A Series Expansion for the Bivariate Normal Integral

KMV Corporation
COPYRIGHT 1996, 1998, KMV CORPORATION, SAN FRANCISCO, CALIFORNIA, USA. All rights reserved. Document Number: 999-0000-043. Revision 2.0.1. KMV Corporation or Kealhofer, McQuown, Vasicek Development, L.P. retain all trade secret, copyright and other proprietary rights in this document. Except for individual use, this document should not be copied without the express written permission of the owner. KMV Corporation and the KMV Logo are registered trademarks of KMV Corporation. Portfolio Manager, Credit Monitor, Global Correlation Model, GCorr, Private Firm Model, EDF Calculator, EDFCalc, Expected Default Frequency and EDF are trademarks of KMV Corporation. All other trademarks are the property of their respective owners.

Published by: KMV Corporation 1620 Montgomery Street, Suite 140 San Francisco, CA 94111 U.S.A. Phone: +1 415-296-9669 FAX: +1 415-296-9458 email: support@kmv.com website: http: // www.kmv.com
Page ii

Authors: Oldrich Alfons Vasicek

Release Date: 1996 Revision: 01-April-1998

A Series Expansion for the Bivariate Normal Integral

Abstract
An infinite series expansion is given for the bivariate normal cumulative distribution function. This expansion converges as a series of powers of 1 2 , where is the correlation coefficient, and thus represents a good alternative to the tetrachoric series when is large in absolute value.

Page iii

Release Date: 1996 Revision: 01-April-1998

A Series Expansion for the Bivariate Normal Integral

Introduction
The cumulative normal distribution function Nx = with n u = 2

af

z af
x

n u du

af a f

exp u 2

appears frequently in modern finance: Essentially all explicit equations of options pricing, starting with the Black/Scholes formula, involve the function in one form or another. Increasingly, however, there is also a need for the bivariate cumulative normal distribution function N 2 x, y , =

zz
x y

n 2 u , v , dudv

(1)

where the bivariate normal density is given by n 2 u , v , = 2

g a f d1 i
1

exp

FG H

u 2 2 uv + v 2 1 2

IJ K

(2)

This need arises in at least the following areas: 1. Pricing exotic options. Option with payout depending on the prices of two lognormally distributed assets, or two normally distributed factors, involve the bivariate normal distribution function in the pricing formula. Examples include the so-called rainbow options (such as calls on maximum or minimum of two assets), extendible options, spread and cross-country swaps, etc. Correlation of derivatives. While the instantaneous correlation of two derivatives is the same as the correlation of the underlying assets, calculation of the correlation over noninfinitesimal intervals often requires the bivariate normal function. Loan loss correlation. If a loan default occurs when the borrower's assets fall below a certain point, the covariance of defaults on two loans is given by a bivariate normal formula. This covariance is needed when evaluating the variance of loan portfolio losses.

2.

3.

A standard procedure for calculating the bivariate normal distribution function is the tetrachoric series,

Page 1

Release Date: 1996 Revision: 01-April-1998

KMV Corporation N 2 x, y , = N x N y + n x n y where He k x =

g a f b g a f b g ak +1 1f! He axfHe byg

k=0 k k k2

k+1

(3)

af

i !a k 2if! a1f 2 i x k 2 i
k!
i i=0

are the Hermite polynomials. For a comprehensive review of the literature, see Gupta (1963). The tetrachoric series (3) converges only slightly faster than a geometric series with quotient , and it is therefore not very practical to use when is large in absolute value. In this note, we give an alternative series that converges approximately as a geometric series with quotient 1 2 .

