QCS – Quantum Credit Scoring

The Quantum Credit Scoring project was launched to investigate whether quantum computing can improve certain stages of the credit assessment process. The aim is to identify which stages of credit scoring could benefit from a quantum approach and to test an initial algorithm that demonstrates the feasibility of this innovative approach.

The initiative aims to assess whether quantum computing or ‘quantum-inspired’ methods can make the process of establishing rating scales more efficient – a crucial step in determining the level of risk associated with an applicant. The long-term objective is to promote more accurate and inclusive processes, thereby contributing to better financial assessment.

Credit scoring involves analysing many variables and ordering them in a consistent manner. Certain steps, such as creating an optimal rating scale, can be formulated as combinatorial optimisation problems, which are difficult to solve using traditional methods. The project investigates whether quantum approaches might offer advantages in scenarios of this kind.

The project identified the definition of the rating scale as one of the most suitable steps for a quantum approach. A formulation based on QUBO (Quadratic Unconstrained Binary Optimisation) was therefore proposed, and an initial quantum-inspired algorithm was developed, which can be run on both classical hardware and, in the future, on quantum computers. The implementation was developed in Python and tested using advanced optimisers, enabling an assessment of the method’s limitations and potential. Experiments were also conducted on HPC infrastructure to compare the classical and quantum approaches.

Looking ahead, more efficient credit scoring could facilitate access to credit, promoting greater financial inclusion and a more accurate assessment of risks. Approaches of this kind can help to prevent imbalances in the financial system and strengthen the robustness of decision-making processes.

The next steps in the initiative involve testing on various quantum architectures and further developing the QUBO formulation, with a view to building a more robust model. The methodology can be replicated in other sectors facing problems similar to those involved in financial scoring, and the project’s code is already publicly available for anyone wishing to experiment with it.