(MASTER THESIS)

DISTRIBUTIONS OF THE PARAMETERS IN

VASICEK MODEL

Gönül AYRANCI

Supervisor: Yrd. Doç. Dr. Banu ÖZGÜREL

Department of Actuarial Science

Bornova-İZMİR

2013
 i

YASAR UNIVERSITY
GRADUATE SCHOOL OF NATURAL AND APPLIED SCIENCE

DISTRIBUTIONS OF THE PARAMETERS IN

VASICEK MODEL

Gönül AYRANCI

Supervisor: Yrd. Doç. Dr. Banu ÖZGÜREL

Department of Actuarial Science

Bornova-İZMİR

2013
 ii

This study titled “ Distributions of the Parameters in Vasicek Model” and

presented as Master Thesis by Gönül Ayrancı has been evaluated in compliance
with the relevant provisions of Y.U Graduate Education and Training Regulation
and Y.U Institute of Science Education and Training Direction and jury members
written below have decided for the defence of this thesis and it has been declared
by consensus / majority of votes that the candidate has succeeded in thesis
defence examination dated May 13, 2013.

Jury Members: Signature:

Yrd. Doç. Dr. Banu ÖZGÜREL ………………...

Yrd. Doç. Dr. Serkan ALBAYRAK …………………

Doç. Dr. Özlem Ege ORUÇ …………………

 iii

ÖZET

VASICEK MODELİNDE PARAMETRELERİN

DAĞILIMLARI

AYRANCI, Gönül

Yüksek Lisans Tezi, Aktüerya Bilimleri Bölümü

Tez Danışmanı: Assist. Prof. Dr. Banu ÖZGÜREL

Mayıs 2013

Bu çalışmada LIBOR TR faiz oranı verilerinin 2 Ocak 2008 – 5 Aralık 2012

tarihleri arasındaki günlük zaman serilerinin stokastik davranışlarını incelemek için
Vasicek faiz oranı modeli kullanılmıştır. Modelde yer alan parametrelerin
tahimininde en küçük kareler tahmin yönteminden faydalanılmıştır. Monte Carlo
simulasyonu kullanılarak parametrelerin dağılımları incelenmiştir. Bu metod
sayesinde sadece parametre tahmini yapılmamış, parametrelere ilişkin güven
aralıkları elde edilmiştir.

Anahtar Kelimeler: Vasicek modeli, En Küçük Kareler Yöntemi, Monte Carlo

Simülasyonu
 iv

ABSTRACT

DISTRIBUTIONS OF THE PARAMETERS IN VASICEK

MODEL

AYRANCI, Gönül

MSc in Actuarial Science

Supervisor: Assist. Prof. Dr. Banu ÖZGÜREL

Mayıs 2013

In this study time series of TRLIBOR interest rates in different maturities are
modeled with Vaiscek Model. OLS method is used for calibrating the model
parameters of Vasicek to TRLIBOR rates data which is between 2.01.2008 and
5.12.2012. Then distributions of parameters are obtained by using Monte Carlo
Simulation ragarding the Vaicek Model. Thus, not only parameters are estimated but
also confidience intervals are given.

Keymords: Vasicek Model, Ordinary Least Square Method, Monte Carlo

Simulation
 v

ACKNOWLEDGEMENTS

I would not have been able to complete this journey without the aid and support of
countless people.

First I must express my gratitude to my supervisor Assist. Prof. Dr. Banu Özgürel
for the guidance, useful comments, remarks and engagement through the learning
process of this master thesis.

I want to express the separate gratitude to Assist. Prof. Dr. Serkan Albayrak for
his advices, irreplaceable contributions, support and especially for his helps in
application of this study.

I am also grateful to Prof. Dr. Behzat Gürkan for encouragement and support in
many ways to the success of this study.

Last but not least, I would like to thank my parents and friends, especially my
Mother and Father for their unconditional support throughout my study.
 vi

TEXT OF OATH

I declare and honesly confirm that my study titled “Distributions of Parameters in

Vasicek Model”, and presented as Master’s Doctorate Thesis has been written
without applying to any assistance inconsistent with scientific ethics and traditions
and all sources I have benefited from are listed in bibliography and I have
benefited from these sources by means of making references.