The Expansion
The starting point of this note is the formula d N 2 x, y , = n 2 x, y , d proven in the appendix. Because of the identity N 2 x , y , = N 2 x ,0 ,

(4)

F GG H

x y
x 2 2 xy + y 2

signx + N 2 y ,0 ,

I JJ K

F GG H

y x
x 2 2 xy + y 2

signy

I L0 if xy > 0 O JJ MN if xy < 0PQ K

for xy 0 , we can limit ourselves to calculation of N 2 x ,0 , . Suppose first that > 0 . Then by integrating equation (4) with y = 0 from to 1 we get
N 2 x ,0 , = min N x , Q

b af g

(5)

where Q = n 2 x ,0 , r dr = 2

z
1

a f z d1 r i
1 1

exp

FG H

x2 dr 1 r2

IJ K

Page 2

Release Date: 1996 Revision: 01-April-1998

A Series Expansion for the Bivariate Normal Integral To evaluate the integral, substitute
r = 1 s

to obtain Q = 2

a f z a 1 sf
1 1 2 0

exp

FG H

x2 ds s

IJ K

(6)

Using the expansion

a 1 sf
we get

k=0

a2 k f ! 2 ak !f
2

2 k k

Q = 2 Because

a f

1 2 0

a k !f2 2 2 k s k expG H
k =0

a2 k f!

x2 ds s

IJ K

(7)

a2 k f! 2 ak !f
2

2 k k

exp

FG H

x2 sk 1 2 s

IJ K

(8)

for k > 0 , 0 s 1 2 , the series in equation (7) converges uniformly in the interval 0 ,1 2 and can be integrated term by term. It can be easily established that

z
t 0

s k exp

FG H

x2 k! 2 k+1 k 1 2 k + 1 x ds = s 2k + 1 !

IJ K

fa f L F x I a2if! a1f 2 . MexpG J MN H t K i! a

2 k i i =0

2 i 1 i +

a f NFGH xt IJK OPP Q

for k 0 . Substitution into (7) then gives Q = Ak

k=0

(9)

where

Page 3

Release Date: 1996 Revision: 01-April-1998

KMV Corporation Ak = 1 1 2 k+1 k 1 2 k x k! 2k + 1

1 2 2

F x I OP (10) x d1 i a2 f NGGH 1 JJK P Q Equations (5), (9) and (10) give an infinite series expansion for N b x ,0 , g, > 0 . When < 0 ,
k i i 2 i 1 2 i+ i =0 2 2

a f LM F x IJ a2if! a1f 2 2 f expG a MN H 1 K i!

integration of equation (4) from -1 to yields

N 2 x ,0 , = max N x ,0 + Q
with Q still given by (9) and (10).

b af

(11)

A convenient procedure for computing the terms in the expansion (9) is using the recursive relationships Ak = Bk = with B0 = 2 A0 x I a f d1 i expFGH 1 J K F x I = a2 f x NG GH 1 JJK + B
1 2 2 2 2 0

a2k 1f d1 iB 2 ka2 k + 1f
2 2

2k 1 x 2 Ak 1 + Bk 2k 2k + 1

k1

To determine the speed of convergence of (9), integrate the first half of inequality (8) from 0 to 1 2 . This results in the bound 0 < Ak 2

a f

1 1 2 k+

k+

for k > 0 , and therefore the series (9) converges approximately as 1 2 tetrachoric series for N 2 x ,0 ,

k . As the

N 2 x ,0 , = N x + 2

a f a f naxf k1! 2k1+ 1 a1f 2

k k=0

He 2 k x 2 k + 1

af

(12)

Page 4

Release Date: 1996 Revision: 01-April-1998

A Series Expansion for the Bivariate Normal Integral converges approximately as 2 k k , a reasonable method for calculating N 2 x ,0 , is to use the tetrachoric series (12) when and the expressions (5) and (11) with the series (9) when
2

2 > .
The error in the calculation of N 2 x ,0 , resulting from using m terms in the expansion (9) is bound in absolute value by

k=m

Ak

< 2

a f

1 1 1 2 2 m+

m+

Numerical Results
A comparison of the convergence of the tetrachoric series (12) and the alternative calculation (5) or (11) with the series (9) in calculating N 2 x ,0 , for high values of the correlation coefficient is given in the following tables:

