future policy action eral funds target on changes of the slope of the yield curve,
from the federal funds futures market. Before looking at we are able to confirm that there exists a significant neg-
these more elaborate macro rules, we define a simple rule ative relationship between the two variables (see Exhibit
based on a straightforward measure of economic activity. 5 for the results). Indeed, our results indicate that for every
The slope of the yield curve and recessions. Recessions 100 basis points increase in the Fed funds rate, there is a
are often associated with a comparatively steep term struc- corresponding 25 basis point flattening of the term struc-
ture. As inflationary pressures are limited during such periods ture as measured by the 10-2-year yield differential.
of reduced economic activity, central banks are generally We now link central bank behavior with changes
lowering their policy rates in order to stimulate the economy. in the slope of the yield curve by following Taylor’s orig-
We define a riding strategy that engages in trades inal specification, which relates the federal funds target
whenever the economy has entered into a recessionary rate to the inflation rate and the output gap as follows:
period. We use different definitions for recessions, depend-
ing on the country in question. For the United States, ( )
itTR = p t + r * + 0.5 p t − p t* + 0.5 yt (22)
recessions are defined according to the NBER’s Business
Cycle Dating Committee methodology, whereby “a reces- where
sion is a significant decline in economic activity spread
across the economy, lasting more than a few months, nor- itTR = federal funds rate recommended by the
mally visible in real GDP, real income, employment, indus- Taylor rule;
trial production, and wholesale-retail sales.”23 For the r∗ = equilibrium real federal funds rate;
United Kingdom and Germany, recessions are defined in π = average inflation rate over current and prior
terms of at least two consecutive quarters during which three quarters (GDP deflator);
real (seasonally adjusted) GDP is declining. π∗ = target inflation rate;
Using recessions as a trigger to ride the yield curve— y = output gap (100 × (real GDP – potential
while theoretically very appealing—suffers from a prac- GDP)/potential GDP).
tical drawback: agents do not know in real time when a
recession begins and ends due to the reporting lag of One of the main criticisms of this specification is
macroeconomic data. This problem may be addressed by that Taylor did not econometrically estimate this equa-
tion, but assumed that the Fed attached fixed weights of

16 RIDING THE YIELD CURVE: A VARIETY OF STRATEGIES SEPTEMBER 2005

0.5 to deviations of both inflation and output.25 An addi- of the target rate, we are interested in predicting target rate
tional problem with Taylor’s original work is that the changes by employing the Taylor rule. For this purpose, we
output gap is estimated in-sample. This shortcoming can regress the actual changes in the federal funds target
be addressed by estimating the Taylor rule out-of-sample ∆FFTRt on changes of the target rate, as recommended
with no look-ahead bias (see panel 2 of Exhibit 6).26 by the Taylor rule ∆Taylort as opposed to the difference
As a response to the criticism that the weights on between the target rate estimate and the actual rate.27 In
inflation and the output gap in Equation (22) are not esti- order to see whether the Taylor signal is particularly pre-
mated, we also consider a dynamic version of the Taylor dictive prior to an interest rate decision, we add a dummy
Rule, following the work of Judd and Rudebusch [1998]. variable FOMCt that has a value only in the month prior
In this specification, Equation (22) is restated as an error- to an FOMC meeting.
correction mechanism that allows for the possibility that The results of these regressions are summarized in
the federal funds rate adjusts gradually to achieve the rate Exhibit 7. For both versions of the Taylor rule (the out-
recommended by the rule. In particular, by adding a lagged of-sample estimation of the original specification and the
output gap term along with the contemporaneous gap, dynamically estimated Judd-Rudebusch version), there is
Equation (22) is replaced with strong significance on the predictive power of the Taylor
rule with regards to target rate changes over the entire
π t TR = π t + r * + λ1 π t − π t* + λ2 yt + λ3 yt −1
( ) (23) sample period (1988-2003).28 In addition, the respon-
siveness of rate changes with respect to the Taylor rule
The dynamics of adjustment of the actual level of increases by almost 20% before FOMC meetings. This is
the federal funds rate to the recommended rate, itTR, are indicated by the increase in the parameter estimates of
given by regressions 2 and 4 in Exhibit 7. Nonetheless, the estimates
for φ are significantly smaller than unity, suggesting that

Draft
( )
Dit = g itTR − it −1 + rDit −1 (24) the recommended rate needs to change by between 120
and 150 basis points to signal a full quarter percent change
This means that the change in the funds rate at time in the actual target rate.29
t partially corrects the difference between the last period Having established a relatively firm link between
and the current target level as well as displaying some the Taylor rule and changes in the slope of the term struc-
dependency on the funds rate change at time t – 1. By ture, we can devise a simple signal for whether to ride
substituting Equation (23) into Equation (24), we obtain and compare it with alternative strategies. At every month
the full ECM to be estimated: end, we estimate itTR by re-estimating yt and πt. The change
in the “equilibrium” federal funds target rate suggested
Dit = ga − g it −1 + g (1 + l1 ) p t + gl2 yt + gl3 yt −1 + rDit −1 (25) by the Taylor rule ∆itTR is then used as the basis for a
simple decision rule:
where α = r ∗ – λ1π∗. This equation provides estimates of
policy weights on inflation and output and on the speed • If ∆itTR > 0, then riding the yield curve is less
of adjustment to the rule. Judging by the plot of our Judd- favorable as there is a strong likelihood that short
Rudebusch estimates of the Taylor rule alone (see panel 3 rates will increase.
of Exhibit 6), it is difficult to conclude whether we are • If ∆itTR < 0 then riding the yield curve is more
able to obtain an improved forecast of the federal funds favorable as there is a strong likelihood that short
rate, compared to the two static methods. rates will decrease.
In order to determine whether the Taylor rule is
a useful means for devising different riding strategies, In order to translate this decision-making rule into
we need to see whether the Taylor rule at time t – 1 can a signal that indicates whether to ride the yield curve, we
predict changes in the federal funds rate at time t. If this construct a variable TaylorSignalt that takes a value of 1
is indeed the case, we can use the Taylor rule as a signal (or –1) whenever the relevant specification of the Taylor
to determine when to ride the yield curve, since we rule indicates a rate rise (cut) and is 0 otherwise. We
have already established that the target rate can predict employ a riding strategy whenever the signal is different
slope changes. from 1 and therefore does not indicate an impending
Rather than determination of the equilibrium level increase in the target rate.

SEPTEMBER 2005 THE JOURNAL OF FIXED INCOME 17

EXHIBIT 6
Modeling the Fed Funds Rate Using the Taylor Rule

Panel 1: In-Sample Estimation Panel 3: Judd-Rudebusch Estimation

10.0 10.0

7.5 7.5
Rate (%)

Rate (%)
5.0 5.0

2.5 2.5

Fed Funds Rate Fed Funds Rate

Taylor Rule Taylor Rule
0.0 0.0
1987 1989 1991 1993 1995 1997 1999 2001 2003 1987 1989 1991 1993 1995 1997 1999 2001 2003

Panel 2: Out-of-Sample Estimation Panel 4: Rate Changes and the Taylor Rule Signal
10.0 0.75

0.50

Draft
7.5
0.25

0.00
Change (%)
Rate (%)

5.0

-0.25

-0.50
2.5

-0.75

Fed Funds Rate Target

Taylor Rule Signal
0.0 -1.00
1987 1989 1991 1993 1995 1997 1999 2001 2003 1987 1989 1991 1993 1995 1997 1999 2001 2003

The slope of the yield curve and expectations from fed futures at time t – 1 are a reliable predictor of movements
funds futures. In theory, federal funds futures should reflect in the yield curve (via implied target rate changes) at time
market expectations of near-term movements in the (effec- t. Should this indeed be the case, we would be able to
tive) fed funds rate and thus the target rate. A growing construct an additional decision-making rule for riding
strand of literature has demonstrated the usefulness of fed the yield curve. Thus, if market expectations implied by
funds futures contracts in predicting monetary policy moves the futures contracts can be used to forecast the changes
one to three months ahead. In particular, using daily data in the federal funds target, we can construct an additional
Söderström [2001] shows that futures-based proxies for decision rule for riding the yield curve. As before, we
market expectations are a successful predictor of the target compare the equilibrium rate implied by the futures con-
rate around target changes and FOMC meetings. In line tracts itExp with the observed rate itActual. This forms the basis
with this literature, this section investigates the relation- for a simple decision rule along the following lines:
ship between market expectations from federal funds futures
and changes in the slope of the yield curve as triggered by • If itExp > itActual, then riding the yield curve is less
changes in the target rate. As with the Taylor rule in the favorable as there is a strong likelihood that short
previous section, we want to see whether the federal fund rates will increase.

18 RIDING THE YIELD CURVE: A VARIETY OF STRATEGIES SEPTEMBER 2005

EXHIBIT 7 meeting and expected post-meeting
The Federal Funds Rate and the Taylor Rule target rate.30 This can be expressed as
In order to assess the predictive power of the Taylor rule with regard to changes in the Fed funds rate, actual
target rate changes ( FFTR) are regressed on rate changes implied by the Taylor rule ( Taylor). Assuming no d1 d
itf = iipre +  pi post + (1 − p) iipre  2 (26)
B  t
inter-meeting rate changes, the dummy variable FOMC tests whether the relationship is particularly strong prior to
a potential target rate decision. Thus, FOMC only has a value in the month prior to an FOMC meeting, when it is B
equal to Taylor. Estimates are multiplied by a factor of 10 2 for ease of interpretation. Standard errors appear
below the coefficient estimates in parentheses and are corrected for serial correlation and heteroskedasticity
using Newey et al. [1994]. Asterisks *,** indicate significance at the 90% and 95% level (two-sided test). All
where i f = futures rate implied by rel-
variables are stationary according to augmented Dickey-Fuller unit root tests. evant contract;31 i pre = target rate pre-
vailing before the FOMC meeting; i post
FFT R t = 0
+ 1,3
aylort( Dynamic)
1
+ 2
FO Ct 1 + t
= target rate expected to prevail after
(1) (2) (3) (4) the FOMC meeting; p = probability
Constant –2.98* –2.99 –2.56 –2.50
of a target rate change; d1 = number of
0
(1.84) (1.83) (1.77) (1.74) days between previous month end and
Taylort-1 15.39** 0.02
1 FOMC meeting; d2 = number of days
(7.23) (2.05)
FOMCt-1 2 21.29** 31.34** between FOMC meeting and current
(9.79) (10.44) month end; B = number of days in
TaylortDynamic 3 28.92** 2.67
1 (8.32) (3.83) month.
Sample period 88:04–03:12 88:04–03:12 88:04–03:12 88:04–03:12
No. obs. 189 189 190 190
Solving Equation (26) for p, the
Adjusted R
2 1.53 1.80 6.79 7.68 probability of a change in the target
Durbin-Watson 1.29 1.31 1.34 1.35 rate can thus be expressed as

d d 

Draft
• If itExp < itActual then riding the yield curve is more itf − itpre  1 + 2 
 B B
favorable as there is a strong likelihood that short p= (27)
rates will decrease. ( d
itpost − itpre 2
B
)
In addition to assuming no inter-meeting changes,
A first visual inspection of plotting the target rate this specification also assumes that the Fed has only two
against the rate implied by the nearest futures contract policy options: either shift the target rate by a prespeci-
(see panel 1 in Exhibit 8) strongly suggests that market par- fied amount or leave it unchanged. For ease of compu-
ticipants indeed do “get it right.” In order to gauge the tation, we can reasonably assume that this amount is
predictive power of futures-based expectations, we test (multiples of ) 25 basis points, since the Fed has not
whether target rate changes can be forecast given the changed rates by any other amount since August 1989.
implied probability of a rate change has passed a certain If market expectations indeed provide useful infor-
threshold (i.e. 50%). mation with regards to riding the yield curve, we need
In order to translate this hypothesis into a trading to test whether market expectations are a good indicator
signal, we start by computing the implied probabilities of of future changes in the federal funds target rate. For this
a change in the federal funds target rate. Futures-based purpose, we construct the variable MarketSignalt, which
expectations before an FOMC meeting can be interpreted has a non-zero value whenever the implied probability
as a meaningful measure of the target rate expected to of a rate rise (cut) is greater than 50%.31 In line with the
prevail after the meeting only if the target rate is not previous section, we employ the dummy variable FOMCt
changed between meetings and never twice in the same to assess whether the predictive power of federal funds
month. Although federal funds futures were first intro- futures is particularly high prior to an FOMC meeting.
duced at the Chicago Board of Trade in October 1988, Our results on the informative content of futures
it was not until 1994 that the FOMC began announcing with regards to target rate indicate that Fed funds futures
changes in its policy stance and abandoned inter-meeting are indeed a useful means of predicting target rate changes,
rate changes (see Chicago Board of Trade [2003]). For using both daily and end-of-the-month monthly data. This
this reason, we do not consider any observations prior to is broadly in line with the existing literature (e.g. Rude-
that date and define the rate implied by the fed funds busch [1995]; Söderström [2001]). Regressing daily and
futures contract as a time-weighted average of a pre- monthly changes in the target rate on the market signal

