Stochastic Calculus

Steve Lalley

http://www.stat.uchicago.edu/ lalley/Courses/390/

Stochastic Calculus – p. 1/2

Tonight —
Foreign Exchange & Exchange Rate Fluctuations
Linear Stochastic Differential Equations
Cameron-Martin-Girsanov Formula

Stochastic Calculus – p. 2/2

Foreign Exchange
Stochastic Models for Exchange Rates
Interest Rates and Exchange Rates
Options on Currency Exchange

Stochastic Calculus – p. 3/2

Basic Principles
Share price processes of tradeable assets are
martingales under any risk-neutral probability
measure.
Risk-neutrality of a probability measure depends on
the numeraire.
Currencies are not tradeable assets!
Money market shares are!

Stochastic Calculus – p. 4/2

Exchange Rate Model
Let Yt denote the exchange rate at time t between US
Dollars $ and UK Pounds Sterling £, i.e., the number of
pounds that one dollar will buy. A simple model:
dYt = µYt dt + σYt dWt
where Wt is a standard Wiener process under the risk
neutral measure for £ investors, and µ and σ are
constants.

Stochastic Calculus – p. 5/2

Exchange Rate Model
Let Yt denote the exchange rate at time t between US
Dollars $ and UK Pounds Sterling £, i.e., the number of
pounds that one dollar will buy. A simple model:
dYt = µYt dt + σYt dWt
where Wt is a standard Wiener process under the risk
neutral measure for £ investors, and µ and σ are
constants.

In a more realistic model, the drift and/or diffusion

coefficients might be time-varying but deterministic:
dYt = µt Yt dt + σt Yt dWt
Stochastic Calculus – p. 5/2
Itô Processes
An Itô process is a stochastic process that satisfies a
stochastic differential equation of the form
dZt = At dt + Bt dWt

Stochastic Calculus – p. 6/2

Itô Processes
An Itô process is a stochastic process that satisfies a
stochastic differential equation of the form
dZt = At dt + Bt dWt
Here Wt is a standard Wiener process (Brownian
motion), and At , Bt are adapted process, that is,
processes such that for any time t, the current values
At , Bt are independent of the future increments of the
Wiener process.

Stochastic Calculus – p. 6/2

Itô Processes
An Itô process is a stochastic process that satisfies a
stochastic differential equation of the form
dZt = At dt + Bt dWt
Here Wt is a standard Wiener process (Brownian
motion), and At , Bt are adapted process, that is,
processes such that for any time t, the current values
At , Bt are independent of the future increments of the
Wiener process.
The local quadratic variation of the Itô process Zt is
defined by
d[Z, Z]t = Bt2 dt
Stochastic Calculus – p. 6/2
Itô’s Formula
If Zt is an Itô process, and if f (x) is a smooth function,
then f (Zt ) is also an Itô process whose Itô SDE is

0 1 00
df (Zt ) = f (Zt ) dZt + f (Zt ) d[Z, Z]t
2

Stochastic Calculus – p. 7/2

Itô’s Formula
If Zt is an Itô process, and if f (x) is a smooth function,
then f (Zt ) is also an Itô process whose Itô SDE is

0 1 00
df (Zt ) = f (Zt ) dZt + f (Zt ) d[Z, Z]t
2
Itô’s formula has a number of important generalizations.
Here is one which is sometimes useful in solving SDEs
with time-dependent coefficients: If u(x, t) is a smooth
function of two variables, then
1
du(Zt , t) = ut dt + ux dZt + uxx d[Z, Z]t
2
Stochastic Calculus – p. 7/2
Solving the SDE
The idea is to guess a solution by applying the Itô
formula to the right process. Assume that under the
probability measure P the exchange rate Yt satisfies
dYt = µYt dt + σYt dWt

Stochastic Calculus – p. 8/2

Solving the SDE
The idea is to guess a solution by applying the Itô
formula to the right process. Assume that under the
probability measure P the exchange rate Yt satisfies
dYt = µYt dt + σYt dWt
Try Itô with f (x) = log x:

d log(Yt ) = µ dt + σ dWt − (σ 2 /2) dt

Stochastic Calculus – p. 8/2

Solving the SDE
The idea is to guess a solution by applying the Itô
formula to the right process. Assume that under the
probability measure P the exchange rate Yt satisfies
dYt = µYt dt + σYt dWt
Try Itô with f (x) = log x:

d log(Yt ) = µ dt + σ dWt − (σ 2 /2) dt

Since µ and σ are constants, this is easily integrated to
give the general solution to the SDE:

