Nonparametric kernel estimation for error density

Summary form only given. Consider a linear model, y_i=x´ _iβ+e_i, i=1,2,..., x´_is are p(⩾1) dimension known vectors and β(∈R°) is an unknown parametric vector and e_i are assumed i.i.d.r.v.´s from a common unknown density function f(x) with med (e_i)=0. Based on LAD (least absolute deviations) estimator β˜ of β, we propose a nonparametric method to estimate unknown f(x). A kernel estimator f˜_n(x) is obtained. Large sample properties of f˜_n(x) are studied. Some computational examples are also given