Académique Documents
Professionnel Documents
Culture Documents
May 2009
Contents
1 Preface 4
6 Yield to maturity 14
6.1 In continuous time . . . . . . . . . . . . . . . . . . . . . . . . . . 14
6.2 In discrete time . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
2
8 Derivatives on interest rates 15
8.1 Forward risk adjusted Probability Measure . . . . . . . . . . . . 15
8.2 European bond options . . . . . . . . . . . . . . . . . . . . . . . 16
8.2.1 Valuation of an European put option . . . . . . . . . . . . 16
8.2.2 Valuation of an European call option . . . . . . . . . . . . 16
8.3 Swaps . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
8.4 Swaptions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
8.5 Caps and Floors . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
8.5.1 Cap . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
8.5.2 Floor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
8.6 Binomial based interest rate models . . . . . . . . . . . . . . . . 18
8.6.1 One period model . . . . . . . . . . . . . . . . . . . . . . 18
10 Calibration 21
11 References 25
3
1 Preface
I have gound this project really rewarding and challenging. I would like to
thank my advisor Norbert Frank Proske for giving great advices and motivation
during the process of studying and writing on the topic of interest rate markets
and interest rate modelling. I also would like to thanks my sister and her
boyfriend for reading through my papers several times, and helping me out
with agrumentation and translation discrepancies etc.
I wanted by writing this project, getting a better knowledge about the inter-
est markets. Lately the topic have been of great importance in connectin with
the credit crisis and later financial crisis we have seen lately and also takes part
today. I have throughout the project taken into consideration the no-arbitrage
condition on the markets studied.
In the beginning of the project I have given a summary of the interest market,
and the pricing procedures of the derivatives connected to the interest rate
markets. I have mostly focused on discrete time in the beginning section, and
in the later sections I am also touching continious modelling. In the end I have
done a calibration of the HJM framework to fit canadian yield to maturities in
the market to the HJM model.
4
2 Interest rates & Bond Markets
Interest rates define the rate of cost the borrower agrees to pay the lender over
a certain time interval. Eg. when you want to buy a house, you may need to
borrow part of the money needed to pay the transaction. You (the borrower)
will then make an agreement with the bank that you will pay back the loan in
20 years, with an eg fixed interest rate of 5%. Let us say you borrow 1 million
NOK to pay the house, you will then have to pay 106 ∗ (1 + 0.05) = 50k in
interest, plus the principle of the loan.
In a different case, a large company may need to borrow short term cash,
and therefore issues a corporate bond. The bond then receives an amount of
money, and agrees to pay back the principal(face value) of the bond loan. eg. in
an annually zero coupon bond at interest rate of 5%, and face value of 1 million
NOK, the bond issuer receives 106 ∗ exp(−.05) = 9510 229. When at maturity,
the bond issuer pays back the face value of 1million NOK. In general the level
of credit risk involved in the loan, sets the level of interest rate, eg. a private
household has a bigger risk of default than a large company, and therefore the
interest rate is higher for the private borrower than for the company.
There are several reasons why we need management of interest rate mod-
elling. In for example calculating the risk neutral price of a stock process, we
discount the future value by the interest rate process. Different banks and
financial institutions may also buy a contract that allows them to freeze the
interest rate on the loan for example, by buying a payer swap, that gives them
a fixed interest rate k, and where they receive the floating rate r(s). This allows
the banks to plan the future expences of the loans. By doing this, they must
however pay a premium in order to get this contract.
Discrete time modelling of the bond market and interest rates, is often easier
to calculate, and also explicitly calculate the risk neutral probability measures,
which we need in order to have a market without arbitrage opportunities. This
kind of modelling gives us however not the whole picture of the world, as we
get closer to by continuous time modelling. In this paper I will take into con-
sideration both discrete time modelling and continuous time modelling. I will
use some different notations in discrete and continuous time.
