Monitoring Residual Autocorrelations in Dynamic Linear Models

Abstract

In this article we analyze the problem of sequential monitoring residual autocorrelations in DLM's. To that end, we propose a specific algorithm to detect and correctly identify them, given that the monitoring schemes of level and variance changes proposed in West and Harrison (1997) are generally not capable of detecting this kind of deterioration. We also study the frequentist behavior of the three algorithms, providing guidelines on how to choose the values of their parameters.

Keywords:

• Dynamic linear models
• Residual autocorrelations
• Bayes factor
• Monitoring
• Monte Carlo
• AR

Acknowledgments

The work reported in this article has been supported by Grant 2FD97-2091 of the European Regional Development Fund (ERDF) under the title “Financial analysis of diversification and similarity of productive structures in the European Union,” administered by the University of Zaragoza, Spain. The authors would like to express their thanks to an anonymous referee for his helpful observations on an earlier version of this article. Similarly, they are grateful to Stephen Wilkins for his help in preparing the final version of the text.