You are on page 1of 39

Volatility-Smile Modeling with Density-Mixture Stochastic Dierential Equations

Damiano Brigo Credit Models Banca IMI , SanPaolo IMI Group Corso Matteotti 6 20121 Milano, Italy Fax: + 39 02 76019324 E-mail: damiano.brigo@bancaimi.it http://www.damianobrigo.it Based on several joint works listed in the references with Fabio Mercurio, Francesco Rapisarda, Giulio Sartorelli Winter School in Mathematical Finance
(Kasteel Oud Poelgeest, Oestgeest, Amsterdam, December 16-18, 2002);

Quantitative Finance Risk Conference (London, November 25-26, 2002); Hitotsubashi University Conference (Tokyo, March 2002); University of Padua Probability Seminar (October 26, 2001)

- Introduction to the mixture diusion dynamics for the volatility smile

Damiano Brigo

Banca IMI

Overview

Brief intro to the smile problem Brief intro to general no-arbitrage mixture diusions for single smile The no-arbitrage lognormal-mixture dynamics and variants Analytical tractability and calibration Decorrelation between average volatility and underlying asset, and brief comparison with stochastic volatility Examples of calibration to market data Example: Good nested structure of the parameterization Consistent generalization to multi-asset, basket smile Basket of two smiley assets: numerical results Conclusions and references to present/future work

- Introduction to the mixture diusion dynamics for the volatility smile

Damiano Brigo

Banca IMI

Short intro to the smile

Sketchy version of the smile problem. Financial (risky) asset (Black& Scholes, stock, FX rate, etc.)

dSt = r(t)Stdt + (t)St dWt, s0, t [0, T ],

R(t) :=
0

r(s)ds, V (t) :=
0

(s) ds

European call option with maturity T and strike K pays

(ST K)+ at time T . E0 [(ST K) /B(T )] = BSCall(S0, K, T, R(T ), V (T )). V (T )/ T is the (average) volatility of the option and does not depend on K .
In this formulation, volatility is a characteristic of stock S underlying the contract, and has nothing to do with the nature of the contract itself. In particular, it has nothing to do with K .
Q +

- Introduction to the mixture diusion dynamics for the volatility smile

Damiano Brigo

Banca IMI

Short intro to the smile (contd)

Now take two dierent strikes K1 and K2. Suppose that the market provides us with the prices MKTCall(S0, K1, T ) and MKTCall(S0, K2, T ). Does there exist a single volatility V (T ) such that MKTCall(S0, K1, T ) = BSCall(S0, K1, T, R(T ), V (T )), MKTCall(S0, K2, T ) = BSCall(S0, K2, T, R(T ), V (T ))? The answer is a resounding NO!!! Market option prices do not behave like this. Instead two dierent implied volatilities V (T, K1) and V (T, K2) are required to match the observed market prices if one is to use Black & Scholes (BS) formula: MKTCall(S0, K1, T ) = BSCall(S0, K1, T, R(T ), V (T, K1)), MKTCall(S0, K2, T ) = BSCall(S0, K2, T, R(T ), V (T, K2)). In other terms, each market option price requires its own Black and Scholes implied volatility V MKT(T, K)/ T depending on the option strike K .

- Introduction to the mixture diusion dynamics for the volatility smile

Damiano Brigo

Banca IMI

Short intro to the smile (contd)


The market therefore uses BS formula simply as a metric to express option prices as volatilities. The curve MKT K V (T, K)/ T is the so called volatility smile of the T -maturity option. If BSs model were consistent along dierent strikes, this curve would be at, since volatility should not depend on K . Instead, this curve exhibits smiley or skewed shapes. BS Geometric Brownian Motion (GBM) assumption for dS is no longer sucient, need richer dynamics to account for the smile. Alternative diusion dynamics is just one possible means to model the smile (local volatility models, here). There are many other possibilities (stochastic vol models, jump diusions, lattices... not here). Example of tractable diusions: CEV and Shifted GBM (one can also combine the two)

dSt = r Stdt + St dWt, S0 = s0. St = e


Rt
0 r(u)du

+ Xt, dXt = r(t)Xtdt + (t)Xt dWt.

CEV or Shifted-GBM: Just one additional parameter or , not exible enough. Shifted CEV has two additional parameters but still largely insucient. Local vol models are too poor or not tractable. Our proposal: tractable and exible loc vol model.

