In Focus Consumer Credit
Hurting those it seeks to help?
Research reveals that equality law can disadvantage women in algorithmic credit decisions
Galina Andreeva Senior lecturer in management science, University of Edinburgh Business School
galina.andreeva@ed.ac.uk
Research from University of Edinburgh Business School has revealed that including gender information in credit-scoring models increases the proportion of women accepted for consumer credit without altering the predictive accuracy of these models. Therefore, women can benefit from gender
being taken into consideration in automated decision-making processes, increasing their chances of being accepted for credit.
Prohibited Currently, using gender in decision making is prohibited by law in all major developed countries, with the EU’s European Equal Treatment in Goods and Services Directive also following this principle. Equal treatment is generally understood as
not using certain protected characteristics, including gender. However, the study found that by not considering gender as a variable, the outcome of credit decisions is not equal for men and women, and has a negative effect on women. The research analysed a sample of
79,386 customers who were issued a car loan from a major European bank between 2003 and 2009, and built four different models following a standard credit-scoring methodology. Using these models, researchers found that
gender is a statistically significant variable, yet its removal does not affect the predictive power of the models. This means that there is no effect on lenders as they can maintain similar levels of bad debt for a given acceptance level irrespective, whether gender is used or not. However, when including gender in one of the models, the effects for consumers changed.
26
accepted for credit decreased for women but rose for men – although, in general, women had lower rejection rates than men. This proved that following the principle
of equal treatment by omitting gender from the modes, as current regulation implies, did not lead to equality in the outcome.
Female loan applicants, having lower default rates in the past, benefitted from a model which included gender, giving them extra points for being good risks and as such, increasing their chances of being accepted for credit. Following this model, their rejection rates were lower compared with those for men
Female loan applicants, having lower
default rates in the past, benefitted from a model which included gender, giving them extra points for being good risks and as such, increasing their chances of being accepted for credit. Following this model, their rejection rates were lower compared with those for men. On the other hand, when applying a unisex model (without gender), the chances of being
www.CCRMagazine.com
Consequences What we set out to explore with our research were the consequences that legal restrictions on information, such as gender, can have in situations of automated decision making, for example when applying for a loan. Our findings highlight the inconsistencies
of the existing regulations as they do not ensure equality of outcome for consumers. From the data that we analysed we were able to prove that women can benefit from gender being included in credit-scoring models as this improves chances for them to be granted credit. Nevertheless, we would caution against
extrapolating our findings to all protected groups and all credit products. It is possible that in other credit portfolios protected characteristics show different patterns. Our research is an illustration to show
that equal treatment does not automatically translate into equal outcome. It also demonstrates that this happens because of correlation between gender and other predictors that remain in the models. We hope that our study will inspire the
development of better and more effective regulations for automated decisions and solutions for both consumers and lenders to make sure that everyone has equal and fair opportunities when applying for credit, no matter their gender or background. CCR
December 2019
Page 1 |
Page 2 |
Page 3 |
Page 4 |
Page 5 |
Page 6 |
Page 7 |
Page 8 |
Page 9 |
Page 10 |
Page 11 |
Page 12 |
Page 13 |
Page 14 |
Page 15 |
Page 16 |
Page 17 |
Page 18 |
Page 19 |
Page 20 |
Page 21 |
Page 22 |
Page 23 |
Page 24 |
Page 25 |
Page 26 |
Page 27 |
Page 28 |
Page 29 |
Page 30 |
Page 31 |
Page 32 |
Page 33 |
Page 34 |
Page 35 |
Page 36 |
Page 37 |
Page 38 |
Page 39 |
Page 40 |
Page 41 |
Page 42 |
Page 43 |
Page 44 |
Page 45 |
Page 46 |
Page 47 |
Page 48 |
Page 49 |
Page 50 |
Page 51 |
Page 52