Title: Constant horizon implied volatility (Constant horizon implied statistics with the GEV option pricing model)
1Constant horizon implied volatility(Constant
horizon implied statistics with the GEV option
pricing model)
3rd 7th September 2007, CCFEA Summer School,
University of Essex
- Amadeo Alentorn
- Old Mutual Asset Managers
- Papers available at
- http//privatewww.essex.ac.uk/aalent
2Contents
- Implied volatility
- Implied volatility surfaces
- Implied risk neutral densities
- The GEV option pricing model
- Empirical results
- Applications
3Implied volatility
- The implied volatility of an option contract is
the volatility implied by the market price of the
option based on an option pricing model. - Traders often quote options in terms of implied
volatility rather than price, as its a more
useful measure of an options relative value. - Different ways to calculate it
- Parametric models like, Black-Scholes
- non-parametric methods, like the method to
calculate the implied volatility index (VIX)
4Implied volatility smile
- In general, options on the same underlying but
with different strikes will yield different
implied volatilities. - In the Black-Scholes model Implied volatility
is the wrong number to put into wrong formulae to
obtain the correct price.
5Implied volatility term structure
- Implied volatilities also vary across time to
maturity (Dumas et al, 1998).
6Beyond Black-Scholes
- Most efforts in the literature at addressing
these Black-Scholes anomalies have centred around
assuming - either a more suitable return process, such as
- Introducing jumps
- Stochastic volatility models
- or fitting parametric models to traded option
prices - implied volatility surfaces
- Implied risk neutral densities
7Implied volatility surfaces
- An implied volatility surface is a 3-D plot that
combines volatility smile and term structure of
volatility into a consolidated view of all
options for an underlying. - Dumas, Fleming and Whaley (JoF, 1998) propose
several parametric models that attempt to capture
variation attributable to both strike level and
time. - A similar approach was taken by Tompkins (2001),
but using a cubic function of time to maturity,
instead of quadratic.
8Implied volatility surfaces
9Implied risk neutral densities
- Another approach to option pricing is based on
assuming a particular distribution for the
underlying price at maturity. - There is a large number of models, studying
different distributional assumptions - Mixture models
- Mixture of lognormals (Bahra 1996 and 1997)
- Specific distributions,
- Weibull distribution (Savickas 2002, 2005)
- Generalized distributions,
- Generalized Lambda Distribution (Corrado 2001)
10The implied risk neutral density
- It can be interpreted as the markets aggregate
risk neutral probability distribution for the
price of the underlying asset at maturity T. - It can be extracted it using a cross section of
traded option prices, with different strikes but
same maturity (Breeden and Litzenberger, 1978).
11Time to maturity effects
- When calculating a time series of implied RNDs or
RND related statistics we encounter the problem
with the time to maturity effect. - This happens due to the fixed maturity of
options. - As we approach maturity, the time horizon of the
implied RND shortens, the degree of uncertainty
decreases, and thus, densities of consecutive
days are not directly comparable. - Clews et al. (2000) propose a method to solve
this problem, by calculating a constant time
horizon implied statistics by linear
interpolating between two maturities that bracket
the horizon of interest. - The GEV model we present here is more general, as
we estimate the implied density across all
maturities.
12Term structure of implied RNDs
- In most markets, there are option contracts
trading for different maturities T1, T2, TN. - For example, for the FTSE 100, there are traded
options with maturities on the closest 3 months,
and also quarterly (Mar, Jun, Sep, and Dec). - At any given day, we can extract a term structure
of RNDs from traded option prices.
13Term structure of implied RNDs
14The Implied RND surface
- Usually, when estimating implied RNDs, a separate
density is obtained for each maturity. For a
given day, we obtained a different set of
parameters for each of the maturities. - Here, we will consider an implied RND surface for
the GEV model. For a given day, we want to obtain
a unique set of parameters, consistent with
option prices for all strikes and maturities. - This is similar to implied volatility surfaces,
where the aim is to have a parametric model that
fits option prices across both dimensions.
15The implied GEV RND surface
- The implied RND surface as a function of time and
price.
Index level
16The GEV option pricing model
- It is an option pricing model based on the
Generalized Extreme Value (GEV) distribution.
This model - Removes pricing biases associated with
Black-Scholes - Capture the stylized facts of the implied RNDs
- Left skewness
- Excess kurtosis (fat tails)
- Has a closed form solution for European options
- Delivers constant time horizon implied statistics
- Delivers the market implied tail shape
- We make no assumption about the price process,
only on the functional form of the terminal
distribution.
