MATH 361: Financial Mathematics for Actuaries Iusers.math.msu.edu/users/albert/math361fall17.pdf · 2017-11-08 · MATH 361: Financial Mathematics for Actuaries I Albert Cohen Actuarial

MATH 361: Financial Mathematics for Actuaries I

Albert Cohen

Actuarial Sciences ProgramDepartment of Mathematics

Department of Statistics and ProbabilityC336 Wells Hall

Michigan State UniversityEast Lansing MI

[email protected]@stt.msu.edu

Albert Cohen (MSU) Financial Mathematics for Actuaries I MSU Fall 2017 1 / 164

Course Information

Syllabus to be posted on class page in first week of classes

Homework assignments will posted there as well

Page can be found at https://math.msu.edu/classpages/


Course Information

Many examples within these slides are used with kind permission ofProf. Dmitry Kramkov, Dept. of Mathematics, Carnegie MellonUniversity.

Book for course: Marcel Finan’s A Discussion of Financial Economicsin Actuarial Models: A Preparation for the Actuarial Exam MFE/3F.Some proofs from there will be referenced as well. Please find thesenotes here

Some examples here will be similar to those practice questionspublicly released by the SOA. Please note the SOA owns thecopyright to these questions.


http://faculty.atu.edu/mfinan/actuarieshall/DFEM.pdf

What are financial securities?

Traded Securities - price given by market.

For example:

StocksCommodities

Non-Traded Securities - price remains to be computed.

Is this always true?

We will focus on pricing non-traded securities.




For example:

StocksCommodities







For example:

StocksCommodities







For example:

StocksCommodities





How does one fairly price non-traded securities?

By eliminating all unfair prices

Unfair prices arise from Arbitrage Strategies

Start with zero capitalEnd with non-zero wealth

We will search for arbitrage-free strategies to replicate the payoff of anon-traded security

This replication is at the heart of the engineering of financial products






























More Questions

Existence - Does such a fair price always exist?

If not, what is needed of our financial model to guarantee at least onearbitrage-free price?

Uniqueness - are there conditions where exactly one arbitrage-freeprice exists?


And What About...

Does the replicating strategy and price computed reflect uncertaintyin the market?

Mathematically, if P is a probabilty measure attached to a series ofprice movements in underlying asset, is P used in computing theprice?


And What About...

Does the replicating strategy and price computed reflect uncertaintyin the market?

Mathematically, if P is a probabilty measure attached to a series ofprice movements in underlying asset, is P used in computing theprice?


Notation

Forward Contract:

A financial instrument whose initial value is zero, and whose finalvalue is derived from another asset. Namely, the difference of thefinal asset price and forward price:

V (0) = 0,V (T ) = S(T )− F (1)

Value at end of term can be negative - buyer accepts this in exchangefor no premium up front


Notation

Forward Contract:


V (0) = 0,V (T ) = S(T )− F (1)



Notation

Forward Contract:


V (0) = 0,V (T ) = S(T )− F (1)



Notation

Interest Rate:

The rate r at which money grows. Also used to discount the valuetoday of one unit of currency one unit of time from the present

V (0) =1

1 + r,V (1) = 1 (2)


Notation

Interest Rate:

The rate r at which money grows. Also used to discount the valuetoday of one unit of currency one unit of time from the present

V (0) =1

1 + r,V (1) = 1 (2)


An Example of Replication

Forward Exchange Rate: There are two currencies, foreign anddomestic:

SBA = 4 is the spot exchange rate - one unit of B is worth SB

A of Atoday (time 0)

rA = 0.1 is the domestic borrow/lend rate

rB = 0.2 is the foreign borrow/lend rate

Compute the forward exchange rate FBA . This is the value of one unit

of B in terms of A at time 1.






































An Example of Replication: Solution

At time 1, we deliver 1 unit of B in exchange for FBA units of domestic

currency A.

This is a forward contract - we pay nothing up front to achieve this.

Initially borrow some amount foreign currency B, in foreign market togrow to one unit of B at time 1. This is achieved by the initial

amountSBA

1+rB(valued in domestic currency)

Invest the amountFBA

1+rAin domestic market (valued in domestic

currency)




currency A.



amountSBA




currency)




currency A.



amountSBA




currency)




currency A.



amountSBA




currency)



This results in the initial value

V (0) =FBA

1 + rA−

SBA

1 + rB(3)

Since the initial value is 0, this means

FBA = SB

A

1 + rA

1 + rB= 3.667 (4)



This results in the initial value

V (0) =FBA

1 + rA−

SBA

1 + rB(3)

Since the initial value is 0, this means

FBA = SB

A

1 + rA

1 + rB= 3.667 (4)


Discrete Probability Space

Let us define an event as a point ω in the set of all possible outcomes Ω.This includes the events ”The stock doubled in price over two tradingperiods” or ”the average stock price over ten years was 10 dollars”.

In our initial case, we will consider the simple binary spaceΩ = H,T for a one-period asset evolution. So, given an initialvalue S0, we have the final value S1(ω), with

S1(H) = uS0,S1(T ) = dS0 (5)

with d < 1 < u. Hence, a stock increases or decreases in price,according to the flip of a coin.

Let P be the probability measure associated with these events:

P[H] = p = 1− P[T ] (6)





S1(H) = uS0,S1(T ) = dS0 (5)



P[H] = p = 1− P[T ] (6)





S1(H) = uS0,S1(T ) = dS0 (5)



P[H] = p = 1− P[T ] (6)





S1(H) = uS0,S1(T ) = dS0 (5)



P[H] = p = 1− P[T ] (6)





S1(H) = uS0,S1(T ) = dS0 (5)



P[H] = p = 1− P[T ] (6)


Arbitrage

Assume that S0(1 + r) > uS0

Where is the risk involved with investing in the asset S ?

Assume that S0(1 + r) < dS0

Why would anyone hold a bank account (zero-coupon bond)?

Lemma Arbitrage free ⇒ d < 1 + r < u


Arbitrage







Arbitrage







Arbitrage







Arbitrage







Derivative Pricing

Let S1(ω) be the price of an underlying asset at time 1. Define thefollowing instruments:

Zero-Coupon Bond : V B0 = 1

1+r ,VB1 (ω) = 1

Forward Contract : V F0 = 0,V F

1 = S1(ω)− F

Call Option : V C1 (ω) = max(S1(ω)− K , 0)

Put Option : V P1 (ω) = max(K − S1(ω), 0)

In both the Call and Put option, K is known as the Strike.Once again, a Forward Contract is a deal that is locked in at time 0 forinitial price 0, but requires at time 1 the buyer to purchase the asset forprice F .

What is the value V0 of the above put and call options?


Derivative Pricing



1+r ,VB1 (ω) = 1


1 = S1(ω)− F






Derivative Pricing



1+r ,VB1 (ω) = 1


1 = S1(ω)− F



In both the Call and Put option, K is known as the Strike.

Once again, a Forward Contract is a deal that is locked in at time 0 forinitial price 0, but requires at time 1 the buyer to purchase the asset forprice F .



Derivative Pricing



1+r ,VB1 (ω) = 1


1 = S1(ω)− F






Derivative Pricing



1+r ,VB1 (ω) = 1


1 = S1(ω)− F






Put-Call Parity

Can we replicate a forward contract using zero coupon bonds and put andcall options?

Yes: The final value of a replicating strategy X has value

V C1 − V P

1 + (K − F ) = S1 − F = X1(ω) (7)

This is achieved (replicated) by

Purchasing one call option

Selling one put option

Purchasing K − F zero coupon bonds with value 1 at maturity.

all at time 0.Since this strategy must have zero initial value, we obtain

V C0 − V P

0 =F − K

1 + r(8)

Question: How would this change in a multi-period model?


Put-Call Parity

Can we replicate a forward contract using zero coupon bonds and put andcall options?Yes: The final value of a replicating strategy X has value

V C1 − V P

1 + (K − F ) = S1 − F = X1(ω) (7)






V C0 − V P

0 =F − K

1 + r(8)



Put-Call Parity


V C1 − V P

1 + (K − F ) = S1 − F = X1(ω) (7)






V C0 − V P

0 =F − K

1 + r(8)



Put-Call Parity


V C1 − V P

1 + (K − F ) = S1 − F = X1(ω) (7)





all at time 0.

Since this strategy must have zero initial value, we obtain

V C0 − V P

0 =F − K

1 + r(8)



Put-Call Parity


V C1 − V P

1 + (K − F ) = S1 − F = X1(ω) (7)






V C0 − V P

0 =F − K

1 + r(8)



Put-Call Parity


V C1 − V P

1 + (K − F ) = S1 − F = X1(ω) (7)






V C0 − V P

0 =F − K

1 + r(8)



General Derivative Pricing -One period model

If we begin with some initial capital X0, then we end with X1(ω). To pricea derivative, we need to match

X1(ω) = V1(ω) ∀ ω ∈ Ω (9)

to have X0 = V0, the price of the derivative we seek.

A strategy by the pair (X0,∆0) wherein

X0 is the initial capital

∆0 is the initial number of shares (units of underlying asset.)

What does the sign of ∆0 indicate?


Replicating Strategy

Initial holding in bond (bank account) is X0 −∆0S0

Value of portfolio at maturity is

X1(ω) = (X0 −∆0S0)(1 + r) + ∆0S1(ω) (10)

Pathwise, we compute

V1(H) = (X0 −∆0S0)(1 + r) + ∆0uS0

V1(T ) = (X0 −∆0S0)(1 + r) + ∆0dS0

Algebra yields

∆0 =V1(H)− V1(T )

(u − d)S0(11)





X1(ω) = (X0 −∆0S0)(1 + r) + ∆0S1(ω) (10)


V1(H) = (X0 −∆0S0)(1 + r) + ∆0uS0

V1(T ) = (X0 −∆0S0)(1 + r) + ∆0dS0

Algebra yields

∆0 =V1(H)− V1(T )

(u − d)S0(11)





X1(ω) = (X0 −∆0S0)(1 + r) + ∆0S1(ω) (10)


V1(H) = (X0 −∆0S0)(1 + r) + ∆0uS0

V1(T ) = (X0 −∆0S0)(1 + r) + ∆0dS0

Algebra yields

∆0 =V1(H)− V1(T )

(u − d)S0(11)





X1(ω) = (X0 −∆0S0)(1 + r) + ∆0S1(ω) (10)


V1(H) = (X0 −∆0S0)(1 + r) + ∆0uS0

V1(T ) = (X0 −∆0S0)(1 + r) + ∆0dS0

Algebra yields

∆0 =V1(H)− V1(T )

(u − d)S0(11)


Risk Neutral Probability

Let us assume the existence of a pair (p, q) of positive numbers, and usethese to multiply our pricing equation(s):

pV1(H) = p(X0 −∆0S0)(1 + r) + p∆0uS0

qV1(T ) = q(X0 −∆0S0)(1 + r) + q∆0dS0

Addition yields

X0(1 + r) + ∆0S0(pu + qd − (1 + r)) = pV1(H) + qV1(T ) (12)



Let us assume the existence of a pair (p, q) of positive numbers, and usethese to multiply our pricing equation(s):

pV1(H) = p(X0 −∆0S0)(1 + r) + p∆0uS0

qV1(T ) = q(X0 −∆0S0)(1 + r) + q∆0dS0

Addition yields

X0(1 + r) + ∆0S0(pu + qd − (1 + r)) = pV1(H) + qV1(T ) (12)



If we constrain

0 = pu + qd − (1 + r)

1 = p + q

0 ≤ p

0 ≤ q

then we have a risk neutral probability P where

V0 = X0 =1

1 + rE[V1] =

pV1(H) + qV1(T )

1 + r

p = P[ω = H] =1 + r − d

u − d

q = P[ω = T ] =u − (1 + r)

u − d



If we constrain

0 = pu + qd − (1 + r)

1 = p + q

0 ≤ p

0 ≤ q

then we have a risk neutral probability P where

V0 = X0 =1

1 + rE[V1] =

pV1(H) + qV1(T )

1 + r

p = P[ω = H] =1 + r − d

u − d

q = P[ω = T ] =u − (1 + r)

u − d


Example: Pricing a forward contract

Consider the case of a stock with

S0 = 400

u = 1.25

d = 0.75

r = 0.05

Then the forward price is computed via

0 =1

1 + rE[S1 − F ]⇒ F = E[S1] (13)



Consider the case of a stock with

S0 = 400

u = 1.25

d = 0.75

r = 0.05

Then the forward price is computed via

0 =1

1 + rE[S1 − F ]⇒ F = E[S1] (13)



This leads to the explicit price

F = puS0 + qdS0

= (p)(1.25)(400) + (1− p)(0.75)(400)

= 500p + 300− 300p = 300 + 200p

= 300 + 200 · 1 + 0.05− 0.75

1.25− 0.75= 300 + 200 · 3

5

= 420

Homework Question: What is the price of a call option in the caseabove,with strike K = 375?



