AI in Business and Society
Loading...
Paper Number
1518
Paper Type
Completed
Description
Financial distress prediction is a prominent research topic in information systems, with two primary modelling categories: stationary and dynamic modelling. Recent stationary modelling works have leveraged company interactions to improve prediction performance, considering the heterogeneity of interactions while ignoring the dynamicity. However, few dynamic modelling works utilized interactions. To address the inconsistency and limitation of stationary and dynamic modelling works in leveraging interactions, we propose the Spatio-Temporal Financial Graph Attention Network with Meta-learning (STFGAN-Meta). STFGAN-Meta leverages interactions' spatial heterogeneity via the Spatial Aggregation module and temporal dynamicity via the Temporal Aggregation module. STFGAN-Meta introduces the Meta-learning Optimization module to unify stationary and dynamic modelling. Our experimental evaluation demonstrates that leveraging dynamicity and heterogeneity of interactions outperforms leveraging dynamicity or heterogeneity alone. Meta-learning succeeds in providing a generalized approach between stationary and dynamic modelling. STFGAN-Meta can be a promising risk assessment and decision-making tool in the financial industry.
Recommended Citation
Qi, Qi and Xing, Frank, "Leveraging Interactions for Stationary and Dynamic Financial Distress Prediction: A Spatio-Temporal Financial Graph Attention Network" (2023). ICIS 2023 Proceedings. 5.
https://aisel.aisnet.org/icis2023/aiinbus/aiinbus/5
Leveraging Interactions for Stationary and Dynamic Financial Distress Prediction: A Spatio-Temporal Financial Graph Attention Network
Financial distress prediction is a prominent research topic in information systems, with two primary modelling categories: stationary and dynamic modelling. Recent stationary modelling works have leveraged company interactions to improve prediction performance, considering the heterogeneity of interactions while ignoring the dynamicity. However, few dynamic modelling works utilized interactions. To address the inconsistency and limitation of stationary and dynamic modelling works in leveraging interactions, we propose the Spatio-Temporal Financial Graph Attention Network with Meta-learning (STFGAN-Meta). STFGAN-Meta leverages interactions' spatial heterogeneity via the Spatial Aggregation module and temporal dynamicity via the Temporal Aggregation module. STFGAN-Meta introduces the Meta-learning Optimization module to unify stationary and dynamic modelling. Our experimental evaluation demonstrates that leveraging dynamicity and heterogeneity of interactions outperforms leveraging dynamicity or heterogeneity alone. Meta-learning succeeds in providing a generalized approach between stationary and dynamic modelling. STFGAN-Meta can be a promising risk assessment and decision-making tool in the financial industry.
When commenting on articles, please be friendly, welcoming, respectful and abide by the AIS eLibrary Discussion Thread Code of Conduct posted here.
Comments
10-AI