Micro, small and medium-sized enterprises (MSMEs) constitute over 98% of the business establishments and employ about 46% of the overall workforce in Hong Kong. However, MSMEs generally encounter difficulties in borrowing money from banks due to the lack of credit information infrastructure and huge burden in managing credit assessment.
ASTRI is commissioned by the HKMA to explore the use of alternative data to perform credit scoring to support MSME loan applications. A panel of industry experts was invited to participate in the study to contribute their insights and discuss the benefits and challenges of the proposed framework of alternative credit scoring. Some of the key findings of this project are published in the HKMA white paper.
There are two software portals that support alternative credit scoring. MSMEs can submit loan applications online through the Online Loan Application Portal, and banks can process loan applications by the alternative credit scoring platform.
Main features of the alternative credit scoring platform
An alternative credit scoring platform that produces credit scores for MSME based on both alternative and conventional data is developed with the following features:
- Due to data privacy regulations, alternative data from different data sources cannot be gathered into a single data for machine learning. Federated learning is a kind of privacy preserving technology that aims to enable machine learning using distributed data sets at different data sources without infringing data privacy
- ASTRI is collaborating with Standard Chartered Bank (Hong Kong) Limited, PAOB, OpenRice Limited, and FreightAmigo for exploring the use of federated learning in alternative credit scoring
- To facilitate sharing of insights from alternative data between banks and data partners while complying with the requirements of the privacy regulations in Hong Kong
- To develop alternative credit scoring models using federated learning
- To verify the performance of the alternative credit scoring models and monitor the privacy control during the process of federated learning
To address the above, ASTRI developed a federated learning platform to support the deployment of alternative credit scoring. The platform can support sharing of machine learning model information among data partners and address the concerns of data privacy.
- Innovative model frameworks for credit scoring based on federated learning
- Algorithmic support for privacy enhancement
- Performance evaluation and privacy control for the model frameworks