Non ergodicity as partial identification
The ergodic axiom lies at the core of economics. The fact that ergodicity is seldom testable in the data favours assuming it from the onset. However, ergodic models with multiple stable equilibria approach non-ergodic behaviour for increasing level of persistence, when the probability of transition from one equilibrium to another goes to zero. In this paper we present a general approach to estimation of computational models with multiple equilibria. We show that non-ergodicity can be usefully thought of as an additional source of uncertainty which can be quantified in empirical applications. We test our method by means of a Montecarlo experiment on an agent-based model of the labor market.
(Matteo Richiardi & Jakob Grazzini)