Financial Applications in

Continuous Time

”Successful investing is anticipating the anticipations of others.”

John Maynard Keynes (1883-1946).

 Investments

A financial investment is an allocation of money with contracts in order to

make its value increase over time. ”Sacrifice of current cash for future cash”.
We name some possible investment opportunities

• Risk-free bank account with given interest

• Interest rate related investments
- Treasury Bills
- Bonds (Government Bonds and Corporate Bonds)
• Participation and profit oriented investments
- Stocks (or shares): ownership in a part of a corporation.
- Indices and funds: collections of stocks (and other investments)
• Derivatives on underlying products
- Options, Futures, Swaps and all kind of derivatives.

Stochastic Systems, 2010 2

 Interest: Continuous Compounding

At time t we can invest an amount St at a given, fixed interest r .

St+1 −St
The return is defined as r = St ⇒ St+1 = St(1 + r)

The value after k periods is: St+k = St(1 + r)k .

Dividing the sample period into n equal intervals

r nk
St+k = St(1 + )
n
for continuous compounding we regard the case for n → ∞
r nk rk
St+k = St lim (1 + ) = St e .
n→∞ n

Stochastic Systems, 2010 3

 The dynamics of asset prices

Given the equations for continuous compounding, we can derive the dynamics for a risk
free bank account B(t) with a deterministic rate of return, rB (t)

dB(t) = B(t)rB (t)dt. (1)

Similarly, we can create the dynamics for a risky asset, by assuming that it’s wealth
follows the same kind of interest dynamics however with a stochastic ”interest rate”
with constant drift µ and diffusion σ .

S(t + 1) − S(t)
rS (t) = µ + σξ(t) = (2)
S(t)

in differential notation we write the geometric Brownian motion

dS(t) = S(t)µdt + S(t)σdW (t). (3)

Stochastic Systems, 2010 4

 The dynamics of asset prices

More generally, there are n risky assets in the market. The asset price processes
Si(t), i = 1, . . . , n, of the risk-bearing investments satisfy the stochastic
differential equation (Si(0) > 0)):

dSi(t)
= µi(S(t), x(t), t) dt + σi(S(t), x(t), t) dWS (t)
Si(t)

• WS (t) ∈ Rm×1: Standard Brownian motion (m-dimensional)

• µi ∈ R is the relative change in price of Si (i.e. drift).
• µi(S(t), x(t), t) is no more a constant, but a function of time, the asset
value S(t) and of further external factors x(t).
• Diffusion σi(S(t), x(t), t) ∈ R1×m is a function of time, asset value S(t)
and of further external factors x(t).
• Σ = σiσiT : the variance per unit time

Stochastic Systems, 2010 5

 Utility Functions

Problem: Which stochastic investment opportunity should I take? Utility func-

tions provide a ranking to judge uncertain situations.
Risk averse investor: Always prefers more wealth to less wealth. When given
two opportunities with the same return a risk averse investor always chooses
the one with less risk.

• The utility function is an increasing continuous function: U 0 > 0

• The utility function must be concave: U 00 < 0

Stochastic Systems, 2010 6

 Utility Functions

1. Exponential function (a > 0)

−ax
U (x) = −e

2. Logarithmic function
U (x) = ln(x)
3. Power functions (b < 1 and b 6= 0)

1 b
U (x) = (x)
b
a
4. Quadratic functions (x ≤ 2b )

2
U (x) = ax − bx

Stochastic Systems, 2010 7

 Utility Functions

Utility U
6
U (x2 )
1 (x + x )) r
U(2 1r 2
r
1 U (x ) + 1 U (x )
U (x1 ) 2 1 2 2
r

- x
x1 1 x2
2 (x1 + x2 )