13/ 05/ 2013 Gönül AYRANCI

 vii

TABLE OF CONTENTS

1 INTRODUCTION ...........................................................................................1
2 METHODS ......................................................................................................5
2.1 Model Description ....................................................................................5
2.1.1 Vasicek Interest Rate Model..............................................................6
2.2 Parameter Estimation ................................................................................9
2.2.1 Ordinary Least Square Estimation for Vasicek Model......................9
2.2.2 Maksimum Likelihood Estimation for The Vasicek Model ............10
2.3 Monte Carlo Simulation..........................................................................11
3 APPLICATION .............................................................................................16
4 CONCLUSION .............................................................................................24
5 BIBLIOGRAPHY .........................................................................................25
 viii

INDEX OF FIGURE

Figure 1.1 Risk Modules of SCR.............................................................................2

Figure 2.1 Mean reversion process..........................................................................7
Figure 2.2 Quarter circle of radius r=1 ..................................................................13
Figure 2.3 value estimation with Monte Carlo Method....................................15
Figure 3.1 Time series of TRLIBOR 1M from 2.01.2008 to 5.12.2012 ...............17
Figure 3.2 Time series of TRLIBOR 6M from 2.01.2008 to 5.12.2012 ...............17
Figure 3.3 Time series of TRLIBOR 1Y from 2.01.2008 to 5.12.2012 ................18
Figure 3.4 Time series of TRLIBOR overnight from 2.01.2008 to 5.12.2012......18
Figure 3.5 Plots of the parameters in simulation step...........................................21
Figure 3.6 Histograms of the parameters in simulation step...............21
Figure 3.7 Plots of the parameters in simulation step..........................................22
Figure 3.8 Histograms of the parameters in simulation step ................................23
 ix

INDEX OF TABLE

Table 3.1 Summary of Vasicek Estimators ...........................................................20

Table 3.2 %95 Confidence Intervals for the parameters .......................................23
Table 3.3 %99 Confidence Intervals for the parameters .......................................23
 x

INDEX OF SYMBOLS AND ABBREVIATIONS

Abbreviations Explanation

CIR Cox-Ingersoll-Ross

MLE Maximum Likelihood Estimation

OLS Ordinary Least Squares

SCR Solvency Capital Requirement

SDE Stochastic Differential Equation

 1

1 INTRODUCTION

Interest rates are an indicator for the global cost of the money. There are
financial instruments which directly link with interest rates in portfolio of
financial institutions and especially of insurance companies. Interest rates and the
volatility of interest rates play a crucial role in estimation of losses which is
caused for holding the money. Moreover there are some government regulations
about estimation of those losses for insurance companies (Solvency II). Insurance
companies have to calculate the capital requirement for the market risk under the
Solvency II requirements. There is a committee that is established by Treasury on
2009 for inform the sector about Solvency II directives and assess the impact of
these directives on the insurance sector. Ten insurance companies, of which 5
operate in life, 4 non-life and 1 in reinsurance, participated in these studies. The
Solvency Capital Requirement (SCR) is the new solvency standard for firms in
Solvency II directives. SCR is based on the idea that an insurance firm should
have the amount of capital that is sufficient with a 99.5% confidence-level to
guarantee that the firm will have enough assets to cover its liabilities at the end of
one year period. In other words, SCR calibrated to the Value-at-Risk of basic own
funds subject to a confidence level 99.5% over a 1 year time horizon. The SCR is
intended to reflect all quantifiable risks that the firm might face. A basic SCR
calculation divided into modules per risks, with adjustment for the loss-absorbing
capacity of technical provisions and deferred taxes. These calculations must be
done at least once per year using a standard formula or an internal model. But
internal models improve risk sensivity of SCR and provide better risk
management, which also improves policyholder protection (CP20 final advice1).
Martin (2012) applies the model of Gatzert and Martin (2012) for deriving a
partial internal model for the market risk module regarding the underlying interest
rate process and parameters. Solvency II Directive gives an option to model only
certain risk modules or business units and to use the standard formula for the
remaining parts.

