Abstract: This paper introduces a new method for analyzing time-series of stock market indexes (Dow Jones, NASDAQ etc) in order to reveal the quantitative level of their interference. In literature, the interference between stock indexes is usually called ‘contagion’. Strictly, contagion – is transmission of shocks from one region to another. In order to prevent destructive effects of crisis it’s very important for the government to predict them. The proposed method consists of two steps: transformation of time series to the space of independent components and clustering in this space using the principle of stability. The contents of clusters just define companies, countries, and regions related by the effect of contagion. To complete the mentioned steps we use: the IcaLab? tool from MatLab? package to construct independent components, MajorClust? method and DEM index for clustering and testing its quality. MajorClust? and DEM were developed in Benno Stein research group. The results of experiments showed the essential advantage of the proposed method over the traditional approach based on formal relations in correlation matrix. Notably, we marked out the groups of market indexes, which were most likely connected via ‘contagion’. Such a conclusion was made using external data like geographical position, level of country development, macroeconomic rates, etc.

Keywords: independent component analysis, clustering with MajorClust?, contagion effect, financial market

ACM Classification Keywords: I.2M Miscellaneous

Link:

CLUSTERING IN THE SPACE OF INDEPENDENT COMPONENTS AS A TOOL FOR REVEALING CONTAGION EFFECT ON FINANCE MARKETS

Konstantin Dylko, Mikhail Alexandrov

http://foibg.com/ibs_isc/ibs-24/ibs-24-p10.pdf