The Credit Research Centre Seminar series featured Professor Edward I. Altman, the creator of the famous z-score model. Prof Altman spoke about 'global zombies'. The report was co-authored by Rui Dai from the Wharton Research Data Services of the University of Pennsylvania and Wei Wang from the Smith School of Business, Queen's University.

The seminar proposed a two-step filtering process based on interest coverage ratio and default prediction models to assess the global extent of "zombieism": the existence of companies that were insolvent but continued to exist due to unusual market conditions, financial institutions, and investor or government support. The seminar also discussed how the average fraction of publicly traded zombie firms in the world's 20 largest economies has increased significantly over the last 30 years but has barely changed during the pandemic period in 2020, concluding that democratic accountability, financial market development, creditor rights, and debt enforcement efficiency can help to explain cross-country variations in zombie ratios. The seminar additionally addressed adopting staggered bankruptcy reforms in eight countries after 2000 as an exogenous variation to bankruptcy code modernization and discovered that the zombie ratio decreases by an average of 1.4 percentage points after the reforms in those countries.

Read the complete paper for a more in-depth understanding of the seminar and its significance.

• Global Zombies

Galina gained her PhD in credit scoring in 2004 and has published papers in world leading journals on international scorecards, modelling risk of default in SMEs, and algorithmic fairness.

• Dr Galina Andreeva's Profile

The seminar highlighted the significance of Credit scorecard modelling in banks and agencies, which has attained a high level of good practice over many years. Many of the early pioneers' and practitioners' beliefs are now well-tested and even required by the profession:

• Logistic regression
• Input variable classification
• Default over-weighting
• Reject inference
• Delta method for scorecard modification
• Use of AUC metrics and much more

These practices may come with warnings or limitations in their application, or they may be recognised as approximations or to have potential biases or errors, but guidance is frequently simply cautionary — "it's better to do it, but watch out in X circumstances!" — and in need of quantification — "add some conservatism to cover Y." Determining when to abandon one traditional technique or how to correct another is still a substantial concern in model management.

Alan offered a geometric description of scorecard adjustment and selection in this session to help address this issue. The seminar's goal was to unite and simplify a complicated menu of excellent practices as examples of geometric connection or misunderstanding in a high-dimensional model and data space.

This point of view is based on statistical information geometry, a lively area of mathematical statistics that began with Rao's differential geometric approach to Fisher information. By applying this basic theory to the situation of scorecards - regressions on contingency tables — the mathematics is considerably simplified, creating a structure that facilitates practical computation and insight.

To demonstrate this, Alan quantified three traditional scorecard practices: model selection, the delta approach in the presence of heavy correlation, and sample weighting. This provided new insights, often confirming the classical approach's correctness while also correcting or simplifying it. It also makes this powerful idea available to scoreboard creators for use in solving other comparable challenges.

The seminar highlighted the use of Scoring models to support resource allocation decisions in finance. Credit scorecards were seen as a prominent example. As they are trained using data from past accepted applicants whose repayment conduct has been observed. This results in sampling bias because the training data only represent a subset of the borrowers to whom the model is applied in operations.

The presentation also emphasised how the research contributed to the field of reject inference in the following ways. First, the report quantified loss due to sample bias using a simulated study and demonstrated how the bias affects scorecard training and evaluation. Second, the research proposed a shallow self-learning approach that addresses training bias and mitigates performance loss by inferring labels from selected rejected applications. Finally, the article developed a new evaluation tool known as the kick-out measure. This measure lowers the bias associated with evaluating a scorecard using a biased training sample of accepted clients and a test sample indicative of the scorecard's operating conditions. Experiments using real credit scoring data proved the proposals' superiority in terms of prediction performance and profitability over existing bias-correcting approaches.

View the full paper

The seminar addressed significant findings from two articles that focused on the effects of energy efficiency and extreme weather in mortgage credit risk. Both examined at a one-of-a-kind micro-level data set on home mortgages in the United Kingdom. According to the findings reported in the first study, mortgages secured by energy-efficient properties are less commonly in arrears than mortgages secured by inefficient properties. As a result, the presentation indicated that energy efficiency is a significant predictor of mortgage defaults.

Key findings from a second paper were also used to supplement the presentation. It investigated whether lenders take extreme weather events into account. It was determined that lender values do not "mark-to-market" in response to local price reductions. As a result, valuations are skewed to the upside. Second, lenders do not modify interest rates or loan amounts to compensate for the valuation bias. Third, low-credit-risk borrowers self-select into high-risk flood zones.

Global investment firm Baillie Gifford has dedicated funding to support research into the ethical challenges posed by the growing use of data and artificial intelligence. As part of this initiative, the University of Edinburgh welcomes applications in the topic of ‘Fair and Ethical Credit Decisions’, supervised by Galina Andreeva (Business School) and Michael Rovatsos (School of Informatics).

While the selected student will be supervised by the University of Edinburgh Business School, they will also have the opportunity to take part in collaborative cohort activities as one of five PhD students in the Edinburgh Futures Institute’s (EFI) Baillie Gifford scholarship programme in the Ethics of Data and Artificial Intelligence.

We are looking for someone with a good background in one or more of the following areas:

• Statistics
• Data Science
• Data Analytics
• Computer Science
• Informatics
• Financial Modelling

This particular project employs mixed methods of quantitative and qualitative research. Yet candidates with background in only one area (qual or quants) are encouraged to apply if they are willing to learn additional skills.

