Approximating Prices of Bonds With Log - Normal Interest Rate
Approximating Prices of Bonds With Log - Normal Interest Rate
interest rate
Sankarshan Basu
and
Angelos Dassios 1
Abstract
This paper calculates bounds for the prices of bonds for various interest rate models resulting in a
log-normal distribution for the interest rate, in two dierent ways. In the rst method we employ
a conditioning factor as was done by Rogers and Shi, to approximate the prices of Asian options.
The second method is via a direct expansion. The lower bounds obtained are so accurate that
they are essentially the true prices. The paper also looks at values of contingent payments on the
interest rate.
Keywords : Gaussian processes, log-normal interest rates, Brownian motion, Ornstein Uhlenbeck
process, bonds, options
AMS 1991 subject classication : 90A09, 60G15
1 Introduction
Let the instantaneous rate of interest rt be given by
rt = bet +Yt
where Yt is a Gaussian process with zero mean and a variance - covariance
Cov(Yu ; Yv ) = uv ;
t is the drift of Xt and b is a scaling factor whose importance will become apparent in the next
section. In our examples it will represent the value of rt at specic time points. Our model can
be put in the framework of the work set out by Heath, Jarrow and Morton (1992); see also Baxter
1
Department Of Statistics,
The London School of Economics and Political Science,
Houghton Street,
London WC 2A 2AE,
U.K.
1
and Rennie (1996). It is an extension of the work by Black and Karasinski (1991), as well as Black,
Derman and Toy (1990). When Yt is a Brownian motion, the model is known as the Black-Derman-
Toy model. In the last two works, a binomial tree approach has been used to calculate the prices.
A similar approach is used by Hull and White (1993) who essentially use a trinomial tree. The Hull
and White model is quite similar to the Vasicek (1977) model. In the Hull and White model as in
our case, the interest rate cannot become negative; this would have had undesired consequences as
has been shown by Rogers (1995).
In the course of this work, we look into the problems of calculating bounds for the price of the
bond, in two ways - rst by the use of a suitable conditioning factor as in Rogers and Shi (1995)
and also alternatively by direct expansion. We also look at the valuing contingent payments on the
interest rate. A more general problem is the calculation of
Z1
E [f ( e(Ys +s ) ds)];
0
where f is a convex function. Thus, in particular the price of the bond (f(x) = e bx ) is given by
b
R 1
e(Ys +s ) ds )
E (e 0 (1)
and the value of the contingent payment (f(x) = [e bx c]+ ) is given by
b
R 1
e(Ys +s ) ds
E (e 0 c)+ : (2)
Rogers and Shi look at f(x) = x+ = max(x; 0). They also assume that Yt is a Brownian motion.
To calculate the price of the bond and call option, we make use of a conditioning factor Z. A lower
bound can be obtained using the fact that
E (f (X )) = E (E (f (X )jZ )) E (f (E (X jZ ))):
Several choices of Z as a conditioning factor have been tried by Rogers and Shi, asR well as us. In
fact, we have found, empirically, that the best choice of Z , in all cases is Z = q YRs ds . Z in
1
0
Var( Ys ds) 1
0
this case follows a standard normal distribution.
The direct expansion method used in here is only concerned with the calculation of the prices for the
bond. The primary reason for using the two methods is that it provides a measure of comparability.
In the valuation of contingent payments on the interest rate, the direct method cannot be used and
hence we resort to the use of simulated values to check the accuracy of the conditioning.
We are interested in calculating a lower bound (LB1 in tables 1,2 and 3) and the corresponding
upper bound (UB1 in tables 1,2 and 3). We do that by considering the following argument. There
exists some variable such that
E (f (X )) = E [f (E (X jZ ))] + E [(X E (X jZ ))f 0 (E (X jZ ))] + 21 E [(X E (X jZ ))2 f 00( )];
Z1 Z 1Z 1 Z1
E [Var( eYs +s dsjZ )] = E [E ( eYu eYv dudvjZ ) (E ( eYs dsjZ ))2 ]
0 0 0 0
Z 1Z 1
= e (ku+kv ) + (wu+wv )(ewuv 1)dudv:
1
2
2 1
2 (8)
0 0
Let us dene Z1 Z1
h(z) = E ( eYs+s dsjZ = z) = ekuz+ 1
2
wuu du: (9)
0 0
3
In the case of calculating the value of the contingent payment,
b
R 1
eYs +s ds
E [e 0 c]+
(c is the strike price of the option), the lower bound is given by
Z1
[e bh(z) c]+ p1 e dz:
z 2
2 (12)
1 2
We just present the lower bounds in this case (see Tables 4 - 6) as also the corresponding simulated
values. We can employ a combination of the argument used above in the calculation of upper
bounds for bond prices and a similar idea due to Rogers and Shi to calculate the upper bounds for
the option price. But, as the calculated lower bounds were extremely close to the simulated values,
this was not deemed necessary.
In the following examples, we present the explicit results for two special cases; the Geometric
Brownian Motion and the exponential function of an Ornstein Uhlenbeck process.
