EURO 2024 Copenhagen
Abstract Submission

EURO-Online login

3156. Explainable AI methods for early warning system of financial crisis prediction

Invited abstract in session TD-2: New Tools in Insurance Risk Management , stream OR in Banking, Finance and Insurance: New Tools for Risk Management.

Tuesday, 14:30-16:00
Room: Glassalen (building: 101)

Authors (first author is the speaker)

1. Jurgita Cerneviciene
Mathematical Modelling, Kaunas University of Technology
2. Audrius Kabasinskas
Department of Mathematical Modeling, Kaunas University of Technology

Abstract

Accurately predicting financial crises holds immense significance for the national economy and the financial sector. The market's instability has significant implications for decision-making processes, emphasizing the need to accurately predict and anticipate economic fluctuations at the earliest possible stage. Early warning methods can be a valuable tool for financial risk managers in reducing the likelihood of extreme losses. With the advent of modern technology and the availability of real-time data, it has become feasible to create forecasting models that can provide almost instantaneous predictions of financial crises. In this study, we present a technique that combines Artificial Neural Network, Random Forest, and XGBoost to analyze and predict the possibility of a financial crisis based on various performance indicators of pension funds, such as return on investment, risk level, and other related ratios. By employing the rolling window technique and methods of Explainable Artificial Intelligence (XAI), such as SHAP (Shapley Additive explanations) and LIME (Local Interpretable Model-agnostic Explanations), a comprehensive analysis was conducted. According to our results hidden market regime (forthcoming financial crisis) within one to ten days can be predicted with a high degree of accuracy. The risk-adjusted performance metrics, such as the Bernardo and Ledoit ratio, Sterling Ratio, Sortino ratio, and Pain index, turnout to be the most important measures.

Keywords

Status: accepted

Back to the list of papers