A1891
Title: A sparse Kalman filter: A non-recursive approach
Authors: Jan Bruha - CNB (Czech Republic) [presenting]
Abstract: An algorithm is proposed to estimate unobserved states and shocks in a state space model under sparsity constraints. Many economic models have a linear state space from linearized DSGE models, VARs, time-varying VARs, or dynamic factor models. Under the conventional Kalman filter, which is essentially a recursive OLS algorithm, all estimated shocks are non-zero. But often the true shocks are zero for multiple periods, and the non-zero estimate is due to noisy data or the ill-conditioning of the model. Applications are shown where sparsity is the natural solution. The sparsity of filtered shocks is achieved by an elastic-net penalty to the least-squares problem and improves statistical efficiency. The algorithm can also be adapted for non-convex penalties or estimates robust to outliers.