This action might not be possible to undo. Are you sure you want to continue?
c 2007 Prof. YuhDauh Lyuu, National Taiwan University Page 627
To set up a philosophy against physics is rash;
philosophers who have done so
have always ended in disaster.
— Bertrand Russell
c 2007 Prof. YuhDauh Lyuu, National Taiwan University Page 628
Deﬁnitions and Basic Results
• Let A ≡ [ a
ij
]
1≤i≤m,1≤j≤n
, or simply A ∈ R
m×n
,
denote an mn matrix.
• It can also be represented as [ a
1
, a
2
, . . . , a
n
] where
a
i
∈ R
m
are vectors.
– Vectors are column vectors unless stated otherwise.
• A is a square matrix when m = n.
• The rank of a matrix is the largest number of linearly
independent columns.
c 2007 Prof. YuhDauh Lyuu, National Taiwan University Page 629
Deﬁnitions and Basic Results (continued)
• A square matrix A is said to be symmetric if A
T
= A.
• A real n n matrix
A ≡ [ a
ij
]
i,j
is diagonally dominant if [ a
ii
[ >
j=i
[ a
ij
[ for
1 ≤ i ≤ n.
– Such matrices are nonsingular.
• The identity matrix is the square matrix
I ≡ diag[ 1, 1, . . . , 1 ].
c 2007 Prof. YuhDauh Lyuu, National Taiwan University Page 630
Deﬁnitions and Basic Results (concluded)
• A matrix has full column rank if its columns are linearly
independent.
• A real symmetric matrix A is positive deﬁnite if
x
T
Ax =
i,j
a
ij
x
i
x
j
> 0
for any nonzero vector x.
• A matrix A is positive deﬁnite if and only if there exists
a matrix W such that A = W
T
W and W has full
column rank.
c 2007 Prof. YuhDauh Lyuu, National Taiwan University Page 631
Cholesky Decomposition
• Positive deﬁnite matrices can be factored as
A = LL
T
,
called the Cholesky decomposition.
– Above, L is a lower triangular matrix.
c 2007 Prof. YuhDauh Lyuu, National Taiwan University Page 632
Generation of Multivariate Normal Distribution
• Let x ≡ [ x
1
, x
2
, . . . , x
n
]
T
be a vector random variable
with a positive deﬁnite covariance matrix C.
• As usual, assume E[ x] = 0.
• This distribution can be generated by Py.
– C = PP
T
is the Cholesky decomposition of C.
a
– y ≡ [ y
1
, y
2
, . . . , y
n
]
T
is a vector random variable
with a covariance matrix equal to the identity matrix.
a
What if C is not positive deﬁnite? See Lai and Lyuu (2007).
c 2007 Prof. YuhDauh Lyuu, National Taiwan University Page 633
Generation of Multivariate Normal Distribution
(concluded)
• Suppose we want to generate the multivariate normal
distribution with a covariance matrix C = PP
T
.
• We start with independent standard normal
distributions y
1
, y
2
, . . . , y
n
.
• Then P[ y
1
, y
2
, . . . , y
n
]
T
has the desired distribution.
c 2007 Prof. YuhDauh Lyuu, National Taiwan University Page 634
Multivariate Derivatives Pricing
• Generating the multivariate normal distribution is
essential for the Monte Carlo pricing of multivariate
derivatives (p. 556).
• For example, the rainbow option on k assets has payoﬀ
max(max(S
1
, S
2
, . . . , S
k
) −X, 0)
at maturity.
• The closedform formula is a multidimensional integral.
a
a
Johnson (1987).
c 2007 Prof. YuhDauh Lyuu, National Taiwan University Page 635
Multivariate Derivatives Pricing (concluded)
• Suppose dS
j
/S
j
= r dt + σ
j
dW
j
, 1 ≤ j ≤ n, where C is
the correlation matrix for dW
1
, dW
2
, . . . , dW
k
.
• Let C = PP
T
.
• Let ξ consist of k independent random variables from
N(0, 1).
• Let ξ
= Pξ.
• Similar to Eq. (63) on p. 595,
S
i+1
= S
i
e
(r−σ
2
j
/2) ∆t+σ
j
√
∆t ξ
j
, 1 ≤ j ≤ n.
c 2007 Prof. YuhDauh Lyuu, National Taiwan University Page 636
LeastSquares Problems
• The leastsquares (LS) problem is concerned with
min
x∈R
n  Ax −b , where A ∈ R
m×n
, b ∈ R
m
, m ≥ n.
• The LS problem is called regression analysis in statistics
and is equivalent to minimizing the meansquare error.
• Often written as
Ax = b.
c 2007 Prof. YuhDauh Lyuu, National Taiwan University Page 637
Polynomial Regression
• In polynomial regression, x
0
+ x
1
x + + x
n
x
n
is used
to ﬁt the data ¦ (a
1
, b
1
), (a
2
, b
2
), . . . , (a
m
, b
m
) ¦.
• This leads to the LS problem,
1 a
1
a
2
1
a
n
1
1 a
2
a
2
2
a
n
2
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
1 a
m
a
2
m
a
n
m
x
0
x
1
.
.
.
x
n
=
b
1
b
2
.
.
.
b
m
.
• Consult the text for solutions.
c 2007 Prof. YuhDauh Lyuu, National Taiwan University Page 638
American Option Pricing by Simulation
• The continuation value of an American option is the
conditional expectation of the payoﬀ from keeping the
option alive now.
• The option holder must compare the immediate exercise
value and the continuation value.
• In standard Monte Carlo simulation, each path is
treated independently of other paths.
• But the decision to exercise the option cannot be
reached by looking at only one path alone.
c 2007 Prof. YuhDauh Lyuu, National Taiwan University Page 639
The LeastSquares Monte Carlo Approach
• The continuation value can be estimated from the
crosssectional information in the simulation by using
least squares.
a
• The result is a function of the state for estimating the
continuation values.
• Use the function to estimate the continuation value for
each path to determine its cash ﬂow.
• This is called the leastsquares Monte Carlo (LSM)
approach and is provably convergent.
b
a
Longstaﬀ and Schwartz (2001).
b
Cl´ement, Lamberton, and Protter (2002).
c 2007 Prof. YuhDauh Lyuu, National Taiwan University Page 640
A Numerical Example
• Consider a 3year American put on a
nondividendpaying stock.
• The put is exercisable at years 0, 1, 2, and 3.
• The strike price X = 105.
• The annualized riskless rate is r = 5%.
• The spot stock price is 101.
– The annual discount factor hence equals 0.951229.
• We use only 8 price paths to illustrate the algorithm.
c 2007 Prof. YuhDauh Lyuu, National Taiwan University Page 641
A Numerical Example (continued)
Stock price paths
Path Year 0 Year 1 Year 2 Year 3
1 101 97.6424 92.5815 107.5178
2 101 101.2103 105.1763 102.4524
3 101 105.7802 103.6010 124.5115
4 101 96.4411 98.7120 108.3600
5 101 124.2345 101.0564 104.5315
6 101 95.8375 93.7270 99.3788
7 101 108.9554 102.4177 100.9225
8 101 104.1475 113.2516 115.0994
c 2007 Prof. YuhDauh Lyuu, National Taiwan University Page 642
0 0.5 1 1.5 2 2.5 3
95
100
105
110
115
120
125
1
2
3
4
5
6
7
8
c 2007 Prof. YuhDauh Lyuu, National Taiwan University Page 643
A Numerical Example (continued)
• We use the basis functions 1, x, x
2
.
– Other basis functions are possible.
a
• The plot next page shows the ﬁnal estimated optimal
exercise strategy given by LSM.
• We now proceed to tackle our problem.
• Our concrete problem is to calculate the cash ﬂow along
each path, using information from all paths.
a
Laguerre polynomials, Hermite polynomials, Legendre polynomials,
Chebyshev polynomials, Gedenbauer polynomials, and Jacobi polynomi
als.
c 2007 Prof. YuhDauh Lyuu, National Taiwan University Page 644
0 0.5 1 1.5 2 2.5 3
95
100
105
110
115
120
125
1
2
3
4
5
6
7
8
c 2007 Prof. YuhDauh Lyuu, National Taiwan University Page 645
A Numerical Example (continued)
Cash ﬂows at year 3
Path Year 0 Year 1 Year 2 Year 3
1 — — — 0
2 — — — 2.5476
3 — — — 0
4 — — — 0
5 — — — 0.4685
6 — — — 5.6212
7 — — — 4.0775
8 — — — 0
c 2007 Prof. YuhDauh Lyuu, National Taiwan University Page 646
A Numerical Example (continued)
• The cash ﬂows at year 3 are the exercise value if the put
is in the money.
• Only 4 paths are in the money: 2, 5, 6, 7.
• Some of the cash ﬂows may not occur if the put is
exercised earlier, which we will ﬁnd out step by step.
• Incidentally, the European counterpart has a value of
0.951229
3
2.5476 + 0.4685 + 5.6212 + 4.0775
8
= 1.3680.
c 2007 Prof. YuhDauh Lyuu, National Taiwan University Page 647
A Numerical Example (continued)
• We move on to year 2.
• For each state that is in the money at year 2, we must
decide whether to exercise it.
• There are 6 paths for which the put is in the money: 1,
3, 4, 5, 6, 7.
• Only inthemoney paths will be used in the regression
because they are where early exercise is relevant.
– If there were none, we would move on to year 1.
c 2007 Prof. YuhDauh Lyuu, National Taiwan University Page 648
A Numerical Example (continued)
• Let x denote the stock prices at year 2 for those 6 paths.
• Let y denote the corresponding discounted future cash
ﬂows (at year 3) if the put is not exercised at year 2.
c 2007 Prof. YuhDauh Lyuu, National Taiwan University Page 649
A Numerical Example (continued)
Regression at year 2
Path x y
1 92.5815 0 0.951229
2 — —
3 103.6010 0 0.951229
4 98.7120 0 0.951229
5 101.0564 0.4685 0.951229
6 93.7270 5.6212 0.951229
7 102.4177 4.0775 0.951229
8 — —
c 2007 Prof. YuhDauh Lyuu, National Taiwan University Page 650
A Numerical Example (continued)
• We regress y on 1, x, and x
2
.
• The result is
f(x) = 22.08 −0.313114 x + 0.00106918 x
2
.
• f estimates the continuation value conditional on the
stock price at year 2.
• We next compare the immediate exercise value and the
continuation value.
c 2007 Prof. YuhDauh Lyuu, National Taiwan University Page 651
A Numerical Example (continued)
Optimal early exercise decision at year 2
Path Exercise Continuation
1 12.4185 f(92.5815) = 2.2558
2 — —
3 1.3990 f(103.6010) = 1.1168
4 6.2880 f(98.7120) = 1.5901
5 3.9436 f(101.0564) = 1.3568
6 11.2730 f(93.7270) = 2.1253
7 2.5823 f(102.4177) = 0.3326
8 — —
c 2007 Prof. YuhDauh Lyuu, National Taiwan University Page 652
A Numerical Example (continued)
• Amazingly, the put should be exercised in all 6 paths: 1,
3, 4, 5, 6, 7.
• Now, any positive cash ﬂow at year 3 should be set to
zero for these paths as the put is exercised before year 3.
– They are paths 5, 6, 7.
• Hence the cash ﬂows on p. 646 become the next ones.
c 2007 Prof. YuhDauh Lyuu, National Taiwan University Page 653
A Numerical Example (continued)
Cash ﬂows at years 2 & 3
Path Year 0 Year 1 Year 2 Year 3
1 — — 12.4185 0
2 — — 0 2.5476
3 — — 1.3990 0
4 — — 6.2880 0
5 — — 3.9436 0
6 — — 11.2730 0
7 — — 2.5823 0
8 — — 0 0
c 2007 Prof. YuhDauh Lyuu, National Taiwan University Page 654
A Numerical Example (continued)
• We move on to year 1.
• For each state that is in the money at year 1, we must
decide whether to exercise it.
• There are 5 paths for which the put is in the money: 1,
2, 4, 6, 8.
• Only inthemoney paths will be used in the regression
because they are where early exercise is relevant.
– If there were none, we would move on to year 0.
c 2007 Prof. YuhDauh Lyuu, National Taiwan University Page 655
A Numerical Example (continued)
• Let x denote the stock prices at year 1 for those 5 paths.
• Let y denote the corresponding discounted future cash
ﬂows if the put is not exercised at year 1.
• From p. 654, we have the following table.
c 2007 Prof. YuhDauh Lyuu, National Taiwan University Page 656
A Numerical Example (continued)
Regression at year 1
Path x y
1 97.6424 12.4185 0.951229
2 101.2103 2.5476 0.951229
2
3 — —
4 96.4411 6.2880 0.951229
5 — —
6 95.8375 11.2730 0.951229
7 — —
8 104.1475 0
c 2007 Prof. YuhDauh Lyuu, National Taiwan University Page 657
A Numerical Example (continued)
• We regress y on 1, x, and x
2
.
• The result is
f(x) = −420.964 + 9.78113 x −0.0551567 x
2
.
• f estimates the continuation value conditional on the
stock price at year 1.
• We next compare the immediate exercise value and the
continuation value.
c 2007 Prof. YuhDauh Lyuu, National Taiwan University Page 658
A Numerical Example (continued)
Optimal early exercise decision at year 1
Path Exercise Continuation
1 7.3576 f(97.6424) = 8.2230
2 3.7897 f(101.2103) = 3.9882
3 — —
4 8.5589 f(96.4411) = 9.3329
5 — —
6 9.1625 f(95.8375) = 9.83042
7 — —
8 0.8525 f(104.1475) = −0.551885
c 2007 Prof. YuhDauh Lyuu, National Taiwan University Page 659
A Numerical Example (continued)
• The put should be exercised for 1 path only: 8.
