CT1 Chapter 15
Stochastic Interest Rate Models
1
Mayur Ankolekar
15th & 22nd March, 2019
FYBSc, DSActEd
Stochastic nature of interest rates
Interest rates in the future could move randomly
This means the rate itself could have a variance
Unlike what was assumed earlier in the course that interest rates
are static
Variation in interest rates are allowed for based on statistical
theory e.g. interest rates could be i.i.d.r.v. or have a serial correlation
Introducing the idea implicit in
stochastic interest rate models
A. Accumulated value at mean interest rate
≠
B. Mean accumulated value
1. Identify the random variable in A and B
2. Write the formula for the accumulated value for A and B.
3. Assume interest rates over 10 years would be 5% pa (p =
0.2), 7% pa (p = 0.5), 9% pa (p = 0.3), find A and B.
4. Calculate the variance of A.
An and Sn as r.v.
4
Sn = (1+i1) (1+i2) …. (1+in) Note:
An here is the
An = (1+i1) (1+i2) …. (1+in) equivalent of s due n
+(1+i2) …. (1+in)
Sn is just the
+… accumulation of 1
over the period
+ (1+in-1)(1+in)
+(1+in)
Q 15.3: Find the probability that Sn will take value between 1.02 x 1.04n-1, if
yield of 2% pa, 4% pa and 6% pa are equally likely. (Read the note above!)
Answer:
n . 1/3 . (1/3)n-1
Deriving mean and variance of Sn, the acc value
Parameters of interest rate over the investment term E(it) = j, V(it) = s2
5
If the kth moment of Sn is (Sn)k =
then E(Sn)k = E [ ]
=
so, E(Sn) = = = (1+j)n
where E(it) = j
E(Sn2) =
Deriving mean and variance of Sn, the acc value … contd.
6
E(Sn2) =
As E (it)2 = [E(it)]2 + V(it), and
E(it)] = j,V(it) = s2
we get E(Sn2) = [1 + 2j + j2 + s2 ]n
and
V(Sn) = E(Sn2) - {E(Sn)}2
= [(1+j)2 + s2]n - (1+j)2n
Practise problems to compute E(Sn) and V(Sn) under various distributions of interest rates.
Expectation of An or “s due n”
7
An = (1+i1) (1+i2) …. (1+in) An and An-1 follow a
+(1+i2) …. (1+in) recursive relationship
such that:
+…
+ (1+in-1)(1+in) An = (1 + in) (1 + An-1)
+(1+in)
If E(An ) is denoted by μn,
then μn = (1 + in) (1 + μn-1) … n ≥ 2
= (1 + j) (1 + μn-1)
= (1 + j) + (1 + j) (1 + μn-1)
and expands to (1+j) + (1+j)2 + (1+j)3 + … + (1+j)n
i.e., “s due n”
Variance of An or “s due n”
8
E(A12) = E(1 + i1)2 = 1 + 2E(i1) + E(i12)
Define mn = E(An2), so m1 = 1 + 2E(i1) + E(i12)
As E (it)2 = [E(it)]2 + V(it), and E(it)] = j, V(it) = s2
we get m1 = E(A12) = 1 + 2j + j2 + s2
Also, mn = An2 = (1 + 2in + in2) (1 + An-1 + An-12)
V (An) = E (An2) – [E (An)]2
= mn - μn2
Lognormal distribution of interest rate
9
Log normal distribution
A special case where the exact analysis of the distribution function of
Sn is particularly simple.
The log-normal variable of interest rate i.e. LN (1 + it) can take a
positive value and has the multiplicative property
if X1 ~ LN (μ1 , σ12) and X2 ~ LN (μ2 , σ22)
then
X1 X2 ~ LN (μ1 + μ2 , σ12 + σ 22)
Lognormal distribution of an accumulation of 1
10
As the kth moment of Sn is (Sn)k =
If (1 + it) ~ LN (μ , σ2) ,
(1 + i1) ~ LN (μ1 , σ12), (1 + i2) ~ LN (μ2, σ22) and so on
then Sn = (1 + i1) (1 + i2) … (1 + in) ~ LN (nμ , nσ2) .. using independence
property
Note: parameters μ and σ need to be found from solving simultaneously
E (1 + it) = exp (μ + σ2/2) and V (1 + it) = exp (2μ + σ2). [exp (σ2) – 1]
Lognormal distribution of an accumulation of 1 .. Contd.
11
i.e.,
log (Sn) ~ N (nμ , nσ2)
Sn ~ LN (nμ , nσ2)
alternatively,
P (logSn ≤ log s) = Φ [(log s – nμ)/ σ√n]
and,
(log s – nμ)/ σ√n ~ N (0,1)
Lognormal distribution of the PV of 1
12
Vn = (1 + i1)-1 (1 + i2)-1 …. (1 + in)-1
ln Vn = - ln (1 + i1) – ln (1 + i2) - …. - ln (1 + in)
If (1 + it) ~ LN (μ , σ2) ,
then Vn ~ LN (-nμ , nσ2),
i.e.,
log (Vn) ~ N (-nμ , nσ2)
alternatively,
{log v – (-nμ)}/ σ√n ~ N (0,1)
and,
P (Vn ≤ v) = Φ [(log v – (-nμ))/ σ√n]
Applications of Lognormal distribution for stochastic
interest rates
13
If Y ~ LN (μ , σ2), then
E(Y) = exp(μ+σ2/2)
V(Y) = exp(2μ+σ2). exp(σ2 – 1)
If E(1 + it) and V(1 + it) are given and (1+ it) ~LN (μ , σ2) , then the
parameters μ and σ2 can be estimated.
The following example can be solved (Q2, exam style):
A lump sum of $14,000 will be invested at time 0 for 4 years at a constant annual rate
of interest i.e. (1 + i) has a log-normal distribution with mean 1.05 and variance
0.007. What is the probability that the investment will accumulate to more than
$20,000 in 4 years’ time?
Hint: Estimate μ and σ2; S4 ~ LN (4 μ , 4σ2), and the r.v. = 20000/14000.
Find 1 - P(S4 ≤ 10/7)
CT1, IFoA, Sep 2018, Q6
14
In a particular investment fund, it is the rate of return in the tth year. Let Sn be the
accumulation of ₤1 invested over a period of n years. Assume the mean and standard
deviation of it is 0.08 and 0.07 respectively, and that 1 + it is log-normally distributed.
(i) Determine the distribution of S10 [5 marks]
An investor is considering investing ₤6,000 in the fund for 10 years.
(ii) Determine the amount of the accumulated value after 10 yr such that there is a 97.5%
probability of the investor actually achieving an amount greater than this. [3 marks]
Approach:
(i) Calculate the parameters μ and σ in Sn ~ LN (nμ , nσ2) using E (1 + it) = 1 + E(it) =
exp (μ + σ2/2) and V (1 + it) = V (it) = exp (2μ + σ2). [exp (σ2) – 1]
(ii) Calculate P (6,000 S10 > x) = 0.975, i.e. 1 - P (6,000 S10 ≤ x) = 0.975
P(ln S10 ≤ ln x/6,000) = 0.025
Φ [ln x/6,000 – 10μ)/ σ√n] = 0.025
(ln x/6,000 – 10μ)/ σ√n = - 1.96
