“…In recent work, Galvão, Giraitis, Kapetanios and Petrova (2015a) have provided a new approach that allows time varying estimation of Bayesian models, used for the time varying estimation of the Smets and Wouters (2007) DSGE model in Galvão, Giraitis, Kapetanios and Petrova (2015b). Their approach is an extension and formalisation of rolling window estimation, generalised by combining kernel-generated local likelihoods with appropriately chosen priors to generate a sequence of posterior distributions for the objects of interest over time, following the methodology developed in Giraitis, Kapetanios and Yates (2014) and Giraitis, Kapetanios, Wetherilt and Zikes (2016). Both the kernel and the rolling window approaches, when applied to structural models, assume that, instead of being endowed with perfect knowledge about the economy's data generating process, agents take parameter variation as exogenous when forming their expectations about the future.…”