This leads to the explicit price

F = puS0 + qdS0

= (p)(1.25)(400) + (1− p)(0.75)(400)

= 500p + 300− 300p = 300 + 200p

= 300 + 200 · 1 + 0.05− 0.75

1.25− 0.75= 300 + 200 · 3

5

= 420

Homework Question: What is the price of a call option in the caseabove,with strike K = 375?


Example: Pricing a call (and put) option contract

In this case, we have

V c0 =

1

1 + rE[V c

1 (ω)]

=1

1 + r(pV c

1 (H) + qV c1 (T ))

=1

1.05

(3

5(500− 375)+ +

2

5(300− 375)+

)=

1

1.05

(3

5(125) +

2

5(0)

)=

75

1.05= 71.43

V p0 =

1

1 + rE[V p

1 (ω)]

=1

1.05

(3

5(375− 500)+ +

2

5(375− 300)+

)=

1

1.05

(3

5(0) +

2

5(75)

)= 28.57.

(14)


General one period risk neutral measure

We define a finite set of outcomes Ω ≡ ω1, ω2, ..., ωn and anysubcollection of outcomes A ∈ F1 := 2Ω an event.

Furthermore, we define a probability measure P, not necessarily thephysical measure P to be risk neutral if

P[ω] > 0 ∀ ω ∈ ΩX0 = 1

1+r E[X1]

for all strategies X .



The measure is indifferent to investing in a zero-coupon bond, or arisky asset X

The same initial capital X0 in both cases produces the same”‘average”’ return after one period.

Not the physical measure attached by observation, experts, etc..

In fact, physical measure has no impact on pricing




















Example: Risk Neutral measure for trinomial case

Assume that Ω = ω1, ω2, ω3 with

S1(ω1) = uS0

S1(ω2) = S0

S1(ω3) = dS0

Given a payoff V1(ω) to replicate, are we assured that an initial price andreplicating strategy exists? By this, we mean 3 equations for 2 unknowns:

V1(ω1) = (V0 −∆0S0)(1 + r) + ∆0S1(ω1).

V1(ω2) = (V0 −∆0S0)(1 + r) + ∆0S1(ω2)

V1(ω3) = (V0 −∆0S0)(1 + r) + ∆0S1(ω3).

(15)

Is this too much to solve for (V0,∆0)?




S1(ω1) = uS0

S1(ω2) = S0

S1(ω3) = dS0


V1(ω1) = (V0 −∆0S0)(1 + r) + ∆0S1(ω1).

V1(ω2) = (V0 −∆0S0)(1 + r) + ∆0S1(ω2)

V1(ω3) = (V0 −∆0S0)(1 + r) + ∆0S1(ω3).

(15)





S1(ω1) = uS0

S1(ω2) = S0

S1(ω3) = dS0


V1(ω1) = (V0 −∆0S0)(1 + r) + ∆0S1(ω1).

V1(ω2) = (V0 −∆0S0)(1 + r) + ∆0S1(ω2)

V1(ω3) = (V0 −∆0S0)(1 + r) + ∆0S1(ω3).

(15)




Try our first example with

(S0, u, d , r) = (400, 1.25.0.75, 0.05)

V digital1 (ω) = 1S1(ω)>450(ω).

This translates to

1 = (V0 − 400∆0)(1.05) + 500∆0

0 = (V0 − 400∆0)(1.05) + 400∆0

0 = (V0 − 400∆0)(1.05) + 300∆0.

(16)

Now, assume you are observe the price on the market to be

V digital0 =

1

1 + rE[V digital

1 ] = 0.25. (17)

Use this extra information to price a call option with strike K = 420.


http://www.wolframalpha.com



(S0, u, d , r) = (400, 1.25.0.75, 0.05)

V digital1 (ω) = 1S1(ω)>450(ω).

This translates to

1 = (V0 − 400∆0)(1.05) + 500∆0

0 = (V0 − 400∆0)(1.05) + 400∆0

0 = (V0 − 400∆0)(1.05) + 300∆0.

(16)


V digital0 =

1

1 + rE[V digital

1 ] = 0.25. (17)






(S0, u, d , r) = (400, 1.25.0.75, 0.05)

V digital1 (ω) = 1S1(ω)>450(ω).

This translates to

1 = (V0 − 400∆0)(1.05) + 500∆0

0 = (V0 − 400∆0)(1.05) + 400∆0

0 = (V0 − 400∆0)(1.05) + 300∆0.

(16)


V digital0 =

1

1 + rE[V digital

1 ] = 0.25. (17)




Solution: Risk Neutral measure for trinomial case

The above scenario is reduced to finding the risk-neutral measure(p1, p2, p3). This can be done by finding the rref of the matrix M:

M =

1 1 1 1500 400 300 420

1 0 0 0.25(1.05)

(18)

which results in

rref (M) =

1 0 0 0.26250 1 0 0.6750 0 1 0.0625

. (19)



It follows that (p1, p2, p3) = (0.2625, 0.675, 0.0625), and so

V C0 =

1

1.05E[(S1 − 420)+ | S0 = 400]

=1

1.05(80p1 + 0p2 + 0p3)

=0.2625

1.05× (500− 420) = 20.

(20)

Could we perhaps find a set of digital options as a basis setV d1

1 (ω),V d21 (ω),V d3

1 (ω)

= 1A1(ω), 1A2(ω), 1A3(ω) (21)

with A1,A2,A3 ∈ F1 to span all possible payoffs at time 1?

How about (A1,A2,A3) = (ω1 , ω2 , ω3) ?



It follows that (p1, p2, p3) = (0.2625, 0.675, 0.0625), and so

V C0 =

1

1.05E[(S1 − 420)+ | S0 = 400]

=1

1.05(80p1 + 0p2 + 0p3)

=0.2625

1.05× (500− 420) = 20.

(20)

Could we perhaps find a set of digital options as a basis setV d1

1 (ω),V d21 (ω),V d3

1 (ω)

= 1A1(ω), 1A2(ω), 1A3(ω) (21)

with A1,A2,A3 ∈ F1 to span all possible payoffs at time 1?

How about (A1,A2,A3) = (ω1 , ω2 , ω3) ?


Exchange one stock for another

Assume now an economy with two stocks, X and Y . Assume that

(X0,Y0, r) = (100, 100, 0.01) (22)

and

(X1(ω),Y1(ω)) =

(110, 105) : ω = ω1

(100, 100) : ω = ω2

(80, 95) : ω = ω3.

Consider two contracts, V and W , with payoffs

V1(ω) = max Y1(ω)− X1(ω), 0W1(ω) = Y1(ω)− X1(ω).

(23)

Price V0 and W0.


Exchange one stock for another

In this case, our matrix M is such that

M =

1 1 1 1110 100 80 101105 100 95 101

(24)

which results in

rref (M) =

1 0 0 310

0 1 0 610

0 0 1 110

. (25)

It follows that

W0 =E[Y1]− E[X1]

1.01= Y0 − X0 = 0

V0 =1

1.01· (15p3) =

1.5

1.01= 1.49.

(26)


Homework

From Finan:

Problems 14.1, 14.3, 14.4, 14.5, 14.6, 14.7, 14.11

Problems 15.1, 15.3, 15.4, 15.6, 15.7, 15.10, 15.11


Existence of Risk Neutral measure

Let P be a probability measure on a finite space Ω. The following areequivalent:

P is a risk neutral measure

For all traded securities S i , S i0 = 1

1+r E[S i

1

]Proof: Homework (Hint: One direction is much easier than others. Also,strategies are linear in the underlying asset.)






1+r E[S i

1







1+r E[S i

1

]

Proof: Homework (Hint: One direction is much easier than others. Also,strategies are linear in the underlying asset.)






1+r E[S i

1



Complete Markets

A market is complete if it is arbitrage free and every non-traded asset canbe replicated.

Fundamental Theorem of Asset Pricing 1: A market is arbitrage freeiff there exists a risk neutral measure

Fundamental Theorem of Asset Pricing 2: A market is complete iffthere exists exactly one risk neutral measure

Proof(s): We will go over these in detail later!


Complete Markets






Complete Markets






Behold the Power of Replication!!

Consider a financial model, where one can buy or sell a forward contract,standard call option and standard put option on the same stock and withthe same maturity.

We assume that both options have the same strike K = 120 and theircurrent prices equal V Put

0 = 14 for the put and V Call0 = 7 for the call.

We also assume that the price of the stock at the maturity is equal toone of two possible values: 90 or 140.

We know that the model is arbitrage free.

Compute the forward price F .


Behold the Power of Replication!!

It’s clear that we need to solve for the triple (p, q, r).

However, we can map to another triple via replication of the forwardvia a strategy that invests in puts and calls.

If we initially purchase ∆P put contracts and ∆C call contracts, thenthe initial value of our portfolio is 0, as the initial value of a forward is0. This means

∆PV P0 + ∆CV C

0 = 0 (27)

and furthermore, we also have

∆PV P1 (ω) + ∆CV C

1 (ω) = S1(ω)− F . (28)

Together, this leads to a system of three equations and threeunknowns,

14∆P + 7∆C = 0

30∆P + 0∆C = 90− F

0∆P + 20∆C = 140− F .

(29)

Solving for F , we get F = 111.43.


Optimal Investment for a Strictly Risk Averse Investor

In the previous setting, we have assumed that the final payoff wasknown, and so we looked for the initial price of this instrument.

Now, we assume that we know our initial capital x as well as set ofbeliefs P and want to design the final payoff X1 that makes us thehappiest

All that is left to do is define happiness...


An Example: Minimizing Variance?

As an introductory example, consider an investor who has initialcapital X0 = x , and wishes to invest in a portfolio such that X1 is nottoo far away from her desired final value c = x(1 + α).

In this case, she knows that it isn’t possible that X1 = c for allpossible outcomes, so she decides to invest her x such that heroutcome X1(ω) minimizes the variance away from c :

u(x) = E[

1

2

(X1 − c

)2]

= minX1:E[X1]=x(1+r)

E[

1

2(X1 − c)2

]= − maxX1:E[X1]=x(1+r)

E[−1

2(X1 − c)2

] (30)

Now, how do we find what the optimal final payoff X1 should be?



Assume a complete market, with a unique risk-neutral measure P.

Characterize an investor by her pair (x ,U) of initial capital x ∈ X andutility function U : X → R+.

Assume U ′(x) > 0.

Assume U ′′(x) < 0.