The studentship includes full tuition fee coverage for up to 4 years, with a yearly stipend at UKRI rates (estimated to be approximately £15,245) and an annual research budget of £2,000.

Deadline is 12 noon, 15 May 2020.

• More on the Edinburgh Futures Institute PhD Studentship

Interested candidates may also want to read a recent paper relevant to the project topic:

• The law of equal opportunities or unintended consequences? The effect of unisex risk assessment in consumer credit

Overview

This lecture discussed how to leverage social networks effectively and efficiently for predictive modelling. Three practical applications were chosen: churn prediction in the telecommunications industry, credit scoring, and insurance fraud detection.

In the first two situations, networks were built using call detail records (CDR), which gave an accurate depiction of people's activity and so became a valuable source of data for researchers in fields such as physics, sociology, epidemiology, transportation, and networking.

The performance of network learning techniques for churn prediction was compared to that of normal binary classifiers (such as logistic regression and random forests) with network characteristics extracted, which was a more traditional approach. According to the findings, churn influence does not extend widely in the social network, and churn status inside a customer's ego-net is strongly predictive of churn.

The conference highlighted the added value of integrating cell phone data, or CDR, in credit risk modelling in the second application, credit scoring. The results showed that variables reflecting calling behaviour were the best predictive in terms of statistical and economic model performance. These findings also revealed significant regulatory, privacy, and ethical consequences.

Finally, the purpose was to discover groups of collaborating fraudsters by connecting claims and the parties involved in a massive social network to detect fraudulent insurance claims. As a result, rather than just the traditional components of the claim, the policyholder, and the policy, the social structures of fraudsters have also been established in insurance fraud detection approaches and models.

Dr McVittie’s seminar dealt with the topic of ‘Macroeconomic Forecasting Uncertainty and IFRS 9 ECL’. The presentation was well-attended by practitioners and academics keen to expand their understanding on the subject. Dr McVittie is an economist and data scientist with extensive experience with the use of economic data and forecasts in credit risk analytics, including IFRS 9. Following an academic career, laterally as Head of Economics at the University of Plymouth, he spent 13 years as Lead Consultant and Research Director with Experian, where he advised, designed products, and delivered projects across a range of areas related to macroeconomics, credit risk, regional economics, and household finances.

The second presentation in the series was to be held on 28 February on ‘Loan Default Analysis in Europe: Tracking Regional Variations Using Big Data’ with Dr Luca Barbaglia of the Joint Research Centre, European Commission. Unfortunately the presentation was postponed but we hope this will be reinstated later in the year.

Upcoming seminars

• 17 April 2020, Dr Benjamin Guin: The roles of energy efficiency and extreme weather for credit risk of residential mortgage lending
• 8 May 2020, Dr Stefan Lessman: Fighting the sampling bias: A novel framework for training and evaluating credit scorecards

The seminar was packed with panel attendees and delegates.

Panel Attendees were as follows:

• Jonathan Crook (Chair), Professor of Business Economics, Deputy Dean and Director of Research, University of Edinburgh Business School, Director of the Credit Research Centre
• Christian Burgin, CFA, Four Two Strategy Ltd
• Colin Garland, Director, Remedies, Business and Financial Analysis, Competition and Markets Authority
• Manuel Peleteiro, Founder, Inbestai

The conference addressed the Open Banking regulation, which went into effect in the United Kingdom in January 2018. The rule compels banks to allow third-party access to an account holder's banking data in an accessible format.

The seminar highlighted the fact that retail and business customers could now have access to new competitively priced products and services and the providers would be regulated by the Financial Conduct Authority (FCA) and European counterparts.

Unfortunately, this concept was not universally supported. Some anticipated that it would benefit just the technologically competent and exacerbate financial exclusion for low-income people. The question raised was whether it was reasonable to expect consumers to own their data and receive better deals from banks and other financial service providers and whether personal data revealed in places like social media could be misused. How could the bank be benefited from the regulation?

The successful applicant will work on research using transactions data for credit risk modelling with the ID Co. and be supervised by Jonathan Crook and Galina Andreeva.

Project Background

The traditional credit scoring models have used application form (and behavioural) variables with credit reference agency variables giving additional information on accounts at other lenders. However, these predictors are, at the most frequent, measured monthly; the application variables (for example income, address, and so on) are not updated; and crucially, they do not give an accurate direct indication of the ability of the account holder to repay any loans granted. Essentially, these variables do not give an indication of an account holder’s cash flow.

On the other hand, account-level transactions data provides daily information on all receipts and expenditures for an account holder for each account for which data is obtained. This information allows a very accurate daily measure of income (stable and volatile) and a fine classification of expenditures by service/product type and by merchant. Following expenditure categorisation and income aggregation across sources and classification into stable and volatile components, a full cash flow analysis for each account holder may be obtained on a daily basis. When used as covariates in a probability of default (PD) model, such covariates are expected to provide a much more accurate prediction of PD for each account holder than current models.

This project will develop a methodology for incorporating a novel type of digital data (financial transactions) into credit risk and affordability models. Transactional data provides more accurate and up-to-date information about the financial status and behaviour of the borrower, compared to traditional data, which is static and often outdated. Despite the great potential of transactional data, its current use is limited because of technical problems which this project will overcome. The project will experiment with innovative categorisation/aggregation algorithms. It will also estimate application and behavioural credit risk models using a range of advanced statistical and machine-learning algorithms.