Numerical option pricing in the Barndorff-Nielsen - Shephard stochastic volatility model. Martin...

20
Numerical option pricing in the Barndorff-Nielsen - Shephard stochastic volatility model. Martin Groth Centre of Mathematics for Applications University of Oslo Joint work with Fred Espen Benth

Transcript of Numerical option pricing in the Barndorff-Nielsen - Shephard stochastic volatility model. Martin...

Page 1: Numerical option pricing in the Barndorff-Nielsen - Shephard stochastic volatility model. Martin Groth Centre of Mathematics for Applications University.

Numerical option pricingin the Barndorff-Nielsen - Shephard

stochastic volatility model.

Martin GrothCentre of Mathematics for Applications

University of Oslo

Joint work with Fred Espen Benth

Page 2: Numerical option pricing in the Barndorff-Nielsen - Shephard stochastic volatility model. Martin Groth Centre of Mathematics for Applications University.

The ModelThe Model

Consider the market with a bond Rt with risk free rate of return r

and a risky asset St, the latter evolving according to the

stochastic volatility model:

dSt = (+Yt)Stdt + (Yt)StdBt , S0 = s > 0

dYt = -Ytdt + dLt , Y0 = y > 0.

where Bt is Brownian motion, Lt is a subordinator and

2(y)= y. Hence, the volatility is modelled as a mean reverting Ornstein-Uhlenbeck process with positive jumps given by a Lévy process. The model has the advantage of capturing both the heavy tails and the dependency structure observed in financial return data. Tell me more about

the backgroundSo, where’s the numerics?

Page 3: Numerical option pricing in the Barndorff-Nielsen - Shephard stochastic volatility model. Martin Groth Centre of Mathematics for Applications University.

Indifference PricingIndifference Pricing

Let us consider an investor who two has choices, trying to maximize her exponential utility. She can either• Enter in to the market by her own account, or• Issue a derivative and invest her incremental wealth after

collecting the premium.

The utility indifference price is then the price of the claim, for a given risk aversion , at which the investor is indifferent between the two different alternatives.

Hey, do I spot a PDE down there?

Page 4: Numerical option pricing in the Barndorff-Nielsen - Shephard stochastic volatility model. Martin Groth Centre of Mathematics for Applications University.

ReferencesReferences

• Benth, F.E. and Meyer-Brandis, T., The density process of the minimal entropy martingale measure in a stochastic volatility model with jumps. Finance and Stochastics (2005), vol IX, no. 4.

• Benth, F.E. and Groth, M., The minimal entropy martingale measure and numerical option pricing for the Barndoff-Nielsen - Shephard stochastic volatility model. To be submitted.

• Nicolato, E. and Venardos, E., Option pricing in stochastic volatility models of the Ornstein-Uhlenbeck type, Mathematical Finance (2003), Vol. 13, No. 4,

Page 5: Numerical option pricing in the Barndorff-Nielsen - Shephard stochastic volatility model. Martin Groth Centre of Mathematics for Applications University.

The MeasureThe Measure

The price in the zero risk aversion limit is known to coincide with the price under the Minimal entropy martingale measure (MEMM). Under this measure the dynamics of the processes St

and Yt are changed to

dSt* = (Yt

*)St dBt* , S0

* = s > 0

dYt* = -Yt

* dt + dL*

t , Y0* = y > 0.

where the subordinator is transformed to a pure jump Markov process with jump measure

Back to the PDEs

ν *(ω,dz,dt) =H(t,Yt

*(ω) + z)

H(t,Yt*(ω))

ν (dz)dt

Page 6: Numerical option pricing in the Barndorff-Nielsen - Shephard stochastic volatility model. Martin Groth Centre of Mathematics for Applications University.

The PDEThe PDE

The option prices we calculate is given by a parabolic PDE with an integral term, here assuming r = 0:

where is the option price. The non-local integral term comes from the subordinator Lt having jump measure ν (dz). We

observe that H(t,y) appears as a measure change in the integral, giving a coupled system of PDEs.

What is H here?Coupled system?

Prices under whatmeasure?

Now, where is the numerics?

t +1

2σ 2(y)s2Λss − λyΛy + λ Λ(t,y + z,s) − Λ(t,y,s){ }

H(t, y + s)

H(t,y)ν (dz) = 0

0

Page 7: Numerical option pricing in the Barndorff-Nielsen - Shephard stochastic volatility model. Martin Groth Centre of Mathematics for Applications University.

