HIERARCHICAL CLUSTERING IN ECONOMIC CRISIS EARLY WARNING SYSTEMS

Abstract

Introduction. Economic crises of recent decades have exposed the vulnerability of global stability and demonstrated the lagging nature of traditional macroeconomic indicators. Unlike GDP or unemployment rates, which capture changes only during the active phase of a recession, stock market indicators reflect investor expectations in real time. This provides a rationale for using the indices of bond trading volumes, IPO market activity, and capitalization dynamics as leading signals to identify systemic imbalances before they transform into a deep recession.

Purpose. The aim of the study is to assess the predictive value of stock market indicators for the early detection of economic crisis signs based on a comparative analysis of the global shocks of 2008, 2020, and 2022. The work is aimed at testing the hypothesis regarding the leading nature of the financial market relative to key macroeconomic parameters and developing an integral early warning index on this basis.

Materials and methods. The statistical database of the study comprises a dataset for the period 2007–2024, covering the world’s leading economies (the US, the UK, Germany, China, Japan, Switzerland, Hong Kong, Poland) and global aggregate indicators. The methodology is based on utilizing Python tools for data processing and applying multivariate statistical methods, specifically hierarchical clustering using Ward’s method and z-score standardization.

Results. Based on the results of the cluster analysis, the categorization of bond trading volumes, IPO activity, and market capitalization into a stable block of signaling variables was mathematically justified. On their basis, an integral early warning index was developed, utilizing the maximum deviation function and thresholds to identify risks. The testing of the model confirmed its ability to capture systemic failures, as in 2008, 2020, and 2022, the index values exceeded critical limits, demonstrating predictive signals a year before the onset of the crisis.

Discussion. Further research may be directed toward expanding the list of indicators and implementing machine learning methods, such as K-Means and anomaly detection algorithms, for the automatic segmentation of time series into regimes and the creation of an adaptive monitoring system. This will allow for the transformation of the model into a dynamic tool capable of accounting for the individual characteristics of national economies and the specifics of concrete systemic shocks.