Template-Type: ReDIF-Paper 1.0
Series: Tinbergen Institute Discussion Papers
Creation-Date: 2023-24-03
Revisions-Date: 2023-01-12
Number: 23-016/III
Author-Name: Francisco Blasques
Author-Workplace-Name: Vrije Universiteit Amsterdam
Author-Name: Siem Jan Koopman
Author-Workplace-Name: Vrije Universiteit Amsterdam
Author-Name: Karim Moussa
Author-Workplace-Name: Vrije Universiteit Amsterdam
Title: Extremum Monte Carlo Filters: Real-Time Signal Extraction via Simulation and Regression
Abstract: This paper introduces a novel simulation-based filtering method for general state space models. It can be used to compute time-varying conditional means, modes, and quantiles, and for predicting latent variables. The method consists of generating artificial data sets from the model and estimating quantities of interest via extremum estimation. We call this procedure extremum Monte Carlo. The approach is conceptually simple and easy to implement. It can be applied to any model from which samples of data can be simulated. Given that most of the computations can be performed in advance, the method is particularly suited for real-time applications. The filter is stable over time under mild assumptions, which remains valid under model misspecification. Conditions are provided for convergence to an optimal filter as the number of draws diverges. The linear version of the filter converges to the Kalman filter. Various other features of the filter are illustrated via examples related to nonlinearity, missing data, and intractable densities. An empirical application to exchange rates demonstrates that, despite a setting of limited tractability, the method is able to efficiently extract the time-varying volatility.
Classification-JEL: 
Keywords: Intractable densities, Least squares Monte Carlo, Nonlinear non-Gaussian state space models, Hidden Markov models, Real-time filtering
File-URL: https://papers.tinbergen.nl/23016.pdf
File-Format: application/pdf
File-Size: 1.032.501 bytes
Handle: RePEc:tin:wpaper:20230016