• Now, any positive future cash ﬂow should be set to zero
for this path as the put is exercised before years 2 and 3.
– But there is none.
• Hence the cash ﬂows on p. 654 become the next ones.
• They also conﬁrm the plot on p. 645.
c 2007 Prof. YuhDauh Lyuu, National Taiwan University Page 660
A Numerical Example (continued)
Cash ﬂows at years 1, 2, & 3
Path Year 0 Year 1 Year 2 Year 3
1 — 0 12.4185 0
2 — 0 0 2.5476
3 — 0 1.3990 0
4 — 0 6.2880 0
5 — 0 3.9436 0
6 — 0 11.2730 0
7 — 0 2.5823 0
8 — 0.8525 0 0
c 2007 Prof. YuhDauh Lyuu, National Taiwan University Page 661
A Numerical Example (continued)
• We move on to year 0.
• The continuation value is, from p 661,
(12.4185 0.951229
2
+ 2.5476 0.951229
3
+1.3990 0.951229
2
+ 6.2880 0.951229
2
+3.9436 0.951229
2
+ 11.2730 0.951229
2
+2.5823 0.951229
2
+ 0.8525 0.951229)/8
= 4.66263.
c 2007 Prof. YuhDauh Lyuu, National Taiwan University Page 662
A Numerical Example (concluded)
• As this is larger than the immediate exercise value of
105 −101 = 4, the put should not be exercised at year 0.
• Hence the put’s value is estimated to be 4.66263.
• Compare this to the European put’s value of 1.3680
(p. 647).
c 2007 Prof. YuhDauh Lyuu, National Taiwan University Page 663
Time Series Analysis
c 2007 Prof. YuhDauh Lyuu, National Taiwan University Page 664
The historian is a prophet in reverse.
— Friedrich von Schlegel (1772–1829)
c 2007 Prof. YuhDauh Lyuu, National Taiwan University Page 665
Conditional Variance Models for Price Volatility
• Although a stationary model (see text for deﬁnition) has
constant variance, its conditional variance may vary.
• Take for example an AR(1) process X
t
= aX
t−1
+
t
with [ a [ < 1.
– Here,
t
is a stationary, uncorrelated process with
zero mean and constant variance σ
2
.
• The conditional variance,
Var[ X
t
[ X
t−1
, X
t−2
, . . . ],
equals σ
2
, which is smaller than the unconditional
variance Var[ X
t
] = σ
2
/(1 −a
2
).
c 2007 Prof. YuhDauh Lyuu, National Taiwan University Page 666
Conditional Variance Models for Price Volatility
(concluded)
• In the lognormal model, the conditional variance evolves
independently of past returns.
• Suppose we assume that conditional variances are
deterministic functions of past returns:
V
t
= f(X
t−1
, X
t−2
, . . . )
for some function f.
• Then V
t
can be computed given the information set of
past returns:
I
t−1
≡ ¦ X
t−1
, X
t−2
, . . . ¦.
c 2007 Prof. YuhDauh Lyuu, National Taiwan University Page 667
ARCH Models
a
• An inﬂuential model in this direction is the
autoregressive conditional heteroskedastic (ARCH)
model.
• Assume that ¦ U
t
¦ is a Gaussian stationary,
uncorrelated process.
a
Engle (1982), cowinner of the 2003 Nobel Prize in Economic Sci
ences.
c 2007 Prof. YuhDauh Lyuu, National Taiwan University Page 668
ARCH Models (continued)
• The ARCH(p) process is deﬁned by
X
t
−µ =
a
0
+
p
i=1
a
i
(X
t−i
−µ)
2
1/2
U
t
,
where a
1
, . . . , a
p
≥ 0 and a
0
> 0.
– Thus X
t
[ I
t−1
∼ N(µ, V
2
t
).
• The variance V
2
t
satisﬁes
V
2
t
= a
0
+
p
i=1
a
i
(X
t−i
−µ)
2
.
• The volatility at time t as estimated at time t −1
depends on the p most recent observations on squared
returns.
c 2007 Prof. YuhDauh Lyuu, National Taiwan University Page 669
ARCH Models (concluded)
• The ARCH(1) process
X
t
−µ = (a
0
+ a
1
(X
t−1
−µ)
2
)
1/2
U
t
is the simplest.
• For it,
Var[ X
t
[ X
t−1
= x
t−1
] = a
0
+ a
1
(x
t−1
−µ)
2
.
• The process ¦ X
t
¦ is stationary with ﬁnite variance if
and only if a
1
< 1, in which case Var[ X
t
] = a
0
/(1 −a
1
).
c 2007 Prof. YuhDauh Lyuu, National Taiwan University Page 670
GARCH Models
a
• A very popular extension of the ARCH model is the
generalized autoregressive conditional heteroskedastic
(GARCH) process.
• The simplest GARCH(1, 1) process adds a
2
V
2
t−1
to the
ARCH(1) process, resulting in
V
2
t
= a
0
+ a
1
(X
t−1
−µ)
2
+ a
2
V
2
t−1
.
• The volatility at time t as estimated at time t −1
depends on the squared return and the estimated
volatility at time t −1.
a
Bollerslev (1986); Taylor (1986).
c 2007 Prof. YuhDauh Lyuu, National Taiwan University Page 671
GARCH Models (concluded)
• The estimate of volatility averages past squared returns
by giving heavier weights to recent squared returns (see
text).
• It is usually assumed that a
1
+ a
2
< 1 and a
0
> 0, in
which case the unconditional, longrun variance is given
by a
0
/(1 −a
1
−a
2
).
• A popular special case of GARCH(1, 1) is the
exponentially weighted moving average process, which
sets a
0
to zero and a
2
to 1 −a
1
.
• This model is used in J.P. Morgan’s RiskMetrics
TM
.
c 2007 Prof. YuhDauh Lyuu, National Taiwan University Page 672
GARCH Option Pricing
• Options can be priced when the underlying asset’s
return follows a GARCH process.
• Let S
t
denote the asset price at date t.
• Let h
2
t
be the conditional variance of the return over
the period [ t, t + 1 ] given the information at date t.
– “One day” is merely a convenient term for any
elapsed time ∆t.
c 2007 Prof. YuhDauh Lyuu, National Taiwan University Page 673
GARCH Option Pricing (continued)
• Adopt the following riskneutral process for the price
dynamics:
a
ln
S
t+1
S
t
= r −
h
2
t
2
+ h
t
t+1
, (66)
where
h
2
t+1
= β
0
+ β
1
h
2
t
+ β
2
h
2
t
(
t+1
−c)
2
, (67)
t+1
∼ N(0, 1) given information at date t,
r = daily riskless return,
c ≥ 0.
a
Duan (1995).
c 2007 Prof. YuhDauh Lyuu, National Taiwan University Page 674
GARCH Option Pricing (continued)
• The ﬁve unknown parameters of the model are c, h
0
, β
0
,
β
1
, and β
2
.
• It is postulated that β
0
, β
1
, β
2
≥ 0 to make the
conditional variance positive.
• The above process, called the nonlinear asymmetric
GARCH model, generalizes the GARCH(1, 1) model (see
text).
c 2007 Prof. YuhDauh Lyuu, National Taiwan University Page 675
GARCH Option Pricing (continued)
• It captures the volatility clustering in asset returns ﬁrst
noted by Mandelbrot (1963).
a
– When c = 0, a large
t+1
results in a large h
t+1
,
which in turns tends to yield a large h
t+2
, and so on.
• It also captures the negative correlation between the
asset return and changes in its (conditional) volatility.
b
– For c > 0, a positive
t+1
(good news) tends to
decrease h
t+1
, whereas a negative
t+1
(bad news)
tends to do the opposite.
a
“. . . large changes tend to be followed by large changes—of either
sign—and small changes tend to be followed by small changes . . . ”
b
Noted by Black (1976): Volatility tends to rise in response to “bad
news” and fall in response to “good news.”
c 2007 Prof. YuhDauh Lyuu, National Taiwan University Page 676
GARCH Option Pricing (concluded)
• With y
t
≡ ln S
t
denoting the logarithmic price, the
model becomes
y
t+1
= y
t
+ r −
h
2
t
2
+ h
t
t+1
. (68)
• The pair (y
t
, h
2
t
) completely describes the current state.
• The conditional mean and variance of y
t+1
are clearly
E[ y
t+1
[ y
t
, h
2
t
] = y
t
+ r −
h
2
t
2
, (69)
Var[ y
t+1
[ y
t
, h
2
t
] = h
2
t
. (70)
c 2007 Prof. YuhDauh Lyuu, National Taiwan University Page 677
The RitchkenTrevor (RT) Algorithm
a
• The GARCH model is a continuousstate model.
• To approximate it, we turn to trees with discrete states.
• Path dependence in GARCH makes the tree for asset
prices explode exponentially (why?).
• We need to mitigate this combinatorial explosion.
a
Ritchken and Trevor (1999).
c 2007 Prof. YuhDauh Lyuu, National Taiwan University Page 678
The RitchkenTrevor Algorithm (continued)
• Partition a day into n periods.
• Three states follow each state (y
t
, h
2
t
) after a period.
• As the trinomial model combines, 2n + 1 states at date
t + 1 follow each state at date t (recall p. 539).
• These 2n + 1 values must approximate the distribution
of (y
t+1
, h
2
t+1
).
• So the conditional moments (69)–(70) at date t + 1 on
p. 677 must be matched by the trinomial model to
guarantee convergence to the continuousstate model.
c 2007 Prof. YuhDauh Lyuu, National Taiwan University Page 679
The RitchkenTrevor Algorithm (continued)
• It remains to pick the jump size and the three branching
probabilities.
• The role of σ in the BlackScholes option pricing model
is played by h
t
in the GARCH model.
• As a jump size proportional to σ/
√
n is picked in the
BOPM, a comparable magnitude will be chosen here.
• Deﬁne γ ≡ h
0
, though other multiples of h
0
are
possible, and
γ
n
≡
γ
√
n
.
• The jump size will be some integer multiple η of γ
n
.
• We call η the jump parameter (p. 681).
c 2007 Prof. YuhDauh Lyuu, National Taiwan University Page 680
(0, 0)
y
t
(1, 1)
(1, 0)
(1, −1)
6
?
ηγ
n

1 day
The seven values on the right approximate the distribution
of logarithmic price y
t+1
.
c 2007 Prof. YuhDauh Lyuu, National Taiwan University Page 681
The RitchkenTrevor Algorithm (continued)
• The middle branch does not change the underlying
asset’s price.
• The probabilities for the up, middle, and down branches
are
p
u
=
h
2
t
2η
2
γ
2
+
r −(h
2
t
/2)
2ηγ
√
n
, (71)
p
m
= 1 −
h
2
t
η
2
γ
2
, (72)
p
d
=
h
2
t
2η
2
γ
2
−
r −(h
2
t
/2)
2ηγ
√
n
. (73)
c 2007 Prof. YuhDauh Lyuu, National Taiwan University Page 682
The RitchkenTrevor Algorithm (continued)
• It can be shown that:
– The trinomial model takes on 2n + 1 values at date
t + 1 for y
t+1
.
– These values have a matching mean for y
t+1
.
– These values have an asymptotically matching
variance for y
t+1
.
• The central limit theorem thus guarantees the desired
convergence as n increases.
c 2007 Prof. YuhDauh Lyuu, National Taiwan University Page 683
The RitchkenTrevor Algorithm (continued)
• We can dispense with the intermediate nodes between
dates to create a (2n + 1)nomial tree (p. 685).
• The resulting model is multinomial with 2n + 1
branches from any state (y
t
, h
2
t
).
• There are two reasons behind this manipulation.
– Interdate nodes are created merely to approximate
the continuousstate model after one day.
– Keeping the interdate nodes results in a tree that can
be as much as n times larger.
c 2007 Prof. YuhDauh Lyuu, National Taiwan University Page 684
y
t
6
?
ηγ
n

1 day
This heptanomial tree is the outcome of the trinomial tree
on p. 681 after its intermediate nodes are removed.
c 2007 Prof. YuhDauh Lyuu, National Taiwan University Page 685
The RitchkenTrevor Algorithm (continued)
• A node with logarithmic price y
t
+ ηγ
n
at date t + 1
follows the current node at date t with price y
t
for
some −n ≤ ≤ n.
• To reach that price in n periods, the number of up
moves must exceed that of down moves by exactly .
• The probability that this happens is
P() ≡
j
u
,j
m
,j
d
n!
j
u
! j
m
! j
d
!
p
j
u
u
p
j
m
m
p
j
d
d
,
with j
u
, j
m
, j
d
≥ 0, n = j
u
+ j
m
+ j
d
, and = j
u
−j
d
.
c 2007 Prof. YuhDauh Lyuu, National Taiwan University Page 686
The RitchkenTrevor Algorithm (continued)
• A particularly simple way to calculate the P()s starts
by noting that
(p
u
x + p
m
+ p
d
x
−1
)
n
=
n
=−n
P() x
. (74)
• So we expand (p
u
x + p
m
+ p
d
x
−1
)
n
and retrieve the
probabilities by reading oﬀ the coeﬃcients.
• It can be computed in O(n
2
) time.
c 2007 Prof. YuhDauh Lyuu, National Taiwan University Page 687
The RitchkenTrevor Algorithm (continued)
• The updating rule (67) on p. 674 must be modiﬁed to
account for the adoption of the discretestate model.
