Jiti Gao, Monasch University

"Unified Estimation in Time-Varying Models"

Abstract

This paper proposes time varying models in which unknown parameters may vary stochastically or deterministically over time or as a mixture of both. These novel features are distinct from existing formulations which typically separate stochastic and deterministic time variation. Estimation methods for the former often rely on Bayesian resampling algorithms whereas nonparametric estimation methods are usually employed for fitting unknown deterministic functional forms. The present work develops instead a unified approach using orthonormal series specially designed to estimate time variation irrespective of whether it is stochastic or deterministic. The proposed approach has wide applicability for both linear and nonlinear time series models and relevance in modeling data where general trend specifications are desirable. Limit theory for valid inference concerning the time varying structures is established. A notable outcome is that stochastic trend parameter evolution can be estimated with asymptotic validity and fast rates of convergence. Other advantages include flexibility and convenience in practical implementation. Simulations are reported on finite sample performance and the procedures are illustrated in several real data examples.

Contact person: Anders Rahbek