SEPTEMBER 2005 THE JOURNAL OF FIXED INCOME 19

EXHIBIT 8
Fed Fund Futures, Market Expectations, and Slope Changes
Panel 1: Target and Futures (Near Contract) Panel 3: Slope and Fed Funds Rate Changes
10.0 1.5

1.0

7.5
0.5

0.0

Change (%)
Rate (%)

5.0

-0.5

-1.0
2.5

-1.5

Target Target
Futures ChgSlope
0.0 -2.0
1987 1989 1991 1993 1995 1997 1999 2001 2003 1982 1984 1986 1988 1990 1992 1994 1996 1998 2000 2002

Panel 2: Target Rate Changes and Expectations

1.00
Target
Signal

0.75

Draft
0.50

0.25
Change (%)

0.00

-0.25

-0.50

-0.75

-1.00
1987 1989 1991 1993 1995 1997 1999 2001 2003

indicates that whenever the market thinks that there is at presented in the previous section. In addition, we present
least a 50% chance of a 25 basis point cut (rise), the target a simple framework that allows investors who are bound
rate indeed decreases (increases) subsequently. As regressions by more conservative investment guidelines to exploit the
2 and 3 in Exhibit 9 indicate, this signal is particularly strong concept of riding the yield curve without incurring a sub-
in the period immediately prior to an FOMC meeting stantial amount of additional interest rate risk.
Thus, using futures closing prices before an FOMC
meeting, we are able to reliably anticipate the FOMC deci- Government Securities
sion. The robustness of this result can also be seen visually
by plotting the changes in the target rate against the signal With a few exceptions, the riding strategies using
from market expectations in panel 2 of Exhibit 8. government securities display superior performance com-
pared with buying and holding across all holding horizons
EMPIRICAL RESULTS and all currencies. In contrast to previous empirical evi-
dence, our results provide surprisingly strong evidence for
This section reports empirical findings for the various the existence of exploitable risk premia in these markets.
riding strategies across instruments and currencies and In general, our results indicate that the excess returns
reviews the effectiveness of the different conditioning rules from riding increase with the maturity of the riding instru-

20 RIDING THE YIELD CURVE: A VARIETY OF STRATEGIES SEPTEMBER 2005

EXHIBIT 9 target rate, and Treasury auctions.
The Federal Funds Rate and Market Expectations For U.K. Gilts, the riding returns
Fed funds futures contracts provide a useful tool for measuring market participants’ expectations with respect to
increase with both the maturity of the
target rate changes. The accuracy of these expectations is gauged by regressing actual changes ( FFTR) on a riding instrument and the length of
conditional measure of expected changes. The variable MarketSignal serves as such a measure and is non-zero
whenever the implied probability a target rate change exceeds 50% (i.e., the signal strength is positive and the holding horizon. The mean riding
increases as the implied probability of a rate rise exceeds 50%; negative as the probability of a cut exceeds 50%;
and 0 otherwise). As in Exhibit 7, the dummy variable FOMC tests whether the relationship is particularly strong
returns are approximately at the same
prior to a potential target rate decision and is equal to MarketSignal before an FOMC meeting. Estimates are levels as those for the dollar market,
multiplied by a factor of 102 for ease of interpretation. Standard errors appear below the coefficient estimates in
parentheses and are corrected for serial correlation and heteroskedasticity using Newey et al. [1994]. Asterisks whereas mean excess returns are on
*,** indicate significance at the 90% and 95% level (two-sided test). All variables are stationary according to
augmented Dickey-Fuller unit root tests.
average only about half those achieved
with U.S. Treasuries. The highest and
simultaneously least volatile excess
FFTRt = 0 + 1 MarketSignalt-1 + 2 FOMCt-1 + t
(1) (2) (3) returns of 3.7% arise from riding the
longest-dated Gilts for the 18-month
Constant 0 –3.66* –2.87 –0.030
(2.01) (1.96) (0.076) holding horizon. However, at the
MarketSignalt-1 1 8.99** 0.97 0.001 other end of the scale, riding the yield
(2.25) (0.72) (0.000)
FOMCt-1 2 10.90** 1.981** curve with U.K. T-bills for short hori-
(3.03) (0.812) zons does worse than holding to matu-
Sample period 94:01–03:12 94:01–03:12 94:01:03–03:12:31
No. obs. 116 116 2480 rity. This may indeed be related to the
Adjusted R
2 21.01 26.91 21.51 fact that the money-market sector of
Durbin-Watson 1.81 1.72 2.00
the Gilt curve is sparsely populated and
T-bills tend to be relatively illiquid

Draft
ment. This is very much in line with the results of other instruments.
studies, such as Dyl and Joehnk [1981] and Grieves et al. The results for German government paper are broadly
[1999], and is a direct consequence of the increased risk- in line with those for U.S. Treasuries, where returns
return trade-off for longer-maturity instruments. Although increase with the maturity of the riding instruments but
riding with longer-dated instruments increases excess decrease with the holding horizon. Similarly, riding the 2-
returns, these strategies tend to do slightly worse on a year Federal Treasury notes (referred to as “Schätze”) is the
risk-adjusted basis because of the increased interest rate most effective strategy on a risk-adjusted basis across holding
risk across all currencies. periods. The mean riding returns are lower than those for
For U.S. Treasuries, excess riding returns are the both Treasuries and Gilts and the maximum mean excess
highest across all instruments for the shortest, 3-month returns of 3.7% are obtained from riding 10-year paper,
holding horizon. Riding the yield curve with a 10-year the so-called Bunds, for 12 and 18 months. Because there
Treasury for 3 months produces an annualized average is no continuous spectrum of on-the-run German T-bills,
return of 12.0%, which is 6.2% in excess of the corre- we are unable to compute any riding strategies with a
sponding buy-and-hold strategy. Riding for 6 months with holding horizon of less than 12 months.32
a 12-month instrument yields the lowest excess mean
return of only 44 basis points. This contrasts with the find- LIBOR/Swaps
ings of Dyl and Joehnk, however, who observe that the
riding returns increase uniformly with the holding horizon. The riding returns and excess returns from using
With the exception of riding 6-month T-bills for 3 months, commercial bank liabilities, namely, LIBOR deposits and
the most efficient rides are consistently performed with 2- swaps, are largely similar to those from using government
year instruments, independent of the holding horizon. instruments.33 As before, riding returns generally tend to
This corresponds to the well-documented fact that this increase with the maturity of the instrument and the
sector of the U.S. Treasury yield curve offers the highest holding horizon. This is not true for dollar and pound
risk premia because it shows the biggest yield volatility sterling excess returns, where the largest return pickups
historically. According to Fleming and Remolona [1999a, are achieved by riding long-maturity instruments at shorter
1999b], U.S. Treasury securities in the 2-year sector of holding horizons.
the yield curve show the strongest responses to macro- Riding a 10-year USD swap for 2 years yields 12.9%
economic announcements, changes in the federal fund per annum, the highest mean riding returns for dollar

SEPTEMBER 2005 THE JOURNAL OF FIXED INCOME 21

EXHIBIT 10 instruments for short holding hori-
U.S. Treasuries: HPR Statistics for Different Riding Strategies zons is the least attractive strategy,
with riding a 6-month deposit for 3
The table summarizes returns and excess returns for different riding strategies across selected horizons. The first
column lists the maturity of the riding instrument m. In order to compute the holding period returns (HPRs) of months offering no excess returns.
riding an m-maturity instrument for h months, the (m - h) rate must also be available. XHPR represents the excess
riding returns over the buy-and-hold strategy. S.R. is the Sharpe ratio of the excess returns. Returns and standard Unlike for government paper, how-
deviations are annualized for ease of comparison. The standard deviations of the various mean returns were ever, none of the riding strategies does
corrected for overlapping data by using a Newey et al. [1994] correction on the standard errors of the respective
mean, where the lags are set equal to the length of the holding horizon. worse than the corresponding buy-
m m and-hold investment.
HPR H t + h (%) XHPR XH t + h (%)
Instrument Mean S.D. Min Max Mean S.D. Min Max S.R.
This is not the case for strategies
3-month with euro-denominated deposits,
6-month 6.3 2.8 0.3 5.0 0.5 0.7 –0.3 1.7 0.71
2-year 7.8 4.9 –1.5 8.4 2.0 3.9 –2.5 5.2 0.52 where money-market rides over a 3-
5-year 9.8 10.1 –5.5 14.8 4.0 9.5 –7.0 12.1 0.42
7-year 10.8 13.2 –7.6 20.0 5.0 12.7 –10.0 17.4 0.39 month period either offer no return
10-year 12.0 17.9 –116 27.4 6.2 17.4 –13.9 24.7 0.35 enhancement or do worse than
6-month matching maturity and investment
1-year 6.5 5.1 0.6 9.3 0.4 1.6 –1.1 2.3 0.28
2-year 7.7 7.4 –0.3 13.9 1.7 4.8 –2.4 6.8 0.35 horizon. In addition, euro credit rides
5-year 9.9 14.3 –4.7 24.4 3.8 12.7 –7.8 17.3 0.30
7-year 10.9 18.5 –8.1 29.5 4.8 17.1 –11.3 22.4 0.28 show slightly lower mean returns than
10-year 12.1 25.1 –12.7 37.6 6.0 23.8 –175 31.0 0.25 government rides (10.1% versus 10.0%
12-month for riding the respective 10-year instru-
2-year 7.6 10.4 1.2 18.7 1.2 4.7 –2.4 56 0.25
5-year 9.9 18.9 –4.2 29.5 3.5 15.7 –8.9 17.8 0.22 ment for 2 years), whereas mean excess
7-year 10.9 24.9 –7.2 39.6 4.5 22.2 –14.1 30.1 0.20
10-year 12.2 33.4 –13.3 54.4 5.8 31.3 –224 45.0 0.18 returns are on average only marginally
18-month
higher than for the risk-free rides. This

Draft
2-year 7.4 12.5 4.4 23.7 0.5 2.8 –1.1 3.10 0.19 follows directly from the historical
5-year 9.8 22.0 –1.2 40.9 3.0 16.0 –8.5 22.1 0.19
7-year 10.9 28.5 –3.5 52.2 4.1 23.3 –14.9 35.4 0.18 behavior of euro deposit and swap
10-year 12.2 37.3 –108 69.7 5.4 33.0 –24.7 51.8 0.16
spreads, which display high levels of
24-month
5-year 9.8 25.3 4.4 51.1 2.6 15.0 –6.6 24.6 0.17
volatility throughout the entire sample
7-year 10.9 31.8 1.6 67.1 3.7 22.3 –9.6 40.6 0.17 period, despite their very low levels.
10-year 12.2 40.1 –1.9 95.3 5.0 31.5 –12.0 68.9 0.16
Despite the fact that the euro swaps
market has a higher notional amount
outstanding than any other currency,34
instruments. This is a mere 70 basis points more than the the absence of any significant swap spreads suggests that
same riding strategy using Treasuries instead. The highest eurozone credit is more expensive than credit elsewhere.
excess returns (6.6% p.a.) are obtained by riding the same This phenomenon, sometimes referred to as the “euro
maturity instrument, but over only a 3-month horizon. credit puzzle,” is illustrated in Exhibit 4.
As with Treasuries, shorter holding horizons perform best
on a risk-adjusted basis, and the 2-year maturity bucket
Conditioned Riding
offers the most attractive reward-to-variability ratios. The
strategy of riding a 2-year dollar swap for 3 months has a This section reports the results from applying a
Sharpe ratio of 0.54, the highest ratio across all credit variety of statistical and macro-based decision-making
strategies. Riding only 6-month U.S. T-bills over the same rules to the different riding strategies. Overall we find
horizon offers a superior risk-adjusted profit with a Sharpe strong evidence that the excess returns of a large number
ratio of 0.71. of riding strategies can be enhanced significantly by relying
Sterling mean riding returns are consistently higher on these rules. This in itself points to the existence of siz-
than the ones for U.S. dollars and peak at 13.0% for able risk premia that can be exploited successfully.
riding a 10-year swap for both 18 months and 2 years. Positive slope. This simplest of ex ante filtering
Mean excess returns are at similar levels as the ones in mechanism produces mixed results at improving mean
dollars, albeit marginally more volatile, which stands in excess riding returns across most of the instruments,
stark contrast to riding government instruments, where holding horizons, and currencies. Generally, the amount
sterling excess returns were only half the size of dollar by which the excess returns rise tends to be highest for
returns. Riding the yield curve with short-maturity the shortest available holding horizons.

