Arbitrage opportunities in misspecied stochastic volatility models

Rudra P. Jena Peter Tankov

cole Polytechnique - CMAP

Modeling and Managing Financial Risks, Paris, 12th Jan 2011

Rudra P. Jena - jena@cmapx.polytechnique.fr

misspecication of SV models

Outline

1. Motivation/ Introduction/Background 2. Set up of the problem 3. General solution 4. The Black Scholes case 5. The Stochastic Volatility case: A perturbative approach 6. Numerics/Results

Rudra P. Jena - jena@cmapx.polytechnique.fr

misspecication of SV models

Introduction
The motivation

Widely documented phenomenon of option mispricing. Given set of assumptions on the real-world dynamics of an asset, the European options on this asset are not efciently priced in options markets. [Y-Ait Sahaliya et. al, Bakshi et. al]

Rudra P. Jena - jena@cmapx.polytechnique.fr

misspecication of SV models

Introduction
Discrepancies between the implied volatility and historical volatility levels
60
Realized vol (2month average)

50 40 30 20 10

Implied vol (VIX)

0 Nov 1997

May 2009

Substantial differences between historical and option-based measures of skewness and kurtosis [Bakshi et. al] have been documented.
Rudra P. Jena - jena@cmapx.polytechnique.fr misspecication of SV models

Background

Misspecication studied extensively in Black Scholes model with misspecied volatility El Karoui, Jeanblanc & Shreve We address the question of Misspecied stochastic volatility models.

Rudra P. Jena - jena@cmapx.polytechnique.fr

misspecication of SV models

Background

Misspecication studied extensively in Black Scholes model with misspecied volatility El Karoui, Jeanblanc & Shreve We address the question of Misspecied stochastic volatility models. Financial engineering folklore ! misspecied correlation a risk reversal misspecied volatility of volatility a buttery spread

Rudra P. Jena - jena@cmapx.polytechnique.fr

misspecication of SV models

Background

Misspecication studied extensively in Black Scholes model with misspecied volatility El Karoui, Jeanblanc & Shreve We address the question of Misspecied stochastic volatility models. Financial engineering folklore ! misspecied correlation a risk reversal misspecied volatility of volatility a buttery spread We concentrate on arbitrage strategies involving underlying asset liquid European options.

Rudra P. Jena - jena@cmapx.polytechnique.fr

misspecication of SV models

Setting
Real world dynamics

Under real-world probability P, the underlying price S follows a stochastic volatility model 1 2 dWt1 + (Yt )t dWt2 t

dSt /St = t dt + (Yt ) dYt = at dt + bt dWt2 ,

: R (0, ) is a Lipschitz C 1 -diffeomorphism (y ) > 0 for all y R; , a, b > 0 and [1, 1] are adapted (W 1 , W 2 ) is a standard 2-dimensional Brownian motion.

Rudra P. Jena - jena@cmapx.polytechnique.fr

misspecication of SV models

Setting

Process t , which represents the instantaneous volatility used by the options market for all pricing purposes. We assume that t = (Yt ) dYt = at dt + bt dWt2 , where at and bt > 0 are adapted. : R (0, ) is a Lipschitz C 1 -diffeomorphism with 0 < (y ) < and (y ) > 0 for all y R; (1)

Rudra P. Jena - jena@cmapx.polytechnique.fr

misspecication of SV models

Setting
The Market model dynamics

Assumptions, Another probability measure Q, called market or pricing probability All traded assets are martingales under Q The interest rate is assumed to be zero Under Q, the underlying asset and its volatility form a 2-dimensional Markovian diffusion: dSt /St = (Yt ) 1 2 (Yt , t)dWt1 + (Yt )(Yt , t)dWt2

dYt = a(Yt , t)dt + b(Yt , t)dWt2 , a, b and are deterministic functions.

Rudra P. Jena - jena@cmapx.polytechnique.fr

misspecication of SV models

Setting Contd.
Suppose that a continuum of European options for all strikes and at least one maturity, quoted in the market. The price of an option with maturity date T and pay-off H(ST ) of St , Yt and t: P(St , Yt , t) = E Q [H(ST )|Ft ]. For every such option, the pricing function P belongs to the class C 2,2,1 ((0, ) R [0, T )) and satises the PDE a where we dene f S 2 (y )2 2 f b2 2 f 2f Lf = + + + S (y )b . t 2 2 y 2 Sy S 2
Rudra P. Jena - jena@cmapx.polytechnique.fr misspecication of SV models

P + LP = 0, y

Under our assumptions any such European option can be used to complete the Q-market. (Romano, Touzi) And price satises P > 0, y

(S, y , t) (0, ) R [0, T ).

