1: \begin{abstract}

2:   We consider the high-dimensional heteroscedastic regression model,

3:   where the mean and the log variance are modeled as a linear

4:   combination of input variables. Existing literature on

5:   high-dimensional linear regression models has largely ignored

6:   non-constant error variances, even though they commonly occur in a

7:   variety of applications ranging from biostatistics to finance. In

8:   this paper we study a class of non-convex penalized pseudolikelihood

9:   estimators for both the mean and variance parameters. We show that

10:   the Heteroscedastic Iterative Penalized Pseudolikelihood Optimizer

11:   (HIPPO) achieves the oracle property, that is, we prove that the

12:   rates of convergence are the same as if the true model was known. We

13:   demonstrate numerical properties of the procedure on a simulation

14:   study and real world data.

15: \end{abstract}

16: