Dynamic Variable Selection with Spike-and-Slab Process Priors

We address the problem of dynamic variable selection in time series regression with unknown residual variances, where set active predictors is allowed to evolve over time. To capture time-varying uncertainty, we introduce new shrinkage priors for coefficients. These are characterized by two main ingredients: smooth parameter evolutions and intermittent zeroes modeling predictive breaks. More formally, our proposed Dynamic Spike-and-Slab (DSS) constructed as mixtures processes: a spike process irrelevant coefficients slab autoregressive The mixing weights themselves depend on lagged values series. Our DSS probabilistically coherent sense that their stationary distribution fully known spike-and-slab marginals. For posterior sampling coefficients, model indicators well propose SSVS algorithm based forward-filtering backward-sampling. scale method large data sets, develop EMVS MAP smoothing. demonstrate, through simulation topical macroeconomic dataset, very effective at separating noisy fast implementation significantly extends reach methods big data.