1
CP20 is a consaltation process initiated by CEIOPS (Committee of European Insurance and
Occupational Pensions Supervisors) on the “advice to the European Commision in the Framework
of the Solveny II”
 2

There is a figure below that shows the risk modules of SCR (Figure 1.1).

Figure 1.1 Risk Modules of SCR

As we can see in the Figure 1.1, interest rate risk plays a significant role
for calculating SCR (Martin, 2012). The aim of this study is to evaluate model
with focus on market risk module of Solvency II regarding the interest rate risk.
Already it is an inevitable fact that insurance companies must have fitting interest
rate model while they estimate the value at risk (VaR) on a specific portfolio of
financial assets. Also Martin (2012) shows that the change in SCR is sustainable
when changing the underlying interest rate model.

Modeling of interest rates is especially used for investment and financial

decisions, portfolio management and insurance. Future interest rates can be
predicted with a suitable interest rate model and those interest rates predictions
can be used for the estimation of price of the financial instruments. An increasing
number of insurance companies and other financial institutions are using
stochastically generated interest rates to measure interest rate risk and to value
financial assets. The use of stochastic interest rates for insurance system is both
appropriate— given the stochastic nature of interest rates— and increasingly
mandated by regulators at the larger more complex financial institutions. An
interest rate model is a probabilistic description of the future evolution of interest
rates. It characterizes the uncertainty involved in interest rates. Interest rate
modeling is a very important part of any insurance. This is because the valuation
of the assets and liabilities of insurance companies depends crucially on interest
 3

rates. Also enormous literature in finance such as Vasicek (1977) and Cox,
Ingersoll and Ross (CIR) (1985) have documented that interest rate should be
followed by a stochastic process.

In the literature various models have been proposed such as Merton

(1973), Vasicek (1977), Dothan (1978), Richard (1979), Brennan and Schwarz
(1979, 1982), Cox, Ingersoll and Ross (1985), Rendleman and Bartter (1980),
Longstaff (1989), Hull and White (1990), Pearson and Sun (1994). These models
can be used to value all interest rate contingent claims in different cases. Major
difficulty in the modeling process is to provide parameters that fit to the initial
term structure of interest rates. In that case several papers focused on the
estimation of the parameters in a consistent way. Chan et al. (1992) estimate and
compare continuous time models of the short term riskless rate using the
Generalized Method of Moments. Kladivko (2007) investigates Maximum
Likelihood Estimation of the CIR process for interest rate time series. While
comparing the Vasicek and CIR models, Munnik and Schotman (1992) used
Ordinary Least Square Estimation method. Berg (2011) describe OLS and MLE
methods calibrating the model parameters of Vasicek. Also Zeytun and Gupta
(2007) analyze Vasicek and CIR models to give results for the parameters.
Carriere (2000) investigate long-term yield rates for valuing a life insurance and
used OLS method parameter estimation of the models such as Vasicek (1977),
CIR (1985), Nelson-Siegel (1987). Önalan (2009) explains the structure of
Vasicek and CIR models, then give parameter estimation methods OLS for
Vaiscek, Matingale method for CIR. Moreover Şahin and Genç (2009) investigate
which short term interest rate model appropriate for Turkish data.

In this paper we examine the application of the Vasicek Model with

TRLIBOR data to study the term structure of interest rates. Simply the aim of this
study is to determine parameters of the Vasicek model and give confidence
intervals regarding to estimation of parameters. After model parameters are
estimated by using OLS method, the distribution of these estimated parameters is
obtained by using the Monte Carlo Simulation.
 4

The rest of this thesis is organized as follows. Chapter 2 describes our

model and methods that we used. This chapter is explained in three main sections.
First general diffusion model description for interest rates is given and Vasicek
Model examined in detail. Then OLS and MLE methods are applied for the
parameters of the Vasicek Model. Last section in chapter 2 Monte Carlo method is
explained with examples. Chapter 3 shows an application of parameter estimation
of Vasicek Model for TRLIBOR rates. Chapter 4 summarizes the paper and gives
results that we obtain.
 5