Yt = Y0 exp{(µ − σ 2 /2)t + σWt }

Stochastic Calculus – p. 8/2
Time-Dependent SDEs
A similar strategy works for equations with
time-dependent coefficients, for example:
dYt = µYt dt + σYt dWt

Stochastic Calculus – p. 9/2

Time-Dependent SDEs
A similar strategy works for equations with
time-dependent coefficients, for example:
dYt = µYt dt + σYt dWt
Itô:
d log(Yt ) = µt dt + σ dWt − (σ 2 /2) dt

Stochastic Calculus – p. 9/2

Time-Dependent SDEs
A similar strategy works for equations with
time-dependent coefficients, for example:
dYt = µYt dt + σYt dWt
Itô:
d log(Yt ) = µt dt + σ dWt − (σ 2 /2) dt
and so

Yt = Y0 exp{(µ̄t − σ 2 /2)t + σWt }

where
t
1
Z
µ̄t = µs ds
t 0 Stochastic Calculus – p. 9/2
Interest Rates
Assume that for each of the two currencies US Dollar
and UK Pound Sterling there is a riskless Money Market.
Let At and Bt be the “share prices” of US Money Market
and UK Money Market, respectively, and for simplicity
assume that the time-zero share prices are both 1.

Stochastic Calculus – p. 10/2

Interest Rates
Assume that for each of the two currencies US Dollar
and UK Pound Sterling there is a riskless Money Market.
Let At and Bt be the “share prices” of US Money Market
and UK Money Market, respectively, and for simplicity
assume that the time-zero share prices are both 1.
Assume that the riskless rates of return rA , rB in the two
currencies are constant, but not necessarily equal. Then
At = exp{rA t} dollars
Bt = exp{rB t} pounds

Stochastic Calculus – p. 10/2

Exchange and Interest Rates
The asset US Money Market is riskless to a Dollar
investor, but not to a Pound Sterling investor. Evaluated
in Pounds Sterling, the share price of the US Money
Market asset is
At Yt = Y0 exp{rA t + µt − σ 2 t/2 + σWt }
where Wt is a standard Wiener Process under the risk
neutral probability measure QB for Pound investors.

Stochastic Calculus – p. 11/2

Exchange and Interest Rates
The asset US Money Market is riskless to a Dollar
investor, but not to a Pound Sterling investor. Evaluated
in Pounds Sterling, the share price of the US Money
Market asset is
At Yt = Y0 exp{rA t + µt − σ 2 t/2 + σWt }
where Wt is a standard Wiener Process under the risk
neutral probability measure QB for Pound investors.
Theorem: µ = rB − rA .

Stochastic Calculus – p. 11/2

Proof
Since US Money Market is a tradeable asset, its share
price Y0 at time t = 0 must be the expected value of its
discounted share price At Yt (in £) at time t, where
the discount rate is rB , and

Stochastic Calculus – p. 12/2

Proof
Since US Money Market is a tradeable asset, its share
price Y0 at time t = 0 must be the expected value of its
discounted share price At Yt (in £) at time t, where
the discount rate is rB , and
the expectation is taken under QB .

Stochastic Calculus – p. 12/2

Proof
Since US Money Market is a tradeable asset, its share
price Y0 at time t = 0 must be the expected value of its
discounted share price At Yt (in £) at time t, where
the discount rate is rB , and
the expectation is taken under QB .
Thus
Y0 = EQB e−rB t At Yt

Stochastic Calculus – p. 12/2

Proof
Since US Money Market is a tradeable asset, its share
price Y0 at time t = 0 must be the expected value of its
discounted share price At Yt (in £) at time t, where
the discount rate is rB , and
the expectation is taken under QB .
Thus
Y0 = EQB e−rB t At Yt
= EQB e−rB t Y0 exp{rA t + µt − σ 2 t + σWt }

Stochastic Calculus – p. 12/2

Proof
Since US Money Market is a tradeable asset, its share
price Y0 at time t = 0 must be the expected value of its
discounted share price At Yt (in £) at time t, where
the discount rate is rB , and
the expectation is taken under QB .
Thus
Y0 = EQB e−rB t At Yt
= EQB e−rB t Y0 exp{rA t + µt − σ 2 t + σWt }
= Y0 exp{(rA − rB + µ − σ 2 /2)t}EQB exp{σWt }

Stochastic Calculus – p. 12/2

Proof
Since US Money Market is a tradeable asset, its share
price Y0 at time t = 0 must be the expected value of its
discounted share price At Yt (in £) at time t, where
the discount rate is rB , and
the expectation is taken under QB .
Thus
Y0 = EQB e−rB t At Yt
= EQB e−rB t Y0 exp{rA t + µt − σ 2 t + σWt }
= Y0 exp{(rA − rB + µ − σ 2 /2)t}EQB exp{σWt }
= Y0 exp{(rA − rB + µ)t}
Stochastic Calculus – p. 12/2
Currency Options
Consider an option Call that gives the owner the right to
buy $1 for £K at time T . What is the arbitrage price at
time 0?

