A Bond Option Pricing Formula in the ExtendedCox-Ingersoll-Ross Model
Joint work with Zheng Liu, Henry Schellhorn
Presented by Qidi Peng
Institute of Mathematical Sciences, Claremont Graduate University, U.S.
Conference on Stochastic Asymptotics & Applications, Sixth Western
Conference on Mathematical Finance
Sept. 25, 2014
Q.Peng (CGU) Extended CIR Model Sept. 25, 2014 1 / 21
1 The extended CIR (ECIR) modelFrom Vasicek to ECIR modelMotivation
2 First Main Result: Distribution of r(t)
3 Second Main Result: Bond Option Pricing in ECIR
4 Future Work
Q.Peng (CGU) Extended CIR Model Sept. 25, 2014 2 / 21
The extended CIR (ECIR) model
Plan
1 The extended CIR (ECIR) modelFrom Vasicek to ECIR modelMotivation
2 First Main Result: Distribution of r(t)
3 Second Main Result: Bond Option Pricing in ECIR
4 Future Work
Q.Peng (CGU) Extended CIR Model Sept. 25, 2014 3 / 21
The extended CIR (ECIR) model From Vasicek to ECIR model
Vasicek model, start of our travel...
In finance, the Vasicek model (1977) is a mathematical model whichdescribes the evolution of interest rates.
{dr(t) = b(a− r(t))dt + σ dW (t)r(0) = r0 > 0.
Advantages:
- The first one to capture the ”mean reversion” property.- a > 0 presents the long run equilibrium value. b > 0denotes the speed of ”mean reversion”.
Inconvenience:
- The interest rate might be negative.
Q.Peng (CGU) Extended CIR Model Sept. 25, 2014 4 / 21
The extended CIR (ECIR) model From Vasicek to ECIR model
This inconvenience can be fixed by Cox-Ingersoll-Ross model,exponential Vasicek model, Black-Derman-Toy model andBlack-Karasinski model, etc.
We study CIR model, because it is difficult and challenging.
Q.Peng (CGU) Extended CIR Model Sept. 25, 2014 5 / 21
The extended CIR (ECIR) model From Vasicek to ECIR model
CIR Model, ”better” than Vasicek....
The CIR model was first introduced by Cox, Ingersoll and Ross (1985).
{dr(t) = b(a− r(t))dt + σ
√
r(t)dW (t)r(0) = r0 > 0.
Advantages:
− When the dimension (integer) 4abσ2 ≥ 2, the solution is strictly
positive.− The CIR model is not a generalization of Vasicek model, since it isa sum of independent squared Ornstein-Uhlenbeck process (providedthat 4ab
σ2 is integer), while the Vasicek model is anOrnstein-Uhlenbeck process.
Inconveniences:
− The volatility is constant (=⇒ time-dependent model needed).− The assumption that 4ab
σ2 is integer generally doesn’t fit real data(=⇒ extension of noncentral chi-square distribution).
Q.Peng (CGU) Extended CIR Model Sept. 25, 2014 6 / 21
The extended CIR (ECIR) model From Vasicek to ECIR model
ECIR model, more generalized than CIR...
To remedy the inconveniences of CIR model, we consider an extended CIRmodel (Hull and White, 1990)
{dr(t) =
(θ(t)− b(t)r(t)
)dt + σ(t)
√
r(t)dW (t)r(0) = r0 > 0.
Some known results:
(Maghsoodi, 1996) If4θ(t)σ2(t) ≡ d ∈ N/{0}, then
(Σ(0, t)
)−1r(t) ∼ χ2
(
d ,r0e
−∫ t
0 b(u)du
Σ(0, t)
)
,
where Σ(0, t) := 14
∫ t
0 e−∫ t
vb(u)duσ2(v )dv and χ2(a,λ) denotes a
noncentral chi-square distribution with d degrees of freedom andnoncentrality parameter λ.
(Maghsoodi, 1996) When d = 1, r(t) visits 0 a finite number oftimes over any compact time interval; when d > 1, r(t) > 0.
Q.Peng (CGU) Extended CIR Model Sept. 25, 2014 7 / 21
The extended CIR (ECIR) model Motivation
Begin to participate in the project :)
Question 1: The distribution of (Σ(0, t))−1r(t), if 4θ(t)σ2(t)
≡ d ∈ R+?
Question 2: The distribution of (Σ(0, t))−1r(t), if4θ(t)σ2(t) = d(t) ∈ R
+?
