Asset Pricing with Realistic Crises Dynamics

What causes deep recessions and slow recovery? I revisit this question and develop a macro-finance asset pricing model that quantitatively matches the salient empirical features of financial crises such as a large drop in the output, a high risk premium, reduced financial intermediation, and a long duration of economic distress. The model features leveraged intermediaries who are subjected to both capital and productivity shocks, and face a regime-dependent exit rate. I show that the model without time varying intermediary productivity and exit, which reduces to Brunnermeier and Sannikov [2016], suffers from a tension between the amplification and the persistence of financial crises. In particular, there is a trade-off between the unconditional risk premium, the conditional risk premium, and the probability and duration of crisis.
Features that generate high financial amplification also induce faster recovery, at odds with the data. I show that my model resolves this tension and generates realistic crises dynamics. The model is solved using a novel numerical method with active machine learning that is scalable and overcomes the curse of dimensionality.