ECONOMETRICS: Dr Shin Kanaya (University of Essex)

Universal inference with time series data

We develop a likelihood-based inference method for time series data by extending the universal inference method of Wasserman, Ramdas, and Balakrishnan (2020, PNAS), originally proposed for independent data. We focus on parametric Markov and hidden Markov models, which encompass a wide range of models used in macroeconomics and finance, including autoregressive, (G)ARCH, and state-space models. Our method retains the original spirit of universal inference: the resulting confidence sets and hypothesis tests are valid without regularity conditions and enjoy finite-sample guarantees. Our approach relies crucially on a novel use of the Markov property. The proposed method is illustrated through Monte Carlo simulations and empirical applications.