Data Scientist, Quantitative Analyst and Matematician

Research items

Option valuation with IG-GARCH model and a U-shaped pricing kernel

DOI: 10.1007/s00500-019-04236-4

Journal of Soft Computing

Published: 23 July 2019

The main subject of this articleis to question the use of such a U-shaped pricing kernel to improve option pricing performances in a non-Gaussian setting. Starting from the so-called inverse Gaussian GARCH model (IG-GARCH), known to provide semi-closed-form formulas for classical European derivatives when an exponential-affine pricing kernel is used, we build a new pricing kernel that is non-monotonic and that still has this remarkable property. Using a daily dataset of call options written on the S&P500 index, we compare the pricing performances of these two IG-GARCH models proving, in this framework, that the new exponential U-shaped stochastic discount factor clearly outperforms the classical exponential-affine one.

Discriminating between GARCH models for option pricing by their ability to compute accurate VIX measures

HAL : halshs-02323959

Journal of Financial Econometrics

Submitted: 16 December 2019

This project attempts to fill several gaps in the GARCH option pricing literature, in particular, from an empirical point of view. Firstly, in the spirit of Christoffersen et al. (2004) the aim of our study is to provide an intensive comparison analysis of empirical performances, in VIX index or options valuation, between different GARCH type models using Gaussian or non-Gaussian distributions under different classes of risk neutral measures. Furthermore, particular attention is granted on the choice of the information set (VIX, options, returns) in the estimation process. As a natural non-Gaussian alternative we favor the so-called NIG distribution not only because it is known to fit the statistical properties of asset returns remarkably but also because, combined with the Esscher and EGP SDF, the pricing equations may be solved explicitly. What is more, monotonic and non-monotonic pricing kernels performances, but few of them consider all these factors at the same time. Our study is a mean of making a contribution to understand the global impact of these complementary aspects (24 combinations of GARCH/distribution/SDF/estimation are tested). Secondly, inspired by the work of Hao & Zhang (2013) that proposes to explain the poor pricing performances of Gaussian GARCH models by their inefficiency to capture the variance risk premium, we also explore in this paper if it is possible to partly classify GARCH option pricing models by their ability to simply reproduce the VIX index. From purely numerical aspects, such a conclusion would be very interesting to backtest these models in an efficient way only using VIX information, when available, instead of complex option datasets.

A new two-step estimation strategy for non-Gaussian GARCH models

DOI: 10.13140/RG.2.2.10299.85282

This working-paper derives from a very simple finding: under Gaussian hypotheses, some GARCH-type models have outstanding properties (closed-form expressions for the VIX and/or option prices) that fail when NIG innovations are involved. Nevertheless, it is now well documented that Gaussian GARCH option pricing models produce poor pricing errors when compared with skewed and fat-tailed counterparts. Thus, inspired by the so-called quasi-maximum likelihood estimator, a new two-steps approach is provided to both take benefit of these remarkable features in Gaussian environment and work with more realistic distributions. This strategy estimates separately the volatility and the distribution parameters supposing Gaussian innovations in the first step to incorporate VIX or options information in the estimation process. In a second step, the NIG distribution is fitted from the residuals obtained in the previous stage. What is more, we provide an empirical test for our new estimation methodology on a large dataset of options written on the S&P500.

Featured research

Reinforcement Learning & portfolio allocation

The role of portfolio managers is to optimize the allocation of capital across various financial assets to maximize a chosen performance metric like expected returns or risk-adjusted returns. Reinforcement learning is a branch of Machine Learning where actions are taken in an environment to maximize the notion of a cumulative reward from arandom underlying environment. Using reinforcement learning, the task of portfolio selection and allocation can be automated wherein the system will provide you with an optimum portfolio which will most likely give you maximum returns.

Logo  data_driven

Logo  visualisation

Malagasy & Natural Language Processing NLP

Think of natural language processing here, which allows computers to interpret human language and helps decode meaning. The goal is to ensure that the conversational AI revolution benefits everyone in emerging markets, regardless of gender, geography, income or education. Machines are becoming more and more brilliant for understanding English, but what about the 25 million Madagascans who need to communicate in their mother tongue?

Social network & graph theory

Social network analysis focuses on the structure of ties within a set of social entities, e. g., persons, groups,organizations, and nations, and so on. Network analysis provides a versatile framework for modeling complex systems and because of its universal applicability it has been invented and rediscovered in many different disciplines. Each of these disciplines enriches the field by providing its own perspective and its own approaches. In a graph theoretical approach, a social network is conceptualized as a graph , that is, a set of vertices (or nodes, units, points) representing social actorsand a set of lines representing one or more social relations among them.

Logo  data_driven
Home
Home
Home
CV
Home
Case Studies
Home
Researchs