The Normal and Lognormal Distributions

John Norstad
j-norstad@northwestern.edu http://www.norstad.org

February 2, 1999 Updated: November 3, 2011

Abstract The basic properties of the normal and lognormal distributions, with full proofs. We assume familiarity with elementary probability theory and with college-level calculus.

DEFINITIONS AND SUMMARY OF THE PROPOSITIONS

Denitions and Summary of the Propositions

Proposition 1:

2 2 1 e(x) /2 dx = 1 2 2 2 1 e(x) /2 dx = 2 2 2 1 e(x) /2 dx = 2 + 2 2

Proposition 2:

Proposition 3:

x2

Denition 1 The normal distribution N [, 2 ] is the probability distribution dened by the following density function:
2 2 1 e(x) /2 2

Note that Proposition 1 veries that this is a valid density function (its integral from to is 1). Denition 2 The lognormal distribution LN [, 2 ] is the distribution of eX where X is N [, 2 ]. Proposition 4: If X is N [, 2 ] then E(X) = and Var(X) = 2 . Proposition 5: If Y is LN [, 2 ] then E(Y ) = e+ 2 and 2 2 Var(Y ) = e2+ (e 1). Proposition 6: If X is N [, 2 ] then aX + b is N [a + b, a2 2 ].
2 2 Proposition 7: If X is N [1 , 1 ], Y is N [2 , 2 ], and X and Y are indepen2 2 dent, then X + Y is N [1 + 2 , 1 + 2 ]. n
1 2

Corollary 1: N [n, n 2 ].

If Xi are independent N [, ] for i = 1 . . . n then

Xi is

Corollary 2: LN [n, n 2 ].

If Yi are independent LN [, ] for i = 1 . . . n then

Yi is

Proposition 8: The probability density function of LN [, 2 ] is:

PROOFS OF THE PROPOSITIONS

Proofs of the Propositions

Proposition 1

2 2 1 e(x) /2 dx = 1 2

Proof: First assume that = 0 and = 1. Let:

a=

2 1 ex /2 dx 2

Then:

a2

1 2

2 2 1 1 ex /2 ey /2 dxdy 2 2

e(x

+y 2 )/2

dxdy

Apply the polar transformation x = r cos , y = r sin , dxdy = rdrd: a2 = = = 1 2 1 2

er er
0
2

/2

rdrd d

/2

2 1 [0 (1)]d 2 0 1 2 = 2 = 1

a > 0, and we just showed that a2 = 1, so we must have a = 1. For the general case, apply the transformation y =

x dx , dy = :

2 2 1 e(x) /2 dx 2

2 1 ey /2 dy 2 1 y2 /2 e dy 2

PROOFS OF THE PROPOSITIONS

Proposition 2

2 2 1 e(x) /2 dx = 2

Proof: First assume that = 0 and = 1.

= = = = =

n 2 1 xex /2 dx lim n 2 n n 2 1 lim ex /2 n 2 n 2 2 1 lim [(en /2 ) (en /2 )] 2 n 1 lim 0 2 n 0

For the general case, apply the transformation y =

x dx , dy = :

2 2 1 e(x) /2 dx 2

= = = =

( + y)

2 1 ey /2 dy 2

2 2 1 1 ey /2 dy + y ey /2 dy 2 2 1 + 0 (by Proposition 1)

Proposition 3

x2

2 2 1 e(x) /2 dx = 2 + 2 2

Proof: First assume that = 0 and = 1. Integrate by parts using f = x, f = 1, g = ex

/2

, g = xex

/2

x2 ex
n

n /2

dx

=
n

f g dx
n

= =

f (n)g(n) f (n)g(n) (nen

f gdx
n n

/2

) (nen

/2

)
n

ex

/2

dx

PROOFS OF THE PROPOSITIONS

= =

2nen
n

ex
2

/2

dx

ex
n

/2

dx 2nen

/2

Then:
2 1 x2 ex /2 dx 2

= = = =

n 2 1 x2 ex /2 dx lim 2 n n n 2 2 1 ex /2 dx 2nen /2 lim n 2 n 2 1 x2 /2 e dx lim 2nen /2 n 2

1 lim 2nen
n

/2

(by Proposition 1)

All that remains is to show that the last limit above is 0. We do this using LHpitals rule: o lim 2nen
2

/2

= lim

2 2n = lim =0 2 n nen /2 en2 /2 dx x , dy = :

For the general case, apply the transformation y =

x2

2 2 1 e(x) /2 dx 2

( + y)2

= =

2 1 ey /2 dy + 2 2 1 2 y ey /2 dy + 2 1 y2 /2 2 dy y2 e 2 2 1 + 2 0 + 2 1

(by Propositions 1 and 2) 2 + 2

PROOFS OF THE PROPOSITIONS

Proposition 4 If X is N [, 2 ] then E(X) = and Var(X) = 2 . Proof: E(X) Var(X) = = = = =