Abbildung 1: Risk averse utility functions

Stochastic Systems, 2010 8

 Self-financing portfolios and wealth dynamics

n
X
X(t) = Ni(t)Si(t)
i=0

X(t) : wealth of the investor at time t

Ni(t) : number of shares of the i-th security in the portfolio
Si(t) : share price of the i-th security. S0(t) risk free with return r(t).
C(t) > 0 : consumption per unit time in the interval [t, t + ∆t]
n
X n
X n
X
dX(t) = Ni(t)dSi(t) + dNi(t)Si(t) + dNi(t)dSi(t)
i=0 i=0 i=0
n
X
= Ni(t)dSi(t) − C(t)dt
i=0

Stochastic Systems, 2010 9

 Self-financing portfolios and wealth dynamics

Replace dSi(t) = Si(t)(µi(t)dt + σi(t)dW (t)),

Ni (t)Si (t) Pn
and define ui(t) = X(t) , i=0 ui (t) = 1.

n
X n
X
dX(t) = ui(t)X(t)µi(t)dt + ui(t)X(t)σi(t)dW (t) − C(t)dt
i=0 i=0

Pn
As u0(t) = 1 − i=1 ui(t) denotes the fraction of wealth invested into the risk free
asset S0(t) with return r(t) we can write
n
X ¡ ¢ ¡ ¢
dX(t) = ui(t) µi(t) − r(t) X(t)dt + rX(t) − C(t) dt
i=1
n
X
+ ui(t)X(t)σi(t)dW (t)
i=1

Stochastic Systems, 2010 10

 Self-financing portfolios and wealth dynamics

In vector notation
· ³ ´ ¸
T ¡ ¢
dX(t) = X(t) u(t) µ(t) − er(t) + r(t) − C(t) dt

T
+X(t)u (t)σ(t)dW (t)

with
u(t) = (u1(t), . . . , un(t)T ∈ Rn
e = (1, . . . , 1)T ∈ Rn
µ(t) = (µ1(t), . . . , µn(t)T ∈ Rn
σ(t) ∈ Rn×m
dW (t) ∈ Rm m-dimensional Brownian motion
X(t), r(t) and C(t) are scalar.

Stochastic Systems, 2010 11

 Portfolio Models and Stochastic Optimal Control

With knowledge of utility functions and wealth dynamics we can write the
stochastic optimal control problem in which we regard:

• a market with n + 1 assets

• one deterministic (risk free) asset: dB = r(t)B(t)dt : (B(0) = b0 > 0)
• n risky assets: dSi(t) = µi(t)Si(t) dt + σi(t)T Si(t) dW (t)

• wealth-equation:
dX(t) = [X(t)(uT (µ − e r) + r) − C] dt + X(t)uT σdW (t)

• cost functional: ( )
RT1 −ρt γ
J(u) = maxu(t) E 0 γ
e C(t) dt + γ1 πX(T )γ , (π < 1, γ < 1)

Stochastic Systems, 2010 12

 Portfolio Models and Stochastic Optimal Control

We can state the optimization problem:

(Z )
T
1 −ρt γ 1 γ
max E e C(t) dt + πX(T )
u 0 γ γ
s.t.
T T
dX(t) = [X(t)(u (µ − e r) + r) − C] dt + X(t)u σdW (t)

cost-to-go function:
(Z )
T
1 −ρt γ 1 γ
J(X(t), t) = max E e C(t) dt + πX(T ) .
C,u(t) 0 γ γ

The stochastic optimal control problem can be solved with HJB-theory.

Stochastic Systems, 2010 13

 Reminder: HJB equation

Stochastic optimal control problem:

nZ T o
max E L(s, x, u)ds + K(x(T ), T )
u 0
s.t.
dx = f (t, x, u)dt + g(t, x, u)dW

cost-to-go function J(t, x)

nZ T o
J(t, x) = max E L(s, x, u)ds + K(x(T ), T ) .
u t

HJB-equation (with terminal condition J(T, x(T )) = K(T, x(T ))):

h 1 n oi
T T
Jt (t, x) + max L(t, x, u) + f (t, x, u)Jx (t, x) + tr Jxx (t, x)g(t, x, u)g (t, x, u) = 0
u 2