Table 1 Partial sums of the tetrachoric and alternative series x = 1, =.95

Number of terms 1 2 3 5 10 20 30 50 100 200 300 Exact Tetrachoric .171033 .171033 .167298 .161764 .157961 .158466 .158660 .158632 .158631 .158631 .158631 Alternative .158632 .158631 .158631 .158631 .158631 .158631 .158631 .158631 .158631 .158631 .158631

x = 1, =.99
Tetrachoric Alternative .174894 .174894 .170304 .162651 .156068 .158068 .159374 .158599 .158711 .158657 .158654 .158655 .158655 .158655 .158655 .158655 .158655 .158655 .158655 .158655 .158655 .158655

.158631

.158655

Page 5

Release Date: 1996 Revision: 01-April-1998

KMV Corporation Table 2 Number of terms necessary for precision 10-4

=
x=0 x=1 x=2

.8 8 7 6 4 3 1

.9 16 14 11 3 1 1

.95 30 22 18 2 1 1

.99 121 75 42 1 1 1

Tetrachoric series

Alternative series x=0 x=1 x=2

Page 6

Release Date: 1996 Revision: 01-April-1998

A Series Expansion for the Bivariate Normal Integral

Appendix
We prove equation (4) by stating a slightly more general result. Let n p , S = 2 Np ,S =

b g a f b g

p 2

exp ' S 1

n p , S d

a f i

be the p-variate normal density function and cumulative distribution function, respectively, where = x1 , x2 ,! , x p is a p 1 vector and S = 11 , 12 ,! , pp is a p p symmetric positive definite matrix. We prove the following lemma: Lemma. Let p 2 . Then for i j , 2 Np ,S = Np ,S ij xi x j

b g i

b g = n e a f , Sa f jN e a f Sa
2 1 11 p2 2

21

fSa11f a1f , Sa22 f Sa21fSa11fSa12 f j

1 1

Here a 1f = xi , x j is the 2 1 vector of xi , x j , a2 f is the p 2 1 vector of the remaining

components of , and Sa 11f , Sa 12 f , Sa 21f , Sa 22 f is the 2 2 , 2 p 2 , p 2 2 , and p 2 p 2 decomposition of S into the i -th and j -th row and column and the remaining rows and columns. Proof. We have S 1 S n p , S = tr S 1 S n p ,S . + S 1 ij ij ij Define = v11 , v12 ,! , v pp by = S 1 and put = y 1 , y 2 ,! , y p =Vx. Since for i j S = ij + ji ij where ij is the matrix having unity for the ij -th element and zeros elsewhere, we get on substitution n p , S = vij + y i y j n p , S ij

gb

g b

F b g G H

F GH

I JK

I JK b g
i

b g d

i b g
Page 7 Release Date: 1996 Revision: 01-April-1998

KMV Corporation On the other hand, n p ,S = y j n p ,S x j

b g

b g i b g

2 n p , S = vij + y i y j n p , S xi x j

b g d

and therefore 2 n p ,S = n p ,S ij xi x j

b g

b g

Integrating with respect to and exchanging the order of integration and differentiation yields the first equality of the lemma. The second equality follows from the factorization
1 1 n p , S = n 2 a 1f , Sa 11f n p 2 a 2 f Sa 21f Sa11 f xa1f , Sa22 f Sa21fSa11fSa12 f

b g

j e

Page 8

Release Date: 1996 Revision: 01-April-1998

A Series Expansion for the Bivariate Normal Integral

References
Gupta, Shanti S., 1963, Probability integrals of multivariate normal and multivariate t, Annals of Mathematical Statistics 34, 792-828. Johnson, N.L. and S. Kotz, 1972, Distributions in Statistics, Continuous Multivariate Distributions, Wiley, New York. Vasicek, Oldrich A., 1997, The loan loss distribution, working paper, KMV Corporation.

Page 9

Release Date: 1996 Revision: 01-April-1998