See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/316130616

The cox-ingersoll-ross interest rate model revisited: Some motivations and

applications

Chapter · January 2016

CITATION READS

1 4,996

3 authors, including:

Juan C. Cortés Rafael Jacinto Villanueva

Universitat Politècnica de València Universitat Politècnica de València
267 PUBLICATIONS 2,156 CITATIONS 197 PUBLICATIONS 1,484 CITATIONS

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

2nd International Conference on Mathematical Modelling in Applied Sciences, ICMMAS’19 View project

Special Issue "Advances on Uncertainty Quantification: Theory and Modelling". (Mathematics, MDPI). ISIT THE WEBSITE:
https://www.mdpi.com/journal/mathematics/special_issues/Uncertain_Quantif View project

All content following this page was uploaded by Rafael Jacinto Villanueva on 12 February 2019.

The user has requested enhancement of the downloaded file.

 Article Name 1

Chapter

THE COX-INGERSOLL-ROSS INTEREST

RATE MODEL REVISITED: SOME
MOTIVATIONS AND APPLICATIONS
R.-V. Arévalo1 , J.-C. Cortés1 , R.-J. Villanueva1∗
1
Instituto Universitario de Matemática Multidisciplinar,
Universitat Politècnica de València, Valencia, (Spain)

Abstract

We revisit the so-called Cox-Ingersoll-Ross (CIR) interest rate model. We pay par-
ticular attention to two different ways of motivating this stochastic model starting from
its deterministic model counterpart. Afterwards, we explain two powerful techniques
to estimate model parameters. Finally, the results are applied to model Euribor interest
rate from a real sample including some measures of goodness-of-fit.

Key Words: Euribor interest rate, Geometric Brownian Motion, Cox-Ingersoll-Ross model,
Itô-type stochastic differential equations, maximum likelihood method.

∗
E-mail address: [email protected]; [email protected]; [email protected]
2 Authors

1. Introduction
The study of interest rates is a very significant issue whose modelling is not trivial since its
behaviour depends on numerous key economic factors. Interest rates have a major influence
on the economy. Some important examples include changes in the investment companies,
fluctuations on the stock prices, level of savings in both companies and families, just to
mention a few.
One of the most important effects can be observed when the interest rate remains high
for a long time. This causes the reduction of companies’ profits resulting in a decrease
in the stock market as well. However, this also leads to the rise in prices for both loans
and credits and therefore, a reduction in the consumption by companies and families. To
the contrary, as the interest rate decreases for an extended period of time it encourages
domestic consumption and, boosts economy.
The aforementioned points necessitate reasons to motivate the search for mathematical
models which are able to predict interest rate dynamics. In this contribution, we revisit the
so-called Cox-Ingersoll-Ross (CIR) interest rate model [1] to show how it can be applied
to describe the dynamics of Euribor interest rate. CIR model is based on Itô stochastic
differential equations (s.d.e.’s) [2].
The main contribution of this chapter is the development of a tailor-made parameter
estimation method to the Cox-Ingersoll-Ross model. This method can be considered as an
adaptation of the technique shown in [3] for another short-term interest rate model, usually
referred to as Vasicek method [4].
This chapter is organized as follows: Section 2. is addressed to motivate the CIR model
from its deterministic counterpart. Section 3. is addressed to show how the main statistical
properties, such as mean and variance functions of the solution to the CIR model can be
obtained even though an explicit solution of the CIR model is not available. Both the mean
and variance functions can be obtained taking advantage of the so-called Itô Lemma. In
Section 4., two statistical methods are developed to provide model parameter estimation,
namely a maximum likelihood estimation method and a non-parametric technique. Model
validation is shown in Section 5.. Section 6. illustrates how the theoretical results intro-
duced in previous sections can be applied to model and predict Euribor interest rate. This
application is validated using some goodness-of-fit measures. Conclusions are drawn in
Section 7..

