1: \begin{abstract}

2: This paper shows that large nonparametric classes of conditional

3: multivariate densities can be approximated in the Kullback--Leibler

4: distance by different specifications of finite mixtures of normal

5: regressions in which normal means and variances and mixing

6: probabilities can depend on variables in the conditioning set

7: (covariates). These models are a special case of models known as

8: ``mixtures of experts'' in statistics and computer science literature.

9: Flexible specifications include models in which only mixing

10: probabilities, modeled by multinomial logit, depend on the covariates

11: and, in the univariate case, models in which only means of the mixed

12: normals depend flexibly on the covariates. Modeling the variance of the

13: mixed normals by flexible functions of the covariates can weaken

14: restrictions on the class of the approximable densities. Obtained

15: results can be generalized to mixtures of general location scale

16: densities. Rates of convergence and easy to interpret bounds are also

17: obtained for different model specifications. These approximation

18: results can be useful for proving consistency of Bayesian and maximum

19: likelihood density estimators based on these models. The results also

20: have interesting implications for applied researchers.

21: 

22: %Obtained characterizations of approximable classes of densities for

23: %different specifications of the model can be useful for choosing

24: %appropriate specifications in applications.

25: \end{abstract}