Tải bản đầy đủ

6 The Efficient Market Hypothesis (EMH), A Guide To Modeling Prices

530

OPTIONS

One thing to keep in mind here is the difference between derivatives pricing

and looking for the best process to describe the underlying stock price process.

All that matters for derivatives pricing is that there exist an equivalent martingale

measure under which the discounted underlying price process, while not

necessarily already a martingale, becomes a martingale.

Derivatives pricing does not require that the discounted underlying stock price

process actually be a martingale under the physical probability measure. The

physical probability measure is the measure that deﬁnes actual prices, not risk neutral

prices. The question that naturally arises is, which discounted underlying price

processes admit at least one EMM? What do these processes look like?

One seemingly easy way to answer this question is simply to hypothesize

that original, actual discounted underlying price processes are already

martingales under the physical probability measure! This is a very nice trick, because

in this case we would already have an EMM. (Provided we can describe its

probability structure, which may still be quite a challenge, we could implement

it.) If the above was literally true, then the no-arbitrage condition would also

automatically hold.

If discounted prices were already martingales under the physical probability

measure, then an EMM already exists and therefore there can be no-arbitrage,

due to FTAP1.

We could use that EMM to price contingent claims as well. The only

question that would remain is uniqueness of the EMM or, what is equivalent,

market completeness.

What this all amounts to in the context of the BOPM, Nу1 is that p′=p,

which may be a fairly remote possibility. The BOPM also assumes that stock

prices are well described by a Binomial process.

Turning to the empirical side, the early empirical literature focused on

something very close to this hypothesis, which we will call the martingale

hypothesis for the physical probability measure. It wasn’t exactly this hypothesis

because actual prices were not discounted. However, for the short intervals

of time over which price changes were measured, this would make little

difference since the discount factor is small.

Commodity prices and stock prices have been intensively studied for a very

long time. Many of the early empirical studies found little or no signiﬁcant

correlation between past changes in stock prices and subsequent price changes. This

ﬁnding was also conﬁrmed for futures prices. This suggested that empirical

stock price processes meet the condition of ‘independent increments’.

The ﬁrst model of prices consistent with the empirical ﬁndings was the

random walk model or what we will call, in continuous time, the arithmetic

EQUIVALENT MARTINGALE MEASURES

531

Brownian motion model (ABM). We will be discussing this model in detail in

Chapter 16.

Later, it was found that independence was too strong a condition, because

the results of the empirical tests seemed only to establish the weaker uncorrelated

increments property (see the Appendix, section 15.7, for this distinction).

This is when martingales entered the picture, because Samuelson (1965)

produced his theoretical paper explaining the empirical ﬁndings that many

researchers had observed up to that time.

As established in the Appendix, section 15.7, if the underlying martingale

process has ﬁnite ﬁrst and second moments, then martingales will have

uncorrelated increments. Therefore, martingales with ﬁnite ﬁrst and second

moments are consistent with the empirical evidence.

After this initial research, a vast amount of further research formulated and

tested the efficient market hypothesis (EMH). There are several variations of this

model, namely weak, semi-strong, and strong form efﬁciency.

The EMH relates the expectations of future prices to information sets.

Let IN denote the information available at time N for N=0,1,2,…, and note

that IN is an evolving information set in the sense that, I0⊆I1⊆I2⊆I3⊆… .

This just means that as time evolves, the currently available information

set—which includes all previous information—gets bigger, or at least it doesn’t

get smaller. Previously available information is also currently available.

Now we can state the EMH in its three forms in terms of martingales. We

are standing at time t=N and trying to predict the random stock price one

period ahead, at time t=N+1. We have available the information set at time

t=N, IN. What can we say?

According to the EMH, the best predictor of future stock prices is the current

price SN . The EMH is just the martingale hypothesis, where the conditioning

variable is information itself. The EMH says that the sequence of prices is a

martingale with respect to IN. That is,

E(SN+1)|IN)=SN for all N=0,1,2,3,…

The currently available information set, IN , can be in at least one of three

forms:

1. Weak-Form EMH, IN={all historical stock prices S0, S1, S2,…SN}.

2. Semi-Strong Form EMH, IN={all historical stock prices and other publicly

available information, such as accounting statements as of time t=N}.

532

OPTIONS

3. Strong-Form EMH, IN={all publicly available information and also all

private information, for example, information held by insiders}.

There are a number of interesting features of the EMH. First, it ties price

formation in with information and information ﬂows. By analyzing the nature

of information ﬂows, one can say quite a bit about the structure of equilibrium

prices. For example, are price paths continuous, or do they have predictable

or unpredictable jumps?

Second, we know that square-integrable martingales have increments that

are uncorrelated (see section 15.7). This means that past information IN is useless

in predicting future prices or even subsequent price changes.