Stochastic Calculus – p. 13/2

Currency Options
Consider an option Call that gives the owner the right to
buy $1 for £K at time T . What is the arbitrage price at
time 0?
Solution: The option is identical to a call on e−rAT
shares of the US Money Market. To a £ investor, the US
Money Market is a risky asset with price process e−rAt Yt .
Thus, the call option may be priced using the
Black-Sholes Formula.

Stochastic Calculus – p. 13/2

Currency Options
Consider an option Call that gives the owner the right to
buy $1 for £K at time T . What is the arbitrage price at
time 0?
Solution: The option is identical to a call on e−rAT
shares of the US Money Market. To a £ investor, the US
Money Market is a risky asset with price process e−rAt Yt .
Thus, the call option may be priced using the
Black-Sholes Formula.
Exercise: Do it! While you’re at it, show how to hedge
the option.

Stochastic Calculus – p. 13/2

Risk-Neutral Measure for $

Theorem: Let QA be the risk-neutral probability

measure for the US Dollar investor, and QB the
risk-neutral measure for the UK Pound Sterling investor.
Unless σ = 0 (that is, unless the exchange rate is purely
deterministic), it must be the case that
QA 6= QB

Stochastic Calculus – p. 14/2

Risk-Neutral Measure for $

Theorem: Let QA be the risk-neutral probability

measure for the US Dollar investor, and QB the
risk-neutral measure for the UK Pound Sterling investor.
Unless σ = 0 (that is, unless the exchange rate is purely
deterministic), it must be the case that
QA 6= QB
This is a special case of a more general phenomenon:

Stochastic Calculus – p. 14/2

Numeraire Change
Suppose that a market has tradeable assets A, B with
share price processes StA and StB (evaluated in a common
numeraire C). Let QA and QB be risk-neutral measures
for numeraires A, B, respectively.

Stochastic Calculus – p. 15/2

Numeraire Change
Suppose that a market has tradeable assets A, B with
share price processes StA and StB (evaluated in a common
numeraire C). Let QA and QB be risk-neutral measures
for numeraires A, B, respectively.
Theorem: QA = QB if and only if StA /StB is a constant
random variable. Furthermore, in general, for any finite
time T ,
 B  B A
dQ ST S0
=
A
dQ FT STA S0B

Stochastic Calculus – p. 15/2

Consequence
In the foreign exchange context, the riskless assets for
the two numeraires are US Money Market and UK
Money Market, with share prices (in $)
At = exp{rA t}
Bt = exp{rB t}/Yt

Stochastic Calculus – p. 16/2

Consequence
In the foreign exchange context, the riskless assets for
the two numeraires are US Money Market and UK
Money Market, with share prices (in $)
At = exp{rA t}
Bt = exp{rB t}/Yt

Therefore, the likelihood ratio between the risk-neutral

measures for £ and $ investors is
 B  −1
dQ YT
A
= exp{(rB − rA )T }
dQ FT Y0

Stochastic Calculus – p. 16/2

Consequence
In the foreign exchange context, the riskless assets for
the two numeraires are US Money Market and UK
Money Market, with share prices (in $)
At = exp{rA t}
Bt = exp{rB t}/Yt

Therefore, the likelihood ratio between the risk-neutral

measures for £ and $ investors is
 B  −1
dQ YT
A
= exp{(rB − rA )T }
dQ FT Y0

Stochastic Calculus – p. 16/2

Likelihood Ratio Identity
Let Vti be the time-t share price of any contingent claim
in numeraire i = A, B, C. These share prices satisfy:

Stochastic Calculus – p. 17/2

Likelihood Ratio Identity
Let Vti be the time-t share price of any contingent claim
in numeraire i = A, B, C. These share prices satisfy:
A C A
Vt = Vt /St
VtB = VtC /StB

