Second order representations of the least absolute deviation regression estimator (Q1388164): Difference between revisions

For the regression coefficients in a classical linear regression model, the least absolute deviation estimators (LADs) are robust alternatives to the famous least squares estimators. There is no, however, explicit expression for LAD estimators, which makes the discussion of their statistical properties more difficult. Some authors [e.g., \textit{C. R. Rao} and \textit{L. C. Zhao}, Can. J. Stat. 20, No. 4, 359-368 (1992; Zbl 0767.62056), and \textit{X. He} and \textit{Q.-M. Shao}, Ann. Stat. 24, No. 6, 2608-2630 (1996; Zbl 0867.62012)] derived weak or strong Bahadur-Kiefer representations for a LAD estimator, which provides an approximation to the LAD estimator by the sum of a sequence of i.i.d. random variables. With these representations, asymptotic properties of LAD estimators would be easily derived from the classical results for the sum of i.i.d. random variables. Based on some new asymptotic results obtained by the author for some processes of sums of independent, but not necessarily identically distributed random variables and the Talagrand isoperimetric inequality [\textit{M. Ledoux} and \textit{M. Talagrand}, Probability in Banach spaces. Isoperimetry and processes. (1991; Zbl 0748.60004)], some new weak or strong Bahadur-Kiefer representations for a LAD estimator are derived under minimal conditions.