A dynamic factor approach to nonlinear stability analysis

A method of principal components is employed to investigate nonlinear dynamic factor structure using a large panel data. Imposing a common factor structure has practical merit in reducing dimension for the nonparametric stability analysis of a large system. Under some conditions, replacing unobservable common factors by principal components in the nonparametric estimation is theoretically justified. The validity of this approach is also supported by simulation analysis even if the true lag order of the autoregressive process of a common component is unknown. When the method is applied to the U.S. business cycles, the class of nonlinearity that can generate endogenous fluctuation or chaos is not supported by the data.