[Paper Review] Cluster formation and evolution in networks of financial market indices

Junior, Leonidas Sandoval|arXiv (Cornell University)|Nov 22, 2011

Complex Systems and Time Series Analysis27 citations

TL;DR

This paper proposes a threshold-based network analysis of global financial market indices using Spearman rank correlation to construct dynamic clusters and study their evolution during financial crises. By applying distance thresholds to correlation-based networks and analyzing the second eigenvector of correlation matrices, the study reveals a persistent two-cluster structure tied to time-zone differences—American and European indices form one group, while Pacific-Asian indices form a second, anti-correlated group—highlighting how market timing shapes systemic risk and network topology during volatility. - meta_description: Analyzes cluster formation in global financial indices using Spearman correlation and threshold-based networks, revealing time-zone-driven market structures and crisis-induced network shrinkage. - objective: - To model dynamic clustering in global financial market indices using correlation-based networks. - To investigate how network structure evolves during periods of financial stress such as Black Monday (1987), the Asian Crisis (1997), and the 2001 dot-com crash. - To identify structural patterns in the second eigenvector of correlation matrices linked to time-zone operations of exchanges. - To filter noise in correlation matrices by establishing statistically significant distance thresholds. - method: - Uses daily log-returns of 16 to 79 international stock indices from 1986 to 2001, with data adjustments for non-trading days and time-zone mismatches. - Computes Spearman rank correlation between indices and defines a distance measure as $ d_{ij} = 1 - c_{ij} $, ensuring Euclidean properties. - Constructs three-dimensional network visualizations using principal component analysis to minimize distortion in inter-node distances. - Applies threshold-based filtering to generate asset trees, revealing cluster formation at varying correlation levels. - Uses randomized data (1000 simulations per period) to calibrate noise thresholds above which spurious connections emerge. - Analyzes the eigenvector corresponding to the second largest eigenvalue of the correlation matrix to detect time-zone-related market structure. - research_questions: - How do clusters of financial market indices form and evolve across different time periods, especially during financial crises? - What is the role of time-zone differences in shaping the second eigenvector of correlation matrices among global indices? - At what threshold values does random noise begin to dominate the network structure of financial indices? - How does network topology change during periods of high market volatility compared to stable periods? - key_findings: - A persistent two-cluster structure emerges in the second eigenvector: one group comprises U.S. and European indices, the other Pacific-Asian indices, with the latter showing negative loadings, indicating anti-correlated movement due to time-zone differences. - The network structure shrinks in size during financial crises, reflecting increased correlation and reduced diversification. - Thresholds above which noise dominates were empirically calibrated using randomized data, enabling reliable identification of real network structures. - The Pacific-Asian cluster becomes more distinct in the 1990s, suggesting growing integration of emerging markets into global risk transmission. - The second eigenvector reveals a unique, time-zone-dependent market structure not observable in standard correlation matrices or single-market data. - The third eigenvalue occasionally shows structure but is heavily contaminated by noise, limiting its interpretability.

ABSTRACT

Using data from world stock exchange indices prior to and during periods of global financial crises, clusters and networks of indices are built for different thresholds and diverse periods of time, so that it is then possible to analyze how clusters are formed according to correlations among indices and how they evolve in time, particularly during times of financial crises. Further analysis is made on the eigenvectors corresponding to the second highest eigenvalues of the correlation matrices, revealing a structure peculiar to markets that operate in different time zones.

Motivation & Objective

• To model dynamic clustering in global financial market indices using correlation-based networks.
• To investigate how network structure evolves during periods of financial stress such as Black Monday (1987), the Asian Crisis (1997), and the 2001 dot-com crash.
• To identify structural patterns in the second eigenvector of correlation matrices linked to time-zone operations of exchanges.
• To filter noise in correlation matrices by establishing statistically significant distance thresholds.

Proposed method

• Uses daily log-returns of 16 to 79 international stock indices from 1986 to 2001, with data adjustments for non-trading days and time-zone mismatches.
• Computes Spearman rank correlation between indices and defines a distance measure as $ d_{ij} = 1 - c_{ij} $, ensuring Euclidean properties.
• Constructs three-dimensional network visualizations using principal component analysis to minimize distortion in inter-node distances.
• Applies threshold-based filtering to generate asset trees, revealing cluster formation at varying correlation levels.
• Uses randomized data (1000 simulations per period) to calibrate noise thresholds above which spurious connections emerge.
• Analyzes the eigenvector corresponding to the second largest eigenvalue of the correlation matrix to detect time-zone-related market structure.

Experimental results

Research questions

• RQ1 How do clusters of financial market indices form and evolve across different time periods, especially during financial crises?
• RQ2 What is the role of time-zone differences in shaping the second eigenvector of correlation matrices among global indices?
• RQ3 At what threshold values does random noise begin to dominate the network structure of financial indices?
• RQ4 How does network topology change during periods of high market volatility compared to stable periods?

Key findings

• A persistent two-cluster structure emerges in the second eigenvector: one group comprises U.S. and European indices, the other Pacific-Asian indices, with the latter showing negative loadings, indicating anti-correlated movement due to time-zone differences.
• The network structure shrinks in size during financial crises, reflecting increased correlation and reduced diversification.
• Thresholds above which noise dominates were empirically calibrated using randomized data, enabling reliable identification of real network structures.
• The Pacific-Asian cluster becomes more distinct in the 1990s, suggesting growing integration of emerging markets into global risk transmission.
• The second eigenvector reveals a unique, time-zone-dependent market structure not observable in standard correlation matrices or single-market data.
• The third eigenvalue occasionally shows structure but is heavily contaminated by noise, limiting its interpretability.