Stochastic

CT1-15: Stochastic interest rate models
Page 1
Chapter 15
Stochastic interest rate models
Syllabus objective
(xiv)
Show an understanding of simple stochastic models for investment returns.

1.
Describe the concept of a stochastic interest rate model and the

fundamental distinction between this and a deterministic model.
2.
Derive algebraically, for the model in which the annual rates of return
are independently and identically distributed and for other simple
models, expressions for the mean value and the variance of the
accumulated amount of a single premium.
3.
Derive algebraically, for the model in which the annual rates of return
are independently and identically distributed, recursive relationships
which permit the evaluation of the mean value and the variance of the
accumulated amount of an annual premium.
4.
Derive analytically, for the model in which each year the random
variable (1 i ) has an independent log-normal distribution, the
distribution functions for the accumulated amount of a single premium
and for the present value of a sum due at a given specified future time.
5.
Apply the above results to the calculation of the probability that a simple
sequence of payments will accumulate to a given amount at a specific
future time.
The Actuarial Education Company
IFE: 2016 Examinations
Page 2
Introduction
The calculations we have done so far have been based on the assumption that future
interest rates will take definite values that are known in advance. This is the
deterministic approach.
In this chapter we will study stochastic interest rate models where future interest rates
are assumed to be random. We cannot specify in advance precisely what interest rates
will apply. Instead, we can make an assumption about the statistical distribution of
future interest rates considered as a random variable.
The deterministic approach can provide only a single fixed answer to a problem. This
answer will be correct only if the assumptions made about future interest rates turn out
to be correct. In practice, an interest rate that errs on the cautious side may be chosen to
allow for uncertainty.
The stochastic approach is a more general method that allows us to determine both the
expected value, as a best estimate of the quantity of interest, and the variance, which
gives an indication of the likely spread of values. The stochastic approach can give
unreliable results if the statistical distribution used is not appropriate.
This chapter requires you to know many statistical techniques including means,
variances, probabilities and the normal distribution. We have included a very brief
summary of the main results you need in the appendix to this chapter. However, if you
have not studied statistics before or are not simultaneously studying Subject CT3 then
we strongly recommend that you spend some time looking at statistical techniques
before you study this chapter. If you have no material to study from, you can either
purchase StatsPack from ActEd or any A-Level (or Higher Level) statistics text book.
Simple models
1.1
Preliminary remarks
Page 3
Financial contracts are often of a long-term nature. Accordingly, at the outset of

many contracts there may be considerable uncertainty about the economic and
investment conditions which will prevail over the duration of the contract. Thus,
for example, if it is desired to determine premium rates on the basis of one fixed
rate of interest, it is nearly always necessary to adopt a conservative basis for
the rate to be used in any calculations.
An alternative approach to recognising the uncertainty that in reality exists is
provided by the use of stochastic interest rate models. In such models no single
interest rate is used. Variations in the rate of interest are allowed for by the
application of probability theory. Possibly one of the simplest models is that in
which each year the rate of interest obtained is independent of the rates of
interest in all previous years and takes one of a finite set of values, each value
having a constant probability of being the actual rate for the year.
For example, the effective annual rates of return that will apply during each of the next n
years might be i1 , i2 ,..., in , where ik , k 1,2,..., n are random variables with the
following discrete distribution:
0.06
i k = 0.08
0.10
with probability 0.2

Question 15.1
Calculate the mean, j, and the standard deviation, s, of ik .
Alternatively, the rate of interest may take any value within a specified range, the
actual value for the year being determined by some given probability density
function.
For example we might assume that the annual rates of return are uniformly distributed
between 5% and 10%.
Page 4
1.2
Fixed interest rate model

At this stage we consider briefly an elementary example, which although
necessarily artificial provides a simple introduction to the probabilistic ideas
implicit in the use of stochastic interest rate models.
Suppose that an investor wishes to invest a lump sum of P into a fund which
grows under the action of compound interest at a constant rate for n years. This
constant rate of interest is not known now, but will be determined immediately
after the investment has been made.
The accumulated value of the sum will, of course, be dependent on the rate of
interest. In assessing this value before the interest rate is known, it could be
assumed that the mean interest rate will apply. However, the accumulated value
using the mean rate of interest will not equal the mean accumulated value. In
algebraic terms:
n
P 1
(i j p j ) P
p j (1 i j )n
j 1
j 1
where:
i j is the jth of k possible rates of interest

p j is the probability of the rate of interest i j
The above result is easily demonstrated with a simple numerical example:

Example
Calculate the expected accumulated value at the end of 5 years of an initial investment
of 5,000 if the returns from the investment are assumed to conform to the fixed interest
rate model with the distribution of interest rates specified in Question 15.1.
Also calculate the accumulated value at the mean rate of interest.
Page 5
Solution
The expected accumulated value must be calculated directly:

5, 000 E ( S5 ) = 5, 000 E[(1 + i )5 ]
= 5, 000(0.2 1.065 + 0.7 1.085 + 0.1 1.105 )
= 5, 000 1.4572
= 7, 286
The mean rate of interest was found in Question 15.1 to be 7.8%.