3 Examples
3.1 Geometric Brownian Motion
In this case, we have,
rt = beat+Yt :
Yt = Bt ; (13)
where Bt is a standard Brownian motion and b = r0 is the initial value of the interest rate. The
one year bond price is R B as
E (e( b e s ds) ):
1
(14)
+
0
R B ds
1
u 2 (15)
0
Thus, we have
Z1 a
Ysds) = a a + ea2 1 = V say;
2
Var( (19)
0
au a(1 u)
ku = p1 2a [ 1 ae + 1 e a
2
]: (20)
V
Once again, we have that given Z, Yu is a Gaussian process with
E (Yu jZ ) = ku Z;
and Cov(Yu ; Yv jZ ) = 2a e aju vj ku kv = wuv ; say.
2
(21)
Once we have these values, we can then easily calculate the price of the bond by substituting
in equation(10) and (11) (results shown in Table 2) and the value of the contingent payment for
various strike prices by substituting in equation (12) (results shown in Table 5).
3.2.2 Non-stationary Ornstein Uhlenbeck Process
Now, let us consider the case where the interest rate is governed by a non-stationary Ornstein
Uhlenbeck process. The interest rate model is dened in the same way as in the case of the
stationary Ornstein Uhlenbeck process. The only dierence in this case being, that Y0 is assumed
5
R
to take the value 0, so Yt = 0t e a(t s) dBs . Here, as in the case of the stationary Ornstein
Uhlenbeck process, b = r0 and be a is the long term value of the interest rate, r0 being the initial
1 2
2 2
We observe that
Z1 Z1 1 e a(1 u) 2 2a + 4e a e 2a 3 = V; say
Var( Ysds) = 2 ( a ) 2
du = 2a a2 (22)
0 0
au a(1 u) e au e a(1+u)
ku = p1 2a f 1 ae + 1 e a
2
a g: (23)
V
So, we then have that given Z, Yu is a Gaussian process with
E (YujZ ) = ku Z;
and Cov(Yu ; Yv jZ ) = 2a [eaju vj e a(u+v) ] ku kv = wuv :
2
(24)
Once we have these values, we can then easily calculate the price of the bond by substituting in
equations (10) and (11) (results shown in Table 3) and the value of the contingent payment for
various strike prices by substituting in equation (12) (results shown in Table 6).
We will use the last two inequalities as the bounds suggested are very close to each other. Here,
we have, R Y
1 bI1 + 21 b2 I2 61 b3 I3 E [e b e s s ds ] 1 bI1 + 12 b2 I2 ;
1
(25)
+
0
where, Z1 Z1
I1 = E [ eYs +s ds] = es + ss ds; 1
2
Z1 Z 1Z 1
0 0
Z1 Z 1Z 1Z 1
0 0 0
I3 = E [ e Y s +s
ds] =
3
es +v +u+ (uu+vv +ss +2uv +2us +2vs) dsdvdu:
1
2
0 0 0 0
6
In the following examples, we present the exact form for two special cases; the Geometric Brownian
Motion and the exponential function of an Ornstein Uhlenbeck process. We use the same cases as
the ones used on conditioning for comparability purposes.
5 Examples
5.1 The Simple Brownian Motion case
In this case, ss = 2 s and us = 2 (u ^ s) and also s = as. Thus,
Z1
I1 = eas+ 1
2
2 sds;
0
Z 1Z u
I2 = 2 eas+au+ s+ u dsdu; 3
2
2 1
2
2
0 0
Z 1Z uZ v
I3 = 6 eau+av+as+ u+ v+ s+ v+2 s dsdvdu:
1
2
2 1
2
2 1
2
2 2 2
0 0 0
a+ 2(a+ ) a+
= 1 b e 1 2 1 + b2 13 2 [ e 2(a + 2 )1 e 1 2 1 ];
1 2 2 1 2
(28)
2 2
a + 2 a + 2 a + 2
and the corresponding lower bound by
Z 1Z uZ v
LB2 = UB2 b3 eau+av+as+ 1
2
2 u+ 21 2 v+ 12 2 s+2 v+22 s dsdvdu
0 0 0
(
= UB2 b 3 e3(a+ ) 1 3
2
+
2
3(e(a+ ) 1) 1
2
2
(29)
(a + 25 2 )(a + 32 2 )(a + 2 ) (a + 25 2 )(a + 32 2 )(a + 12 2 )
Calculations are given in Table 1.