2
1 2

Proposition 5 If Y is LN [, 2 ] then E(Y ) = e+ 2 and 2 2 Var(Y ) = e2+ (e 1). Proof: Y = eX where X is N [, 2 ]. First assume that = 0:

E(Y )

= E(eX ) =

ex

2 2 1 ex /2 dx 2

= e2
1

2x 2 x2 1 e 22 dx 2 (x 2 )2 + 4 1 2 2 e dx 2 2 2 2 1 e(x ) /2 dx 2

= e2 1 = e E(Y )
2
1 2 2

(by Proposition 1)

= E(e2X ) =

e2x

2 2 1 ex /2 dx 2

= e

2 2
2

4x 2 x2 1 e 22 dx 2 (x2 2 )2 +4 4 1 2 2 e dx 2 2 2 2 1 e(x2 ) /2 dx 2

= e2 1 = e Var(Y ) =
2 2

(by Proposition 1) (by Proposition 1 in reference [1])

E(Y 2 ) E(Y )2
2 1 2 2

= e2 (e

)2 = e2 e = e (e 1)

PROOFS OF THE PROPOSITIONS

For the general case, apply the transformation y = x , dy = dx:

E(Y )

= E(eX ) =

ex

2 2 1 e(x) /2 dx 2

2 2 1 ey /2 dy 2 2 2 1 = e ey ey /2 dy 2 = (by the same reasoning as for the case = 0 above)

e+y

e e 2
1

2 2

= e+ 2 E(Y 2 )

= E(e2X ) =

e2x

2 2 1 e(x) /2 dx 2

2 2 1 ey /2 dy 2 2 2 1 2 = e ey /2 dy e2y 2 = (by the same reasoning as for the case = 0 above)

2(y+)

e2 e2 Var(Y ) =

2 2

= e2+2

E(Y 2 ) E(Y )2
2 1 + 2 2 2 2

(by Proposition 1 in reference [1]) )2

= e2+2 (e
2 2

= e2+2 e2+ = e2+ (e 1)

Proposition 6 If X is N [, 2 ] then aX + b is N [a + b, a2 2 ]. Proof: Prob(aX + b < k) = = = Prob(X < (k b)/a)

(kb)/a

2 yb 2 1 1 e( a ) /2 dy a 2 2 2 2 1 e(y(a+b)) /2a dy 2(a)

The last term above is the cumulative density function for N [a + b, a2 2 ], so we have our result.

PROOFS OF THE PROPOSITIONS

2 2 Proposition 7 If X is N [1 , 1 ], Y is N [2 , 2 ], and X and Y are indepen2 2 dent, then X + Y is N [1 + 2 , 1 + 2 ].

Proof: First assume that X is N [0, 1] and Y is N [0, 2 ]. Then: Prob(X + Y < k) =
x+u<k
2 2 2 1 1 ex /2 eu /2 dxdu 2 2

(apply the transformation u = y) 1 2 1 2 e(x

+y 2 )/2

dxdy (1)

e(x
x+y<k

+y 2 )/2

dxdy

At this point we temporarily make the assumption that k 0. Figure 1 shows the area over which we are integrating. It is the half of the plane below and to the left of the line x + y = k.

Figure 1: Area of Integration Note the following relationships: x = r cos y = r sin r cos + r sin = x + y = k r= k cos + sin

tan = k/(k/) = = arctan

PROOFS OF THE PROPOSITIONS

Well use the following technique to prove the result. First well convert the double integral above to polar coordinates. Then well rotate the result by so that the graph above becomes the one shown in 2 below. Then well show that the resulting integral is the same as the one for Prob(Z < k) where Z is N [0, 1 + 2 ].

Figure 2: Area of Integration Rotated Note that r cos = x = k, and r = k/cos. Apply the polar transformation x = r cos , y = r sin , dxdy = rdrd to equation (1): Prob(X + Y < k) = = = 1 2 1 2 1 2 1 2 e(x
x+y<k
2

+y 2 )/2

dxdy

rer
x+y<k 3/2+ /2+ 3/2+ /2+ 0 0

/2

drd
/2

rer

drd + rer
2

k cos + sin

/2

drd

(2)

Note that we have made use of the assumption that k 0 at this point to split the area over which we are integrating into two regions. In the rst region, varies from /2 + to 3/2 + , and the vector at the origin with angle does not intersect the line x + y, so r varies from 0 to . In the second region,

PROOFS OF THE PROPOSITIONS

varies from /2 + to /2 + , and the vector does intersect the line, so r varies from 0 to k/(cos + sin ). We want to apply the transformation = to rotate. We rst must calculate what happens to the upper limit of integration in the last integral above under this transformation. k cos + sin = = k cos( + ) + sin( + ) k cos( + arctan ) + sin( + arctan )

(3)

