Beruflich Dokumente
Kultur Dokumente
Kai Detlefsen Wolfgang Hrdle Institut fr Statistik and konometrie CASE Humboldt-Universitt zu Berlin http://ise.wiwi.hu-berlin.de/ detlefs http://ise.wiwi.hu-berlin.de/ haerdle
Motivation
1-2
Motivation
In March 1997 Bank of Tokyo suered a $ 83 million loss in the derivative markets because of a wrongpricing model. (Cont (2005)) In 1999 the derivative losses because of model risk summed up to $ 5 billion in the banking industry. (Williams (1999))
Motivation
1-3
Model risk
The choice of a good pricing model is essential. Schoutens et al. (2004) calibrate dierent stochastic volatility models (Heston, Bates, ...) and exponential Levy models (Variance Gamma, CGMY, ...) to an implied volatility surface and price exotic options (barriers, cliquets, ...). They nd huge price dierences among the models (up to 200%). The risk of using a wrong model for pricing options is called model risk.
Motivation
1-4
Calibration risk
After a model is chosen it has to be calibrated to observed data. Normally (e.g. Schoutens (2004)) option prices are used as input in the calibration. But other choices like the implied volatility surface are intuitively preferred by traders. We analyze how the prices of exotic options change with the calibration criterion. We observe signicant prices changes and conclude that there is also calibration risk.
Motivation
1-5
Aims
We analyze calibration risk in the popular option pricing model of Heston for exotic options. In particular, we analyze the price dierences with respect to option type time to maturity goodness of t
Motivation
1-6
1. motivation 2. introduction 3. models and data 4. calibration 5. exotic options 6. conclusions 7. outlook
introduction
2-7
Model uncertainty
Dierent parametric forms for the process of the underlying lead to dierent prices of exotic options although the plain vanilla prices coincide:
lookback 15 %
barrier 200 %
cliquet 40 %
Table 1: Price dierences of exotic options for dierent models (Schoutens et al. (2004)).
introduction
2-8
Risk measures
Financial risks are quantied by risk measures. Artzner et al. (1999) have introduced risk measures as monotone, translation invariant, subadditive, positive homogene functions. Such a risk measure represents the worst case expected payo for a class P of probabilistic models: (X ) = sup EP (X )
PP
introduction
2-9
Cont (2005) proposes a quantitative framework for measuring model risk that takes into account special features of model risk like bid-ask spreads or the existence of hedging instruments. Conts simplest measure for model risk is given by P (X ) = sup EP (X ) inf EP (X )
PP PP def
introduction
2-10
Risk measures
As this risk measure is given in terms of probability distributions we can use it also for quantifying calibration risk. Then the class of probabilistic models P will result from dierent calibration methods applied to one model. Thus we identify the price range for several exotic options using a time series of implied volatility surfaces.
introduction
2-11
Calibration methods
Calibration of option pricing models can be done in two ways: by Monte Carlo Markov Chains (see e.g. Eraker (2001)) by minimization of an error functional (see e.g. Bakshi et al. (1997)) We focus on the minimization of a quadratic error functional because it is most common in practice.
introduction
2-12
Calibration risk
Calibration risk arises from dierent specications of the error functional. We consider here as error measures dierences of prices or implied volatilities in absolute or relative terms. These measures have been applied before (see e.g. Schoutens et al. (2004), Mikhailov et al. (2003)) but never been analyzed together.
introduction
2-13
Options
We analyze calibration risk for the following options: up and out calls down and out puts cliquets Moreover, we look at the relation between calibration risk and model risk (understood as the choice of a parametric model).
3-14
Models
We focus on the Heston model because of its popularity in practice. In addition, we consider the Bates model, an extension of the Heston model. It gives a better t to data and hence we can analyze the impact of goodness of t for the same type of model. Moreover, these two models allow us to consider model risk and analyze its relation to calibration risk.
3-15
where the volatility process is modelled by a square-root process: dVt = ( Vt )dt + Vt dWt ,
(2)
3-16
The mean-reversion speed , the long vol and the short vol V0 control the term structure of the implied volatility surface (i.e. time to maturity direction). The correlation and the vol of vol control the smile/skew (i.e. moneyness direction).
3-17
(0.8,0.0,0.4)
Vola
Maturity
0.4
Moneyness
0.4 (0.8,1.0,0.3) 0.7 0.3 0.2 (0.8,0.0,0.3) Calibration Risk for Exotic 0.9 1.0 1.1 in the Heston model Options (1.2,0.0,0.3) 0.5
3-18
+ dZt Vt dWt
(2)
= ( Vt )dt +
where Z is a compound Poisson process. This model extends the Heston model and has three more parameters.