Answer to Question 1:
(Σ(0, t)
)−1r(t) ∼ χ2
(
d ,r0e
−∫ t
0b(u)du
Σ(0, t)
)
,
where χ2(a,λ) denotes a noncentral chi-square distribution with d
(real-valued) degrees of freedom and noncentrality parameter λ, i.e.,
E(e iωχ2(a,λ)) =exp( iλω
1−2iω )
(1− 2iω)a/2.
Remark: r(t) ≥ 0, if r0 > 0 and d > 0.
The answer to Question 2 is our first main result.
Q.Peng (CGU) Extended CIR Model Sept. 25, 2014 8 / 21
First Main Result: Distribution of r(t)
Plan
1 The extended CIR (ECIR) modelFrom Vasicek to ECIR modelMotivation
2 First Main Result: Distribution of r(t)
3 Second Main Result: Bond Option Pricing in ECIR
4 Future Work
Q.Peng (CGU) Extended CIR Model Sept. 25, 2014 9 / 21
First Main Result: Distribution of r(t)
”It is difficult to derive the probability distribution of the squared Besselprocesses with time-varying dimensions explicitly.” ——By Shirakawa(2002).
Q.Peng (CGU) Extended CIR Model Sept. 25, 2014 10 / 21
First Main Result: Distribution of r(t)
Only have characteristic function :(
Theorem 1: Assume θ(·), σ(·) are continuous positive-valued and d(·) iscontinuously differentiable and positive-valued. The characteristic functionof the interest rate r(t) in the ECIR model is:
E(e iωr(t)) = exp
(
iω( r0e
−∫ t
0 b(u)du
1− 2iωΣ(0, t)−∫ t
0
θ(s)e−∫ t
sb(u)du
1− 2iωΣ(s, t)ds))
,
where:
Σ(s, t) :=1
4
∫ t
se−
∫ t
vb(u)duσ2(v )dv .
Remark: r(t) ≥ 0, if r0 > 0 and d(s) > 0 for all s ≥ 0.
Q.Peng (CGU) Extended CIR Model Sept. 25, 2014 11 / 21
First Main Result: Distribution of r(t)
Our result extends Maghsoodi (1996)’s
An equivalent version of E(e iωr(t)):
E(e iωr(t)) =
exp
(
iωr0e−∫ t0 b(u)du
1−2iωΣ(0,t)− 1
2
∫ t
0d ′(s) log
(
1− 2iωΣ(s, t)ds))
(1− 2iωΣ(0, t))d(0)/2.
(1)
When d(t) ≡ d ∈ N/{0},
E(e iωr(t)) =
exp
(
iωr0e−∫ t0 b(u)du
1−2iωΣ(0,t)
)
(1− 2iωΣ(0, t))d(0)/2⇐⇒
(Σ(0, t)
)−1r(t) ∼ χ2
(
d(0),r0e
−∫ t
0b(u)du
Σ(0, t)
)
,
which generalizes Maghsoodi (1996)’s result.
When d(t) ≡ d ∈ R+, (1) generalizes the answer to Question 1.
Q.Peng (CGU) Extended CIR Model Sept. 25, 2014 12 / 21
First Main Result: Distribution of r(t)
Our result fits Fokker-Planck equation
Fokker-Planck equation for the density f (x , t) of r(t):
∂f (x , t)
∂t= b(t)f (x , t)+
(σ2(t)+ b(t)x− θ(t)
)∂f (x , t)
∂x+
xσ2(t)
2
∂2f (x , t)
∂x2.
-”Don’t know f (x , t)!”-”Try with Fourier transform of Fokker-Planck.”
The Fourier transform of Fokker-Planck equation with respect to x is
∂f (ω, t)
∂t= −θ(t)iωf (ω, t)−
(b(t)ω +
iω2σ2(t)
2
)∂f (ω, t)
∂ω. (2)
E(e−iωr(t)) =√2π f (ω, t) fits (2).
Q.Peng (CGU) Extended CIR Model Sept. 25, 2014 13 / 21
Second Main Result: Bond Option Pricing in ECIR
Plan
1 The extended CIR (ECIR) modelFrom Vasicek to ECIR modelMotivation
2 First Main Result: Distribution of r(t)
3 Second Main Result: Bond Option Pricing in ECIR
4 Future Work
Q.Peng (CGU) Extended CIR Model Sept. 25, 2014 14 / 21
Second Main Result: Bond Option Pricing in ECIR
”Although these researches (Maghsoodi, 2006; Roger, 1995) enable us tostudy the special case of extended CIR model, they do not permit thestudy of the general case.” —–By Shirakawa (2002).