5
i(m) = nominal interest rate compounded m times annualy.
i(m) m
(1 + ) =1+i (1)
m
1. Federal funds rate: Rate at which private financial institutions lend bal-
ances at the the federal reserve(USA)
2. Discount rate: Rate at which banks borrow short term funds from central
banks
6
2.2 Bond market
The bond market is a financial market for debt securities like bonds. There
are several different kind of bonds, such as corporate bonds, government bonds
(ex. T-bonds US. Treasury bonds), Mortgage backed securities, such as CDO’s
collateralized debt obligations, which is the ones that went sour in the housing
market, and led to huge losses. Most of the traded bonds are done OTC (over the
counter), but there also exists a bond market at the New York Stock Exchange,
where mostly corporate bonds are traded. The bond markets were estimated to
be 45 $ trillion in 20061 .
Zero coupon bonds are bonds that do not pay interest or coupons during the life
of the contract. The buyer instead buys the bond discounted over the years, and
receives at maturity the face value, or principal. Even though the zero coupon
bond doesn’t pay out interests, investors in the US, will still have to pay income
tax on the ”‘hidden”’ interest rate. Example of zero coupon bond is US treasury
bills, or municipal zero coupon bond. Municipal bonds have also the advantage
of no federal taxes, and also in many US. states no local tax, especially if you
live in the state of which you hold the bond.2
A coupon bearing bond is similar to a zero coupon bond, but this contract
also has a payout of coupon amount. This coupon amount is often paid semi-
annually, and must be taken into consideration when pricing the bond.
The term structure or yield curve is the relationship between interest rates and
time to maturity.
7
2.2.4 Types of Yield curve
1. Normal
2. Flat
3. Inverted
4. Humped
8
2.2.6 Flat Yield Curve
here we have the same interest rate for short term interest rate, as for long term.
Market expectations of decreased interest rates. This kind of yield curve often
appears when in a transition between normal yield curve and the inverted yield
curve.
Higher interest rate for short-term loans than for longer terms. The market ex-
pects strong decrease in interest rates over time. Possible reasons why investors
would still buy the longer maturity bonds would be that the expectations are
that the rates will be even lower in the future, and that they prefer to lock in a
long term interest rate.
Yield curves may be constructed from prices of bonds, the yield to maturity,
and the forward spot interest rate. We assume that the market has a perfect
competition, and that the information is available to all investors at the same
time, and that all previous information is known to all the investors. From these
assumptions, we introduce the term risk neutral probability measure.
1. The sarket expects decreasing rates. Increased demand for long-term rates
shapes the inverted yield curve.
2. The market expects increasing rates leads to a higher demand for short-
term rates, and this results a normal yield curve
fault: The theory doesn’t explain why normal yield curves are more com-
mon than inverted yield curves
9
2.3.2 Liquidity preference theory
1. The yield curve is shaped due to the market expectations, and the liquidity
premium. Investors expect a premium for tigh up capital over longer
periods.
In discrete time, we need a probability measure that makes sure that there are
no arbitrage opportunities. If there exists an arbitrage opportunity, one can
take a position in the market, and lock in a risk free profit.
In the single period market, we have 2 states in our model, the time zero and
the time 1. In this market the risk neutral probability measure is quite straight
forward calculating. Sn is the stochastic price process of eg. a stock. Bt is the
bank account or discount process, defined as Bt = 1+r1
t
, B0 = 1.rt ≡ BtB−B t−1
t−1
is
the interest rate from the time span [t−1, t]. The measure Q needs the following
criteria to hold:
K
X
Qk = 1 (2)
k=1
Qk > 0∀k
Sn+1
Eq [ ] = Sn (3)
Bt
10
Sn
Eq [∆ ]=0
Bt
For a single period market, we must solve a set of linear equations in where
the stock process can lead to two different states at time t=1 in order to de-
termine the risk neutral probability measure. In this example we use the stock
process Sn (t) which takes the values S(0) = 5, S(1, ω1 ) = 20 40
3 S(1, ω2 ) = 9 , and
10
r = 1/9, which gives the value 9 to Bt . By using equation (3), we get equation
(4), and by condition (2) we get (5).
q1 + q2 = 1 (5)
1
This gives the unique solution q1 = q2 = 2 , which satisfies the criterias
above.
We also say that when the market has a unique risk neutral probability
measure ⇒ there are no arbitrage opportunities ⇒ there exist no dominant
trading strategies ⇒ the market is complete.
