Abstract
A method for estimating parameters in non-Gaussian moving average models is proposed based on Bayesian analysis. In the conventional approach, the Gaussian likelihood is used for parameter estimation (QMLE). However, it may not be appropriate when the process is non-invertible. Huang and Pawitan (2000) proposed another likelihood-based estimation method using Laplace likelihood (H&P). Comparing among these three methods, the empirical results show that the Bayesian analysis performs better than QMLE and H&P in terms of smaller root mean square error no matter the MA process is invertible or non-invertible.