Machine Learning Projection Methods for Macro-Finance Models

This paper develops a global simulation-based solution method to solve large states space macro-finance models using machine learning. We use an artificial neural network (ANN) to approximate the expectations in the optimality conditions in the spirit of the parameterized expectations algorithm (PEA). Because our method can process the entire information set at once, it is easily scalable to handle models with large state spaces that are highly collinear. We demonstrate these computational gains in two applications. First, we extend the optimal government debt problem studied by Faraglia et. al. (forthcoming) to ten maturities and we find that, when borrowing and lending constraints are tight, the optimal policy prescribes an active role for the medium-term maturities. Second, we reassess the solution of Kehoe and Perri (2002) for the international business cycle puzzles documented in Backus et al. (BKK1992). We show that extending their two-country framework to three countries (US, Europe, China ) can change the risk-sharing properties of the economy significantly