Singular spectrum analysis for real-time financial cycles measurement

Abstract

This paper provides a new statistical methodology based on Singular Spectrum Analysis to extract the cycle component of an economic time series in real-time, addressing several criticisms towards classical ones. I measure the Credit-to-GDP cycles of 21 developed countries and run a horse race to compare it with the Hodrick-Prescott filter, the Hamilton regression filter, the 8-quarters growth rates, a Bartlett window, and variants. The SSA methodology performs best as an Early Warning Indicator for banking vulnerabilities and banking crises up to three years ahead. These conclusions about methodologies have practical implications for the measure of the systemic risk and the conduct of macroprudential policies.