17The GEV distribution
- It is a distribution from Extreme Value Theory,
and is defined as the limiting distribution of
block maxima. - The standardized GEV distribution is given by
- where
- µ is the location parameter
- s is the scale parameter
- ? is the tail shape parameter
18The GEV for different values of ?
19Option pricing approach
- Our option pricing approach is based on the
estimating the implied Risk Neutral Density (RND)
using traded option prices. - Following Harrison and Pliska (1981) there
exists a risk neutral density (RND) function,
g(ST), such that the call option price can be
written as - where EQ is the risk-neutral expectation
operator, conditional on all information
available at time t.
20Moments of the GEV distribution
- The first two moments of the GEV distribution
are - In order to model the implied RND surface across
time, we will allow the first two moments to vary
with time to maturity T.
21Time scaling of volatility
- Back-Scholes assumes that volatility scales with
T1/2 - We explicitly model the scaling of implied
volatility with time to maturity by introducing a
scaling parameter, similarly to parametric models
for IV surfaces. - The volatility of a GEV distribution is given by
- We allow for time scaling of volatility with Tb
22The forward no-arbitrage condition
- The forward no-arbitrage condition states that
- The mean of the GEV distribution is
- We can rewrite the location parameter µ to be a
function of time T, the other parameters, and the
known Futures price Ft,T, thus removing one
degree of freedom
23Solving the pricing equation
- The price density function, (when negative
returns RT are GEV distributed) is given by - After some rearranging, the call option pricing
equation that needs to be solved is
24The GEV closed form solution
- Using the Gamma function, we can obtain a closed
form solution for the call option pricing
equation under GEV returns - where
- Similarly, we obtain an equation for put options.
25Estimation of the RND surface
- For a given day, we have M maturities available,
with Nj traded option prices each. - Using non-linear least squares, we can estimate
the set of parameters that minimizes the sum of
squared errors. - We price both calls and puts simultaneously.
- Note how the optimization problem is across both
strikes and maturities
26Pricing performance
- We estimate an implied GEV RND surface on a daily
basis from 1997 to 2003, using closing prices for
all traded option. - We also estimate the Black-Scholes (BS) and
Mixture of lognormals (MLN) models. - The GEV model exhibits lower pricing errors (in
sample).
27Time series of pricing errors
28Hedging performance
- The GEV model also outperforms the Black-Scholes,
and IV model in most cases (out of sample
performance).
29Volatility scaling parameter
- Recall that the parameter b gives the scaling law
for the implied volatility (assumed to be 0.5 in
Black-Scholes). - We find that in 93 of days, H0 b 0.5 can be
rejected. - To test the economic significance of having b, we
re-estimated the GEV model fixing b 0.5 and
found that the average pricing error doubled. - Even in that case, the GEV model still
outperformed the MLN model, despite of the MLN
model now having two extra degrees of freedom.
30Volatility scaling parameter
31Applications
32Implied tail shape parameter ?
33Implied volatility (VIX)
34Implied skewness (SIX?)
35Implied kurtosis (KIX?)
36Event studies Asian crisis
- We can compare the implied RNDs before and after
a crisis, for a constant horizon, to try an asses
the change in market sentiment reflected by
option prices.
37Event studies statistics
- In all three events, substantial increase in
implied volatility and in higher moments. - Note that ? only substantially increased in the
first two cases (due to 9/11 not being a
financial event?)
38Constant horizon E-VaR
- Another interesting constant horizon implied
measure we can obtain from the implied RND
surface is Ecomic VaR. - Alentorn and Markose (2006) showed how to obtain
constant horizon E-VaR using a two step process - First, estimation of the implied RND term
structure - Second, a linear regression on the log-log plot.
- With the implied RND surface we can obtain a
E-VaR value for any maturity and any confidence
level. - E-VaR values from the implied surface were found
slightly less volatile, probably due to using a
more robust estimation method.
39Conclusions
- We showed how the flexibility of the GEV
distribution allows us to capture different
levels of skewness and kurtosis. Unlike other
models, we dont have to specify the type of
distribution a priori (i.e. Weibull, Fréchet,
Gumbel). - The GEV option pricing model removes the well
known Black-Scholes pricing biases. It also
outperforms the mixture of lognormals method. It
also exhibits superior hedging performance. - By estimating an implied RND surface across both
strikes and maturities, we can easily obtain
constant horizon implied statistics. - The model also delivers the implied tail shape
parameter, which controls the implied skewness
and the fatness of the tails, and can be used to
asses (risk neutral) market expectations of
extreme outcomes. We find it increase after
crisis events.