Strong representations for LAD estimators in linear models (Q1116225): Difference between revisions

Consider the standard linear model \(y_ i=z_ i\beta +e_ i\), \(i=1,2,...,n\), where \(z_ i\) denotes the i th row of an \(n\times p\) design matrix, \(\beta \in {\mathbb{R}}^ p\) is an unknown parameter to be estimated and \(e_ i\) are independent random variables with a common distribution function F. The least absolute deviation (LAD) estimate \({\tilde \beta}\) of \(\beta\) is defined as any solution of the minimization problem \[ \sum^{n}_{i=1}| y_ i-z_ i{\tilde \beta}| =\inf \{\sum^{n}_{i=1}| y_ i-z_ i\beta |:\quad \beta \in {\mathbb{R}}^ p\}. \] Bahadur type representations are obtained for \({\tilde \beta}\) under very mild conditions on F near zero and on \(z_ i\), \(i=1,...,n\). These results are extended to the case, when \(\{e_ i\}\) is a mixing sequence. In particular, the results are applicable when the residuals \(e_ i\) form a simple autoregressive process.