Case 1: No Claim IssuedCase 1: No Claim Issued

Through a dynamic programming approach Benth & Meyer-Brandis derived an integro-PDE for the value function of the investor in the case she enter directly into the market:

with (t,y) [0,T)R+. We can assume that H(t,0) = 0 and also that

the solution will approach the explicit solution, given by Benth & Meyer-Brandis, for the special case = 0, as y .

Do you solve this ghastly thing?

But this is not theoption prices, right?

OK, but what about the other case?

H t −μ + βy( )

2

2σ 2(y)H − λHy + λ H(t, y + z) − H(t, y){ }ν (dz) = 0

0

Page 8: Numerical option pricing in the Barndorff-Nielsen - Shephard stochastic volatility model. Martin Groth Centre of Mathematics for Applications University.

The Plot, H(T,Y)The Plot, H(T,Y)

We solve this equation with the finite difference method giving results looking like this:

Did I misssomething?

But why do you care about H(t,y)?

Page 9: Numerical option pricing in the Barndorff-Nielsen - Shephard stochastic volatility model. Martin Groth Centre of Mathematics for Applications University.

Indifference PricingIndifference Pricing

Text här

Åt högerÅt vänster Nedåt

Page 10: Numerical option pricing in the Barndorff-Nielsen - Shephard stochastic volatility model. Martin Groth Centre of Mathematics for Applications University.

Discretizing the PDEsDiscretizing the PDEs

To solve the coupled system of PDEs we use the finite difference method. We discretize the equation and restrict the problem to a finite domain. We use dimensional splitting for the two spatial dimensions of the option prices and derive an implicit Lax-Wendroff scheme for the equation in y.

We first have to solve for H(t,y) since this appears in the integrand of the option price. We also have to numerically evaluate the non-local integral term on the whole solution space using a simple trapezoidal rule.

Maybe I need to know about the problem anyway.

What about the integral then?

Page 11: Numerical option pricing in the Barndorff-Nielsen - Shephard stochastic volatility model. Martin Groth Centre of Mathematics for Applications University.

Case 2: Claim IssuedCase 2: Claim Issued

The indifference price for the option comes from the value function in the case the investor issues a claim. Going to the zero risk aversion limit Benth & Meyer-Brandis derived the following PDE for the option price:

with terminal conditions (t,y,s) = f(s), where f(s) is the payoff function of the option. Observe that H(t,y) appears in the integrand.

Nice, but what is H doing here?

Let’s get our handsdirty with numerics!

t +1

2σ 2(y)s2Λss − λyΛy + λ Λ(t,y + z,s) − Λ(t,y,s){ }

H(t, y + s)

H(t,y)ν (dz) = 0

0

Page 12: Numerical option pricing in the Barndorff-Nielsen - Shephard stochastic volatility model. Martin Groth Centre of Mathematics for Applications University.

The Role of H(t,y)The Role of H(t,y)

The fraction H(t,y+z)/H(t,y) appears as a measure change in the integrand of the option price. It will scale the jumps from the subordinator, making the jump measure of the subordinator time-inhomogeneous and state-dependent. We see that as we approach zero the fraction will approach infinity. Hence for small volatilities the volatility process will have a large probability to jump up. The smaller the volatility the higher intensity and size of the jumps, and thus, we will quickly jump to higher volatilities.

How does this effect the option?

What does this H-function look like?

Page 13: Numerical option pricing in the Barndorff-Nielsen - Shephard stochastic volatility model. Martin Groth Centre of Mathematics for Applications University.

ResultsResults

We simulate prices for a European call option using parameters from Nicolato & Venardos. The stationary distribution of the volatility process is assumed to be inverse Gaussian, giving option prices which are approximately normal inverse Gaussian distributed. From Eberlein we know that prices in a exponential Lévy model with a normal inverse Gaussian Lévy process give a characteristic “W”-shape compared to Black& Scholes prices. To compare we need to choose volatility for the to Black & Scholes prices. We compare prices under MEMM at y equal to the expected value of the stationary distribution with Black & Scholes prices with the same volatility.

Hit me with a plot!

Page 14: Numerical option pricing in the Barndorff-Nielsen - Shephard stochastic volatility model. Martin Groth Centre of Mathematics for Applications University.

The IntegralThe Integral

There are a few issues concerning the integral we need to address. Restricting to a finite domain we need to handle the cut-off for large y. This is taken care of by realizing that the option prices will adjust toward the Black & Scholes price for large volatilities. We can use this to integrate beyond the boundary.

