Path integrals for option pricing Theory of Quantum and Complex systems Statistical modeling, financial data analysis and applications Venice, 11-14 september.

Path integrals for option pricingTheory of

Quantum andComplex systems

Statistical modeling, financial data analysis and applicationsVenice, 11-14 september 2013

From left to right: dr. Kai Ji, Maarten Baeten, dr. Serghei Klimin, Stijn Ceuppens, Dries Sels, prof. Jacques Tempere, Ben Anthonis, prof. Michiel Wouters, dr. Jeroen Devreese, Enya Vermeyen, Giovanni Lombardi, Selma Koghee, dr. Onur Umucalilar, Nick Van den Broeck.Not shown: prof.em. Jozef Devreese, prof.em. Fons Brosens, dr. Vladimir Gladilin, dr. Wim Casteels

January 2013

Financial support by the Fund for Scientific Research-Flanders

J. Tempere, S.N. Klimin, J.T. Devreese, TQC, Universiteit Antwerpen

Part I: path integrals in quantum mechanics

AB

Two alternatives: add the amplitudes

1

2

Introduction: quantum mechanics with path integrals

AB

Many alternatives: add the amplitudes

Introduction: quantum mechanics with path integrals

x

AB

Many alternatives: add the amplitudes

Introduction: quantum mechanics with path integrals

x

t

AB

Many alternatives: add the amplitudes

Introduction: quantum mechanics with path integrals

x

t

x(t)The amplitude corresponding to a given path x(t) is

Here, S is the action functional:

With L the Lagrangian, eg.

H. Kleinert, Path Integrals in Quantum Mechanics, Statistics, Polymer Physics, and Financial Markets , 5th ed. (World Scientific, Singapore, 2009)

is called the path integral propagator

Part II: Path integrals in finance: option pricing

Main question in option pricing: how much is this ‘right to buy in the future’ worth?

One year from now, you will require 100 tonne of steel; how do you deal with possible price fluctuations?

1. Decide a price now, say 630 EUR/tonne.

Rather than a contract to get steel in one year at 630 EUR/tonne, it is better to obtain the right, not the obligation to buy steel at 630 EUR/tonne one year from now.

This is called an ‘option contract’. All types exist, eg.also the right to sell, with different times, and for different underlying assets.

sT=730

sT=530

K=630s0=610

Options: the right to buy/sell in the future at a fixed price

p

1-p

Expected payoff:p100+(1p)0

The ‘standard model’ of option pricing: Black-Scholes

We work with the logreturn and model the changes in the valueof the underlying asset as a Brownian random walk

time

x

From the payoff at the finalposition, work a step back toobtain a differential equation

p1-p

p1-p

Binomial tree method: see eg. Options, Futures, and Other Derivatives by John C. Hull (Prentice Hall publ.)

xRather than 2 possible futures, there are many – but they can bee seen as a limit of many small ‘binomial’ steps according to the central limit theorem, the outcome is Gaussian fluctuations

The probability to end up in xT is also given by the sum over all paths thatend up there, weighed by the probability of these paths

p1-p

The ‘path-centered’ point of view on Black-Scholes

We work with the logreturn and model the changes in the valueof the underlying asset as a Brownian random walk

time

x

Take the sum over all paths: this is the price propagator from t=0 to t=T.

(note: in this example there would already be 212 possible paths…)

etc.

The ‘path-centered’ point of view on Black-Scholes

time

x

x0

x(t)

xT

Many alternatives: add the amplitudes

this Feynman path integral determines the propagator

We work with the logreturn and model the changes in the valueof the underlying asset as a Brownian random walk

Quantum

The amplitude for a given path is a phase factor :

where S is the action functional ,

fixed by integrating the Lagrangian along the path, eg. for a free particle:

xRather than 2 possible futures, there are many – but they can bee seen as a limit of many small ‘binomial’ steps according to the central limit theorem, the outcome is Gaussian fluctuations

The probability to end up in xT is also given by the sum over all paths thatend up there, weighed by the probability of these paths

The ‘path-centered’ point of view on Black-Scholes

time

x

x0

x(t)

xT

We work with the logreturn and model the changes in the valueof the underlying asset as a Brownian random walk

Stochastic

The amplitude for a given path is a phase factor :

where S is the action functional ,

fixed by integrating the Lagrangian along the path, eg. for a free particle:

Many alternatives: add the probabilities

this Wiener path integral determines the propagator

xRather than 2 possible futures, there are many – but they can bee seen as a limit of many small ‘binomial’ steps according to the central limit theorem, the outcome is Gaussian fluctuations

The probability to end up in xT is also given by the sum over all paths thatend up there, weighed by the probability of these paths

The ‘path-centered’ point of view on Black-Scholes

time

x

x0

x(t)

xT

We work with the logreturn and model the changes in the valueof the underlying asset as a Brownian random walk

