Does Learning Lead to Endogenous Business Cycles? - CDS Colloquium

Speaker: Nathan Palmer, Federal Reserve Board (speaking via Zoom)

Title: Does Learning Lead to Endogenous Business Cycles? (Work in progress)

Abstract: We use a reinforcement learning (RL) algorithm to solve the household problem in a standard

Ramsey growth model. We show that the exploration-exploitation decision in RL changes the convergence
path traveled by any macroeconomy in which agents using RL to obtain a near-optimal policy, and that
it disciplines which sets of non-optimal policies RL agents use in equilibrium. We expect that the RL
economy will experience endogenous boom-bust cycles due to the interaction of the exploration-exploitation
dynamics and the saddle-point stability of the Ramsey model. The exploration-exploitation parameter
offers a way to quantify costs of optimization. If time permits we will also present estimations of this from the data.

Joint work with Michael Funke and Nicole Glanemann.

For Zoom link contact Robert Axtell