- Introduction to the mixture diusion dynamics for the volatility smile

Damiano Brigo

Banca IMI

Smile modeling through alternative diusions

Alternative dSt can model a non-at smile: 1. Set K to a starting value; 2. Compute the model option price
RT 0 r(s)ds Q e E0 [(ST

(T, K) =

K) ]

with S modeled through an alternative dynamics

dSt = r(t)Stdt + (t, St) St dWt, S0 = s0


(Harrison and Pliskas Risk Neutral valuation theory still stands) 3. Invert BS formula for this strike, i.e. solve

(K) = BSCall(S0, K1, T, R(T ), V (T, K))


in V (T, K), thus obtaining the model implied volatility V (T, K). 4. Change K and restart from point 2.

- Introduction to the mixture diusion dynamics for the volatility smile

Damiano Brigo

Banca IMI

Smile problem: summary

Traders use the BS metric to price plain-vanilla options; Options are priced (quoted) with a (implied) volatility that varies with the option strike; The term structure of implied volatilities is skewed or smiley The BS model cannot consistently price all options quoted in a market (the real risk-neutral distribution is not lognormal); Need for an alternative asset price dynamics (not just terminal distribution) to price exotics or non quoted plain-vanilla options; This model should feature, among other qualities: explicit dynamics with known marginal distrib.; analytical formulas for European options (analytic and rapid calibration to plain vanilla); good tting of market data (reasonable number of parameters in the dynamics).

- Introduction to the mixture diusion dynamics for the volatility smile

Damiano Brigo

Banca IMI

The no-arbitrage mixture diusion dynamics

dSt = Stdt + (t, St) St dWt, S0 = s0


(assume r(t) = with no loss of generality).

(1)

is no longer a determinisitic function of time, but depends now on the underlying S itself.
We propose a class of analytically tractable models for an asset-price dynamics that are exible enough to reproduce a large variety of market volatility structures. The asset under consideration underlies a given option market (not necessarily a tradeable asset). We can think of an FX rate, a stock index, or a forward LIBOR rate. The asset dynamics follows from assuming that

The risk-neutral measure Q exists; The dynamics of the asset price S under Q is (1). The marginal density of S under Q is a weighted average of the known densities of some given diusion processes.
More specically:
- Introduction to the mixture diusion dynamics for the volatility smile 7

Damiano Brigo

Banca IMI

The no-arbitrage mixture diusion dynamics


Let us then consider N instrumental diusions

dSt = St dt + vi(t, St ) St dWt, s0, i = 1, . . . , N,


i We denote by pi () the density function of St t

Problem. Derive the local volatility (t, St) in

dSt = Stdt + (t, St) St dWt, s0


such that the Q-density of S satises

pt(y) =

N X i=1

i ipt(y),

i > 0,

N X i=1

i = 1.

Solution. Back out from the Fokker-Planck equation for S . We end up with the following SDE under Q:

dSt

v u PN 2 u i=1 ivi (t, St)pi (St) t t = Stdt + StdWt PN i


i=1

ipt(St)

- Introduction to the mixture diusion dynamics for the volatility smile

Damiano Brigo

Banca IMI

Option pricing with the mixture dynamics.


Let us give for granted that the previous SDE has a unique strong solution and consider a European Call option with maturity T , strike K . The option value at t = 0 is : Call = P (0, T )E
T

(ST K) (y K)

Z
= P (0, T )
0 N X i=1

N X i=1

ipT (y)dy
N X i=1

Z
iP (0, T )

(y K) pT (y)dy =

+ i

iCalli.

Remark [Mixtures without dynamics.] Earlier authors used to postulate a lognormal mixture for the risk neutral density, but did not provide any consistent arbitrage-free dynamics. Remark [Greeks]. Due to linearity of dierentiation, the same convex combination applies to all option Greeks (sensitivities). Remark [Complete market, Hedging]. As for all local volatility models, the mixture diusion dynamics yields a complete market (contrary to stochastic vol) and a delta-hedging strategy.
- Introduction to the mixture diusion dynamics for the volatility smile 9

Damiano Brigo

Banca IMI

The no-arbitrage lognormal mixture dynamics

vi(t, y) = i(t)y, dSt = St dt + i(t) St dWt, s0,


where is are det. bounded from above and below, continuous and an such that we have a common initial value i(t) = 0 PN 2 2 for each t in [0, ]. Set mix(t, y) = i=1 i (t, y)i (t)

i(t, y) = PN

i pN (ln s

2 2 (ln y) 0 +tVi (t) /2, Vi (t) ) 2 2 (ln y) 0 +tVj (t) /2, Vj (t) )

j=1

j pN (ln s

for (t, y) > (0, 0); mix(t, y) = 0 for (t, y) = (0, s0). Then the SDE dSt = Stdt + mix(t, St)StdWt , has a unique strong solution whose marginal density is the mixture of the lognormal S is densities

pSt (y) =

N X j=1

j pN (ln s

2 2 (ln y) 0 +tVj (t) /2, Vj (t) )

