31.05.2014 Views

Processus de Lévy en Finance - Laboratoire de Probabilités et ...

Processus de Lévy en Finance - Laboratoire de Probabilités et ...

Processus de Lévy en Finance - Laboratoire de Probabilités et ...

SHOW MORE
SHOW LESS

You also want an ePaper? Increase the reach of your titles

YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.

100 CHAPTER 3. NUMERICAL IMPLEMENTATION<br />

mo<strong>de</strong>l from the time series of ass<strong>et</strong> r<strong>et</strong>urns. Since, un<strong>de</strong>r the conditions of Theorem 2.14,<br />

the calibration procedure yields a martingale probability equival<strong>en</strong>t to the prior, the choice<br />

of historical probability as prior <strong>en</strong>sures the abs<strong>en</strong>ce of arbitrage opportunity involving stock<br />

and options in the calibrated mo<strong>de</strong>l. Estimation of expon<strong>en</strong>tial Lévy mo<strong>de</strong>ls is not discussed<br />

here and interested rea<strong>de</strong>r is referred to [27, Chapter 7].<br />

Specific expon<strong>en</strong>tial Lévy mo<strong>de</strong>ls<br />

are discussed in [57] (Merton mo<strong>de</strong>l), [79] (g<strong>en</strong>eralized hyperbolic mo<strong>de</strong>l) and [21] (CGMY or<br />

tempered stable mo<strong>de</strong>l). [9] and [12] treat more g<strong>en</strong>eral jump-diffusion type mo<strong>de</strong>ls.<br />

To <strong>en</strong>sure the stability of calibrated Lévy measures over time, one can systematically choose<br />

the calibrated expon<strong>en</strong>tial Lévy mo<strong>de</strong>l of the previous day as today’s prior. This choice guarantees<br />

that the curr<strong>en</strong>t calibrated Lévy measure will be altered in the least possible way, to<br />

accommodate the arrival of new option pricing information.<br />

Alternatively, the prior can simply correspond to a mo<strong>de</strong>l that seems “reasonable” to the<br />

user (e.g. an analyst). In this case our calibration m<strong>et</strong>hod may be se<strong>en</strong> as a way to make the<br />

smallest possible change in this mo<strong>de</strong>l to take into account the observed option prices.<br />

Wh<strong>en</strong> the user does not have such <strong>de</strong>tailed views, it should be possible to g<strong>en</strong>erate a refer<strong>en</strong>ce<br />

measure P automatically from options data. To do this we choose an auxiliary mo<strong>de</strong>l Q(θ) (e.g.<br />

Merton mo<strong>de</strong>l) <strong>de</strong>scribed by a finite-dim<strong>en</strong>sional param<strong>et</strong>er vector θ and calibrate it to data<br />

using the least squares procedure:<br />

θ ∗ = arg min<br />

θ<br />

‖C Q(θ) − C M ‖ 2 w (3.9)<br />

It is g<strong>en</strong>erally not a good i<strong>de</strong>a to recalibrate this param<strong>et</strong>ric mo<strong>de</strong>l every day, because in this<br />

case the prior will compl<strong>et</strong>ely lose its stabilizing role. On the contrary, one should try to find<br />

typical param<strong>et</strong>er values for a particular mark<strong>et</strong> (e.g.<br />

averages over a long period) and fix<br />

them once and for all. Since the objective function in (3.9) is usually not convex, a simple<br />

gradi<strong>en</strong>t <strong>de</strong>sc<strong>en</strong>t procedure may not give the global minimum. However, the solution Q(θ ∗ ) will<br />

be corrected at later stages and should only be viewed as a way to regularize the optimization<br />

problem (2.4) so the minimization procedure at this stage need not be precise.<br />

Theorem 3.2 shows that the calibrated solutions are continuous with respect to the prior,<br />

that is, small changes in the prior process induce small changes in the solutions.<br />

To assess<br />

empirically the influ<strong>en</strong>ce of finite changes in the prior on the result of calibration, we have<br />

carried out two series of numerical tests. In the first series of tests, Lévy measure was calibrated

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!