Detecting contemporaneous mean co-breaking via ART

In this talk we present a ART-based procedure to detect contemporaneous menan co-breaking. In particular, ART is employed for the preliminary identification of breaks in the mean of each one-dimensional component of a multidimensional time series. The corresponding break dates represents candidate co-breaking dates and thus they are employed to delimit an interval on which to Principal Component Analysis (PCA) is recursively performed to generate the linear combinations of the series. On each of these linear combinations ART is performed again to identify the best linear combination that "hides" the co-breaking date. Although the use of PCA is not new in the field of common feature analysis, our approach represents an enhancement as it enables the date of the contemporaneous mean co-breaking to bea priori unknown i.e. to be estimated along the procedure. According to the current literature on the topic we focus on a single level shift and we present the results of a simulation study carried out considering contemporaneous mean co-breaking in various models.