Title: Finite Markov mixture modeling to cluster turning points
Authors: Maximo Camacho - Universidad de Murcia (Spain)
Lola Gadea - University of Zaragoza (Spain)
Ana Gomez-Loscos - Bank of Spain (Spain) [presenting]
Abstract: A new date-then-average approach is proposed to date business cycle turning points from a large set of economic indicators. Each individual turning point date is viewed a realization of a mixture of an unspecified number of separate Gaussian distributions, which determine the number and the dates of the aggregated turning points of the reference cycle. Finite mixture models techniques are used to determine the number of turning points of the reference cycle, to estimate the parameters for the separate distributions, to determine the mixing proportions and to perform statistical inference for the estimated reference cycle. By means of a Monte Carlo experiment, we show the high accuracy of the method to determine turning points. Using a set of US economic indicators, we show that the method is able to identify the NBER turning points with a maximum deviation of one quarter.