• The logarithmic price y
t
+ ηγ
n
at date t + 1 following
state (y
t
, h
2
t
) at date t has a variance equal to
h
2
t+1
= β
0
+ β
1
h
2
t
+ β
2
h
2
t
(
t+1
−c)
2
, (75)
– Above,
t+1
=
ηγ
n
−(r −h
2
t
/2)
h
t
, = 0, ±1, ±2, . . . , ±n,
is a discrete random variable with 2n + 1 values.
c 2007 Prof. YuhDauh Lyuu, National Taiwan University Page 688
The RitchkenTrevor Algorithm (continued)
• Diﬀerent conditional variances h
2
t
may require diﬀerent
η so that the probabilities calculated by Eqs. (71)–(73)
on p. 682 lie between 0 and 1.
• This implies varying jump sizes.
• The necessary requirement p
m
≥ 0 implies η ≥ h
t
/γ.
• Hence we try
η = ¸ h
t
/γ , ¸ h
t
/γ  + 1, ¸ h
t
/γ  + 2, . . .
until valid probabilities are obtained or until their
nonexistence is conﬁrmed.
c 2007 Prof. YuhDauh Lyuu, National Taiwan University Page 689
The RitchkenTrevor Algorithm (continued)
• The suﬃcient and necessary condition for valid
probabilities to exist is
a
[ r −(h
2
t
/2) [
2ηγ
√
n
≤
h
2
t
2η
2
γ
2
≤ min
1 −
[ r −(h
2
t
/2) [
2ηγ
√
n
,
1
2
.
• Obviously, the magnitude of η tends to grow with h
t
.
• The plot on p. 691 uses n = 1 to illustrate our points
for a 3day model.
• For example, node (1, 1) of date 1 and node (2, 3) of
date 2 pick η = 2.
a
Lyuu and Wu (2003).
c 2007 Prof. YuhDauh Lyuu, National Taiwan University Page 690
y
0
(1, 1)
(2, 3)
(2, 0)
(2, −1)
6
?
γ
n
= γ
1

3 days
c 2007 Prof. YuhDauh Lyuu, National Taiwan University Page 691
The RitchkenTrevor Algorithm (continued)
• The topology of the tree is not a standard combining
multinomial tree.
• For example, a few nodes on p. 691 such as nodes (2, 0)
and (2, −1) have multiple jump sizes.
• The reason is the path dependence of the model.
– Two paths can reach node (2, 0) from the root node,
each with a diﬀerent variance for the node.
– One of the variances results in η = 1, whereas the
other results in η = 2.
c 2007 Prof. YuhDauh Lyuu, National Taiwan University Page 692
The RitchkenTrevor Algorithm (concluded)
• The possible values of h
2
t
at a node are exponential
nature.
• To address this problem, we record only the maximum
and minimum h
2
t
at each node.
a
• Therefore, each node on the tree contains only two
states (y
t
, h
2
max
) and (y
t
, h
2
min
).
• Each of (y
t
, h
2
max
) and (y
t
, h
2
min
) carries its own η and
set of 2n + 1 branching probabilities.
a
Cakici and Topyan (2000).
c 2007 Prof. YuhDauh Lyuu, National Taiwan University Page 693
Negative Aspects of the RitchkenTrevor Algorithm
a
• A small n may yield inaccurate option prices.
• But the tree will grow exponentially if n is large enough.
– Speciﬁcally, n > (1 −β
1
)/β
2
when r = c = 0.
• A large n has another serious problem: The tree cannot
grow beyond a certain date.
• Thus the choice of n may be limited in practice.
• The RT algorithm can be modiﬁed to be free of
shortened maturity and (to some extent) exponential
complexity.
b
a
Lyuu and Wu (2003); Lyuu and Wu (2005).
b
It is only quadratic if n is not too large!
c 2007 Prof. YuhDauh Lyuu, National Taiwan University Page 694
Numerical Examples
• Assume S
0
= 100, y
0
= ln S
0
= 4.60517, r = 0,
h
2
0
= 0.0001096, γ = h
0
= 0.010469, n = 1,
γ
n
= γ/
√
n = 0.010469, β
0
= 0.000006575, β
1
= 0.9,
β
2
= 0.04, and c = 0.
• A daily variance of 0.0001096 corresponds to an annual
volatility of
√
365 0.0001096 ≈ 20%.
• Let h
2
(i, j) denote the variance at node (i, j).
• Initially, h
2
(0, 0) = h
2
0
= 0.0001096.
c 2007 Prof. YuhDauh Lyuu, National Taiwan University Page 695
Numerical Examples (continued)
• Let h
2
max
(i, j) denote the maximum variance at node
(i, j).
• Let h
2
min
(i, j) denote the minimum variance at node
(i, j).
• Initially, h
2
max
(0, 0) = h
2
min
(0, 0) = h
2
0
.
• The resulting threeday tree is depicted on p. 697.
c 2007 Prof. YuhDauh Lyuu, National Taiwan University Page 696
13.4809
13.4809
12.2883
12.2883
11.7170
11.7170
12.2846
10.5733
10.9645
10.9645
10.5697
10.5256
13.4644
10.1305
10.9600
10.9600
10.5215
10.5215
10.9603
10.1269
10.6042
09.7717
10.9553
10.9553
12.2700
10.5173
11.7005
10.1231
10.9511
10.9511
12.2662
10.5135
13.4438
10.9473
y
t
4.60517
4.61564
4.62611
4.63658
4.64705
4.65752
4.59470
4.58423
4.57376
0.01047
1
1
2
2
1
1
1
1
2
2
1
1
2
1
2
1
1
1
c 2007 Prof. YuhDauh Lyuu, National Taiwan University Page 697
A top (bottom) number inside a gray box refers to the
minimum (maximum, respectively) variance h
2
min
(h
2
max
,
respectively) for the node. Variances are multiplied by
100,000 for readability. A top (bottom) number inside a
white box refers to η corresponding to h
2
min
(h
2
max
,
respectively).
c 2007 Prof. YuhDauh Lyuu, National Taiwan University Page 698
Numerical Examples (continued)
• Let us see how the numbers are calculated.
• Start with the root node, node (0, 0).
• Try η = 1 in Eqs. (71)–(73) on p. 682 ﬁrst to obtain
p
u
= 0.4974,
p
m
= 0,
p
d
= 0.5026.
• As they are valid probabilities, the three branches from
the root node use single jumps.
c 2007 Prof. YuhDauh Lyuu, National Taiwan University Page 699
Numerical Examples (continued)
• Move on to node (1, 1).
• It has one predecessor node—node (0, 0)—and it takes
an up move to reach the current node.
• So apply updating rule (75) on p. 688 with = 1 and
h
2
t
= h
2
(0, 0).
• The result is h
2
(1, 1) = 0.000109645.
c 2007 Prof. YuhDauh Lyuu, National Taiwan University Page 700
Numerical Examples (continued)
• Because ¸ h(1, 1)/γ  = 2, we try η = 2 in
Eqs. (71)–(73) on p. 682 ﬁrst to obtain
p
u
= 0.1237,
p
m
= 0.7499,
p
d
= 0.1264.
• As they are valid probabilities, the three branches from
node (1, 1) use double jumps.
c 2007 Prof. YuhDauh Lyuu, National Taiwan University Page 701
Numerical Examples (continued)
• Carry out similar calculations for node (1, 0) with
= 0 in updating rule (75) on p. 688.
• Carry out similar calculations for node (1, −1) with
= −1 in updating rule (75).
• Single jump η = 1 works in both nodes.
• The resulting variances are
h
2
(1, 0) = 0.000105215,
h
2
(1, −1) = 0.000109553.
c 2007 Prof. YuhDauh Lyuu, National Taiwan University Page 702
Numerical Examples (continued)
• Node (2, 0) has 2 predecessor nodes, (1, 0) and (1, −1).
• Both have to be considered in deriving the variances.
• Let us start with node (1, 0).
• Because it takes a middle move to reach the current
node, we apply updating rule (75) on p. 688 with = 0
and h
2
t
= h
2
(1, 0).
• The result is h
2
t+1
= 0.000101269.
c 2007 Prof. YuhDauh Lyuu, National Taiwan University Page 703
Numerical Examples (continued)
• Now move on to the other predecessor node (1, −1).
• Because it takes an up move to reach the current node,
apply updating rule (75) on p. 688 with = 1 and
h
2
t
= h
2
(1, −1).
• The result is h
2
t+1
= 0.000109603.
• We hence record
h
2
min
(2, 0) = 0.000101269,
h
2
max
(2, 0) = 0.000109603.
c 2007 Prof. YuhDauh Lyuu, National Taiwan University Page 704
Numerical Examples (continued)
• Consider state h
2
max
(2, 0) ﬁrst.
• Because ¸ h
max
(2, 0)/γ  = 2, we ﬁrst try η = 2 in
Eqs. (71)–(73) on p. 682 to obtain
p
u
= 0.1237,
p
m
= 0.7500,
p
d
= 0.1263.
• As they are valid probabilities, the three branches from
node (2, 0) with the maximum variance use double
jumps.
c 2007 Prof. YuhDauh Lyuu, National Taiwan University Page 705
Numerical Examples (continued)
• Now consider state h
2
min
(2, 0).
• Because ¸ h
min
(2, 0)/γ  = 1, we ﬁrst try η = 1 in
Eqs. (71)–(73) on p. 682 to obtain
p
u
= 0.4596,
p
m
= 0.0760,
p
d
= 0.4644.
• As they are valid probabilities, the three branches from
node (2, 0) with the minimum variance use single jumps.
c 2007 Prof. YuhDauh Lyuu, National Taiwan University Page 706
Numerical Examples (continued)
• Node (2, −1) has 3 predecessor nodes.
• Start with node (1, 1).
• Because it takes a down move to reach the current node,
we apply updating rule (75) on p. 688 with = −1 and
h
2
t
= h
2
(1, 1).
• The result is h
2
t+1
= 0.0001227.
c 2007 Prof. YuhDauh Lyuu, National Taiwan University Page 707
Numerical Examples (continued)
• Now move on to predecessor node (1, 0).
• Because it also takes a down move to reach the current
node, we apply updating rule (75) on p. 688 with
= −1 and h
2
t
= h
2
(1, 0).
• The result is h
2
t+1
= 0.000105609.
c 2007 Prof. YuhDauh Lyuu, National Taiwan University Page 708
Numerical Examples (continued)
• Finally, consider predecessor node (1, −1).
• Because it takes a middle move to reach the current
node, we apply updating rule (75) on p. 688 with = 0
and h
2
t
= h
2
(1, −1).
• The result is h
2
t+1
= 0.000105173.
• We hence record
h
2
min
(2, −1) = 0.000105173,
h
2
max
(2, −1) = 0.0001227.
c 2007 Prof. YuhDauh Lyuu, National Taiwan University Page 709
Numerical Examples (continued)
• Consider state h
2
max
(2, −1).
• Because ¸ h
max
(2, −1)/γ  = 2, we ﬁrst try η = 2 in
Eqs. (71)–(73) on p. 682 to obtain
p
u
= 0.1385,
p
m
= 0.7201,
p
d
= 0.1414.
• As they are valid probabilities, the three branches from
node (2, −1) with the maximum variance use double
jumps.
c 2007 Prof. YuhDauh Lyuu, National Taiwan University Page 710
Numerical Examples (continued)
• Next, consider state h
2
min
(2, −1).
• Because ¸ h
min
(2, −1)/γ  = 1, we ﬁrst try η = 1 in
Eqs. (71)–(73) on p. 682 to obtain
p
u
= 0.4773,
p
m
= 0.0404,
p
d
= 0.4823.
• As they are valid probabilities, the three branches from
node (2, −1) with the minimum variance use single
jumps.
c 2007 Prof. YuhDauh Lyuu, National Taiwan University Page 711
Numerical Examples (concluded)
• Other nodes at dates 2 and 3 can be handled similarly.
• In general, if a node has k predecessor nodes, then 2k
variances will be calculated using the updating rule.
– This is because each predecessor node keeps two
variance numbers.
• But only the maximum and minimum variances will be
kept.
c 2007 Prof. YuhDauh Lyuu, National Taiwan University Page 712
Negative Aspects of the RT Algorithm Revisited
a
• Recall the problems mentioned on p. 694.
• In our case, combinatorial explosion occurs when
n >
1 −β
1
β
2
=
1 −0.9
0.04
= 2.5.
• Suppose we are willing to accept the exponential
running time and pick n = 100 to seek accuracy.
• But the problem of shortened maturity forces the tree to
stop at date 9!
a
Lyuu and Wu (2003).
c 2007 Prof. YuhDauh Lyuu, National Taiwan University Page 713
25 50 75 100 125 150 175
Date
5000
10000
15000
20000
25000
Dotted line: n = 3; dashed line: n = 4; solid line: n = 5.
c 2007 Prof. YuhDauh Lyuu, National Taiwan University Page 714
Backward Induction on the RT Tree
• After the RT tree is constructed, it can be used to price
options by backward induction.
• Recall that each node keeps two variances h
2
max
and
h
2
min
.
• We now increase that number to K equally spaced
variances between h
2
max
and h
2
min
at each node.
• Besides the minimum and maximum variances, the other
K −2 variances in between are linearly interpolated.
a
a
In practice, loglinear interpolation works better (Lyuu and Wu
(2005)). Logcubic interpolation works even better (Liu (2005)).
c 2007 Prof. YuhDauh Lyuu, National Taiwan University Page 715
Backward Induction on the RT Tree (continued)
• For example, if K = 3, then a variance of
10.5436 10
−6
will be added between the maximum
and minimum variances at node (2, 0) on p. 697.
