A Transformer-Based Event-Aware Model for Oil Price Uncertainty and Sovereign Credit Risk Forecasting

Evolutionary and Institutional Economics Review · Springer Nature · 2026-11-19 · DOI: 10.1007/s10614-026-11476-8

Аннотация

This study proposes an Event-Aware Transformer for forecasting WTI and Brent crude oil prices and analyzing their relationship with credit default swap (CDS) spreads. The model integrates a learnable event embedding module that encodes major economic events including geopolitical shocks, monetary policy decisions, and pandemic related disruptions, directly into the Transformer architecture alongside standard price and CDS spread features. Trained on daily financial data spanning 2015 to 2021, the proposed model achieves a mean absolute error (MAE) of 1.2432 for WTI and 1.3194 for Brent, outperforming all tested baselines including LSTM, GRU, LTSF-Linear, and SVM variants across all reported metrics. An ablation study confirms that the event embedding component alone accounts for a reduction of approximately 52% in mean squared error compared to a standard Transformer without event context. These results demonstrate that explicitly incorporating economic event information into attention-based architectures substantially improves forecasting accuracy for event-driven commodity markets. The findings also suggest that CDS spread dynamics are closely linked to oil price shocks during periods of heightened economic uncertainty, with implications for credit risk monitoring and financial stability analysis.

Читать у издателя →

Похожие статьи