- Introduction to the mixture diusion dynamics for the volatility smile

10

Damiano Brigo

Banca IMI

Why a mixture of lognormals?

analytically tractable (calibration!) and linked to BS model; log-returns ln(St/S0) are leptokurtic; Mixtures of lognormals work well in many practical situations: Ritchey (1990), Melick and Thomas (1997), Bhupinder (1998) and Guo (1998) found a good tting quality to market options data.
Proposition. Consider a Call option with maturity T , strike K and written on the asset. The model yields Call = P (0, T )
N X i=1

"
i S0e
T

ln

S0 K

+ + i T

1 2 2 i

ln

S0 K

+ i T

1 2 2 i

!#
,

Vi(T ) i(T ) := T

This leads to smiles with a minimum at K = s0eT . Can shift the dynamics to t asymmetric smiles (skews) by adding a new parameter .

At = (A0 S0)e

+ St,

where is a real constant. By Itos formula,

dA = Adt+(t, A(A0S0)e )(A(A0S0)e )dW


- Introduction to the mixture diusion dynamics for the volatility smile 11

Damiano Brigo

Banca IMI

No arbitrage lognormal-mix dyn: Drifts variants

Consider the instrumental processes

dSt = i St dt + i(t)St dWt, s0


and look for a diusion coecient (t, y) such that

dSt = Stdt + dmix(t, St)StdWt, s0


has marginal density pSt (y) =

Pm
i=1

ipS i (y).
t

Call mix(t, y)2 the solution of the analogous problem when all instrumental processes share the same drift (found earlier). It is possible to show that

(dmix(t, y)y) := (mix(t, y)y) +

PN
i=1

i(i )

R +
y t

PN

xpS i (x)dx
t

j=1

j pS j (y)

It is possible to nd conditions under which this diusion coecient has linear growth and does not explode in nite time (Sartorelli, (2002)). The integral in the numerator is just the Black and Scholes price of an asset or nothing option for the instrumental process S i, which is readily available in terms of the Gaussian cumulative distribution function.
- Introduction to the mixture diusion dynamics for the volatility smile 12

Damiano Brigo

Banca IMI

No arbitrage lognormal-mix dyn: Shifts Variants

Can also shift single basic distributions:

At = i e

+ St , dSt = St dt + i(t)St dWt

that leads to instrumental processes

dAt = Atdt + i(t)(At i e )dWt,


and look for an SDE for S with pSt =
N X j=1

P
j

j pAj , i.e.
t

pSt (y) =

j pN (ln(s


2 2 0 j )+tVj (t) /2, Vj (t) )

ln y j e



We nd (but here no a-priori results available for the SDE)

dSt = Stdt+vsmix(t, St)dWt, vsmix(t, y) =


t i

N X i=1

i(t, y)vi(t, y)

i(t, y) = PN

vi(t, y) = i(t)(y ie ). i pN (ln(s )+tV (t)2/2, V (t)2)(ln(y iet))


0 i i

j=1

j pN (ln(s

2 2 (ln(y 0 j )+R(t)Vj (t) /2, Vj (t) )

j
13

et))

- Introduction to the mixture diusion dynamics for the volatility smile

Damiano Brigo

Banca IMI

Mixture model vs Stochastic volatility: Correlation(average volatility,underlying)


dSt = r(t)Stdt + (t, St) St dWt, S0 = s0,
(2)

in general can be either a deterministic or a stochastic function of St. In the latter case we have a stochastic-volatility model (SVM), for example (t, S) = (t),

d((t) ) = b(t, (t) )dt + (t, (t) )dZt,


with the important specication

dZtdWt = dt.

It is usually said that SVM are better than local volatility models (LVM), because the instantaneous correlation is: Corr(dSt, d (t, St)) = < 1 (e.g. = 0 in Hull-White SVM) Corr(dSt, dmix(t, St)) = 1, and the same holds for all LVMs But what about terminal correlations?
2 2

- Introduction to the mixture diusion dynamics for the volatility smile

14

Damiano Brigo

Banca IMI

Mixture model vs Stochastic volatility: Correlation(average volatility,underlying)


Corr(dSt, dHW(t, St)) = 0, Corr(dSt, dmix(t, St)) = 1.
2 2

Z
VHW(T ) :=
0

2 HW(t, St)dt,

Z
Vmix(T ) :=
0

mix(t, St)dt

the average variances of the process S in the Hull-White model and in our mixture model, respectively. Then Corr(ST , VHW(T )) = 0, Corr(ST , Vmix(T )) = 0 all T Correlation

