This paper compares the properties of two particle filters – the Bootstrap Filter and the Auxiliary Particle Filter – applied to the computation of the likelihood of artificial data simulated from a basic DSGE model with nominal and real rigidities. Particle filters are compared in terms of speed, quality of the approximation of the probability density function of data and tracking of state variables. Results show that there is a case for the use of the Auxiliary Particle Filter only when the researcher uses a large number of observable variables and the number of particles used to characterize the likelihood is relatively low. Simulations also show that the largest gains in tracking state variables in the model are found when the number of particles is between 20,000 and 30,000, suggesting a boundary for this number.