2 METHODS

2.1 Model Description

Interest rate processes have been defined with stochastic differential

equations. Generally models divided into equilibrium models and no-arbitrage
models. We only focus on equilibrium models and from those only the Vasicek
model describe here. The equilibrium models based on some assumptions about
economic variables and give us process for the short rate, . (Hull and White
1993). To achieve a balance between supply of bonds rate and demand of the
investors is the aim of the equilibrium models. The most well-known equilibrium
models are Merton(1973), Vasicek(1977) and CIR(1985).

There are also single or multi-factor models for the interest rate modeling
and the Vasicek model is classified as one-factor model and the interest rate is
called instantaneous interest rate. In one-factor models it is accepted that there is
only one source of the risk. These models may be preliminary among the recent
multi-factor models but they provide a good introduction to the study of interest
rates. Moreover they can be easily solved and Sorwar (2007) states that the
existence of analytical solutions leads to quick valuation of the bond, option,
prices, and the hedge parameters such as delta for risk management purposes.

The general formula for one-factor diffusion model is:

(1)

where is a Brownian motion process. Different forms of functions and

give various diffusion models.
 6

2.1.1 Vasicek Interest Rate Model

Vasicek model is a one-factor model describing an evolution of interest

rates. It describes movements of interest rates as driven by source of market risk.
Also it can be used for the derivatives of interest rates. It was introduce in 1977 by
Oldrich Vasicek. The spot rate follows so-called Ornstein-Uhlenbeck process
and the equation has following form

(2)

(3)

where Wt is a Wiener process (also called Brownian Motion) under the risk
neutral measure modelling the random market risk factor. With ,
corresponding to the choice , in equation (1). This
description of the spot rate process has been proposed by Merton (1971).

According to the Vasicek O. (1977), Journal of Financial Economics, 5,

177-188, “The Ornstein-Uhlenbeck process with is sometimes called the elastic
random walk. It is a Markov process with normally distributed increments. In contrast to
the random walk (the Wiener process), which is an unstable process and after long time
will diverge to infinite values, the Ornstein-Uhlenbeck process possesses a stationary
distribution. The instantaneous drift represents a force that keeps pulling the
process towards its long-term mean with magnitude proportional to the deviation of the
process from the mean. The stochastic element, which has a constant instantaneous
variance, causes the process fluctuate around the level in an erratic, but continuous,
fashion”

The most important feature which this model exhibits and explained by
Vasicek is the mean reversion, which means that if the interest rate is bigger than
the long run mean ( then the coefficient makes the drift become
negative so that the rate will be pulled down in the direction of . Similarly, if
the interest rate is smaller than the long run mean ( , then the coefficient
makes the drift term become positive so that the rate will be pulled up in
 7

the direction of . Therefore, the coefficient is the speed of adjustment of the

interest rate towards its long run level.

Figure 2.1 Mean reversion process

One of the most attractive feature of the Vasicek Model is that there is
close form solution for . We start solution of the Stochastic Differential
Equation (2) by taking the derivative of , which yields:

(4)

Substitute equation (2) into equation (4) yields;

(5)

Rearrange the equation (5) gives:

(6)

Taking integral from to for both sides of the equation (6):

(7)
 8

(8)

(9)

If we multiply both sides of the equation with the solution of the differential
equation (2) is:

(10)

process given in equation (10) follows a Gaussian Distribution with mean:2

(11)

and variance:

(12)

If time increases the mean tends to the long-term value and the variance
remains bounded, implying mean reversion. The long-term distribution of the
Ornstein-Uhlenbeck process is stationary and is Gaussian with mean and

variance . (Sypkens 2010)

2
Önalan Ö., Vasicek ve CIR Modelleri Kullanılarak Oynaklık ve Faiz Oranlarının Modellenmesi,
Marmara Üniversitesi, İİBF dergisi, II, S. 329-344, 2009.
 9

2.2 Parameter Estimation

2.2.1 Ordinary Least Square Estimation for Vasicek Model

The explicit solution of the Ornstein-Uhlenbeck SDE is obtained in

previous section, namely:

(13)

is a random variable with mean and variance and

respectively. It is assumed that follows a process as

follow:

(14)

The discrete version of the equation (13) on time grid

(equal time intervals) and

(15)

where is a Gauission white noise ( .

