Abstract
Stochastic volatility (SV) model is a popular model for characterizing time-varying variance for return data. Due to some regulation rules in the financial market, observed returns for assets sometimes have truncations. Adapted the idea of Park, Genton and Ghosh (2007) to deal with the truncated data in fitting an ARMA model, this thesis suggests an estimation method to deal with the truncated return data in fitting SV model, which incorporates an imputation step in the maximum likelihood estimation. We demonstrate the efficiency gain of the truncation-adjusted estimator over the unadjusted estimator by comparing their trace of the inverse Fisher information matrices via simulations. The applications to HTC and WINTEK returns are provided for illustration. Keywords: stochastic volatility model, truncated data, ARMA, data augment, estimation, Fisher information matrices.