Molin Zhong, Federal Reserve Board
Project Description/Abstract
We propose a new dynamic factor model that allows nonlinear dynamics in the state and measurement equations. The proposed nonlinear factor model 1) can generate asymmetric, state-dependent, and size-dependent responses of observables to shocks; 2) can produce time-varying volatility, skewness, and tail risks in the predictive distributions; and 3) fits the data better than a linear factor model. Using macroeconomic and financial variables, we show how to take the model to the data. We find overwhelming evidence in favor of the nonlinear factor model over its linear counterpart in applications that include interest rates with zero lower bounds, credit default swap spreads for European countries, and nonfinancial corporate credit default swap spreads in the U.S.
Co-authors
Pablo Guerron-Quintana, Boston College
Alexey Khazanov, Boston College