Q.Peng (CGU) Extended CIR Model Sept. 25, 2014 15 / 21
Second Main Result: Bond Option Pricing in ECIR
Bond option pricing in ECIR
Let P(t,T ) be the price of a zero-coupon bond at time t, withmaturity T .A classical result (Shreve, 2004):
P(t,T ) = exp(
− r(t)C (t,T ) + A(t,T ))
,
where C (t,T ) satisfies the Riccati equation:{
∂C (t,T )∂t
− b(t)C (t,T )− 12σ2(t)C 2(t,T ) + 1 = 0;
C (T ,T ) = 0.
A(t,T ) = −∫ T
tθ(u)C (u,T )du.
Let K be the strike price of the option. The price C (0) at time 0 of aEuropean option is:
C (0) = E
[
exp(−∫ t
0r(u)du
)max
(P(t,T )−K , 0
)]
.
Q.Peng (CGU) Extended CIR Model Sept. 25, 2014 16 / 21
Second Main Result: Bond Option Pricing in ECIR
Change of measure and Laplace transform of density
Fix t as in P(t,T ). By a change of measure (t-forward measure Pt),
there exists a Brownian motion W t(s) such that
dr(s) =(
(−b(s) + σ2(s)C (s, t))︸ ︷︷ ︸
bt (s)
r(s)+ θ(s))
ds+ σ(s)√
r(s)dW t(s).
The Laplace transform of the density of the interest rate r(t) underthe t-forward measure P
t can be expressed as: for some suitable p,
Fr(t)(p) = exp
(
−p( r0e
−∫ t
0 bt (u)du
1+ 2pΣt(0, t)−∫ t
0
θ(s)e−∫ t
sbt (u)du
1+ 2pΣt(s, t)ds))
,
where
Σt(s, t) :=
1
4
∫ t
se−
∫ t
vbt (u)duσ2(v )dv .
Q.Peng (CGU) Extended CIR Model Sept. 25, 2014 17 / 21
Second Main Result: Bond Option Pricing in ECIR
Pricing C (0)
Theorem 2: The price C (0) is given by Laplace inversion:
C (0) =P(0, t)
2πilim
b→+∞
∫ a+ib
a−ibepr0 Ct(p)Fr(t)(p)dp,
where, Ct(p) is the Laplace transform of the payoff max(P(t,T )− K , 0):
Ct(p) =pK−p−C (t,T )eA(t,T ) − (p + C (t,T ))Kp+1
C (t,T )p − p2.
Proof: it is well-known (see, e.g., Lewis (2000)) that the price of anoption is given by the discounted inverse Laplace transform of the productof the Laplace transform of the payoff and the Laplace transform of thedistribution. �
Q.Peng (CGU) Extended CIR Model Sept. 25, 2014 18 / 21
Future Work
Plan
1 The extended CIR (ECIR) modelFrom Vasicek to ECIR modelMotivation
2 First Main Result: Distribution of r(t)
3 Second Main Result: Bond Option Pricing in ECIR
4 Future Work
Q.Peng (CGU) Extended CIR Model Sept. 25, 2014 19 / 21
Future Work
Next stop...
Conjecture: In ECIR model, r(t) > 0 for all t ≥ 0, providedd(0) > 1; r(t) visits 0 a finite number of times over any compacttime interval, if d(0) ∈ (0, 1].
Unsolved problem: The density of r(t)? Which should beautifullygeneralize the density of noncentral chi-square distribution, and willlargely simplify Theorem 2.
Q.Peng (CGU) Extended CIR Model Sept. 25, 2014 20 / 21
Future Work
References
Vasicek, Oldrich (1977). ”An Equilibrium Characterisation of theTerm Structure”. Journal of Financial Economics 5 (2): 177C188.
Cox, J.C., J.E. Ingersoll and S.A. Ross (1985). ”A Theory of theTerm Structure of Interest Rates”. Econometrica 53: 385C407.
Maghsoodi, Y. (1996). ”Solution of the Extended CIR Term Structureand Bond Option Valuation”. Mathematical Finance (6): 89C109.
Shirakawa, H. (2002). ”Squared Bessel Processes and TheirApplications to the Square Root Interest Rate Model”. Asia-PacificFinancial Markets 9 (3-4): 169-190.
Lewis, A. L. (2000). ”Option Valuation under Stochastic Volatility:with Mathematica Code.” Finance Press.
Q.Peng (CGU) Extended CIR Model Sept. 25, 2014 21 / 21