Stochastic Systems, 2010 14

 Derivatives

• The value of a derivative security (derivative) depends explicitly on other

variables, e.g. on the value of another financial security (underlying security )
• forward contracts (forwards): over-the-counter
• futures: e.g. CBOT
• options: e.g. CBOT
• most import concept: arbitrage

Stochastic Systems, 2010 15

 Forward contracts

• A contract to purchase or sell an asset at a specific price and at a specific

time in the future.
• spot price: price of an asset for immediate delivery
• buyer → long position, seller → short position
• specified price in the contract: delivery price K .
• specified price in the contract if the contract would be established at that
time t: forward price Ft
• initial price of a forward contract is zero → no money is transferred at the
beginning.
• example: Currency Risk

Stochastic Systems, 2010 16

 Forward contracts

• payoff of the long position: ST − K (maximum possible loss: −K )

• payoff of the short position: K − ST (maximum possible loss: −∞)

Payoff
6

Long position

0 - ST
K

Short position

Stochastic Systems, 2010 17

 Delivery price of a forward contract

• current price of a security: S0

• risk-free interest rate is r
delivery price in a forward contract with time to delivery T :
rT
K = e S0

or
1
K= S0,
(1 + r)T
respectively.

Stochastic Systems, 2010 18

 Value of a forward contract

• forward price is Ft
• delivery price of the forward is K
• risk-free interest rate is r
• delivery at time T
current value of the forward contract (long position)
−r(T −t)
ft = e (Ft − K)

or
1
ft = T −t
(Ft − K),
(1 + r)
respectively.

Stochastic Systems, 2010 19

 Futures

Forwards are traded over the counter, futures overcome this drawback, therefore
they need standardization. A future is specified by:
• The underlying asset (usually commodities or shares)
• The contract size (the quantity of the asset)
• Delivery arrangements
• Delivery date (month)
• Quality of delivered goods
• Price quotes
• Daily price movement limits (prevention of large movements)
• Position limits (maximum of contracts a speculator may hold)

Stochastic Systems, 2010 20

 Futures

Remaining problems: how to avoid defaults and how to specify the forward
price.
Solution: margin account
• initial margin
• maintenance margin
• margin call

Stochastic Systems, 2010 21

 Convergence of futures and spot prices

6
Price

Futures

Spot

- t
T

Stochastic Systems, 2010 22

 Options

• The right, but not the obligation, to buy (or sell) an asset (underlying) in
the future
– call option: right to buy
– put option: right to sell
• Specification
– underlying (what can be bought or sold)
– the exercise (strike) price K
– expiration date T of the option
• Two types of options regarded here
– American option
– European option
• The party that grants the option is said to write an option
• The purchaser of an option faces only the risk of the original purchase price

Stochastic Systems, 2010 23

 Options

Intrinsic value of options

• C(S, t = T ) = max(0, S(T ) − K)
• P (S, t = T ) = max(K − S(T ), 0)

P C
6 6

0 - S 0 - S
K K
Put option Call option

• call option where S > K : in the money

• call options where S < K : out of the money.

Stochastic Systems, 2010 24

 Pricing of Options

Underlying concepts:
• arbitrage (equilibrium of prices)
• writer of the option should not undertake any risks by writing the option
asset price dynamics

dS(t) = µS(t)dt + σS(t)dW

where µ and σ are constant values.

Bond dynamics:

dB = rB(t)dt

Stochastic Systems, 2010 25

 Black Scholes equation

The Black Scholes equation is a PDE which explains the price dynamics of
derivative on a security.
• f = f (t, S): price of an arbitrary derivative on S
• r : risk free interest rate
• main idea: build a replicating portfolio with investments at the risk-free rate
and in the stock

Stochastic Systems, 2010 26

 Black Scholes equation

The call option is a derivative security of the underlying stock S and therefore,
it changes its value dC according to Itôs formula
³ ∂C ∂C 1 ∂ 2C 2 2´ ∂C
dC = + µS + σ S dt + σSdW (4)
∂t ∂S 2 ∂S 2 ∂S
The writer of the option does not want to undertake any risks by writing the
option and therefore forms a portfolio G which is comprised of the amount x
of the underlying stock and of the amount y of a bond B , which should match
the option value at all times.