Define the Radon-Nikodym derivative of P to P as the randomvariable

Z (ω) :=P(ω)

P(ω). (31)

Note that Z is used to map expectations under P to expectationsunder P: For any random variable X , it follows that

E[X ] = E[ZX ]. (32)



A strictly risk-averse investor now wishes to maximize her expected utilityof a portfolio at time 1, given initial capital at time 0:

u(x) := maxX1∈Ax

E[U(X1)]

Ax := all portfolio values at time 1 with initial capital x .(33)

In linear algebra terms, define

~Q := (p1, ..., pn)

~X1 := (X1(ω1), ...,X1(ωn)).(34)

Then for fixed scalars (x , r), we can rewrite

Ax =~X1 ∈ Rn | ~Q · ~X1 = x(1 + r)

. (35)



A strictly risk-averse investor now wishes to maximize her expected utilityof a portfolio at time 1, given initial capital at time 0:

u(x) := maxX1∈Ax

E[U(X1)]

Ax := all portfolio values at time 1 with initial capital x .(33)

In linear algebra terms, define

~Q := (p1, ..., pn)

~X1 := (X1(ω1), ...,X1(ωn)).(34)

Then for fixed scalars (x , r), we can rewrite

Ax =~X1 ∈ Rn | ~Q · ~X1 = x(1 + r)

. (35)



Theorem

Define X1 via the relationship

U ′(

X1

):= λZ (36)

where λ sets X1 as a strategy with an average return of r under P:

E[X1] = x(1 + r). (37)

Then X1 is the optimal strategy.



Proof.

Assume X1 ∈ Ax to be an arbitrary strategy with initial capital x , suchthat X1 6= X1 . Then for

f (y) := E[U(

X1 + y(X1 − X1))]

(38)

it follows that

f ′(0) = E[U ′(X1)

(X1 − X1

)]= E

[λZ(

X1 − X1

)]= λE

[(X1 − X1

)]= 0

f ′′(y) = E[

U ′′(

X1 + y(X1 − X1))(

X1 − X1

)2]< 0

(39)

and so f attains its maximum at y = 0. We conclude thatE[U(X1)] < E[U(X1)] for any admissible strategy X1.


Optimal Investment: U(x) = ln (x)

Assume an investor and economy defined by

U(x) = ln (x)

(S0, u, d , p, q, r) = (400, 1.25.0.75, 0.5, 0.5, 0.05).(40)

It follows that

(p, q) =

(3

5,

2

5

)(Z (H),Z (T )) =

(6

5,

4

5

).

(41)

Since U ′(x) = 1x , we have

X1(ω) =1

λ

1

Z (ω)

x = X0 =1

1 + rE[X1] =

1

1 + r

(p · 1

λ

p

p+ q · 1

λ

q

q

)=

1

λ

1

1 + r.

(42)



Combining the previous results, we see that

X1(ω) =x(1 + r)

Z (ω)

u(x) = p ln X1(H) + (1− p) ln X1(T )

= p ln

(x(1 + r)

Z (H)

)+ (1− p) ln

(x(1 + r)

Z (T )

)= ln

((1 + r)

Z (H)pZ (T )1−p x

)= ln (1.0717x) > ln (1.05x).

(43)



In terms of her actual strategy, we see that

π0 :=∆0S0

X0=

S0

x

X1(H)− X1(T )

S1(H)− S1(T )=

1 + r

u − d

(1

Z (H)− 1

Z (T )

)=

1 + r

u − d

(p

p− 1− p

1− p

)=

1.05

0.5

(5

6− 5

4

)= −0.875.

(44)

Therefore, the optimal strategy is to sell a stock portfolio worth 87.5% ofher initial wealth x and invest the proceeds into a safe bank account.



In fact, since π0 = 1+ru−d

(pp −

1−p1−p

), we see that qualitatively, her optimal

strategy involves

π0 =

> 0 : p > p= 0 : p = p< 0 : p < p.

This links with her strategy via

1 + r1(ω) :=X1(ω)

X0= (1− π0)(1 + r) + π0

S1(ω)

S0(45)

and so for our specific case where (r , u, d , π0) = (0.05, 1.25, 0.75,−0.875),we have

1 + r1(H) = (1− π0)(1 + r) + π0u = 0.875

1 + r1(T ) = (1− π0)(1 + r) + π0d = 1.3125.(46)


Optimal Investment: U(x) =√x

Consider now the same set-up as before, only that the utility functionchanges to U(x) =

√x .

It follows that

U ′(X1) =1

2

1√X1

⇒ X1 =1

4λ2

1

Z 2

(47)

Solving for λ returns

x(1 + r) = E[X1]

= E[Z X1]

= E[

Z1

4λ2

1

Z 2

]=

1

4λ2E[

1

Z

].

(48)


Optimal Investment: U(x) =√x

Combining the results above, we see that

X1 =x(1 + r)

Z 2E[

1Z

]⇒ u(x) = E

[√X1

]= E

[√x(1 + r)

Z 2E[

1Z

]]

=√

x(1 + r)

√E[

1

Z

].

(49)

Question: Is it true for all (p, p) ∈ (0, 1)× (0, 1) that√E[

1

Z

]=

√p2

p+

(1− p)2

1− p≥ 1? (50)


Optimal Investment: U(x) = −12(x − c)2

Consider now the same set-up as before, only that the utility functionchanges to U(x) = −1

2 (x − c)2.

It follows thatU ′(X1) = c − X1 = λZ

⇒ X1 = c − λZ(51)

Solving for λ returns

x(1 + r) = E[X1]

= E[Z X1]

= E [Z (c − λZ )]

= c − λE[Z 2]

⇒ λ =c − x(1 + r)

E [Z 2].

(52)


Optimal Investment: U(x) = −12(x − c)2

Combining the results above, we see that

X1 = c − c − x(1 + r)

E [Z 2]Z

⇒ u(x) = −E[−1

2

(X1 − c

)2]

= E

[1

2

(c − x(1 + r)

E [Z 2]Z

)2]

=1

2

[c − x(1 + r)2

]=

1

2x2(α− r)2.

(53)

HW: What are

the optimal proportion of wealth π0 invested in stock S0, and

the optimal return r1(ω)

for an economy with (S0, u, d , p, q, r) = (400, 1.25.0.75, 0.5, 0.5, 0.05)?


Optimal Betting at the Omega Horse Track!

Imagine our investor with U(x) = ln (x) visits a horse track.

There are three horses: ω1, ω2 and ω3.

She can bet on any of the horses to come in 1st .

The payoff is 1 per whole bet made.

She observes the price of each bet with payoff 1 right before the raceto be

(B10 ,B

20 ,B

30 ) = (0.5, 0.3, 0.2). (54)

Symbolically,B i

1(ω) = 1ωi(ω). (55)

Our investor feels the physical probabilities of each horse winning is

(p1, p2, p3) = (0.6, 0.35, 0.05). (56)

How should she bet if the race is about to start?



In this setting, we can assume r = 0.

This directly implies that

(p1, p2, p3) = (0.5, 0.3, 0.2). (57)

Our Radon-Nikodym derivative of P to P is now

(Z (ω1),Z (ω2),Z (ω3)) =

(0.5

0.6,

0.3

0.35,

0.2

0.05

). (58)



Her optimal strategy X1 reflects her betting strategy, and satisfies

X1(ω) =X0

Z (ω)

∴

(X1(ω1)

X0,

X1(ω2)

X0,

X3(ω1)

X0

)=

(6

5,

7

6,

1

4

).

(59)

So, per dollar of wealth, she buys 65 of a bet for Horse 1 to win, 7

6 ofa bet for Horse 2 to win, and 1

4 of a bet for Horse 3 to win.

The total price (per dollar of wealth) is thus

6

5· 0.5 +

7

6· 0.3 +

1

4· 0.2 = 0.6 + 0.35 + 0.05 = 1. (60)


Dividends

What about dividends? How do they affect the risk neutral pricing ofexchange and non-exchange traded assets? What if they are paid atdiscrete times? Continuously paid?

Recall that if dividends are paid continuously at rate δ, then 1 share attime 0 will accumulate to eδT shares upon reinvestment of dividends intothe stock until time T .

It follows that to deliver one share of stock S with initial price S0 at timeT , only e−δT shares are needed. Correspondingly,

Fprepaid = e−δTS0

F = erT e−δTS0 = e(r−δ)TS0.(61)


Dividends







Dividends







Binomial Option Pricing w/ cts Dividends and Interest

Over a period of length h, interest increases the value of a bond by afactor erh and dividends the value of a stock by a factor of eδh.

Once again, we compute pathwise,

V1(H) = (X0 −∆0S0)erh + ∆0eδhuS0

V1(T ) = (X0 −∆0S0)erh + ∆0eδhdS0

and this results in the modified quantities

∆0 = e−δhV1(H)− V1(T )

(u − d)S0

p =e(r−δ)h − d

u − d

q =u − e(r−δ)h

u − d


Binomial Option Pricing w/ cts Dividends and Interest

Over a period of length h, interest increases the value of a bond by afactor erh and dividends the value of a stock by a factor of eδh.

Once again, we compute pathwise,

V1(H) = (X0 −∆0S0)erh + ∆0eδhuS0

V1(T ) = (X0 −∆0S0)erh + ∆0eδhdS0

and this results in the modified quantities

∆0 = e−δhV1(H)− V1(T )

(u − d)S0

p =e(r−δ)h − d

u − d

q =u − e(r−δ)h

u − dAlbert Cohen (MSU) Financial Mathematics for Actuaries I MSU Fall 2017 55 / 164

Binomial Models w/ cts Dividends and Interest

For σ, the annualized standard deviation of continuously compoundedstock return, the following models hold:

Futures - Cox (1979)

u = eσ√h

d = e−σ√h.

General Stock Model

u = e(r−δ)h+σ√h

d = e(r−δ)h−σ√h.

Currencies with rf the foreign interest rate, which acts as a dividend:

u = e(r−rf )h+σ√h

d = e(r−rf )h−σ√h.


1- and 2-period pricing

We can solve for 2-period problems

on a case-by-case basis, or

by developing a general theory for multi-period asset pricing.

In the latter method, we need a general framework to carry out ourcomputations














Risk Neutral Pricing Formula

Assume now that we have the ”regular assumptions” on our coin flipspace, and that at time N we are asked to deliver a path dependentderivative value VN . Then for times 0 ≤ n ≤ N, the value of thisderivative is computed via

Vn = e−rhEn [Vn+1] (62)

and so

X0 = E0 [XN ]

Xn :=Vn

enh.

(63)




Vn = e−rhEn [Vn+1] (62)

and so

X0 = E0 [XN ]

Xn :=Vn

enh.

(63)




Vn = e−rhEn [Vn+1] (62)

and so

X0 = E0 [XN ]

Xn :=Vn

enh.

(63)




Vn = e−rhEn [Vn+1] (62)

and so

X0 = E0 [XN ]

Xn :=Vn

enh.

(63)


Computational Complexity

Consider the case

p = q =1

2

S0 = 4, u =4

3, d =

3

4

(64)

but now with term n = 3.

There are 23 = 8 paths to consider.

However, there are 3 + 1 = 4 unique final values of S3 to consider.

In the general term N, there would be 2N paths to generate SN , butonly N + 1 distinct values.

At any node n units of time into the asset’s evolution, there are n + 1distinct values.


Computational Complexity

At each value s for Sn, we know that Sn+1 = 43 s or Sn+1 = 3

4 s.

Using multi-period risk-neutral pricing, we can generate forvn(s) := Vn(Sn(ω1, ..., ωn)) on the node (event) Sn(ω1, ..., ωn) = s:

vn(s) = e−rh[pvn+1

(4

3s)

+ qvn+1

(3

4s)]. (65)


An Example:

Assume r , δ, and h are such that

p =1

2= q, e−rh =

9

10

S0 = 4, u = 2, d =1

2V3 := max 10− S3, 0 .

(66)

It follows thatv3(32) = 0

v3(8) = 2

v3(2) = 8

v3(0.50) = 9.50.

(67)

Compute V0.


Another Example

Consider the case r = 0.10, δ = 0.05, h = 0.01, σ = 0.1,S0 = 10 = K .

Now price two digital options, using the

1 General Stock Model

2 Futures-Cox Model

with respective payoffs

V K1 (ω) := 1S1≥K(ω)

V K2 (ω) := 1S2≥K(ω).


Another Example

Consider the case r = 0.10, δ = 0.05, h = 0.01, σ = 0.1,S0 = 10 = K .

