Improving SME Credit Risk Management with Advanced Predictive Analytics
Lead Research Organisation:
University of Southampton
Department Name: Southampton Business School
Abstract
One of the key issues in risk management for micro, small and medium firms (SME) is the scarcity of reliable data for credit risk evaluations1. A typical loan evaluation includes an in-depth financial profiling, followed by a free-text recommendation by a specialized analyst. A second analyst complements this report with all available socioeconomic and behavioural data, and then decides whether the entrepreneur is credit-worthy. This process is cumbersome, inefficient, and bias-prone, resulting in underfunding for small businesses2, and subsequent economic inefficiencies.
Credit risk modelling has always been one of the key strategic areas within DDAR. In previous research, we have explored the challenges and limitations of current statistical models for SME3,4 risk management, and we believe there is great potential in the new advances in data science for improved risk evaluations. These methodologies have shown great results when dealing with diverse data, such as a free-form text, detailed financial evaluations, or social networks in B2B markets, key components of SME credit data. We seek support for a PhD student that would explore this potential answering these questions, each one potentially conducting to a publication:
1. What is the most efficient design of a deep neural network that can process the diverse data created during SME evaluations?
2. Can this automated model reach better efficiencies in statistical accuracy, monetary cost, bias reduction, and evaluation time over simulated conditions, and in real data?
3. What would be the impact of this new method in regulatory and financial terms?
Credit risk modelling has always been one of the key strategic areas within DDAR. In previous research, we have explored the challenges and limitations of current statistical models for SME3,4 risk management, and we believe there is great potential in the new advances in data science for improved risk evaluations. These methodologies have shown great results when dealing with diverse data, such as a free-form text, detailed financial evaluations, or social networks in B2B markets, key components of SME credit data. We seek support for a PhD student that would explore this potential answering these questions, each one potentially conducting to a publication:
1. What is the most efficient design of a deep neural network that can process the diverse data created during SME evaluations?
2. Can this automated model reach better efficiencies in statistical accuracy, monetary cost, bias reduction, and evaluation time over simulated conditions, and in real data?
3. What would be the impact of this new method in regulatory and financial terms?
Organisations
People |
ORCID iD |
| Cristián Bravo (Primary Supervisor) | |
| Matthew Stevenson (Student) |
Studentship Projects
| Project Reference | Relationship | Related To | Start | End | Student Name |
|---|---|---|---|---|---|
| ES/P000673/1 | 01/10/2017 | 30/09/2027 | |||
| 1992301 | Studentship | ES/P000673/1 | 01/01/2018 | 30/11/2023 | Matthew Stevenson |
| Description | Stevenson, M. and Bravo, C., 2019. Advanced turbidity prediction for operational water supply planning. Decision Support Systems, 119, pp.72-84. - The aim of this project was to identify the potential variables causing daily turbidity peaking events at six groundwater sources for a water supply company, and, to utilise machine learning techniques to predict turbidity levels exceeding the regulatory limits up to seven days in advance. - We show that machine learning techniques are better than traditional statistical models. - We demonstrate an effective methodology using machine learning for predicting future turbidity peaking events which can be used by water supply organisations to reduce financial and water quality risk. Stevenson, M., Mues, C. and Bravo, C., 2021. The value of text for small business default prediction: A deep learning approach. European Journal of Operational Research, 295(2), pp.758-771. - Applies Deep Learning (Artificial Intelligence) to predict small business loan default using loan officer text statements with standard credit scoring features. - We find the text predictive, but less relevant when combined with structured data. - We explore how the text length and content influence the default predictions. This has implications for the case study organisation and the wider industry with regards to using text to improve data capturing exercises to improve credit risk assessment and reduce the cost to assess SME loans. |
| Exploitation Route | Stevenson, M., Mues, C. and Bravo, C., 2021. The value of text for small business default prediction: A deep learning approach. European Journal of Operational Research, 295(2), pp.758-771. - It provides a reference for credit lenders as to how text data might be used to provide more robust credit assessments. - We also show how AI techniques can be used to inform operational data capturing procedures that may also lead to better decision making for lenders. |
| Sectors | Environment,Financial Services, and Management Consultancy,Other |
| Description | Stevenson, M. and Bravo, C., 2019. Advanced turbidity prediction for operational water supply planning. Decision Support Systems, 119, pp.72-84. This was my first paper which was written during my award while awaiting data, though it was not directly related to my research subject. This was one of the first papers in the literature that uses machine learning to predict water quality events, an area that has since seen a number of related applications developed. Stevenson, M., Mues, C. and Bravo, C., 2021. The value of text for small business default prediction: A deep learning approach. European Journal of Operational Research, 295(2), pp.758-771. This paper used Deep Learning (artificial intelligence) approaches to predict loan default for small businesses. As part of this project, we were able to use textual loan request data to extract useful topics to improve operational procedures for collecting data for the case study organisation with wider insights for the SME credit industry. The intention is that this will lower the cost for lenders and subsequently improve credit accessibility. |
| First Year Of Impact | 2018 |
| Sector | Agriculture, Food and Drink,Environment,Other |
| Description | Conference: EURO2018 |
| Form Of Engagement Activity | A talk or presentation |
| Part Of Official Scheme? | No |
| Geographic Reach | International |
| Primary Audience | Other audiences |
| Results and Impact | Presentation of preliminary results at an international research conference to an audience of 30 members |
| Year(s) Of Engagement Activity | 2018 |
| URL | https://www.euro-online.org/conf/admin/tmp/program-euro29.pdf |
| Description | Conference: EURO2019 |
| Form Of Engagement Activity | A talk or presentation |
| Part Of Official Scheme? | No |
| Geographic Reach | International |
| Primary Audience | Other audiences |
| Results and Impact | Presentation at the EURO2019 conference to circa 40 persons, predominantly from Academia. |
| Year(s) Of Engagement Activity | 2019 |
| URL | https://www.euro-online.org/conf/euro30/treat_abstract?frompage=search&paperid=2255 |
| Description | Credit Research Conference 2019 |
| Form Of Engagement Activity | A talk or presentation |
| Part Of Official Scheme? | No |
| Geographic Reach | International |
| Primary Audience | Professional Practitioners |
| Results and Impact | Presentation at the CRC Conference 2019 presenting preliminary research findings. I had several academic and business professionals ask questions about my research following the presentation. I have been contacted twice since by email requesting further details |
| Year(s) Of Engagement Activity | 2019 |
| URL | https://crc.business-school.ed.ac.uk/wp-content/uploads/sites/55/2019/07/E30-Deep-Learning-for-Credi... |