(Volatility)2 Asset-Value:
Hull-White SVM 0 0 lognormal mix dyn 1 0

Model Intantaneous correlation Terminal correlation

Yet, correlation is not a satisfactory measure of dependence outside the Gaussian world... Finally, in variants of the basic lognormal mixture model (e.g. mixtures with dierent drifts) one can impose correlation patterns as part of the calibration procedure

- Introduction to the mixture diusion dynamics for the volatility smile

15

Damiano Brigo

Banca IMI

Example 1 of calibration to market data, single smile

0.158 0.157 0.156 0.155 0.154 0.153 0.152 0.151 0.15 0.04 Market volatilities Calibrated volatilities

0.045

0.05

0.055

0.06

0.065

Data: Two-year Euro caplet volatilities as of November 14th, 2000 (Libor resetting at 1.5 years). We set: N = 2, T = 1.5, 2 = 1 1. We minimize the squared percentage dierence between model and market (mid) prices: 1 = 0.241, 2 = 0.759, 1(T ) = 0.125, 2(T ) = 0.194, = 0.147.
- Introduction to the mixture diusion dynamics for the volatility smile 16

Damiano Brigo

Banca IMI

Example 2 of calibration to market data, single smile

0.4 0.375 0.35 0.325 0.3 0.275 0.25 0.225 0.2 39000 41000 43000 45000 47000 48500 bid volatilities ask volatilities calibrated volatilities

Data: Italian MIB30 equity index on March 29, 2000, at 3,21pm (most liquid puts with the shortest maturity). We set N = 3, T = 0.063014, 3 = 1 1 2. We minimize the squared percentage dierence between model and market mid prices. We get: 1 = 0.201, 2 = 0.757, 1(T ) = 0.019, 2(T ) = 0.095, 3(T ) = 0.229, = 1.852.
- Introduction to the mixture diusion dynamics for the volatility smile 17

Damiano Brigo

Banca IMI

Example 3 of calibration to market data, single smile

0.1155 0.115 0.1145 0.114 0.1135 0.113 0.1125 0.112 0.1115 0.82 market volatilities calibrated volatilities

0.84

0.86

0.88

0.9

0.92

0.94

Data: USD/Euro two-month implied volatilities as of May 21, 2001. We set N = 2, T = 0.167, 2 = 1 1. We minimize the squared percentage dierence between model and market mid prices. We get: 1 = 0.451, 1(T ) = 0.129, 2(T ) = 0.114, = 0.076.
- Introduction to the mixture diusion dynamics for the volatility smile 18

Damiano Brigo

Banca IMI

Example 4, calibration of a whole FX vol surface


EUR/USD market volatility surface May 17, 2001. Vol surface function of (T t) and . FX market: quote volatilities in terms of ATM vol (AT M ), riskreversal (r ), strangle (s) (e.g. Malz (1997)). Common assumption for a interpolating functional form


(, T ) = AT M (T )2r(T )

1 +16s(T ) 2 + (rd rf + T

1 2

2
,

2
(T ) = e ln(St/X) rd T 4

2 2 )T

3 5

being the delta of a call option.


T O/N 1W 2W 1M 2M 3M 6M 9M 1Y 2Y AT M 13.50% 10.50% 10.40% 11.00% 11.15% 11.50% 11.85% 12.00% 12.05% 12.05% r 0.60% 0.60% 0.40% 0.40% -0.05% -0.05% -0.10% -0.14% -0.15% -0.15% s 0.29 % 0.29% 0.29% 0.30% 0.30% 0.30% 0.30% 0.30% 0.30% 0.30%
Bid/ask spread 2%-2.5% 2%-2.5% 1%-1.5% 0.35%-0.85% 0.30%-0.80% 0.30%-0.80% 0.30%-0.68% 0.30%-0.55% 0.25%-0.45% 0.25%-0.45%

Table 1: Market data for ATM implied vols, riskreversal and strangle prices as of May 17, 2001.
- Introduction to the mixture diusion dynamics for the volatility smile 19

Damiano Brigo

Banca IMI

0.15 0.14 0.13 0.12 0.11 1 0.9 0.8 0.7 0.6 0.5 0.4 Delta 0.3 0.2 0.1

0.2

0.4

1 1.2 Time to maturity (Y)

0.6

0.8

1.4

1.6

1.8

2 0

0.015 0.01 0.005 0 -0.005 1 0.9 0.8 0.7 0.6 0.5 0.4 Option Delta 0.3 0.2 0.1

0.2

0.4

0.6

0.8 T (Y)

1.2

1.4

1.6

1.8

2 0

Figure 1: The market implied volatility surface (above) and absolute dierence in implied volatility after calibration of the model with N = 3 (below) for the May 17, 2001 market data.
- Introduction to the mixture diusion dynamics for the volatility smile 20

