FIXED-INCOME SECURITIES

Chapter 6

Beyond Duration
 Outline

• Accounting for Larger Changes in Yield

• Accounting for a Non Flat Yield Curve

• Accounting for Non Parallel Shits

 Beyond Duration
Limits of Duration

• Duration hedging is
– Very simple
– Built on very restrictive assumptions
• Assumption 1: small changes in yield
– The value of the portfolio could be approximated by its first order Taylor
expansion
– OK when changes in yield are small, not OK otherwise
– This is why the hedge portfolio should be re-adjusted reasonably often
• Assumption 2: the yield curve is flat at the origin
– In particular we suppose that all bonds have the same yield rate
– In other words, the interest rate risk is simply considered as a risk on the
general level of interest rates
• Assumption 3: the yield curve is flat at each point in time
– In other words, we have assumed that the yield curve is only affected only
by a parallel shift
 Accounting for Larger Changes in Yield
Duration and Interest Rate Risk

Bond Price vs Yield

175
Bond Price (% of Par)

155

135
Actual
115 Duration Est.

95

75

55
6 7 8 9 10 11 12 13 14
Yield(%)
 Accounting for Larger Changes in Yield
Hedging Error

• Let us consider a 10 year maturity bond, with a 6% annual

coupon rate, a 7.36 modified duration, and which sells at par
• What happens if
– Case 1: yield increases from 6% to 6.01% (small increase)
– Case 2: yield increases from 6% to 8% (large increase)
• Case 1:
– Discount future cash-flows with new yield and obtain $99.267
– Absolute change : - 0.733 = (99.267-100)
– Use modified duration and find that change in price is
-100x7.36x0.001= - $0.736
– Very good approximation
• Case 2:
– Discount future cash-flows with new yield and obtain $86.58
– Absolute change : - 13.42 = (86.58 -100)
– Use modified duration and find that change in price is
-100x7.36x0.02= - $14.72
– Lousy approximation
 Accounting for Larger Changes in Yield
Convexity

• Relationship between price and yield is convex:

V 2 m
C  V ' '  y   2   i  (i  1) 
Fi
i2
0
 y i 1 (1  y )
• Taylor approximation:
V 1 V 2
V  V  y  y   V  y   y    2
y
y 2 2 y
• Relative change
V V ' ( y)
 y 
1 V " ( y)
y   Sens  y  Conv  y 
2 1 2

V V ( y) 2 V ( y) 2
• Conv is relative convexity, i.e., the second derivative
of value with respect to yield divided by value
 Accounting for Larger Changes in Yield
Convexity and $ Convexity

• (Relative) convexity is
V " y  1 m
i(i  1) Fi 
Conv     i 
 V y
V y   1  y 
2
i 1 1  i  

• $ Convexity = V’’(y) = Conv x V(y)

• Example (back to previous)
– 10 year maturity bond, with a 6% annual coupon rate, a 7.36
modified duration, a 6974 $ convexity and which sells at par
– Case 2: yields go from 6% to 8%
• Second order approximation to change in price
– Find: -14.72 + (6974.(0.02)²/2) = -$13.33
– Exact solution is -$13.42 and first order approximation is -$14.72
 Accounting for Larger Changes in Yield
Properties of Convexity

• Convexity is always positive

• For a given maturity and yield, convexity increases as
coupon rate
– Decreases
• For a given coupon rate and yield, convexity
increases as maturity
– Increases
• For a given maturity and coupon rate, convexity
increases as yield rate
– Decreases
Accounting for Larger Changes in Yield
Properties of Convexity

Bond Maturity Coupon YTM Price Conv

Bond 1 1 7% 6% 100.94 1.78
Bond 2 1 6% 6% 100 1.78
Bond 3 5 7% 6% 104.21 22.47
Bond 4 5 6% 6% 100 22.92
Bond 5 10 4% 6% 85.28 75.89
Bond 6 10 8% 6% 114.72 65.17
Bond 7 20 4% 6% 77.06 211.53
Bond 8 20 8% 7% 110.59 157.93
Bond 9 50 6% 6% 100 440.04
Bond 10 50 0% 6% 5.43 2269.5
 Accounting for Larger Changes in Yield
Properties of Convexity - Linearity

• Duration of a portfolio of n bonds

n
ConvP   Convi  wi
i 1
where wi is the weight of bond i in the portfolio, and:
n

w 1
i 1
i

• This is true if and only if all bonds have same yield,

i.e., if yield curve is flat
 Accounting for Larger Changes in Yield
Duration-Convexity Hedging

