Joint DSC/CIS seminar series - talk by Dr. Jing Wu of URI
CCB-243
Donghui Yan, Yuchou Chang, Josh Carberry
dyan@umassd.edu
We develop a variational Bayes approach to approximate the posterior distributions under the exponential tilted empirical likelihood framework for stationary time series data. Moment conditions are obtained using Whittle’s estimation method that transforms the original dependent observations to approximately independent periodogram ordinates. Two reparameterization techniques are adopted to ensure the stationarity and invertibility of the parameters under the Markov Chain Monte Calo sampling and to facilitate a more accurate posterior approximation under the variational Bayes algorithm. A calibration technique is further proposed to guarantee the posterior coverage probability attains the nominal level. Extensive simulation studies and a real data analysis are conducted to
investigate the empirical performance of the proposed method.