3.1.2 Filtration
1. Ø ∈ Ω
2. A ∈ F ⇒ Ac ∈ F
S
3. A1 , A2 ,. . . ∈ F ⇒ i≥1 Ai ∈ F
11
3.1.3 Martingale
E[Xt+1 |Ft ] = Xt
eg.3
Eq [Sn∗ (t + s)|Ft ] = Sn∗ (t) (6)
Equation 2 and 3, are subject to the conditional expectation w.r.t the filtra-
tion Ft
A zero coupon bond, or discounted bond, is a loan contract, where the bond
issuer agrees to pay at maturity the principal or face value to the bond holder
at certain interest rate.
eg. the bond issuer agrees to pay back in one year 1$. There are no payments
during the time period, but at maturity, the bond holder receives 1$.
3 See [1]example 3.3 continued p.93
12
P(T,T) is the principal amount, or the value of a Zero coupon bond at
maturity.
Bs ∗ Ztτ
Zsτ = Eq [ |Fs ], 0 ≤ s ≤ t ≤ τ (8)
Bt
with the q as the risk neutral probability measure, and the stochastic bond price
process Ztτ , which is Ft -adapted.
A coupon bearing bond is a bond that pays out a coupon amount m times a
year in addition to the principle at maturity, eg. semi-annual coupon. The
coupon payment can be a fixed price, or a percentage of the par value. The
word coupon comes from earlier days, when there was attached a coupon to the
bonds, which would be collected at certain dates.
N
X
B(t) = Zttn ∗ Cn (9)
n=1
13
5.2 Continuous time pricing
m
X
B(t) = ci ∗ e−(Ti −t)∗R(t,Ti ) + X ∗ e−(Tm −t)∗R(t,Tm ) (10)
i=1
Where m is the m-times paying coupon per year, and X is the principal.
6 Yield to maturity
The definition of the yield to maturity is Ytτ is to be the one period overnight
interest rate of the sum of money equal to the current price of the zero coupon
bond, Ztτ will be Zττ = 1. Ytτ , is the stochastic interest rate process from time
t to T, t ≤ T . Denoted in continuous time by R(t,T), and in discrete time by
Ytτ . Note that the Ytt+1 = rt+1 , the current spot interest rate. The yield to
maturity can be directly found by using bond prices.
By solving equation (7) w.r.t the yield to maturity R(t,T) we get the following
equation:
1
R(t, T ) = − ∗ log(P (t, T )) (11)
(T − t)
1
Ytτ = [Ztτ ] −1 (12)
t−τ
Forward interest rates are the interest rates for a future period of time, im-
plied by the rates existing in the market today. For example 100$ invested
at a fixed rate of 3% the first year, continuous compounded = 100*e0.03∗1 =
103.5. Assuming that the two year fixed interest is 4% the investment return
is 100*e0.04∗2)=108.3287 . The forward rate, calculated with respect to the initial
overnight rate of 3% is then; 100*e0.03 ∗ ef (1,2) = 108.3287 4
4 See[6]
14
7.1 in discrete time
Zst
f (s, t) = − 1, s ≤ t (13)
Zst+s
P (t, s)
f (t, T ) = ,t ≤ s (14)
P (t, s + 1)
Eq [Mt ] = 1
we now introduce the new probability measure Pt (ω) = Mt (ω) ∗ Q(ω) This
new variable is a probability measure, since Pt > 0 ∀ω, and that since Eq [Mt ] =
1, Pt (ω) = 1.
We can then by using equation(16) calculate the forward risk adjusted prob-
ability measure Pτ as follows.
Q(ω)
Pτ = Mτ (ω) ∗ Q(ω) =
Z0τ ∗ Bτ (ω)
15
8.2 European bond options
An European put option give the buyer the right, to sell the underlying bond
at maturity T at the strike price k. If the bond value is above the strike price
k, the option should not be exercised.
The initial value of an European put option, must be Ft adapted, so that there
are no arbitrage opportunities. Therefore the V0 value of the put must be:
X
Eq [ |Ft ]
Bt
The same also applies to the call option. The initial value of an European
call option, must be Ft adapted, so that there is no arbitrage opportunities.