The real-world market may be incomplete in our setting.

Rudra P. Jena - jena@cmapx.polytechnique.fr

misspecication of SV models

Decay Properties of the Greeks

Lemma Let P be the price of a call or a put option with strike K and maturity date T . Then P(S, y , t) P(S, y , t) = lim = 0, y y K 0 2 P(S, y , t) 2 P(S, y , t) = lim = 0, lim K 0 K + y 2 y 2 2 P(S, y , t) 2 P(S, y , t) and lim = lim =0 Sy Sy K + K 0
K +

lim

for all (y , t) R [0, T ). All the above derivatives are continuous in K and the limits are uniform in S, y , t on any compact subset of (0, ) R [0, T ).

Rudra P. Jena - jena@cmapx.polytechnique.fr

misspecication of SV models

Sketch of the Proof

The option price satises, a P + LP = 0, y

Differentiate w.r.t. y and S, Use Feynman Kac representation to relate the various greeks to the fundamental solutions of pde. Using the classical bounds for fundamental solutions of parabolic equations.

Rudra P. Jena - jena@cmapx.polytechnique.fr

misspecication of SV models

Formulation of the Problem

The arbitrage problem is set up from the perspective of a trader, Who knows market is using misspecied model Wants to construct a strategy to benet from this misspecication. The rst step, sets up a dynamic self nancing delta and vega-neutral portfolio Xt with zero initial value.
at each date t, a stripe of European call or put options with a common time to expiry Tt . t (dK ) : quantity of options with strikes between K and K + dK

t of stock Bt of cash. |t (dK )| = 1

Rudra P. Jena - jena@cmapx.polytechnique.fr misspecication of SV models

Formulation contd.
The value of the resulting portfolio is,
Xt = PK (St , Yt , t)t (dK ) t St + Bt ,

The dynamics of this portfolio is given by, P K P K dSt + dYt S y

dXt = where,

t (dK ) LP K dt +

t dSt

Lf =

S 2 (Yt )2 2 f b2 2 f 2f f + t + t + St (Yt )bt t t 2 S 2 2 y 2 Sy

Rudra P. Jena - jena@cmapx.polytechnique.fr

misspecication of SV models

choose, t (dK ) P K = 0, y t (dK ) P K = t S

to eliminate the dYt and dSt terms. The resulting portfolio is risk free. The portfolio dynamics reduces to, dXt = t (dK )LP K dt,

Rudra P. Jena - jena@cmapx.polytechnique.fr

misspecication of SV models

Now we can write down the risk free prot from model misspecication as, dXt = At the liquidation date T ,
T

t (dK )(L L)P K dt.

XT =
0

t (dK )(L L)P K dt,

where, S 2 ( 2 2 (Yt )) 2 P K (b2 bt2 ) 2 P K (L L)P K = t t + t 2 2 S 2 y 2 2 K P + St (t bt t (Yt )bt t ) Sy

Rudra P. Jena - jena@cmapx.polytechnique.fr

misspecication of SV models

The problem in a Nutshell

The trader needs to maximize this aribtrage prot. Taking advantage of arbitrage opportunity to the following optimisation problem, Maximize Pt = subject to t (dK )(L L)P K and t (dK ) P K = 0. y

|t (dK )| = 1

ANSWER: Spread of only two options is sufcient to solve this problem.

Rudra P. Jena - jena@cmapx.polytechnique.fr

misspecication of SV models

General Result
Proposition The instantaneous arbitrage prot is maximized by t (dK ) = wt1 K 1 (dK ) wt2 K 2 (dK ),
t t

where K (dK ) denotes the unit point mass at K , (wt1 , wt2 ) are time-dependent optimal weights given by wt1 =
P K2 y P K1 y

P K2 y

wt2

P K1 y P K1 y

P K2 y

and (Kt1 , Kt2 ) are time-dependent optimal strikes given by (Kt1 , Kt2 ) = arg max
P K y
2 1 (L L)P K

P K y

2 (L L)P K

K 1 ,K 2

P K 1 y

P K 2 y

Rudra P. Jena - jena@cmapx.polytechnique.fr

misspecication of SV models

Sketch of Proof

The proof is done in two steps, First show that the optimization problem is well-posed, i.e., the maximum is attained for two distinct strike values. show that the two-point solution suggested by this proposition is indeed the optimal one.