Parameters of eqaution (14) is:

(16)

(17)

(18)

Rearrange equations (16), (17) and (18) in terms of the parameters α, β, and
which yield:
 10

(19)
, ,

The formulas for the least squares estimates a, b, and are:

(20)

(21)

(22)

2.2.2 Maksimum Likelihood Estimation for The Vasicek Model

MLE method is consistent for the Vasicek model since its error term is
normally distributed (Ren Raw Chen, 1996). So, the likelihood function can be
derived for the MLE. We need to define the likelihood function and find the
paremeters that maximize the likelihood function. There is different methods in
literature for maximize the methods likelihood function (Sypens, 2010).
Optimization method is explained for this problem. The conditional probability
function is derived by combining the solution of our model with the normal
distribution function (Calibrating the Vasicek Thijs Von den Berg, 2011).

(23)

If is given, the conditional density function of on the time grid

, with equal time step by using equation (11) and
(12).
 11

(24)

Since the logarithmic function is monotonically increasing, maximizing the log-

likelihood function also maximizes the likelihood function. The log-likelihood
function is;

(25)

We can find the maximum likelihood estimates of parameter vecor by

maximizing maximizing the log likelihood function (Equation 25):

(26)

2.3 Monte Carlo Simulation

Monte Carlo means using random numbers in scientific computing. More

precisely, it means using random numbers as a tool to compute something that is
not random. One of the first uses of this method is suggested by Buffon (1777)
with an experiment that calculates the probability that the needle intersect one of
the lines on a horizontal plane, now known as the famous Buffon’s Needle
Problem. Later Laplace (1886) suggested the idea of evaluating with Monte
 12

Carlo Simulation. Also it is used in kinetic theory of gases by Kelvin (1901). But
the method today labeled as “Monte Carlo Method” is formally introduced to the
literature by Ulam (1940) where he advocated the use of computers in an
integration problem related to Manhattan Project which has no formal solution.

The process of method can be explained as follows: During a Monte Carlo

Simulation, values are chosen randomly from the input probability distributions.
These generated samples are called iterations and results from the samples are
recorded for each iteration. Through Monte Carlo simulation, this process can be
done hundreds and thousands times. In this way, Monte Carlo simulation provides
much more extensive information about what may happen and how likely it is to
happen.

We can explain Monte Carlo method further with a nice and well-known
example of value estimation. is a constant that is the ratio of a circle’
circumference to its diameter. It is simply accepted as 3.14 but true value cannot
represent as a finite decimal fraction. Let us consider a unit circle with a radius 1.
If is a point on the unit circle, by the Pythagorean Theorem, that satisfy the
following equation;

(27)

By choosing random points on the plane, we need to decide each point that we
have chosen whether is within the circle or not. The quarter unit circle is
considered as shown in the following figure.
 13

Figure 2.2 Quarter circle of radius r=1

Since we know the area of a circle with radius .

(28)

The unit circle area is equal to and the area of the quarter circle that we

interested is equal to . We have to look at the unit circle (radius = 1) within a

square with sides equal to 1 (see Figure 2.2). Now if we pick a random points
where both and are between from 0 to 1. The probability of that this
random point lies inside the unit circle is given as the proportion between the area
of the unit circle and the square :

(29)

If we pick a random point times and of those times the point lies
inside the unit circle, the probability of that a random point lies inside the unit
circle is given as :

(30)
 14

where the dot indicates that this is a discreet distribution (because M and N are
integers). But if N becomes very large (theoretically infinite), the two
probabilities will become equal and we can write :

(31)

Monte Carlo method applied and simulation run 10000 times to

approximating the value of . Obtaining results and the codes relating to the
process are given (Figure 2.3).