Therefore the accumulated value at the mean rate of interest is:
5,000 1078
. 5 7,279
Question 15.2
What is the variance of the accumulated value of this investment?
This model, where the effective annual interest rate of return is a single unknown rate i
and will apply throughout the next n years, is often known as the fixed interest rate
model.
For the fixed interest rate model, the mean and variance of the accumulated value of an
investment must be calculated from first principles.
1.3
Varying interest rate model

In our previous example the effective annual rate of interest was fixed
throughout the duration of the investment. A more flexible model is provided by
assuming that over each single year the annual yield on invested funds will be
one of a specified set of values or lie within some specified range of values, the
yield in any particular year being independent of the yields in all previous years
and being determined by a given probability distribution.
This model is often called the varying interest rate model. The main difference between
this and the fixed interest rate model is that, in the varying rate model, the interest rates
can be different in each future year, whereas, in the fixed rate model, the same
(unknown) interest rate will apply in each future year.
Page 6
Measure time in years.
Consider the time interval [0, n ] subdivided into
successive periods [0,1],[1,2], ,[ n - 1, n ] .
For t = 1,2, , n let it be the yield
obtainable over the tth year, ie the period [t - 1, t ] .
Assume that money is
invested only at the beginning of each year. Let Ft denote the accumulated
amount at time t of all money invested before time t and let Pt be the amount of
money invested at time t. Then, for t = 1,2,3, :
Ft = (1 + it )(Ft - 1 + Pt - 1 )
(1.1)
It follows from this equation that a single investment of 1 at time 0 will

accumulate at time n to:
Sn = (1 + i1 )(1 + i2 ) (1 + in )
(1.2)
Similarly a series of annual investments, each of amount 1, at times

0,1,2, , n - 1 will accumulate at time n to:
An = (1 + i1 )(1 + i2 )(1 + i3 ) (1 + in )
+ (1 + i2 )(1 + i3 ) (1 + in )
+
(1.3)
+ (1 + in - 1 )(1 + in )
+ (1 + in )
Note that An and Sn are random variables, each with its own probability
distribution function.
For example, if the yield each year is 0.02, 0.04, or 0.06 and each value is equally
likely, the value of Sn will be between 1.02n and 1.06n . Each of these extreme
values will occur with probability (1 3)n .
Question 15.3
What is the probability that Sn will take the value 102
. 104
. n 1 ?
Page 7
In general, a theoretical analysis of the distribution functions for An and Sn is

somewhat difficult. It is often more useful to use simulation techniques in the
study of practical problems. However, it is perhaps worth noting that the
moments of the random variables An and Sn can be found relatively simply in
terms of the moments of the distribution for the yield each year. This may be
seen as follows.
Moments of Sn
Lets consider the kth moment of Sn .
From Equation (1.2) we obtain:
(Sn )k
(1 it )k
t 1
and hence:
n
E [Snk ] = E
(1 + it )k
t =1
E [(1 + it )k ]
(1.4)
t =1
since (by hypothesis) i1 , i2 , , in are independent. Using this last expression

and given the moments of the annual yield distribution, we may easily find the
moments of Sn .
For example, suppose that the yield each year has mean j and variance s 2 .
Then, letting k = 1 in Equation (1.4), we have:
E [Sn ] =
E [(1 + it )]
t =1
n
(1 + E [it ])
t =1
= (1 + j )n
(1.5)
since, for each value of t, E [ it ] = j .
Page 8
With k = 2 in Equation (1.4) we obtain:

E [Sn2 ] =
E [(1 + 2it + it2 )]

t =1
n
(1 + 2E [it ] + E [it2 ])
t =1
= (1 + 2 j + j 2 + s 2 )n
(1.6)
since, for each value of t:
E [ it2 ] = (E [ it ])2 + var[ it ] = j 2 + s 2

The variance of Sn is:
var[Sn ] = E [Sn2 ] - (E [Sn ])2
= (1 + 2 j + j 2 + s 2 )n - (1 + j )2n
(1.7)
from Equations (1.5) and (1.6).
or equivalently:
n
E[ Sn2 ] = E[(1 + it ) 2 ]
t =1
n
= var[(1 + it )] + E 2 [(1 + it )]
t =1
n
= s 2 + (1 + j ) 2
t =1
= s 2 + (1 + j ) 2
Question 15.4
Explain the steps in the proof above.
Page 9
This means that:

var[ Sn ] [(1 j ) 2 s2 ]n (1 j ) 2 n
These arguments are readily extended to the derivation of the higher moments of
Sn in terms of the higher moments of the distribution of the annual rate of
interest.
Example
Calculate the variance of the accumulated value of the investment in the example on
page 4, assuming the returns conform to the varying interest rate model with the
specified distribution.
Solution
Using the formula gives:

var(5, 000 S5 ) = 5, 0002 var( S5 )
= 5, 0002 [((1 + j ) 2 + s 2 )5 - (1 + j )10 ]
= 5, 0002 [(1.0782 + 0.000116)5 - (1.078)10 ]
= (163) 2
Question 15.5
Calculate the mean and variance of the accumulated value of an initial investment of
40,000 at the end of 25 years if the annual rates of return are assumed to conform to
the varying interest rate model and follow a Gamma (16,200) distribution. (You can
find formulae for the mean and variance of a Gamma ( , ) distribution on page 12 of
the Tables.)
Page 10
Moments of An
Remember that An is a random variable that represents the accumulated value at time n
of a series of annual investments, each of amount 1, at times 0,1, 2, , n - 1 .
i1 , i2 , , in are independent random variables, each with a mean j and a variance s2 .
From Equation (1.3):
An -1 = (1 + i1 )(1 + i2 ) (1 + in -1 )
+ (1 + i2 ) (1 + in -1 )
+ (1 + in - 2 )(1 + in -1 )
+ (1 + in -1 )
and:
An = (1 + i1 )(1 + i2 ) (1 + in -1 )(1 + in )
+ (1 + i2 ) (1 + in )
+ (1 + in -1 )(1 + in )
+ (1 + in )
It follows from Equation (1.3) (or from Equation (1.1)) that, for n 2 :
An (1 i n )(1 An 1 )
(1.8)
Equation (1.8) can also be deduced easily by general reasoning.

