Title: Likelihood evaluation through particle filter methods for high-frequency stochastic volatility models
Authors: Antonio Santos - University of Coimbra (Portugal) [presenting]
Abstract: Volatility analysis is crucial for financial decision-making. High-frequency volatility analysis is feasible from intraday data availability, and through the application of stochastic volatility models to such data. New elements appeared as critical in the full characterization of high-frequency volatility evolution, and examples are the two-factor model and the presence of jumps in volatility. The variety of models increased, and using a Bayesian paradigm to estimate the models' parameters; direct likelihood evaluations are not feasible. The log-likelihood function value is essential to calculate the Bayes' factors, which is a measure used to choosing between competing models. A strategy is applied for evaluating the log-likelihood function of such complex models. The log-likelihood function evaluation makes use of techniques like importance sampling and particle filter methods. The results allow defining the best type of stochastic volatility model to use for volatility evolution analysis on two equities traded in US markets.