Panel data modeling: Identifying and handling outliers with the VSOM approach

Abstract

While modeling panel data, identifying outliers is crucial, yet visually spotting them can be challenging. However, not many methods exist for outlier detection in panel data models other than robust estimation methods. This study uses the variance shift outlier model (VSOM) approach to identify and handle outliers in panel data modeling, both in single and simultaneous equation modeling. This study applies the VSOM approach to GDP and FDI data from ten countries that are part of the ASEANChina Free Trade Area (ACFTA). The results demonstrate that the VSOM approach effectively handle outliers in panel data modeling, in both single and simultaneous equation model. Specifically, the VSOM model shows a lower sum of squared residuals (SSR) compared to the null model. This indicates that VSOM successfully reduces the impact of outliers, leading to improved model fit and accuracy in capturing the underlying relationships within the data. Some of the highlights of the proposed approach are:•Identifying the square of the standardized residual.•Generate the distribution of the squared standardized residuals by parametric bootstrapping.•Reducing the impact of observations identified as outliers by downweighting their variance through the addition of a D matrix.