كنفرانس بين المللی مهندسی صنايع و سيستم ها

صفحه اصلی / دوازدهمین كنفرانس بين المللی مهندسی صنايع و سيستم ها

Forecasting Rare Events and Systemic Crises in Financial Markets A Hybrid Econometric Filtering and Cost-Sensitive Machine Learning Framework

نویسندگان :

Madine Azad¹ Mohammadmehdi Kazemi² Marziye Khakestari³

1- دانشگاه خوارزمی 2- دانشگاه خوارزمی 3- دانشگاه خوارزمی

کلمات کلیدی :

systemic risk،cost-sensitive machine learning،ARIMA-XGBoost hybrid model،asset-liability mismatch،explainable AI،inancial markets

چکیده :

Forecasting rare events and systemic crises in financial markets requires tools capable of capturing linear trends and nonlinear volatility simultaneously. Financial markets - and the insurance sector in particular - remain persistently exposed to macroeconomic shocks that give rise to asset-liability mismatches. This study proposes a novel hybrid framework that couples an econometric filtering stage with cost-sensitive machine learning to forecast crisis regimes in an insurance-sector exchange-traded fund (ETF). In this architecture, an Autoregressive Integrated Moving Average (ARIMA) model first serves as a low-pass filter to isolate the linear component of market behavior; its residuals are extracted as structural shock signals. These residuals, together with macroeconomic variables - including the CBOE Volatility Index (VIX) and changes in the 10-year U.S. Treasury yield - are fed into a gradient-boosted decision tree classifier (XGBoost). To address the severe class imbalance inherent in rare-event prediction, cost-sensitive learning is used in place of synthetic oversampling, thereby preserving the genuine geometric structure of the data. Out-of-sample evaluation shows that the model achieves an overall accuracy of 93.85% and an area under the ROC curve (AUC) of 0.8423. Finally, SHapley Additive exPlanations (SHAP) analysis confirms that the VIX and interest-rate shocks are the dominant drivers of the regime shift toward crisis. This framework provides an efficient tool for risk managers to adaptively adjust hedging strategies based on empirical statistical evidence before market collapses materialize.