• In general, the kth variance at node (i, j) is
h
2
min
(i, j) + k
h
2
max
(i, j) −h
2
min
(i, j)
K −1
,
k = 0, 1, . . . , K −1.
• Each interpolated variance’s jump parameter and
branching probabilities can be computed as before.
c 2007 Prof. YuhDauh Lyuu, National Taiwan University Page 716
Backward Induction on the RT Tree (concluded)
• During backward induction, if a variance falls between
two of the K variances, linear interpolation of the
option prices corresponding to the two bracketing
variances will be used as the approximate option price.
• The above ideas are reminiscent of the ones on p. 332,
where we dealt with arithmetic averagerate options.
c 2007 Prof. YuhDauh Lyuu, National Taiwan University Page 717
Numerical Examples
• We next use the numerical example on p. 697 to price a
European call option with a strike price of 100 and
expiring at date 3.
• Recall that the riskless interest rate is zero.
• Assume K = 2; hence there are no interpolated
variances.
• The pricing tree is shown on p. 719 with a call price of
0.66346.
– The branching probabilities needed in backward
induction can be found on p. 720.
c 2007 Prof. YuhDauh Lyuu, National Taiwan University Page 718
5.37392
5.37392
3.19054
3.19054
3.19054
3.19054
2.11587
2.11587
1.20241
1.20241
1.05240
1.05240
1.05240
1.05240
0.66346
0.66346
0.52360
0.52360
0.26172
0.48366
0.00000
0.00000
0.13012
0.13012
0.14573
0.00000
0.00000
0.00000
0.00000
0.00000
0.00000
0.00000
0.00000
0.00000
S
t
100.00000
101.05240
102.11587
103.19054
104.27652
105.37392
98.95856
97.92797
96.90811
1
1
2
2
1
1
1
1
2
2
1
1
2
1
2
1
1
1
c 2007 Prof. YuhDauh Lyuu, National Taiwan University Page 719
h
2
[ i][ j][0]
η [ i][ j][0]
rb[ i][0]
10.9600
10.9600
10.9553
10.9553
10.5215
10.5215
10.9645
10.9645
10.9511
10.9511
12.2700
10.5173
10.9603
10.1269
10.5697
10.5256
12.2883
12.2883
13.4438
10.9473
12.2662
10.5135
11.7005
10.1231
10.6042
09.7717
13.4644
10.1305
12.2846
10.5733
11.7170
11.7170
13.4809
13.4809
h
2
[0] [ ][ ]
1
1
1
1
1
1
2
2
1
1
2
1
2
1
1
1
2
2
η [0][ ][ ]
p[0][ ][ ][ ]
0
0
1
− 1
3
− 2
5
− 3
rb[0][ ]
0.4974 0.4974
0.0000 0.0000
0.5026 0.5026
0.4972 0.4972
0.0004 0.0004
0.5024 0.5024
0.4775 0.4775
0.0400 0.0400
0.4825 0.4825
0.1237 0.1237
0.7499 0.7499
0.1264 0.1264
0.4970 0.4970
0.0008 0.0008
0.5022 0.5022
0.4773 0.1385
0.0404 0.7201
0.4823 0.1414
0.4596 0.1237
0.0760 0.7500
0.4644 0.1263
0.4777 0.4797
0.0396 0.0356
0.4827 0.4847
0.1387 0.1387
0.7197 0.7197
0.1416 0.1416
h
2
[1] [ ][ ]
h
2
[2] [ ][ ]
h
2
[3] [ ][ ]
η [1][ ][ ]
η [2][ ][ ]
p[1][ ][ ][ ]
p[2][ ][ ][ ]
rb[ i][1]
h
2
[ i][ j][1]
η[i][j][1]
p[ i][ j][0 ][ 1] p[ i][ j] [ 1][ 1]
p[ i][ j][0][0] p[ i][ j][1][0]
p[ i][ j][0][ − 1] p[ i][ j][1][ − 1]
rb[1][ ] rb[2][ ] rb[3][ ]
j
3
2
1
0
− 1
− 2
3
2
1
0
− 1
− 2
j
5
4
3
2
0
j
− 3
− 2
− 1
1
c 2007 Prof. YuhDauh Lyuu, National Taiwan University Page 720
Numerical Examples (continued)
• Let us derive some of the numbers on p. 719.
• The option price for a terminal node at date 3 equals
max(S
3
−100, 0), independent of the variance level.
• Now move on to nodes at date 2.
• The option price at node (2, 3) depends on those at
nodes (3, 5), (3, 3), and (3, 1).
• It therefore equals
0.1387 ×5.37392 + 0.7197 ×3.19054 + 0.1416 ×1.05240 = 3.19054.
• Option prices for other nodes at date 2 can be computed
similarly.
c 2007 Prof. YuhDauh Lyuu, National Taiwan University Page 721
Numerical Examples (continued)
• For node (1, 1), the option price for both variances is
0.1237 ×3.19054 + 0.7499 ×1.05240 + 0.1264 ×0.14573 = 1.20241.
• Node (1, 0) is most interesting.
• We knew that a down move from it gives a variance of
0.000105609.
• This number falls between the minimum variance
0.000105173 and the maximum variance 0.0001227 at
node (2, −1) on p. 697.
c 2007 Prof. YuhDauh Lyuu, National Taiwan University Page 722
Numerical Examples (continued)
• The option price corresponding to the minimum
variance is 0.
• The option price corresponding to the maximum
variance is 0.14573.
• The equation
x 0.000105173 + (1 −x) 0.0001227 = 0.000105609
is satisﬁed by x = 0.9751.
• So the option for the down state is approximated by
x 0 + (1 −x) 0.14573 = 0.00362.
c 2007 Prof. YuhDauh Lyuu, National Taiwan University Page 723
Numerical Examples (continued)
• The up move leads to the state with option price
1.05240.
• The middle move leads to the state with option price
0.48366.
• The option price at node (1, 0) is ﬁnally calculated as
0.4775 ×1.05240 + 0.0400 ×0.48366 + 0.4825 ×0.00362 = 0.52360.
c 2007 Prof. YuhDauh Lyuu, National Taiwan University Page 724
Numerical Examples (concluded)
• It is possible for some of the three variances following an
interpolated variance to exceed the maximum variance
or be exceeded by the minimum variance.
• When this happens, the option price corresponding to
the maximum or minimum variance will be used during
backward induction.
• An interpolated variance may choose a branch that goes
into a node that is not reached in the forwardinduction
treebuilding phase.
a
• In this case, the algorithm fails.
a
Lyuu and Wu (2005).
c 2007 Prof. YuhDauh Lyuu, National Taiwan University Page 725
Interest Rate Derivative Securities
c 2007 Prof. YuhDauh Lyuu, National Taiwan University Page 726
What you are, you are only by contracts.
— Richard Wagner (1813–1883),
Der Ring des Nibelungen
Which shows that gambling’s not a sin
provided that you always win.
— Roald Dahl (1916–1990),
“SnowWhite and the Seven Dwarfs”
c 2007 Prof. YuhDauh Lyuu, National Taiwan University Page 727
Term Structure Fitting
c 2007 Prof. YuhDauh Lyuu, National Taiwan University Page 728
That’s an old besetting sin;
they think calculating is inventing.
— Johann Wolfgang Goethe (1749–1832)
c 2007 Prof. YuhDauh Lyuu, National Taiwan University Page 729
Introduction to Term Structure Modeling
c 2007 Prof. YuhDauh Lyuu, National Taiwan University Page 730
The fox often ran to the hole
by which they had come in,
to ﬁnd out if his body was still thin enough
to slip through it.
— Grimm’s Fairy Tales
c 2007 Prof. YuhDauh Lyuu, National Taiwan University Page 731
Outline
• Use the binomial interest rate tree to model stochastic
term structure.
– Illustrates the basic ideas underlying future models.
– Applications are generic in that pricing and hedging
methodologies can be easily adapted to other models.
• Although the idea is similar to the earlier one used in
option pricing, the current task is more complicated.
– The evolution of an entire term structure, not just a
single stock price, is to be modeled.
– Interest rates of various maturities cannot evolve
arbitrarily or arbitrage proﬁts may occur.
c 2007 Prof. YuhDauh Lyuu, National Taiwan University Page 732
Issues
• A stochastic interest rate model performs two tasks.
– Provides a stochastic process that deﬁnes future term
structures without arbitrage proﬁts.
– “Consistent” with the observed term structures.
• The unbiased expectations theory, the liquidity
preference theory, and the market segmentation theory
can all be made consistent with the model.
c 2007 Prof. YuhDauh Lyuu, National Taiwan University Page 733
History
• Methodology founded by Merton (1970).
• Modern interest rate modeling is often traced to 1977
when Vasicek and Cox, Ingersoll, and Ross developed
simultaneously their inﬂuential models.
• Early models have ﬁtting problems because they may
not price today’s benchmark bonds correctly.
• An alternative approach pioneered by Ho and Lee (1986)
makes ﬁtting the market yield curve mandatory.
• Models based on such a paradigm are called (somewhat
misleadingly) arbitragefree or noarbitrage models.
c 2007 Prof. YuhDauh Lyuu, National Taiwan University Page 734
Binomial Interest Rate Tree
• Goal is to construct a noarbitrage interest rate tree
consistent with the yields and/or yield volatilities of
zerocoupon bonds of all maturities.
– This procedure is called calibration.
a
• Pick a binomial tree model in which the logarithm of the
future short rate obeys the binomial distribution.
– Exactly like the CRR tree.
• The limiting distribution of the short rate at any future
time is hence lognormal.
a
Derman (2004), “complexity without calibration is pointless.”
c 2007 Prof. YuhDauh Lyuu, National Taiwan University Page 735
Binomial Interest Rate Tree (continued)
• A binomial tree of future short rates is constructed.
• Every short rate is followed by two short rates in the
following period (see next page).
• In the ﬁgure on p. 737 node A coincides with the start of
period j during which the short rate r is in eﬀect.
c 2007 Prof. YuhDauh Lyuu, National Taiwan University Page 736
r
*
r
0.5
j
r
h 0.5
A
B
C
period j −1 period j period j + 1
c 2007 Prof. YuhDauh Lyuu, National Taiwan University Page 737
Binomial Interest Rate Tree (continued)
• At the conclusion of period j, a new short rate goes into
eﬀect for period j + 1.
• This may take one of two possible values:
– r
: the “low” shortrate outcome at node B.
– r
h
: the “high” shortrate outcome at node C.
• Each branch has a ﬁfty percent chance of occurring in a
riskneutral economy.
c 2007 Prof. YuhDauh Lyuu, National Taiwan University Page 738
Binomial Interest Rate Tree (continued)
• We shall require that the paths combine as the binomial
process unfolds.
• The short rate r can go to r
h
and r
with equal
riskneutral probability 1/2 in a period of length ∆t.
• Hence the volatility of ln r after ∆t time is
σ =
1
2
1
√
∆t
ln
r
h
r
(see Exercise 23.2.3 in text).
• Above, σ is annualized, whereas r
and r
h
are period
based.
c 2007 Prof. YuhDauh Lyuu, National Taiwan University Page 739
Binomial Interest Rate Tree (continued)
• Note that
r
h
r
= e
2σ
√
∆t
.
• Thus greater volatility, hence uncertainty, leads to larger
r
h
/r
and wider ranges of possible short rates.
• The ratio r
h
/r
may depend on time if the volatility is a
function of time.
• Note that r
h
/r
has nothing to do with the current
short rate r if σ is independent of r.
c 2007 Prof. YuhDauh Lyuu, National Taiwan University Page 740
Binomial Interest Rate Tree (continued)
• In general there are j possible rates in period j,
r
j
, r
j
v
j
, r
j
v
2
j
, . . . , r
j
v
j−1
j
,
where
v
j
≡ e
2σ
j
√
∆t
(76)
is the multiplicative ratio for the rates in period j (see
ﬁgure on next page).
• We shall call r
j
the baseline rates.
• The subscript j in σ
j
is meant to emphasize that the
short rate volatility may be time dependent.
c 2007 Prof. YuhDauh Lyuu, National Taiwan University Page 741
Baseline rates
A C
B
B
C
C
D
D
D
D
H
H L
H L
! !
H L
! !
H
H
!
c 2007 Prof. YuhDauh Lyuu, National Taiwan University Page 742
Binomial Interest Rate Tree (concluded)
• In the limit, the short rate follows the following process,
r(t) = µ(t) e
σ(t) W(t)
, (77)
in which the (percent) short rate volatility σ(t) is a
deterministic function of time.
• As the expected value of r(t) equals µ(t) e
σ(t)
2
(t/2)
, a
declining short rate volatility is usually imposed to
preclude the short rate from assuming implausibly high
values.
• Incidentally, this is how the binomial interest rate tree
achieves mean reversion.
c 2007 Prof. YuhDauh Lyuu, National Taiwan University Page 743
Memory Issues
• Path independency: The term structure at any node is
independent of the path taken to reach it.
• So only the baseline rates r
i
and the multiplicative
ratios v
i
need to be stored in computer memory.
• This takes up only O(n) space.
a
• Storing the whole tree would have taken up O(n
2
)
space.
– Daily interest rate movements for 30 years require
roughly (30 365)
2
/2 ≈ 6 10
7
doubleprecision
ﬂoatingpoint numbers (half a gigabyte!).
a
Throughout this chapter, n denotes the depth of the tree.
c 2007 Prof. YuhDauh Lyuu, National Taiwan University Page 744
Set Things in Motion
• The abstract process is now in place.
• Now need the annualized rates of return associated with
the various riskless bonds that make up the benchmark
yield curve and their volatilities.