That’s why martingales predict using the current value, SN . That’s also

why martingales can do no better in predicting than the current market price.

There are no trends in prices, martingales are driftless. Any kind of trend or

correlation could be used to forecast prices and could lead to arbitrage

opportunities. Thus the martingale formulation is one that is consistent with

no-arbitrage.

All forms of the EMH conclude that it is effectively impossible to predict

future price changes from past information. The different forms of the EMH

specify what information one has access to in attempting to predict.

It is pretty clear that if a market is informationally efficient in any reasonable

sense, then there exists a positive, linear pricing mechanism (this would be

implied by efﬁciency). Then, the no-arbitrage condition must hold by the

FTAP1. That is, there can be no-arbitrage opportunities in an informationally

efﬁcient market.

Another interesting aspect of the EMH is that it says that one need not

look far and wide for at least one EMM. If actual prices form a martingale,

and discounting has minimal effect on the actual prices, then the most obvious

example of an EMM is the actual price process! The actual price process is

equivalent to itself. Of course, this can only happen if there is no-arbitrage.

What could stand in the way of the actual price process being an EMM?

The risk-free rate r >0 is one candidate. Discounting prices by the numeraire

eliminates that part of drift caused by the risk-free rate and thereby makes the

discounted price process (but not the actual price process) a martingale.

If there is an empirically measurable risk premium in stock prices, then actual

prices would form a sub-martingale, not a martingale. A risk premium in stock

prices would not create arbitrage, but rather is a reward for taking on nondiversiﬁable risk(s). If these are the only two factors contributing to positive

EQUIVALENT MARTINGALE MEASURES

533

drift, which is not consistent with martingale behavior, then by de-trending

the price series, one should be left with a martingale.

Even if actual (de-trended) prices do not form a martingale, the EMH

suggests a very interesting hypothesis. That hypothesis is that the actual price

process cannot be very far from being a martingale, as long as no-arbitrage

holds. Because, if there is no-arbitrage, we know that there is at least one

EMM for the discounted price process.

So the question is, if no-arbitrage holds, then do actual price processes (while

not necessarily martingales themselves) have a martingale component?

This takes us into the research literature. Instead of going that route, we

will examine some actual continuous time price processes that are consistent

with the EMH, and for which we can price options and therefore, for which

there is at least one EMM for the discounted price process (FTAP1). We will

also look for martingale components in the actual price processes. Chapter 16

begins this program.

15.7 APPENDIX: ESSENTIAL MARTINGALE PROPERTIES

Here we collect a few of the many properties of martingales that are used in

proving results that make martingales useful in applied ﬁnance. We restrict

attention to discrete-time martingales and sometimes even choose N=2. No

attempt at mathematical rigor is claimed. The intuition behind these results

is the primary concern.

We start with a discrete-time stochastic process (Xn()n=0,1,2,3,… with ﬁnite

ﬁrst and second moments E(Xn )<∞ and E(Xn2 )<∞ for all n=0,1,2,3,… and the

martingale property,

E(Xn+1()|Xn)=Xn for all n=0,1,2,3,…

(MP1)

1. Tower Property (TP)

E(X2|X0)=E{E(X2|X1)|X0}

(TP)

This is a general property of conditional expectations and doesn’t require

a martingale. The Tower Property says that, whenever we condition a

martingale process by an event earlier than the immediately preceding one,

like X0, we can break this down into a two-step process. First, condition by

the immediately preceding event. Then condition that conditional expectation

by the earlier one.

534

OPTIONS

One implication of TP is that the martingale property MP also holds for

all earlier events. Let’s see. Suppose (Xn())n=0,1,2 is a martingale.

E(X2|X0)=E{E(X2|X1)|X0}

=E{X1|X0}

=X0.

Here we used the martingale property twice. First, E(X2|X1)=X1. Second,

E(X1|X0)=X0. The TP allows us to rephrase the martingale property as,

E(Xn+1()|Xm)=Xm for all mр n=0,1,2,3,…

(MP2)

Sometimes you will see the (MP1) property stated as (MP2). By the tower

property, (MP2) follows from (MP1).

2. Double Expectations (DE)

This is another general property of conditional and unconditional expectations,

with important applications to martingales,

E{E(X2|X0)}=E(X2)

(DE)

This says that we ﬁrst calculate the conditional expectation of a random variable,

say X2, with respect to another random variable, say X0 . If we integrate out

the conditioning variable, X0, then we obtain the unconditional expectation

E(X2).

One immediate application of DE is to show that the means of a martingale

process are constant. Start with a martingale process (Xn()), n=0,1,2.

E{E(X2|X1)}=E(X2)

by (DE)

E(X2|X1)=X1

by (MP1)

Therefore,

E{E(X2|X1)}=E(X1).

So we conclude that E(X2)=E(X1).