The modelling of the volatility process might give a jump measure of the subordinator which has infinite activity. To integrate numerically from zero is then futile and to compensate for the influence from the small jumps we add a drift term to the PDE.

Did anyone sayboundary condition?

Come on, results now!

Page 15: Numerical option pricing in the Barndorff-Nielsen - Shephard stochastic volatility model. Martin Groth Centre of Mathematics for Applications University.

Boundary ConditionsBoundary Conditions

For the finite domain we derive appropriate boundary conditions. We show that for large values of s and y the prices will approach Black & Scholes prices with integrated volatility.

If s = 0 the asset will be worthless throughout the whole lifetime of the asset and the option value will equal the payoff.

On the last side, where y = 0 we show heuristically that it is reasonable to assume that we have a Neumann condition on the boundary.

Nice, but does it converge?

Let’s go straight to the results!

Page 16: Numerical option pricing in the Barndorff-Nielsen - Shephard stochastic volatility model. Martin Groth Centre of Mathematics for Applications University.

The Plot, ConvergenceThe Plot, Convergence

From numerical test it appears to converge:

B.C.But you can’t show it, right?

Title: /Volumes/work/Users/martijg/Projekt/Integro-PDE/Resultat/Resultat v.0.4/Convergens_yta.epsCreator: MATLAB, The Mathworks, Inc.Preview: This EPS picture was not saved with a preview (TIFF or PICT) included in itComment: This EPS picture will print to a postscript printer but not to other types of printers

Page 17: Numerical option pricing in the Barndorff-Nielsen - Shephard stochastic volatility model. Martin Groth Centre of Mathematics for Applications University.

The Plot, Option Prices IThe Plot, Option Prices I

Our results displays the characteristic “w”-shape as expected:

Does it look like option prices?

Title: /Volumes/work/Users/martijg/Projekt/Integro-PDE/Resultat/Resultat v.0.4/BS-BNS.epsCreator: MATLAB, The Mathworks, Inc.Preview: This EPS picture was not saved with a preview (TIFF or PICT) included in itComment: This EPS picture will print to a postscript printer but not to other types of printers

Explain a bit more!

Page 18: Numerical option pricing in the Barndorff-Nielsen - Shephard stochastic volatility model. Martin Groth Centre of Mathematics for Applications University.

The Plot, Option Prices IIThe Plot, Option Prices II

Plotting the option price as a function of first t and s and then s and y we see that the result in much resembles the B & S price:

Does it smile?And compared toB & S prices?

Title: /Volumes/work/Users/martijg/Projekt/Integro-PDE/Resultat/Resultat v.0.4/Indiffprice_ST_yta_vol0_1528.epsCreator: MATLAB, The Mathworks, Inc.Preview: This EPS picture was not saved with a preview (TIFF or PICT) included in itComment: This EPS picture will print to a postscript printer but not to other types of printers

Title: /Volumes/work/Users/martijg/Projekt/Integro-PDE/Resultat/Resultat v.0.4/Indiffprice_SY_yta.epsCreator: MATLAB, The Mathworks, Inc.Preview: This EPS picture was not saved with a preview (TIFF or PICT) included in itComment: This EPS picture will print to a postscript printer but not to other types of printers

Page 19: Numerical option pricing in the Barndorff-Nielsen - Shephard stochastic volatility model. Martin Groth Centre of Mathematics for Applications University.

The Volatility SmileThe Volatility Smile

Using a stochastic volatility model we expect a better fit to observed volatility smiles than given by the Black & Scholes model. Indeed, the results produce a smile in the implied volatility

Nice plot! But whatabout convergence?

Title: /Volumes/work/Users/martijg/Projekt/Integro-PDE/Kod/Integro-PDE 0.4/build/volatility_smile.epsCreator: MATLAB, The Mathworks, Inc.Preview: This EPS picture was not saved with a preview (TIFF or PICT) included in itComment: This EPS picture will print to a postscript printer but not to other types of printers

Can I see what theresults look like?

Page 20: Numerical option pricing in the Barndorff-Nielsen - Shephard stochastic volatility model. Martin Groth Centre of Mathematics for Applications University.

ConvergenceConvergence

We have so far not proved that our numerical schemes are converging. This analysis will be carried out in further research. We are confident that our solver will fit into a larger framework of convergence analysis for integro-PDEs.

At his point we can only rely on numerical justification of the convergence, illustrated in the plot above.

Look, what amarvellous plot! Show me some numerical evidence