Stochastic Systems, 2010 27

 Black Scholes equation

The bond dynamics are

dB = rBdt (5)

The portfolio G = xS + yB should be self-financing, i.e., no money should

be needed except for the initial capital outlay. The portfolio dynamics are given
by

dG = xdS + ydB + dxS + dyB = xdS + ydB (6)

The term dxS + dyB equals zero, because a change in the amount of stocks
and bonds we hold at constant stock and bond prices equals an in- or outflow
of money.

Stochastic Systems, 2010 28

 Black Scholes equation

The portfolio dynamics are extended by substituting dS from (??) into (6).
This yields

dG = x(µS(t)dt + σS(t)dw) + y(rB)dt = (xµS + yrB)dt + xσSdW . (7)

Since the portfolio dynamics should match the dynamics of the option, we
match the coefficients of dt and dw in (7) and (4).

Stochastic Systems, 2010 29

 Black Scholes equation

To do this, we first match the coefficient of dw by setting

∂C
x= . (8)
∂S
Since we want G = C and using G = xS + yB yields

∂C
G = yB + S=C
∂S
1³ ∂C ´
y= C− S . (9)
B ∂S

Stochastic Systems, 2010 30

 Black Scholes equation

Substituting the (8) and (9) into (7) and matching the coefficient of dt in (4)
gives

∂C 1³ ∂C ´ ∂C ∂C 1 ∂ 2C 2 2
µS + C− S rB = + µS + 2
σ S .
∂S B ∂S ∂t ∂S 2 ∂S
Finally, we arrive at the celebrated Black-Scholes PDE

∂C ∂C 1 ∂ 2C 2 2
rC = + rS + σ S . (10)
∂t ∂S 2 ∂S 2
The solution of the PDE together with the appropriate boundary conditions
gives the pricing of the option.

Stochastic Systems, 2010 31

 Black Scholes equation

Let us consider the stock itself. It is (in a trivial way) a derivative of the stock
itself, so C(t, S) = S(t) should satisfy the PDE. With this choice we have
∂C
• ∂S = 1
∂ 2C
• ∂S 2
=0
∂C
• ∂t = 0
and therefore rS(t) = rS(t).

This proves that C(t, S) = S(t) is a possible solution. There are uncountably
many more solutions to this PDE.

Stochastic Systems, 2010 32

 Black-Scholes formula for European call options

Pricing formula of a European call option: C = max(0, ST − K)

−r(T −t)
C(t, S) = S(t)N (d1) − Ke N (d2).

where N (d) is the cumulative normal distribution and

Z d y2
1 − 2
N (d) = √ e dy
2π −∞

with d1 und d2
2
S
ln( K ) + (r + σ2 )(T − t) p
d1 = √ , d2 = d1 − σ T −t
σ T −t

Stochastic Systems, 2010 33

 Put-Call parity

Relationship between the put and the call price of a European option → arbitrage
argument:
• long position in call with strike price K with value C
• short position in put with strike price K with value P
• payoff at maturity:
C − P = max(S(T ) − K, 0) − max(K − S(T ), 0) = S(T ) − K
This yields (at maturity): C − P + K = S(T )
→ Lend Ke−r(T −t), which will be worth K at maturity:
−r(T −t)
C − P + Ke = S(t)

Stochastic Systems, 2010 34

 Black-Scholes formula for European put options

→ use put call parity:

