Heterogeneity and the MPC: Observations from Machine Learning
Authors: Satyajit Dutt and Jan Radermacher (Goethe University and SAFE)
Title: Heterogeneity and the MPC: Observations from Machine Learning
Abstract: We replicate Christelis et al. (2019) and extend the tool-set used in the paper to supervised and unsupervised machine learning (ML) techniques. We compare algorithms by predictive power and find that traditional econometric techniques and ML methods compliment each other well. Particularly, we are able to identify more heterogeneity in the data and come up with a richer model specification to better capture the data generating process. When households are asked about how much they will consume from negative transitory income shocks, we find that that household characteristics related to saving/investing perceptions matter while life cycle demographics are the key factors in positive transitory income shocks.