Calibrating Stochastic Volatility Models with Time-based Rescaling of Option Data

A key issue in practical option pricing with stochastic volatility models is the choice of feasible model parameters, which can be realized by calibration. Aiming to improve the reliability of calibration results for out-of-sample option pricing without changing original model setting, we introduce a Bayesian approach for calibration with time-based rescaling of option data, which is realized by assuming pricing errors with time-varying variances in the state-space model. The calibration procedure is implemented with Markov chain Monte Carlo (MCMC) algorithm. Empirical tests in SPX option market show strong preference for models calibrated with our method in out-of-sample option pricing, compared with models calibrated without any rescaling of data.