|
Abstract:
|
We propose a new set of mixture models which can estimate causal relationships in a wide range of observable economic phenomena. Many economic problems deal with censored data whose distribution follows a power law distribution once it is above some hurdle. For example, wages follow a power law once they above zero. As of yet, there are limited modelling options to capture causal relationships in these distributions. We extend two hurdle models which are popular in econometrics (Cragg and Lognormal) to include instrumental variables. These models allow the researcher to take advantage of the properties of maximum likelihood estimation, rather than rely on GMM procedures as is currently common practice. The Lognormal IV also has a number of convenient linearity properties which we will discuss. We apply these models to a study of the impact of Chinese import competition on Portuguese firms from 1995-2007. Using detailed data on Portuguese firms and workers, we find that the trade shock from China caused a large increase in firm shutdown in Portugal.
|