• In the U.S., for example, the ontherun yield curve
obtained by the most recently issued Treasury securities
may be used as the benchmark curve.
c 2007 Prof. YuhDauh Lyuu, National Taiwan University Page 745
Set Things in Motion (concluded)
• The term structure of (yield) volatilities
a
can be
estimated from either the historical data (historical
volatility) or interest rate option prices such as cap
prices (implied volatility).
• The binomial tree should be consistent with both term
structures.
• Here we focus on the term structure of interest rates.
a
Or simply the volatility (term) structure.
c 2007 Prof. YuhDauh Lyuu, National Taiwan University Page 746
Model Term Structures
• The model price is computed by backward induction.
• Refer back to the ﬁgure on p. 737.
• Given that the values at nodes B and C are P
B
and P
C
,
respectively, the value at node A is then
P
B
+ P
C
2(1 + r)
+ cash ﬂow at node A.
• We compute the values column by column without
explicitly expanding the binomial interest rate tree (see
ﬁgure next page).
• This takes quadratic time and linear space.
c 2007 Prof. YuhDauh Lyuu, National Taiwan University Page 747
HL
A C
B
Cash flows:
B
C
C
D
D
D
D
+
+
2 2
H
= B
+
2 2
HL
!
= B
H
HL
+
2 2
HL
! "
? D
2
2
2
!
2
"
c 2007 Prof. YuhDauh Lyuu, National Taiwan University Page 748
Term Structure Dynamics
• An nperiod zerocoupon bond’s price can be computed
by assigning $1 to every node at period n and then
applying backward induction.
• Repeating this step for n = 1, 2, . . . , one obtains the
market discount function implied by the tree.
• The tree therefore determines a term structure.
• It also contains a term structure dynamics.
– Taking any node in the tree as the current state
induces a binomial interest rate tree and, again, a
term structure.
c 2007 Prof. YuhDauh Lyuu, National Taiwan University Page 749
To set up a philosophy against physics is rash; philosophers who have done so have always ended in disaster. — Bertrand Russell
c 2007 Prof. YuhDauh Lyuu, National Taiwan University
Page 628
Deﬁnitions and Basic Results
• Let A ≡ [ aij ]1≤i≤m,1≤j≤n , or simply A ∈ Rm×n , denote an m × n matrix. • It can also be represented as [ a1 , a2 , . . . , an ] where ai ∈ Rm are vectors. – Vectors are column vectors unless stated otherwise. • A is a square matrix when m = n. • The rank of a matrix is the largest number of linearly independent columns.
c 2007 Prof. YuhDauh Lyuu, National Taiwan University
Page 629
1 ]. . • A real n × n matrix A ≡ [ aij ]i. 1. YuhDauh Lyuu. National Taiwan University Page 630 .j is diagonally dominant if  aii  > 1 ≤ i ≤ n. j=i  aij  for c 2007 Prof. • The identity matrix is the square matrix I ≡ diag[ 1. . .Deﬁnitions and Basic Results (continued) • A square matrix A is said to be symmetric if AT = A. – Such matrices are nonsingular. .
National Taiwan University Page 631 . YuhDauh Lyuu. c 2007 Prof.Deﬁnitions and Basic Results (concluded) • A matrix has full column rank if its columns are linearly independent. • A real symmetric matrix A is positive deﬁnite if xT Ax = i. • A matrix A is positive deﬁnite if and only if there exists a matrix W such that A = W T W and W has full column rank.j aij xi xj > 0 for any nonzero vector x.
YuhDauh Lyuu. c 2007 Prof.Cholesky Decomposition • Positive deﬁnite matrices can be factored as A = LLT . National Taiwan University Page 632 . called the Cholesky decomposition. L is a lower triangular matrix. – Above.
. National Taiwan University Page 633 .Generation of Multivariate Normal Distribution • Let x ≡ [ x1 . . . • As usual. . YuhDauh Lyuu. . . a What if C is not positive deﬁnite? See Lai and Lyuu (2007). . assume E[ x ] = 0. y2 .a – y ≡ [ y1 . . x2 . c 2007 Prof. – C = P P T is the Cholesky decomposition of C. yn ]T is a vector random variable with a covariance matrix equal to the identity matrix. • This distribution can be generated by P y. xn ]T be a vector random variable with a positive deﬁnite covariance matrix C.
y2 . National Taiwan University Page 634 . . . . yn ]T has the desired distribution. • Then P [ y1 . . . yn .Generation of Multivariate Normal Distribution (concluded) • Suppose we want to generate the multivariate normal distribution with a covariance matrix C = P P T . . . YuhDauh Lyuu. • We start with independent standard normal distributions y1 . . y2 . c 2007 Prof.
.Multivariate Derivatives Pricing • Generating the multivariate normal distribution is essential for the Monte Carlo pricing of multivariate derivatives (p. National Taiwan University Page 635 . Sk ) − X. • The closedform formula is a multidimensional integral. c 2007 Prof. • For example. YuhDauh Lyuu.a a Johnson (1987). the rainbow option on k assets has payoﬀ max(max(S1 . . . . 556). 0) at maturity. S2 .
• Similar to Eq.Multivariate Derivatives Pricing (concluded) • Suppose dSj /Sj = r dt + σj dWj . dW2 . • Let ξ = P ξ. • Let C = P P T . Si+1 = Si e 2 (r−σj /2) ∆t+σj √ ∆t ξj . dWk . (63) on p. • Let ξ consist of k independent random variables from N (0. . National Taiwan University Page 636 . . 595. . 1 ≤ j ≤ n. where C is the correlation matrix for dW1 . 1 ≤ j ≤ n. . 1). c 2007 Prof. YuhDauh Lyuu.
b ∈ Rm . National Taiwan University Page 637 . YuhDauh Lyuu. • The LS problem is called regression analysis in statistics and is equivalent to minimizing the meansquare error. c 2007 Prof. • Often written as Ax = b. where A ∈ Rm×n .LeastSquares Problems • The leastsquares (LS) problem is concerned with minx∈Rn Ax − b . m ≥ n.
··· an 1 an 2 .Polynomial Regression • In polynomial regression. 1 to the LS problem. . . b2 ). an m x0 x1 . . National Taiwan University Page 638 . . a2 m ··· ··· . x0 + x1 x + · · · + xn xn is used to ﬁt the data { (a1 . . . .. • This leads 1 1 . . bm ) }. . bm = • Consult the text for solutions. . . b1 ). . . c 2007 Prof. . YuhDauh Lyuu. . . . a1 a2 . (a2 . am a2 1 a2 2 . . (am . xn b1 b2 .
• The option holder must compare the immediate exercise value and the continuation value. National Taiwan University Page 639 . • In standard Monte Carlo simulation. c 2007 Prof. each path is treated independently of other paths. YuhDauh Lyuu. • But the decision to exercise the option cannot be reached by looking at only one path alone.American Option Pricing by Simulation • The continuation value of an American option is the conditional expectation of the payoﬀ from keeping the option alive now.
• This is called the leastsquares Monte Carlo (LSM) approach and is provably convergent. and Protter (2002). Lamberton.b and Schwartz (2001). e a Longstaﬀ c 2007 Prof.The LeastSquares Monte Carlo Approach • The continuation value can be estimated from the crosssectional information in the simulation by using least squares. • Use the function to estimate the continuation value for each path to determine its cash ﬂow. YuhDauh Lyuu. b Cl´ment. National Taiwan University Page 640 .a • The result is a function of the state for estimating the continuation values.
• The annualized riskless rate is r = 5%. National Taiwan University Page 641 . YuhDauh Lyuu. • The put is exercisable at years 0. • We use only 8 price paths to illustrate the algorithm. – The annual discount factor hence equals 0.951229.A Numerical Example • Consider a 3year American put on a nondividendpaying stock. 1. and 3. • The spot stock price is 101. c 2007 Prof. 2. • The strike price X = 105.
3600 104.3788 100.2103 105.A Numerical Example (continued) Stock price paths Path 1 2 3 4 5 6 7 8 Year 0 101 101 101 101 101 101 101 101 Year 1 97.6010 98.1763 103.1475 Year 2 92.7802 96.5815 105.5315 99.0564 93.7120 101.8375 108.9225 115.2345 95.0994 c 2007 Prof.4411 124.5115 108.4524 124.9554 104. National Taiwan University Page 642 .4177 113.2516 Year 3 107.5178 102.7270 102.6424 101. YuhDauh Lyuu.
5 3 95 c 2007 Prof.5 1 1. YuhDauh Lyuu.125 5 3 120 115 7 8 4 2 110 105 100 1 6 0 0.5 2 2. National Taiwan University Page 643 .
A Numerical Example (continued) • We use the basis functions 1. Legendre polynomials. National Taiwan University Page 644 . • We now proceed to tackle our problem. using information from all paths. Hermite polynomials. x. • Our concrete problem is to calculate the cash ﬂow along each path. and Jacobi polynomials.a • The plot next page shows the ﬁnal estimated optimal exercise strategy given by LSM. x2 . YuhDauh Lyuu. Gedenbauer polynomials. a Laguerre c 2007 Prof. Chebyshev polynomials. – Other basis functions are possible. polynomials.
5 1 1.5 2 6 1 2.5 3 c 2007 Prof. National Taiwan University Page 645 .125 120 115 110 8 105 3 7 5 4 2 100 95 0 0. YuhDauh Lyuu.
0775 0 c 2007 Prof.A Numerical Example (continued) Cash ﬂows at year 3 Path 1 2 3 4 5 6 7 8 Year 0 — — — — — — — — Year 1 — — — — — — — — Year 2 — — — — — — — — Year 3 0 2. National Taiwan University Page 646 .4685 5. YuhDauh Lyuu.6212 4.5476 0 0 0.
the European counterpart has a value of 2. National Taiwan University Page 647 . • Only 4 paths are in the money: 2.3680. YuhDauh Lyuu. 5.6212 + 4.5476 + 0. 7. 0.0775 = 1. 6.A Numerical Example (continued) • The cash ﬂows at year 3 are the exercise value if the put is in the money. which we will ﬁnd out step by step. • Some of the cash ﬂows may not occur if the put is exercised earlier.4685 + 5.951229 × 8 3 c 2007 Prof. • Incidentally.
A Numerical Example (continued) • We move on to year 2. we must decide whether to exercise it. • Only inthemoney paths will be used in the regression because they are where early exercise is relevant. • For each state that is in the money at year 2. we would move on to year 1. • There are 6 paths for which the put is in the money: 1. 3. 6. YuhDauh Lyuu. 4. 7. c 2007 Prof. – If there were none. National Taiwan University Page 648 . 5.
YuhDauh Lyuu. c 2007 Prof. • Let y denote the corresponding discounted future cash ﬂows (at year 3) if the put is not exercised at year 2. National Taiwan University Page 649 .A Numerical Example (continued) • Let x denote the stock prices at year 2 for those 6 paths.
951229 0.7270 102. YuhDauh Lyuu.951229 — 0 × 0.951229 5.4177 — y 0 × 0.951229 0 × 0.4685 × 0.951229 — c 2007 Prof. National Taiwan University Page 650 .951229 4.6212 × 0.0564 93.7120 101.0775 × 0.A Numerical Example (continued) Regression at year 2 Path 1 2 3 4 5 6 7 8 x 92.5815 — 103.6010 98.
00106918 × x2 . National Taiwan University Page 651 . x.313114 × x + 0. • The result is f (x) = 22. and x2 . c 2007 Prof.A Numerical Example (continued) • We regress y on 1. • We next compare the immediate exercise value and the continuation value. YuhDauh Lyuu. • f estimates the continuation value conditional on the stock price at year 2.08 − 0.
2558 — f (103.4185 — 1.0564) = 1.3990 6. YuhDauh Lyuu.1168 f (98.3568 f (93.9436 11.6010) = 1.7270) = 2.5901 f (101.1253 f (102. National Taiwan University Page 652 .2880 3.5815) = 2.5823 — Continuation f (92.A Numerical Example (continued) Optimal early exercise decision at year 2 Path 1 2 3 4 5 6 7 8 Exercise 12.7120) = 1.4177) = 0.3326 — c 2007 Prof.2730 2.
7. 4. c 2007 Prof. 6. 7. • Hence the cash ﬂows on p. YuhDauh Lyuu. 646 become the next ones. the put should be exercised in all 6 paths: 1. National Taiwan University Page 653 .A Numerical Example (continued) • Amazingly. any positive cash ﬂow at year 3 should be set to zero for these paths as the put is exercised before year 3. 5. 6. – They are paths 5. • Now. 3.
YuhDauh Lyuu.A Numerical Example (continued) Cash ﬂows at years 2 & 3 Path 1 2 3 4 5 6 7 8 Year 0 — — — — — — — — Year 1 — — — — — — — — Year 2 12.4185 0 1.5476 0 0 0 0 0 0 c 2007 Prof.2730 2.5823 0 Year 3 0 2.3990 6.9436 11. National Taiwan University Page 654 .2880 3.
c 2007 Prof. • For each state that is in the money at year 1. we must decide whether to exercise it. 6.A Numerical Example (continued) • We move on to year 1. 2. YuhDauh Lyuu. we would move on to year 0. – If there were none. • There are 5 paths for which the put is in the money: 1. 8. • Only inthemoney paths will be used in the regression because they are where early exercise is relevant. National Taiwan University Page 655 . 4.
A Numerical Example (continued) • Let x denote the stock prices at year 1 for those 5 paths. • From p. 654. YuhDauh Lyuu. we have the following table. c 2007 Prof. • Let y denote the corresponding discounted future cash ﬂows if the put is not exercised at year 1. National Taiwan University Page 656 .
YuhDauh Lyuu.8375 — 104.2103 — 96.5476 × 0.2880 × 0.9512292 — 6.951229 — 0 c 2007 Prof.4185 × 0.1475 y 12.A Numerical Example (continued) Regression at year 1 Path 1 2 3 4 5 6 7 8 x 97. National Taiwan University Page 657 .2730 × 0.4411 — 95.951229 2.951229 — 11.6424 101.
c 2007 Prof. National Taiwan University Page 658 . x. and x2 . • f estimates the continuation value conditional on the stock price at year 1.0551567 × x2 . • We next compare the immediate exercise value and the continuation value. • The result is f (x) = −420.964 + 9. YuhDauh Lyuu.78113 × x − 0.A Numerical Example (continued) • We regress y on 1.
1475) = −0.551885 c 2007 Prof.6424) = 8.A Numerical Example (continued) Optimal early exercise decision at year 1 Path 1 2 3 4 5 6 7 8 Exercise 7.2230 f (101.8525 Continuation f (97.7897 — 8.1625 — 0.8375) = 9.5589 — 9. National Taiwan University Page 659 .9882 — f (96.2103) = 3.3576 3.3329 — f (95.83042 — f (104.4411) = 9. YuhDauh Lyuu.
654 become the next ones. • Now. • Hence the cash ﬂows on p. – But there is none. 645. National Taiwan University Page 660 . c 2007 Prof. • They also conﬁrm the plot on p.A Numerical Example (continued) • The put should be exercised for 1 path only: 8. any positive future cash ﬂow should be set to zero for this path as the put is exercised before years 2 and 3. YuhDauh Lyuu.
4185 0 1.5823 0 Year 3 0 2.A Numerical Example (continued) Cash ﬂows at years 1. YuhDauh Lyuu.5476 0 0 0 0 0 0 c 2007 Prof.2730 2. National Taiwan University Page 661 .2880 3. 2.8525 Year 2 12.3990 6. & 3 Path 1 2 3 4 5 6 7 8 Year 0 — — — — — — — — Year 1 0 0 0 0 0 0 0 0.9436 11.
• The continuation value is. YuhDauh Lyuu.2880 × 0.5476 × 0.9512292 + 0. from p 661.9436 × 0.3990 × 0.9512292 + 11.9512293 +1.66263.9512292 + 2. National Taiwan University Page 662 .5823 × 0.9512292 + 6.2730 × 0.9512292 +2. c 2007 Prof.4185 × 0.951229)/8 = 4.8525 × 0.A Numerical Example (continued) • We move on to year 0. (12.9512292 +3.
66263. • Compare this to the European put’s value of 1.A Numerical Example (concluded) • As this is larger than the immediate exercise value of 105 − 101 = 4. 647). YuhDauh Lyuu. National Taiwan University Page 663 . c 2007 Prof.3680 (p. the put should not be exercised at year 0. • Hence the put’s value is estimated to be 4.
Time Series Analysis c 2007 Prof. YuhDauh Lyuu. National Taiwan University Page 664 .
The historian is a prophet in reverse. — Friedrich von Schlegel (1772–1829) c 2007 Prof. National Taiwan University Page 665 . YuhDauh Lyuu.
. t is a stationary. its conditional variance may vary. ]. . National Taiwan University Page 666 . uncorrelated process with zero mean and constant variance σ 2 . equals σ 2 . Var[ Xt  Xt−1 .Conditional Variance Models for Price Volatility • Although a stationary model (see text for deﬁnition) has constant variance. . YuhDauh Lyuu. Xt−2 . which is smaller than the unconditional variance Var[ Xt ] = σ 2 /(1 − a2 ). c 2007 Prof. t – Here. • The conditional variance. • Take for example an AR(1) process Xt = aXt−1 + with  a  < 1.
. • Then Vt can be computed given the information set of past returns: It−1 ≡ { Xt−1 . . }. Xt−2 . YuhDauh Lyuu. .Conditional Variance Models for Price Volatility (concluded) • In the lognormal model. . • Suppose we assume that conditional variances are deterministic functions of past returns: Vt = f (Xt−1 . ) for some function f . Xt−2 . . c 2007 Prof. National Taiwan University Page 667 . . the conditional variance evolves independently of past returns.
uncorrelated process. YuhDauh Lyuu. National Taiwan University Page 668 . • Assume that { Ut } is a Gaussian stationary. c 2007 Prof. cowinner of the 2003 Nobel Prize in Economic Sci ences. a Engle (1982).ARCH Modelsa • An inﬂuential model in this direction is the autoregressive conditional heteroskedastic (ARCH) model.
• The variance Vt2 satisﬁes p Vt2 = a0 + i=1 ai (Xt−i − µ)2 .ARCH Models (continued) • The ARCH(p) process is deﬁned by p 1/2 Xt − µ = a0 + i=1 ai (Xt−i − µ)2 Ut . where a1 . . Vt2 ). . – Thus Xt  It−1 ∼ N (µ. National Taiwan University Page 669 . YuhDauh Lyuu. . c 2007 Prof. • The volatility at time t as estimated at time t − 1 depends on the p most recent observations on squared returns. ap ≥ 0 and a0 > 0. .
c 2007 Prof. • For it. National Taiwan University Page 670 .ARCH Models (concluded) • The ARCH(1) process Xt − µ = (a0 + a1 (Xt−1 − µ)2 )1/2 Ut is the simplest. • The process { Xt } is stationary with ﬁnite variance if and only if a1 < 1. Var[ Xt  Xt−1 = xt−1 ] = a0 + a1 (xt−1 − µ)2 . YuhDauh Lyuu. in which case Var[ Xt ] = a0 /(1 − a1 ).
Taylor (1986). 2 • The simplest GARCH(1. resulting in 2 Vt2 = a0 + a1 (Xt−1 − µ)2 + a2 Vt−1 . National Taiwan University Page 671 . • The volatility at time t as estimated at time t − 1 depends on the squared return and the estimated volatility at time t − 1. a Bollerslev (1986). c 2007 Prof. 1) process adds a2 Vt−1 to the ARCH(1) process.GARCH Modelsa • A very popular extension of the ARCH model is the generalized autoregressive conditional heteroskedastic (GARCH) process. YuhDauh Lyuu.
• A popular special case of GARCH(1. in which case the unconditional. YuhDauh Lyuu. • It is usually assumed that a1 + a2 < 1 and a0 > 0. • This model is used in J. c 2007 Prof.P. which sets a0 to zero and a2 to 1 − a1 .GARCH Models (concluded) • The estimate of volatility averages past squared returns by giving heavier weights to recent squared returns (see text). 1) is the exponentially weighted moving average process. Morgan’s RiskMetricsTM. National Taiwan University Page 672 . longrun variance is given by a0 /(1 − a1 − a2 ).
t + 1 ] given the information at date t. • Let h2 be the conditional variance of the return over t the period [ t. National Taiwan University Page 673 . – “One day” is merely a convenient term for any elapsed time ∆t. • Let St denote the asset price at date t. c 2007 Prof.GARCH Option Pricing • Options can be priced when the underlying asset’s return follows a GARCH process. YuhDauh Lyuu.
YuhDauh Lyuu. 0. (66) = ∼ = β0 + β1 h2 + β2 h2 ( t t t+1 − c)2 . daily riskless return. c 2007 Prof.GARCH Option Pricing (continued) • Adopt the following riskneutral process for the price dynamics:a h2 St+1 = r − t + ht ln St 2 where h2 t+1 t+1 t+1 . (67) N (0. National Taiwan University Page 674 . 1) given information at date t. r c ≥ a Duan (1995).
1) model (see text). • The above process. National Taiwan University Page 675 . β1 . YuhDauh Lyuu. called the nonlinear asymmetric GARCH model.GARCH Option Pricing (continued) • The ﬁve unknown parameters of the model are c. h0 . • It is postulated that β0 . c 2007 Prof. β1 . generalizes the GARCH(1. β0 . and β2 . β2 ≥ 0 to make the conditional variance positive.
c 2007 Prof. ” b Noted by Black (1976): Volatility tends to rise in response to “bad news” and fall in response to “good news. and so on. . National Taiwan University Page 676 . YuhDauh Lyuu.a – When c = 0. a large t+1 results in a large ht+1 . which in turns tends to yield a large ht+2 .b – For c > 0. whereas a negative t+1 (bad news) tends to do the opposite. a positive t+1 (good news) tends to decrease ht+1 .GARCH Option Pricing (continued) • It captures the volatility clustering in asset returns ﬁrst noted by Mandelbrot (1963). . .” a “. • It also captures the negative correlation between the asset return and changes in its (conditional) volatility. . large changes tend to be followed by large changes—of either sign—and small changes tend to be followed by small changes .
t 2 Var[ yt+1  yt . t t (69) (70) c 2007 Prof. the model becomes yt+1 h2 = y t + r − t + ht 2 t+1 . h2 ) completely describes the current state. National Taiwan University Page 677 . h2 ] = yt + r − t . YuhDauh Lyuu. h2 ] = h2 . (68) • The pair (yt .GARCH Option Pricing (concluded) • With yt ≡ ln St denoting the logarithmic price. t • The conditional mean and variance of yt+1 are clearly h2 E[ yt+1  yt .
• Path dependence in GARCH makes the tree for asset prices explode exponentially (why?). YuhDauh Lyuu. • We need to mitigate this combinatorial explosion. a Ritchken and Trevor (1999). we turn to trees with discrete states. c 2007 Prof. National Taiwan University Page 678 . • To approximate it.The RitchkenTrevor (RT) Algorithma • The GARCH model is a continuousstate model.
• These 2n + 1 values must approximate the distribution of (yt+1 . h2 ). 539). t+1 • So the conditional moments (69)–(70) at date t + 1 on p. • Three states follow each state (yt . National Taiwan University Page 679 . h2 ) after a period. c 2007 Prof.The RitchkenTrevor Algorithm (continued) • Partition a day into n periods. YuhDauh Lyuu. 2n + 1 states at date t + 1 follow each state at date t (recall p. t • As the trinomial model combines. 677 must be matched by the trinomial model to guarantee convergence to the continuousstate model.
though other multiples of h0 are possible. c 2007 Prof. National Taiwan University Page 680 . • The role of σ in the BlackScholes option pricing model is played by ht in the GARCH model. a comparable magnitude will be chosen here. n • The jump size will be some integer multiple η of γn . YuhDauh Lyuu. • Deﬁne γ ≡ h0 .The RitchkenTrevor Algorithm (continued) • It remains to pick the jump size and the three branching probabilities. • We call η the jump parameter (p. and γ γn ≡ √ . √ • As a jump size proportional to σ/ n is picked in the BOPM. 681).
YuhDauh Lyuu.(1. 0) yt (1. c 2007 Prof. 0) 6n ηγ ? (1. −1) 1 day The seven values on the right approximate the distribution of logarithmic price yt+1 . 1) (0. National Taiwan University Page 681 .
National Taiwan University Page 682 . = − 2η 2 γ 2 2ηγ n c 2007 Prof. YuhDauh Lyuu. and down branches are pu pm pd = h2 r − (h2 /2) t t √ . middle. + 2η 2 γ 2 2ηγ n (71) (72) (73) h2 = 1 − 2t 2.The RitchkenTrevor Algorithm (continued) • The middle branch does not change the underlying asset’s price. • The probabilities for the up. η γ h2 r − (h2 /2) t t √ .
The RitchkenTrevor Algorithm (continued) • It can be shown that: – The trinomial model takes on 2n + 1 values at date t + 1 for yt+1 . – These values have an asymptotically matching variance for yt+1 . National Taiwan University Page 683 . • The central limit theorem thus guarantees the desired convergence as n increases. YuhDauh Lyuu. c 2007 Prof. – These values have a matching mean for yt+1 .
• The resulting model is multinomial with 2n + 1 branches from any state (yt . – Keeping the interdate nodes results in a tree that can be as much as n times larger. h2 ). c 2007 Prof. t • There are two reasons behind this manipulation. National Taiwan University Page 684 . – Interdate nodes are created merely to approximate the continuousstate model after one day. 685). YuhDauh Lyuu.The RitchkenTrevor Algorithm (continued) • We can dispense with the intermediate nodes between dates to create a (2n + 1)nomial tree (p.
c 2007 Prof. National Taiwan University Page 685 .6n ηγ ? yt 1 day This heptanomial tree is the outcome of the trinomial tree on p. 681 after its intermediate nodes are removed. YuhDauh Lyuu.
ju ! jm ! jd ! u m d = ju − jd . the number of up moves must exceed that of down moves by exactly . with ju . • To reach that price in n periods. and c 2007 Prof.jd n! pju pjm pjd . n = ju + jm + jd . National Taiwan University Page 686 .The RitchkenTrevor Algorithm (continued) • A node with logarithmic price yt + ηγn at date t + 1 follows the current node at date t with price yt for some −n ≤ ≤ n. YuhDauh Lyuu.jm . • The probability that this happens is P( ) ≡ ju . jm . jd ≥ 0.
• It can be computed in O(n2 ) time. c 2007 Prof.The RitchkenTrevor Algorithm (continued) • A particularly simple way to calculate the P ( )s starts by noting that n (pu x + pm + pd x−1 )n = =−n P( )x . (74) • So we expand (pu x + pm + pd x−1 )n and retrieve the probabilities by reading oﬀ the coeﬃcients. National Taiwan University Page 687 . YuhDauh Lyuu.
h2 ) at date t has a variance equal to t h2 = β0 + β1 h2 + β2 h2 ( t+1 t t – Above. . ±1. t+1 t+1 − c)2 .The RitchkenTrevor Algorithm (continued) • The updating rule (67) on p. . (75) ηγn − (r − h2 /2) t . YuhDauh Lyuu. National Taiwan University Page 688 . . ±n. . = ht = 0. ±2. c 2007 Prof. 674 must be modiﬁed to account for the adoption of the discretestate model. is a discrete random variable with 2n + 1 values. • The logarithmic price yt + ηγn at date t + 1 following state (yt .