An1 is the
accumulated value at time n 1 of a series of annual payments, each of amount 1, at
times 0, 1, 2, , n 2 . The value, at time n 1 , of the same series of payments
together with an extra payment at time n 1 is 1 An1 . Accumulating this value
forward to time n gives (1 in )(1 An 1 ) and this is equivalent to An .
Page 11
The usefulness of Equation (1.8) lies in the fact that, since An - 1 depends only on
the values i1, i2 , , in - 1 , the random variables in and An - 1 are independent. (By
assumption the yields each year are independent of one another.) Accordingly,
Equation (1.7) permits the development of a recurrence relation from which may
be found the moments of An . We illustrate this approach by obtaining the mean
and variance of An .
Let:
mn = E [ An ]
and let:
mn = E [ An2 ]
Since:
A1 = 1 + i1
it follows that:
E[ A1 ] = E[1 + i1 ] = 1 + E[i1 ] = 1 + j
1 1 j
and:
m1 = E[ A12 ] = E[(1 + i1 ) 2 ] = 1 + 2 E[i1 ] + E[i12 ]

m1 1 2 j j 2 s 2
where, as before, j and s 2 are the mean and variance of the yield each year.
Taking expectations of Equation (1.8), we obtain (since in and An - 1 are
independent):
n (1 j )(1 n 1)
n2
Page 12
Applying the recursive formula repeatedly for each year, gives:
mn = (1 + j )[1 + mn -1 )] = (1 + j ) + (1 + j ) mn -1
= (1 + j ) + (1 + j ) 2 [1 + mn - 2 ] = (1 + j ) + (1 + j ) 2 + (1 + j ) 2 mn - 2
=
= (1 + j ) + (1 + j ) 2 + (1 + j )3 + + (1 + j ) n
which is the formula for sn| , calculated at the expected interest rate j .
This equation, combined with initial value 1 , implies that, for all values of n:
n sn|
at rate j
(1.9)
Thus the expected value of An is simply sn| , calculated at the mean rate of
interest.
Since:
An2 (1 2i n i n2 )(1 2An 1 An2 1)

by taking expectations we obtain, for n 2 :
mn (1 2 j j 2 s 2 )(1 2 n 1 mn 1)
(1.10)
As the value of n 1 is known (by Equation (1.9)), Equation (1.10) provides a

recurrence relation for the calculation successively of m2 , m3 , m4 , . The
variance of An may be obtained as:
var[ An ] = E [ An2 ] - (E [ An ])2 = mn - mn2
(1.11)
In principle the above arguments are fairly readily extended to provide

recurrence relations for the higher moments of An .
The arguments are also easily extended to provide recurrence relations for other series
of investments. Letting Ft again represent the accumulated amount at time t of all
money invested before time t and let Pt be the amount of money invested at time t. We
stated in Equation (1.1) that:
Ft (1 it )( Ft 1 Pt 1 )
Page 13
Therefore the mean of Ft can be found from the recursive relationship:

E ( F0 ) = 0
E ( Ft ) = (1 + j )( E ( Ft -1 ) + Pt -1 )
Question 15.6
Explain what is wrong with the following derivation a student has used.
We know that: Ak (1 ik )(1 Ak 1 )
(1)
So, finding the variance of both sides gives:
var( Ak ) var[(1 ik )(1 Ak 1 )]
(2)
By independence, this is:

var( Ak ) var[(1 ik )] var[(1 Ak 1 )]
(3)
But we know that var[(1 ik )] var(ik ) s2 and var[(1 Ak 1 )] var( Ak 1 )
(4)
So this gives:
var( Ak ) s2 var( Ak 1 )
(5)
Applying this equation recursively gives:

var( An ) = s 2 var( An -1 ) = s 4 var( An - 2 ) = = s 2n var( A0 )
(6)
Since there is definitely no money in the fund at the outset, we know that:
var( A0 ) 0
(7)
var( An ) s2 n 0 0
(8)
So:
Page 14
Example
A company considers that on average it will earn interest on its funds at the rate
of 4% pa. However, the investment policy is such that in any one year the yield
on the companys funds is equally likely to take any value between 2% and 6%.
For both single and annual premium accumulations with terms of 5, 10, 15, 20,
and 25 years and single (or annual) investment of 1, find the mean
accumulation and the standard deviation of the accumulation at the maturity
date. (Ignore expenses.)
Solution
The annual rate of interest is uniformly distributed on the interval [0.02,0.06] .
The corresponding probability density function is constant and equal to 25 (ie
1 (0.06 - 0.02) ). The mean annual rate of interest is clearly:
j = 0.04
and the variance of the annual rate of interest is:

s2
1
(0.06 0.02)2 34 10 4
12
This formula for the variance of a uniform random variable is given on page 13 of the
Tables.
We are required to find
E [ An ] ,
(var[ An ]) 2 ,
E [Sn ] , and
(var[Sn ]) 2
for
n = 5,10,15,20 and 25 .
Substituting the above values of j and s 2 in Equations (1.5) and (1.7), we

immediately obtain the results for the single premiums.
For example:
E[ S5 ] = 1.045 = 1.21665
var[ S5 ] = (1 + 0.08 + 0.042 + 43 10 -4 )5 - 1.0410 = 0.000913
standard deviation [ S5 ] = 0.03021
Page 15
For the annual premiums we must use the recurrence relation (1.10) (with
n 1 s | at 4%) together with Equation (1.11).
n 1
Equation (1.9) is used to calculate E[ An ] .

