Finite Sample Performance of Standard Error Estimators for Dynamic Factor Analysis of Non-Normal Data Using the Kalman Filter Algorithm

This master thesis is concerned with the finite sample properties of four standard error (SE) estimators for dynamic factor analysis using the Kalman filter algorithm with both normal and nonnormal data. The estimators considered are the observed information based SE estimator, Harvey's SE estimator, and the two sandwich type SE estimators. Statistical properties of these estimators are assessed using a simulation study. Results indicate that the sandwich type SE estimator proposed by Papanastassiou (2006) generally outperforms other SE estimators. However, the observed information SE estimator is still valuable in that the advantage of the sandwich type SE estimator proposed by Papanastassiou (2006) over the observed information SE estimator for non-covariance component parameters is limited.