−r(T −t)
P (t, S) = Ke N (−d2) − S(t)N (−d1)

where N (d) is the cumulative normal distribution and

Z d y2
1 − 2
N (d) = √ e dy
2π −∞

with d1 und d2
2
S
ln( K ) + (r + σ2 )(T − t) p
d1 = √ , d2 = d1 − σ T −t
σ T −t

Stochastic Systems, 2010 35

 Black-Scholes formula for American options

∂P ∂P 1 ∂ 2P 2 2
+ rS + σ S − rP ≤ 0
∂t ∂S 2 ∂S 2

P (t, S) > max(K − S(t), 0) P (T, S) = max(K − S(T ), 0)

boundary conditions for the PDE:

• limS→∞ P (T, S) = 0
• P (T, 0) = e−r(T −t)K (if S(t) = 0)

Stochastic Systems, 2010 36

 General Option Pricing

Consider a general model for the underlying:

S(t) = µ(t, S)dt + σ(t, S)dw

dB = rBdt

→ the PDE for pricing any possible option f (t, S) is given by

∂f ∂f 1 ∂ 2f 2
rf = + rS + σ(t, S)
∂t ∂S 2 ∂S 2

Stochastic Systems, 2010 37

 Stochastic returns I

Portfolio example: Geometric Brownian motion model with σ as a constant and µ

as a stochastic process (mean reverting)

dµ(t) = κ(θ − µ(t)) dt + σµ dWµ

The stock market portfolio process:

dP (t) = µ(t)P (t) dt + σP P (t) dWP

The correlation between Wµ and WP is denoted by ρ. The variable r denotes the

short term risk free interest rate, u(t) the fraction of the wealth invested in the
stock market

Stochastic Systems, 2010 38

 Stochastic returns II

The optimal control problem

( )
1 γ
max E X(T )
u(t) γ
s.t.
dX(t) = X(t)[u(t)(µ(t) − r) + r] dt + X(t)u(t)σP dWP
p
dµ(t) = κ(θ − µ(t))dt + σµ(ρdWP + 1 − ρ2dWµ).

The optimal cost-to-go function by

( )
1 γ
J(t, X, µ) = max E X(T ) .
u(t) γ

Stochastic Systems, 2010 39

 Stochastic returns III

HJB equation:
h
Jt + max JxX(t)[(µ − r)u(t) + r] + Jµκ(θ − µ(t))
u(t)

1 1 i
2 2 2 2
+ JxxX (t)u (t)σP + JxµX(t)u(t)ρσP σµ + Jµµσµ = 0
2 2
1 γ
J(T ) = X(T )
γ

Stochastic Systems, 2010 40

 Stochastic returns IV

The optimal u(t) (first order condition):

Jx(µ(t) − r) + JxµρσP σµ
u(t) = − .
JxxX(t)σP2

The optimal u(t) is plugged back into the HJB equation and rearranged, which
yields:

1 2
Jt + JxX(t)r + Jµ κ(θ − µ(t)) + Jµµ σµ
2
2 2 2
1 Jx2(µ(t) − r)2 JµxJx(µ(t) − r)ρσµ 1 Jµxρ σµ
− − − = 0.
2 JxxσP2 JxxσP 2 Jxx

Stochastic Systems, 2010 41

 Stochastic returns V

We now guess that the value function J has a form

1 r(T −t) γ a(t)+b(t)µ(t)+c(t)µ2 (t)

J(t, X, µ) = (X(t)e ) e
γ
1 γ
In order to satisfy the terminal condition J(T, X, µ) = γ X (T ), the following
conditions are imposed

a(T ) = b(T ) = c(T ) = 0.

Stochastic Systems, 2010 42

 Stochastic returns VI

1 2 2 2 1 2 ρ σµ c(t) 2 c2(t) ρ2σµ2

ċ(t) − κc(t) + c (t) − 2
− − = 0
γ γ γ 2(γ − 1)σP (γ − 1)σP (γ − 1)

1 κ b(t) c(t) 2r
ḃ(t) + (2 θ c(t) − b(t)) + +
γ γ γ (γ − 1) σP2
ρ σµ(a(t) − r c(t) 2 ρ2 σµ2 b(t) c(t)
− − = 0
(γ − 1) σP (γ − 1)

1 κθ b2(t) + 2c2(t) r2 ρ σµ r a(t) ρ2σµ2 b2(t)

ȧ(t) + b(t) + − − − = 0
γ γ 2γ 2(γ − 1)σ 2 (γ − 1) σP2 2(γ − 1)

with terminal conditions a(T ) = b(T ) = c(T ) = 0.