We now apply some trigonometric identities: cos( + arctan ) sin( + arctan ) cos(arctan ) sin(arctan ) cos( + arctan ) sin( + arctan ) = = = = = = cos cos(arctan ) sin sin(arctan ) sin cos(arctan ) + cos sin(arctan ) 1 1 + 2 1 + 2 cos sin 1 + 2 sin + 2 cos 1 + 2

(4) (5)

Adding equations (4) and (5) gives: cos( + arctan ) + sin( + arctan ) = = = cos + 2 cos 1 + 2 cos (1 + 2 ) 1 + 2 cos 1 + 2 (6)

We can now do our rotation under the transformation = . Equations (2), (3) and (6) give: Prob(X + Y < k) = 1 2 1 2 = 1 2 1 2
3/2 /2 /2
cos

rer
0
k

/2

drd +
2

1+ 2

rer

/2

drd

/2 3/2

er
/2 /2

/2

d +
0
k cos 1+ 2

er
/2

/2

PROOFS OF THE PROPOSITIONS

10

= =

1 2 1

[1] d +
/2

1 2

e 2 cos2 (1+2 ) (1) d d (7)

1 2

/2

e
/2

k2 2 cos2 (1+ 2 )

Now we turn our attention to evaluating Prob(Z < k) where Z is N [0, 1 + 2 ]. Let W be another random variable which is also N [0, 1+ 2 ]. We use a sequence of steps similar to the one above, only without the rotation: Prob(Z < k) = = = = = Prob(Z < k, < W < ) 2 2 2 2 1 1 ez /2(1+ ) ew /2(1+ ) dzdw 2 2 2 1 + 2 1 + z<k 2 2 2 1 e(z +w )/2(1+ ) dzdw 2(1 + 2 ) z<k 2 2 1 rer /2(1+ ) drd 2(1 + 2 ) z<k 1 2(1 + 2 ) 1 2(1 + 2 ) = 1 2(1 + 2 ) 1 2(1 + 2 ) = 1 1 2
/2 /2 3/2 /2 /2 /2 3/2 0 0 k/cos

rer

/2(1+ 2 )

drd + drd

rer

/2(1+ 2 )

(1 + 2 )er
/2 /2

/2(1+ 2 )

d +
0 k/ cos

(1 + 2 )er
/2
k2

/2(1+ 2 )

d
0

e 2 cos2 (1+2 ) d

(8)

Equations (7) and (8) are the same, so at this point we have completed our proof that Prob(X + Y < k) = Prob(Z < k) when k 0. Now suppose that k < 0. Propositon 6 implies that for any normal random variable X with mean 0, X is also normally distributed with mean 0 and the same variance as X. Let A = X, B = Y , and W = Z. Then A is N [0, 1], B is N [0, 2 ], and W is N [0, 1 + 2 ]. So we have: Prob(X + Y < k) = Prob((X + Y ) > k) = Prob(A + B > k) = = = = = 1 Prob(A + B < k) 1 Prob(W < k) Prob(W > k) Prob(Z > k) Prob(Z < k) (because k > 0)

PROOFS OF THE PROPOSITIONS

11

At this point we have shown that Prob(X + Y < k) = Prob(Z < k) for all k. Thus the random variables X + Y and Z have the same cumulative density function. Z is N [0, 1 + 2 ], so X + Y is also N [0, 1 + 2 ]. This completes our proof for the case that X is N [0, 1] and Y is N [0, 2 ].
2 2 For the general case where X is N [1 , 1 ] and Y is N [2 , 2 ], let A = (X1 )/1 2 2 and B = (Y 2 )/1 . By Property 6, A is N [0, 1] and B is N [0, 2 /1 ]. Thus:

Prob(X + Y < k)

= = =

Prob(A + B < (k 1 2 )/1 )

2 1 + (apply the transformation y = 1 x + 1 + 2 )

1
2 2 2 /1

ex

2 2 /2(1+2 /1 )

dx

1 2
2 1

2 2

e(y(1 +2 ))

2 2 /2(1 +2 )

dy

2 2 This is the cumulative density function for N [1 + 2 , 1 + 2 ], so we have our full result. n

Corollary 1 If Xi are independent N [, 2 ] for i = 1 . . . n then

Xi is

N [n, n 2 ]. Proof: This corollary follows immediately from Proposition 7.

Corollary 2 If Yi are independent LN [, 2 ] for i = 1 . . . n then

Yi is

LN [n, n 2 ]. Proof: This corollary follows immediately from Denition 2 and Proposition 7.

PROOFS OF THE PROPOSITIONS

12

Proposition 8 The probability density function of LN [, 2 ] is:

Proof: Suppose X is LN [, 2 ]. Thev X = eY where Y is N [, 2 ]. Then: Prob(X < k) = = = = Prob(eY < k) Prob(Y < log(k))
2 2 1 e(y) /2 dy 2 1 (apply the transformation x = ey , y = log(x), dy = x dx)

log(k)

2 2 1 e(log(x)) /2 dx x 2

REFERENCES

13

References
[1] John Norstad. Probability review. http://www.norstad.org/nance, Sep 2002.