Clustering economic and financial time series: Exploring the existence of stable correlation conditions

Clustering plays an important role in extracting information from the noise in economic and financial time series; it is one way to perform a coarse-graining of the empirical data sets to extract stable dependence information from the surrounding noise. This discussion paper explores clustering as an econometric tool for studying dependencies between variables. It places clustering in the context of the classical statistical tools for dependence analysis (i.e., correlation, regression, cointegration and structural change); discusses similarity, the key concept in time-series clustering; and reviews the principal approaches to quantifying similarity with the distance function. It argues that concepts of similarity, such as dynamic time-warping, used in a wide range of application domains (e.g., speech recognition, medicine, biomathematics, seismology) might be useful in the context of finance and economics.