Bayesian Semiparametric Approach for Stochastic Volatility Model

• Activity Number: 601
• Type: Contributed
• Date/Time: Wednesday, August 12, 2015 : 2:00 PM to 3:50 PM
• Sponsor: Section on Bayesian Statistical Science
• Author(s): Peng Sun* and Inyoung Kim and Kiahm Lee
• Companies: Virginia Tech and Virginia Tech and Seoul National University
• Keywords: Bayes factor ; Weighted Dirichlet process ; Metropolis-Hastings ; Model selection ; Parameterization ; Stochastic volatility

Abstract

We propose weighted Stochastic Volatility (SV) model which is developed under the semiparametric Bayesian framework to study the stock daily return. Traditional SV models specify the distribution of the error term which is the difference between the target variable and log-volatility. However, when it comes to daily returns of individual stock, the distribution should be discreetly chosen because a non-ignorable proportion of zero returns can make the data deviate a lot from the specified distribution. Our weighted SV model does not specify the distribution of error. Each observation contain a vector of weights which is equivalent to the distribution of what candidate prior can be selected as this observation's actual prior of error term. The value of a weight is determined by the distance between the related observation and candidate in terms of their predictive information. Our approach has a good property that a nearer candidate is always more likely to be chosen than a farther one, and it is able to greatly reduce computational burden compared with canonical weighted method while producing better marginal likelihoods under several circumstances based on our empirical results.