Damiano Brigo

Banca IMI

Example 4 (contd)

Use Shifts variant of the lognormal mixture dynamics explore the eect of varying the number of basis densities n

i(T ) =
i

q R T 1
T 0

2 i (s)ds are taken as

(T ) = ai + bi 1 exp

T i



i + ci exp T

T i


,

(Nelson and Siegel (1987) for yield curves) Model parameters to calibrate: x = (1:N , 1:N , a1:N , b1:N , c1:N , 1:N ) has dimensionality 4N + N + (N 1) = 6N 1. minimize the sum of the relative squared discrepancies The resulting rootmeansquare error is 3 104 and 7 105 for calibrations with N = 2 and N = 4 respectively. The maximum error for any maturity is well below the corresponding bidask spread already with N = 3.

- Introduction to the mixture diusion dynamics for the volatility smile

21

Damiano Brigo

Banca IMI

0.6

0.5

0.4

0.3

0.2

(-0.0756, 0.57)

0.1 (-1.75, 0.43)

0 0 0.2 0.4 0.6 0.8 1

Figure 2: T i(T ; a, b, c, ) after calibration of the model with N = 2; We show (i, i) for each component.

- Introduction to the mixture diusion dynamics for the volatility smile

22

Damiano Brigo

Banca IMI

0.6

0.5

0.4

0.3 (0.68, 0.17)

0.2 (-0.424, 0.293)

0.1

(-0.783, 0.536) 0 0 0.2 0.4 0.6 0.8 1

Figure 3: N = 3
0.6

0.5

0.4

0.3

(0.577, 0.144)

0.2 (0.181, 0.0679) (-0.657, 0.568) 0.1 (0.134, 0.220) 0 0 0.2 0.4 0.6 0.8 1

Figure 4: N = 4. Notice the nested structure


- Introduction to the mixture diusion dynamics for the volatility smile 23

Damiano Brigo

Banca IMI

Baskets: Consistent multivariate extension

Single assets: S(t) = [S1(t), S2(t), . . . , S(t), . . . , Sn(t)] .


Mixture diusion model for each asset S: Find mix with dS = S [ dt + mix (t, S )dW ] pS (t) =

PN

k k=1 pS k (t)

1 k N where S , . . . , S , . . . , S are instrumental procs for S :

dS (t) = S dt + S dW , dW dW = , dt
Multivariate extension: def n n matrix C by C(t, x) C (t, x)

PN
=

k1 kn k1 ,..,kn (t) p k1 ,...,kn =1 1 n V

PN

(x) k k [S1 1 (t),...,Snn (t)]

k1 kn k1 ,...,kn =1 1 n

p k1

k ,..,kn V 1 (t) =

k k (t) , (t)

(x) k [S1 (t),...,Snn (t)]


,=1,...,n

Our extension is (Rapisarda (2001), Multi Variate Mixture Dynamics)

dS(t) = diag()S(t)dt + diag(S(t))C(t, S(t))d[W1 , .., Wn ]

and satises pS(t) =

PN
k1

1 1 k n p n ,...,kn =1

k k [S1 1 (t),...,Snn (t)]

- Introduction to the mixture diusion dynamics for the volatility smile

24

Damiano Brigo

Banca IMI

Multivariate Mixture Dynamics (MVMD)


dS(t) = diag()S(t)dt + diag(S(t))C(t, S(t))d[W1 , .., Wn ]

PN
(CC )(t, x) =

kn k1 ,..,kn 1 (t) p k1 (x) k1 ,...,kn =1 1 n V k [S1 ,...,Snn ]

PN

k1 kn k1 ,...,kn =1 1 n k ,..,kn V 1 (t) = N X k1 ,...,kn =1

p k1 (x) k [S1 (t),...,Snn (t)]

k k (t) , (t) k

,=1,...,n

dWi dWj = i,j dt, pS(t) =

1 1 nn

p k1 k [S1 (t),...,Snn (t)]

Can evaluate simple claims (e.g. call option) on basket Pn A(t) = i=1 ai Si (t) = a S one shot, since we know pS . Compare with naive numerical Euler or Milstein scheme for Monte Carlo, consisting of a time-discretization of the Simply-Correlated Mixture Dynamics, obtained by single mixture sdes by instantaneously correlated Brownian motions

dS = diag()Sdt + diag((mix(t, S)))diag(S)d[W1, .., Wn] pS(t) =?!? Both are consistent with single mixture smiles
dS = S [ dt + mix (t, S )dW ] pS =