2. Motivating the CIR Stochastic Model

Throughout this section we introduce two different approaches in order to motivate the CIR
model. As we shall see later, the first approach is based on considering perturbations in a
classical deterministic interest rate model. The second approach leads to CIR interest rate
model as the limit of a discrete stochastic model with adequate transition probabilities.
Numerous models have been proposed to describe interest rate dynamics. The majority
of them are referred to as mean-reverting models, i.e. they tend to stabilize in the long-
term. The following deterministic interest rate model can be considered as the cornerstone
 Article Name 3

of mean-reverting models

dr(t) = α(re − r(t)) dt, α > 0,
(1)
r(0) = r0 ,

where r(t) denotes the interest rate at the time instant t > 0, r0 is the initial interest rate and
α can be interpreted as the mean-reversion velocity of adjustment of r(t) to the long-term
interest rate re . This latter assertion is based on the fact that

r(t) = re + (r0 − re )e−α(t−t0 ) −→ re . (2)

t→∞

Since the long-term interest rate re is not known in a deterministic way, in the seminal
contribution [4] the following stochastic model was proposed

dr(t) = α(re − r(t)) dt + σdB(t), α, σ > 0,
(3)
r(0) = r0 ,

where B(t) denotes the brownian motion also termed the standard Wiener stochastic pro-
cess. This formulation can be justified introducing the following perturbation re →
re + σB 0 (t) in the deterministic model (3). Here, σ parameter is referred to as the local
volatility and B 0 (t) denotes the white noise stochastic process [2].
Since volatility hardly ever remains constant another alternative to model (3) has been
proposed. In this regard, the Cox-Ingersoll-Ross model is likely the most popular
p in ap-
plications. In this model it is assumed that σ depends on t, namely σ(t) = σ r(t). This √
hypothesis is based on the fact that brownian motion B(t) has as standard deviation t
on the time interval [0, t], t > 0. Thus, it makes sense that the volatility behaves like the
brownian motion. This yields the so-called Cox-Ingersoll-Ross model [1]
 p
dr(t) = α(re − r(t)) dt + σ r(t)dB(t), α, σ > 0,
(4)
r(0) = r0 ,

Many generalizations of the CIR model have been proposed since the seminal contri-
bution by [1]. In [5] a comparative study of 8 short-term interest rate models can be found.
Here, we point out the CEV model where it is assumed that σ(t) = σr(t)λ , λ > 0. This
generalization offers more flexibility when fitting real interest rate samples. In [6], a com-
parative study to several interest rate models is shown. The study is done by comparing
implied parametric density with the same density estimated nonparametrically. A compre-
hensive study of short-term interest rate models can be found in [7, 8].
In [2] one can find a good motivation for the CIR interest rate model. As we shall show
below, it is based on interest rate’s changes for small time intervals ∆t. Let us suppose
that r(t) represents the instantaneous interest rate for the time t + ∆t, and let us denote
by ∆r = r(t + ∆t) − r(t). Then, there are three possibilities regarding these changes or
increments of interest rate: increase, remain constant or decrease. These changes consist
of (∆r)1 = 1, (∆r)2 = 0, and (∆r)3 = −1, respectively. The next step is to find out
their probabilities which thereafter will be denoted by p1 , p2 and p3 , respectively. For
4 Authors

convenience, let us assume that

  2 
σ r α(re −r)


 (∆r)1 = 1 ⇒ p1 = 2 − 2 ∆t,


(∆r)2 = 0 ⇒ p2 = 1 − p1 − p3 , (5)

  2 
α(re −r)

 (∆r)3 = −1 ⇒ p3 = σ r
+ ∆t.

2 2

p p1 and p3 are
The assignment of these probabilities has an intuitive interpretation, namely,
centred about the average value of square of the diffusion coefficient, σ r(t), being its
displacement the average value of the deterministic coefficient, α(re − r(t)). Naturally, the
value of p2 is determined taking into account that p1 + p2 + p3 = 1.
Therefore, the first and second statistical moments of increments ∆r are directly com-
puted from probabilities pi , 1 ≤ i ≤ 3,

E[∆r] = α(re − r)∆t, and E[(∆r)2 ] = σ 2 r∆t. (6)

This is consistent with the general results exhibited inp

[2, p.140] where it is proven that
E[(∆r)/(∆t)] corresponds to the coefficient of dt and E[(∆r)2 /∆t] corresponds to the
coefficient of dBt.