(71)–(73) on p. National Taiwan University Page 689 . c 2007 Prof. ht /γ + 2. • The necessary requirement pm ≥ 0 implies η ≥ ht /γ. YuhDauh Lyuu. . until valid probabilities are obtained or until their nonexistence is conﬁrmed. ht /γ + 1. . . • This implies varying jump sizes. • Hence we try η = ht /γ .The RitchkenTrevor Algorithm (continued) • Diﬀerent conditional variances h2 may require diﬀerent t η so that the probabilities calculated by Eqs. 682 lie between 0 and 1.
National Taiwan University Page 690 . 691 uses n = 1 to illustrate our points for a 3day model. • The plot on p. YuhDauh Lyuu. the magnitude of η tends to grow with ht .The RitchkenTrevor Algorithm (continued) • The suﬃcient and necessary condition for valid probabilities to exist isa h2  r − (h2 /2)  1  r − (h2 /2)  t t t √ √ ≤ 2 2 ≤ min 1 − . node (1. • Obviously. 1) of date 1 and node (2. a Lyuu and Wu (2003). • For example. 3) of date 2 pick η = 2. 2η γ 2 2ηγ n 2ηγ n . c 2007 Prof.
0) 6 n = γ1 γ ? (2. 3) (1. −1) 3 days  c 2007 Prof. YuhDauh Lyuu.(2. 1) y0 (2. National Taiwan University Page 691 .
National Taiwan University Page 692 . – Two paths can reach node (2. 0) from the root node. 0) and (2. c 2007 Prof. • For example. – One of the variances results in η = 1.The RitchkenTrevor Algorithm (continued) • The topology of the tree is not a standard combining multinomial tree. 691 such as nodes (2. −1) have multiple jump sizes. a few nodes on p. YuhDauh Lyuu. • The reason is the path dependence of the model. each with a diﬀerent variance for the node. whereas the other results in η = 2.
a Cakici and Topyan (2000). • To address this problem. c 2007 Prof. National Taiwan University Page 693 . we record only the maximum and minimum h2 at each node. h2 ) carries its own η and max min set of 2n + 1 branching probabilities. YuhDauh Lyuu. max min • Each of (yt . h2 ) and (yt .The RitchkenTrevor Algorithm (concluded) • The possible values of h2 at a node are exponential t nature. h2 ) and (yt . h2 ).a t • Therefore. each node on the tree contains only two states (yt .
b and Wu (2003). National Taiwan University Page 694 .Negative Aspects of the RitchkenTrevor Algorithma • A small n may yield inaccurate option prices. • Thus the choice of n may be limited in practice. – Speciﬁcally. Lyuu and Wu (2005). • A large n has another serious problem: The tree cannot grow beyond a certain date. YuhDauh Lyuu. • But the tree will grow exponentially if n is large enough. n > (1 − β1 )/β2 when r = c = 0. • The RT algorithm can be modiﬁed to be free of shortened maturity and (to some extent) exponential complexity. b It is only quadratic if n is not too large! a Lyuu c 2007 Prof.
β0 = 0. YuhDauh Lyuu. • A daily variance of 0. and c = 0. n = 1.000006575. h2 = 0.04. 0 c 2007 Prof.0001096 corresponds to an annual √ volatility of 365 × 0. h2 (0. j).60517. j) denote the variance at node (i.9.0001096. y0 = ln S0 = 4. • Initially. β1 = 0. β2 = 0. r = 0.010469.0001096 ≈ 20%. • Let h2 (i. 0) = h2 = 0. National Taiwan University Page 695 .0001096.010469. γ = h0 = 0.Numerical Examples • Assume S0 = 100. 0 √ γn = γ/ n = 0.
Numerical Examples (continued)
• Let h2 (i, j) denote the maximum variance at node max (i, j). • Let h2 (i, j) denote the minimum variance at node min (i, j). • Initially, h2 (0, 0) = h2 (0, 0) = h2 . max 0 min • The resulting threeday tree is depicted on p. 697.
c 2007 Prof. YuhDauh Lyuu, National Taiwan University
Page 696
yt
4.65752 13.4809 13.4809
4.64705
4.63658
12.2883 2 12.2883 2
11.7170 11.7170 10.5733 12.2846
4.62611
4.61564 0.01047 4.60517 10.9600 1 10.9600 1
10.9645 2 10.9645 2 10.5215 1 10.5215 1 10.9553 1 10.9553 1
10.5256 1 10.5697 1 10.1269 1 10.9603 2 10.5173 1 12.2700 2 10.9511 1 10.9511 1
10.1305 13.4644 09.7717 10.6042 10.1231 11.7005 10.5135 12.2662 10.9473 13.4438
4.59470
4.58423
4.57376
c 2007 Prof. YuhDauh Lyuu, National Taiwan University
Page 697
A top (bottom) number inside a gray box refers to the minimum (maximum, respectively) variance h2 (h2 , max min respectively) for the node. Variances are multiplied by 100,000 for readability. A top (bottom) number inside a white box refers to η corresponding to h2 (h2 , max min respectively).
c 2007 Prof. YuhDauh Lyuu, National Taiwan University
Page 698
0. • Start with the root node. 0. 0). (71)–(73) on p. YuhDauh Lyuu.4974.Numerical Examples (continued) • Let us see how the numbers are calculated. National Taiwan University Page 699 . node (0. • Try η = 1 in Eqs. the three branches from the root node use single jumps. c 2007 Prof. • As they are valid probabilities.5026. 682 ﬁrst to obtain pu pm pd = = = 0.
National Taiwan University Page 700 .Numerical Examples (continued) • Move on to node (1. • It has one predecessor node—node (0. 1). 0)—and it takes an up move to reach the current node. 0). = 1 and c 2007 Prof. 1) = 0.000109645. t • The result is h2 (1. YuhDauh Lyuu. • So apply updating rule (75) on p. 688 with h2 = h2 (0.
YuhDauh Lyuu. 0.1237. we try η = 2 in Eqs. 682 ﬁrst to obtain pu pm pd = = = 0. the three branches from node (1.7499. 1)/γ = 2. • As they are valid probabilities. c 2007 Prof. 1) use double jumps.1264. National Taiwan University Page 701 .Numerical Examples (continued) • Because h(1. (71)–(73) on p. 0.
−1) = 0.Numerical Examples (continued) • Carry out similar calculations for node (1.000109553. • The resulting variances are h2 (1. c 2007 Prof. −1) with = −1 in updating rule (75). 0) with = 0 in updating rule (75) on p. National Taiwan University Page 702 . YuhDauh Lyuu.000105215. 688. • Carry out similar calculations for node (1. 0) = h2 (1. • Single jump η = 1 works in both nodes. 0.
t+1 c 2007 Prof. • Because it takes a middle move to reach the current node. 0) has 2 predecessor nodes.000101269. 0). National Taiwan University Page 703 . we apply updating rule (75) on p. • Let us start with node (1. −1).Numerical Examples (continued) • Node (2. • Both have to be considered in deriving the variances. 688 with = 0 and h2 = h2 (1. YuhDauh Lyuu. (1. 0) and (1. t • The result is h2 = 0. 0).
−1). • Because it takes an up move to reach the current node. apply updating rule (75) on p.000109603. 0.Numerical Examples (continued) • Now move on to the other predecessor node (1. t+1 • We hence record h2 (2. 0) = max 0. c 2007 Prof. 688 with = 1 and h2 = h2 (1. YuhDauh Lyuu.000101269. National Taiwan University Page 704 . −1). 0) = min h2 (2.000109603. t • The result is h2 = 0.
National Taiwan University Page 705 . c 2007 Prof.1263. the three branches from node (2. max • Because hmax (2. (71)–(73) on p. 0) ﬁrst.1237. 0. • As they are valid probabilities. YuhDauh Lyuu. we ﬁrst try η = 2 in Eqs. 682 to obtain pu pm pd = = = 0. 0) with the maximum variance use double jumps.Numerical Examples (continued) • Consider state h2 (2. 0)/γ = 2. 0.7500.
0. YuhDauh Lyuu. c 2007 Prof. 0). we ﬁrst try η = 1 in Eqs. National Taiwan University Page 706 .4596.0760. 682 to obtain pu pm pd = = = 0. • As they are valid probabilities.Numerical Examples (continued) • Now consider state h2 (2. 0)/γ = 1. 0) with the minimum variance use single jumps. the three branches from node (2. 0. min • Because hmin (2.4644. (71)–(73) on p.
• Because it takes a down move to reach the current node. t+1 c 2007 Prof. YuhDauh Lyuu. t • The result is h2 = 0. we apply updating rule (75) on p. −1) has 3 predecessor nodes. 688 with = −1 and h2 = h2 (1.0001227. 1). National Taiwan University Page 707 . 1).Numerical Examples (continued) • Node (2. • Start with node (1.
National Taiwan University Page 708 . • Because it also takes a down move to reach the current node.000105609. 688 with = −1 and h2 = h2 (1. 0). t • The result is h2 = 0.Numerical Examples (continued) • Now move on to predecessor node (1. YuhDauh Lyuu. 0). t+1 c 2007 Prof. we apply updating rule (75) on p.
t+1 • We hence record h2 (2. c 2007 Prof. −1) max = = 0. −1). YuhDauh Lyuu. t • The result is h2 = 0. we apply updating rule (75) on p. 0.0001227. −1).Numerical Examples (continued) • Finally. 688 with = 0 and h2 = h2 (1.000105173. National Taiwan University Page 709 . • Because it takes a middle move to reach the current node. −1) min h2 (2. consider predecessor node (1.000105173.
the three branches from node (2. 682 to obtain pu pm pd = = = 0.Numerical Examples (continued) • Consider state h2 (2. max • Because hmax (2. −1). we ﬁrst try η = 2 in Eqs. YuhDauh Lyuu.1414. −1) with the maximum variance use double jumps. 0. −1)/γ = 2.1385. (71)–(73) on p.7201. c 2007 Prof. 0. • As they are valid probabilities. National Taiwan University Page 710 .
0404. −1) with the minimum variance use single jumps. • As they are valid probabilities. consider state h2 (2. 0. −1)/γ = 1. (71)–(73) on p. 0. min • Because hmin (2. 682 to obtain pu pm pd = = = 0. the three branches from node (2.Numerical Examples (continued) • Next. YuhDauh Lyuu. National Taiwan University Page 711 .4823.4773. we ﬁrst try η = 1 in Eqs. c 2007 Prof. −1).
• But only the maximum and minimum variances will be kept. then 2k variances will be calculated using the updating rule. National Taiwan University Page 712 . YuhDauh Lyuu.Numerical Examples (concluded) • Other nodes at dates 2 and 3 can be handled similarly. c 2007 Prof. – This is because each predecessor node keeps two variance numbers. • In general. if a node has k predecessor nodes.
Negative Aspects of the RT Algorithm Revisiteda • Recall the problems mentioned on p. National Taiwan University Page 713 .9 n> = = 2. 694.5.04 • Suppose we are willing to accept the exponential running time and pick n = 100 to seek accuracy. c 2007 Prof. YuhDauh Lyuu. • But the problem of shortened maturity forces the tree to stop at date 9! a Lyuu and Wu (2003). combinatorial explosion occurs when 1 − β1 1 − 0. β2 0. • In our case.
dashed line: n = 4. solid line: n = 5. YuhDauh Lyuu. c 2007 Prof. National Taiwan University Page 714 .25000 20000 15000 10000 5000 25 50 75 100 125 150 175 Date Dotted line: n = 3.
Backward Induction on the RT Tree • After the RT tree is constructed. it can be used to price options by backward induction. a In c 2007 Prof. max min • Besides the minimum and maximum variances. YuhDauh Lyuu. National Taiwan University Page 715 . the other K − 2 variances in between are linearly interpolated. Logcubic interpolation works even better (Liu (2005)). loglinear interpolation works better (Lyuu and Wu (2005)). min • We now increase that number to K equally spaced variances between h2 and h2 at each node.a practice. • Recall that each node keeps two variances h2 max and h2 .
j) is h2 (i. K −1 c 2007 Prof. j) − h2 (i.Backward Induction on the RT Tree (continued) • For example. YuhDauh Lyuu. • Each interpolated variance’s jump parameter and branching probabilities can be computed as before. 1. 697. h2 (i. then a variance of 10. the kth variance at node (i. K − 1.5436 × 10−6 will be added between the maximum and minimum variances at node (2. National Taiwan University Page 716 . • In general. if K = 3. . . . 0) on p. j) min + k max . . j) min k = 0.
Backward Induction on the RT Tree (concluded) • During backward induction. National Taiwan University Page 717 . c 2007 Prof. linear interpolation of the option prices corresponding to the two bracketing variances will be used as the approximate option price. YuhDauh Lyuu. • The above ideas are reminiscent of the ones on p. 332. if a variance falls between two of the K variances. where we dealt with arithmetic averagerate options.
• The pricing tree is shown on p. • Recall that the riskless interest rate is zero.Numerical Examples • We next use the numerical example on p. 720.66346. National Taiwan University Page 718 . c 2007 Prof. 697 to price a European call option with a strike price of 100 and expiring at date 3. 719 with a call price of 0. YuhDauh Lyuu. – The branching probabilities needed in backward induction can be found on p. • Assume K = 2. hence there are no interpolated variances.