The results are summarised in Table 1. It should be noted that, for both annual
and single premiums, the standard deviation of the accumulation increases
rapidly with the term.
Table 1
Term
(years)
Single premium 1
Annual premium 1
Mean
accumulation
()
1.21665
Standard
deviation
()
0.03021
Mean
accumulation
()
5.63298
Standard
deviation
()
0.09443
10
1.48024
0.05198
12.48635
0.28353
15
1.80094
0.07748
20.82453
0.57899
20
2.19112
0.10886
30.96920
1.00476
25
2.66584
0.14810
43.31174
1.59392
Question 15.7
Check the values given in the table for E[ A5 ] and standard deviation [ A5 ] .
Question 15.8
An investor invests 1 unit at time t 0 and a further 2 units at time t 2 . Use
recursive formulae to calculate the accumulated value of the fund at time t 5 ,
assuming the varying interest rate model applies and that the expected interest rate for
each year is 10%.
Calculate the accumulated value based on the corresponding deterministic model and
comment on your answer.
Page 16
The log-normal distribution

In general a theoretical analysis of the distribution functions for An and Sn is
somewhat difficult, even in the relatively simple situation when the yields each
year are independent and identically distributed. There is, however, one special
case for which an exact analysis of the distribution function for Sn is particularly
simple.
Because of the compounding effect of interest, the accumulated value of an investment

bond grows multiplicatively. This makes the log-normal distribution a natural choice
for modelling the annual growth factors 1 i , since a log-normal random variable can
take any positive value and has the following multiplicative property:
If X 1 ~ log N ( 1, 12 ) and X 2 ~ log N ( 2, 22 ) are independent random variables, then:
X 1 X 2 ~ log N ( 1 2, 12 22 )
The graph below illustrates the shape of the PDF of a typical log-normal distribution
used to model annual growth rates.
log N (0.075, 0.12 )
Question 15.9
What are the mean and standard deviation of the log-normal distribution shown in the
graph?
Suppose that the random variable log(1 + it ) is normally distributed with mean
and variance 2 . In this case, the variable (1 + it ) is said to have a log-normal
distribution with parameters and 2 .
Page 17
Equation (1.2) is equivalent to:

log Sn
log(1 it )
t 1
The sum of a set of independent normal random variables is itself a normal

random variable.
Hence, when the random variables (1 it ) (t 1) are
independent and each has a log-normal distribution with parameters and and
2 , the random variable Sn has a log-normal distribution with parameters n

and n 2 .
ie
log Sn ~ N (n , n 2 )
or
log Sn n
~ N (0,1)
n
Since the distribution function of a log-normal variable is readily written down in

terms of its two parameters, in the particular case when the distribution function
for the yield each year is log-normal we have a simple expression for the
distribution function of Sn .
Similarly for the present value of a sum of 1 due at the end of n years:
Vn = (1 + i1 )-1 (1 + in )-1
logVn = - log(1 + i1 ) - - log(1 + in )
Since, for each value of t, log(1 it ) is normally distributed with mean m and
variance 2 , each term on the right hand side of the above equation is normally
distributed with mean - m and variance s 2 . Also the terms are independently
distributed. So, logVn is normally distributed with mean - n m and variance n 2 .
That is, Vn has log-normal distribution with parameters n and n 2 .
By statistically modelling Vn , it is possible to answer questions such as:
to a given point in time, for a specified confidence interval, what is the

range of values for an accumulated investment
what is the maximum loss which will be incurred with a given level of
probability
Page 18
Although outside the scope of this Subject, it is interesting to note that such
techniques may be extended readily to model and predict the behaviour of
portfolios of investments. These techniques are referred to as Value at Risk or
VaR methods and are covered in more detail in subjects CT8, Financial
Economics, and ST6, Finance and Investment Specialist Technical B. One
possible definition of Value at Risk is a portfolios maximum loss from an
adverse market movement, within a specified confidence interval and over a
defined period of time.
As with all statistical modelling techniques, the results of VaR can only be as
good as the statistical model of the performance of the underlying investments.
In all investment markets, even seemingly efficient ones, it continues to prove
very difficult to choose a reliable statistical model which is robust over even
short periods of time.
Example
If the annual growth factors 1 ik for individual years have a log N ( , 2 ) distribution,
then the distribution functions for the accumulated value Sn and the discounted value
Vn can be expressed in terms of ( x ) , the distribution function of the standard normal
distribution:
log s n
P ( Sn s)
and
n
log s n
P (Vn s) 1
Proof
The formula for the distribution function of Sn follows immediately from the result that
Sn has a log-normal distribution with parameters n and n 2 .