Stochastic Systems, 2010 43

 Stochastic returns VII

The optimal portfolio control law is given by

∗ (µ(t) − r) + ρ σP σµ[b(t) + 2 c(t) µ(t)]

u (t, X, µ) = .
(1 − γ) σP2

The three ODEs can be solved off line before the controller is used to make the
investment decisions.

Stochastic Systems, 2010 44

 Model: Asset price dynamics

We regard a market in which n ≥ 1 risk-bearing investments exist. The asset

price processes (P1(t), P2(t), . . . , Pn(t)) of the risk-bearing investments satisfy the
stochastic differential equation

dPi(t)
= µi(t, x(t)) dt + σi(t)dZP (t) , (11)
Pi(t)
Pi(0) = pi0 > 0 .

Here µi(t, x(t)) ∈ R is expected return, σi ∈ R1×n is the i-th row of the volatility
matrix σ(t) ∈ Rn×n, and ZP (t) ∈ Rn is a Brownian motion. Additionally, we consider
a “risk-freeässet with a short-term interest rate, obtained by

dP0(t)
= r(t, x(t))dt , (12)
P0(t)
P0(0) = p00 > 0 .

Stochastic Systems, 2010 45

 Model: Mean Returns and Factor dynamics

The factors affect the mean return of the risk-bearing assets and the interest rate of
the risk-free asset. Furthermore, we assume that the drift terms are affine functions of
the factor levels as given by

µi(t, x(t)) = Gi(t)x(t) + gi(t) , (13)

r(t, x(t)) = F0(t)x(t) + f (t) , (14)

where x(t) = (x1(t), x1(t), . . . xm(t))T ∈ Rm, Gi(t), F0(t) ∈ R1×m, and
gi(t), f0(t) ∈ R. The factors are modelled as Gaussian processes obtained by

dx(t) = (A(t)x(t) + a(t))dt + ν(t)dZx(t) , (15)

where A(t) ∈ Rm×m, a(t) ∈ Rm, ν(t) ∈ Rm×m, and Zx(t) ∈ Rm is a Brownian
motion. The correlation matrix between ZP (t) and Zx(t) is ρ(t) ∈ Rn×m. The factors
allow us to connect the mean returns with the underlying economic environment.

Stochastic Systems, 2010 46

 Model: Wealth dynamics

The wealth dynamics of our portfolio optimization problem is given as

T
dW (t) = u(t) (F (t)x(t) + f (t))W (t) dt + (F0(t)x(t) + f0(t))W (t) dt
T
+ W (t)u(t) σ(t)dZP (t) + q(t)dt , (16)

where u(t) = (u1(t), u2(t), . . . , un(t))T denotes the fraction of wealth (port-
folio value) invested in the i-th asset at time t, and q(t) denotes and in-
Pn
or outflows of money. The weights fulfil the constraints i=0 ui (t) = 1 and
T
u(t) = (u1(t), u2(t), . . . , un(t)) ∈ U where U is a convex set. We use the
following abbreviations

u(t) = (u1(t), u2(t), . . . , un(t))T ∈ U ,

G(t) = [GT1 (t), GT2 (t), . . . , GTn (t)] ∈ Rn×m, F (t) = G(t) − eF0(t) ∈ Rn×m
g(t) = (g1(t), . . . , gn(t))T ∈ Rn×1, f (t) = f (t) − ef0(t) ∈ Rn×1
σ = [σiT , . . . , σnT ] ∈ Rn×n, e = (1, 1, . . . , 1)T ∈ Rn×1 .