3. Solving the Stochastic Interest Rate Model

As it happens in dealing with deterministic differential equations, closed form solution
stochastic process of many s.d.e.’s are not available. However, sometimes it is possible to
calculate the mean and variance functions of the solution. Fortunately, this is the case for
the CIR model. As we shall show below, this can be done taking advantage of Itô lemma
and the crucial properties of Itô integral.
In order to the compute the expectation, E[r(t)], of the solution stochastic process to
model (4), let us write it in the following equivalent integral form
Z t Z t rp Z
r(t) = r0 + α re ds − α r(s) ds + σ r(s)dB(s)
0 0 r0
Z t Z rp (7)
= r0 + αre t − α r(s) ds + σ r(s)dB(s).
0 r0

We take the expectation operator in this latter expression and use the following properties
of Itô integrals
Z b  Z b Z b 
E g(t)dt = E [g(t)] dt, E g(t)dB(t) = 0, (8)
a a a

this leads Z t
E [r(t)] = r0 + αre t − α E [r(s)] ds. (9)
0
 Article Name 5

Differentiating this expression with respect to t, the following initial value problem is ob-
tained
 d (E [r(t)]) = αr − αE [r(t)] ,

e
dt (10)
r(0) = r0 .


The solution of this deterministic initial value problem is well-known

m1 (t) = E [r(t)] = re + (r0 − re ) e−αt −→ re . (11)

t→∞

Here, it is interesting to observe that the averaged value of the solution stochastic model
(4) matches the solution of the deterministic model (1) and, both share the same long-term
behaviour (see (2)).
Since
V[r(t)] = E (r(t))2 − (E [r(t)])2 ,
 
(12)
in order to compute the variance
 V[r(t)]  of r(t), it will first be necessary to compute the
second statistical moment E (r(t))2 . This will be done by applying the cornerstone of
stochastic calculus, namely, the Itô lemma.
For the sake of completeness, now we state the Itô lemma. Let us consider the general
s.d.e. (
dX(t) = f (t, X(t))dt + g(t, X(t))dB(t),
(13)
X(0) = X0 .
∂F (t,x)
Let F (t, x) be a deterministic mapping so that the following partial derivatives ∂t ,
∂F (t,x) ∂ 2 F (t,x)
∂x , and ∂x2
exist and are continuous. Then, F (t, X(t)) satisfies the following Itô
s.d.e

∂F (t, x) ∂F (t, x)
dF (t, X(t)) = + f (t, X)
∂t x=X(t) ∂x x=X(t)

∂ 2 F (t, x)

1
+ (g(t, X))2 dt (14)
2 ∂x2 x=X(t)

∂F (t, x)
+ g(t, X) dB(t).
∂x x=X(t)

Identifying general model (13) and (4) CIR model, one gets
p
X(t) = r(t), f (t, X(t)) = α(re − r(t)), g(t, X(t)) = σ r(t), X0 = r0 .

Now we apply (14) to F (t, x) = x2 with the identification x = r ≡ r(t), taking into
account that
∂F (t, r) ∂F (t, r) ∂ 2 F (t, r)
= 0, = 2r, = 2.
∂t ∂r ∂r2
This yields

d(r(t))2 = 2α(re − r(t))r(t) + σ 2 r(t) dt + 2σ(r(t))3/2 dB(t).

(15)
6 Authors

Integrating on the interval [0, t], one gets

Z t Z t Z t
2 2 2 2
(r(t)) = (r0 ) + 2αre r(s)ds − 2α (r(s)) ds + σ r(s)ds
0 0 0
Z t (16)
+ 2σ (r(s))3/2 dB(s).
0

Now, we take the expectation operator and then we use the properties given in (8) to obtain
Z t Z t
2
2 E (r(s))2 ds
   
E (r(t)) = (r0 ) + 2αre E [r(s)] ds − 2α
0 0
Z t (17)
+ σ2 E [r(s)] ds.
0

Let us denote
m2 (t) = E (r(t))2 ,
 
(18)
then (17) can be written as
Z t Z t Z t
2 2
m2 (t) = (r0 ) + 2αre m1 (s)ds − 2α m2 (s)ds + σ m1 (s)ds, (19)
0 0 0

where m1 (t) = E [r(t)]. Differentiating with respect to t this integral equation one sets the
following initial value problem

 d (m2 (t)) = −2αm (t) + 2αr + σ 2
m (t),


2 e 1
dt (20)

m2 (0) = (r0 ) ,2

where m1 (t) is given by (11). It can be checked that the solution to (20) is given by
1 −2αt 
m2 (t) = e 2(r0 )2 α − 4r0 re α + 2(re )2 α − 2r0 σ 2 + re σ 2
2α
+ 4r0 re α − 4α(re )2 + 2r0 σ 2 − 2re σ 2 eαt (21)

+ 2α(re )2 + re σ 2 e2αt .