Vn is log-normally distributed with parameters n and n 2 and so:
log s (n )
P (Vn s )
log s n
1
These results enable us to calculate probabilities for the range of the accumulated
amount of a single payment.
Page 19
Example
Find the upper and lower quartiles for the accumulated value at the end of 5 years of an
initial investment of 1,000, using the varying interest rate model and assuming that the
annual growth rate has a log-normal distribution with parameters 0.075 and
2 0.0252 .
Solution
By definition, the accumulated amount X 1,000S5 will exceed the upper quartile u
with probability 25%, ie:
0.75 P( X u) P(1,000S5 u) P( S5 u 1,000)

So, from the formula for the distribution function:
log(u 1, 000) n
0.75 P( S5 u 1, 000)
From the tables of the distribution function of the normal distribution on page 162 of
the Tables, we find that F(0.6745) = 0.75 . So, we must have:
log(u 1, 000) - n m
= 0.6745
s n
ie u = 1, 000e5 m +0.6745s
= 1,511
Similarly, the lower quartile is:
l = 1, 000e5 m -0.6745s
= 1, 401
Page 20
Question 15.10
A man now aged exactly 50 has built up a savings fund of 400,000. In order to retire
at age 60, he will require a fund of at least 600,000 at that time. Calculate the
probability that, if he makes no further contributions to the fund, he will be able to retire
at age 60. Assume that annual growth rates vary independently from year to year and
have the log-normal distribution shown in the graph on page 16.
Question 15.11
Derive expressions for the mean and variance of the accumulated value of 1 unit after n
years for the fixed interest rate model, assuming that the annual growth rate has a lognormal distribution with parameters and 2 .
Question 15.12
A lump sum of $14,000 will be invested at time 0 for 4 years at an annual rate of
interest i . The interest rate, once determined, will be the same in each of the four years.
1 i has a log-normal distribution with mean 1.05 and variance 0.007. Calculate the
probability that the investment will accumulate to more than $20,000 in 4 years time.
Page 21
Exam-style questions
Before you finish, try the following two exam-style questions on stochastic interest
rates. The solutions are over the page.
Question 1
Interest rates over the next five years are fixed at either 4% pa with probability 0.2 or
5% pa with probability 0.8. What is the standard deviation of the present value of a
payment of 25,000 in 5 years time?
[3]
Question 2
The annual returns, i , on a fund are independent and identically distributed, with a
mean of 6% and a standard deviation of 3%. Each year, the distribution of 1 i is
log-normal with parameters and 2 .
(i)
Calculate the values of and 2 .
(ii)
Calculate the probability that the accumulation of a single investment of 1 will

be greater than 110% of its expected value after 10 years.
[4]
[Total 8]
[4]
Page 22
Solution 1
The present value will either be

25, 000
1.055
25, 000
1.045
= 20548.18 with probability 0.2, or
= 19588.15 , with probability 0.8.
The mean of these values is 20,548.18 0.2 + 19,588.15 0.8 = 19, 780.16 .
The standard deviation of these values is:
20,548.182 0.2 + 19,588.152 0.8 - 19780.162 = 384
Solution 2
(i)
Parameters of the log-normal distribution
We know that 1 i log N , 2 , and that E i 0.06 and var i 0.032 . Using
the formulae for the mean and variance of the log-normal distribution from the Tables:
E 1 i 1 E i 1.06 e
and:
12 2
var 1 i var i 0.032 e2
(Equation 1)
2
e 1
(Equation 2)
Squaring Equation 1, we have:

2
12 2
2 2
1.062
e
e
Substituting this into Equation 2 gives:
0.032
2
1 0.00080068
1.062 e 1 0.032 2 ln
1.062
Page 23
So, using Equation 1:
ln 1.06 12 2 0.057869
(ii)
Probability
Let S10 denote the accumulated value at time 10 of an investment of 1 at time 0. We

require the probability P S10 1.1E S10 .
As the interest rate in each year is independent of that in other years, using the varying
rate model, we have:
E S10 1 j 1.0610 since j E i 0.06
10
The distribution of S10 under this model is:
S10 log N 10 ,10 2 log N 0.57869,0.0080068
So:
P S10 1.1E S10 P S10 1.11.0610 P ln S10 ln 1.11.0610
Using Z N 0,1 , we have:
10
P ln S10 ln 1.11.06
ln 1.1 1.0610 0.57869
P Z
0.0080068
P Z 1.110
1 1.110
1 0.86650
0.1335
ie a probability of 13.35%.
Page 24
Appendix basic statistical results

This appendix gives a brief summary of the statistical results that are used in this
chapter. All of these can be found in the CT3 course notes, the StatsPack or any A level
(or equivalent) statistics textbook.
4.1
Expectation
The expectation (or mean) of a discrete random variable which takes values x with
probabilities P( X = x) is given by:
E ( X ) = xP( X = x)
x
We can also find the expectations of functions of X :
E ( X 2 ) = x 2 P( X = x)
x
In general:
E[ g ( X )] = g ( x) P ( X = x)
x
The following relationships hold:

E (aX + b) = aE ( X ) + b
E ( X1 + X 2 ) = E ( X1 ) + E ( X 2 )
E ( X 1 X 2 ) = E ( X1 ) E ( X 2 )
if X1 , X 2 are independent
4.2
Page 25
Variance
The variance of a discrete random variable which takes values x with probabilities
P( X = x) is given by:
var( X ) = E ( X 2 ) - [ E ( X )]2
The variance measures the average square distance of each value from the mean.
Hence, the variance is a measure of the square of the spread of the distribution.
The following relationships hold:
var(aX + b) = a 2 var( X )
var( X1 + X 2 ) = var( X1 ) + var( X 2 ) if X1 , X 2 are independent
Standard deviation
The standard deviation is the square root of the variance and therefore measures the
spread of the distribution.
4.3
Lognormal distribution
A random variable Y has a lognormal distribution if:
log Y ~ N ( m ,s 2 )
This is often abbreviated to Y ~ log N ( m ,s 2 ) .

ie the natural log ( log e = ln ) of Y has a normal distribution with mean m and variance
s 2.
Note that m and s 2 are the mean and variance of the normal distribution and not the
lognormal distribution. As such they are called the parameters of the lognormal
distribution. The actual mean and variance of the lognormal can be found on page 14 of
the Tables:
E (Y ) = e m +s
and
var(Y ) = e2 m +s (es - 1)
Page 26
Calculating lognormal probabilities

Y has a lognormal distribution with parameters m = 5 and s 2 = 3 .
P(Y > 400) .
Calculate
First we log both sides to convert it to a normal distribution probability:
P (Y > 400) = P (ln Y > ln 400)

Next we standardise it, that is we convert it to the standard normal Z ~ N (0,1) . We do
this by subtracting m and dividing by s :
ln 400 - 5
P (Y > 400) = P Z >
3
= P( Z > 0.57242)
Finally, we can find the probability by using the standard normal tables on pages 160161 of the Tables. Since these only list less than probabilities, we have:
P (Y > 400) = 1 - P( Z < 0.57242)

From the Tables we see that:
P( Z < 0.57) = 0.71566
P( Z < 0.58) = 0.71904
We could round it and say that 0.57242 is approximately 0.57, hence the probability is
approximately 1 - 0.71566 0.284 . Alternatively, we see that 0.57242 is 24.2%
between 0.57 and 0.58:
P ( Z < 0.57242) = P ( Z < 0.57) + 0.242( P ( Z < 0.58) - P ( Z < 0.57))
= 0.71566 + 0.242(0.71904 - 0.71566)
= 0.71648
Hence:
P (Y > 400) = 1 - 0.71648 = 0.28352
Page 27
End of Part 4
What next?
1.
Briefly review the key areas of Part 4 and/or re-read the summaries at the end
of Chapters 14 and 15.
2.
Attempt some of the questions in Part 4 of the Question and Answer Bank. If
you dont have time to do them all, you could save the remainder for use as part
of your revision.
3.
Attempt Assignment X4.
Time to consider rehearsal products

Mock Exam A / AMP and Marking There are three separate mock exam papers that
you can attempt and get marked. A recent student survey found that students who do a
mock exam of some form have significantly higher pass rates. Students have said:
I wanted an indication of how ready I was for the actual exam.
[The marker gave] Helpful comments, telling me where to improve and how I can
prepare better for the exam. Detailed corrections and handy tips on what I
should concentrate on learning well as it could count for a lot. [They also gave] tips
on exam technique which are useful.
You can find lots more information on our website at www.ActEd.co.uk.

Buy online at www.ActEd.co.uk/estore
And finally ...

Good luck!
Page 28
This page has been left blank so that you can keep the chapter summaries
together for revision purposes.
Page 29
Chapter 15 Summary
A stochastic interest rate model provides information about the distribution of financial
outcomes. This distribution can be used to find best estimates and probabilities.
The varying interest rate model and the fixed interest rate model provide formulae for
the mean and variance of the accumulated amount of a fund or the present value of a
future payment.
Varying interest rate model (single premium):
E ( Sn ) = (1 + j )n
var( Sn ) = [(1 + j )2 + s 2 ]n - (1 + j ) 2n
Fixed interest rate model (single premium):

E ( Sn ) = E[(1 + i ) n ]
var( Sn ) = E[(1 + i ) 2n ] - ( E[(1 + i ) n ]) 2
For the varying interest rate model, the variance and higher moments of the
accumulated amount of a series of payments can be calculated using recursive formulae.
Varying interest rate model (annual premium)
E ( An ) =
sn|
at rate j
Recursive formulae for E ( An ) :