Stochastic Systems, 2010 47

 Portfolio Optimization: CRRA utility

The first portfolio choice problem is to maximize the expected power utility defined
over terminal wealth. Additionally, we assume that q(t) = 0, ZP (t) and Zx(t) are
correlated, and u is unrestricted. Mathematically the problem statement is:
h1 i
γ
max E W (T )
u(·)∈Rn γ
s.t.
T
dW (t) = W (t)[F0(t)x(t) + f0(t) + u (t)(F (t)x(t) + f (t))]dt
T
+W (t)u (t)σ(t)dZP , W (0) = W0
dx(t) = (A(t)x(t) + a(t))dt + ν(t)dZx, x(0) = x0
dZP dZx = ρ(t)dt , (17)

where T denotes the time horizon and γ < 1 denotes the coefficient of risk aversion.

Stochastic Systems, 2010 48

 Portfolio Optimization: CARA utility

The second portfolio choice problem is to maximize the exponential utility defined over
terminal wealth. We assume the risk-free interest rate not to be a function of x(t),
i.e., F0(t) = 0 and that q(t) = H(t)x(t) + h(t). Problem statement is given by
h 1 −γ W (T )i
max E − e
u(·)∈Rn γ
s.t.
T
dW (t) = W (t)[f0(t) + u (t)(F (t)x(t) + f (t))]dt
T
+[H(t)x(t) + h(t)]dt + W (t)u (t)σ(t)dZP , W (0) = W0
dx(t) = (A(t)x(t) + a(t))dt + ν(t)dZx, x(0) = x0
dZP dZx = ρ(t)dt , (18)

where T denotes the time horizon and γ > 0 denotes the coefficient of risk aversion.

Stochastic Systems, 2010 49

 Solution to the Portfolio Optimization: CRRA utility

In order to obtain the solution, we solve the following problem :

h
T
Jt(·) + maxn W (t)(F0(t)x(t) + f0(t) + u (t)(F (t)x(t) + f (t)))JW (·)
u(·)∈R

T 1 2 T
+ (A(t)x(t) + a(t)) Jx(·) + W (t)u (t)Σ(t)u(t)JW W (·)
2
1 i
T T T
+ W (t)u (t)σ(t)ρ(t)ν (t)JW x(·) + tr{Jxx(·)ν(t)ν (t)} = 0,(19)
2

where Σ(t) = σ(t)σ T (t) and with the terminal condition J(T, W (T ), x(T )) =
1 γ
γ W (T ) . Using the first order conditions, we obtain the optimal portfolio control
(composition) vector

∗ −Σ−1(t) ³ T
´
u (·) = JW (·)(F (t)x(t) + f (t)) + σ(t)ρ(t)ν(t) JW x(·) . (20)
W (t)JW W (·)

Stochastic Systems, 2010 50

 Solution to the Portfolio Optimization: CRRA utility

By using the optimal portfolio vector (20) and inserting it back in (19), we obtain
T
Jt(·) + W (t)(F0(t)x(t) + f0(t))JW (·) + (A(t)x(t) + a(t)) Jx(·)
1 T
+ tr{Jxx(·)ν(t)ν(t) }
2
2
1 JW (·) T −1
− (F (t)x(t) + f (t)) Σ (t)(F (t)x(t) + f (t))
2 JW W (·)
T −1 T JW (·)
− (F (t)x(t) + f (t)) Σ (t)σ(t)ρ(t)ν (t) JW x(·)
JW W (·)
T
JW x (·) T T
− ν(t)ρ (t)ρ(t)ν (t)JW x(·) = 0 , (21)
2JW W (·)

with the terminal condition J(T, W (T ), x(T )) = γ1 W (T )γ .