n=10000

x=runif(n)

y=runif(n)

counter=0

plot(NA,type="n",xlim=c(0,1),ylim=c(0,1),
xlab="n=10000",ylab="")

for (i in 1:n){

if ((x[i]^2+y[i]^2)<=1) {

counter=counter+1

points(x[i],y[i],type="p")

4*counter/n
 15

Figure 2.3 value estimation with Monte Carlo Method

( )
Monte Carlo simulation can be used to approximate probability distributions
of functions of random variables. For example, let be a random variable and its
expected value be If we generate independent random
variables from the same distribution then we can approximate by

(32)

The strong law of large numbers states that . Albeit

the generated random numbers would be different each time and therefore so
does the sample average, , the target number, , is fixed and nonrandom.
Using thousands of Monte Carlo runs one can approximate the distribution of ,
around

Monte Carlo Methods allows us to account for risk in quantitive analysis

and decision making. The technique is used in various fields such as finance
(Haugh, 2004), project management (Kwak and Ingall, 2007), manufacturing
(Pica et al., (2006), engineering (Amar, 2006), insurance (Collins, 1962), the
environmental and resource economics (Scarpa and Alberini, 2005).
 16

3 APPLICATION

In this thesis, we are going to use TRLIBOR data with different time to maturities.
The Turkish Lira Interbank Offered Rate, or TRLIBOR, is the average interest
rate at which term deposits are offered between prime banks in the Turkish
wholesale money market or interbank market.
 17

TRLIBOR initiated on August 1, 2002 by the Banks Association of Turkey and

has resulted in the formation of benchmark interest rates that the economic agents
base their projections. The formation of benchmark interest rates is an important
development as they will increase transparency of the markets, form a sound yield
curve in money markets and improve forward transactions by pricing financial
assets more effectively.

Our study sample from the web page of the Banks Association of Turkey
TRLIBOR3 data period is from 2.01.2008 to 5.12.2012 and contains 1230 daily
observations for each time maturity. The rates are shown in the following figures
in different maturities, 1 month (Figure 3.1), 6 month (Figure 3.2), 1 year (Figure
3.3) and overnight (Figure 3.4).

Figure 3.1 Time series of TRLIBOR 1M from 2.01.2008 to 5.12.2012

Figure 3.2 Time series of TRLIBOR 6M from 2.01.2008 to 5.12.2012

3
http://www.tbb.org.tr
 18

Figure 3.3 Time series of TRLIBOR 1Y from 2.01.2008 to 5.12.2012

It can clearly be seen from the figures (Figure 3.1, Figure 3.2 and Figure 3.3) in 1
month, six month and 1 year maturities same pattern is observed. TRLIBOR rates
were extremely high at the start of monitored period, especially in last months of
2008. After that there is a dramatic fall occurred in earlier months of 2009 and
then it seems stable from 2010 to 2011. Finally, there is an increasing trend
between the year 2011 and 2012.

Figure 3.4 Time series of TRLIBOR overnight from 2.01.2008 to 5.12.2012

In the case of overnight maturity, we observed a pattern that more sensitive than
the others especially on the last interval from 2011 to 2012. So to estimate the
parameters of Vasicek Model we use the daily TRLIBOR data with overnight
maturity.
 19

The time series plots of our sample data is shown in the figures above. Since we
can observe from the figures, modeling interest rates is a very complex and hard
task. So Vasicek Model parameters will be estimated using TRLIBOR data.

(33)

(34)

where speed of mean reversion long run mean

instantaneous volatility Weiner process

In order to estimate the development of the instantaneous interest rate the Euler-
discretisation of the Vasicek Model (Equation 33) is used and is given:

(35)

where is generated random number from (Josef, 2009).

