Recommender system for the labor market: identify sorting with collaborative filtering and deep neuron networks

More on Jianhong Xin

Abstract:

We propose a non-parametric method to uncover the sorting pattern between workers and firms in the labor market, using the matched employer-employee data on wages. Our approach can be applied to identify very wide range of models of sorting, either based on absolute advantage or comparative advantage. In particular, we collaborative filtering to detect the latent productivity types of workers and firms, make inference on the counterfactual wage and output for a potential (welfare enhancing) reallocation. With less restricted framework, our approach better estimate the potential output loss from search friction. Method-wise, we contribute a deep-learning-accelerated spectral bi-clustering that display scalability and performance for sparse and high dimensional data. Preliminary and incomplete, all comments are welcome.