abstract
- In Econometrics, the Breusch-Pagan test-statistic has become an iconic application of the Lagrange multipliers (LM) test. We shall introduce beta-score LM tests for heteroscedasticity in linear regression models, which trades-off the degree of robustness and efficiency is through a tuning parameter beta>=0, being beta =0 the classical Breusch-Pagan test-statistic, the most efficient one under absence of outliers. A very elegant expression is obtained, with an appealing least squares interpretation. The construction of the test-statistic is performed extending the methodology of Basu et al. (2022) from identically distributed to non-identically distributed individuals, for composite null hypotheses. Detailed theoretical justifications about robustness and efficiency properties are given, all of them under normality. A modified version is derived, the Koenker's beta-score test-statistic.