Therefore the V0 value of the call must be:
X
Eq[ |Ft ]
Bt
8.3 Swaps
Swaps are agreements between two parties where the first pays a floating rate to
the second, while the second pays a fixed rate to the first, with both payments
16
being based on a common principal amount.5 . A payer swap is a contract where
the payer pays a fixed interest rate k, and receives the floating rate r(t). A
receiver swap is a contract where the receiver receives the fixed interest k, and
pays floating rate r(t). The value of the either the payer or receiver is equal to
the negative value of the other.
payerswap = −receiverswap
A payer swap is the value from the perspective of the party who pays the
fixed rate, and receives the floating rate. Time t-value:
s
X Bt
πt = EQ [ (ru − K)|Ft ] (19)
u=τ
Bu
A receiver swap is the value from the perspective of the party who receives the
fixed rate, and pays the floating rate. Time t-value:
s
X Bt
πt = EQ [ (K − ru )|Ft ] (20)
u=τ
Bu
The forward swap rate κ is the value of the fixed rate K which makes the
time t value of the forward swap zero.
Ztτ −1 − Zts
κ = κ(t, τ, s) ≡ (21)
Ztτ + . . . + Zts
8.4 Swaptions
A reveicer swaption, is an option on the swap contract to pay floating price and
receive fixed price k. Time t-value:
s
Bt X Bt
EQ [ (EQ [ (k − ru )|Fτ −1 ])+ |Ft ] (23)
Bτ −1 u=τ
B u
5 See [1]
17
8.5 Caps and Floors
8.5.1 Cap
A caplet is a European call option on the spot interest rate r(t) with the following
payout function:
A cap is a collection of caplets over time, with the same exercise price.
8.5.2 Floor
A floorlet is a European put option on the spot interest rate r(t) with a payout
function given by:
We have an asset with a deterministic zero value S0, and we have two possible
outcomes at time 1 (S1). We may represent this by a lattice diagram as follows.6
In order to guarantee that this one period binomial model risk neutral, the
following condition must hold.
6 See [7]
18
Figure 2: Lattice of the stock process S(t)
0 < d < 1+r < u, where r is the one step short rate, u is upward movement,
and d is downward movement
n S0 S ∗ (1, ω1 ) S ∗ (1, ω2
u d
1 1 1+r 1+r
Sn
Sn∗ =
Bt
Eq [∆Sn∗ ] = 0
u d
I : q1 ( − 1) + q2 ( − 1) = 0
(1 + r) (1 + r)
II : q1 + q2 = 1
which yields:
1+r−d
q1 =
u−d
0<1+r−d<u−d
d<1+r <u
19
9 More on continuous time modelling of interest
rates
In a world where everything is certain, the discrete time models are excelent
for modelling interest rate derivatives. However in the real world, the outcome
of the future is not certain, so we must therefore model the interest process by
stochastic processes.
Xt = r(t) − b
eat ∗ Xt
Z t Z t Z t Z t
as as as 1
= (r(0) − b) + ae ∗ Xs ds + e ∗ (−a ∗ Xs )ds + e ∗ σdBs + 0ds
0 0 0 2 0
Z t
≡ (r(0) − b) + σ eas dBs
0
Rt
⇒ r(t) = r(0)e−at + b(1 − e−at ) + σe−at 0 eas dBs
This makes r(t) normally distributed, with mean and var:
E[r(t)] = r(0)e−at + b(a − e−at
Z t
−at
V ar[t] = V ar[σe eas dBs ]
0
20
limt→∞ E[r(t)] = b Which is the mean reverting level, with speed a of the
vasicek model.
In contrast to the Vasicek model and the CIR model,the HJM framework will
model the evolution of the interest rate of the whole yield curve forward in time.
Here the assumption is tht the bondproces P (t, u) take the form:
Ru
(− f (t,s)ds)
P (t, u) = e t (28)
Where f (t, s) is instantaneous forward rate, which can be interpreted as the
interest rate between[s, s + ds] as seen from time t < s. In the HJM framework,
the forward rates are modelled by the SDE(Stochastic differential equation):
Z t Z t
f (t, u) = f (0, u) + α(v, u)dv + σ(v, u)dBv (29)
0 0
⇐⇒
df (t, u) = α(t, u)dt + σ(t, u)dBt , o ≤ t ≤ u (30)
Under the risk neutral probability measure, the dynamics of the forward rate
f (t, u) only depends on σ(t, u) under P ∗ .
