Heteroscedasticity, multiple populations and outliers in trade data

Abstract

International trade data are often affected by multiple linear populations and heteroscedasticity. An immediate consequence is the false declaration of outliers. We propose the monitoring of the White test statistic through the Forward Search as a new robust tool to test the presence of heteroscedasticity. We briefly describe how the regression estimates change when considering a heteroscedastic regression model. We finally show that, if the data are analyzed on a monthly basis, the heteroscedastic problem can be often bypassed.