A contribution to the analysis of historical economic fluctuations (1870-2010): filtering, spurious cycles and unobserved component modelling

Abstract

Time series filtering methods such as the Hodrick-Prescott (HP) filter, with a consensual choice of the smoothing parameter, eliminate the possibility of identifying long swing cycles (e.g., Kondratieff type) or, alternatively, may distort periodicities that are in fact present in the data, giving rise, for example, to spurious Kuznets-type cycles. In this paper, we propose filtering Maddison’s time series for the period 1870-2010 for a selection of developed countries using a less restrictive filtering technique that does not impose but rather estimates the cut-off frequency. In particular, we use unobserved component models that optimally estimate the smoothing parameter. Using this methodology, we identify cycles of periods mainly in the range of 4-7 years (Juglar type cycles), as well as a pattern of cyclical convergence that deepens with globalization processes. After 1950, a common business cycle factor grouping all economies is found.