Now price two digital options, using the

1 General Stock Model

2 Futures-Cox Model

with respective payoffs

V K1 (ω) := 1S1≥K(ω)

V K2 (ω) := 1S2≥K(ω).


Homework

From Notes:

Previous two examples.

Be able to compute the maximum expected utility u(x), optimalproportion of wealth π0 invested in stock S0, and optimal return r1(ω)for

(S0, u, d , p, q, r) = (400, 1.25.0.75, 0.5, 0.5, 0.05), and

Utility functions U(x) = x13 and U(x) = 1− e−x .

From Finan:

Problems 16.8, 16.9, 16.10

Problems 17.1, 17.2, 17.3, 17.4, 17.5, 17.6, 17.7, 17.8, 17.9

Problems 18.7, 18.9, 18.10.


Markov Processes

If we use the above approach for a more exotic option, say a lookbackoption that pays the maximum over the term of a stock, then we findthis approach lacking.

There is not enough information in the tree or the distinct values forS3 as stated. We need more.

Consider our general multi-period binomial model under P.

Definition We say that a process X is adapted if it depends only on thesequence of flips ω := (ω1, ..., ωn)

Definition We say that an adapted process X is Markov if for every0 ≤ n ≤ N − 1 and every function f (x) there exists another function g(x)such that

En [f (Xn+1)] = g(Xn). (68)


Markov Processes

This notion of Markovity is essential to our state-dependent pricingalgorithm.

Indeed, our stock process evolves from time n to time n + 1, usingonly the information in Sn.

We can in fact say that for every f (s) there exists a g(s) such that

g(s) = e−rhEn [f (Sn+1) | Sn = s] . (69)

In fact, that g depends on f :

g(s) = e−rh[pf (us) + qf (ds)

]. (70)

So, for any f (s) := VN(s), we can work our recursive algorithmbackwards to find the gn(s) := Vn(s) for all 0 ≤ n ≤ N − 1


Markov Processes

Some more thoughts on Markovity:

Consider the example of a Lookback Option.

Here, the payoff is dependent on the realized maximumMn := max0≤i≤nSi of the asset.

Mn is not Markov by itself, but the two-factor process (Mn, Sn) is.Why?

Let’s generate the tree!

Homework Can you think of any other processes that are not Markov?


The Interview Process

Consider the following scenario:

After graduating, you go on the job market, and have 4 possible jobinterviews with 4 different companies.

So sure of your prospects that you know that each company will makean offer, with an identically, independently distributed probabilityattached to the 4 possible salary offers:

P [Salary Offer=50, 000] = 0.1



P [Salary Offer=100, 000] = 0.2.

(71)



Questions:

How should you interview?

Specifically, when should you accept an offer and cancel theremaining interviews?

How does your strategy change if you can interview as many times asyou like, but the distribution of offers remains the same as above?



Questions:






Questions:





The Interview Process: Strategy

Some more thoughts...

At any time the student will know only one offer, which she can eitheraccept or reject.

Of course, if the student rejects the first three offers, than she has toaccept the last one.

So, compute the maximal expected salary for the student after thegraduation and the corresponding optimal strategy.


The Interview Process: Optimal Strategy

The solution process Xk4k=1 follows an Optimal Stopping Strategy:

Xk(s) = max

s, E[Xk+1 | kth offer = s

]. (72)

At time 4, the value of this game is X4(s) = s, with s being the salaryoffered.

At time 3, the conditional expected value of this game is

E[X4 | 3rd offer = s

]= E[X4]

= 0.1× 50, 000 + 0.3× 70, 000

+ 0.4× 80, 000 + 0.2× 100, 000

= 78, 000.

(73)

Hence, one should accept an offer of 80, 000 or 100, 000, and rejectthe other two.



This strategy leads to a valuation:

X3(50, 000) = 78, 000

X3(70, 000) = 78, 000

X3(80, 000) = 80, 000

X3(100, 000) = 100, 000.

(74)

At time 2, similar reasoning using E2[X3] leads to the valuation

X2(50, 000) = 83, 200

X2(70, 000) = 83, 200

X2(80, 000) = 83, 200

X2(100, 000) = 100, 000.

(75)



At time 1,X1(50, 000) = 86, 560

X1(70, 000) = 86, 560

X1(80, 000) = 86, 560

X1(100, 000) = 100, 000.

(76)

Finally, at time 0, the value of this optimal strategy is

E0[X1] = E[X1] = 0.8× 86, 560 + 0.2× 100, 000 = 89, 248. (77)

So, the optimal strategy is, for the first two interviews, accept only anoffer of 100, 000. If after the third interview, and offer of 80, 000 or100, 000 is made, then accept. Otherwise continue to the last interviewwhere you should accept whatever is offered.


Review

Let’s review the basic contracts we can write:

Forward Contract Initial Value is 0, because both buyer and sellermay have to pay a balance at maturity.

(European) Put/Call Option Initial Value is > 0, because both onlyseller must pay balance at maturity.

(European) ”Exotic” Option Initial Value is > 0, because both onlyseller must pay balance at maturity.

During the term of the contract, can the value of the contract ever fallbelow the intrinsic value of the payoff? Symbolically, does it ever occurthat

vn(s) < g(s) (78)

where g(s) is of the form of g(S) := max S − K , 0, in the case of a Calloption, for example.


Review






vn(s) < g(s) (78)



Review






vn(s) < g(s) (78)



Review






vn(s) < g(s) (78)



Review






vn(s) < g(s) (78)



Review






vn(s) < g(s) (78)



Review






vn(s) < g(s) (78)



For Freedom! (we must charge extra...)

What happens if we write a contract that allows the purchaser to exercisethe contract whenever she feels it to be in her advantage? By allowing thisextra freedom, we must

Charge more than we would for a European contract that is exercisedonly at the term N.

Hedge our replicating strategy X differently, to allow for thepossibility of early exercise.












American Options

In the end, the option v is valued after the nth value of the stockSn(ω) = s is revealed via the recursive formula along each path ω:

vn(Sn(ω)) = max

g(Sn(ω)), e−rhE[v(Sn+1(ω)) | Sn(ω)

]τ∗(ω) = inf k ∈ 0, 1, ..,N | vk(Sk(ω)) = g(Sk(ω)) .

(79)

Here, τ∗ is the optimal exercise time.

In the Binomial case, we reduce to

vn(s) = max

g(s), e−rh [pvn+1(us) + qvn+1(ds)]

τ∗ = inf k | vk(s) = g(s) .(80)


American Options

Some examples:

”American Bond:” g(s) = 1

”American Digital Option:” g(s) = 16≤s≤10

”American Square Option:” g(s) = s2.

Does an investor exercise any of these options early? Consider again thesetting

p =1

2= q, e−rh =

9

10

S0 = 4, u = 2, d =1

2.

(81)


American Square Options

Consider the American Square Option. We know via Jensen’s Inequalitythat

e−rhE[g(Sn+1) | Sn

]= e−rhE

[S2n+1 | Sn

]≥ e−rh

(E[Sn+1 | Sn

]2)= e−rh

(erhSn

)2

= erhS2n > S2

n = g(Sn).

(82)


American Square Options

It follows that vN(s) = s2 and

vN−1(s) = max

g(s), e−rhE[vN(SN) | SN−1 = s]

= max

s2, e−rhE[S2N | SN−1 = s]

= e−rhE[S2

N | SN−1 = s]

= e−rhE[vN(SN) | SN−1 = s].

(83)

In words, with one period to go, don’t exercise yet!!

The American and European option values coincide. Keep going.

How about with two periods left before expiration?


American Options

Let’s return to the previous European Put example, where

p =1

2= q, e−rh =

9

10

S0 = 4, u = 2, d =1

2V3 := max 10− S3, 0 .

(84)

It follows that S3(ω) ∈

12 , 2, 8, 32

.

Use this to compute v3(s) and the American Put recursively.

HW: FinanEx.19.1, 19.2, 19.3, 19.5, 19.7, 19.8, 19.9, 19.10, 19.12, 19.13 .


Asian Options

In times of high volatility or frequent trading, a company may want toprotect against large price movements over an entire time period, using anaverage. For example, if a company is looking at foreign exchange marketsor markets that may be subject to stock pinning due to large actors.

As an input, the average of an asset is used as an input against a strike,instead of the spot price.

There are two possibilities for the input in the discrete case: h = TN and

Arithmetic Average: IA(T ) := 1N+1

∑Nk=0 Skh .

Geometric Average: IG (T ) :=(

ΠNk=0Skh

) 1N+1

.

HW: Is there an ordering for IA, IG that is independent of T ?


Asian Options





∑Nk=0 Skh .


ΠNk=0Skh

) 1N+1

.



Asian Options





∑Nk=0 Skh .


ΠNk=0Skh

) 1N+1

.



Asian Options





∑Nk=0 Skh .


ΠNk=0Skh

) 1N+1

.



Asian Options: An Example:

Notice that these are path-dependent options, unlike the put and calloptions that we have studied until now. Assume r , δ = 0, and h are suchthat

S0 = 4, u = 2, d =1

2, e−rh =

9

10g(I ) = max I − 2.5, 0 .

(85)

It follows that p = erh−du−d =

109− 1

2

2− 12

= 1127 .



Consider an arithmetic average with N = 2. Then

vE2 (HH) = max

4 + 8 + 16

3− 2.5, 0

=

41

6

vE2 (HT ) = max

4 + 8 + 4

3− 2.5, 0

=

17

6

vE2 (TH) = max

4 + 2 + 4

3− 2.5, 0

=

5

6

vE2 (TT ) = max

4 + 2 + 1

3− 2.5, 0

= 0.

(86)

Compute v0, assuming a European structure. How about an Americanstructure?




vE2 (HH) = max

4 + 8 + 16

3− 2.5, 0

=

41

6

vE2 (HT ) = max

4 + 8 + 4

3− 2.5, 0

=

17

6

vE2 (TH) = max

4 + 2 + 4

3− 2.5, 0

=

5

6

vE2 (TT ) = max

4 + 2 + 1

3− 2.5, 0

= 0.

(86)

Compute v0, assuming a European structure.

How about an Americanstructure?




vE2 (HH) = max

4 + 8 + 16

3− 2.5, 0

=

41

6

vE2 (HT ) = max

4 + 8 + 4

3− 2.5, 0

=

17

6

vE2 (TH) = max

4 + 2 + 4

3− 2.5, 0

=

5

6

vE2 (TT ) = max

4 + 2 + 1

3− 2.5, 0

= 0.

(86)

Compute v0, assuming a European structure. How about an Americanstructure?




vE1 (H) = e−rh

(pvE

2 (HH) + qvE2 (HT )

)=

9

10

(11

27

41

6+

16

27

17

6

)=

241

60

vE1 (T ) = e−rh

(pvE

2 (TH) + qvE2 (TT )

)=

9

10

(11

27

5

6+

16

27

0

6

)=

11

36

∴ vE0 = e−rh

(pvE

1 (H) + qvE1 (T )

)=

9

10

(11

27

241

60+

16

27

11

36

)=

8833

5400

≈ 1.636.(87)




vE1 (H) = e−rh

(pvE

2 (HH) + qvE2 (HT )

)=

9

10

(11

27

41

6+

16

27

17

6

)=

241

60

vE1 (T ) = e−rh

(pvE

2 (TH) + qvE2 (TT )

)=

9

10

(11

27

5

6+

16

27

0

6

)=

11

36

∴ vE0 = e−rh

(pvE

1 (H) + qvE1 (T )

)=

9

10

(11

27

241

60+

16

27

11

36

)=

8833

5400

≈ 1.636.(87)



For the Americanized version:Time 2 :

vA2 (HH) = vE

2 (HH) =41

6

vA2 (HT ) = vE

2 (HT ) =17

6

vA2 (TH) = vE

2 (TH) =5

6

vA2 (TT ) = vE

2 (TT ) = 0.