14573 2 0.00000 1 1. National Taiwan University Page 719 .05240 1 1.19054 3.05240 1.19054 2 3.66346 1 0.11587 101.52360 1 0.00000 100.92797 96.00000 0.05240 1.52360 1 0.St 105.95856 97.19054 2 3.00000 0.00000 1 0.00000 0.00000 1 0.00000 0.00000 0. YuhDauh Lyuu.27652 103.00000 0.11587 2.00000 98.20241 2 1.20241 2 0.00000 0.37392 104.48366 1 0.90811 c 2007 Prof.13012 1 1.05240 1 0.19054 3.66346 1 0.05240 0.19054 2.26172 2 0.11587 102.13012 1 0.37392 5.37392 5.
5215 10.1237 0. YuhDauh Lyuu.1387 0.1387 0.5022 1 0 − 1 −2 c 2007 Prof.4970 0.9511 10.9553 10.1416 0.5024 0.1263 0.0396 0.4972 0.4823 0.1237 0.4775 0.1264 0.9645 10.2883 11.5173 12.1264 0.7170 10.0000 0.9600 p[0][ ][][ ] 0.5733 12.0760 0.0400 0. National Taiwan University Page 720 .4644 0.4773 0.5026 0.6042 2 η[i][j][0] η[i][j][1] η 2][][ ] [ 2 2 j 3 2 η[1][][ ] 2 η[0][ ][ ] 1 1 2 1 1 1 1 1 1 1 2 1 2 1 1 p[2][][][ ] 1 0 − 1 −2 j 3 2 j 5 4 3 2 1 0 − 1 − 2 −3 h2[i][j][0] h2[i][j][1] p[i][j][0 1] ][ p[i][j][0][0] p[i][j][0][−1] p[i][j]1][ ] [ 1 p[i][j][1][0] p[i][j][1][−1] p[1][][][ ] 0.5256 10.9600 10.7201 0.5024 0.4809 13.4809 [ h [2] ][ ] 12.4827 0.0404 0.9511 10.2700 10.1269 10.0004 0.9553 10.4775 0.4777 0.1305 10.4972 0.4797 0.5022 0.rb[i][0] rb[i][1] rb[0][ ] 0 0 rb[1][ ] −1 1 rb[2][ ] −2 3 rb[3][ ] −3 5 [ h2[3] ][ ] 13.7197 0.4974 0.2846 h2[1] ][ ] [ 10.1385 0.5697 13.0400 0.4825 0.4847 0.2662 10.0000 0.0008 0.7170 12.4438 10.9645 10.9603 10.2883 11.7500 0.1414 0.4596 0.1416 h2[0] ][ ] [ 10.0004 0.7717 10.5026 0.4970 0.9473 13.5215 10.7197 0.0356 0.4825 0.7499 0.5135 12.1231 11.1237 0.4974 0.0008 0.7005 09.7499 0.4644 10.
National Taiwan University Page 721 .05240 = 3. • The option price at node (2.1387 × 5. 5).37392 + 0. (3.19054. 0). 1).7197 × 3. YuhDauh Lyuu. 3).19054 + 0.1416 × 1. • It therefore equals 0. • Now move on to nodes at date 2. independent of the variance level. and (3.Numerical Examples (continued) • Let us derive some of the numbers on p. • Option prices for other nodes at date 2 can be computed similarly. 719. c 2007 Prof. 3) depends on those at nodes (3. • The option price for a terminal node at date 3 equals max(S3 − 100.
Numerical Examples (continued)
• For node (1, 1), the option price for both variances is
0.1237 × 3.19054 + 0.7499 × 1.05240 + 0.1264 × 0.14573 = 1.20241.
• Node (1, 0) is most interesting. • We knew that a down move from it gives a variance of 0.000105609. • This number falls between the minimum variance 0.000105173 and the maximum variance 0.0001227 at node (2, −1) on p. 697.
c 2007 Prof. YuhDauh Lyuu, National Taiwan University
Page 722
Numerical Examples (continued)
• The option price corresponding to the minimum variance is 0. • The option price corresponding to the maximum variance is 0.14573. • The equation x × 0.000105173 + (1 − x) × 0.0001227 = 0.000105609 is satisﬁed by x = 0.9751. • So the option for the down state is approximated by x × 0 + (1 − x) × 0.14573 = 0.00362.
c 2007 Prof. YuhDauh Lyuu, National Taiwan University
Page 723
Numerical Examples (continued)
• The up move leads to the state with option price 1.05240. • The middle move leads to the state with option price 0.48366. • The option price at node (1, 0) is ﬁnally calculated as
0.4775 × 1.05240 + 0.0400 × 0.48366 + 0.4825 × 0.00362 = 0.52360.
c 2007 Prof. YuhDauh Lyuu, National Taiwan University
Page 724
the option price corresponding to the maximum or minimum variance will be used during backward induction. YuhDauh Lyuu. a Lyuu and Wu (2005). • An interpolated variance may choose a branch that goes into a node that is not reached in the forwardinduction treebuilding phase. National Taiwan University Page 725 .a • In this case. the algorithm fails. c 2007 Prof. • When this happens.Numerical Examples (concluded) • It is possible for some of the three variances following an interpolated variance to exceed the maximum variance or be exceeded by the minimum variance.
Interest Rate Derivative Securities c 2007 Prof. National Taiwan University Page 726 . YuhDauh Lyuu.
— Richard Wagner (1813–1883). Der Ring des Nibelungen Which shows that gambling’s not a sin provided that you always win. — Roald Dahl (1916–1990). National Taiwan University Page 727 .What you are. you are only by contracts. YuhDauh Lyuu. “SnowWhite and the Seven Dwarfs” c 2007 Prof.
National Taiwan University Page 728 . YuhDauh Lyuu.Term Structure Fitting c 2007 Prof.
YuhDauh Lyuu. they think calculating is inventing. — Johann Wolfgang Goethe (1749–1832) c 2007 Prof. National Taiwan University Page 729 .That’s an old besetting sin.
National Taiwan University Page 730 . YuhDauh Lyuu.Introduction to Term Structure Modeling c 2007 Prof.
The fox often ran to the hole by which they had come in. to ﬁnd out if his body was still thin enough to slip through it. — Grimm’s Fairy Tales c 2007 Prof. National Taiwan University Page 731 . YuhDauh Lyuu.
National Taiwan University Page 732 . • Although the idea is similar to the earlier one used in option pricing.Outline • Use the binomial interest rate tree to model stochastic term structure. YuhDauh Lyuu. – Applications are generic in that pricing and hedging methodologies can be easily adapted to other models. the current task is more complicated. is to be modeled. – Interest rates of various maturities cannot evolve arbitrarily or arbitrage proﬁts may occur. c 2007 Prof. – Illustrates the basic ideas underlying future models. – The evolution of an entire term structure. not just a single stock price.
• The unbiased expectations theory. National Taiwan University Page 733 . – “Consistent” with the observed term structures. the liquidity preference theory. c 2007 Prof. and the market segmentation theory can all be made consistent with the model. – Provides a stochastic process that deﬁnes future term structures without arbitrage proﬁts. YuhDauh Lyuu.Issues • A stochastic interest rate model performs two tasks.
• An alternative approach pioneered by Ho and Lee (1986) makes ﬁtting the market yield curve mandatory. • Early models have ﬁtting problems because they may not price today’s benchmark bonds correctly. National Taiwan University Page 734 .History • Methodology founded by Merton (1970). YuhDauh Lyuu. • Modern interest rate modeling is often traced to 1977 when Vasicek and Cox. Ingersoll. c 2007 Prof. and Ross developed simultaneously their inﬂuential models. • Models based on such a paradigm are called (somewhat misleadingly) arbitragefree or noarbitrage models.
” c 2007 Prof. a Derman (2004). YuhDauh Lyuu. – This procedure is called calibration.Binomial Interest Rate Tree • Goal is to construct a noarbitrage interest rate tree consistent with the yields and/or yield volatilities of zerocoupon bonds of all maturities.a • Pick a binomial tree model in which the logarithm of the future short rate obeys the binomial distribution. National Taiwan University Page 735 . – Exactly like the CRR tree. • The limiting distribution of the short rate at any future time is hence lognormal. “complexity without calibration is pointless.
c 2007 Prof.Binomial Interest Rate Tree (continued) • A binomial tree of future short rates is constructed. • Every short rate is followed by two short rates in the following period (see next page). • In the ﬁgure on p. YuhDauh Lyuu. 737 node A coincides with the start of period j during which the short rate r is in eﬀect. National Taiwan University Page 736 .
0.5 period j − 1 period j B * r j rh C period j + 1 c 2007 Prof. YuhDauh Lyuu.5 r A 0. National Taiwan University Page 737 .
YuhDauh Lyuu. c 2007 Prof. • Each branch has a ﬁfty percent chance of occurring in a riskneutral economy.Binomial Interest Rate Tree (continued) • At the conclusion of period j. National Taiwan University Page 738 . – rh : the “high” shortrate outcome at node C. a new short rate goes into eﬀect for period j + 1. • This may take one of two possible values: – r : the “low” shortrate outcome at node B.
Binomial Interest Rate Tree (continued)
• We shall require that the paths combine as the binomial process unfolds. • The short rate r can go to rh and r with equal riskneutral probability 1/2 in a period of length ∆t. • Hence the volatility of ln r after ∆t time is 1 1 σ= √ ln 2 ∆t (see Exercise 23.2.3 in text). • Above, σ is annualized, whereas r and rh are period based. rh r
c 2007 Prof. YuhDauh Lyuu, National Taiwan University
Page 739
Binomial Interest Rate Tree (continued)
• Note that
√ rh 2σ ∆t =e . r
• Thus greater volatility, hence uncertainty, leads to larger rh /r and wider ranges of possible short rates. • The ratio rh /r may depend on time if the volatility is a function of time. • Note that rh /r has nothing to do with the current short rate r if σ is independent of r.
c 2007 Prof. YuhDauh Lyuu, National Taiwan University
Page 740
Binomial Interest Rate Tree (continued)
• In general there are j possible rates in period j,
j−1 2 rj , rj vj , rj vj , . . . , rj vj ,
where vj ≡ e
2σj √ ∆t
(76)
is the multiplicative ratio for the rates in period j (see ﬁgure on next page). • We shall call rj the baseline rates. • The subscript j in σj is meant to emphasize that the short rate volatility may be time dependent.
c 2007 Prof. YuhDauh Lyuu, National Taiwan University
Page 741
YuhDauh Lyuu. National Taiwan University Page 742 .D Baseline rates B A H C H! D H C H!L! D B HL C H!L! D c 2007 Prof.
r(t) = µ(t) eσ(t) W (t) . National Taiwan University Page 743 . this is how the binomial interest rate tree achieves mean reversion. 2 (77) c 2007 Prof. the short rate follows the following process. a declining short rate volatility is usually imposed to preclude the short rate from assuming implausibly high values. • As the expected value of r(t) equals µ(t) eσ(t) (t/2) .Binomial Interest Rate Tree (concluded) • In the limit. in which the (percent) short rate volatility σ(t) is a deterministic function of time. • Incidentally. YuhDauh Lyuu.
Memory Issues • Path independency: The term structure at any node is independent of the path taken to reach it. – Daily interest rate movements for 30 years require roughly (30 × 365)2 /2 ≈ 6 × 107 doubleprecision ﬂoatingpoint numbers (half a gigabyte!). c 2007 Prof.a • Storing the whole tree would have taken up O(n2 ) space. • This takes up only O(n) space. YuhDauh Lyuu. National Taiwan University Page 744 . • So only the baseline rates ri and the multiplicative ratios vi need to be stored in computer memory. n denotes the depth of the tree. a Throughout this chapter.
. for example. National Taiwan University Page 745 . • Now need the annualized rates of return associated with the various riskless bonds that make up the benchmark yield curve and their volatilities.S.Set Things in Motion • The abstract process is now in place. c 2007 Prof. • In the U. the ontherun yield curve obtained by the most recently issued Treasury securities may be used as the benchmark curve. YuhDauh Lyuu.
• The binomial tree should be consistent with both term structures. • Here we focus on the term structure of interest rates. c 2007 Prof. National Taiwan University Page 746 .Set Things in Motion (concluded) • The term structure of (yield) volatilitiesa can be estimated from either the historical data (historical volatility) or interest rate option prices such as cap prices (implied volatility). a Or simply the volatility (term) structure. YuhDauh Lyuu.
Model Term Structures • The model price is computed by backward induction. National Taiwan University Page 747 . c 2007 Prof. respectively. YuhDauh Lyuu. • This takes quadratic time and linear space. 737. • Refer back to the ﬁgure on p. • Given that the values at nodes B and C are PB and PC . 2(1 + r) • We compute the values column by column without explicitly expanding the binomial interest rate tree (see ﬁgure next page). the value at node A is then PB + PC + cash ﬂow at node A.
D H 2 C B A B HL + 2 2 H = B D 2 HL C + 2 2! HL = B 2! D C + ? 2! 2" HL D D 2" Cash flows: + c 2007 Prof. YuhDauh Lyuu. National Taiwan University Page 748 .
2.Term Structure Dynamics • An nperiod zerocoupon bond’s price can be computed by assigning $1 to every node at period n and then applying backward induction. one obtains the market discount function implied by the tree. . • Repeating this step for n = 1. again. YuhDauh Lyuu. – Taking any node in the tree as the current state induces a binomial interest rate tree and. National Taiwan University Page 749 . . • The tree therefore determines a term structure. . • It also contains a term structure dynamics. a term structure. . c 2007 Prof.