Testing for Heteroskedasticity in The Presence of Outliers

Abstract

Regression analysis is prone to the issue of heteroscedastic data in a variety of real-world cases, including macroeconomic data. Thus, it is crucial to test the data for possible heteroskedasticity. This is important because if the data is found to be heteroskedastic. Then, this may seriously impact the regression analysis's estimation and testing phase. It is emphasized that real data may contain one or more outliers; thus, it is important to use the appropriate test to test for the presence of heteroskedasticity when there is evidence of outliers present in the data. A commonly used test to test for heteroskedasticity is the Goldfeld-Quandt (GQ) test. However, its performance becomes questionable when the data contains one or more outliers. A modified version of GQ (MGQ) is available in the literature that considers the issue of outliers into account while testing for possible heteroskedasticity in the data. Though this is a good addition to an existing stream of heteroskedasticity tests, little attention is given to literature regarding its applicability. The present study takes the lead and makes the case that practitioners should use this newly proposed MGQ test. Various real-world cases using popular data sets are discussed, indicating the superiority of MGQ over the conventional GQ test when there are outliers in the data. The findings based on real-world data indicate that practitioners should use the MGQ test whenever the data contains outliers to avoid misleading conclusions.