(88)

Time 1 :

vA1 (H) = max

3.5,

241

60

=

241

60

vA1 (T ) = max

0.5,

11

36

= 0.5

(89)



For the Americanized version:Time 0 :

vA0 = max

1.5, e−rh

(pvA

1 (H) + qvA1 (T )

)= max

1.5,

9

10

(11

27

241

60+

16

27

1

2

)= max

1.5,

3131

1800

≈ 1.739 > 1.636 = vE

0 .

(90)


American Options - Up-and-In American Put

Consider a binomial model, where N = 2, and

S0 = 4, u = 2, d =1

2, r =

1

4(K ,U) = (5, 7)

(91)

where K is the usual strike price, and U is an upper barrier on the stockprice. Compute the initial price V A,P,U

0 of an option such that

This option is an American Put at the first time the stock price isgreater than, or equal to, U.

If the stock is less than the upper barrier for the term of the option,the option expires worthless.



Denote

V A,Pn as the value of a standard American Put at time n that hasn’t

been exercised before n, and

V A,P,Un the value of the option at time n that hasn’t been knocked-in

before n.

It follows that, at the end of the term,

V A,PN (s) = max K − s, 0 = max 5− s, 0

V A,P,UN (s) = V A,P

N (s)1s≥U = max 5− s, 0 1s≥7 = 0.(92)

Also note that we can compute p = q = 12 .



Working backwards,

V A,Pn (s) = max

max 5− s, 0 , 1

1 + r

(pV A,P

n+1(us) + qV A,Pn+1(ds)

)V A,P,Un (s) = 1s≥7V

A,Pn (s) + 1s<7

1

1 + r

(pV A,P,U

n+1 (us) + qV A,P,Un+1 (ds)

).

(93)



Time Stock Value V E ,Pn (s) V A,P

n (s) V A,P,Un (s)

2 16 0 0 02 4 1 1 02 1 4 4 01 8 0.4 0.4 0.41 2 2 3 00 4 0.96 1.36 0.16

HW:

Redo this example for N = 3 and N = 4.How does equation (93) change if we switch to a 2−factor modelthat tracks both stock s and running maximum m at time n? Whatis the recursion equation for the Up-and-In American Put?Calculate (V0,∆0) for a down-and-rebate option which pays 100 atthe first time when the stock price is below the barrier K = 3. If thestock price is above the barrier for all times then the option expiresworthless. Use N = 3.Albert Cohen (MSU) Financial Mathematics for Actuaries I MSU Fall 2017 89 / 164

Call Options on Zero-Coupon Bonds

Assume an economy where

One period is one year

The one year short term interest rate from time n to time n + 1 is rn.

The rate evolves via a stochastic process rnn≥0 such that

r0 = 0.02

P[rn+1 = 2rn] = P[rn+1 = rn] = P[

rn+1 =1

2rn

]=

1

3.

(94)



Consider now a zero-coupon bond that matures in 3−years withcommon face and redemption value F = 100.

Also consider a call option on this bond that expires in 2−years withstrike K = 97.

Denote Bn and Cn as the bond and call option values, respectively.

Note that we iterate backwards from the values

B3(r) = 100

C2(r) = max B2(r)− 97, 0 .(95)

Compute (B0,C0).



Our general recursive formula is

Bn(r) =1

1 + rE[Bn+1(rn+1) | rn = r ]

Cn(r) =1

1 + rE[Cn+1(rn+1) | rn = r ].

(96)

Iterating backwards, we see that at t = 2,

B2(r) =1

1 + r

1

3

1∑k=−1

B3(2k r). (97)

At time t = 2, we have that

r2 ∈ 0.08, 0.04, 0.02, 0.01, 0.005 . (98)



Our associated Bond and Call Option values at time 2:

r2 B2 C2

0.08 92.59 00.04 96.15 00.02 98.04 1.040.01 99.01 2.01

0.005 99.50 2.50




r1 B1 C1

0.04 91.92 0.330.02 95.82 1.000.01 97.87 1.83


r0 B0 C0

0.02 93.34 1.03

HW Question: Recalculate all bond / call option values assumingthat the delivery time of the option is now 3. Symbolically,

C3(r) = max B3(r)− 97, 0 . (99)

HW: Finan: Ex 81.1, 81.2, 81.5, 81.6, 81.7, 81.8, 81.9.


Matching Interest Rates to Market Conditions

Consider again a series of coin flips (ω1, ..., ωn) where at time n, theinterest rate from n to n + 1 is modeled via

rn = rn(ω1, ..., ωn) (100)

Keep in mind that we will build a recombining binomial tree for thismodel. So, for example,

r2(H,T ) = r2(T ,H). (101)



Furthermore, we can define the yield rate y(t,T , r(t)) for a zero-couponbond B(t,T , r(t)) via

B(t,T , r(t)) =1

(1 + y(t,T , r(t)))T−t(102)

Given yield rates, there is also the corresponding yield rate volatility

σn =1

2ln

(y(1, n, r1(H))

y(1, n, r1(T ))

). (103)

We compare the yield rate volatility with the short rate volatility

σn =1

2ln

(rn(ω1, ..., ωn−1, ωn = H)

rn(ω1, ..., ωn−1, ωn = T )

). (104)



We can in fact see that

y(1, 2, r1(H)) = r1(H)

y(1, 2, r1(T )) = r1(T ).(105)

But, for example, it is not clear how to obtain

(y(1, 3, r1(H)), y(1, 3, r1(T ))) . (106)

Furthermore, how can we match to market conditions and update ourestimates for interest rates rn?



One way forward is the Black Derman Toy Model 1

Consider a market with the following observations:

Maturity Yield to Maturity y(0,T , r) Yield Volatility σT1 0.100 N.A.2 0.110 0.1903 0.120 0.1804 0.125 0.1505 0.130 0.140

(For Daily US Treasury Real Yield Curve Rates click here )

1Black, Fischer, Emanuel Derman, and William Toy. ”A one-factor modelof interest rates and its application to treasury bond options.” FinancialAnalysts Journal 46.1 (1990): 33-39.


https://www.treasury.gov/resource-center/data-chart-center/interest-rates/Pages/TextView.aspx?data=realyield


In this setting, we match market conditions to a model where at eachtime, an interest rate moves up or down with a ”risk-neutral”probability of 1

2 .

It follows that we have from time t = 0 to t = 1, with an initial rater0 = r ,

1

1 + y(0, 1, r)=

1

1 + r

⇒ y(0, 1, r) = r .

(107)



From time t = 0 to t = 2, again with our initial rate r0 = r ,

Connecting our observed two-year yield with yearly interest ratesreturns

1

(1 + y(0, 2, r))2=

1

1 + r

(1

2

1

1 + r1(H)+

1

2

1

1 + r1(T )

)⇒ 1

1.112=

1

1.10

(1

2

1

1 + r1(H)+

1

2

1

1 + r1(T )

).

(108)

Also, connecting our one-year yields with yearly interest rates leads to

σ1 =1

2ln

(r1(H)

r1(T )

)=

1

2ln

(y(1, 2, r1(H))

y(1, 2, r1(T ))

)= σ2 = 0.190.

(109)

Solution leads to the pair

(r1(H), r1(T )) = (0.1432, 0.0979). (110)



From time t = 0 to t = 3, again with our initial rate r0 = r , we try toestimate the matching (r2(H,H), r2(H,T ), r2(T ,T )).

Connecting our observed two-year yield with yearly interest ratesreturns

1

(1 + y(0, 3, r))3=

1

1 + r

(1

4

1

1 + r1(H)

1

1 + r2(H,H)

)+

1

1 + r

(1

4

1

1 + r1(H)

1

1 + r2(H,T )

)+

1

1 + r

(1

4

1

1 + r1(T )

1

1 + r2(T ,H)

)+

1

1 + r

(1

4

1

1 + r1(T )

1

1 + r2(T ,T )

).

(111)

We can now substitute our values(r , r1(H), r1(T ), y(0, 3, r)) = (0.10, 0.1432, 0.0979, 0.120).



Our model will also assume that σ2 6= σ2(ω1, ω2), i.e. short rate volatilityis only time-dependent, and so short rate volatility is not dependent onstate:

σ2 =1

2ln

(r2(H,H)

r2(H,T )

)σ2 =

1

2ln

(r2(T ,H)

r2(T ,T )

)⇒ r2(H,T ) = r2(T ,H) =

√r2(H,H)r2(T ,T ).

(112)



Finally, we have one more matching condition via

0.180 = σ3 =1

2ln

(y(1, 3, r1(H))

y(1, 3, r1(T ))

)

=1

2ln

1√

11+r1(H)

(12

11+r2(H,H)

+ 12

11+r2(H,T )

) − 1

1√1

1+r1(T )

(12

11+r2(T ,H)

+ 12

11+r2(T ,T )

) − 1

(113)

where

1

(1 + y(1, 3, r1(H))2=

1

1 + r1(H)

(1

2

1

1 + r2(H,H)+

1

2

1

1 + r2(H,T )

)1

(1 + y(1, 3, r1(T ))2=

1

1 + r1(T )

(1

2

1

1 + r2(T ,H)+

1

2

1

1 + r2(T ,T )

).

(114)Now solve for (r2(H,H), r2(H,T ), r2(T ,T ))!!


Pricing a Two-year Coupon Bond Using Market Conditions

Consider the previous observations for term structure, and a bond withF = 100 and coupon rate 10%. In this setting, we have

Time Interest Rate Value

0 r0 0.1001 r1(H) 0.14321 r1(T ) 0.09792 r2(H,H) 0.19422 r2(H,T ) 0.13772 r2(T ,H) 0.13772 r2(T ,T ) 0.0976



We can decompose the two-year coupon bond into two componentzero-coupon bonds.

The first is B(1), which has face 10, maturity T = 1, and initial priceB(1)(0,T , r) at t = 0.

The first is B(2), which has face 110, maturity T = 2, and initial priceB(2)(0,T , r) at t = 0.

The total coupon bond price is thus

B(0, 2, r) = B(1)(0, 1, r) + B(2)(0, 2, r). (115)



Our component bonds thus have prices

Time Interest Rate B1(t,T , r)

0 r0 = 0.10 B(1)(0, 1, r0) = 9.09

1 r1(H) = 0.1432 B(1)(1, 1, r1(H)) = 10

1 r1(T ) = 0.0979 B(1)(1, 1, r1(T )) = 10

Time Interest Rate B2(t,T , r)

0 r0 = 0.10 B(2)(0, 2, r0) = 89.28

1 r1(H) = 0.1432 B(2)(1, 2, r1(H)) = 96.22

1 r1(T ) = 0.0979 B(2)(1, 2, r1(T )) = 100.19

2 r2(H,H) = 0.1942 B(2)(2, 2, r2(H,H)) = 110

2 r2(H,T ) = 0.1377 B(2)(2, 2, r2(H,T )) = 110

2 r2(T ,H) = 0.1377 B(2)(2, 2, r2(T ,H)) = 110

2 r2(T ,T ) = 0.0976 B(2)(2, 2, r2(T ,T )) = 110



Finally, we have our coupon bond with price

Time Interest Rate B(t,T , r)

0 r0 = 0.10 B(0, 2, r0) = 98.371 r1(H) = 0.1432 B(1, 2, r1(H)) = 106.221 r1(T ) = 0.0979 B(1, 2, r1(T )) = 110.192 r2(H,H) = 0.1942 B(2, 2, r2(H,H)) = 1102 r2(H,T ) = 0.1377 B(2, 2, r2(H,T )) = 1102 r2(T ,H) = 0.1377 B(2, 2, r2(T ,H)) = 1102 r2(T ,T ) = 0.0976 B(2, 2, r2(T ,T )) = 110

Note: Note that with our coupons, the yield yc(0, 2, r) = 0.1095, which isobtained by solving

98.37 =10

1 + yc(0, 2, r)+

110

(1 + yc(0, 2, r))2. (116)


Pricing a 1-year Call Option on our 2-year Coupon Bond

Consider the previous term structure, and a European Call Option on thetwo year bond with K = 97 and expiry of 1 year.

