Missing observations in observation-driven time series models

Abstract We argue that existing methods for the treatment of missing observations in time-varying parameter observation-driven models lead to inconsistent inference. provide a formal proof this inconsistency Gaussian model with mean. A Monte Carlo simulation study supports theoretical result and illustrates how problem extends score-driven and, more generally, models, which include well-known conditional volatility. To overcome inference, we propose novel estimation procedure based on indirect This easy-to-implement method delivers consistent The asymptotic properties new are formally derived. Our proposed shows promising performance exercise as well an empirical concerning measurement volatility from financial returns data.