3 Conditional Expectation.

Martingales
1. Y and V are independent, E(V ) = 0.
X = c + aY + bV , where a, b, c are constants.

(a) Find E(X|Y ) by using the properties of conditional expectation

Solution
E(X|Y ) = E(c + aY + bV |Y ) = E(c|Y ) + E(aY |Y ) + E(bV |Y ).
Now, E(c|Y ) = c since conditional expectation of a constant is a constant,
E(aY |Y ) = aY since given Y , aY is known,
E(bV |Y ) = bE(V |Y ) = bE(V ) = 0 since conditional expectation of V
given Y is the expectation of V by independence.
Finally
E(X|Y ) = c + aY .
(b) Comment on the best predictor of X based on Y .
Solution
E(X|Y ) = c + aY is the best predictor of X given Y , in the sense that
it minimizes the mean square error E[(X h(Y ))2 ] between X and any
function of Y , h(Y ).

2. Y and W are independent.

X = W g(Y ), where g(y) is some bounded function.

(a) Find E(X|Y ) by using the properties of conditional expectation

Solution
E(X|Y ) = E(W g(Y )|Y ) = g(Y )E(W |Y ) = g(Y )E(W ).
(b) Comment on the best predictor of X based on Y .
Solution
The best predictor of X given Y is given by E(X|Y ) = g(Y )E(W ).

3. Find the following conditional expectations for Brownian motion Bt

(a) of B2 given B1
Solution Use the standard decomposition on B2 , that
B2 = B2 + B1 B1
= (B2 B1 ) + B1
where B1 is independent of B2 B1 by independence of increments.
E[B2 |B1 ] = E[(B2 B1 ) + B1 |B1 ]
= E[B2 B1 |B1 ] + E[B1 |B1 ]
= E[B2 B1 ] + B1
= 0 + B1
 (b) of B1 given B2
Solution
Using Theorem 18 (p. 28, consider example below theorem)
Cov(B1 , B2 )
E[B1 |B2 ] = E[B1 ] + (B2 E[B2 ])
V ar(B2 )
1
= 0 + (B2 0)
2
1
= B2
2
4. Bachelier model for stock price. For t T ,

St = S0 + t + Bt

where Bt is Brownian motion started at zero, and , > 0 are constants.

Find

(a) the marginal distribution of St

Solution
d d
Bt = N (0, 2 t). So St = S0 + t + N (0, 2 t) = N (S0 + t, 2 t).
(b) the joint distribution of St and ST
Solution " #
Bt
From the lecture notes, p.19, we know that is bivariate N (0, )
BT
" #
t t
with = .
t T
" # " # " #" #
St S0 + t 0 Bt
= +
ST S0 + T 0 BT
" #
d T St
Because we have a + AN (0, ) = N (a, AA ), has a bivariate
ST
" #
S0 + t
normal distribution with mean vector and covariance matrix
S0 + T
" #
2t 2t
.
2t 2T
(c) the conditional distribution of ST given St
Solution
From the definition of St and ST we have that

ST St = (T t) + (BT Bt )
ST = St + (T t) + (BT Bt )
 St depends on Bt , which is independent of BT Bt . Hence St and BT Bt
are independent.
BT Bt has N (0, T t) distribution. Therefore given St = x the dis-
tribution of ST is N (x + (T t), 2 (T t)), that is, the conditional
distribution of ST given St is N (St + (T t), 2 (T t)).
(d) the best predictor of the future price ST based on the price St
Solution
The best predictor is E[ST |St ]. Using the expression for ST in terms of
St from the previous part of the question,

E[ST |St ] = E[St + (T t) + (BT Bt )|St ]

= St + (T t) + E[(BT Bt )|St ]
= St + (T t) + E[(BT Bt )] since BT Bt is independent of St
= St + (T t) since BT Bt has zero mean.

5. Black-Scholes model for stock price. For t T ,

St = S0 eBt +t

where Bt is Brownian motion started at zero, and , > 0 are constants. Find

(a) the marginal distribution of St

Solution
Putting S0 into the exponential, we get

St = eln S0 +t+Bt

Since ln S0 + t + Bt has N (ln S0 + t, 2 t) distribution, St is LN (ln S0 +

t, 2 t).
(b) the conditional distribution of ST given St
Solution
Using the expressions for St and ST ,
ST
= e(T t)+(BT Bt )
St
ST = St e(T t)+(BT Bt )
ST = eln St +(T t)+(BT Bt )
d
Conditional on St = x, we have ST = LN (ln x + (T t), 2 (T t)).
(c) the best predictor of the future price ST based on the price St .
Solution
 The best predictor is E[ST |St ]. Using the expression for ST in terms of
St from the previous part of the question,

E[ST |St ] = E[eln St +(T t)+(BT Bt ) |St ]

= St e(T t) E[e(BT Bt ) |St ]
= St e(T t) E[e(BT Bt ) ](since BT Bt is independent of St )
1 2
= St e(T t) e 2 (T t)
see exponential moments of Normal, Theorem 3, p.9.

6. Show that the process e2Bt 2t is a martingale.

Solution
To show Mt = e2Bt 2t is a martingale, we need to show that E(Mt+s |Ft ) = Mt
and E|Mt | < .
For the martingale property,

E(e2Bt+s |Ft ) = E(e2Bt+s 2Bt e2Bt |Ft )

= e2Bt E(e2Bt+s 2Bt |Ft )
= e2Bt E(e2(Bt+s Bt ) )
1
= e2Bt e0(2)+ 2 (s)(4)
= e2Bt +2s

where the second last line comes from Theorem 3 (p.9).