Compute the initial Call price C0(r).

Compute the initial number of bonds to hold to replicate this option.


Pricing a 1-year Call Option on our 2-year Coupon Bond

In this case, we look at the price of the bond minus the accrued interest:

Time Interest Rate B(t,T , r)

0 r0 = 0.10 B(0, 2, r0) = 98.371 r1(H) = 0.1432 B(1, 2, r1(H)) = 96.221 r1(T ) = 0.0979 B(1, 2, r1(T )) = 100.192 r2(H,H) = 0.1942 B(2, 2, r2(H,H)) = 1002 r2(H,T ) = 0.1377 B(2, 2, r2(H,T )) = 1002 r2(T ,H) = 0.1377 B(2, 2, r2(T ,H)) = 1002 r2(T ,T ) = 0.0976 B(2, 2, r2(T ,T )) = 100

Pricing and hedging is accomplished via

C0(0.10) =1

1.10

(1

2· 0 +

1

23.19

)= 1.45

∆0(0.10) =0− 3.19

96.22− 100.19= 0.804.

(117)


Homework

From Finan:

Problems 81.1, 81.2, 81.5, 81.6, 81.7, 81.8, 81.9.

Problems 82.1, 82.2, 82.3, 82.4, 82.5

Problems 82.6, 82.7, 82.8, 82.9, 82.10, 82.11.


Lognormality

Analyzing returns, we assume:

A probability space(

Ω,F ,P).

Our asset St(ω) has an associated return over any period (t, t + u)defined as

rt,u(ω) := ln

(St+u(ω)

St(ω)

). (118)


Lognormality

Partition the interval [t,T ] into n intervals of length h = T−tn , then:

The return over the entire period can be taken as the sum of thereturns over each interval:

rt,T−t(ω) = ln

(ST (ω)

St(ω)

)=

n∑k=1

rtk ,h(ω)

tk = t + kh.

(119)

We model the returns as being independent and possessing a binomialdistribution.

Employing the Central Limit Theorem, it can be shown that asn→∞, this distribution approaches normality.


Binomial Tree and Discrete Dividends

Another issue encountered in elementary credit and investment theoryis the case of different compounding and deposit periods.

This also occurs in the financial setting where a dividend is not paidcontinuously, but rather at specific times.

It follows that the dividend can be modeled as delivered in the middleof a binomial period, at time τ(ω) < T .

This view is due to Schroder and can be summarized as viewing theinherent value of St(ω) as the sum of a prepaid forward PF and thepresent value of the upcoming dividend payment D:

PFt(ω) = St(ω)− De−r(τ(ω)−t)

u = erh+σ√h

d = erh−σ√h.

(120)

Now, the random process that we model as having up and downmoves is PF instead of S .


Back to the Continuous Time Case

Consider the case of a security whose binomial evolution is modeled as anup or down movement at the end of each day. Over the period of oneyear, this amounts to a tree with depth 365. If the tree is not recombining,then this amounts to 2365 branches. Clearly, this is too large to evaluatereasonably, and so an alternative is sought.

Whatever the alternative, the concept of replication must hold. This isthe reasoning behind the famous Black-Scholes-Merton PDE approach.


Back to the Continuous Time Case

Consider the case of a security whose binomial evolution is modeled as anup or down movement at the end of each day. Over the period of oneyear, this amounts to a tree with depth 365. If the tree is not recombining,then this amounts to 2365 branches. Clearly, this is too large to evaluatereasonably, and so an alternative is sought.

Whatever the alternative, the concept of replication must hold. This isthe reasoning behind the famous Black-Scholes-Merton PDE approach.


Monte Carlo Techniques

Our model for asset evolution is

⇒ St = S0e(α−δ− 12σ2)t+σ

√tZ

Z ∼ N(0, 1).(121)

Consider now the possibility of simulating the stock evolution bysimulating the random variable Z , or in fact an i.i.d. sequence

Z (i)

ni=1

.

For a European option with time expiry T , we can simulate the expirytime payoff mulitple times:

V(

S (i),T)

= G(

S (i))

= G(

S0e(α−δ− 12σ2)T+σ

√TZ (i)

). (122)




⇒ St = S0e(α−δ− 12σ2)t+σ

√tZ

Z ∼ N(0, 1).(121)


Z (i)

ni=1

.


V(

S (i),T)

= G(

S (i))

= G(


√TZ (i)

). (122)




⇒ St = S0e(α−δ− 12σ2)t+σ

√tZ

Z ∼ N(0, 1).(121)


Z (i)

ni=1

.


V(

S (i),T)

= G(

S (i))

= G(


√TZ (i)

). (122)



If we sample uniformly from our simulated values

V(

S (i),T)n

i=1we

can appeal to a sampling-convergence theorem with the appoximation

V (S , 0) = e−rT1

n

n∑i=1

V(

S (i),T). (123)

The challenge now is to simulate our lognormally distributed assetevolution.

One can simulate the value ST directly by one random variable Z , or amultiple of them to simulate the path of the evolution until T .

The latter method is necessary for Asian options.




V(

S (i),T)n

i=1we


V (S , 0) = e−rT1

n

n∑i=1

V(

S (i),T). (123)







V(

S (i),T)n

i=1we


V (S , 0) = e−rT1

n

n∑i=1

V(

S (i),T). (123)







V(

S (i),T)n

i=1we


V (S , 0) = e−rT1

n

n∑i=1

V(

S (i),T). (123)






There are multiple ways to simulate Z . One way is to find a randomnumber U taken from a uniform distribution U[0, 1].

It follows that one can now map U → Z via inversion of the Normal cdf N:

Z = N−1(U). (124)

It can be shown that in a sample, the standard deviation of the sampleaverage σsample is related to the standard deviation of an individual drawvia

σsample =σdraw√

n. (125)

If σdraw = σ, then we can see that we must increase our sample size by22k if we wish to cut our σsample by a factor of 2k .





Z = N−1(U). (124)


σsample =σdraw√

n. (125)






Z = N−1(U). (124)


σsample =σdraw√

n. (125)



Calibration Exercise

Assume table below of realized gains & losses over a ten-period cycle.

Use the adjusted values (r , δ, h,S0,K ) = (0.02, 0, 0.10, 10, 10).

Calculate binary options from last slide using these assumptions.

Period Return

1 S1S0

= 1.05

2 S2S1

= 1.02

3 S3S2

= 0.98

4 S4S3

= 1.01

5 S5S4

= 1.02

6 S6S5

= 0.99

7 S7S6

= 1.03

8 S8S7

= 1.05

9 S9S8

= 0.96

10 S10S9

= 0.97


Calibration Exercise: Linear Approximation

We would like to compute σ for the logarithm of returns ln(

SiSi−1

).

Assume the returns per period are all independent.

Q: Can we use a linear (simple) return model instead of a compoundreturn model as an approximation?

If so, then for our observed simple return rate values:

Calculate the sample variance σ2∗.

Estimate that σ ≈ σ∗√h

.



Note that if SiSi−1

= 1 + γ for γ 1, then

ln

(Si

Si−1

)≈ γ =

Si − Si−1

Si−1. (126)

Approximation: Convert our previous table, using simple interest.

Over small time periods h, define linear return values for i th period:

Xih :=Si − Si−1

Si−1. (127)

In other words, for simple rate of return Xi for period i :

Si = Si−1 · (1 + Xih). (128)



Our returns table now looks like

Period Return

1 S1−S0S0

= 0.05

2 S2−S1S1

= 0.02

3 S3−S2S2

= −0.02

4 S4−S3S3

= 0.01

5 S5−S4S4

= 0.02

6 S6−S5S5

= −0.01

7 S7−S6S6

= 0.03

8 S8−S7S7

= 0.05

9 S9−S8S8

= −0.04

10 S10−S9S9

= −0.03

sample standard deviation σ∗ = 0.0319estimated return deviation σ ≈ 0.1001



We estimate, therefore, that under the Futures-Cox model

(u, d) = (e0.0319, e−0.0319) = (1.0324, 0.9686)

p =e0.002 − e−0.0319

e0.0319 − e−0.0319= 0.5234.

(129)



For the one-period digital option:

V0 = e−rhE0[1S1≥10] = e−0.002 · p = 0.5224. (130)

For the two-period digital option:

V0 = e−2rhE0[1S2≥10] = e−0.004 ·[p2 + 2pq

]= 0.7698. (131)




V0 = e−rhE0[1S1≥10] = e−0.002 · p = 0.5224. (130)


V0 = e−2rhE0[1S2≥10] = e−0.004 ·[p2 + 2pq

]= 0.7698. (131)


Calibration Exercise: No Approximation

Without the linear approximation, we can directly estimate

σY√

h = 0.03172

(u, d) = (e0.03172, e−0.03172) = (1.0322, 0.9688)

p =e0.002 − e−0.03172

e0.03172 − e−0.03172= 0.5246.

(132)


V0 = e−rhE0[1S1≥10] = e−0.002 · p = 0.5236. (133)


V0 = e−2rhE0[1S2≥10] = e−0.004 ·[p2 + 2pq

]= 0.7721. (134)


Calibration Exercise: No Approximation

Without the linear approximation, we can directly estimate

σY√

h = 0.03172

(u, d) = (e0.03172, e−0.03172) = (1.0322, 0.9688)

p =e0.002 − e−0.03172

e0.03172 − e−0.03172= 0.5246.

(132)


V0 = e−rhE0[1S1≥10] = e−0.002 · p = 0.5236. (133)


V0 = e−2rhE0[1S2≥10] = e−0.004 ·[p2 + 2pq

]= 0.7721. (134)


Capital Structure Model

As an analyst for an investments firm, you are tasked with advisingwhether a company’s stock and/or bonds are over/under-priced.

You receive a quarterly report from this company on it’s return onassets, and have compiled a table for the last ten quarters below.

Today, just after the last quarter’s report was issued, you see that inbillions of USD, the value of the company’s assets is 10.

There are presently one billion shares of this company that are beingtraded.

The company does not pay any dividends.

Six months from now, the company is required to pay off a billionzero-coupon bonds. Each bond has a face value of 9.5.



Assume Miller-Modigliani holds with At = Bt + St , where the assetsof a company equal the sum of its share and bond price.

Presently, the market values are (B0, S0, r) = (9, 1, 0.02).

The Merton model for corporate bond pricing asserts that atredemption time T ,

Bt = e−r(T−t)E [min AT ,F]St = e−r(T−t)E [max AT − F , 0] .

(135)

With all of this information, your job now is to issue a Buy or Sell onthe stock and the bond issued by this company.



Table of return on assets for Company X, with h = 0.25.

Period Return on Assets

1 A1−A0A0

= 0.05

2 A2−A1A1

= 0.02

3 A3−A2A2

= −0.02

4 A4−A3A3

= 0.01

5 A5−A4A4

= 0.02

6 A6−A5A5

= −0.01

7 A7−A6A6

= 0.03

8 A8−A7A7

= 0.05

9 A9−A8A8

= −0.04

10 A10−A9A9

= −0.03

sample standard deviation σ∗ = 0.0319estimated return deviation σ ≈ 0.0638



We scale all of our calculation in terms of billions ($, shares, bonds).

Using the Futures- Cox model, we have

(u, d) = (e0.0319, e−0.0319) = (1.0324, 0.9686)

p =e0.005 − e−0.0319

e0.0319 − e−0.0319= 0.5706.

(136)

Using this model, the only time the payoff of the bond is less than theface is on the path ω = TT .

The price of the bond and stock are thus modeled to be

B0 = e−0.02·(2·0.25)[p2 · 9.5 + 2pq · 9.5 + q2 · 9.38

]= 9.38 > 9.00

S0 = 10− 9.38 = 0.62 < 1.00.

(137)

It follows that,according to our model, one should Buy the bond as itis under priced and one should Sell the stock as it is overpriced.


Optimal Investment: U(x) = ln (x): Maximized over p?