22 RIDING THE YIELD CURVE: A VARIETY OF STRATEGIES SEPTEMBER 2005

EXHIBIT 11 For rides with either U.S. Trea-
U.K. Gilts: HPR Statistics for Different Riding Strategies suries or German Bunds, a positive
The table summarizes returns and excess returns for different riding strategies across selected horizons. The first
slope is not able to improve the excess
column lists the maturity of the riding instrument m. In order to compute the holding period returns (HPRs) of returns at any horizon. This is in line
riding an m-maturity instrument for h months, the (m - h) rate must also be available. XHPR represents the excess
riding returns over the buy-and-hold strategy. S.R. is the Sharpe ratio of the excess returns. Returns and standard with the results of Grieves et al. [1992],
deviations are annualized for ease of comparison. The standard deviations of the various mean returns were
corrected for overlapping data by using a Newey et al. [1994] correction on the standard errors of the respective
whose study covers a similar sample
mean, where the lags are set equal to the length of the holding horizon. period but uses daily data. For most
other instruments, there are significant
HPR H tm+ h (%) XHPR XH tm+ h (%)
Instrument Mean S.D. Min Max Mean S.D. Min Max S.R.
excess returns at short horizons, but
3-month excess returns fall below the uncondi-
6-month 5.6 1.4 0.7 3.1 –0.1 0.3 –0.2 0.2 –0.29
9-month 7.8 2.9 0.4 4.2 –0.2 0.7 –0.8 05 –0.28 tioned riding returns for holding
1-year 8.0 3.1 0.1 4.8 –0.2 1.1 –1.1 0.8 –0.19
2-year 8.9 5.1 –1.1 7.6 0.2 2.4 –2.3 2.0 0.10 horizons beyond 1 year. Using dollar-
5-year 10.2 11.4 –8.9 17.0 0.9 6.3 –6.7 4.7 0.14
7-year 10.9 14.4 –11.6 20.3 1.4 8.2 –8.7 6.0 0.17
denominated deposits and swaps, for
10-year 11.8 18.1 –139 28.4 2.3 10.6 –10.6 7.8 0.22 example, the mean excess returns are
6-month improved by over 60 basis points, from
9-month 5.9 2.5 1.5 7.0 –0.1 0.5 –0.7 0.3 –0.21
1-year 8.0 4.8 1.2 81 –0.4 1.7 –2.4 1.8 –0.22 4.04 to 4.68% p.a. for 3-month rides.
2-year 9.0 7.2 0.1 12.0 0.2 4.5 –4.9 6.1 0.04
5-year 10.3 15.0 –8.3 23.8 1.3 12.7 –13.8 17.4 0.10
For any longer horizon, however, the
7-year 11.0 18.7 –12.4 29.5 1.9 16.3 –17.0 20.1 0.12 unconditioned returns are higher.
10-year 11.8 23.0 –172 41.8 2.6 20.3 –21.7 21.2 0.13
Euro-deposits perform even bet-
12-month
2-year 9.2 10.8 3.0 21.3 0.5 4.8 –4.2 6.8 0.11 ter, with mean excess returns improv-
5-year 10.6 18.8 –4.6 42.8 1.9 15.6 –13.8 28.3 0.12
7-year 11.4 23.6 –8.4 52.4 2.8 21.0 –17.7 37.9 0.13
ing by over 350 basis points for

Draft
10-year 12.3 30.6 –128 64.2 3.6 28.9 –23.7 49.7 0.13 3-month rides and over 30 basis points
18-month for 2-year rides. Conditioned rides
2-year 9.0 13.7 6.3 25.6 0.2 2.7 –2.8 3.1 0.09
5-year 10.8 21.8 0.9 44.9 1.8 15.2 –12.8 22.6 0.12 with sterling instruments also produce
7-year 11.7 27.4 –2.6 53.2 2.7 21.9 –18.1 30.8 0.12
10-year 12.7 36.1 –103 67.6 3.7 32.4 –25.8 44.6 0.11
higher mean excess returns for holding
24-month
horizons up to 1 year.
5-year 11.0 25.4 5.5 52.5 1.6 14.2 –12.3 21.1 0.11 Positive and 75%ile cushion. In
7-year 11.9 31.3 1.0 63.0 2.6 21.7 –18.4 31.7 0.12
10-year 13.0 40.7 –5.2 79.8 3.6 33.6 –276 47.8 0.11 quantifying how much rates have to
increase before a riding trade loses
money, it comes as no surprise that
EXHIBIT 12 using the cushion as a filter performs
better than just looking at the slope.
German Government Bonds: HPR Statistics for Different Riding Strategies
For all rides except the percentile
The table summarizes returns and excess returns for different riding strategies across selected horizons. The first
column lists the maturity of the riding instrument m. In order to compute the holding period returns (HPRs) of cushion in the case of sterling credit
riding an m-maturity instrument for h months, the (m - h) rate must also be available. XHPR represents the excess instruments, both cushion-based con-
riding returns over the buy-and-hold strategy. S.R. is the Sharpe ratio of the excess returns. Returns and standard
deviations are annualized for ease of comparison. The standard deviations of the various mean returns were ditions increase mean excess returns
corrected for overlapping data by using a Newey et al. [1994] correction on the standard errors of the respective
mean, where the lags are set equal to the length of the holding horizon. significantly.
In fact, of all the filtering strate-
HPR H tm+ h (%) XHPR XH tm+ h (%)
gies presented in this article, the per-
Instrument Mean S.D Min Max Mean S.D. Min Max S.R.
12-month centile cushion is by far the most
2-year 6.7 9.6 0.9 16.0 0.8 5.1 –5.2 5.4 0.15
5-year 8.3 17.7 –5.9 23.0 2.4 16.0 –14.0 14.3 0.15
effective method to enhance riding
7-year 8.9 22.5 –9.6 26.6 3.0 21.4 –17.7 18.8 0.14 returns across all instruments and cur-
10-year 9.6 29.0 –11.9 37.1 3.7 28.5 –20.0 27.8 0.13
rencies. This is again a fairly intuitive,
18-month
5-year 8.4 22.0 –4.0 34.3 2.2 18.0 –13.6 15.2 0.12 yet powerful, result that states that the
7-year 9.1 27.7 –9.0 37.4 2.9 24.8 –18.1 23.1 0.12
10-year 9.9 35.3 –15.6 46.8 3.7 33.4 –22.7 33.0 0.11
higher the break-even interest rate
24-month
change at the beginning of the riding
5-year 8.5 25.4 –3.3 36.7 2.0 18.5 –13.4 14.9 0.11 period, the more profitable it is to ride.
7-year 9.2 32.1 –8.9 40.3 2.7 26.8 –19.3 23.5 0.10
10-year 10.1 41.1 –151 52.4 3.6 37.2 –23.6 35.4 0.10 The biggest increases are obtained
with dollar-based instruments, where

SEPTEMBER 2005 THE JOURNAL OF FIXED INCOME 23

EXHIBIT 13 exceeding its 2-year 75%ile not only
USD LIBOR/Swaps: HPR Statistics for Different Riding Strategies enhances returns in the long run but
The table summarizes returns and excess returns for different riding strategies across selected horizons. The first
also ensures the highest possible
column lists the maturity of the riding instrument m. In order to compute the holding period returns (HPRs) of number of individual trades do not
riding an m-maturity instrument for h months, the (m - h) rate must also be available. XHPR represents the excess
riding returns over the buy-and-hold strategy. S.R. is the Sharpe ratio of the excess returns. Returns and standard suffer a capital loss.
deviations are annualized for ease of comparison. The standard deviations of the various mean returns were
corrected for overlapping data by using a Newey-West [1994] correction on the standard errors of the respective Recessions. The results for using
mean, where the lags are set equal to the length of the holding horizon. a specific measurement of reduced eco-
m
HPR H t + h (%)
m
XHPR XH t + h (%)
nomic activity (i.e., a recession) are
Instrument Mean S.D Min Max Mean S.D. Min Max S.R. quite mixed and vary between curren-
3-month cies, but not instruments. As indicated
6-month 5.8 2.3 0.3 3.2 0.2 0.4 –0.3 0.4 0.45
9-month 5.9 2.3 0.3 3.5 0.3 1.0 –0.9 0.8 0.29 earlier, we use different definitions of
1-year 6.4 2.7 –0.1 4.4 0.8 1.5 –1.0 1.6 0.53
2-year 7.7 4.5 –1.8 6.4 2.1 3.9 –2.7 3.7 0.54 what constitutes a recession for dif-
5-year 9.8 9.5 –6.8 10.3 4.1 9.4 –7.6 8.6 0.44
7-year 10.9 12.2 –9.8 12.5 5.3 12.0 –10.6 10.8 0.44 ferent markets. This does not seem to
10-year 12.2 16.2 –11.9 15.2 6.6 16.0 –12.7 14.4 0.41 matter, since the definition proposed
6-month by the NBER for the U.S. market does
9-month 5.9 3.9 0.6 5.9 0.1 0.7 –0.6 0.7 0.19
1-year 6.2 4.1 0.6 6.5 0.4 1.5 –1.1 1.6 0.29 as well at improving mean excess riding
2-year 7.7 6.1 –0.5 8.7 1.9 4.9 –2.5 5.3 0.39
5-year 9.8 12.6 –5.8 16.6 4.0 12.3 –7.6 13.6 0.33 returns as the more “trivial” defini-
7-year 11.0 16.3 –8.7 19.2 5.2 16.0 –10.5 18.1 0.32 tions used for the United Kingdom
10-year 12.2 21.5 –12.1 24.8 6.5 21.2 –14.4 23.7 0.31
and Euroland.35
12-month
2-year 7.5 8.7 1.1 13.6 1.4 5.1 –2.8 4.9 0.27 For dollar-denominated assets,
5-year 10.0 17.0 –5.7 21.2 3.8 16.2 –9.5 15.2 0.24
7-year 11.2 21.3 –9.2 26.3 5.1 20.8 –12.8 19.8 0.24 riding the yield curve only during an