E ( A0 ) = 0
and:
E ( Ak ) = (1 + j )[1 + E ( Ak -1 )] ( k = 1, 2,..., n )
Recursive formulae for var( An ) :
E ( A02 ) = 0
and:
E ( Ak2 ) = [(1 + j ) 2 + s 2 ][1 + 2 E ( Ak -1 ) + E ( Ak2-1 )]
( k = 1, 2,..., n )
Page 30
Then:
var( An ) = E ( An2 ) - [ E ( An )]2
For the fixed interest rate model with annual premiums, calculate the mean and variance
directly from the definitions.
The log-normal distribution can be used to model the annual growth rate. This allows
probabilities to be determined in terms of the distribution function of the normal
distribution.
The lognormal model formulae for the varying interest rate model are:
log Sn - n m
s n
~ N (0,1)
log s n
P(Sn s)
log s n
P ( Sn1 s ) 1
Page 31
Chapter 15 Solutions
Solution 15.1
The mean is:
j = E (ik ) = 0.2 0.06 + 0.7 0.08 + 0.1 0.10 = 0.078
We can calculate the variance using the formula var(ik ) = E (ik2 ) - [ E (ik )]2 :
E (ik2 ) = 0.2 0.062 + 0.7 0.082 + 0.1 0.102 = 0.0062
s 2 = var(ik ) = E (ik2 ) - [ E (ik )]2 = 0.0062 - 0.0782 = 0.000116 = 0.01082
So, the mean is 7.8% and the standard deviation is 1.08%.
Solution 15.2
The variance of the accumulated value must be calculated directly:
var = 5, 0002 ( E[(1 + i )10 ] - ( E[(1 + i )5 ]) 2 )
= 5, 0002 [(0.2 1.0610 + 0.7 1.0810 + 0.1 1.1010 )
-(0.2 1.065 + 0.7 1.085 + 0.1 1.105 ) 2 ]
= 5, 0002 (2.128791 - 1.4572262 ) = (363)2
(You need to keep a few extra decimals in this last calculation to avoid losing accuracy,
since the calculation involves subtracting two numbers of similar magnitude.)
Solution 15.3
There are n different years in which the 2% could fall and so:
1 1
Probability n
3 3
n 1
n
3n
Page 32
Solution 15.4
The first step, namely:
n
E[ Sn2 ] = E[(1 + it ) 2 ]
t =1
n
= var[(1 + it )] + E 2 [(1 + it )]
t =1
comes straight from the definition of the variance:

var[ g (i )] = E[ g 2 (i )] - E 2 [ g (i )] E[ g 2 (i )] = var[ g (i )] + E 2 [ g (i )]
The second step, namely:
(var[(1 + it )] + E 2[(1 + it )])

n
t =1
n
= s 2 + (1 + j ) 2
t =1
comes from the facts that:

var[(1 + it )] = var[it ] = s 2
and
E 2 [(1 + it )] = (1 + E[it ]) = (1 + j ) 2
2
The last step follows since the factor s 2 + (1 + j ) 2 is independent of n.
Page 33
Solution 15.5
Let ik be the return in year k. Therefore:
ik Gamma(16, 200)
Using the formulae for the mean and variance of the gamma distribution from page 12
of the Tables:
j E (ik )
16
0.08
200
s 2 = var(ik ) =
a
16
=
= (0.02) 2
2
2
l
200
So, the mean of the accumulated amount is:

40, 000 E ( S25 ) = 40, 000(1 + j ) 25 = 40, 000 1.0825 = 273,900
and the variance is:
var(40, 000 S25 ) = 40, 0002 var( S25 )
= 40, 0002 [(1 + j ) 2 + s 2 ]25 - (1 + j )50
= 40, 0002 [(1.082 + 0.022 ) 25 - (1.08)50 ]

= (25, 400) 2
(If you keep exact values during the calculation, you should get 273,939 and
(25, 417) 2 .)
Solution 15.6
Step (3) is invalid. Independence doesnt allow you to factorise variances (which is
why we had to work out E ( Ak2 ) instead).
In fact, this step would only be valid if the terms (1 + ik ) and (1 + Ak -1 ) are constant. If
this was the case, we would be dealing with a deterministic model, and the calculation
then gives the correct answer of zero, since there is no uncertainty in the final amount.
Page 34
Solution 15.7
E[ A5 ] = s5|4% =
1.045 - 1
= 5.63298
0.04 1.04
In order to calculate the standard deviation of A5 , we first need to calculate m5 from

the recursive formula:
mn = (1 + 2 j + j 2 + s 2 )(1 + 2 mn -1 + mn -1 )
The values required are tabulated below:
n
mn
1
2
3
4
5
1.08173
4.50189
10.54158
19.50853
31.73933018
mn
1.04
2.1216
3.24646
4.41632
5.6329755
var[ A5 ] = m5 - ( E[ A5 ])5 = 31.73933018 - 5.63297552 = 0.0089172

The standard deviation is:
0.0089172 = 0.09443
Warning: this answer is very sensitive to rounding. Try to carry forward as many
decimal places as possible.
Page 35
Solution 15.8
Let An be the accumulated amount at the end of n years of the series of cashflow given.
We can use a recursive approach similar to before to obtain:
E ( A0 ) = 0
E ( A1 ) = 1.1 [1 + E ( A0 )] = 1.1 (1 + 0) = 1.1
E ( A2 ) = 1.1 E ( A1 ) = 1.1 1.1 = 1.21
E ( A3 ) = 1.1 [2 + E ( A2 )] = 1.1 (2 + 1.21) = 3.531

E ( A4 ) = 1.1 E ( A3 ) = 1.1 3.531 = 3.8841
E ( A5 ) = 1.1 E ( A4 ) = 1.1 3.8841 = 4.27251
If we had used the corresponding deterministic model (ie an ordinary actuarial
calculation) based on the expected interest rate of 10%, we would have got:
1.15 + 2 1.13 = 1.61051 + 2 1.331 = 4.27251
So the corresponding deterministic model gives the same answer.
Solution 15.9
The mean and variance of the
2
log N ( m , s 2 )
distribution are
e m +s
and
e 2 m +s (es - 1) . These formulae are given in the Tables on page 14.