Thus
I1 = e a ; 1 2
2 2
Z 1Z 1
Z 1Z u
1 2 1 2
uv dvdu = 2 e( a + a e a(u v ) )
2 2
I2 = e a
2 2
+2 2 + a 2 2 dvdu
0 0 0 0
7
2
Z1 aw
(1 w)e a e
2
= 2e 2a dw , where, u-v = w; 2
0
Z 1Z 1Z 1
I3 = e a euv +us+vs3 2
2 2
0 0 0
Z 1Z uZ v au s au v av s
3 2
e a (e +e +e
2
= 6e a2 2
)
dsdvdu:
2
( ) ( ) ( )
0 0 0
Z1 e 2au
I1 = e 1
2
2 1 2a du;
0
Z 1Z 1 au av aju v j a(u+v)
e e a + e a + e e 2 2
I2 = 1
2
21
2 a dvdu 1
2
21
2
2
2
0 0
Z 1Z u au av au v e au v
e e a + e a + e
2 2 ( ) ( + )
=2
1
2
21
2 a dvdu; 1
2
21
2
2
2
0 0
Z 1Z 1Z 1
I3 = e (ss +vv +uu )+uv +us +vs dsdvdu:
1
2
0 0 0
8
Note :
Alternatively, we could make use of the following recursion relation to calculate I1 , I2 and I3 and
thereby the bonds; especially for the case of the Non-stationary Ornstein Uhlenbeck process. Since,
dYt = aYt + dBt
and dening Zt
Zt = eYs ds;
0
Setting
= ke t , we have
Zt t s) )
Hm(k; t) = E (Ztm ekYt ) = m Hm 1 (ke (t s) + 1; s)e 12 2 k2 (1 e 2 (
ds:
0
From this recursive relation, we can get I1 = H1 (0; 1), I2 = H2 (0; 1) and I3 = H3 (0; 1). Thus the
lower and upper bounds can be given as
It should be noted that the expansion technique described in this section is not guaranteed to
work. Indeed, the expansion used might diverge making it impossible to improve accuracy by
calculating more terms. For example, in the case of the Brownian motion we observed that the
model completely breaks down for 1:5. In our case, the method works because we mostly (but
not always) consider small values of .
9
The method of conditioning discussed above to approximate the prices of the bonds can be very
easily extended to the case of coupon paying bonds as well as to the situation where there is a non-
zero probability of the bond defaulting before maturity. In case of default, all payments including
coupon payments stop and a single payment is made at default. This payment is generally a small
fraction of the value of the bond.
7 References
1. M.Baxter and A.Rennie (1996); \ Financial Calculus", Cambridge University Press.
2. F.Black, E.Derman and W.Toy (1990); \ A one factor model of interest rates and its application
to treasury bond options"; Financial Analysts Journal, January/February, 33 - 39.
3. F.Black and P Karasinski (1991); \Bond and option pricing when short rates are log-normal";
Financial Analysts Journal, July/August, 52 - 59.
4. D.Heath, R.Jarrow and A.Morton (1992); \ Bond pricing and the term structure of interest
rates : A new methodology for contingent claims valuation; Econometrica, 60(1), 77 - 105.
5. J.Hull and A.White (1993); \One factor interest rate models and the valuation of interest rate
derivative securities"; Journal of Financial and Quantitative Analysis, 28(2), 235 - 255.
6. L.C.G.Rogers and Z.Shi (1995); \The value of an Asian option"; Journal of Applied Probability,
32(4), 1077 - 1088.
7. L.C.G.Rogers (1995); \Which model for term structure of interest rates should one use"; Math-
ematical Finance, ed. Davis, Due, Fleming, Shreve, Springer - Verlag, 93 - 115.
8. O.Vasicek (1977); \An equilibrium characterisation of the term structure"; Journal of Financial
Economics, 5, 177 - 188.
10
Table 1 : Comparison of bounds calculated by conditioning (LB and UB ) 1 1
with directly calculated bounds (LB and UB ) where the interest rate follows a
2 2
geometric Brownian motion. Here b = 0.07.(All prices are multiplied by 100.)
11
Table 2 : Comparison of bounds calculated by conditioning (LB and UB ) with
1 1
directly calculated bounds (LB and UB ) where the interest rate follows an
2 2
exponential function of a stationary Ornstein Uhlenbeck process. Here a = 1
and b = 0.07. (All prices are multiplied by 100.)
Note : In some cases in tables 1,2 and 3, lower bounds calculated using one approach are
slightly higher than the upper bounds calculated by the other method. This is due to small
inaccuracies in the numerical integration procedures and indicates how close they are to the
actual price.
12
Table 4.1 : Lower bounds and simulated prices for contingent payments where
the interest rate follows a geometric Brownian Motion with no drift. Here b =
0.07. (All prices are multiplied by 100.)
13
Table 4.2 : Lower bounds and simulated prices for contingent payments where
the interest rate follows a geometric Brownian Motion with a drift of -0.5. Here
b = 0.07. (All prices are multiplied by 100.)
14
Table 4.3 : Lower bounds and simulated prices for contingent payments where
the interest rate follows a geometric Brownian Motion with a drift of 0.5. Here
b = 0.07. (All prices are multiplied by 100.)
15
Table 5 : Lower bounds and simulated prices for contingent payments where the
interest rate follows an exponential function of a stationary Ornstein Uhlenbeck
process. Here a = 1 and b = 0.07. (All prices are multiplied by 100.)
16
Table 6 : Lower bounds and simulated prices for contingent payments where
the interest rate follows an exponential function of a non-stationary Ornstein
Uhlenbeck process. Here a = 1 and b = 0.07. (All prices are multiplied by 100.)
17