Panel Data Model Outlier Test

Outliers are frequently present in panel data, which may be caused by the use of inappropriate model or some unusual economic and financial phenomena. Therefore, the identification of outliers in panel data model is a very important work. However, this research was not paid great attention in current literature. In panel data model, because of the introduction of individual effect, the traditional mean-shift model can not be used to detect the ourliers and a new method need to be suggested. In this thesis, a variance-shift model is proposed to study the detection of outliers in panel data, and test statistics based on Lagrang Multiplier method are derived. The main results we obtained are:(1) The LM statistics and corresponding test procedures for multiple outliers are derived in unbalanced static panel data model, which including the balanced static panel data model as a special case.(2) The LM statistic and corresponding test procedures for multiple outliers are derived in static panel data model with time effect parameter.(3) The LM ststistic and corresponding test procedures for multiple outliers are derived in dynamic panel data model.(4) Some simulation studies and real data analysis are used to illustrat the proposed methodology.The panel data model is a widely used statistical model in economic analysisa. However it is crucial to establish a proper model for a given data. The use of inappropriate model may have a misleading for drawing the conclusion. The detection of outliers is one of important issues in statistical diagnostics, which is helpful for model building and the identification of some unusual characteristic appeared in the data. Thus the results obtained in this thesis have an important theoretical and applied values.