Applications of Class of Symmetric Mesokurtic Error Distributions in Auto Regressive Models

This paper examines a family of symmetric mesokurtic distributions proposed by Sebastian and James (2002), employed as error distributions in Auto Regressive models. As maximum likelihood equations become intractable under such settings, the Modified Maximum Likelihood Estimation (MML) approach is adopted to obtain parameter estimates. The asymptotic properties are discussed, and simulation experiments are carried out to evaluate the estimators. Results from the simulation show that the proposed estimates perform better than least squares estimators with respect to standard errors for small sample sizes, and the relative efficiency tends to one as n increases.