10 Calibration
Calibration is the procedure of fitting the desired model to the actual market
prices. I will in this paper use the HJM framework for calibrating the model to
the market.
21
Time-to-maturity t1 Yield% t1 + δ Yield% t2 Yield % t2 + δ Yield %
1 month 4,05% 3,88% 3,89% 3,93%
3 month 4,08% 3,88% 3,98% 3,97%
6 month 4,27% 4,2% 4,21% 4,3%
1 year 4,32% 4,3% 4,23% 4,44%
Because of equation (30), only the volatility function σ(t, u) and the initial
forward curve f (0, u) are required for derivative pricing. We therefore need to
estimate f (0, u) and σ(t, u). From equation (29) we derive
d
f (t, u) = − logP (t, u), 0 ≤ u ≤ T (31)
du
f (tj + δ, tj + τk ) − f (tj , tj + τk )
Dj,k := √ (32)
δ ∗ min{M, f (tj , tj + τk )}
Where:
1
δ= 52 = 1 week (tenor)
k
τk = 5 years , k = 1,. . .,4
t1 = 05.09.07, t1 + δ = 12.09.07
t2 = 0.3.10.07, t2 + δ = 10.10.07
g(ξk+1 ) − g(ξk )
g(x) = g(ξk ) + ∗ (x − ξk )
ξk+1 − ξk
d
f (t, T ) = − dt logP (t, T )
8 See [5]
22
1 1
ξ1 = t1 + 12 , ξ = t1 + 4
1 1
g(ξ1 ) = P (t1 , ξ1 ) = e− 12 ∗0,0405 , g(ξ2 ) = P (t1 , ξ2 ) = e(− 4 ∗0,0408)
f (t, t1 + τ1 ) = f (t, t1 + 15 ) = − dt
d
logP (t, T )|T =t1 + 51 =
d g(ξ2 ) − g(ξ1 )
− log(g(ξ1 ) + ∗ (T − ξ1 )
dt ξ2 − ξ1
=
− g(ξξ22)−g(ξ
−ξ1
1)
g(ξ2 )−g(ξ1 )
|T =t1 + 51 = 0, 0407046
g(ξ1 ) + ξ2 −ξ1 ∗ (T − ξ1 )
by the same arguments we get:
f (tj , tj + k5 )
j k=1 k=2 k=3 k=4
1 0,0407046 0,0446 0,0434 0,0438
2 0,0403 0,0444 0,0422 0,04426
f (tj + δ, tj + k5 )
j k=1 k=2 k=3 k=4
1 0,0388 0,0452 0,0437 0,044
2 0,0399 0,0463 0,0454 0,0458
−0.33741312 0.1062942 0.05314708 0.03543139
Dj,k = .
−0.07157422 0.3399775 0.57259375 0.57259375
23
We know that
σ̃(τ1 )
σ̃(τ2 )
∗ (σ̃(τ1 ), . . . , σ̃(τ4 ))
C =
σ̃(τ3 )
σ̃(τ4 )
√
C=≈ λ1 ∗ e1 by PCA10
10 See [4]
24
11 References
1. S.R Pliska,Introduction to Mathematical Finance,Discrete time models,
Blackwell Publishing, US,UK,Australia, 12.ed, 2005
2. S.E. Shreve,Stochastic Calculus for Finance I(The binomial Asset Pricing
Model, Springer, New York, 2004
25
12 Appendix A(R programming of calibration)
delta=1/52, M=0.08
D=matrix(c(1:8),byrow=T,nrow=2)
D[1,] = (f1d-f1)/(sqrt(delta)*min(m,f1))
D[2,] = (f2d-f2)/(sqrt(delta)*min(M,f2))
C=t(D)%*%D
eigen(C)
$values
$vectors
plot(c(1:4),sigmatau,type =’l’)
26