Stochastic

The probability for a given path is a real number:

where S is the action functional ,

fixed by integrating the Lagrangian along the path, eg. for a free particle:

Many alternatives: add the probabilities

this Wiener path integral determines the propagator

xRather than 2 possible futures, there are many – but they can bee seen as a limit of many small ‘binomial’ steps according to the central limit theorem, the outcome is Gaussian fluctuations

The probability to end up in xT is also given by the sum over all paths thatend up there, weighed by the probability of these paths

The ‘path-centered’ point of view on Black-Scholes

time

x

x0

x(t)

xT

We work with the logreturn and model the changes in the valueof the underlying asset as a Brownian random walk

Stochastic

The probability for a given path is a real number:

where S is the action functional ,

fixed by integrating the Lagrangian along the path, eg. for the BS model:

Many alternatives: add the probabilities

this Wiener path integral determines the propagator

xRather than 2 possible futures, there are many – but they can bee seen as a limit of many small ‘binomial’ steps according to the central limit theorem, the outcome is Gaussian fluctuations

The probability to end up in xT is also given by the sum over all paths thatend up there, weighed by the probability of these paths

The ‘path-centered’ point of view on Black-Scholes

We work with the logreturn and model the changes in the valueof the underlying asset as a Brownian random walk

The probability for a given path is a real number:

where S is the action functional ,

fixed by integrating the Lagrangian along the path, eg. for the BS model:

Many alternatives: add the probabilities

this Wiener path integral determines the propagator

xRather than 2 possible futures, there are many – but they can bee seen as a limit of many small ‘binomial’ steps according to the central limit theorem, the outcome is Gaussian fluctuations

The probability to end up in xT is also given by the sum over all paths thatend up there, weighed by the probability of these paths

Black-Scholes: The Galton board

The application of path integrals to option prices in BS has been pioneered by various authors: Dash, Linetsky, Rosa-Clot, Kleinert.

The ‘path-centered’ point of view on Black-Scholes

We work with the logreturn and model the changes in the valueof the underlying asset as a Brownian random walkx

Rather than 2 possible futures, there are many – but they can bee seen as a limit of many small ‘binomial’ steps according to the central limit theorem, the outcome is Gaussian fluctuations

The probability to end up in xT is also given by the sum over all paths thatend up there, weighed by the probability of these paths

Fisher Black & Myron Scholes, "The Pricing of Options and Corporate Liabilities". Journal of Political Economy 81 (3): 637–654 (1973).Robert C. Merton, "Theory of Rational Option Pricing“, Bell Journal of Economics and Management Science (The RAND Corporation) 4 (1): 141–183 (1973).

Part III: Some advantages of the path integral point of view

Two problems with the standard model

Problem 2: Not all options have a payoff that depends only on x(t=T), many options have a path-dependentpayoff, i.e. payoff is a functional of x(t).

Problem 1: The fluctuations are not Gaussian

Δx

Black-Scholes-Merton model

Δx

Free particle propagator Black-Scholes option price

Two problems with the standard model

Problem 2: Not all options have a payoff that depends only on x(t=T), many options have a path-dependentpayoff, i.e. payoff is a functional of x(t).

Problem 1: The fluctuations are not Gaussian

Free particle propagator Black-Scholes option price

Asian option: payoff is a function of the average of the underlying price?

xA xB

x

t

Improving Black-Scholes : stochastic volatility

Δx

Black-Scholes-Merton model

Δx

t

Heston model

Δx

t

The Heston model treats the variance as a second stochastic variable, satisfying its own stochastic differential equation:

mean reversion rate mean reversion level

the ‘volatility of the volatility’

t

xtvt

t

two particle problem

with z = (v/)1/2

t

xtvt

t

two particle problem

Improving Black-Scholes : stochastic volatility

Δx

From the infinitesimal propagator of the stochastic process we identify the following Lagrangian that corresponds to the same propagator:

with z = (v/)1/2

Improving Black-Scholes : stochastic volatility

Δx

Black-Scholes-Merton model

Δx

t

Heston model

Δx

tt

xtvt

t

two particle problem : free particle strangely coupled to a radial harmonic oscillator

with z = (v/)1/2

Improving Black-Scholes : stochastic volatility

Δx

Black-Scholes-Merton model

Δx

t

Heston model

Δx

tt

xtvt

t

two particle problem : free particle strangely coupled to a radial harmonic oscillator

with z = (v/)1/2

Improving Black-Scholes : stochastic volatility

Δx

Black-Scholes-Merton model

Δx

t

Heston model

Δx

tt

xtvt

t

two particle problem : free particle strangely coupled to a radial harmonic oscillator

with z = (v/)1/2

Details: D.Lemmens, M. Wouters, JT, S. Foulon, Phys. Rev. E 78, 016101 (2008).