Hence, substituting expressions of m1 (t) = E [r(t)] and m2 (t) = E (r(t))2 given by (11)
 

and (21), respectively, into (12), one gets the variance function of the solution stochastic
process to model (4)

σ 2 re σ 2 (r0 − re ) −αt
 2
σ re σ 2 r0 σ 2 re

V[r(t)] = + e + − e−2αt −→ . (22)
2α α 2α α t→∞ 2α

4. Parameter Estimation
After studying the model, one needs to estimate the model parameters, in this case, re α, and
σ from sample interest rate data. This section is addressed to face this question. As it will
be seen later, we will estimate the parameters for a sample of the 1-month Euribor interest
 Article Name 7

rate, although our approach is valid for other short-term interest rates. It is important to
point out that, from a theoretical point of view short-term condition is required since the
stochastic model (4) has been established assuming that the time increment ∆t > 0 is short
enough to legitimate its mathematical formulation.
Model parameter estimations will be done using two different methods. First, maximum
likelihood method and, secondly, a non-parametric moment method.

4.1. Maximum likelihood estimation

Let S = {r0 , r1 , . . . , rN } be a sample of interest rates corresponding to times t0 , t0 +
~ r0 , r1 , . . . , rN ) be the likelihood func-
∆t, . . . , t0 + N ∆t, ∆t > 0, respectively. Let l(θ;
~
tion associated to the sample S, where θ = (re , α, σ) denotes the parameter vector to be
estimated.
In order to obtain the transition probability density functions that will appear involved
in the likelihood function, we apply the Euler-Maruyama stochastic numerical scheme to
the s.d.e. (4) on the time discretization 0 = t0 < t1 = ∆t < t2 = 2∆t < · · · < ti−1 =
(i − 1)∆t < ti = i∆t < · · · < tN = N ∆t, [9, p.305]. This leads to
√
r(ti ) ≈ ri = ri−1 + α(re − ri−1 )∆t + σ ri−1 {Wi − Wi−1 }, 1 ≤ i ≤ N, (23)

being

Wi − Wi−1 = W (ti ) − W (ti−1 ) = W (i∆t) − W ((i − 1)∆t) ∼ N(0; ∆t),

where we recall that

√ the increments of brownian motion are gaussian with zero mean and
standard deviation ∆t. Using the following
√ representation for simulating the increments
of brownian motion, Wi − Wi−1 = ∆tηi , ηi ∼ N(0; 1), the scheme (23) can be written
as
√ √
ri ≈ ri−1 + α(re − ri−1 )∆t + σ ri−1 ∆tηi , ηi ∼ N(0; 1), 1 ≤ i ≤ N. (24)

From this approximation one deduces the probabilistic distribution of ri given ri−1 , since
ri is a linear transformation of the gaussian random variable ηi , and we have

ri |ri−1 ∼ N ri−1 + α(re − ri−1 )∆t; σ 2 ri−1 ∆t , 1 ≤ i ≤ N.

(25)

As a consequence, the transition probability density function of ri |ri−1 is given by

[ri −(ri−1 +α(re −ri−1 )∆t)]2

1 σ 2 ri−1 ∆t
p(ri |ri−1 ; α, re , σ) ≈ p e , 1 ≤ i ≤ N. (26)
2πσ 2 ri−1 ∆t

Taking into account that the solution stochastic process to Itô-type differential equations
(13) are Markov processes of order 1 [10, Th. 5.2.5, p.131] and the total probability theo-
rem, it is easy to check that the likelihood function can be expressed as
N
Y
~ = p(r0 )
l(θ) ~
p(ri |ri−1 ; θ), θ~ = (re , α, σ). (27)
i=1
8 Authors
~ to get the best parameter estimation. However,
Therefore, our goal is to maximize l(θ)
from a numerical standpoint it is more convenient, and equivalent, to handle the following
objective function:
 
~
min L(θ), where L(θ)~ = − ln l(θ) ~ . (28)
~ 3
θ∈R

After replacing (26) into (27)–(28) and performing simplifications, one obtains the follow-
ing expression for the log-likelihood function

