<abstract xmlns="http://www.w3.org/1999/xhtml">

<p>The abrupt and destructive nature of systemic financial risks underscores the imperative for modern societies to prioritize early detection and prevention over post-crisis management. This paper introduces a dual-perspective early warning indicator system for regional systemic financial risks, encompassing both macro and market dimensions. It further enhances this system by incorporating the ReliefF algorithm into the feature selection phase of a modified random forest model aimed at predicting regional systemic financial risks. The model’s efficacy was assessed using data spanning from 2009 to 2023, enabling dynamic early warning evaluations of regional systemic financial risks. The model was also employed to project the risk landscape for 2024. The results demonstrate superior performance metrics for the random forest model, with an accuracy of 0.9909, precision of 0.9847, recall of 0.9871, and an F1 score of 0.9785—outperforming three comparative models. Notably, during years of significant systemic risk (2008, 2010, and 2013), the model’s predictions exceeded 0.8, while in 2015 and 2020, they surpassed 0.9. For 2024, the model predicts a higher likelihood of maintaining a “normal” state of systemic financial risk in China, with probabilities ranging between 0.3 and 0.5. This study thus offers robust theoretical support for forecasting regional systemic financial risks.</p>
</abstract>

The abrupt and destructive nature of systemic financial risks underscores the imperative for modern societies to prioritize early detection and prevention over post-crisis management. This paper introduces a dual-perspective early warning indicator system for regional systemic financial risks, encompassing both macro and market dimensions. It further enhances this system by incorporating the ReliefF algorithm into the feature selection phase of a modified random forest model aimed at predicting regional systemic financial risks. The model’s efficacy was assessed using data spanning from 2009 to 2023, enabling dynamic early warning evaluations of regional systemic financial risks. The model was also employed to project the risk landscape for 2024. The results demonstrate superior performance metrics for the random forest model, with an accuracy of 0.9909, precision of 0.9847, recall of 0.9871, and an F1 score of 0.9785—outperforming three comparative models. Notably, during years of significant systemic risk (2008, 2010, and 2013), the model’s predictions exceeded 0.8, while in 2015 and 2020, they surpassed 0.9. For 2024, the model predicts a higher likelihood of maintaining a “normal” state of systemic financial risk in China, with probabilities ranging between 0.3 and 0.5. This study thus offers robust theoretical support for forecasting regional systemic financial risks.

Digital finance and regional systemic financial risk forecasting

Applied Mathematics and Nonlinear Sciences

This work is licensed under the Creative Commons Attribution-NonCommercial-NoDerivatives 3.0 License.

{"article-title":"Digital finance and regional systemic financial risk forecasting"}

The abrupt and destructive nature of systemic financial risks underscores the imperative for modern societies to prioritize early detection and prevention over...

Digital finance and regional systemic financial risk forecasting

Published Online: Jun 07, 2024

Page range: -

Received: Feb 21, 2024

Accepted: Apr 25, 2024

DOI: https://doi.org/10.2478/amns-2024-1380

Keywords
Machine learning, Random forest, K-neighborhood, Deep neural network, Systemic risk

© 2024 Enlin Tang et al., published by Sciendo

This work is licensed under the Creative Commons Attribution-NonCommercial-NoDerivatives 3.0 License.

Digital finance and regional systemic financial risk forecasting

Published Online: Jun 07, 2024

Page range: -

Received: Feb 21, 2024

Accepted: Apr 25, 2024

DOI: https://doi.org/10.2478/amns-2024-1380

KeywordsMachine learning, Random forest, K-neighborhood, Deep neural network, Systemic risk

© 2024 Enlin Tang et al., published by Sciendo

This work is licensed under the Creative Commons Attribution-NonCommercial-NoDerivatives 3.0 License.

Keywords
Machine learning, Random forest, K-neighborhood, Deep neural network, Systemic risk