Matthias Rottner, Deutsche Bundesbank

“Estimating Nonlinear Heterogeneous Agents Models with Neural Network”


We leverage recent advancements in machine learning to develop a method to solve and to perform likelihood estimation of the parameters of nonlinear, heterogeneous agents models. Neural networks are set up to obtain an accurate approximation of the model's nonlinear transition equations and likelihood function. The proposed method can retrieve the true parameter values of a prototypical model whose solution is worked out analytically. Furthermore, the parameters estimated with our method are consistent with those estimated using a state-of-the-art method, which can, however, only be applied to stylized nonlinear representative agent models. Using simulated data we show that the proposed method provides an accurate estimation of the parameters of a nonlinear Heterogeneous Agent New Keynesian (HANK) model with a zero lower bound (ZLB) constraint and aggregate uncertainty. When we estimate the HANK model using U.S. data, we find that the interplay between aggregate nonlinearities and uninsurable idiosyncratic risk plays a large role in explaining business cycles volatility. The estimation with real data provides a first evaluation of how nonlinear HANK models should be expanded to improve their empirical performance.

For more information about Matthias Rottner and his interesting work - link to his website.

Contact person: Søren Hove Ravn