In the case of a logarithmic investor, if (p, x , r) are fixed, known terms,then it follows from (43) that

u(x ; p) = ln

(1 + r)(pp

)p (1−p1−p

)1−p x

= ln (x(1 + r)) + p ln

(p

p

)+ (1− p) ln

(1− p

1− p

).

∴∂

∂p[u(x ; p)] = ln

(p

p

)− ln

(1− p

1− p

)= 0⇒ p = p.

(138)



In the case of a logarithmic investor, if (p, x , r) are fixed, known terms,then it follows from (43) that

u(x ; p) = ln

(1 + r)(pp

)p (1−p1−p

)1−p x

= ln (x(1 + r)) + p ln

(p

p

)+ (1− p) ln

(1− p

1− p

).

∴∂

∂p[u(x ; p)] = ln

(p

p

)− ln

(1− p

1− p

)= 0⇒ p = p.

(138)



It follows that, if we define (and compute)

u(x ; 0) := limp↓0

u(x ; p) = ln (x(1 + r)) + ln

(1

1− p

)u(x ; 1) := lim

p↑1u(x ; p) = ln (x(1 + r)) + ln

(1

p

)u(x ; p) = ln (x(1 + r))

(139)

then

max0≤p≤1

u(x ; p) = max u(x ; 0), u(x ; p), u(x ; 1)

= u(x ; 0)× 1p>0.5 + u(x ; 1)× 1p≤0.5

min0≤p≤1

u(x ; p) = min u(x ; 0), u(x ; p), u(x ; 1) = u(x ; p)

= ln (x(1 + r)).

(140)

Does this make sense?


Optimal Investment: U(x) = ln (x): Maximized over P?

In general, if there are n distinct states, then the problem of logaritmicsolution has optimal portfolio value X1(ω) = x(1+r)

Z(ω) and so

u(x ; p1, .., pn) = E[

ln

(x(1 + r)

Z (ω)

)]= ln (x(1 + r))− E[ln (Z )]

= ln (x(1 + r)) +n∑

i=1

pi ln

(pi

pi

).

(141)

If we assume the pi are known, then a similar calculation as in (138) leadsto the critical physical probabilities pi = pi . This leads to a min insteadof a max solution.

Furthermore, we can see that shifting all the physical probabilities to 1 forstates i where pi < 0.5 is the maximal solution.

However, this kind of investing results in a Radon-Nikodym derivative thatis in the set 0,∞. We may need to put limits on what beliefs/physicalprobabilities we allow the investor to possess!




Z(ω) and so

u(x ; p1, .., pn) = E[

ln

(x(1 + r)

Z (ω)

)]= ln (x(1 + r))− E[ln (Z )]

= ln (x(1 + r)) +n∑

i=1

pi ln

(pi

pi

).

(141)







Z(ω) and so

u(x ; p1, .., pn) = E[

ln

(x(1 + r)

Z (ω)

)]= ln (x(1 + r))− E[ln (Z )]

= ln (x(1 + r)) +n∑

i=1

pi ln

(pi

pi

).

(141)







Z(ω) and so

u(x ; p1, .., pn) = E[

ln

(x(1 + r)

Z (ω)

)]= ln (x(1 + r))− E[ln (Z )]

= ln (x(1 + r)) +n∑

i=1

pi ln

(pi

pi

).

(141)






If, however, we are in an arbitrage-free market where we don’t necessarilyknow P, then for

I[P] := −n∑

i=1

pi ln (pi )

Ic [P, P] := −n∑

i=1

pi ln (pi )

(142)

we wish to compute

U(x) := ln (x(1 + r))− I[P] + maxP∈AP

Ic [P, P]. (143)

for some set of admissible risk-neutral measures AP. How should we dothis?


Black Scholes Pricing using Underlying Asset

In the next course, we will derive the following solutions to theBlack-Scholes PDE:

V C (S , t) = e−r(T−t)E [(ST − K )+ | St = S ]

= Se−δ(T−t)N(d1)− Ke−r(T−t)N(d2)

V P(S , t) = e−r(T−t)E [(K − ST )+ | St = S ]

= Ke−r(T−t)N(−d2)− Se−δ(T−t)N(−d1)

d1 =ln(

SK

)+ (r − δ + 1

2σ2)(T − t)

σ√

T − t

d2 = d1 − σ√

T − t

N(x) =1√2π

∫ x

−∞e−

z2

2 dz .

(144)

Notice that V C (S , t)− V P(S , t) = Se−δ(T−t) − Ke−r(T−t).Question: What underlying model of stock evolution leads to this value?How can we support such a probability measure?


Lognormal Random Variables

We say that Y ∼ LN(µ, σ) is Lognormal if ln(Y ) ∼ N(µ, σ2).

As sums of normal random variables remain normal, products of lognormalrandom variables remain lognormal.

Recall that the moment-generating function ofX ∼ N(µ, σ2) ∼ µ+ σN(0, 1) is

MX (t) = E[etX ] = eµt+ 12σ2t2

(145)

If Y = eµ+σZ , then, it can be seen that

E[Y n] = E[enX ] = eµn+ 12σ2n2

(146)

and

fY (y) =1

σ√

2πyexp

(− (ln(y)− µ)2

2σ2

)(147)


Lognormal Random Variables

We say that Y ∼ LN(µ, σ) is Lognormal if ln(Y ) ∼ N(µ, σ2).

As sums of normal random variables remain normal, products of lognormalrandom variables remain lognormal.

Recall that the moment-generating function ofX ∼ N(µ, σ2) ∼ µ+ σN(0, 1) is

MX (t) = E[etX ] = eµt+ 12σ2t2

(145)

If Y = eµ+σZ , then, it can be seen that

E[Y n] = E[enX ] = eµn+ 12σ2n2

(146)

and

fY (y) =1

σ√

2πyexp

(− (ln(y)− µ)2

2σ2

)(147)


Stock Evolution and Lognormal Random Variables

One application of lognormal distributions is their use in modeling theevolution of asset prices S . If we assume a physical measure P with α theexpected return on the stock under the physical measure, then

ln

(St

S0

)= N

((α− δ − 1

2σ2)t, σ2t

)⇒ St = S0e(α−δ− 1

2σ2)t+σ

√tZ

(148)



We can use the previous facts to show

E[St ] = S0e(α−δ)t

P[St > K ] = N

(ln S0

K + (α− δ − 0.5σ2)t

σ√

t

).

(149)

Note that under the risk-neutral measure P, we exchange α with r , therisk-free rate:

E[St ] = S0e(r−δ)t

P[St > K ] = N(d2).(150)



We can use the previous facts to show

E[St ] = S0e(α−δ)t

P[St > K ] = N

(ln S0

K + (α− δ − 0.5σ2)t

σ√

t

).

(149)

Note that under the risk-neutral measure P, we exchange α with r , therisk-free rate:

E[St ] = S0e(r−δ)t

P[St > K ] = N(d2).(150)



Risk managers are also interested in Conditional Tail Expectations (CTE’s)of random variables:

CTEX (k) := E[X | X > k] =E[X 1X>k

]P[X > k]

. (151)



In our case,

E[St | St > K ] =

E

[S0e(α−δ− 1

2σ2)t+σ

√tZ1

S0e(α−δ− 1

2σ2)t+σ

√tZ>K

]

P[S0e(α−δ− 1

2σ2)t+σ

√tZ > K

]

= S0e(α−δ)t

N

(ln

S0K

+(α−δ+0.5σ2)t

σ√t

)

N

(ln

S0K

+(α−δ−0.5σ2)t

σ√t

)

⇒ E[St | St > K ] = S0e(r−δ)t N(d1)

N(d2)(152)



In fact, we can use this CTE framework to solve for the European Calloption price in the Black-Scholes framework, where P0[A] = P[A | S0 = S ]and

V C (S , 0) := e−rT E[(ST − K )+ | S0 = S

]= e−rT E0

[ST − K | ST > K

]· P0[ST > K ]

= e−rT E0

[ST | ST > K

]· P0[ST > K ]− Ke−rT P0[ST > K ]

= e−rTSe(r−δ)T N(d1)

N(d2)· N(d2)− Ke−rTN(d2)

= Se−δTN(d1)− Ke−rTN(d2).(153)


Black Scholes Analysis: Option Greeks

For any option price V (S , t), define its various sensitivities as follows:

∆ =∂V

∂S

Γ =∂∆

∂S=∂2V

∂S2

ν =∂V

∂σ

Θ =∂V

∂t

ρ =∂V

∂r

Ψ =∂V

∂δ.

(154)

These are known accordingly as the Option Greeks.



Straightforward partial differentiation leads to

∆C = e−δ(T−t)N(d1)

∆P = −e−δ(T−t)N(−d1)

ΓC = ΓP =e−δ(T−t)N ′(d1)

σS√

T − t

νC = νP = Se−δ(T−t)√

T − tN ′(d1)

(155)



as well as..

ρC = (T − t)Ke−r(T−t)N(d2)

ρP = −(T − t)Ke−r(T−t)N(−d2)

ΨC = −(T − t)Se−δ(T−t)N(d1)

ΨP = (T − t)Se−δ(T−t)N(−d1).

(156)

What do the signs of the Greeks tell us?

HW: Compute Θ for puts and calls.



as well as..

ρC = (T − t)Ke−r(T−t)N(d2)

ρP = −(T − t)Ke−r(T−t)N(−d2)

ΨC = −(T − t)Se−δ(T−t)N(d1)

ΨP = (T − t)Se−δ(T−t)N(−d1).

(156)

What do the signs of the Greeks tell us?

HW: Compute Θ for puts and calls.


Option Elasticity

Define

Ω(S , t) := limε→0

V (S+ε,t)−V (s,t)V (S,t)

S+ε−SS

=S

V (S , t)limε→0

V (S + ε, t)− V (s, t)

S + ε− S

=∆ · S

V (S , t).

(157)

Consequently,

ΩC (S , t) =∆C · S

V C (S , t)=

Se−δ(T−t)

Se−δ(T−t) − Ke−r(T−t)N(d2)≥ 1

ΩP(S , t) =∆P · S

V P(S , t)≤ 0.

(158)


Option Elasticity

Theorem

The volatility of an option is the option elasticity times the volatility of thestock:

σoption = σstock× | Ω | . (159)

The proof comes from Finan: Consider the strategy of hedging a portfolioof shorting an option and purchasing ∆ = ∂V

∂S shares.The initial and final values of this portfolio are

Initally: V (S(t), t)−∆(S(t), t) · S(t)

Finally: V (S(T ),T )−∆(S(t), t) · S(T )


Option Elasticity

Proof.

If this portfolio is self-financing and arbitrage-free requirement, then

er(T−t)(

V (S(t), t)−∆(S(t), t) ·S(t))

= V (S(T ),T )−∆(S(t), t) ·S(T ).

(160)It follows that for κ := er(T−t),

V (S(T ),T )− V (S(t), t)

V (S(t), t)= κ− 1 +

[S(T )− S(t)

S(t)+ 1− κ

]Ω

⇒ Var

[V (S(T ),T )− V (S(t), t)

V (S(t), t)

]= Ω2Var

[S(T )− S(t)

S(t)

]⇒ σoption = σstock× | Ω | .

(161)


Option Elasticity

If γ is the expected rate of return on an option with value V , α is theexpected rate of return on the underlying stock, and r is of course the riskfree rate, then the following equation holds:

γ · V (S , t) = α ·∆(S , t) · S + r ·(

V (S , t)−∆(S , t) · S). (162)

In terms of elasticity Ω, this reduces to

Risk Premium(Option) := γ − r = (α− r)Ω. (163)

Furthermore, we have the Sharpe Ratio for an asset as the ratio of riskpremium to volatility:

Sharpe(Stock) =(α− r)

σ=

(α− r)Ω

σΩ= Sharpe(Call). (164)

HW Sharpe Ratio for a put? How about elasticity for a portfolio ofoptions? Now read about Calendar Spreads, Implied Volatility, andPerpetual American Options.