Other improvements to Black-Scholes

A) Stochastic Volatility

* Heston model:

* Hull-White model

* Exponential Vasicek model

B) Jump Diffusion (and Levy models)

poisson process

BS

Add stoch vol. Add jump diff.

Heston Kou

again a zoo of proposals

H. Kleinert , Option Pricing from Path Integral for Non-Gaussian Fluctuations.Natural Martingale and Application to Truncated Lévy Distributions , Physica A 312, 217 (2002).

Other models and other tricks

Improvements to Black-Scholes ...translate into ...to which quantum mechanics physical actions solving techniques can be applied

A) Stochastic Volatility

1. Heston model free particle coupled to exact solution radial harmonic oscillator

B) Stochastic volatiltiy + Jump Diffusion

2. Exponential Vasicek model particle in an exponential perturbational gauge field generated by (Nozieres – Schmitt-Rink free particle expansion)

3. Kou and Merton’s models particle in complicated variational (Jensen-Feynman potential (not previously variational principle) studied)

References1. D. Lemmens, M. Wouters, J. Tempere, S. Foulon, Phys. Rev. E 78 (2008) 016101.2. L. Z. Liang, D. Lemmens, J. Tempere, European Physical Journal B 75 (2010) 335–342.3. D. Lemmens, L. Z. J. Liang, J. Tempere, A. D. Schepper, Physica A 389 (2010) 5193 – 5207.

More complicated payoffs

‘Plain vanilla’ or simple options have a payoff that only depends on the value of the underlying at expiration, x(t=T). For such options we have:

Many other option contracts have a payoff that depends on the entire path, such as: Asian options: payoff depends on the average price during the option lifetime Timer options: contract duration depends on a volatility budget Barrier options: contract becomes void if price goes above/below some value

For such options, the price is given by

Feynman-Kac ‘interpretation’ include payoff in the path weight:

Part IV: A concrete and recent example: Timer options

Timer options have an uncertain expiry time, equal to the time at which a certain “variance budget” has been used up.

The Duru-Kleinert transformation

t

x

x0

x(t)

xTwith

and F>0

q

qA

q()

qB

final time depends on path

‘clock’ time is a functional of the path followed:

H. Duru and H. Kleinert, Solution of the Path Integral for the H-Atom, Phys. Letters B 84, 185 (1979).H. Duru and N. Unal, Phys. Rev. D 34, 959 (1986).

The Duru-Kleinert transformation

H. Duru and H. Kleinert, Solution of the Path Integral for the H-Atom, Phys. Letters B 84, 185 (1979).H. Duru and N. Unal, Phys. Rev. D 34, 959 (1986).

This transformation results in the equivalency between the following two path integrals:

F(q) can be chosen to regularize a singular potential.

This technique was used to solve the propagator ofthe electron in the hydrogen atom, transformingthe a 3D singular potential into a 4D harmonicoscillator problem.

More complicated payoffs: timer option

Timer options have an uncertain expiry time, equal to the time at which a certain pre-specified “variance budget” has been used up. Their description requires a stochastic volatility model:

L. Z. J. Liang, D. Lemmens, and J. Tempere, Physical Review E 83, 056112 (2011).

The expiry time is determined by the variance budget B :

This now defines a Duru-Kleinertpseudotime !

More complicated payoffs: timer option

Timer options have an uncertain expiry time, equal to the time at which a certain pre-specified “variance budget” has been used up. Their description requires a stochastic volatility model:

L. Z. J. Liang, D. Lemmens, and J. Tempere, Physical Review E 83, 056112 (2011).

The Duru-Kleinert transformationhas a well defined inverse .

Denoting and we find that these obey new SDE’s:

now X and V evolve up to a fixed time,

More complicated payoffs: timer option

L. Z. J. Liang, D. Lemmens, and J. Tempere, Physical Review E 83, 056112 (2011).

The 3/2 model:

results in a particle in a Morse potential.

The Heston model

results here in particle in a Kratzer potential

Conclusions

t

StFluctuating paths in finance are described by stochastic models, which can be translated to Lagrangians for path integration.

Path integrals can solve in a unifying framework the two problems of the ‘standard model of option pricing’:

1/ The real fluctuations are not gaussian 2/ New types of option contracts have path- dependent payoffs

D. Lemmens, M. Wouters, JT, S. Foulon, Phys. Rev. E 78, 016101 (2008); J.P.A. Devreese, D. Lemmens, JT, Physica A 389, 780-788 (2010);L. Z. J. Liang, D. Lemmens, JT, European Physical Journal B 75, 335–342 (2010); L. Z. J. Liang, D. Lemmens, JT, Physical Review E 83, 056112 (2011).D. Lemmens, L. Z. J. Liang, JT, A. D. Schepper, Physica A 389, 5193–5207 (2010);