Draft
10-year 12.6 27.5 –14.4 33.4 6.5 27.0 –18.4 25.6 0.24 economic slump is the second most
18-month profitable of all riding strategies. For
2-year 7.3 10.1 4.9 19.1 0.7 3.2 –1.1 34 0.21
5-year 10.1 20.1 –2.2 29.5 3.5 17.2 –8.4 17.2 0.20 the shortest Treasury riding horizon,
7-year 11.4 24.5 –50 33.3 4.7 22.2 –11.9 22.8 0.21
10-year 12.8 30.1 –9.6 41.6 6.2 28.3 –171 33.0 0.22 mean excess returns are boosted from
24-month
from 3.9 to more than 9.9%, whereas
5-year 10.1 22.7 4.9 37.8 3.0 16.4 –5.5 18.8 0.19 a 1-year holding horizon for credit
7-year 11.4 27.7 2.1 45.9 4.3 22.1 –8.6 26.8 0.19
10-year 12.9 33.3 –2.8 55.3 5.8 28.4 –132 37.2 0.20 instruments augments excess returns
from 4.0 to 8.3% p.a.
Riding the sterling yield curves
mean excess returns jump from 3.8 to 12.3% p.a. for 6- during a recession is the best of all filtering rules, except
month Treasury rides and from 4.0 to 18.5% p.a. riding in the case of short investment horizons for Gilts, where it
deposits and swaps for 3 months. However, although the actually causes substantial underperformance compared
percentile is the most successful riding strategy in most with buying and holding. Recessionary riding with German
instances, it also has the drawback of sending the least assets does not work well with government paper but dis-
frequent riding signal. In addition, this strategy seems plays some return enhancement potential for credit instru-
most effective for shorter horizons, which could be related ments. In line with the results for the U.K. market, the
to the fact that after, say, 18 months the original signal excess returns are largest for the shorter holding horizons.
no longer contains much informational content. As identified earlier, these results might display a
Because excess returns surge so drastically with the simultaneity bias owing to the reporting lag associated with
percentile cushion as a filter, the proportion of individual recession (cf. note 24). However, some preliminary com-
trades with negative returns falls accordingly. This is illus- putations indicate that for most currencies and instru-
trated in Exhibit 18, where we see that for both dollar- ments, excess returns are underestimated rather than
and euro-denominated trades an exceptionally large overstated as a result of this.36
number of the strategies produce positive returns. This is Taylor rule. The results for riding strategies condi-
particularly welcome news for risk-averse investors, such tioned on both the traditional and the dynamically esti-
as central bank portfolio managers, who at all times are mated version of the Taylor rule are less pronounced than
bound by capital preservation constraints. In other words, for other filters but encouraging nonetheless; in particular
riding the yield curve conditional on the cushion the dynamic or Judd-Rudebusch specification of the Taylor

24 RIDING THE YIELD CURVE: A VARIETY OF STRATEGIES SEPTEMBER 2005

EXHIBIT 14 tations-based filter to ride the yield
GBP LIBOR/Swaps: HPR Statistics for Different Riding Strategies curve, however, our empirical results
are mixed as average excess riding
The table summarizes returns and excess returns for different riding strategies across selected horizons. The first
column lists the maturity of the riding instrument m. In order to compute the holding period returns (HPRs) of returns cannot be increased across all
riding an m-maturity instrument for h months, the ( m - h) rate must also be available. XHPR represents the excess
riding returns over the buy-and-hold strategy. S.R. is the Sharpe ratio of the excess returns. Returns and standard holding horizons.
deviations are annualized for ease of comparison. The standard deviations of the various mean returns were The strategy works well at the 3-
corrected for overlapping data by using a Newey et al. [1994] correction on the standard errors of the respective
mean, where the lags are set equal to the length of the holding horizon. month and 6-month holding horizons,
m m
roughly increasing excess returns in the
HPR H t + h (%) XHPR XH t + h (%)
Instrument Mean S.D. Min Max Mean S.D. Min Max S.R.
same order of magnitude as the Taylor
3-month rule for the same horizons. Excess
6-month 6.7 2.2 0.7 3.6 0.0 0.4 –0.3 0.8 0.00
9-month 6.8 2.3 0.5 4.4 0.2 0.8 –0.6 1.8 0.25 returns can be pushed up by close to 50
1-year 7.0 2.6 0.3 5.4 0.4 1.3 –1.0 2.7 0.31
2-year 8.0 4.3 –0.8 8.0 1.4 3.2 –2.1 5.3 0.42
basis points from 2.6 to 3.1% (or
5-year 9.7 9.0 –7.3 12.9 3.1 8.1 –8.7 10.3 0.38 +18.5%) over a 3-month period and
7-year 10.8 11.3 –10.7 13.6 4.2 10.6 –12.1 11.0 0.39
10-year 12.4 14.5 –15.0 17.4 5.8 13.8 –16.4 15.9 0.42 increase by 30 basis points over a 6-
6-month month horizon. For these horizons,
9-month 6.8 3.8 1.6 6.7 0.1 0.5 –0.4 1.0 0.11
1-year 7.0 4.0 1.3 76 0.3 1.2 –0.7 2.1 0.22 expectations-based riding also repre-
2-year 8.0 6.2 0.1 11.9 1.3 4.0 –2.7 6.5 0.32 sents a superior strategy on a risk-
5-year 9.9 12.5 –8.5 19.4 3.1 10.6 –11.1 14.0 0.29
7-year 10.9 15.6 –13.7 21.1 4.2 13.9 –16.4 15.6 0.30 adjusted basis as the conditioned excess
10-year 12.6 20.1 –20.0 21.0 5.9 18.5 –22.7 18.0 0.32
returns have higher Sharpe ratios than
12-month
2-year 7.9 8.2 2.7 15.5 1.0 3.6 –2.7 4.6 0.28 unrestricted riding alternatives. For
5-year 10.0 16.8 –54 25.8 3.2 12.7 –10.8 15.1 0.25 holding horizons beyond 6 months,
7-year 11.2 21.2 –96 30.5 4.3 17.2 –15.0 19.7 0.25

Draft
10-year 12.9 27.5 –15.1 36.3 6.1 23.7 –20.5 25.5 0.26 however, market expectations are not
18-month able to enhance excess returns—on the
2-year 7.7 9.1 7.0 19.5 0.5 2.0 –1.0 2.5 0.27
5-year 10.1 19.2 0.8 33.0 2.9 12.8 –7.2 16.5 0.22 contrary, this strategy even dampens
7-year 11.2 24.5 –32 38.5 4.0 18.3 –11.2 22.5 0.22 returns while not reducing volatility
10-year 13.0 32.6 –8.5 47.3 5.8 27.0 –165 30.9 0.21
accordingly. This should hardly be sur-
24-month
5-year 10.0 19.5 8.3 41.8 2.4 11.4 –2.6 17.8 0.21 prising, taking into account that the
7-year 11.2 25.0 6.4 49.1 3.6 17.4 –4.4 27.0 0.21
10-year 13.0 34.3 4.2 62.8 5.4 27.5 –6.9 40.7 0.20 informational content of a short-term
instrument such as federal funds futures
is unlikely to be relevant for much
rule increases mean excess riding returns by as much as 40
beyond the instrument’s maturity.
basis points p.a. for a 3-month holding horizon. In line
A more detailed investigation into a possible “term
with the majority of alternative riding conditions, the addi-
structure of market expectations” as implied by Fed funds
tional return pickup for this type of ride steadily declines
futures could investigate whether deferred month futures
over longer investment horizons. Nevertheless, for a 1-year
contracts are able to provide an improved signal for longer-
investment period, Taylor rule riding still offers an improve-
dated investment horizons.
ment of 3.2% p.a. compared with buy-and-hold strategies.
In this article, we apply the Taylor Rule only to the
U.S. market since specification issues of estimating the Government versus Credit
Taylor rule for other currencies are beyond the current
The effectiveness of riding credit instruments instead
scope. Given its relative success as a return enhancement
of risk-free government paper generally increases with
strategy for U.S. Treasury rides, however, extending the
the maturity of the instrument and the holding horizon.
application to other markets could be an interesting area
This strategy appears to work best for dollar-denominated
for further research.
assets, where excess returns can be improved by as much
Market expectations. As reported earlier, market par-
as 1.61% p.a. by riding with 10-year swaps as opposed to
ticipants are fairly good at forecasting changes in the fed-
Treasuries. For euro assets, the success of such trades is at
eral funds rate, which implies that futures-based proxies
best very modest, whereas for sterling-based trades riding
for market expectations are a useful predictor of changes
the credit curve instead of the government curve does
in monetary policy stance. When employing this expec-
not seem advisable.

SEPTEMBER 2005 THE JOURNAL OF FIXED INCOME 25

EXHIBIT 15 risk-management guidelines, may not
EURIBOR/Swaps: HPR Statistics for Different Riding Strategies be able to engage in longer-maturity
rides without being able to control for
The table summarizes returns and excess returns for different riding strategies across selected horizons. The first
column lists the maturity of the riding instrument m. In order to compute the holding period returns (HPRs) of duration.
riding an m-maturity instrument for h months, the (m - h) rate must also be available. XHPR represents the excess
riding returns over the buy-and-hold strategy. S.R. is the Sharpe ratio of the excess returns. Returns and standard Adjusting for duration. By defi-
deviations are annualized for ease of comparison. The standard deviations of the various mean returns were
corrected for overlapping data by using a Newey et al. [1994] correction on the standard errors of the respective
nition, any riding strategy is implicitly
mean, where the lags are set equal to the length of the holding horizon. not only taking a position on the slope
m m of the term structure but also entails
HPR H t + h (%) XHPR XH t + h (%)
Mean S.D. Min Max Mean S.D. Min Max S.R.
some exposure to the level of interest
Instrument rates. By adjusting for duration, the
3-month
6-month 5.3 2.3 0.5 2.7 0.0 0.3 –0.5 0.2 0.00 element of placing an outright bet on
9-month 5.4 2.3 0.4 3.1 0.0 0.7 –0.9 0.6 0.06
1-year 5.2 2.2 –0.1 3.4 –0.2 1.2 –1.4 1.0 –0.14 the future direction of interest rates is
2-year 6.0 3.5 –1.5 5.5 0.6 3.1 –2.7 3.0 0.21 removed and the investor is left with
5-year 7.2 7.9 –4.0 8.2 1.9 7.8 –5.0 5.7 0.24
7-year 7.8 9.7 –5.7 8.6 2.5 9.6 –7.1 6.3 0.26 her primary objective of taking advan-
10-year 8.5 11.9 –8.2 9.0 3.2 11.9 –10.3 7.5 0.27
tage of a specific yield curve environ-
6-month
9-month 5.4 4.2 1.0 5.3 0.1 0.5 –0.5 0.4 0.08 ment. This may be particularly relevant
1-year 5.4 4.0 0.9 56 0.1 1.1 –1.1 0.9 –0.03 in our case, since for all currencies
2-year 6.0 5.2 –0.1 8.3 0.6 4.0 –3.1 3.3 0.16
5-year 7.5 11.3 –5.2 14.2 2.1 11.1 –8.9 9.2 0.19 there has been a clear downtrend in
7-year 8.1 14.0 –8.4 16.1 2.7 14.0 –12.1 11.1 0.19
10-year 8.8 17.5 –13.2 17.2 3.4 17.6 –16.9 12.2 0.19 interest rates over the entire 25-30-
12-month year sample period.
2-year 6.1 7.8 1.0 12.8 0.6 4.1 –2.8 4.0 0.14 In the context of this article, the
5-year 7.8 15.4 –3.6 20.3 2.3 14.7 –8.7 13.2 0.15

Draft
7-year 8.5 19.4 –6.8 22.5 3.0 19.4 –11.7 14.7 0.15 most meaningful duration target is the
10-year 9.3 25.1 –11.6 25.5 3.8 25.3 –168 16.9 0.15
duration of the different buy-and-hold
18-month
2-year 6.0 10.5 3.0 16.5 0.3 2.4 –1.8 22 0.11 strategies (i.e., 3, 6, 9, 12, 18, and 24
5-year 7.9 16.9 –3.2 27.8 2.2 14.7 –10.8 13.0 0.15
7-year 8.7 21.2 –6.8 32.1 3.0 20.4 –14.4 17.3 0.15
months). For this purpose, we match
10-year 9.7 27.8 –10.9 35.4 3.9 27.8 –184 24.6 0.14 the duration of the holding horizon by
24-month constructing a duration-neutral barbell
5-year 8.0 18.0 1.5 31.9 2.0 13.1 –9.3 14.0 0.15
7-year 8.9 21.8 –1.9 37.2 2.9 19.3 –12.6 18.1 0.15 portfolio using a weighted combina-
10-year 10.0 28.1 –6.3 41.3 4.0 27.2 –170 27.9 0.15
tion of the respective riding instrument
and an overnight deposit. For instance,
In the case of euro assets, the poor performance of in the case of riding a 12-month instru-
credit relative to government rides is easily explained by ment for 3 months, the duration of a portfolio invested in
the virtual absence of a positive credit spread (cf. lower an overnight deposit plus the 12-month instrument should,
graph of Exhibit 4). In the case of sterling assets, however, ex ante, be equal to the duration of the 3-month instru-
any attempts at explanation seem less straightforward but ment. This is expressed as:
are most likely linked to the fact that, on balance, the Gilt
curve tends to be steeper than the GBP LIBOR/swap D H = v D R + (1 − v ) D O (28)
curve (cf. Exhibits 2 and 3).
where DH is the target duration of the holding horizon,
Duration-Neutral Riding DR is the duration of the riding instrument, DO is the
duration of an overnight deposit, and w is the proportion
Although we have seen that riding the yield curve invested in the instrument such that the portfolio is dura-
may indeed offer an attractive means to enhance returns, tion neutral. Solving Equation (28) for ω gives
there are some practical drawbacks to this strategy. In par-
ticular, riding the yield curve instead of buying and holding DH − DO
v= R (29)
exposes the investor to a higher amount of interest rate risk D − DO
because of the duration extension implicit in riding the
yield curve. Indeed, bond portfolio managers, especially For practical purposes we can assume in the above
reserve managers at central banks who operate within strict example that DH = 0.25, DR = 1, and DO = 0; thus ω =

