PhD Studentship (ESRC South Coast DTP) - Image-Enriched Credit Risk Modelling for Micro and SME Lending (1065818TR)

Image-Enriched Credit Risk Modelling for Micro and SME Lending

A fully funded studentship awarded by the Economic and Social Research Council (ESRC) South Coast Doctoral Training Partnership (SCDTP) commencing in 2018/19 Academic Year if PhD only (+3) or 2019/20 Academic Year if Masters + PhD (1+3 funding).

Supervisory Team:

Dr. Cristián Bravo (Southampton Business School, Lead Supervisor) (C.Bravo@soton.ac.uk), Prof. Christophe Mues (Southampton Business School)

The Department of Decision Analytics and Risk, within the Southampton Business School, offers a fully funded studentship awarded by the Economic and Social Research Council (ESRC) South Coast Doctoral Training Partnership (SCDTP). The offer can be for either a PhD (+3) or MSc + PhD (1+3) position.

One of the key issues in risk management for micro, small and medium businesses (mSME) is the scarcity of reliable data for credit risk evaluations. The current evaluation processes are cumbersome, inefficient, and bias-prone, resulting in widespread underfunding for small businesses, and subsequent reduced productivity. Following the strategic research priority of the ESRC to improve productivity, this project is focused on creating a fair, efficient, and transparent methodology to improve lending to SME standards by increasing available information for evaluations. One of our industry partners has made available to us images of the places of work of a large number of mSME, along with commonly-used structured information on the businesses, and repayment information covering several years of lending practice. This unique dataset is currently being used by human specialists to override automatic decisions, which hints at a great potential to improve the decision-making using the latest image-processing developments in AI.

The general aim of the PhD project is to create novel AI methodologies to generate image-enriched credit scores modelling the probability of default for micro and small companies, using a mixture of images and structured data. These models should satisfy the legal requirements of banking regulators, in particular those regarding the unfair discrimination regulations, and those regarding explainability of the models.

While this PhD project is an exciting opportunity to research complex models in credit risk, extreme care must be taken to not include any information that would unfairly discriminate against some sectors of the population, by including e.g. gender or race information. Transparency standards in the lending industry require clear explanations of the reasons why an application is rejected, resulting thus far in the use of simpler models. This project will develop novel techniques to control the unfair biases that may occur, and to distil knowledge from the images to identify what exactly is being considered to discriminate between good or bad borrowers. 

During the PhD the candidate will conduct research to achieve the following:

Develop new methodologies to remove potential implicit or explicit biases present in the unstructured (image) data. This includes gender, race, religion, and any other identity-related information. Construct and apply deep learning architectures that can process this unbiased data and can generate a prediction regarding repayment probability of a loan. We will evaluate the impact of using non-structured data in credit risk models both in statistical terms and financial terms.Generate and apply context-dependent knowledge distillation approaches to evaluate what parts of the image or image information is being used when estimating this probability.

Skills Required

We are looking for an excellent candidate with a background in either Data Science, Financial Inclusion, Business Analytics, or Econometrics, and offer the possibility to conduct state of the art research in the area of Credit Risk for Micro and SME, working with companies and institutions in several countries, with the ultimate goal of providing better access to funding for new and existing small businesses.  

Funding

South Coast DTP Funding provides an annual maintenance grant (tax free) of £14777 (2018/19 RCUK rate), plus payment of all programme fees.  Other funding available for SCDTP funded students can be found on the SCDTP website (www.southcoastdtp.ac.uk).

Funding is provided for 3 years full-time PhD study (pro-rata for part-time students) and must commence in the 2018/19 Academic Year.  Applications for 1+3 funding for students completing a Master's year prior to the commencement of PhD study are also welcome (details available at www.southcoastdtp.ac.uk) and will commence in the 2019/20 Academic Year.

Application Procedure

The closing date and time for applications is initially 31st December 2018, however the advertisement will be withdrawn if a suitable applicant is identified.  The full application procedure, the funding application form, and more information on the South Coast Doctoral Training Partnership can be found at: http://southcoastdtp.ac.uk/apply/

For further information about this project, please contact the lead supervisor detailed above.  For questions relating to the application procedure, or for more information about the SCDTP, please visit the SCDTP website or contact us at scdtp@soton.ac.uk

Closing Date: 31 Dec 2018
Post Type: PhD Studentship (Funded)