A Deep Learning Approach to Competing Risks Representation in Peer-to-Peer Lending.

Area of Science:

• Fintech and Computational Finance
• Machine Learning in Finance
• Risk Management

Background:

• Online peer-to-peer (P2P) lending offers lower transaction costs and disintermediates traditional finance.
• Maximizing investor return on investment is a primary objective in P2P lending.
• Existing models often fail to address the competing nature of charge-off and prepayment risks simultaneously.

Purpose of the Study:

• To develop a unified framework for modeling competing charge-off and prepayment risks in P2P loans.
• To leverage deep learning for simultaneous risk assessment and feature representation.
• To enhance investor decision-making and investment performance in P2P lending.

Main Methods:

• Development of a hierarchical grading framework to integrate qualitative and quantitative risk factors.
• Implementation of an end-to-end deep learning approach using deep neural networks.
• Decomposition of the problem into multiple binary classification subproblems for joint risk learning.

Main Results:

• The proposed deep learning methodology effectively models competing risks in P2P loans.
• Achieved appealing investment performance by explicitly addressing risk interactions.
• Saliency map analysis provided intuitive insights into loan payment dynamics.

Conclusions:

• This research presents the first deep learning approach to characterize competing risks in P2P lending.
• The methodology offers a significant advancement in understanding and managing P2P lending risks.
• The findings empower investors with data-driven insights for improved investment strategies.