26 RIDING THE YIELD CURVE: A VARIETY OF STRATEGIES SEPTEMBER 2005

EXHIBIT 16 ments will yield few additional insights.
Government Securities: Mean Excess Holding Period Returns The results of these duration-neutral
Mean excess holding period returns for a given riding strategy are aggregated by holding period across all instruments. The
riding strategies are reported in Exhibit
first column lists the various conditions for implementing a given riding strategy. All returns are annualized for ease of 20, which also contains the non-adjusted
comparison. Numbers in parentheses are standard errors, which are corrected for serial correlation and heteroskedasticity
using Newey et al. [1994], where the lags are set equal to the length of the holding horizon (e.g., lags = 3 for 3-month riding returns for ease of comparison.
returns). Asterisks *,** indicate significance at the 90% and 95% level (two-sided test). Dashes (—) indicate that no results
were obtained for a given strategy; blanks indicate that no observations exist for a given holding period. Most striking, but nonetheless
Holding period
expected, is the dramatic decline in the
Riding condition 3 months 6 months 12 months 18 months 24 months mean excess returns when comparing the
XH tm+ Obs. XH tm+ h Obs. XH tm+ h Obs. XH tm+ h Obs.
XH tm+ h Obs.
standard rides with the duration-neutral
U.S. rates ones. Since the interest rate exposure of
Unconditioned 3.88** 261 3.76** 258 3.62** 252 3.33** 246 3.10** 240
(0.58) (0.81) (1.08) (1.17) (1.17) the standard rides is a linear function of the
Slope > 0 bps 3.89** 126 3.20** 127 2.46** 131 1.67** 125 2.05** 131
(0.72) (0.88) (1.08) (0.88) (0.67) duration of the riding instrument, the dura-
Cushion > 0 bps 4.20** 175 4.06** 179 4.39** 189 4.03** 185 3.24** 179
(0.70) (0.89) (1.20) (1.41) (1.40)
tion-neutral excess returns are reduced by
Cushion 75%ile 11.48* 4 12.34** 9 10.20** 13 6.13** 19 2.37** 15 a factor roughly equivalent to the duration
*
(2.94) (1.85) (0.49) (1.04) (1.07) of the riding instrument. In other words,
Recession 9.92** 28 902** 28 7.24** 28 5.35** 28 4.10** 28
(1.98) (2.40) (1.44) (1.09) (1.63) the duration-adjusted excess riding returns
Unconditioned 3.16** 186 3.08** 183 3.18** 177 2.98** 171 2.72** 165
(0.56) (0.75) (0.96) (1.00) (0.95) of the 10-year Treasury are approximately
Taylor rule 3.40** 100 2.74** 99 2.46** 94 2.39** 90 2.35** 86
(0.58) (0.82) (1.08) (1.07) (0.95) 10 times smaller than the non-adjusted
Dynamic Taylor rule 3.52** 144 3.34** 142 3.22** 137 3.08** 131 2.80** 125 ones, independent of the holding horizon.
(0.62) (0.78) (1.03) (1.10) (1.05) For a given holding horizon, how-
Unconditioned‡ 2.60** 117 2.72** 114 3.00** 108 2.85** 102 2.43** 96
(0.72) (0.93) (1.16) (1.22) (1.14) ever, the relative riskiness of the different

Draft
Expectations 3.08** 91 3.01** 88 2.71** 82 2.49* 76 2.21 70
(0.74) (0.95) (1.23) (1.39) (1.39)
riding instruments remains unchanged.
U.K. rates For example, with the exception of the
Unconditioned 1.08 73 1.62** 84 212** 89 1.99* 89 2.04** 87
(0.70) (0.83) (0.92) (1.08) (1.19) 3-month holding horizon, using 2-year
Slope > 0 bps 5.76** 5 0.44 4 3.66** 4 –3.15** 4 –0.02 5
(0.82) (2.22) (0.37) (0.11) (0.95) Treasuries as a riding instrument is the
Cushion > 0 bps 3.52** 11 5.80** 13 3.92** 17 3.71** 13 3.12** 8
(1.62) (0.92) (0.70) (1.67) (2.47) most effective riding strategy, whereas
Cushion 75%ile 3.04** 11 — — — — — — — —
(1.12) — — — — using 10-year Treasuries invariably seems
Recession — — –4.06** 1 6.57** 1 6.71** 2 — —
— (0.01) (0.01) (1.01) —
to be the most risky strategy.
German rates
Unconditioned 2.33** 360 2.29** 354 2.19* 348
Although the risk-adjusted rankings
(0.87) (1.04) (1.16) of different riding strategies seem to be
Slope > 0 bps 2.06* 224 1.38 221 0.91 212
(1.09) (1.38) (1.29) transitive between the two scenarios, the
Cushion > 0 bps 3.06** 252 2.81** 247 2.46* 242
(0.93) (1.16) (1.39) duration-adjusted strategies are signifi-
Cushion 75%ile 7.21** 17 10.23** 21 14.22** 26
(0.98) (0.90) (0.73)
cantly more efficient on a risk-adjusted
Recession 3.18 55 3.31* 53 3.32* 50 basis. With almost no exceptions, the
(2.03) (1.84) (1.92)
† Excess returns conditioned on the Taylor rule use a shorter sample period (1988:04–2003:12) since a minimum of five years of out-of-sample duration-neutral strategies display a higher
data are needed for the first estimate. ‡ Excess returns conditioned on market expectations use a sample period from 1994:01 to 2003:12 since
the Fed effective rate targeted by the FOMC was not announced prior to 1994. Sharpe ratio than the unadjusted strategies.
This result confirms earlier findings that
0.25. In line with the notation of Equation (13), the dura- duration is a good proxy for interest rate risk as up to 90%
tion-neutral riding excess returns are now defined as of yield curve changes are explained by a level change
across rates. Thus, as with other investment strategies, an
¶
XR[ m, h] = v H[ m, h ] + (1 − v ) H[ h] − H[ h ]
R O
(30) investor is likely to increase her returns by assuming a dura-
tion exposure when riding the yield curve—but she does
so at the cost of increased relative volatility (cf. Ilmanen
where HR[m,h] is the riding return, H O[h] is the return of an [1996a,1996b, 2002]). Duration-neutral riding may there-
overnight deposit compounded over the holding horizon fore provide fixed-income managers with an additional
h, and H[h] is the return of the buy-and-hold strategy.37 tool to increase their portfolio returns without unduly
Results. We compute the duration-neutral excess increasing the interest rate risk of their investments.
holding period returns for U.S. Treasuries only since the
extension of this concept to other currencies and instru-

SEPTEMBER 2005 THE JOURNAL OF FIXED INCOME 27

EXHIBIT 17 nificantly enhanced their returns by
LIBOR/Swaps: Mean
LIBOR/Swaps: MeanExcess
ExcessHolding
HoldingPeriod Returns
Period Returns riding the yield curve instead of buying
Mean excess holding period returns for a given riding strategy are aggregated by holding period across all instruments. The
and holding. Furthermore, employing
first column lists the various conditions for implementing a given riding strategy. All returns are annualized for ease of relatively straight forward filter rules
comparison. Numbers in parentheses are standard errors, which are corrected for serial correlation and heteroskedasticity
using Newey et al. [1994], where the lags are set equal to the length of the holding horizon (e.g., lags = 3 for 3-month riding would have increased these excess
returns). Asterisks *,** indicate significance at the 90% and 95% level (two-sided test).
returns even more. Since not all con-
Holding period ditional rides perform equally well
Riding condition 3 months 6 months 12 months 18 months 24 months
across currencies and instruments,
XH m Obs.
t +h XH m Obs.
t +h XH mt +h
Obs. XH tm+ h Obs. XH tm+ hObs.
diversification among various strategies
U.S. rates may present an additional approach to
Unconditioned 4.04** 195 4.02** 192 3.99** 186 3.84** 180 3.57** 174
(0.64) (0.88) (1.19) (1.33) (1.38) improve returns over the longer term.
Slope > 0 bps 4.68** 103 3.86** 117 3.50** 127 2.75** 120 3.06** 124
(0.84) (1.10) (1.44) (1.40) (1.29) By introducing the concept of dura-
Cushion > 0 bps 5.28** 102 5.40** 113 4.41** 162 4.16** 166 3.53** 162
(0.90) (1.12) (1.19) (1.31) (1.41)
tion-neutral riding, we are able to
Cushion 75%ile 18.48** 8 15.48** 14 11.92** 16 8.01** 17 4.00** 19 show that riding the yield curve is also
(1.20) (0.74) (0.62) (1.18) (1.90)
Recession 6.84** 17 6.30** 17 8.30** 17 9.00** 17 7.90** 17 a superior investment strategy on a
(1.62) (1.20) (0.87) (0.24) (1.15)
risk-adjusted basis.
U.K. rates
Unconditioned 3.16** 155 3.26** 152 3.43** 146 3.29** 140 3.04** 134
(0.62) (0.83) (1.09) (1.22) (1.23)
Slope > 0 bps 5.60** 44 3.14** 46 4.28** 41 2.64* 41 1.94 48
APPENDIX 1
(0.50) (0.96) (0.31) (1.49) (1.84)
Cushion > 0 bps 3.52** 40 5.04** 51 3.92** 64 2.93** 72 2.94** 66
(0.74) (0.72) (0.62) (1.17) (1.17) Derivation of Formula for
Cushion 75%ile 1.92 13 4.06 12 3.52 11 0.92 12 0.65 10 Riding Returns
(2.98) (3.42) (2.02) (1.76) (0.57)

Draft
Recession 4.04** 15 680** 15 6.55** 15 8.86** 15 7.47** 15
(0.88) (1.13) (0.73) (1.04) (1.48) This section provides a detailed
German rates derivation of Equations (18) and (19).
Unconditioned 1.80** 181 2.00** 178 2.29** 172 2.37* 166 2.39* 160
(0.52) (0.79) (1.14) (1.25) (1.24)
Recall that these equations provide an intu-
Slope > 0 bps 5.32** 66 4.32** 74 2.71** 76 2.99** 80 2.69* 69 itive approximation to calculate the excess
(0.60) (0.75) (1.38) (1.40) (1.43)
Cushion > 0 bps 3.20** 69 4.08** 81 4.03** 93 3.40** 95 3.06** 94 riding returns from selecting one strategy
(0.72) (0.95) (1.16) (1.40) (1.48) over another. In our case, we are calcu-
Cushion 75%ile 8.24** 6 7.76** 11 7.15** 22 6.23** 28 5.60** 30
(2.04) (1.16) (0.58) (0.28) (0.21) lating the excess returns from riding down
Recession 4.92** 41 3.72** 41 3.59** 35 3.26** 33 2.58** 30
(0.82) (0.95) (1.22) (1.71) (1.41)
the government curve instead of the
(LIBOR-based) credit curve.
CONCLUSION

Riding the yield curve, a conceptually simple Money-Market Version

trading strategy, relies on the existence of exploitable risk Our starting point is the explicit money-market version
premia. If market participants are able to earn risk- of the excess riding returns, Equation (18):
adjusted excess profits from riding the yield curve, this
contradicts at least the weak form of the efficient mar-
  m 
kets hypothesis. This article explores to what extent this   1 + ym , t  
z
proposition holds for two main asset classes across three XH[M,ride = − 1
m, h ]
 1 + y m − h  
major fixed-income markets.   m − h, t + h
z 
 
The article adds to the existing literature by looking   m 
at riding strategies for maturities beyond one year, by   1 + yˆm, t  
z
− − 1
focusing on non-dollar currencies, and by comparing rides   1 + yˆ m − h 
  m − h, t + h
z 
  (A-1)
between risk-free government securities and instruments
that contain a limited amount of credit risk. In addition,
we propose and test various ex ante rules to improve the suc- where the hats over the variables indicate the corresponding
cess rate of different riding strategies. rates for the credit instrument at the respective times and z is
With a sample period covering several interest rate the currency-specific day count basis. We now introduce the
cycles, our findings confirm that investors could have sig- following notation:

28 RIDING THE YIELD CURVE: A VARIETY OF STRATEGIES SEPTEMBER 2005

 1. At time t, the interest rate of the m-maturity credit instru- According to Equation (A-2), the excess returns from
ment ŷm,t can be expressed as the government rate ym,t riding the government instead of the credit yield curve are a
plus a yield spread ε. This is written as ŷm,t = ym,t + ε. linear combination of the initial yield pickup, ε, and the rela-
2. Between time t and time t+h, the interest rate of the tive slope difference of the instruments’ yield curve, c – η.
credit instrument has changed by an amount η. This is
expressed as ŷm–h,t+h = ŷm,t – η. Bond-Market Version
3. Similarly, between time t and time t + h, the interest rate
of the government instrument ym,t has changed by an As before, we begin with the explicit version of the excess
amount c. This is expressed as ym–h, t+h = ym,t – c. riding returns, Equation (19):

Noting that for small x and y it can be assumed that

(1+x)/(1+y) ≈ 1 + x – y, Equation (A-1) can now be stated as:  (
1 + ym , t )m / z − 1
m, h ] = 
XH[B,ride
(
 1 + ym − h, t + h )(m − h) / z 
   (1 + yˆ )m / z 
  m  m − h −
m, t
− 1
XH[M,ride ≈ y − y
 m  z  12 m − h ,t + h 
3  z    (1 + yˆm − h, t + h )m−h/ z
m,h]
4 4 
 ym ,t − c  (A-3)
 
  m  m − h where the hats over the variables indicate the corresponding
−  yˆm , t   − yˆm − h, t + h 
{  z  12 4 4 3  z  
 ym ,t +e y + e −h  rates for the credit instrument at the respective times. Again,
assuming that, ŷm,t = ym,t + ε, ŷm–h,t+h = ŷm,t – h, and ym–h,t+h =
m ,t

Ä
1
( )( ) (
=  − me − ym ,t − c m − h + ym ,t + e − h m − h 
z  )( ) ŷm,t – c, we can substitute these conditions into Equation (A-
3). Recalling that Pt+h/Pt – 1 ≈ ym,th/z – ∆yt Dt+h, we can derive
1
( )( )

Draft
Ä =  − me + c + e − h m − h  an approximate expression for the excess riding returns from
z
 
selecting one strategy over another
1  
= −
{
z  initialÄ
he + (
c − h
144244
m −)(
h
3
)
 spread
slopeÄ effect  (A-2)  
 h −
XH[B,ride ≈ y − Dy D
m,h]
 m z {t m − h ,t + h 
EXHIBIT 18  −c 
Positive Mean Excess HPRs:  
h
Government Bonds versus LIBOR/Swaps Ä yˆm , t − Dyˆm , t Dm − h ,t + h 

Positive Mean Excess HPRs: Government Bonds versus LIBOR/Swaps { z { 
Aggregated positive mean excess returns are expressed as a percentage of total excess returns. For example, riding U.S.  y h 
m ,t+e
Treasuries for 6 months conditional on a 75%ile cushion, on average 88.9% of the excess returns were positive. All returns
are annualized for ease of comparison. Dashes (—) indicate that no results were obtained for a given strategy; blanks indicate
 h h
that no observations exist for a given holding period.
y
= m z
(
+ c Dm − h ,t + h − ym ,t + e  ) z
Holding period
Riding condition 3 months 6 months 12 months 18 months 24 months  − h Dm − h ,t + h 
Govt Corp Govt Corp Govt Corp Govt Corp Govt Corp
 
U.S. rates
Unconditioned 61.3 63.6 67.1 68.2 68.3 74.7 74.4 80.6 78.8 82.8
1
=  − {
z inittialÄ
he + c
1
(
−
44
h
4
)(
D
2
z)
m − h ,t + h 
444 3

Slope > 0 bps 63.5 64.1 68.5 66.7 60.3 68.5 69.6 73.3 79.4 79.0  spread 
Cushion > 0 bps 62.9 67.6 68.7 71.7 75.7 77.2 77.3 82.5 79.3 82.1  slopeÄ effect 
Cushion 75%ile 75.0 100.0 88.9 100.0 100.0 100.0 100.0 100.0 73.3 78.9
Recession 82.1 82.4 89.3 88.2 89.3 100.0 89.3 100.0 89.3 100.0 (A-4)
Taylor rule 61.1 61.6 66.0 72.2 77.9
Dynamic Taylor 63.9 68.3 70.8 78.6 81.6
rule This way of expressing the excess returns
Expectations 62.9 67.9 72.4 77.3 79.2
of different investment strategies may particu-
U.K. rates larly appeal to market practitioners for two rea-
Unconditioned 54.8 67.7 65.5 75.7 74.2 81.5 75.3 77.9 71.3 75.4
Slope > 0 bps 100.0 81.8 75.0 80.4 100.0 976 — 73.2 20.0 58.3 sons. First, because it relies only on inputs that
Cushion > 0 bps 54.5 65.0 92.3 90.2 88.2 90.6 100.0 80.6 75.0 80.3
Cushion 75%ile 54.6 61.5 — 75.0 — 72.7 — 41.7 — 50.0 can easily be observed, the formula is straight-
Recession — 66.7 — 93.3 100.0 100.0 100.0 100.0 — 100.0 forward to compute. Second, excess returns are
German rates expressed as a function of two theoretically mean-
Unconditioned 58.6 64.6 70.3 71.5 701 74.1 72.4 76.9 ingful factors. This means that the formula is par-
Slope > 0 bps 80.3 81.1 67.4 76.3 60.2 83.8 58.5 78.3
Cushion > 0 bps 66.7 75.3 74.6 82.8 73.7 83.2 72.3 80.9 ticularly useful for performing ad hoc scenario
Cushion 75%ile 83.3 90.9 100.0 100.0 100.0 100.0 100.0 100.0
Recession 65.9 73.2 78.2 80.0 71.7 75.8 80.0 86.7
analyses. Furthermore, its use as a decision-

SEPTEMBER 2005 THE JOURNAL OF FIXED INCOME 29

 Svensson Method
EXHIBIT 19
Mean Excess HPRs: Government Bonds versus LIBOR/Swaps
Mean Excess HPRs: Government Bonds versus LIBOR/Swaps Whereas for zero-coupon bonds,
Mean excess holding period returns for a given riding strategy are aggregated by holding period across all instruments. The spot rates can be derived directly from
first column lists the various conditions for implementing a given riding strategy. All returns are annualized for ease of
comparison. Numbers in parentheses are standard errors, which are corrected for serial correlation and heteroskedasticity observed prices, for coupon-bearing bonds
using Newey et al. [1994], where the lags are set equal to the length of the holding horizon (e.g., lags = 3 for 3-month riding usually only their “yield to maturity” is
returns). Asterisks *,** indicate significance at the 90% and 95% level (two-sided test). Blanks indicate that no observations
exist for a given holding period. quoted. Let Pi,t be the price38 of a bond
with maturity i = 1,2,…,n and a stream of
Holding Period, XH tswap(
+h
m)
XH tGovt(
+h
m)
cash flows CFij at times mij. The yield to
Instrument 6 months 12 months 18 months 24 months
Mean Min Max Mean Min Max Mean Min Max Mean Min Max
maturity is the constant interest rate yt that
sets the present value of a bond equal to
U.S. rates
1-year 0.10** –0.7 1.0 its price:
(0.02)
0.56** –0.8 2.9 0.41** –0.3 1.3 0.20** –0.1 0.6
n
(0.04) (0.03) (0.02) CFi
5-year 0.94** –4.6 7.6 0.94** –2.2 4.7 0.87** –1.5 3.7 0.72** –1.2 3.2 Pi ,t = ∑1 (A-5)
7-year
(0.10)
1.20** –7.3 7.4
(0.03)
1.25** –3.0 5.6
(0.10)
1.22** –1.8 5.7
(0.09)
1.07** –2.1 4.8
( +y)
i =1 t
ti

(0.14) (0.10) (0.14) (0.14)

10-year 1.39** –13.9 14.9 1.56** –6.2 8.7 1.62** –3.8 8.1 1.49** –3.9 6.3 The yield to maturity is therefore
(0.21) (0.14) (0.22) (0.23) an average of the spot rates—and conse-
U.K. rates quently also the discount rates—across dif-
1-year –0.02 –1.6 1.7 ferent maturities. Consequently, the vector
(0.02)
–0.56** –8.0 3.9 1.97** –23.4 15.5 3.48** –13.6 15.7 of discount bonds corresponding to the
(0.10) (0.52) (0.59)
5-year –0.73** –13.5 16.1 –1.90** –10.1 7.2 –2.10** –8.3 3.8 1.88** –8.3 10.5
coupon-bearing bonds can be estimated
(0.28) (0.52) (0.29) (0.46) from the following non-linear model:
7-year 0.44* –8.2 22.7 –0.83** –12.9 3.9

Draft
–0.98** –11.8 2.7 –1.00** –11.2 2.4
(0.24) (0.28) (0.20) (0.20)
10-year 0.46 –13.5 22.8 –1.07** –19.7 7.1 –1.30** –17.2 51 –1.34** –15.6 4.6 n r
(0.34) (0.19) (0.35) (0.36)
Pi ,t = ∑ CF d ( m , b ) + ε
ij ij i, j
, i =1,2,...,n (A-6)
German j =1
rates
–0.05** –0.6 0.4 where δ(mij,β) is a parametric discount
(0.02)
5-year 0.08** –1.7 3.9 0.18** –1.2 5.3 0.24** –0.8 3.7 function with the parameter vector
β = ( β 0 , β1 , β 2 , β3 , τ 1 , τ 2 ) .
r
(0.03) (0.07) (0.08)
7-year 0.09 –2.5 3.0 0.22** –1.7 7.2 0.34** –1.0 5.2
(0.06) (0.09) (0.11) In attempting to estimate this dis-
10-year 0.12 –5.8 4.3 0.27** –4.2 8.6 0.48** –2.6 6.9
(0.08) (0.13) (0.15) count function, Nelson and Siegel [1987]
assume an explicit functional form for the
term structure of interest rates. To improve
the flexibility of the curves and the fit,
making tool can easily be extended to many other investment Svensson [1994] extended Nelson and
strategies. Siegel’s function and, according to this model, the zero-coupon
rates are given by

APPENDIX 2
ur 1 − exp − m / t1 ( )
Estimation of Zero-Coupon Yields
( )
s m, b = b0 + b1
m
+

t1
This section follows closely an unpublished technical  
manual on the implementation of zero-coupon curve estima-  1 − exp − m / t
+b2  1(  m
− exp  −  
)
tion techniques at central banks compiled by the BIS [1999].  m  t1  
The non-parametric estimation of a zero-coupon yield curve  t1 
is based on an assumed functional relationship between either  
par yields, spot rates, forward rates, or discount factors on one  1 − exp − m / t
+b3 
(2  m 

− exp  −  
)
hand and maturities on the other hand. Discount factors are the  m  t2  
quantities used at a given point in time to obtain the present   (A-7)
t2
value of future cash flows. A discount function dt,m is the collec-
tion of discount factors at time t for all maturities m.