• Principle: immunize the value of a bond portfolio with

respect to changes in yield
– Denote by P the value of the portfolio
– Denote by H1 and H2 the value of two hedging instruments
– Needs two hedging instrument because want to hedge one risk
factor (still assume a flat yield curve) up to the second order
• Changes in value
– Portfolio P' ' ( y ) 2
dP  P' ( y )dy  dy
2
– Hedging instruments  1
 dH 1  H 1 ' ( y ) dy  H 1 ' ' ( y ) dy 2

 2
1
dH 2  H 2 ' ( y )dy  H 2 ' ' ( y )dy 2
 2
 Accounting for Larger Changes in Yield
Duration-Convexity Hedging

• Strategy: hold q1 (resp. q2) units of the first (resp.

second) hedging instrument so that
dP  q1  dH1  q2  dH 2  0
• Solution (under the assumption of unique dy –
parallel shifts)
 P' ( y )  q1 H1 ' ( y )  q2 H 2 ' ( y )  0

P' ' ( y )  q1 H1 ' ' ( y )  q2 H 2 ' ' ( y )  0
– Or (under the assumption of a unique y – flat yield curve)

 q1H1 ( y) Dur1  q2 H 2 ( y ) Dur2   P( y) Durp


q1H1 ( y )Conv1  q2 H 2 ( y)Conv2   P( y )Conv p
 Accounting for a Non Flat Yield Curve
Allowing for a Term Structure

• Problem with the previous method: we have

assumed a unique yield for all instrument, i.e., we
have assumed a flat yield curve
• We now relax this simplifying assumption and
consider 3 potentially different yields y, y1, y2
• On the other hand, we maintain the assumption of
parallel shifts, i.e., we assume dy = dy1 = dy2
• We are still looking for q1 and q2 such that
dP  q1  dH1  q2  dH 2  0
 Accounting for a Non Flat Yield Curve

• Solution (under the assumption of unique dy –

parallel shifts)  P' ( y)  q H ' ( y )  q H ' ( y )  0
1 1 1 2 2 2

P' ' ( y )  q1 H1 ' ' ( y1 )  q2 H 2 ' ' ( y2 )  0
– Or (relaxing the assumption of a flat yield curve)

 q1 H1 ( y1 ) Sens1  q2 H 2 ( y2 ) Sens2   P( y ) Sens p


q1H1 ( y1 )Conv1  q2 H 2 ( y2 )Conv2   P( y )Conv p
– Just replace (Macaulay) duration by sensitivity or modified duration
in the first equation
 Accounting for a Non Flat Yield Curve
Time for an Example!

• Portfolio at date t
– Price P = $ 32863.5
– Yield y = 5.143%
– Modified duration Sens = 6.76
– Convexity Conv =85.329
• Hedging instrument 1
– Price H1 = $ 97.962
– Yield y1 = 5.232 %
– Modified duration Sens1 = 8.813
– Convexity Conv1 = 99.081
• Hedging instrument 2:
– Price H2 = $ 108.039
– Yield y2 = 4.097%
– Modified duration Sens2 = 2.704
– Convexity Conv2 = 10.168
 Accounting for a Non Flat Yield Curve
Time for an Example!

• Optimal quantities q1 and q2 of each hedging

instrument are given by
 q1  8.813  97.962  q2  2.704  108.039  32863.5  6.76
q  99.081 97.962  q  10.168  108.039  32863.5  85.329
 1 2
– Or q1 = -305 and q2 = 140
• If you hold the portfolio, you should sell 305 units of
H1 and buy 140 units of H2
 Accounting for Non Parallel Shifts
Accounting for Changes in Shape of the TS

• Bad news is: not only the yield curve is not flat, but also it
changes shape!
• Afore mentioned methods do not allow to account for such
deformations
– Additional risk factors
– One has to regroup different risk factors to reduce the dimensionality of the
problem: e.g., a short, medium and long maturity factors
• Systematic approach: factor analysis on historical data has
shed some light on the dynamics of the yield curve
• 3 factors account for more than 90% of the variations
– Level factor
– Slope factor
– Curvature factor
 Accounting for Non Parallel Shifts
Accounting for Non Parallel Shits

