Econometric modeling of business Telecommunications demand using Retina and Finite Mixtues

In this paper we estimate the business telecommunications demands for local, intra-LATA and inter-LATA services, using US data from a Bill Harvesting R survey carried out during 1997. We model heterogeneity, which is present among firms due to a variety of different business telecommunication needs, by estimating normal heteroskedastic mixture regressions. The results show that a three-component mixture model fits the demand for local services well, while a two-component structure is used to model intra-LATA and inter-LATA demand. We characterize the groups in terms of their differences among the coefficients, and then use Retina to perform automatic model selection over an expanded candidate regressor set which includes heterogeneity parameters as well as transformations of the original variables. Our models improve substantially the in-sample fit as well the out-ofsample predictive ability over alternative candidate models. Retina suggests that the final demand specification should include telephone equipment variables as relevant regressors. On the other hand, the output of the firm, as well as its physical extension, have second order, yet significant effects on the demand for telecommunication services. Estimated elasticities are different for the three demands but always positive for access form (single-line or private network)