30 RIDING THE YIELD CURVE: A VARIETY OF STRATEGIES SEPTEMBER 2005

and the discount function is assumptions about the HPRs. For example, the liquidity pref-
erence hypothesis assumes that HRPs also depend on a con-
ur stant term premium that monotonically increases with the term
(
ur
)
d m, b = exp  −
( )
 s m, b 
m
(A-8)
to maturity. Other variations of the REHTS include the market
 100 
segmentation hypothesis and the preferred habitat hypothesis.
See Mishkin [1990] and Cuthbertson [1996] for a thorough
overview.
2
Equations (A-7) and (A-8)r are substituted into equation See Cook and Hahn [1990] for a comprehensive review
(A-6) and the parameter vector β is estimated via a non-linear of the literature. Since then a number of authors have claimed
maximization algorithm. to have found evidence in support of the hypothesis (Rudebusch
[1995]; Gerlach and Smets [1997]). Other authors, however,
continue to reject the hypothesis either fully (M. Taylor [1992])
Spline-Based Method
or only for short-dated maturities (Campbell and Shiller [1991]).
3
The “smoothing splines” method developed by Fisher et A notable exception is provided by Ang et al. [1998],
al. [1995] represents an extension of the more traditional cubic who assess the efficacy of riding the yield curve with longer-
spline techniques.39 A cubic spline is a piecewise cubic poly- dated holding periods for Australia, Canada, and the United
nomial joined at “knot points.” The polynomials are then Kingdom. Unlike previous researchers, they are unable to find
restricted at the knot points such that their level and first two that riding strategies consistently outperform buy-and-hold
derivatives are identical. To each knot in the spline corresponds strategies.
4
one parameter. In the case of “smoothing splines” the number The terms “riding” and “rolling” down the yield curve
of parameters to be estimated is not fixed in advance. Instead, are often used interchangeably. Although they are similar, they
one starts from a model that is initially over-parameterized. are not exactly the same. Rolling refers to funding a long-term
Allowing for a large number of knot points guarantees sufficient asset with a short-term liability, for example, by borrowing

Draft
flexibility for curvature throughout the spline. The optimal money at the one-month LIBID rate and investing into a one-
number of knot points is then determined by minimizing the year. T-bill. It is essentially a leveraged ride of the yield curve.
ratio of a goodness-of-fit measure to the number of parame- In this article, the two concepts are kept separate.
5
ters. This approach penalizes the presence of parameters that do Campbell and Shiller [1991] provide an extensive treat-
not contribute significantly to the fit. ment of this point.
6
There are a broad range of spline-based models that use See Fehle [2003] for a recent overview of the literature
the “smoothing” method pioneered by Fisher et al. The main and He [2000] for a concise summary of the main drivers of
difference among the various approaches simply lies in the swap spreads.
7
extent to, and fashion by, which the smoothing criterion is Duffie and Huang [1996] examine the effects of credit
applied to obtain a better fix. The “variable-penalty rough- risk on swap rates. They conclude that the credit quality dif-
ness” approach recently implemented by the Bank of England ferential between the swap counterparties increases the swap
allows the “roughness” parameter to vary with the maturity, rate by as little as 1 basis point per 100 basis points difference
permitting more curvature at the short end (see Anderson and in the bond yields of the two counterparties.
8
Sleath [1999]). See Dignan [2003] for a recent exposition. Brandt and
Generally, the estimation method largely depends on the Kravajecz [2003] find that liquidity can account for as much as
intended use of the data: no-arbitrage pricing and valuation of 26% of the day-to-day variation in U.S. Treasury yields.
9
fixed-income and derivative instruments or information extrac- Throughout this article, simple compounding is used
tion for investment analytical and monetary policy purposes. for interest rates and yields are expressed in percentage rather
One of the main advantages of spline-based techniques over para- than decimal format, whereby ym,t = 0.035 is written as ym,t =
metric forms, such as the Svensson method, is that rather than 3.5%. T-bill rates can be converted from discount yield to
specifying a single functional form to describe spot rates, they money-market yield using the conversion yM 360 yd /(360-dyd).
10
fit a curve to the data that is composed of many segments, with Different currencies and different fixed-income instru-
the constraint that the overall curve is continuous and smooth.40 ments have different methods of counting days. Money-market
instruments generally count the actual number of days per
month and use a 360-day calendar year. Thus, the convention
ENDNOTES is m/z = ACT/360 except for GBP, where z = 365. Corpo-
1 rate bonds generally count 30 days to each month and 360 days
Apart from the simple or pure REHTS, there exist var-
per year (30/360), while Treasury bonds and swaps count the
ious other theories of the term structure of interest rates. These
actual days per month and year (ACT/ACT).
theories distinguish themselves by being based on different 11
This approximation of returns ignores convexity effects.

SEPTEMBER 2005 THE JOURNAL OF FIXED INCOME 31

EXHIBIT 20 smoothing splines is given by Anderson et
al. [1999] and Brooke et al. [2000].
U.S. Treasuries: Statistics for Duration-Neutral Riding Strategies 17
From January 1999 the DEM
The table summarizes duration-neutral returns and excess returns for different riding strategies across selected
horizons, where the duration target is set equal to the holding horizon. The first column in this table lists the
LIBOR and swap rates are replaced by euro
maturity of the riding instrument m. In order to compute the holding period returns (HPRs) of riding an m-maturity interest rates.
instrument for h months, the ( m - h) rate must also be available. XHPR represents the excess riding returns over 18
the buy-and-hold strategy. Hats indicate the relevant duration-neutral variables. S.R. is the Sharpe ratio of the This categorization of different
R O curve types is often applied inconsistently
excess returns. H[m,h] is the weighted ride return and (1 ) H[h] = (1 )Hh H[h] is the weighted return
of the overnight rate minus the return of the buy-and-hold strategy, as defined in Equation (30). Returns and in the literature as non-parametric curves
standard deviations are annualized for ease of comparison. The standard deviations of the various mean returns also depend upon a set of parameters.
were corrected for overlapping data using a Newey et al. [1994] correction on the standard errors of the
19
respective mean, where the lags are set equal to the length of the holding horizon. Dai and Singleton [2000] explore
the structural differences and relative good-
HPR [m, h] (%) XHPR XH[m, h] (%)
XHPR XH m,h (%) ness-of-fit of so-called affine term structure
models. Given that for such models there
Instrument Mean S.D. R ) H[h] Mean S.D. S.R.
H[m,h] (1 Mean S.D S.R is a trade-off between flexibility in mod-
eling the conditional correlations and the
3-month
6-month 6.3 2.8 0.500 3.15 –2.67 0.49 0.68 0.71 0.48 0.40 1.20 volatilities of the risk factors, they identify
2-year 7.8 4.9 0.125 0.98 –0.34 2.00 3.88 0.52 0.64 0.56 1.14 some models that are better suited than
5-year 9.8 10.1 0.050 0.49 0.11 4.00 9.54 0.42 0.60 0.60 1.00
7-year 10.8 13.2 0.036 0.39 0.21 4.96 12.66 0.39 0.60 0.58 1.03 others to explain historical interest rate
10-year 12.0 17.9 0.025 0.30 0.30 6.16 17.44 0.35 0.60 0.58 1.03
behavior.
6-month 20
Similar results are reported by
1-year 6.5 5.1 0.500 3.25 –2.93 0.44 1.60 0.28 0.32 0.55 0.58
2-year 7.7 7.4 0.250 1.93 –1.37 1.66 4.78 0.35 0.56 0.89 0.63 Dewachter and Lyrio [2002], who find that
5-year 9.9 14.3 0.100 0.99 –0.43 3.80 12.66 0.30 0.56 1.05 0.54
7-year 10.9 18.5 0.071 0.77 –0.25 4.76 17.08 0.28 0.52 1.07 0.48
the level factor is highly correlated to long-
10-year 12.1 25.1 0.050 0.61 –0.13 6.00 23.84 0.25 0.48 1.10 0.44 run inflation expectations, the slope factor

Draft
12-month captures temporary business and conditions,
2-year 7.6 10.4 0.500 3.80 –3.21 1.19 4.72 0.25 0.59 1.55 0.38 and the curvature factor appears to repre-
5-year 9.9 18.9 0.200 1.98 –1.30 3.47 15.68 0.22 0.68 2.13 0.32
7-year 10.9 24.9 0.143 1.56 –0.92 4.53 22.18 0.20 0.64 2.31 0.28 sent an independent monetary policy factor.
10-year 12.2 33.4 0.100 1.22 –0.65 5.78 31.25 0.18 0.57 2.44 0.23 21
This is consistent with a number of
18-month empirical studies that report a positive rela-
2-year 7.4 12.5 0.750 5.55 –5.22 0.55 2.83 0.19 0.33 1.26 0.26
5-year 9.8 22.0 0.300 2.94 –2.26 3.01 16.04 0.19 0.68 2.81 0.24 tionship between the volatility of short-
7-year 10.9 28.5 0.214 2.33 –1.70 4.17 23.27 0.18 0.63 3.18 0.20
10-year 12.2 37.3 0.150 1.83 –1.30 5.37 33.00 0.16 0.53 3.53 0.15
term interest rates and the shape of the yield
curve (e.g., Christiansen and Lund [2002]).
24-month 22
5-year 9.8 25.3 0.400 3.92 –3.23 2.61 14.96 0.17 0.70 3.34 0.21 McCallum [1994] shows the theo-
7-year 10.9 31.8 0.286 3.12 –2.47 3.75 22.30 0.17 0.65 3.78 0.17 retical linkage between the Federal
10-year 12.2 40.1 0.200 2.44 –1.90 5.04 31.54 0.16 0.55 4.24 0.14
Reserve’s policy and various tests of the
REHTS. For a comprehensive set of results,
It can be improved by including convexity such that [(Pt+h)/Pt] see Dotsey and Otrok [1995] and Rudebusch [1995].
– 1 ≈ ym,th/z – ∆yt Dt+h + 1/2 (Ct+h ∆yt2). See Fabozzi [1997] 23
See http://www.nber.org/cycles/ main.html for infor-
or Grabade [1996] for a derivation of this approximation. mation on recessions and recoveries, the NBER Business Cycle
12
Although only two types of instruments (government Dating Committee, and related topics.
and swaps) are considered in this article, the following analysis 24
In the case of the NBER, there are some curious
can easily be extended to other fixed-income asset classes. announcement asymmetries: the peaks of business cycles are
13
A detailed description of the intuition behind the new generally declared with a lag of 7-8 months, whereas troughs
notation and the derivation of Equations (18) and (19) is pro- take up to 18 months to report. For example, the most recent
vided in Appendix 1. recession lasting from March to November 2001 was announced
14
It is important to note that Equations (20) and (21) on 26 November 2001 and officially declared over only on 17
assume no change in the yield curve between time t and time July 2003. In the case of the United Kingdom and Germany,
t + h. there are no official statements that help identify recessions.
15
Selected Interest Rates (Table H.15 in Statistics: Releases Thus, taking the standard definition of 2 quarters of declining
and Historic Data) published by the Board of Governors of the GDP, recessions become known only with a lag of 6 months.
Federal Reserve System. 25
J. Taylor [1993] used a log-linear trend of real GDP
16
Until 1999, the Bank of England also employed the over 1984:Q1 to 1992:Q3 as a measure of potential GDP. As
Svensson method for yield curve estimation. A detailed account discussed later, Judd and Rudebusch [1998] use a more flex-
of the motivation for adopting a new approach based on ible estimate.

32 RIDING THE YIELD CURVE: A VARIETY OF STRATEGIES SEPTEMBER 2005

 26 35
Look-ahead bias arises because of the use of informa- Over the respective sample periods for the different cur-
tion in a simulation that would not be available during the time rencies, there are 28 months of recession in the United States,
period being simulated. Using lags of variables as they would 25 months in the United Kingdom, and 61 months in Ger-
have been available at the time of the simulation, itTR is estimated many.
as πt–3 + r * + 0.5(πt–3 – πt*) + 0.5yt–3. 36
For both Treasuries and USD Swaps, using lagged
27
In an alternative specification, ∆Taylort is defined as the NBER recessions increases excess returns even more—across
difference between the Taylor rule estimate and the actual target all holding horizons. E.g., for 3-month rides, mean excess
rate, which implies that the Taylor rule not only is useful to pre- returns increase from 9.92 to 13.12% for Treasuries and from
dict changes in the federal funds target but also sets the optimal 6.84 to 13.72% for Swaps. For Gilts, lagged recessions do slightly
level. In this instance, there is only mild significance on the pre- worse and for Swaps the results are broadly unchanged (some
dictive power of the Taylor rule. In particular, the Taylor rule horizons improve; others get marginally worse). For German
does well prior to 2000, but then seems to be breaking down. Bunds, lagged recessions increase riding returns marginally
Running the regression from 1989:01 (when the Federal Reserve across all holding horizons compared with the “simultaneous”
started moving in multiples of 25 basis points) to 2000:01 (just recessions. For EUR Swaps thinqs get worse across the board,
before the target rate peaked), the predictive power of the dynam- though there are still positive excess returns. For some horizons,
ically estimated Taylor rule is highest. See Exhibit 6. however, the excess returns become lower than the uncondi-
28
Since a minimum of 5 years of out-of-sample data are tional ones.
37
required for a first reasonable Taylor rule estimate, the overall The returns of the overnight deposits are computed by
sample size for U.S. government data is reduced by approxi- geometrically linking daily returns of overnight LIBID rates
mately 60 observations. for each month of the sample period. Although we ignore trans-
29
One possible explanation for the observation that φ < action costs, the duration-neutral riding strategies may incur
1 may be stem from the fact that the parameter estimates suffer higher transaction costs owing the daily rebalancing of the
from a downward bias owing to the implied “target rate stick- overnight deposit.