PN

k k=1 pS k

- Introduction to the mixture diusion dynamics for the volatility smile

25

Damiano Brigo

Banca IMI

MVMD vs SCMD
dS(t) = diag()S(t)dt + diag(S(t))C(t, S(t))d[W1 , .., Wn ] pS(t) =
N X k1 ,...,kn =1

1 1 nn

p k1 k [S1 (t),...,Snn (t)]

VS

dS = diag()Sdt + diag((mix (t, S )) )diag(S)d[W1 , .., Wn ] dWi dWj = i,j dt, pS(t) =?!?,

Pros MVMD: 1) Terminal distribution of S T (and of basket AT = a S T ) can be simulated one-shot, no time discretization. 2) Explicit multivariate distribution that is the most natural non-trivial generalization of the scalar case, i.e. a multivariate mixture Cons MVMD: 1) Combinatorial explosion: A possibly large number of densities to mix in the multivariate mixture (typically N n , e.g. 310 = 59049). BUT... Typically N = 2, 3, and there is a hierarchy in the base densities: 1 > 2 >> 3 , so that with weights given by 1 2 3 only few multivariate densities have appreciable weights, thus easing the simulations. 2) No immediate statistical interpretation of if not as cross-sectional tting parameters... Pros SCMD: 1) A clear interpretation for as instantaneous correlation among the single names. 2) Number of densities to mix does not increase with n but remains equal to N . Cons SCMD: 1) Need time discretization to price simple claims on

AT = a S T . As T increases, we need more time steps.


- Introduction to the mixture diusion dynamics for the volatility smile 26

Damiano Brigo

Banca IMI

Dening the basket smile

B(t) =

b X i=1

wiSi(t); dSi(t) = (r(t)qi)Si(t)dt+(...)dWi

Moment-matching paradigm. Call qi the continuous dividends of Si, so that its risk-neutral drift is i = r qi. Find q in

dBt = (r(t) q)Btdt + (....)dWt, B0 = B(0)


such that
R(T )qT E BT = B(0)e = b X i=1

Si(0)e

R(T )qi T

= EB(T ).

Now that we have q , we may decide to quote basket implied volatilities by inverting BSs formula, by solving the following equation in V (T, K):
BSCall(Basket0 , K, T, R(T ), q, V (T, K)) = Model-Basket-Call(Basket0 , T, K).

From the multivariate model prices of basket European options on the right hand side, back out the basket implied volatilities V (K, T )/ T such that BlackScholes formulas with the synthetic dividend q reproduce such prices.
- Introduction to the mixture diusion dynamics for the volatility smile 27

Damiano Brigo

Banca IMI

Basket of 2 smiley assets, each modeled with a 2-mixture

We now give some examples with a basket of 2 assets.


1 2 Asset S1 , mixture with (1 , 1 (t)), (2 , 1 (t)) 1 1 1 2 Asset S2 , mixture with (1 , 2 (t)), (2 , 2 (t)) 2 2

Dene
j V (t) :=

 Z

1 t j 2 (u) du t 0 0 11 q t

1/2 0 1

B t CC j j j B V (t) = A + B @1 exp @ q AA j D
In all examples we will take: First Asset S1 : S1 (0) = 1, 1 = 5%

j D

B t C j + C exp @ q A j D

1 1 1 A1 = 0.3, B1 = 0.01, C1 = 0.01, D1 = 5, 1 = 0.6 1 1 2 2 2 A2 = .2, B1 = .001, C1 = .001, D1 = 3, 2 = 0.4 1 1

Second Asset S2 : S2 (0) = 1, 2 = 3%


1 1 1 A1 = 0.25, B2 = 0.008, C2 = 0.008, D2 = 4.8, 1 = 0.7 2 2 2 2 2 A2 = 0.35, B2 = 0.0008, C2 = 0.008, D2 = 2.8, 2 = 0.3 2 2

- Introduction to the mixture diusion dynamics for the volatility smile

28

Damiano Brigo

Banca IMI

Average volatilities in the mixtures components


0.32 0.3

0.28

0.26

0.24

0.22

0.2

10

12

14

16

18

20

Figure 5: Average vols V11 (below) and V12 (above) for the mixture
concurring to S1
0.36

0.34

0.32

0.3

0.28

0.26

0.24

10

12

14

16

18

20

Figure 6: Average vols V21 (below) and V22 (above) for the mixture
concurring to S2

