1: \begin{abstract}

2: We propose a new class of estimators of the multivariate response linear regression 

3: coefficient matrix that exploits the assumption that the response and 

4: predictors have a joint multivariate Normal distribution. This allows 

5: us to indirectly estimate the regression coefficient matrix through 

6: shrinkage estimation of the parameters of the inverse regression, 

7: or the conditional distribution of the predictors given the responses. 

8: We establish a convergence rate bound for estimators in our class

9: and we study two examples. The first example estimator exploits an 

10: assumption that the inverse regression's coefficient matrix is sparse. 

11: The second example estimator exploits an assumption that the inverse regression's coefficient 

12: matrix is rank deficient. These estimators do not require the popular 

13: assumption that the forward regression coefficient matrix is sparse or has 

14: small Frobenius norm.  Using simulation studies, we show that our example estimators

15: outperform relevant competitors for some data generating models. 

16: \end{abstract}

17: