A1217
Title: How to improve outliers detection in panel data
Authors: M Cristina Miranda - University of Aveiro (Portugal)
Manuela Souto de Miranda - University of Aveiro (Portugal)
Anabela Rocha - University of Aveiro (Portugal) [presenting]
Abstract: Panel data is the framework of many econometric studies. However, real-world data sets often include atypical values far away from the main pattern suggested by the majority of the data. Traditional methods for parameter estimation, such as least squares, can be seriously affected by violation of the model assumptions, producing unreliable estimates. The focus is on adapting recent techniques of cellwise and case-wise outlier detection to panel data, presenting a robust estimate applied to a random effects model. It consists of a modified version of the feasible generalized least squares estimate that incorporates robust choices in the sequential estimation process. The performance of the new robust proposal is presented with a simulation study and also evaluated with a real-world panel data set.