- Introduction to the mixture diusion dynamics for the volatility smile

29

Damiano Brigo

Banca IMI

Example: 2 2 basket. Single smiles


1 2 S1 (0) = 1, 1 = 5%, 1 (t), 1 (t), 1 = 0.6, 2 = 1 1 1 1 1 1 2 S2 (0) = 1, 2 = 3%, 2 (t), 2 (t), 1 = 0.7, 2 = 1 1 2 2 2 (1) (2) AT = w1 S + w2 S , w1 = w2 = 0.5, T =1Y. T T
Implied vol. for asset no. 1

0.285 0.28 0.275 0.27 0.265 0.26

0.1

0.3 Implied vol. for asset no. 2

0.2

0.4

0.5

0.6

Maturity (y) 0.32 0.315 0.31 0.305 0.3 0.295 0.29 0.285 0.28

0.7

0.8

0.9

1 0.95 0.9 0.85 0.8

1.3 1.25 1.2 1.15 1.1 1.05 Strike

0.1

0.2

0.3

0.4

0.5

0.6

Maturity (y)

0.7

0.8

0.9

1 0.95 0.9 0.85 0.8

1.3 1.25 1.2 1.15 1.1 1.05 Strike

- Introduction to the mixture diusion dynamics for the volatility smile

30

Damiano Brigo

Banca IMI

2 2 MVMD vs SCMD basket densit, = 0.9


1 2 S1 (0) = 1, 1 = 5%, 1 (t), 1 (t), 1 = 0.6, 2 = 1 1 1 1 1 1 2 S2 (0) = 1, 2 = 3%, 2 (t), 2 (t), 1 = 0.7, 2 = 1 1 2 2 2 (1) (2) AT = w1 S + w2 S , w1 = w2 = 0.5, T =1Y. T T = 90%; the basket density pA(T ) (; ) under the SCMD-EMCscheme, continuous line; pA(T ) (; ) under the MVMD scheme, dashed line.

rho=-0.9 0.06 Euler dynamics Fast dynamics 0.05

0.04

0.03

0.02

0.01

0 40 60 80 100 120 140 160 180

- Introduction to the mixture diusion dynamics for the volatility smile

31

Damiano Brigo

Banca IMI

2 2 MVMD vs SCMD basket densities, = 0


1 2 S1 (0) = 1, 1 = 5%, 1 (t), 1 (t), 1 = 0.6, 2 = 1 1 1 1 1 1 2 S2 (0) = 1, 2 = 3%, 2 (t), 2 (t), 1 = 0.7, 2 = 1 1 2 2 2 (1) (2) AT = w1 S + w2 S , w1 = w2 = 0.5, T =1Y. T T = 0; the basket density pA(T ) (; ) under the SCMD-EMCscheme, continuous line; pA(T ) (; ) under the MVMD scheme, dashed line.

rho=0 0.025 Euler dynamics Fast dynamics

0.02

0.015

0.01

0.005

0 40 60 80 100 120 140 160 180

- Introduction to the mixture diusion dynamics for the volatility smile

32

Damiano Brigo

Banca IMI

2 2 MVMD vs SCMD basket densities, = 0.9


1 2 S1 (0) = 1, 1 = 5%, 1 (t), 1 (t), 1 = 0.6, 2 = 1 1 1 1 1 1 2 S2 (0) = 1, 2 = 3%, 2 (t), 2 (t), 1 = 0.7, 2 = 1 1 2 2 2 (1) (2) AT = w1 S + w2 S , w1 = w2 = 0.5, T =1Y. T T = 90%; the basket density pA(T ) (; ) under the SCMD-EMCscheme, continuous line; pA(T ) (; ) under the MVMD scheme, dashed line.

rho=0.9 0.016 Euler dynamics Fast dynamics 0.014

0.012

0.01

0.008

0.006

0.004

0.002

0 40 60 80 100 120 140 160 180

- Introduction to the mixture diusion dynamics for the volatility smile

33

Damiano Brigo

Banca IMI

2 2 calibrated MVMD basket smile, = 0.3


1 2 S1 (0) = 1, 1 = 5%, 1 (t), 1 (t), 1 = 0.6, 2 = 1 1 1 1 1 1 2 S2 (0) = 1, 2 = 3%, 2 (t), 2 (t), 1 = 0.7, 2 = 1 1 2 2 2 (1) (2) AT = w1 S + w2 S , w1 = w2 = 0.5, T =1Y. T T
The basket smile for = 30%.