So, with m = 0.075 and s 2 = 0.12 , the mean is 1.0833 and the variance is 0.10862 ie
the annual rate of growth will have a mean value of 8.33% and a standard deviation of
10.86%.
Page 36
Solution 15.10
If he makes no further contributions, the accumulated fund at age 60 will be
400,000S10 .
So, the probability that the fund will be sufficient for him to retire is:
600, 000
P (400, 000 S10 600, 000) 1 P S10

400, 000
log1.5 10
1
10
1 (1.0895) (1.0895) 0.862
Solution 15.11
For the fixed interest rate model:
Sn = (1 + i ) n
If 1 + i ~ log N ( m , s 2 ) , then:
log(1 + i ) ~ N ( m , s 2 )
So:
log(1 + i ) n = n log(1 + i ) ~ N (n m , n 2s 2 )
So:
Sn = (1 + i )n ~ log N (n m , n 2s 2 )
Note that the distribution of Sn obtained here is different to that derived in the chapter
of Sn ~ log N (n m , ns 2 ) , which applies to the varying rate model, where the returns in
each year are independent.
Page 37
Using the formulae for the mean and variance of the log-normal distribution:
E ( Sn ) = e
n m + 12 n 2s 2
2 2
2 2
var( S n ) = e2 n m + n s (e n s - 1)
Solution 15.12
We first need to find the values of the parameters for the log-normal distribution. Using
the formulae for the mean and variance from the Tables:
e
and:
12 2
e 2
1.05
(Equation 1)
e 1 0.007
(Equation 2)
Squaring Equation 1, we have:

2
12 2
2 2
1.052
e
e
Substituting this into Equation 2 gives:
2
0.007
1 0.006329
1.052 e 1 0.007 2 ln
1.052
So, using Equation 1:
ln(1.05) 12 2 0.04563
We know that 1 i log N , 2 . Since we have a constant interest rate over the 4
years, the fixed rate model applies. Letting Sn denote the accumulated value at time n
of an investment of 1 made at time 0, we have S n log N n , n 2 2 , as derived in

Question 15.11.
Page 38
We need the probability P 14,000 S 4 20,000 where:
S 4 log N 4 ,16 2 log N 0.1825,0.1013
So:
P 14,000 S4 20,000 P S 4 1.429 P ln S4 0.3567
Letting Z N 0,1 , we have:

0.3567 0.1825
P ln S 4 0.3567 P Z
0.1013
P Z 0.547
1 0.547
1 0.7078
0.2922

Stochastic

Hochgeladen von

Dokumentinformationen

Originalbeschreibung:

Copyright

Verfügbare Formate

Dieses Dokument teilen

Dokument teilen oder einbetten

Freigabeoptionen

Stufen Sie dieses Dokument als nützlich ein?

Sind diese Inhalte unangemessen?

Copyright:

Verfügbare Formate

Stochastic

Hochgeladen von

Copyright:

Verfügbare Formate

CT1-15: Stochastic interest rate models

Show an understanding of simple stochastic models for investment returns.

Describe the concept of a stochastic interest rate model and the

The Actuarial Education Company

IFE: 2016 Examinations

CT1-15: Stochastic interest rate models

IFE: 2016 Examinations

The Actuarial Education Company

CT1-15: Stochastic interest rate models

Financial contracts are often of a long-term nature. Accordingly, at the outset of

with probability 0.2

The Actuarial Education Company

IFE: 2016 Examinations

CT1-15: Stochastic interest rate models

Fixed interest rate model

i j is the jth of k possible rates of interest

The above result is easily demonstrated with a simple numerical example:

IFE: 2016 Examinations

The Actuarial Education Company

CT1-15: Stochastic interest rate models

The expected accumulated value must be calculated directly:

The mean rate of interest was found in Question 15.1 to be 7.8%.

Varying interest rate model

The Actuarial Education Company

IFE: 2016 Examinations

CT1-15: Stochastic interest rate models

Measure time in years.

Consider the time interval [0, n ] subdivided into

successive periods [0,1],[1,2], ,[ n - 1, n ] .

For t = 1,2, , n let it be the yield

obtainable over the tth year, ie the period [t - 1, t ] .

Assume that money is

It follows from this equation that a single investment of 1 at time 0 will

Similarly a series of annual investments, each of amount 1, at times

IFE: 2016 Examinations

The Actuarial Education Company

CT1-15: Stochastic interest rate models

In general, a theoretical analysis of the distribution functions for An and Sn is

since (by hypothesis) i1 , i2 , , in are independent. Using this last expression

since, for each value of t, E [ it ] = j .

The Actuarial Education Company

IFE: 2016 Examinations

CT1-15: Stochastic interest rate models

With k = 2 in Equation (1.4) we obtain:

E [(1 + 2it + it2 )]

since, for each value of t:

E [ it2 ] = (E [ it ])2 + var[ it ] = j 2 + s 2

from Equations (1.5) and (1.6).

IFE: 2016 Examinations

The Actuarial Education Company

CT1-15: Stochastic interest rate models

This means that:

Using the formula gives:

The Actuarial Education Company

IFE: 2016 Examinations

CT1-15: Stochastic interest rate models

Equation (1.8) can also be deduced easily by general reasoning.

IFE: 2016 Examinations

The Actuarial Education Company

CT1-15: Stochastic interest rate models

m1 = E[ A12 ] = E[(1 + i1 ) 2 ] = 1 + 2 E[i1 ] + E[i12 ]

The Actuarial Education Company

IFE: 2016 Examinations