Encoding Behavior Commonalities In Global Stock Market Indexes: Unsupervised Machine Learning Approach

In recent times there is a consensus that the stock market is a dynamic and complex system, with some factors difficult to assess and are highly unpredictable that can cause disruptions. While the influence of crises and uncertainties on individual stock markets has been well-studied, a systematic understanding of their impact on global market relationships remains limited. This paper explores Machine Learning techniques over traditional econometric techniques to analyze stock market behavior of select countries over a period of time of twenty-one years. Specifically, we utilize a novel end-to-end hierarchical clustering method and proximity analysis to uncover changes in global stock market behavior across various crisis periods (2001, 2002, 2007-2009, 2016, and 2020). Daily time series data for global stock indices from 2002, to 2023, is analyzed. The proposed clustering method effectively identifies groups of countries with distinct risk profiles. These clusters, combined with an inference strategy, have the potential to inform investment decisions by aiding in the selection of outperforming or underperforming stocks. The results led to four clusters out of 26 countries depicting some countries consistently showing similarity in the behavior of stock market dynamics. Countries like India, Japan, Australia, New Zealand, Sweden, Israel, the US, and South Korea have been considered as balancing out volatility and hedging risks. The study paves the way for further exploration by incorporating macroeconomic variables to investigate their influence on the stock indices within each identified cluster. Additionally, the analysis of common characteristics within each cluster can be further explored.