Alternative dynamics, rho=0.3 Alternative dynamics

0.255 0.25 0.245 0.24 0.235 0.23 0.225 0.22 0.215 1.3 1.25 1.2 1.15 1.1 1.05 1 Strike/Spot 0.95 0.9 0.85 1 0.8

0.1

0.2

0.3

0.4

0.5

0.6

Maturity (y)

0.7

0.8

0.9

- Introduction to the mixture diusion dynamics for the volatility smile

34

Damiano Brigo

Banca IMI

2 2 calibrated MVMD basket smile, = 0.9


1 2 S1 (0) = 1, 1 = 5%, 1 (t), 1 (t), 1 = 0.6, 2 = 1 1 1 1 1 1 2 S2 (0) = 1, 2 = 3%, 2 (t), 2 (t), 1 = 0.7, 2 = 1 1 2 2 2 (1) (2) AT = w1 S + w2 S , w1 = w2 = 0.5, T =1Y. T T
The basket smile for = 90%.

Alternative dynamics, rho=0.9 Alternative dynamics

0.285 0.28 0.275 0.27 0.265 0.26 1.3 1.25 1.2 1.15 1.1 1.05 1 Strike/Spot 0.95 0.9 0.85 1 0.8

0.1

0.2

0.3

0.4

0.5

0.6

Maturity (y)

0.7

0.8

0.9

- Introduction to the mixture diusion dynamics for the volatility smile

35

Damiano Brigo

Banca IMI

2 2 MVMD basket smile at T = 6m against


1 2 S1 (0) = 1, 1 = 5%, 1 (t), 1 (t), 1 = 0.6, 2 = 1 1 1 1 1 1 2 S2 (0) = 1, 2 = 3%, 2 (t), 2 (t), 1 = 0.7, 2 = 1 1 2 2 2 (1) (2) AT = w1 S + w2 S , w1 = w2 = 0.5, T =6m. T T
The basket smile for T =6m for dierent s.

0.25

rho=-90% rho=-60% rho=-30% rho=0 rho=30% rho=60% rho=90%

0.2

0.15

0.1

0.8

0.85

0.9

0.95

1.05 Strike/Spot

1.1

1.15

1.2

1.25

1.3

- Introduction to the mixture diusion dynamics for the volatility smile

36

Damiano Brigo

Banca IMI

Conclusions and perspectives


Mixture diusion local volatility model: lognormal mixture and variants

Good and analytic calibration to market data Explicit dynamics, results, complete market, delta hedging Known marginal densities, unknown transitions Possible Monte Carlo (Euler scheme) Decorrelation between average volatility and underlying asset Nested structure for parameterization; Extension to multivariate diusion A model to coherently link a basket smile to the smiles of its single names
Current/Future work:

Stochastic volatility versions of the mixture dynamics lose market completeness but have known transition densities, more tractable for exotic options and diagnostics Develop analytical approximations for exotic options with smile (e.g. barrier); Diagnostic tests on future volatility structures following calibration; More tests on baskets and the role of correlation;

- Introduction to the mixture diusion dynamics for the volatility smile

37

Damiano Brigo

Banca IMI

Literature

Brigo, D., Mercurio, F. (2000) A Mixed-up Smile. Risk, September, 123-126. Brigo, D., Mercurio, F. (2001) Displaced and Mixture Diusions for Analytically-Tractable Smile Models. In Mathematical Finance - Bachelier Congress 2000, Geman, H., Madan, D.B., Pliska, S.R., Vorst, A.C.F., eds. Springer Finance, Springer, Berlin Brigo, D., Mercurio, F. (2001) Lognormal-Mixture Dynamics and Calibration to Market Volatility Smiles, International Journal of Theoretical and Applied Finance. Vol. 5, No. 4 (2002), 427-446. Extended version with F. Rapisarda featuring surface calibration presented by F. Rapisarda at the Annual Research Conference in Financial Risk, July 12-14, 2001 - Budapest, Hungary, downloadable at F. Rapisardas web site http://it.geocities.com/rapix/frames.html and at www.damianobrigo.it Brigo, D., Mercurio, F. (2001) Interest Rate Models: Theory and Practice. Springer Finance. Springer. Brigo, D., Mercurio, F., and Rapisarda, F. (2002), An Alternative Correlated Dynamics for Multivariate Option Pricing, Working paper presented by F. Rapisarda at the 2nd World Congress of the Bachelier Finance Soc., june 1215, 2002. Brigo, D., Mercurio, F., (2002), Analytical pricing of the smile in a forward LIBOR market model, Accepted for publication, extended version available at http://www.fabiomercurio.it Brigo, D., Mercurio, F., (2002), The general mixture-diusion dynamics for stochastic dierential equations with a result on the volatilityasset covariance, working paper, http://www.damianobrigo.it

- Introduction to the mixture diusion dynamics for the volatility smile

38

You might also like