Title: A New Class of Asset Pricing Models with Lvy processes: Theory and Applications Technical Slides Gen
1A New Class of Asset Pricing Models with Lévy
processes Theory and Applications(Technical
Slides General Framework)
www.ornthanalai.com
Nov 2008
2 Methodological Contribution New Framework
- I provide a general solution for the pricing
transform of a large class of asset return
dynamics - The setup allows for a wide variety of asset
return specifications - Closed-form valuation of various derivatives
What do returns look like?
Shocks to return
Lévy processes Normal, Poisson, Gamma, VG,
NIG, CGMY, IG, LS, Meixner, TS
How does the return volatility behave?
Affine GARCH dynamics
- Persistence long memory?
- Jumps in volatility?
3 The General Framework
- Assume d-dimensional contemporaneously
independent Levy shocks in the return - Each of the Levy shocks consists of a
parameter that controls the dynamic of
its variance and higher moments - I use affine GARCH dynamics to drive
for i 1..d
The Return Dynamic under the Physical Measure
4 The General Solution to the Pricing Transform
- The MGF of asset return at time T conditional on
current period t is
- The general solution to the MGF has an
exponential affine form
5 The General Solution to the Pricing Transform
- The MGF of asset return at time T conditional on
current period t is
- The general solution to the MGF has an
exponential affine form
- I provide a general solution for the affine
coefficients in Proposition 1
6A New Class of Asset Pricing Models with Lévy
processes Theory and Applications(Technical
Slides MLIS/Estimation)
www.ornthanalai.com
October 2008
7 Affine GARCH(1,1) dynamic
- The return dynamic
- The affine GARCH(1,1) dynamic
-
- We need to filter out zt expost of Rt ,
equivalently - Note that the filtering problem of
are equivalent
8 Why is MLE easier with Lévy GARCH ?
Continuous-time
The density function involves triple integration
over three sources of randomness
Lévy GARCH
The density function involves one integration as
diffusive variance and the jump intensity are
deterministic
9 Estimation of Lévy GARCH Particle Filter
- I use Particle Filtering (PF) in the estimation
- It can handle nonlinearity such as jumps
- It is relatively fast and lends itself naturally
to MLE - MLE PF gt Maximum Likelihood Importance
Sampling, see Pitt (2002), and Neil and Sheppard
(2000) - The sampling technique is advantageous because
many Lévy processes do not have analytical
density function
10 Estimation of Lévy GARCH Particle Filter
- Note that the filtering problem of
- are equivalent
- The filtering problem of yt involves computing
- The sampling density of yt is
- The sampling technique is advantageous because
many Lévy processes do not have analytical
density function
Sampling
Re-sampling
11 Estimation of Lévy GARCH MLIS
- The log likelihood function is computed from
- This probability function is the same as the one
used in the particle filtering of yt (see last
slide) - The likelihood is already computed as part of
the PF procedure - Relatively fast, efficient, and does not rely on
the analytical likelihood function