Stochastic Systems, 2010 51

 Solution to the Portfolio Optimization: CRRA utility

The solution for J(t, W (t), x(t)) needs to be found in order to calculate the optimal
feedback controller. Hence we conjecture a functional form of the solution

1 γ T (t)x(t)+ 1 xT (t)K (t)x(t)

k1 (t)+k2 3
J(t, W, e) = W (t) l(t, x), l(t, x) = e 2 , (22)
γ

with terminal conditions k1(T ) = 0, k2(T ) = 0, and K3(T ) = 0. Furthermore, one

may notice that K3(t) is a symmetric m × m matrix, k2(t) is a m × 1 vector, and
k1(t) is a scalar.
We compute the partial derivatives of (22) and insert it into (21) in order to derive the
conditions for the solution. The solution consists of three ordinary differential equations
(ODEs) for k1(t), k2(t), and K3(t), and an algebraic relationship between the control
vector u(·) and the parameters k1(t), etc.

Stochastic Systems, 2010 52

 Solution to the Portfolio Optimization: CRRA utility

We obtain the following three systems of nonlinear ODEs. For the scalar k1(t):

1
k˙1
T T T
+ tr{ν (k2k2 + K3))ν} + a k2 + γf0
2
γ ³ ´
T −1 T −1 T T T T
− f Σ f + 2f Σ σρν k2 + k2 νρ ρν k2 = 0
2(γ − 1)
k1(T ) = 0. (23)

For the vector k2(t):

k˙2 + γF0 + K3νν k2 + A k2 + K3a

T T T

γ ³ ´
T −1 T −1 T T −1 T T
− F Σ f + F Σ σρνk2 + K3νρ σ Σ f + K3νρ ρν k2 = 0
(γ − 1)
k2(T ) = 0 . (24)

Stochastic Systems, 2010 53

 Solution to the Portfolio Optimization: CRRA utility

For the matrix K3(t):

T T
K̇3 + K3νν K3 + K3A + A K3
γ ³
T −1 T −1 T
− F Σ F + F Σ σρν K3
(γ − 1)
´
T T −1 T T
+ K3νρ σ Σ F + K3νρ ρν K3 = 0

K3(T ) = 0. (25)

The optimal feedback controller is calculated by using the value function conjecture
(22) and, plugging it in (20), we obtain

∗ Σ−1(t) ³ T ¡ ¢´
u (·) = F (t)x(t) + f (t) + σ(t)ρ(t)ν (t) K3(t)x(t) + k2(t) . (26)
1−γ

Stochastic Systems, 2010 54

 Solution to the Portfolio Optimization: CRRA utility

In order to compute the optimal controller, we need to calculate the solutions of (24)
and (25). This results in solving 12 m2 + 23 m ODEs, where m is the dimension of the
vector x(t).
In the case, that the all the parameters are time-invariant and if the following two
conditions are met:
h γ γ i
T T −1 T T T 1
( νρ σ Σ F − A) , (νν − νρ ρν ) 2 is controllable (27)
γ−1 γ−1
h γ 1 γ i
T −1 T T −1
( F Σ F )2 , ( νρ σ Σ F − A) is observable , (28)
γ−1 γ−1

the Riccati equation (25) possesses a positive-definite and finite solution for T → ∞,
γ
see [?]. If the matrix (K3νν T + AT − γ−1 (Σ−1σρν + K3νρT ρν T )) is invertible and
(25) possesses a positive definite finite solution, then (24) possesses a finite solution
for T → ∞.

Stochastic Systems, 2010 55

 Solution to the Portfolio Optimization: CRRA utility

The optimal controller can be split into two parts.

• The first being the myopic demand 1−γ 1
Σ−1(t)(F (t)x(t) + f (t)) which is
independent of the time horizon considered, similar to Merton Problem [?].

• The second part is the inter-temporal hedging demand

1 −1 T
1−γ Σ σ(t)ρ(t)ν (t)(K3(t)x(t) + k2(t)), which depends on the time
horizon and the correlation of the external variables with the asset price dynamics.
The shorter the time horizon is, the smaller the inter-temporal hedging demand
becomes, because of the terminal conditions are K3(T ) = 0 and k2(T ) = 0.

Stochastic Systems, 2010 56