Rudra P. Jena - jena@cmapx.polytechnique.fr

misspecication of SV models

The Black Scholes case

The misspecied model is the Black-Scholes with constant volatility (but the true model is of course a stochastic volatility model). In the Black-Scholes model (r = 0): P = Sn(d1 ) T = Kn(d2 ) T , 2P n(d1 )d2 = , S 2P Sn(d1 )d1 d2 T = , 2
m where d1,2 = T normal density. T 2 ,

m = log(S/K ) and n is the standard

Rudra P. Jena - jena@cmapx.polytechnique.fr

misspecication of SV models

Proposition Let b = = 0. The optimal option portfolio maximizing the instantaneous arbitrage prot is described as follows: The portfolio consists of a long position in an option with 2 log-moneyness m1 = z1 T 2T and a short position in 2 an option with log-moneyness m2 = z2 T 2T , where z1 and z2 are maximizers of the function f (z1 , z2 ) = with w0 =
(bT +2) . b T

(z1 z2 )(z1 + z2 w0 ) ez1 /2 + ez2 /2

2 2

The weights of the two options are chosen to make the portfolio vega-neutral. We dene by Popt the instantaneous arbitrage prot realized by the optimal portfolio.
Rudra P. Jena - jena@cmapx.polytechnique.fr misspecication of SV models

Proof

Substituting the Black-Scholes values for the derivatives of option prices, m change of variable z = + 2 T , T the function to maximize w.r.t. z1 , z2 becomes: n(z1 )n(z2 ) n(z1 ) + n(z2 ) b T 2 bT 2 (z1 z2 ) (z1 z2 ) (z1 z2 ) , 2 2

from which the proposition follows directly.

Rudra P. Jena - jena@cmapx.polytechnique.fr

misspecication of SV models

Role of Butteries and Risk reversals: Part 1

Proposition Let b = = 0, and dene by P opt the instantaneous arbitrage prot realized by the optimal strategy. Consider a portfolio (RR) described as follows: If bT /2 + 0
1 buy 2 units of options with log-moneyness 2 m1 = T 2T , or, equivalently, delta value N(1) 0.16 1 selling 2 units of options with log-moneyness 2 m2 = T 2T , or, equivalently, delta value N(1) 0.84.

if bT /2 + < 0 buy the portfolio with weights of the opposite sign. Then the portfolio (RR) is the solution of the maximization problem under the additional constraint that it is -antisymmetric.
Rudra P. Jena - jena@cmapx.polytechnique.fr misspecication of SV models

Part 2
Proposition Consider a portfolio (BB) consisting in buying x0 units of options with log-moneyness m1 = z0 T 2 T , or, equivalently, delta value N(z0 ) 0.055, where z0 1.6 is a universal constant. buying x0 units of options with log-moneyness m2 = z0 T 2 T , or, equivalently, delta value N(z0 ) 0.945 selling 1 2x0 units of options with log-moneyness 2 m3 = 2T or, equivalently, delta value N(0) = 1 . 2 The quantity x0 is chosen to make the portfolio vega-neutral, that is, x0 0.39. Then, the portfolio (BB) is the solution of the maximization problem under the additional constraint that it is -symmetric.
Rudra P. Jena - jena@cmapx.polytechnique.fr misspecication of SV models

Part 3

Proposition Dene by P RR the instantaneous arbitrage prot realized by the portfolio of part 1 and by P BB that of part 2. Let = |bT + 2| + 2bK0 T |bT + 2|

where K0 is a universal constant, dened below in the proof, and approximately equal to 0.459. Then P RR P opt and P BB (1 )P opt .

Rudra P. Jena - jena@cmapx.polytechnique.fr

misspecication of SV models

Sketch of Proof
The maximization problem can be reduced to, Sb2 T max z 2 n(z) t (dz) Sb(bT /2 + ) 2 subject to n(z) t (dz) = 0, | t (dz)| = 1.

zn(z) t (dz)

Observe that the contract (BB) maximizes the rst term while the contract (RR) maximizes the second term. The values for the contract (BB) and (RR) are given by Sb|bT /2 + | 1 Sb2 T z 2 /2 BB e 0 , P RR = e 2. P = 2 2

Rudra P. Jena - jena@cmapx.polytechnique.fr

misspecication of SV models

therefore P RR |bT + 2| = BB + P RR P |bT + 2| + 2bK0 T with K0 = e 2 2 .