EQUIVALENT MARTINGALE MEASURES

535

This justiﬁes the statement that a martingale process neither increases nor

decreases on average. The same statement is that the means of a martingale

process are constant.

3. Uncorrelated Martingale Increments (UCMI)

This is probably the most important property of martingales from the point

of view of ﬁnance. Recall that we assume that our martingales have ﬁnite

means and variances. Then the result is that the changes of a square-integrable

martingale process, while not necessarily independent, are uncorrelated.

Consider the two martingale increments (changes) X2–X1 and X1–X0. First

note that E(X2–X1)=E(X2)–E(X1)=0 because of the constant means property.

For the same reason, E(X1–X0)=E(X1)–E(X0)=0.

The deﬁnition of the covariance between two random variables X and Y

is Cov(X,Y)=E{(X–E(X))(Y–E(Y))}. In our case X=X2–X1 and Y=X1–X0 so,

for our case, Cov(X,Y)=E{XY} because E(X)=E(Y)=0.

Then,

Cov(X2–X1,X1–X0)=E{(X2–X1)(X1–X0)}

=E{X2–X1)X1}–E{(X2–X1)X0}.

We will consider each term separately. For the ﬁrst term, E{(X2–X1)X1},

E{(X2–X1)X1} =E{E{(X2–X1)X1|X1}} by (DE),

=E{E{(X2X1–X12)|X1}} by expanding all terms,

=E{E(X2X1|X1)–E(X12|X1)} by linearity of

conditional expectation,

2

=E{X1E(X2|X1)–X1 } by factoring out the constant

X1 in E(X2|X1), and noting that

E(X12|X1)=X12,

=E{X1X1–X12} by (MP1)

=E{X12–X12}

=0.

Next we consider the second term E{(X2–X1)X0},

E{(X2–X1)X0} =E{E{(X2–X1)X0|X0}} again by (DE),

=E{E{(X2X0–X1X0)|X0}} by expanding all terms,

536

OPTIONS

=E{E(X2X0|X0)–E(X1X0|X0}} by linearity of

conditional expectations,

=E{X0E(X2|X0)–X0 E(X1|X0)} by factoring out the

constant X0 in E(X1|X0),

=E{X0X0–X0X0} by (MP2)

=E{X02–X02}

=0.

Combining both terms, we have proved that Cov(X2–X1,X1–X0)=0, so that

the martingale increments X2–X1 and X1–X0 are uncorrelated.

4. Degrees of ‘Independence’: Independent, Uncorrelated, and

Orthogonal Random Variables

Understanding the EMH requires an intimate knowledge of the types of

correlation between random variables that can occur. So we collect the three

basic notions here.

A. Two random variables X() and Y() are said to be independent if

E[X()*Y()]=E[X()]*E[Y()].

B. Two random variables X() and Y() are said to be orthogonal if

E[X()*Y()]=0.

C. Two random variables X() and Y() are said to be uncorrelated if

COV(X(),Y())=E[(X()–E(X())(Y()–E(Y())]=0.

This means that, after subtracting their means, X′()=X()–E(X()) and

Y ′()=Y()–E(Y()) are orthogonal.

The strongest form of unrelatedness is independence. Then comes

orthogonality. Finally, there is uncorrelatedness. Thus, martingales enjoy a weak

form of unrelatedness between their increments. Weaker than independence

of increments, which some processes exhibit.

n

n

n

EQUIVALENT MARTINGALE MEASURES

n

1.

2.

3.

4.

5.

6.

7.

8.

9.

10.

11.

12.

13.

14.

15.

n

537

KEY CONCEPTS

Primitive Arrow–Debreu Securities and Option Pricing.

Exercise 1, Pricing B(0,1).

Exercise 2, Pricing ADu() and ADd().

Pricing a European Call Option.

Pricing any Contingent Claim.

Equivalent Martingale Measures (EMMs).

Introduction and Examples.

Deﬁnition of a Discrete-Time Martingale.

Martingales and Stock Prices.

The Equivalent Martingale Representation of Stock Prices.

The Equivalent Martingale Representation of Option Prices.

Discounted Option Prices.

Summary of the EMM Approach.

The Efﬁcient Market Hypothesis (EMH), A Guide to Modeling Prices.

Essential Martingale Properties.

END OF CHAPTER EXERCISES FOR CHAPTER 15

1. (Pricing AD securities in a Binomial model)

Make the usual assumptions of the BOPM, N=1. Suppose that u′=8%,

r′=5% and that d′=–3%. The current stock price is S0=$100. Assume that

the term to expiration is one year, so everything is already annualized.

a. Calculate P0(ADu()).

b. Calculate P0(ADd()).

2. Assume the same scenario as in exercise 1 with the same parameters.

a. Price an at-the-money European call option.

b. Price an at-the-money European put option.

