Comparing generalized least squares - quantile regression with ordinary least squares in handling both heteroscedasticity and outliers simultaneously
Modeling in the presence of outliers and heteroscedasticity is a substantial problem in the realm of robust statistical analysis, both of which are essential components for drawing valid conclusions. This study focuses on the challenge of resolving problems arising from homoscedasticity violations when outliers are present, as conventional regression models can become unstable in these situations. The goal of this study is to create a robust regression model that successfully tackles these issues by simultaneously assessing Generalized Least Squares (GLS) and Quantile Regression (QR). Real life dataset for some macroeconomic indicators were obtained from the website of the Central Bank of Nigeria for period, 1981 to 2020. The regression models were obtained using Statistical Package for Social Sciences, Number Cruncher Statistical Systems and R program. Poverty Reduction is the dependent variable, while Money Supply, Government Expenditure, Government Revenue, and Financial Inclusion are the independent variables. The estimator of GLS and QR was combined in other to have GLS-QR. The proposed method exhibits better resistance against outliers and heteroscedasticity. These strategies are useful in addressing complex data, as demonstrated by real- life applications. This research provides a helpful toolkit for practitioners and academics to model complex datasets more accurately and robustly.
Keywords: GLS-Quantile Regression, Heteroscedasticity, Least Squares, Outliers
*Correspondence: ahmed.olasupo@oouagoiwoye.edu.ng, 08063421603
Download Paper