1 2 z0

Since the maximum of a sum is always no greater than the sum of maxima, P opt P BB + P RR

Rudra P. Jena - jena@cmapx.polytechnique.fr

misspecication of SV models

Remarks

Risk reversals are never optimal and butteries are not optimal unless = bT . 2 Nevertheless, risk reversals and butteries are relatively close to being optimal, and have the additional advantage of being independent from the model parameters, whereas the optimal claim depends on the parameters. This near-optimality is realized by a special universal risk reversal (16-delta risk reversal in the language of foreign exchange markets) and a special universal buttery (5.5-delta vega weighted buttey). When b 0, 1, In this case RR is nearly optimal.

Rudra P. Jena - jena@cmapx.polytechnique.fr

misspecication of SV models

Stochastic Volatility Model

A simple stochastic volatility model, which captures all the desired effects, the SABR = 1 . The dynamics of the underlying asset under Q is dSt = t St ( 1 2 dWt1 + dWt2 ) d t = bt dW 2
t

(2) (3)

The true dynamics of the instantaneous implied volatility is d t = bt dWt2 , and the dynamics of the underlying under the real-world measure is dSt = t St ( 1 2 dWt1 + dWt2 ). (5) (4)

Rudra P. Jena - jena@cmapx.polytechnique.fr

misspecication of SV models

First order correction

Call option price C satisies the following pricing equation, b2 2 C 2C C 2C + Sb + S22 2 + =0 t 2 2 S S stochastic volatility is introduced as a perturbation b = . Look for asymptotic solutions of the form,
C = C0 + C1 +
2

C2 + O( 3 )

Here C0 corresponds to the leading Black Scholes solution. 2 C0 C0 + S2 2 =0 t S 2

The rst leading order to satises the following equation neglecting the higher order terms O( 2 ), C1 = 2 (T t) 2 C0 S 2 S
misspecication of SV models

Rudra P. Jena - jena@cmapx.polytechnique.fr

Perturbation Results

k1 1.0315 1.0310 1.0305 1.36 1.0300 1.34 1.0295 1.32 v 0.1 0.2 0.3 0.4 0.5 0.6 0.1 0.2 0.3 0.4 0.5 0.6 1.0290 v

1.40 1.38

Figure: Optimal Strikes for the set of parameters = .2, S = 1, b = .3, = .3, = .5, t = 1, as a function of the misspecied b [.01, .4].

Rudra P. Jena - jena@cmapx.polytechnique.fr

misspecication of SV models

Numerical Example and Conclusions

Numerical Setup

The trader is aware about the misspecication. Stock price = 100 and volatility = 0.1 Real world parameters: b = .8, = .5 Market or pricing parameters: b = .3, = .7 Demonstration for only one month options. Results are shown for 100 trajectories of the stock and volatility.

Rudra P. Jena - jena@cmapx.polytechnique.fr

misspecication of SV models

0.5

1.0 0.8 misspecified not misspecified

0.4

misspecified not misspecified Portfolio value

0.6 0.4 0.2 0.0 0.2 0.4

0.3

Portfolio value

0.2

0.1

0.0 0.6 0.1 0.0 0.8 0.0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1.0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1.0

time (in yrs)

time (in yrs)

Figure: The evolution of portfolios using options with 1 month

Left: The true parameters are = .2, b = .1. The misspecied or the market parameters are = .3, b = .9. include a bid ask-fork of 0.45% in implied volatility terms for every option transaction.The evolution of the portfolio performance with 32 rebalancing dates.
Rudra P. Jena - jena@cmapx.polytechnique.fr misspecication of SV models

Thank You This research is part of the Chair Financial Risks of the Risk Foundation sponsored by Socit Gnrale, the Chair Derivatives of the Future sponsored by the Fdration Bancaire Franaise, and the Chair Finance and Sustainable Development sponsored by EDF and Calyon.

Rudra P. Jena - jena@cmapx.polytechnique.fr

misspecication of SV models