Stochastic Calculus – p. 17/2

Likelihood Ratio Identity
Let Vti be the time-t share price of any contingent claim
in numeraire i = A, B, C. These share prices satisfy:
A C A
Vt = Vt /St
VtB = VtC /StB

The time-zero share price is the discounted expected

value of the time−t share price for each of the
numeraires A, B. The discount factors are 1, so

Stochastic Calculus – p. 17/2

Likelihood Ratio Identity
Let Vti be the time-t share price of any contingent claim
in numeraire i = A, B, C. These share prices satisfy:
A C A
Vt = Vt /St
VtB = VtC /StB

The time-zero share price is the discounted expected

value of the time−t share price for each of the
numeraires A, B. The discount factors are 1, so

V0A = V0C /S0A = E A VtC /StA

B C B B C B
V0 = V0 /S0 =E Vt /St
Stochastic Calculus – p. 17/2
Likelihood Ratio Identity
It follows that for every contingent claim V with share
price VtC (in numeraire C),

S0A E A (VtC /StA ) = S0B E B (VtC /StB )

Stochastic Calculus – p. 18/2

Likelihood Ratio Identity
It follows that for every contingent claim V with share
price VtC (in numeraire C),

S0A E A (VtC /StA ) = S0B E B (VtC /StB )

Apply this to the contingent claim with payoff VTC STB at

time T to obtain the following identity, valid for all
nonnegative random variables VTC measurable FT :
 B A
B C A C ST S0
E VT = E VT
STA S0B

Stochastic Calculus – p. 18/2

Likelihood Ratio Identity
It follows that for every contingent claim V with share
price VtC (in numeraire C),

S0A E A (VtC /StA ) = S0B E B (VtC /StB )

Apply this to the contingent claim with payoff VTC STB at

time T to obtain the following identity, valid for all
nonnegative random variables VTC measurable FT :
 B A
B C A C ST S0
E VT = E VT
STA S0B
This is the defining property of a likelihood ratio.
Stochastic Calculus – p. 18/2
Exponential Martingales
Let Wt be a standard Wiener process, wth Brownian
filtration Ft , and let θt be a bounded, adapted process.
Define
Z t Z t 
Zt = exp θs dWs − θs2 ds/2
0 0

Stochastic Calculus – p. 19/2

Exponential Martingales
Let Wt be a standard Wiener process, wth Brownian
filtration Ft , and let θt be a bounded, adapted process.
Define
Z t Z t 
Zt = exp θs dWs − θs2 ds/2
0 0

Fact: Zt is a positive martingale.

Stochastic Calculus – p. 19/2

Exponential Martingales
Let Wt be a standard Wiener process, wth Brownian
filtration Ft , and let θt be a bounded, adapted process.
Define
Z t Z t 
Zt = exp θs dWs − θs2 ds/2
0 0

Fact: Zt is a positive martingale. Proof: Itô!

Stochastic Calculus – p. 19/2

Exponential Martingales
Let Wt be a standard Wiener process, wth Brownian
filtration Ft , and let θt be a bounded, adapted process.
Define
Z t Z t 
Zt = exp θs dWs − θs2 ds/2
0 0

Fact: Zt is a positive martingale. Proof: Itô!

dZt = Zt θt dWt − Zt θt2 dt/2+Zt θt2 dt/2

= Zt θt dWt =⇒
Z t
Zt = Z 0 + Zs θs dWs
0
Stochastic Calculus – p. 19/2
Girsanov’s Theorem
Because Zt is a positive martingale under P with initial
value Z0 = 1, for every fixed time T the random variable
ZT is a likelihood ratio: that is,
Q(F ) := EP (IF ZT )
defines a new probability measure on the σ−algebra FT
of events F that are observable by time T .

Stochastic Calculus – p. 20/2

Girsanov’s Theorem
Because Zt is a positive martingale under P with initial
value Z0 = 1, for every fixed time T the random variable
ZT is a likelihood ratio: that is,
Q(F ) := EP (IF ZT )
defines a new probability measure on the σ−algebra FT
of events F that are observable by time T .
Theorem: Under the measure Q, the process
Rt
{Wt − 0 θs ds}0≤t≤T is a standard Wiener process.

Stochastic Calculus – p. 20/2

Exchange Rates
Consider again the $ and £ currencies. Assume that each
has a riskless Money Market, and that the rates of return
rA , rB are constant. Assume that the exchange rate Yt
obeys
dYt = (rB − rA )Yt dt + σYt dWt
where Wt is a standard Wiener process under the
risk-neutral probability QB for £ investors. Thus,

Yt = Y0 exp{(rB − rA − σ 2 /2)t + σWt }.