Draft
38
iness,” i.e., the assumption that the Fed only moves rates in Defined as clean price plus accrued interest up to time t.
39
multiples of 25 basis points. Spline functions, such as basis or B-splines, are used in
30
The implied futures rate is given by itf = 100 − ptf , where the context of yield curve estimation. There is sometimes some
ptf is the price of the contract at time t. Because the expected confusion among practitioners between spline functions and spline-
average funds rate for the entire contract month is a time- based interpolation. Whereas the former technique uses polyno-
weighted average of the observed rates so far and the expected mials in order to approximate (unknown) functions, the latter is
rates for the remaining days, as the month end approaches, simply a specific method to interpolate between two data points.
40
the futures price gets increasingly determined by past daily For example, at the long end of the yield curve, the
movements in the effective funds rate rather than expectations. Svensson model is constrained to converge to a constant level,
Thus, when the FOMC meeting falls on any time after the directly implying that the unbiased expectation hypothesis holds.
middle of the month, we define the next month’s contract as
the “relevant contract.”
31
REFERENCES
As with the signal from the Taylor rule, we put on a
riding trade whenever the market expectations signal does not Anderson, Nicola, and John Sleath. “New Estimates of the UK
indicate a rate hike. Real and Nominal Yield Curves.” Bank of England Quarterly
32
The German Treasury has only recently started auc- Bulletin, No. 4 (1999), pp. 384–392.
tioning six-month discount paper, the Bubills, at regular
monthly intervals. Ang, Andrew, and Monika Piazzesi. “A No-Arbitrage Vector
33
Although a swap is a zero-NPV instrument (i.e., not Autoregression of Term Structure Dynamics with Macroeco-
an investment in the strictest sense), a synthetic asset can be nomic and Latent Variables.” Journal of Monetary Economics, Vol.
created by receiving the fixed rate of the swap and investing 50, No. 4 (2003), pp. 745–787.
the proceeds in a deposit that is continuously rolled over to
meet the floating payments. As such, swaps represent AA credit Ang, Sharon, Lakshman Alles, and Dave Allen. “Riding the
risk and have less correlation with lower credits except during Yield Curve: An Analysis of International Evidence.” The Journal
a “flight to quality” or other Treasury-driven events. of Fixed Income, Vol. 8, No. 4 (1998), pp. 57–74.
34
According to the BIS Triennial Survey [2005], at the
end of June 2004 approximately 37% of the total notional prin- Bank for International Settlements. “Zero-Coupon Yield
cipal outstanding of $59 trillion was denominated in euros, 34% Curves: Technical Documentation.” Basel, March 1999.
in dollars, and 12% in yen.

SEPTEMBER 2005 THE JOURNAL OF FIXED INCOME 33

Bank for International Settlements. “Triennial Central Bank Dyl, Edward A., and Michael D. Joehnk. “Riding the Yield
Survey of Foreign Exchange and Derivatives Market Activity Curve: Does it Work?” The Journal of Portfolio Management, Vol.
2004.” Basel, March 2005. 7, No. 3 (1981), pp. 13–17.

Brandt, Michael W., and Kenneth A. Kravajecz. “Price Dis- Evans, Charles L, and David Marshall. “Economic Determi-
covery in the U.S. Treasury Market: The Impact of Orderflow nants of the Nominal Treasury Yield Curve.” Working Paper
and Liquidity on the Yield Curve.” NBER Working Paper 01–16, Federal Reserve Bank of Chicago, Chicago, IL, Sep-
Series 9529, National Bureau of Economic Research, Cam- tember 2002.
bridge, MA, February 2003.
Fabozzi, Frank J., ed., Handbook of Fixed Income Securities, 5th
Brooke, Martin, Neil Cooper, and Cederic Scholtes. “Inferring ed. Chicago, IL: Irwin Professional Publishing, 1997.
Market Interest Rate Expectations from Money Market Rates.”
Bank of England Quarterly Bulletin, No. 4 (2000), pp. 392–402. Fehle, Frank. “The Components of Interest Rate Swap Spreads:
Theory and International Evidence.” Journal of Futures Markets,
Campbell, John Y., and Robert J. Shiller. “Yield Spreads and Vol. 23, No. 4 (2003), pp. 347–387.
Interest Rate Movements: A Bird’s Eye View.” Review of Eco-
nomic Studies, Vol. 58, No. 2 (1991), pp. 495–514. Fisher, Mark E., Douglas Nychka, and David Zervos. “Fit-
ting the Term Structure of Interest Rates with Smoothing
Chicago Board of Trade. “Reference Guide: CBOT Fed Funds Splines.” Finance and Economics Discussion Series 95–1,
Futures.” Chicago, IL, March 2003. Board of Governors of the Federal Reserve System, Wash-
ington, DC, 1995.
Christiansen, Charlotte, and Jesper Lund. “Revisiting the Shape
of the Yield Curve: The Effect of Interest Rate Volatility.” Fleming, Michael J., and Eli M. Remolona. “The Term Struc-

Draft
Working Paper, Aarhus School of Business, Aarhus, Denmark, ture of Announcement Effects.” BIS Working Paper 71, Bank
February 2002. for International Settlements, Basel, June 1999a.

Cook, Timothy, and Thomas Hahn. “Interest Rate Expectations ——.”What Moves Bond Prices?” The Journal of Portfolio Man-
and the Sope of the Money Market Yield Curve.” Federal Reserve agement, Vol. 25, No. 4 (1999b), pp. 28–38.
Bank of Richmond Economic Review, Vol. 76 (1990), pp. 3–26.
Garbade, Kenneth D. “Fixed Income Analytics.” Cambridge, MA:
Cuthbertson, Keith. Quantitative Financial Economics. Chich- MIT Press, 1996.
ester: John Wiley & Sons, 1996.
Gerlach, Stefan, and Frank Smets. “The Term Structure of
Dai, Qiang, and Kenneth J. Singleton. “Specification Analysis Euro-Rates: Some Evidence in Support of the Expectations
of Affine Term Structure Models.” Journal of Finance, Vol. 55, Hypothesis.” Journal of International Money and Finance, Vol. 16
No. 5 (2000), pp. 1943–1978. (1997), pp. 305–321.

Dewachter, Hans, and Marco Lyrio. “Macro Factors and the Grieves, Robin, and Alan J. Marcus. “Riding the Yield Curve:
Term Strucutre of Interest Rates.” Research in Management ERS- Reprise.” The Journal of Portfolio Management, Vol. 18, No. 4
2003-037, Erasmus Research Institute of Management, Rot- (1992), pp. 67–76.
terdam, Netherlands, October 2002.
Grieves, Robin, Steven V. Mann, Alan J. Marcus, and Pradip-
Dignan, James H. “Nondefault Components of Investment- kumar Ramanlal. “Riding the Bill Curve.” The Journal of Port-
Grade Bond Spreads.” Financial Analysts Journal, Vol. 59, No. folio Management, Vol. 25, No. 3 (1999), pp. 74–82.
3 (2003), pp. 93–102.
He, Hua. “Modelling Term Structure of Swap Spreads.”
Dotsey, Michael, and Christopher Otrok. “The Rational Expec- Working paper, Yale School of Management, New Haven, CT,
tations Hypothesis of the Term Structure, Monetary Policy, May 2000.
and Time-Varying Term Premia.” Federal Reserve Bank of Rich-
mond Economic Quarterly, Vol. 81, No. 1 (1995), pp. 65–81. Ilmanen, Antti. “Does Duration Extension Enhance Long-Term
Expected Returns.” The Journal of Fixed Income, Vol. 6, No. 2
Duffie, Darrel, and Ming Huang. “Swap Rates and Credit (1996), pp. 23–36.
Quality.” Journal of Finance, Vol. 51, No. 3 (1996), pp. 921–949.

34 RIDING THE YIELD CURVE: A VARIETY OF STRATEGIES SEPTEMBER 2005

——. “Market Rate Expectation and Forward Rates.” The Rudebusch, Glenn D. “Federal Reserve Interest Rate Tar-
Journal of Fixed Income, Vol. 6. No. 2 (1996), pp. 8–22. geting, Rational Expectations, and the Term Structure.” Journal
of Monetary Economics, 35 (1995), pp. 245–274.
Ilmanen, Antti, and Rafey Sayood. “Quantitative Forecasting
Models and Active Diversification for International Bonds.” Rudebusch, Glenn D., and Tao Wu. “A Macro-Finance Model
The Journal of Fixed Income, Vol. 12, No. 3 (2002), pp. 40–51. of the Term Structure, Monetary Policy, and the Economy.”
Working paper, Federal Reserve Bank of San Francisco, San
Judd, John P., and Glenn D. Rudebusch. “Taylor’s Rule and the Francisco, CA, September 2003.
Fed: 1970–1997.” Federal Reserve Bank of San Francisco Economic
Review, No. 3 (1998), pp. 3–16. Söderström, Ulf. “Predicting Monetary Policy with Federal
Funds Futures Prices.” Journal of Futures Markets, Vol. 21, No.
Knetz, Peter J., Robert Litterman, and Jose A. Scheinkman. 4 (2001), pp. 377–391.
“Explorations into Factors Explaining Money Market Returns.”
Journal of Finance, Vol. 49, No. 5 (1994), pp. 1861–1882. Svensson, Lars E.O. “Estimating and Interpreting Forward
Interest Rates: Sweden 1992–1994.” NBER Working Paper
Litterman, Robert, and José A. Scheinkman. “Common Fac- Series 4871, National Bureau of Economic Research, Cam-
tors Affecting Bond Returns.” The Journal of Fixed Income, Vol. bridge, MA, September 1994.
1, No. 3 (1991), pp. 49–53.
Taylor, John B. “Discretion versus Policy Rules in Practice.”
Mankiw, N. Gregory, and James A. Miron. “The Changing Carnegie-Rochester Conference Series on Public Policy, 39 (1993),
Behaviour of the Term Structure of Interest Rates.” Quarterly pp. 195–214.
Journal of Economics, Vol. 101, No. 2 (1986), pp. 211–228.
Taylor, Mark P. “Modelling the Yield Curve.” Economic Journal,

Draft
McCallum, Bennett T. “Monetary Policy and the Term Struc- 102 (1992), pp. 524–537.
ture of Interest Rates.” NBER Working Paper Series 4938,
National Bureau of Economic Research, Cambridge, MA, Wu, Tao. “Monetary Policy and the Slope Factor in Empirical
December 1994. Term Structure Estimations.” Working paper, Federal Reserve
Bank of San Francisco, San Francisco, CA, August 2001.
Mishkin, Frederic S. “Yield Curve.” NBER Working Paper
Series 3550, National Bureau of Economic Research, Cam- ——. “What Makes the Yield Curve Move.” Economic Letter,
bridge, MA, December 1990. Federal Reserve Bank of San Francisco, San Francisco, CA,
June 2003.
Nelson, Charles R., and Andrew F. Siegel. “Parsimonious Mod-
eling of Yield Curves.” Journal of Business, Vol. 60, No. 4 (1987),
pp. 473–489. To order reprints of this article, please contact Ajani Malik at
[email protected] or 212-224-3205.
Newey, Whitney K., and Kenneth D. West. “Automatic Lag
Selection in Covariance Matrix Estimation.” Review of Economic
Studies, Vol. 61, No. 4 (1994), pp. 631–653.

SEPTEMBER 2005 THE JOURNAL OF FIXED INCOME 35