TingGuoZheng: Fast Estimation of a Large TVP-VAR Model with Score-Driven Volatilities
Reporter：TingGuoZheng
Abstract：This paper introduces a large time-varying parameter structural vector autoregressive (TVP-SVAR) model and then proposes a fast approach to estimate it. Based on the score-driven modeling framework, we firstly assume that the time-varying variances of structural errors in each equation of the TVP-SVAR are score-driven, and then propose the filtering and smoothing procedures for estimating time-varying parameters and time-varying volatilities. We show that under the Minnesota prior and forgetting factors, the filtering estimation of time-varying parameters is equivalent to an equation-by-equation estimator, which can greatly reduce the dimension of state space and thus is a very fast estimation. Moreover, we find that under forgetting factors, the smoothing estimation is also straightforward and extremely fast, which overcomes the inverse of supra-high dimensional state equation covariance matrix. Our simulation study shows that the proposed method in filtering the data is more accuracy than the existing popular method and illustrates the computational gain from the equation-by-equation estimator. Finally, we carry out an empirical study on dynamic connectedness of global stock markets, which demonstrates the advantages of our method in real-time and ex-post analysis.