• To properly account for the changes in the yield

curve, one has to get back to pure discount rates
m m
 F (i) B(t, t  i)  1  R (t, i)i
F (i)
V (t ) 
i 1 i 1 a
• Or, using continuously compounded rates
m m
V (t )   F (i) B(t, t  i)  F (i) exp i  R(t, i)
i 1 i 1
 Accounting for Non Parallel Shifts
Nelson Siegel Model

• The challenge is that we are now facing m risk factors

• Reduce the dimensionality of the problem by writing discount
rates as a function of 3 parameters
• One classic model is Nelson et Siegel’s
1  exp(  )  1  exp(  ) 
R(0,    0  1     2  exp(   ) 
       
– with R(0,): pure discount rate with maturity 
– 0 : level factor
– 1 : rotation factor
– 2 : curvature factor
–  : fixed scaling parameter
• Hedging principle: immunize the portfolio with respect to
changes in the value of the 3 parameters
 Accounting for Non Parallel Shifts
Nelson Siegel Model

• Mechanics of the model: changes in beta

parameters imply changes in discount rates, which
in turn imply changes in prices
• One may easily compute the sensitivity (partial
derivative) of R(0,) with respect to each parameter
beta (see next slide)
• Very consistent with factor analysis of interest rates
in the sense that they can be regarded as level,
slope and curvature factors, respectively
 Accounting for Non Parallel Shifts
Nelson Siegel
1.2

0.8
Sensitivity of rates

béta 0
0.6 béta 1
béta 2

0.4

0.2

0
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30
Maturity of rates
 Accounting for Non Parallel Shifts
Nelson Siegel Model

• Let us consider at date t=0 a bond with price P delivering the

future cash-flows Fi
• The price is given by
P0   Fi B(0,i )   Fi e i R (0, i )
i i
• Sensitivities of the bond price with respect to each beta
parameter are
 P0
   i Fi e  i R ( 0, i )
  0 i
 P0 1  exp(  i /  )   i R ( 0, i )

1
    i
i /  Fi e
 i  
 P0 1  exp(  i /  )   i R ( 0, i )

  2
    i
i /
 exp(   i /  )  Fi e
i  
 Accounting for Non Parallel Shifts
Example

• At date t=0, parameters are estimated (fitted) to be

Beta 0 Beta 1 Beta 2 Scale parameter

8% -3% -1% 3

• Sensitivities of 3 bonds with respect to each beta parameter, as

well as that of the portfolio invested in the 3 bonds, are

Maturity Coupon Price S0 S1 S2

Bond 1 2 ans 5% 98.627$ -192.51 -141.08 -41.28
Bond 2 7 ans 5% 90.786$ -545.42 -224.78 -156.73
Bond 3 10 ans 5% 79.606$ -812.61 -207.2 -173.03
Portfolio -1550.54 -573.06 -371.04
 Accounting for Non Parallel Shifts
Hedging with Nelson Siegel

• Principle: immunize the value of a bond portfolio with

respect to changes in parameters of the model
– Denote by P the value of the portfolio
– Denote by H1, H2 and H3 the value of three hedging instruments
– Needs 3 hedging instruments because want to hedge 3 risk factors
(up to the first order)
– Can also impose dollar neutrality constraint q0H0 + q1H1 + q2H2 +
q3H3 + q4H4 = - P (need a 4th instrument for that)
• Formally, look for q1, q2 and q3 such that
 P G1 G2 G3
   q1  q 2  q 3 0
 0  0  0  0
 P G1 G2 G3
  q1  q2  q3 0
 1 1 1 1
 P  q G1  q G2  q G3  0
  1
 2
 3
 2
 2 2 2
 Beyond Duration
General Comments

• Whatever the method used, duration, modified duration,

convexity and sensitivity to Nelson and Siegel parameters are
time-varying quantities
– Given that their value directly impact the quantities of hedging instruments,
hedging strategies are dynamic strategies
– Re-balancement should occur to adjust the hedging portfolio so that it
reflects the current market conditions
• In the context of Nelson and Siegel model, one may elect to
partially hedge the portfolio with respect to some beta
parameters
– This is a way to speculate on changes in some factors; it is known as
« semi-hedging » strategies
– For example, a portfolio bond holder who anticipates a decrease in interest
rates may choose to hedge with respect to parameters beta 1 and beta 2
(slope and curvature factors) while remaining voluntarily exposed to a
change in the beta 0 parameter (level factor)