Machine learning-based dynamic risk measurement for white sugar futures under geopolitical risks

Zihao Qiu¹Siyu Chen¹Zixin Feng²Ruitong Luo¹Zhiwei Wang^3*

• ¹Sichuan University, Chengdu, China
• ²Shanghai University of International Business and Economics, Shanghai, China
• ³China Banking Regulatory Commission, Beijing, China

This study integrates traditional VaR theory with machine learning methods to measure market risk in sugar futures under geopolitical uncertainty. It systematically examines the risk characteristics and transmission mechanisms of the sugar futures market against the backdrop of geopolitical conflicts. From a game theory perspective, market risk is not only the result of external shocks and random fluctuations but also the outcome of strategic interactions among various participants, including hedgers, speculators, arbitrageurs, and regulators. Utilizing sugar No. 5 futures trading data from the Zhengzhou Futures Exchange spanning 2015-2019 and 2024, empirical tests reveal that the annual delta values range between 0.26 and 1.16, averaging approximately 35 % below the theoretical values. A comparative analysis of three risk measurement methods demonstrates that the machine learning-based Value at Risk (VaR) at 95% confidence level exhibits a violation rate of 5.00%. By conducting return tests to calculate epsilon values (the relative deviation between actual and estimated tail risk occurrences), the study finds no statistically significant difference between 2024 (0.08) and the 2015-2019 average level (0.14), indicating that despite geopolitical conflicts, the fundamental risk transmission mechanisms of the sugar futures market remain relatively stable. The hybrid "machine learning-traditional theory" risk framework developed in this research provides a theoretical foundation and practical guidance for regulatory bodies to enhance risk prevention and control systems, as well as for market participants to optimize risk management strategies.