Title: A general class of score-driven smoothers
Authors: Giuseppe Buccheri - Scuola Normale Superiore (Italy) [presenting]
Giacomo Bormetti - University of Bologna (Italy)
Fulvio Corsi - University of Pisa and City University London (Italy)
Fabrizio Lillo - Scuola Normale Superiore (Italy)
Abstract: It is first shown that, in the steady state, Kalman filter and smoother recursions can be re-parameterized in terms of the score of the conditional density and the Fisher matrix. Since in the new representation the predictive filter has the form of score-driven models, we introduce, by analogy, a score-driven update filter (SDU) and smoother (SDS). In this new framework, we recover smoothed estimates of observation-driven models, as well as assess filtering uncertainty and construct confidence bands. We test both empirically and through simulations the advantages of SDU and SDS over standard score-driven filters and exact simulation-based methods.