Bereiche | Tage | Auswahl | Suche | Aktualisierungen | Downloads | Hilfe
DY: Fachverband Dynamik und Statistische Physik
DY 28: Extreme Events
DY 28.4: Vortrag
Donnerstag, 3. April 2014, 10:15–10:30, ZEU 146
Optimizing cluster analysis by stochastic methods — •Philip Rinn1, Yuriy Stepanov2, Thomas Guhr2, Joachim Peinke1, and Rudi Schäfer2 — 1ForWind -- Center for Wind Energy Research, Institute of Physics, University of Oldenburg, Germany — 2Faculty of Physics, University of Duisburg-Essen, Germany
A new method to analyze the the results of a clustering algorithm is presented. Using a similarity measure daily prices of S&P 500 stocks are clustered with a top-down clustering scheme to represent states in the financial market. Time series of the distance between each data point and the cluster centers are calculated which describe the evolution of the financial market seen from the respective cluster center.
With methods from stochastic data analysis we separate the stochastic part from the deterministic part of the given time series. From the deterministic part we calculate a potential and find the fixed points of the system. We link the stable fixed points of the deterministic potential to the centers of the aforementioned clusters. Fixed points of the system that do not match with cluster centers can be identified as artificial clusters and ideas for optimizing the clustering to match the systems fixed points can be derived.