Variance estimation in linear models when p>n is an important problem. A good error variance estimator is needed for developing significance tests for the estimated coefficients as well as for variable selection criteria such as AIC, BIC,..

We review some existing estimators in literature and propose a new one.

We will present some simulation which compare the performance of the proposed estimator with other under different sparsity and correlation structures. We will also present some initial theoretical results.

This is a (work in progress) joint work with E. Ahmed, H. Kim and B. Yuzbasi