Stochastic Calculus – p. 21/2

Exchange Rates
Since
A
   
dQ YT
= exp{−(rB − rA )T }
dQB FT Y0
= exp{σWT − σ 2 T /2}

Stochastic Calculus – p. 22/2

Exchange Rates
Since
A
   
dQ YT
= exp{−(rB − rA )T }
dQB FT Y0
= exp{σWT − σ 2 T /2}

Girsanov implies that under QA the process Wt is a

Wiener process with drift σ. Thus, to the $ investor, it
appears that the exchange rate obeys

dYt = (rB − rA − σ 2 )Yt dt + σYt dW̃t

where W̃t is a standard Wiener process under QA .

Stochastic Calculus – p. 22/2
Stochastic Calculus – p. 23/2
Proof of Girsanov 1
The statement that X is a standard Wiener process is an
assertion that the increments of X are independent
Gaussian random variables with the correct variances.
Let’s show that under Q, the distribution of WT − ΘT is
RT
gaussian with var T (where ΘT = 0 θs ds).

Stochastic Calculus – p. 24/2

Proof of Girsanov 1
The statement that X is a standard Wiener process is an
assertion that the increments of X are independent
Gaussian random variables with the correct variances.
Let’s show that under Q, the distribution of WT − ΘT is
RT
gaussian with var T (where ΘT = 0 θs ds). For this, it
suffices to show that for any real λ,

EQ exp{λ(WT − ΘT )} = exp{λ2 T /2}

Stochastic Calculus – p. 24/2

Proof of Girsanov 1
The statement that X is a standard Wiener process is an
assertion that the increments of X are independent
Gaussian random variables with the correct variances.
Let’s show that under Q, the distribution of WT − ΘT is
RT
gaussian with var T (where ΘT = 0 θs ds). For this, it
suffices to show that for any real λ,

EQ exp{λ(WT − ΘT )} = exp{λ2 T /2}

To evaluate the expectation, change measure:
EQ exp{λ(WT − ΘT )} = EP exp{λ(WT − ΘT )}ZT

Stochastic Calculus – p. 24/2

Proof of Girsanov 2
Objective: Show that EP HT = 1, where

Ht = exp{λ(Wt − Θt ) − λ2 t/2}Zt

Stochastic Calculus – p. 25/2

Proof of Girsanov 2
Objective: Show that EP HT = 1, where

Ht = exp{λ(Wt − Θt ) − λ2 t/2}Zt
Z t Z t
= exp{ (θs + λ) dWs + (λθs − θs2 /2 − λ2 /2) ds}
0 0

Stochastic Calculus – p. 25/2

Proof of Girsanov 2
Objective: Show that EP HT = 1, where

Ht = exp{λ(Wt − Θt ) − λ2 t/2}Zt
Z t Z t
= exp{ (θs + λ) dWs + (λθs − θs2 /2 − λ2 /2) ds}
0 0
Z t Z t
= exp{ (θs + λ) dWs − (θs + λ)2 ds/2}
0 0

Stochastic Calculus – p. 25/2

Proof of Girsanov 2
Objective: Show that EP HT = 1, where

Ht = exp{λ(Wt − Θt ) − λ2 t/2}Zt
Z t Z t
= exp{ (θs + λ) dWs + (λθs − θs2 /2 − λ2 /2) ds}
0 0
Z t Z t
= exp{ (θs + λ) dWs − (θs + λ)2 ds/2}
0 0

Thus, Ht is an exponential martingale under P , and so

its expectation is constant over time.

Stochastic Calculus – p. 25/2

Proof of Girsanov 2
Objective: Show that EP HT = 1, where

Ht = exp{λ(Wt − Θt ) − λ2 t/2}Zt
Z t Z t
= exp{ (θs + λ) dWs + (λθs − θs2 /2 − λ2 /2) ds}
0 0
Z t Z t
= exp{ (θs + λ) dWs − (θs + λ)2 ds/2}
0 0

Thus, Ht is an exponential martingale under P , and so

its expectation is constant over time. A similar
calculation establishes the independence of the
increments.
Stochastic Calculus – p. 25/2
Scratch

Stochastic Calculus – p. 26/2

Scratch

Stochastic Calculus – p. 26/2

Scratch

Stochastic Calculus – p. 26/2

Scratch

Stochastic Calculus – p. 27/2

Scratch

Stochastic Calculus – p. 27/2

Scratch

Stochastic Calculus – p. 27/2