N
L(re , α, σ; r0 , . . . , rN ) = (ln(2π∆t) + 2 ln(σ))
2
N
1X
+ ln(ri−1 ) (29)
2
i=1
N
1 X [ri − (ri−1 + α(re − ri−1 )∆t)]2
+ .
2∆tσ 2 ri−1
i=1

To estimate the model parameters (re , α, σ) we have considered a sample of 134 values
corresponding to the 1-month Euribor interest rate from January 2nd, 2012 until July 11th,
2013 taken from [11]. These values have been plotted in Figure 1 (red points). Model pa-
rameter estimation has been computed by Mathematica software [12] using the Differential
Evolution algorithm with time step ∆t = 1 in (29). The results are collected in the second
column of Table 1.

4.2. Non-parametric estimation

In a similar way to Subsection 4.1., let S = {r0 , r1 , . . . , rN } be a sample of interest rates

corresponding to times t0 , t0 + ∆t, . . . , t0 + N ∆t, ∆t > 0, respectively. The problem
lies in how to estimate the parametric vector θ~ = (re , α, σ). This nonparametric estima-
tion method has been studied in [13]. In a first step, this consists of applying the Eu-
ler Maruyama scheme to the general stochastic differential equation (13) similarly as was
shown in the previous subsection. Therefore, one obtain the equation (23).
After these steps, as it is mentioned in [2] we will perform the main part, which is based
on the approximate expectations. One can compute this approximation by using the values
of ri , ∀ 0 ≤ i ≤ N − 1, for a discrete-time process. Obtaining the equations

E [(ri+1 − ri ) /∆t − α(re − ri )] = O(∆t),

h i
E (ri+1 − ri )2 /∆t − σ 2 ri = O(∆t), (30)

E [(ri+1 − ri ) ri /∆t − α(re − ri )ri ] = O(∆t).

Consequently, one could estimate the parametric vector θ~ using the sample counterparts
of the equation (30),
 Article Name 9

N −1 N −1
X 1 X
α (ri+1 − ri ) = (ri+1 − ri ) ,
∆t
i=0 i=0
N −1 N −1
X 1 X
σ 2 ri = (ri+1 − ri )2 , (31)
∆t
i=0 i=0
N −1 N −1
X 1 X
α (ri+1 − ri ) ri = (ri+1 − ri ) ri .
∆t
i=0 i=0

This results are collected in the third column of Table 1.

4.3. Results
In this subsection we summarize the model parameter estimations obtained using the two
methods described in Subsections 4.1. and 4.2.. To facilitate the comparison between both
methods, the numerical results are collected in Table 1. We observe that both estimations
are close. This provides reliability and robustness to our estimations.

Parameter Maximum Likelihood Method Nonparametric Technique

α 0.03172 0.03222
re 0.12043 0.12037
σ 0.00281 0.00286

Table 1. Model parameter estimations to the CIR model using a maximun likelihood method
and a non-parameric technique considering a 1-month Euribor interest rate sample from
January 2nd, 2012 until July 11th, 2013.

5. Validation of the Stochastic Model

The next natural step is to verify if the parameters are good enough to be able to use them.
To that end, we are going to use two different methods of measurement of goodness-of-fit,
MAPE (Mean Absolute Percentage Error) and RMSE (Root Mean Square Error). Both of
them will be calculated for the theoretical estimation of the values, r(t).
For the parameters estimation, we will take the expected value, equation (11). Then, the
values for the RMSE and MAPE measures of goodness-of-fit are show in Table 2. We can
see that with both methods, RMSE, and MAPE validate the model (4), and therefore, they
tell us that we can take this estimation to model and predict the 1-month Euribor rates.

6. Predictions based on the Parameters Estimation

Due to the fact that, the method have been validated, we are able to make predictions for
1-month Euribor interest rate. In Table 3 the exact value of the 1-mount Euribor, our pre-
dictions and confidence interval for the maximum likelihood estimation are shown. In the
10 Authors

Figure 1. Model values (mean), condifence interval (mean plus/minus two standard devia-
tions) and real data for 1-month Euribor interest rate from January 2nd, 2012 to July 11th,
2013.