3-19
The meaning of the parameters , , , and V0 is the same as in the Heston model.
The parameters of the compound Poisson process control the smile/skew especially for short times to maturity.
3-20
Data
Eurex settlement volatilities of European options underlying : dax time period: March 2003 - April 2004 risk free interest rate: Euribor no dividends because dax is performance index Because of computation time we consider only one day each week. Hence, we consider 51 implied volatility surfaces.
3-21
Data
Arbitrage: The implied volatility surfaces have been preprocessed in order to eliminate arbitrage. Illiquidity: Only options with moneyness m [0.75, 1.35] for small times to maturity T 1 have been considered because of illiquidity.
3-22
0.38 0.36 0.34 0.32 0.3 0.28 0.26 0.24 0.22 0.2
3600
DAX
0.18 01-Apr-03
01-Oct-03
31-Mar-04
Figure 1: DAX and ATM implied volatility with 1 year to maturity on the trading days from 01 April 2003 to 31 March 2004.
3-23
calibration
4-24
Error functionals I
For the minimization we consider the four objective functions based on the root weighted square error:
n
ap =
rp =
wi (
calibration
4-25
Error functionals II
ai =
def i=1 n
ri =
def i=1
wi (
where mod refers to a model quantity and mar to a quantity observed on the market, P to a price and IV to an implied volatility.
Calibration Risk for Exotic Options in the Heston model
calibration
4-26
Data design
4.5 0.5
0.5
1.5
0.75
1 moneyness
1.25
Figure 2: Grid of the DAX implied volatility surface on March 1st, 2004. (only for moneyness between 0.5 and 1.5)
Calibration Risk for Exotic Options in the Heston model
calibration
4-27
Data design
The observations per day have a special design: data come in strings strings are not uniformly distributed in time-to-maturity strings have dierent moneyness ranges strings are moving in time strings disappear new strings appear
calibration
4-28
Data design
We use the RMSE to measure the dierence between market and model. In order to take into account the special data design we use weights such that each string get the same weight all observations in a string have the same weight These weights imply an average time to maturity of 2.02. Hence, it is a good weighting to analyze options with 1,2 or 3 years to expiration.
Calibration Risk for Exotic Options in the Heston model
calibration
4-29
Calibration method
The error functionals are minimized with respect to the model parameters by a global stochastic minimization routine. Cont et al. (2004) have shown that these error functionals can have local minima. Hence it is essential to use a stochastic/global optimizer. We apply dierential evolution (see Storn (1997)). Other people use gradient descent methods. But we have found them to be signicantly inferior to dierential evolution.
calibration
4-30
Calibration method
The plain vanilla prices are calculated by a method of Carr et al. (1999): C (K , T ) = exp{ ln(K )}
+
for a damping factor > 0. The function T is given by T (v ) = exp(rT )T {v ( + 1)i} 2 + v 2 + i(2 + 1)v
calibration
4-31
Calibration method
calibration
4-32
Calibration method
calibration
0.5 0.4 0.3 0.2 0.6 0.4 0.35 0.3 0.25 0.2 0.6 0.4 0.35 0.3 0.25 0.2 0.6 0.8 1 1.2 1.4 1.6 0.8 1 1.2 1.4 1.6 0.4 0.35 0.3 0.25 0.8 1 1.2 1.4 1.6 0.2 0.6 0.4 0.35 0.3 0.25 0.2 0.6 0.4 0.35 0.3 0.25 0.2 0.6 0.8 1 1.2 1.4 1.6 0.8 1 1.2 1.4 1.6 0.8 1 1.2 1.4 1.6
4-33
Figure 3: IVS on 25/06/03 for AI in Heston (maturities: 0.26, 0.52, 0.78, 1.04, 1.56, 2.08). (market: blue, model: red; X: moneyness, Y:iv)
Calibration Risk for Exotic Options in the Heston model
calibration
4-34
0.4 0.35 0.3 0.25 0.2 0.6 0.4 0.35 0.3 0.25 0.2 0.6 0.8 1 1.2 1.4 1.6 0.8 1 1.2 1.4 1.6
0.4 0.35 0.3 0.25 0.2 0.6 0.4 0.35 0.3 0.25 0.2 0.6 0.8 1 1.2 1.4 1.6 0.8 1 1.2 1.4 1.6
Figure 4: IVS on 25/06/03 for AI in Heston (maturities: 2.60, 3.12, 3.64, 4.70). (market: blue, model: red; X: moneyness, Y:iv)
calibration
4-35
AP RP AI RI
calibration
4-36
AP RP AI RI
exotic options
5-37
exotic options
5-38
The payos of the barrier options are based on a maximum/minimum in continuous time. In the simulations these extrema are replaced by maxima/minima over 250 days a year. Although there is some discretization bias it can regarded as payo of an approximate instrument.
exotic options
5-39
Barrier options
The prices of up and out calls are given by exp(rT ) E[(ST K )+ 1{MT <B} ] where MT = max St .