Option Elasticity


γ · V (S , t) = α ·∆(S , t) · S + r ·(

V (S , t)−∆(S , t) · S). (162)





σ=

(α− r)Ω




Option Elasticity


γ · V (S , t) = α ·∆(S , t) · S + r ·(

V (S , t)−∆(S , t) · S). (162)





σ=

(α− r)Ω




Example: Hedging

Under a standard framework, assume you write a 4− yr European Calloption a non-dividend paying stock with the following:

S0 = 10 = K

σ = 0.2

r = 0.02.

(165)

Calculate the initial number of shares of the stock for your hedgingprogram.


Example: Hedging

Recall

∆C = e−δ(T−t)N(d1)

d1 =ln(

SK

)+ (r − δ + 1

2σ2)(T − t)

σ√

T − t

d2 = d1 − σ√

T − t.

(166)

It follows that

∆C = N(

0.4)

= 0.6554. (167)


Example: Risk Analysis

Assume that an option is written on an asset S with the followinginformation:

The expected rate of return on the underlying asset is 0.10.

The expected rate of return on a riskless asset is 0.05.

The volatility on the underlying asset is 0.20.

V (S , t) = e−0.05(10−t)(

S2eS)

Compute Ω(S , t) and the Sharpe Ratio for this option.



By definition,

Ω(S , t) =∆ · S

V (S , t)=

S · ∂V (S,t)∂S

V (S , t)

=S d

dS (S2eS)

(S2eS)=

S · (2SeS + S2eS)

S2eS

= 2 + S .

(168)

Furthermore, since Ω = 2 + S ≥ 2, we have

Sharpe =0.10− 0.05

0.20= 0.25. (169)



By definition,

Ω(S , t) =∆ · S

V (S , t)=

S · ∂V (S,t)∂S

V (S , t)

=S d

dS (S2eS)

(S2eS)=

S · (2SeS + S2eS)

S2eS

= 2 + S .

(168)

Furthermore, since Ω = 2 + S ≥ 2, we have

Sharpe =0.10− 0.05

0.20= 0.25. (169)


Example: Black Scholes Pricing

Consider a portfolio of options on a non-dividend paying stock S thatconsists of a put and a call, both with strike K = 5 = S0. What is the Γfor this option as well as the option value at time 0 if the time toexpiration is T = 4, r = 0.02, σ = 0.2.

In this case,

V = V C + V P

Γ =∂2

∂S2

(V C + V P

)= 2ΓC .

(170)



Consider a portfolio of options on a non-dividend paying stock S thatconsists of a put and a call, both with strike K = 5 = S0. What is the Γfor this option as well as the option value at time 0 if the time toexpiration is T = 4, r = 0.02, σ = 0.2.

In this case,

V = V C + V P

Γ =∂2

∂S2

(V C + V P

)= 2ΓC .

(170)



Consequently, d1 = 0.4 and d2 = 0.4− 0.2√

4 = 0, and so

V (5, 0) = V C (5, 0) + V P(5, 0)

= 5(

N(d1) + e−4rN(−d2)− e−4rN(d2)− N(−d1))

= 5(

N(0.4) + e−4rN(0)− e−4rN(0)− N(−0.4))

= 1.5542

Γ(5, 0) =2N ′(0.4)

0.2 · 5 ·√

4= N ′(0.4) =

1√2π

e−0.5·(0.4)2= 0.4322.

(171)


Market Making

On a periodic basis, a Market Maker, services the option buyer byrebalancing the portfolio designed to replicate the payoff written into theoption contract.Define

Vi = Option Value i periods from inception

∆i = Delta required i periods from inception

∴ Pi = ∆iSi − Vi

(172)

Rebalancing at time i requires an extra (∆i+1 −∆i ) shares.


Market Making

On a periodic basis, a Market Maker, services the option buyer byrebalancing the portfolio designed to replicate the payoff written into theoption contract.Define

Vi = Option Value i periods from inception

∆i = Delta required i periods from inception

∴ Pi = ∆iSi − Vi = Cost of Strategy

(173)

Rebalancing at time i requires an extra (∆i+1 −∆i ) shares.


Market Making

Define

∂Si = Si+1 − Si

∂Pi = Pi+1 − Pi

∂∆i = ∆i+1 −∆i

(174)

Then

∂Pi = Net Cash Flow = ∆i∂Si − ∂Vi − rPi

= ∆i∂Si − ∂Vi − r(

∆iSi − Vi

) (175)

Under what conditions is the Net Flow = 0?


Market Making

For a continuous rate r , we can see that if ∆ := ∂V∂S , Pt = ∆tSt − Vt ,

dV := V (St + dSt , t + dt)− V (St , t)

≈ Θdt + ∆ · dSt +1

2Γ · (dSt)

2

⇒ dPt = ∆tdSt − dVt − rPtdt

≈ ∆tdSt −(

Θdt + ∆ · dSt +1

2Γ · (dSt)

2

)− r (∆tSt − Vt) dt

≈ −

(Θdt + r(∆St − V (St , t))dt +

1

2Γ · [dSt ]

2

).

(176)


Market Making

If dt is small, but not infinitessimally small, then on a periodic basis giventhe evolution of St , the periodic jump in value from St → St + dSt may beknown exactly and correspond to a non-zero jump in Market Maker profitdPt .

If dSt · dSt = σ2S2t dt, then if we sample continuously and enforce a zero

net-flow, we retain the BSM PDE for all relevant (S , t):

∂V

∂t+ r(

S∂V

∂S− V

)+

1

2σ2S2∂

2V

∂S2= 0

V (S ,T ) = G (S) for final time payoff G (S).

(177)


Market Making

If dt is small, but not infinitessimally small, then on a periodic basis giventhe evolution of St , the periodic jump in value from St → St + dSt may beknown exactly and correspond to a non-zero jump in Market Maker profitdPt .

If dSt · dSt = σ2S2t dt, then if we sample continuously and enforce a zero

net-flow, we retain the BSM PDE for all relevant (S , t):

∂V

∂t+ r(

S∂V

∂S− V

)+

1

2σ2S2∂

2V

∂S2= 0

V (S ,T ) = G (S) for final time payoff G (S).

(177)


Note: Delta-Gamma Neutrality vs Bond Immunization

In an actuarial analysis of cashflow, a company may wish to immunizeits portfolio. This refers to the relationship between a non-zero valuefor the second derivative with respect to interest rate of the(deterministic) cashflow present value and the subsequent possibilityof a negative PV.

This is similar to the case of market maker with a non-zero Gamma.In the market makers cash flow, a move of dS in the stockcorresponds to a move 1

2 Γ(dS)2 in the portfolio value.

In order to protect against large swings in the stock causing non-lineareffects in the portfolio value, the market maker may choose to offsetpositions in her present holdings to maintain Gamma Neutrality or shewish to maintain Delta Neutrality, although this is only a linear effect.


Option Greeks and Analysis - Some Final Comments

It is important to note the similarities between Market Making andActuarial Reserving. In engineering the portfolio to replicate thepayoff written into the contract, the market maker requires capital.

The idea of Black Scholes Merton pricing is that the portfolio shouldbe self-financing.

One should consider how this compares with the capital required byinsurers to maintain solvency as well as the possibility of obtainingreinsurance.


Exam Practice

Consider an economy where :

The current exchange rate is x0 = 0.011 dollaryen .

A four-year dollar-denominated European put option on yen with astrike price of 0.008$ sells for 0.0005$.

The continuously compounded risk-free interest rate on dollars is 3%.

The continuously compounded risk-free interest rate on yen is 1.5%.

Compute the price of a 4−year dollar-denominated European call optionon yens with a strike price of 0.008$.

ANSWER: By put call parity, and the Black Scholes formula, with theasset S as the exchange rate, and the foreign risk-free rate rf = δ,

V C (x0, 0) = V P(x0, 0) + x0e−rf T − Ke−rT

= 0.0005 + 0.011e−0.015·4 − 0.008e−0.03·4

= 0.003764.

(178)


Exam Practice

Consider an economy where :

The current exchange rate is x0 = 0.011 dollaryen .

A four-year dollar-denominated European put option on yen with astrike price of 0.008$ sells for 0.0005$.

The continuously compounded risk-free interest rate on dollars is 3%.

The continuously compounded risk-free interest rate on yen is 1.5%.

Compute the price of a 4−year dollar-denominated European call optionon yens with a strike price of 0.008$.ANSWER: By put call parity, and the Black Scholes formula, with theasset S as the exchange rate, and the foreign risk-free rate rf = δ,

V C (x0, 0) = V P(x0, 0) + x0e−rf T − Ke−rT

= 0.0005 + 0.011e−0.015·4 − 0.008e−0.03·4

= 0.003764.

(178)


Exam Practice

An investor purchases a 1−year, 50− strike European Call option ona non-dividend paying stock by borrowing at the risk-free rate r .

The investor paid V C (S0, 0) = 10.

Six months later, the investor finds out that the Call option hasincreased in value by one: V C (S0.05, 0.5) = 11.

Assume (σ, r) = (0.2, 0.02).

Should she close out her position after 6 months?

ANSWER: Simply put, her profit if she closes out after 6 months is

11− 10e0.02 12 = 0.8995. (179)

So, yes, she should liquidate her position.


Exam Practice

An investor purchases a 1−year, 50− strike European Call option ona non-dividend paying stock by borrowing at the risk-free rate r .

The investor paid V C (S0, 0) = 10.

Six months later, the investor finds out that the Call option hasincreased in value by one: V C (S0.05, 0.5) = 11.

Assume (σ, r) = (0.2, 0.02).

Should she close out her position after 6 months?

ANSWER: Simply put, her profit if she closes out after 6 months is

11− 10e0.02 12 = 0.8995. (179)

So, yes, she should liquidate her position.


Exam Practice

Consider a 1−year at the money European Call option on anon-dividend paying stock.

You are told that ∆C = 0.65, and the economy bears a 1% rate.

Can you estimate the volatility σ?

ANSWER: By definition,

∆C = e−δTN(d1) = N( r + 1

2σ2

σ

)= N

(0.01 + 12σ

2

σ

)= 0.65

⇒0.01 + 1

2σ2

σ= 0.385

⇒ σ ∈ 0.0269, 0.7431 .

(180)

More information is needed to choose from the two roots computed above.


Exam Practice





∆C = e−δTN(d1) = N( r + 1

2σ2

σ

)= N

(0.01 + 12σ

2

σ

)= 0.65

⇒0.01 + 1

2σ2

σ= 0.385

⇒ σ ∈ 0.0269, 0.7431 .

(180)



Exam Practice





∆C = e−δTN(d1) = N( r + 1

2σ2

σ

)= N

(0.01 + 12σ

2

σ

)= 0.65

⇒0.01 + 1

2σ2

σ= 0.385

⇒ σ ∈ 0.0269, 0.7431 .

(180)



Exam Practice





∆C = e−δTN(d1) = N( r + 1

2σ2

σ

)= N

(0.01 + 12σ

2

σ

)= 0.65

⇒0.01 + 1

2σ2

σ= 0.385

⇒ σ ∈ 0.0269, 0.7431 .

(180)



Exam Pointers

When reviewing the material for the exam, consider the followingmilestones and examples:

The definition of the Black-Scholes pricing formulae for Europeanputs and calls.

What are the Greeks? Given a specific option, could you compute theGreeks?

What is the Option Elasticity? How is it useful? How about theSharpe ratio of an option? Can you compute the Elasticity andSharpe ration of a given option?


Exam Pointers

What is Delta Hedging?

If the Delta and Gamma values of an option are known, can youcalculate the change in option value given a small change in theunderlying asset value?

How does this correspond the Market Maker’s profit?


Documents

MATH 361: Financial Mathematics for Actuaries Iusers.math.msu.edu/users/albert/math361fall17.pdf · 2017-11-08 · MATH 361: Financial Mathematics for Actuaries I Albert Cohen Actuarial