Feature ranking and network analysis of global financial indices

Area of Science:

• Quantitative Finance
• Machine Learning Applications
• Network Analysis in Economics

Background:

• Understanding the interconnectedness of global stock markets is crucial for financial stability.
• Identifying influential stock indices can help in predicting market behavior and contagion effects.
• Traditional methods may not fully capture the dynamic influence between diverse global stock markets.

Purpose of the Study:

• To apply machine learning feature ranking to analyze the influence and network properties of 21 world stock indices.
• To identify dominant stock indices and their influence patterns across different market conditions, especially during crises.
• To quantify market interconnectedness and information flow using network centrality and entropy measures.

Main Methods:

• Utilized machine learning (Random Forest, Gradient Boosting) for feature ranking to determine index influence probabilities.
• Constructed stock market networks based on feature ranking matrices and analyzed network properties like global reaching centrality.
• Calculated Shannon entropy from influence probabilities to assess the distribution of market influence.

Main Results:

• North American and US stock indices exhibit significant influence, particularly during the global financial crisis.
• European indices also show influence during the European sovereign debt crisis, alongside American indices.
• Asian indices (India, China) demonstrated notable influence during periods of market unrest; global reaching centrality increased during crises.

Conclusions:

• Machine learning feature ranking effectively identifies influential stock indices and their dynamic interdependencies.
• Shannon entropy drops during crises indicate the dominance of a few key indices, serving as a crisis indicator.
• The findings provide valuable insights for studying global stock market contagions and identifying key market drivers.