In the simulation process, the most important step is to estimate the parameters of
our model accurately. Parameter estimation methods regarding the
Vasicek Model are described in Chapter 2. Estimation of parameters with OLS
method gives very similiar results to MLE (Sypkens, 2010). To estimate the
parameters of our model we use OLS method and run the R code below which
estimates required parameters (Table 3.1)

Vasicek.OLS = function(data,dt){

N = length(data)

rate = data[2:N]

lagrate= data[1:(N-1)]

a=(N*sum(rate*lagrate)-sum(rate)*sum(lagrate))/

(N*sum(lagrate^2)- sum(lagrate)^2)

alphahat = -log(a)/dt

betahat = sum(rate-a*lagrate ) / (N*(1-a))

 20

v2hat=sum((rate-lagrate*alpha-betahat*(1-a))^2)/N

sigmahat=sqrt(2*alpha_hat*v2hat/(1-a^2))

c(betahat,alphahat,sigmahat)}

Table 3.1 Summary of Vasicek Estimators

OLS 0.0799080219842794 1.55095522339525 0.0635675446843325

To estimate distribution of the parameters that fits Vasicek Model, Monte Carlo
Simulation explained in Chapter 2 is used and the codes of process are given:

init.value=test.data[sizedata[1]]

sizeofsim=1000

sizeofMonteCarlo=1000

delta.t=1/260

set.seed(1)

TEZ=matrix(data=NA,nrow=sizeofMonteCarlo,ncol=3)

for (j in 1:sizeofsim){

r=init.value

for (i in 2:sizeofsim){

delta.r=(outt[2]*(outt[1]-r[(i-
1)])*delta.t)+(outt[3]*rnorm(1)*sqrt(delta.t))

tempo=r[(i-1)]+delta.r

r=c(r,tempo)

TEZ[j,]=Vasicek.OLS(data=r,dt=1/260)

In each run of the Monte Carlo Simulation process, new parameter values are
estimated from the interest rate observations. Each run initiates from a different
 21

random point. Furthermore at each step randomized variates enter the model
distinguishing each run. Monte Carlo does this process thousands of times. This
process runs with two simulation steps. The histograms and plots of the
simulation results are given.

Figure 3.5 Plots of the parameters ( ) in simulation step

Figure 3.6 Histograms of the parameters ( ) in simulation step

As we can see in the Figure 3.5 observations of estimation, which of long run
mean from simulated values of interest rate, have extreme values (outliers)
relative to other observations. Ghosh and Vogt (2012) states that there is three
main approaches to eliminate outlier problems:
 22

1. To keep the outlier and treat it like any other data point
2. To modify its value
3. To eliminate it

Also they state that method 2 and 3 introduce statistical bias and may undervalue
the outlier.

To obtain distribution of parameters, increase our sample size the simulation step
is increased to 10000 times.

Figure 3.7 Plots of the parameters ( ) in simulation step

 23

Figure 3.8 Histograms of the parameters ( ) in simulation step

Now after the simulation with time step n=10000, parameters are normally
distributed as it can be observed in the Figure 3.7 and Figure 3.8. We give our
results in Table 3.2 %95 confidence intervals and Table 3.3 %99 confidence
intervals for the estimated parameters.

Table 3.2 %95 Confidence Intervals for the parameters

Parameters %95 Confidence Interval

Long Term Mean 0.06697832 0.09310074

Speed of Reversion 1.126175 2.326765

Volatility 0.06288429 0.06465596

Table 3.3 %99 Confidence Intervals for the parameters

 24

Parameters %99 Confidence Interval

Long Term Mean 0.06288056 0.09721137

Speed of Reversion 1.015408 2.585334

Volatility 0.06265065 0.06454089

As we can see from Table 3.2 , there is a significant difference between the
confidence interval borders for the long term mean and speed of reversion
parameters. Also increasing confidence level (%99) caused increase in interval
length. Any chosen parameter ignores the confidence intervals may cause over
estimation or under estimation of the model parameters and interest rates.

4 CONCLUSION

This study has been focuses on the distribution of the Vasicek Model
parameters. First we emphasized the importance of interest rate modeling
especially in insurance sector. As we indicate in introduction, with the new
regulations about capital requirements (Solvency II) modeling interest rates arise
one of the most important problem in that field. First we gave a general form of
the stochastic differential equation for the interest rate modeling. The focus is on
the Vasicek Model. The solution of the stochastic differential equation of Vasicek
Model is given and it is shown that two different parameter estimation methods,
 25

Ordinary Least Square and Maximum Likelihood estimation, can be applied to the
model. Then Monte Carlo Simulation Method is explained.

