Bayesian analysis for varying coefficient autoregressive models

The varying coefficient autoregressive model is widely applied in various areas. In the paper we propose three Bayesian approaches to make inference for this model. The first one uses traditional Gaussian process priors for the varying coefficients in the model, and the other two adopt the reconstruction parameterization technique to reduce the computation. The corresponding MCMC algorithms are provided to compute the posterior distributions. Numerical comparisons indicate that, the third approach can yield satisfactory inferential results with the lowest computational cost, and thus is very suitable for massive data. We also apply it to analyse a dataset of Beijing taxi ridership.