0tT def
exotic options
5-40
Barrier options II
We consider for the barrier B and the strike K B = 1 + T 0.2 K = 1 T 0.1 where T = 1, 2, 3 denotes time to maturity.
exotic options
up and out calls, T=3
5-41
1.2
1.15
1.1
1.05
0.95
Figure 5: Relative prices of the up and out calls in the Heston model for 3 years to maturity.
Calibration Risk for Exotic Options in the Heston model
exotic options
up and out calls, T=3
5-42
1.4
1.3
1.2
1.1
0.9
0.8
AP/RP
AP/AI
AP/RI
RP/AI
RP/RI
AI/RI
Figure 6: Relative prices of the up and out calls in the Bates model for 3 years to maturity.
Calibration Risk for Exotic Options in the Heston model
exotic options
5-43
Heston
Bates
T T T T T T
=1 =2 =3 =1 =2 =3
exotic options
5-44
Cliquet options
We consider cliquet options with prices exp(rT ) E[H] where the payo H is given by
N
H = min(cg , max(fg ,
i=1
def
min(cli , max(fl i ,
i Here cg (fg ) is a global cap (oor) and cg (fgi ) is a local cap (oor) for the period [ti1 , ti ].
exotic options
5-45
Here we consider three periods with ti = i T (i = 0, . . . , 3) and the 3 caps and oors are given by cg = fg = 0 cli = 0.08, i = 1, 2, 3 fl i = 0.08, i = 1, 2, 3
exotic options
cliquet options, T=3
5-46
1.06
1.04
1.02
0.98
Figure 7: Relative prices of the cliquet options in the Heston model for 3 years to maturity.
Calibration Risk for Exotic Options in the Heston model
exotic options
cliquet options, T=3 1.3
5-47
1.2
1.1
0.9
0.8
Figure 8: Relative prices of the cliquet options in the Bates model for 3 years to maturity.
Calibration Risk for Exotic Options in the Heston model
exotic options
5-48
Heston
Bates
T T T T T T
=1 =2 =3 =1 =2 =3
exotic options
5-49
Calibration risk leads to signicant price dierences. What is bigger/more important model or calibration risk? Are these risks independent?
exotic options
5-50
barrier 200 % 13 %
cliquet 40 % 6%
(Model risk from Schoutens et al (2004) for a wide range of models) Model risk is bigger than calibration risk.
Calibration Risk for Exotic Options in the Heston model
exotic options
up and out calls, T=3 1.02
5-51
0.98
0.96
0.94
0.92
0.9
0.88
0.86 AP RP AI RI
Figure 9: Bates prices over Heston prices for up and out calls with 3 years to maturity on 51 days.
Calibration Risk for Exotic Options in the Heston model
exotic options
5-52
cliquets
T T T T T T T T T
=1 =2 =3 =1 =2 =3 =1 =2 =3
Conclusion
6-53
Calibration risk
exotic prices from AP dier from exotic prices from RP,AI,RI price dierences grow for longer times to maturity price dierences bigger for barrier options than for cliquets models with better ts do not have less calibration risk
Conclusion
6-54
model risk and calibration risk not independent model risk (between Heston and Bates) smallest for AP and biggest for RI
Conclusion
6-55
If the choice of the parametric model is not clear for the considered option then calibration w.r.t AP minimizes risk. For a given model calibration w.r.t AI gives (relatively) stable parameters, good ts and exotics prices with small variances that lie between the prices from AP and RP calibrations.
Conclusion
6-56
Heston model
The existence of this calibration risk raises the question if the Heston model is appropriate at all. Tests on other models should be conducted to clarify if calibration risk is a general phenomenon in option pricing.
Outlook
7-57
Outlook
Traders in banks are not only interested in good calibrations. It is vital to have stable results, i.e. stable prices and greeks. Possible regularizations: parameters (nite dimensional) distributions of the stock process other liquid markets (e.g. variance swap market) How does regularization inuence calibration risk?
Outlook
7-58
Outlook
7-59
Outlook
7-60
Outlook
7-61
Williams, D. Models vs The Market: Survival of the ttest, Report FIN514, Meridien Research.