Data Filtering Rules, Sample Selection, Sample Composition, Option Pricing, Out-of-Sample Pricing, Loss Function
Abstract :
[en] The choice of data filtering rules are, next to model selection and parameter calibration, an important step in an option pricing exercise. We illustrate the implications of data filtering rules, by investigating three alternative renowned filtering rules in the context of pricing European S&P 500 index options. Different filtering rules result in strongly diverging samples, which carry different information and therefore lead to different parameter estimates. This is illustrated for the Ad Hoc Black-Scholes model.
No filtering rule is, in terms of pricing performance, superior on the whole range of options. Instead, each filtering rule is specialized toward better pricing of options types that were included in the calibration sample at the costs of excluded options. Included options are unable to perfectly represent the properties and characteristics of excluded options.
In particular, option prices are heterogeneous in the maturity dimension, which is a major driving force underlying the impact of exclusion filters on pricing performance. Additionally, small deviations from the put-call parity strongly affect parameter estimates as well as the accompanying pricing performance. These results emphasize the prominent role of filtering rules as an important implicit choice for an option pricing model calibration.
Disciplines :
Finance
Author, co-author :
Bams, Dennis
Blanchard, Gildas
Lehnert, Thorsten ; University of Luxembourg > Faculty of Law, Economics and Finance (FDEF) > Luxembourg School of Finance (LSF)