In the third part of this study is the application of the model on a data set from
TRLIBOR rates. We simulated new data set that is approximated by Euler
discretization with parameters from OLS estimators. This process applied with
Monte Carlo Method for 1000 time steps and 10000 time steps. First simulation
gave us distribution of the long term mean with outliers and also high variance. In
order to achieve greater accuracy, we increased the sample size to 10000. After
we clearly saw that our estimated parameter values from simulations are normally
distributed, we gave our results with the %95 and %99 confidence intervals for
the parameters.

In this study our aim is observing the distribution of the Vasicek Model
parameters. There are already various studies about parameter estimation of the
Vasicek and other diffusion models. But we used Monte Carlo Method for
obtaining the distribution of the estimated parameters. Through this technique we
observed probability distributions of parameters and we also got confidence
intervals. It can be clearly seen major difference between the borders of the
intervals for long term mean and mean reversion parameters. So during the
parameter estimation process confidence intervals are necessary for more
accuracy. This step is the key element of our study because probability
distributions are much more realistic way describing uncertainty in variables of
risk analysis.

5 BIBLIOGRAPHY

Berg T., Calibrating the Ornstein-Uhlenbeck (Vasicek) Model,2011

Sorwar, G., Barone-Adesi, G. and Allegretto, W., (2007)., "Valuation of

Derivatives Based on Single-Factor Interest Rate Models", Global finance
journal, Vol.18
 26

Brigo, D., Neugebauer, M., Dalessandro, A., and Triki, F., (2009). A stochastic
Processes Toolkit for Risk Management, Journal of Risk Management in
Financial Institutions, Vol. 3, 1.

Hsiao C. and Semmler W., Modeling Short Term Interest Rates, (2002), Elsevier
Science

Josef O. (2009), Models of Interst Rate Evolution-Vasicek and CIR Models,

Journal of Applied Mathematics, volume II

Önalan Ö., (2009) Vasicek ve CIR Modelleri Kullanılarak Oynaklık ve Faiz

Oranlarının Modellenmesi, Marmara Üniversitesi, İ.İ.B.F. Dergisi, Sayı II, S. 329-
344.

Şahin H. and Genç İ., (2009), Kısa Dönem Faiz Modellerinin Türkiye için
Ampirik Analizi, BDDK Bankacılık ve Finansal Piyasalar, Cilt 3, Sayı 2.

Vasicek O., (1977), An equilibrium characterization of the term structure, Journal

of Financial Economics, 5, 177.186.

Benninga S. and Wiener Z., (1998), Term Structure of Interest Rates,

Mathematica in Education and Research, Vol. 7 No. 2

Martin M., (2012), Assessing the Model Risk with Respect to the Interest Rate
Term Structure under Solvency II, FAU.

Zeytun S. and Gupta A., (2007), A Comparative Study of the Vasicek and the
CIR Model of the Short Rate, Berichte des Fraunhofer ITWM, Nr. 124

Hull J. and White A., (1990), Pricing Interst Rate Derivative Securities, The
Review of Financial Studies, Volume 3, number 4, p. 573-392.

Kalos H. and Whitlock P., (2008), Monte Carlo Methods, Wiley-Blackwell

 27

Tang C. and Chenb S., (2009), Parameter Estimation and Bias Correction for
Difusion Processes., Journal of Econometrics 149, (2009) 65-81.

Acar O., (2009), Avrupa Birliği Solvency II Projesi, Türkiye Sigorta ve

Reasürans Şirketler Birliği.

Giaccotto C., Stochastic Modeling of Interest Rates: Actuarial vs. Equilibrium

Approach, The Jourmal of Risk and Insurance.

Cairn A., (2004), Interest Rate Models, Princeton University Press.

Semir C. and Taray M., (2005), Türk Lirası Referans Faiz Oranı-TRLIBOR
Uygulaması, Bankacılar Dergisi, Sayı 54.