On Multivariate Methods in Robust Econometrics

This work studies implicitly weighted robust statistical methods suitable for econometric problems. We study robust estimation mainly for the context of heteroscedasticity or high dimension, which are up-to-date topics of current econometrics. We describe a modifi cation of linear regression resistant to heteroscedasticity and study its computational aspects. For a robust version of the instrumental variables estimator we propose an asymptotic test of heteroscedasticity. Further we describe robust statistical methods for dimension reduction and classifi cation analysis. We propose the robust quadratic classifi cation analysis based on a new minimum weighted covariance determinant (MWCD) estimator. In general the robust methods based on down-weighting less reliable observations are resistant to outlying values (outliers) and insensitive to the assumption of Gaussian normal distribution of the data. The methods are illustrated on econometric data examples.