3. Write out the single-period, discrete version of European Put-Call Parity

and conﬁrm that the results of your calculations conform to it.

4. Consider a generalized option with payoff function MAX[ST2 –K, 0]. Use

AD securities to price it in the usual Binomial framework.

538

n

OPTIONS

SELECTED CONCEPT CHECK SOLUTIONS

Concept Check 3

E(W1|W0)=E(W0+X1()|W0)

=E(W0|W0)+E(X1()|W0)

=W0+E(X1()|W0)

=W0+E(X1())

=W0+0

=W0.

We used the fact that X1() is independent of W0 in the fourth equality. Our

conclusion is that the property E(W1()|W0)=W0 holds.

CHAPTER 16

OPTION PRICING IN

CONTINUOUS TIME

16.1 Arithmetic Brownian Motion (ABM)

540

16.2 Shifted Arithmetic Brownian Motion

541

16.3 Pricing European Options under Shifted Arithmetic

Brownian Motion with No Drift (Bachelier)

542

16.3.1 Theory (FTAP1 and FTAP2)

542

16.3.2 Transition Density Functions

543

16.3.3 Deriving the Bachelier Option Pricing Formula 547

16.4 Defining and Pricing a Standard Numeraire

551

16.5 Geometric Brownian Motion (GBM)

553

16.5.1 GBM (Discrete Version)

553

16.5.2 Geometric Brownian Motion (GBM),

Continuous Version

559

16.6 Itô’s Lemma

562

16.7 Black–Scholes Option Pricing

566

16.7.1 Reducing GBM to an ABM with Drift

567

16.7.2 Preliminaries on Generating Unknown

Risk-Neutral Transition Density Functions

from Known Ones

570

16.7.3 Black–Scholes Options Pricing from Bachelier

571

16.7.4 Volatility Estimation in the Black–Scholes

Model

582

16.8 Non-Constant Volatility Models

585

16.8.1 Empirical Features of Volatility

585

16.8.2 Economic Reasons for why Volatility is not

Constant, the Leverage Effect

586

540

OPTIONS

16.8.3 Modeling Changing Volatility, the Deterministic

Volatility Model

586

16.8.4 Modeling Changing Volatility, Stochastic

Volatility Models

16.9 Why Black–Scholes Is Still Important

587

588

In this chapter we are going to give an introduction to continuous-time ﬁnance.

This can be a daunting and rather technical subject but we will try to cut

through all of that and stress the intuition. At the same time, we want to work

through some important examples to get the ﬂavor of the continuous-time

framework.

Many of the same ideas from the discrete-time framework we have been

discussing carry over to the continuous-time case. Risk-neutral valuation carries

over, and dynamic hedging which we discussed in the BOPM, N>1 is the

very essence of continuous-time trading and hedging.

Equivalent martingales measures (EMMs) form the foundation for the

modern approach to pricing derivatives, as we have discussed in Chapter 15.

The fundamental theorems of asset pricing, FTAP1 and FTAP2 , apply in this

continuous-time context as well.

We will begin with the prototype of all continuous time models, and that

is arithmetic Brownian motion (ABM). ABM is the most basic and important

stochastic process in continuous time and continuous space, and it has many

desirable properties including the strong Markov property, the martingale

property, independent increments, normality, and continuous sample paths.

Of course, here we want to focus on options pricing rather than the pure

mathematical theory. The idea here is to partially prepare you for courses in

mathematical ﬁnance. The details we have to leave out are usually covered

in such courses.

16.1 ARITHMETIC BROWNIAN MOTION (ABM)

ABM is a stochastic process {Wt()}tу0 deﬁned on a sample space (⍀,ℑW,℘W ).

We won’t go into all the details as to exactly what (⍀,ℑW,℘W ) represents

but you can think of the probability measure, ℘W, which is called Wiener

measure, to be deﬁned in terms of the transition density function p(T,y;t,x) for

=T–t,

## David goldenberg derivatives markets routledge (2016)

## 7 Mapping Out Spot, Forward, and Futures Prices

## 5 Profits to a Naked (Unhedged) Long Spot Position

## 2 Pricing Zero-Coupon, Unit Discount Bonds in Continuous Time

## 2 Modeling Continuous Yields: An Introduction to Non-Stochastic Differential Equations

## 8 Appendix: Modeling Stock Returns with and without Dividends

## 4 Alternative Types of Orders: Market, Market with Protection, Limit

## 2 Basic Terminology for Interest-Rate Swaps: Paying Fixed and Receiving Floating

## 6 The Dealer’s Problem: Finding the Other Side to the Swap

## 4 A Basic American Call (Put) Option Pricing Model

## 6 Going Beyond the Basic Definitions: Infrastructure to Understand Puts and Calls

Tài liệu liên quan