Multiplying both sides by e2(t+s) ,

e2(t+s) E(e2Bt+s |Ft ) = e2t2s e2Bt +2s

E(e| 2Bt+s{z2(t+s)} |Ft ) = e2Bt 2t = Mt .
Mt+s

Hence Mt has the martingale property.

For the integrability, we have that

E|Mt | = E|e2Bt 2t |
= E(e2Bt 2t ), ex is positive for any real x
= E(e2(Bt t) )
1
= e2t+ 2 4t <

where Ee2Bt is calculated by Theorem 3 (p. 9).

7. (a) It is known that the process Bt2 h(t) is a martingale with mean zero.
Find h(t).
Solution
 For the process to have zero mean we must have
E[Bt2 h(t)] = 0
E[Bt2 ] h(t) = 0
Hence h(t) = E(Bt2 ).
If X has EX = 0 then V ar(X) = E(X 2 ) (EX)2 = E(X 2 ). Because
EBt = 0, using X = Bt , E(Bt2 ) = V ar(Bt ) = t, so that h(t) = t.
For completeness, we check the two properties that make Mt = Bt2 t a
martingale
1. E[Mt+s |Ft ] = Mt (Martingale Property)
2
E[Mt+s |Ft ] = E[Bt+s (t + s)|Ft ]
= E[((Bt+s Bt ) + Bt )2 |Ft ] (t + s)
= E[(Bt+s Bt )2 + 2(Bt+s Bt )Bt + Bt )2 |Ft ] (t + s)
= E[(Bt+s Bt )2 |Ft ] + E[2(Bt+s Bt )Bt |Ft ] + E[Bt2 |Ft ] (t + s)
= E[(Bt+s Bt )2 |Ft ] + 2Bt E[Bt+s Bt |Ft ] + E[Bt2 |Ft ] (t + s)
= E[(Bt+s Bt )2 |Ft ] + 2Bt E[Bt+s Bt |Ft ] + Bt2 (t + s)
= E[(Bt+s Bt )2 ] + 2Bt E[(Bt+s Bt )] + Bt2 (t + s)
= s + 2Bt 0 + Bt2 (t + s)
= Bt2 t
= Mt

2. E|Mt | < (Integrability)

Here we can use the triangle inequality: For any two real numbers x, y
we have that |x + y| |x| + |y|. In this case,
|Bt2 t| |Bt2 | + | t|
= Bt2 + t
So that
E|Bt2 t| E[Bt2 + t]
= t+t
= 2t <

(b) It is known that the process eBt g(t) is a martingale with mean one. Find
g(t).
Solution
For the process to have mean one, we must have
E[eB(t) g(t)] = 1
g(t)E[eB(t) ] = 1
 1 2 1
From Theorem 3 (p. 9), E[e1Bt ] = e0(1)+ 2 (1 )(t) , hence g(t) = e 2 t . For
1
completeness, we check the two properties that make Mt = eBt 2 t a
martingale.
1. E[Mt+s |Ft ] = Mt (Martingale property)
1
E[Mt+s |Ft ] = E[eBt+s 2 (t+s) |Ft ]
1
= e 2 (t+s) E[eBt+s Bt +Bt |Ft ]
1
= e 2 (t+s) eBt E[eBt+s Bt |Ft ]
1
= e 2 (t+s) eBt E[eBt+s Bt ]
1 1
= e 2 (t+s) eBt e 2 s
1
= eBt 2 t
= Mt
2. E|Mt | < (Integrability)
This holds because we know that E[Mt ] = 1 and Mt 0.
8. Show that if Mt is a martingale then aMt + b for any constants a, b is also a
martingale.
Solution
Call Xt = aMt + b. Need to show two properties
1. E[Xt+s |Ft ] = Xt (Martingale Property)
E[Xt+s |Ft ] = E[aMt+s + b|Ft ]
= aE[Mt+s |Ft ] + b
= aMt + b
where the last line follows because Mt has the martingale property.
2. E|Xt | < (Integrability)
Using the triangle inequality, we have
|aMt + b| |aMt | + |b|
= |a||Mt | + |b|
So,
E|aMt + b| E[|a||Mt | + |b|]
= |a|E[|Mt |] + |b|
<
The last inequality comes from the fact that a, b are finite and E|Mt | <
because Mt is a martingale.
Because the process Xt is integrable and has the martingale property, it is a
martingale.
9. Show that if Mt is a martingale then Mt2 is not a martingale. (Excluding
trivial cases.)
Solution
2
We must show that E[Mt+s |Ft ] 6= Mt .
2
E[Mt+s |Ft ] = E[(Mt+s Mt + Mt )2 |Ft ]
= E[(Mt+s Mt )2 |Ft ] + 2Mt E[Mt+s Mt |Ft ] + E[Mt2 |Ft ]

Another way of expressing the martingale property is as follows:

E[Mt+s |Ft ] = Mt
E[Mt+s |Ft ] Mt = 0
E[Mt+s |Ft ] E[Mt |Ft ] = 0
E[Mt+s Mt |Ft ] = 0

So we have
2
E[Mt+s |Ft ] = E[(Mt+s Mt )2 |Ft ] + 2Mt 0 + E[Mt2 |Ft ]
= E[(Mt+s Mt )2 |Ft ] + Mt2 .

Because (Mt+s Mt )2 > 0 (unless Mt = constant which is excluded), E[(Mt+s

Mt )2 |Ft ] > 0, and we have from above that E[Mt+s
2
|Ft ] > Mt2 , so that Mt is
not a martingale.