Publication type: Conference other
Type of review: Peer review (abstract)
Title: Forecasting correlation structures
Authors: Schüle, Martin
Ott, Thomas
Schwendner, Peter
Conference details: NDES 2017, 25th Nonlinear Dynamics of Electronic Systems Conference, Zernez, 5-7 June 2017
Issue Date: 6-Jun-2017
Language: English
Subject (DDC): 510: Mathematics
Abstract: Often the signature of a complex system is a couple of empirically found time series. In many cases the exact processes generating these series are unknown and a merely descriptive data analysis is undertaken. A popular tool to describe the structural behaviour of the complex system is thereby the analysis of the correlation structure, i.e., the system of pairwise correlations. As the correlation coefficients are calculated from a given data set, the analysis usually does not allow to forecast the future correlated behaviour of the system. However, in many examples of complex systems, the dynamic correlation structure shows certain patterns that form and cluster and dissipate again over time. If there are thus certain consistent patterns in the analyzed correlation structure to be found, the joint behaviour of the system may be forecasted, at least for short time periods. We present such an approach to forecast correlation matrices by means of a financial time series example. By analyzing the eigenmodes of the correlation matrices for oscillation patterns the main market dynamics are identified. Then, the principal eigenmode oscillations are forecasted by multivariate autoregressive and mean-reversion models allowing to infer the future correlation structure for certain time periods. The inferred correlation matrices are further regularized and compared to benchmark models. The proposed method can be of use in any field with the need of analyzing empirical correlation structures, e.g. in climate research.
Fulltext version: Published version
License (according to publishing contract): Licence according to publishing contract
Departement: Life Sciences and Facility Management
Organisational Unit: Institute of Computational Life Sciences (ICLS)
Appears in collections:Publikationen Life Sciences und Facility Management

Files in This Item:
There are no files associated with this item.

Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.