Estimation Methods RMSE MAPE

Maximum Likelihood Technique 0.00422968 2.68099
Nonparametric Technique 0.00421973 2.67816

Table 2. Validation of the estimation parameters for the CIR model (4) by using RMSE and
MAPE .
 Article Name 11

Date 1-month Euribor Punctual estimation 95% Confidence interval

12 Jul 2013 0.123 0.120268 [0.112539, 0.127997]
15 Jul 2013 0.123 0.120273 [0.112544, 0.128002]
16 Jul 2013 0.123 0.120278 [0.112548, 0.128008]
17 Jul 2013 0.122 0.120283 [0.112553, 0.128013]
18 Jul 2013 0.122 0.120287 [0.112557, 0.128018]

Table 3. Predictions for the 1-month Euribor using the likelihood estimations.

Date 1-month Euribor Punctual estimation 95% Confidence interval

12 Jul 2013 0.123 0.120227 [0.112403, 0.128052]
15 Jul 2013 0.123 0.120232 [0.112407, 0.128058]
16 Jul 2013 0.123 0.120237 [0.112411, 0.128063]
17 Jul 2013 0.122 0.120241 [0.112416, 0.128067]
18 Jul 2013 0.122 0.120246 [0.112420, 0.128072]

Table 4. Predictions for the 1-month Euribor using the nonparametric estimations.

same way, but for the nonparametric estimation methods are shown in the Table 4. These
confidence intervals and estimations for the 134 days have been plotted in the Figure 1, as
well as, the next 5 days predictions. These latter values have been plotted after the vertical
(black) line.

7. Conclusions
In this paper we have shown several motivations of the so-called stochastic Cox-Ingersoll-
Ross’s model taking as starting point its deterministic counterpart. This model is based
on an Itô type stochastic differential equation. In opposition to Vasicek’s model, in CIR
model, variable local volatility is considered. Nevertheless, the mean keeps having mean
reversion asymptotic behaviour. Afterwards, we have provided several model parameter
estimations and verified that these estimations are reliable. This has enabled us to construct
good predictions since all of the real data are within of the confidence interval. Additionally,
we have obtained that, the results with the CIR model are closer than the results obtained
with Vasicek’s model in [3].

References
[1] J.C. Cox, J.E. Ingersoll, S.A. Ross, A theory of the term structure of interest rate,
Econometrica 53(2), (1985), 385–407.

[2] E. Allen, Modelling with Itô Stochastic Differential Equations, Springer, Series Math-
ematical Modelling: Theory and applications, 2007.
 12 Authors

[3] J.C. Cortés, J.V. Romero, A. Sánchez-Sánchez, R.J. Villanueva, Modelling 1-month
Euribor interest rate by using differential equations with uncertainty, Applied Mathe-
matical and Computational Science 7(3), (2015), 37–50.

[4] O. Vasicek, An equilibrium characterization of the term structure, Journal of Financial

Economics 5(2), (1977), 177–188.

[5] K.C. Chan, G.A. Karolyi, F.A. Longstaff, A.B. Sanders, An empirical comparison
of alternative models of the short-term interest rate, Journal of Finance XLVII(3),
(1992), 1209–1227.

[6] Y. Aı̈t-Sahalia, Testing continuous-time models of the spot rate, The Review of Finan-
cial Studies 9, (1996), 385–426.

[7] M. Baxter, A. Rennie, Financial Calculus: An Introduction to Financial Pricing, Cam-

bridge University Press, 1997.

[8] D. Brigo, F. Mercurio, Interest Rate Models-Theory and Practice with Smile, Inflation
and Credit, Springer Finance, 2006.

[9] P.E. Kloeden, E. Platen, Numerical Solution of Stochastic Differential Equations,

Springer, Series Applications of Mathematics: Stochastic Modelling and Applied
Probability 23, 1992.

[10] T.T. Soong, Random Differential Equations in Science and Engineering, Academic
Press, Series Mathematics in Science and Engineering 103, 1973.

[11] www.emmi-benchmarks.eu/euribor-org/euribor-rates.html

[12] https://www.wolfram.com/mathematica/

[13] Stanton R., A nonparametric model of term structure dynamics and the market price
of interest rate risk, Journal of Finance LII, (1997), 1973–2002.

[14] Y. Maghsoodi, Solution of the extended CIR term structure and bond option valuation,
Mathematical Finance 6(1), (1996), 89–109.

[15] T.A. Marsh, E.R. Rosenfeld, Stochastic processes for interest rates and equilibrium
bond prices, Journal of Finance, 38(2), (1982), 635–646.

View publication stats