BUSINESS

Consumer credit score is key to lending

EVGENIA TZORTZI

TAGS: Banking

Efficient financial data reporting helps give consumers access to lending while limiting moral hazard, says Enrique Velazquez, director general of the European Union’s Association of Consumer Credit Information Suppliers (ACCIS), in an interview with Kathimerini.

Velazquez was in Greece recently for the annual ACCIS general meeting and conference, organized by its Greek member Tiresias.
 

What is the financial system’s approach to the need for responsible lending post-crisis and avoiding over-indebtedness? What has changed and what is the role of risk assessment systems?

The financial system and the regulatory / supervisory apparatus have learned the lessons from the financial crisis. Take, for example, the differences between the regulatory framework for consumer and mortgage credits. For both types of lending, lenders are obliged to lend responsibly. This obligation helps ensure that consumers are protected against over-indebtedness.

On the more general question of the importance of credit bureaus and the information they process in individuals’ lending, the best I can offer is to echo the views of the 2011 World Bank’s General Principles for Credit Reporting, which says that: “Credit reporting systems reduce information asymmetries by making a debtor’s credit history available to potential creditors and are therefore an effective tool in mitigating issues of adverse selection and moral hazard. Through credit reporting information and the tools derived from it (e.g. credit scores), creditors can better predict future repayment prospects based on a debtor’s past and current payment behavior and level of indebtedness, among other factors.”

How is the individuals’ privacy protected against the extended use of personal data analysis?

The legislation is there, in the EU’s General Data Protection Regulation (GDPR). We are working tirelessly to make sure that we are at all times compliant. Let me be very clear on the GDPR. Credit reference agencies – and their data users – have taken GDPR readiness very seriously. CRAs have been running multi-annual compliance programs before the date of application of GDPR (May 2018). CRAs have undertaken root-and-branch reviews of businesses, including products, services and internal systems. In this process, there have been – and still are – some challenges. For example, CRAs had to identify all their treatments of data and their data assets. This included giving consideration to the legal grounds for processing data, to the time the data could be retained in the systems, etc. CRAs are also constantly reexamining their procedures, to be able to better cope with the range of the data subjects’ new rights.

I am sure that all the work that the industry has put in will eventually pay off. It will however still take time for all stakeholders in the credit-ecosystem – including supervisors – to adjust to the ‘new’ normal.

What is the role of technology and new methods in risk assessment? What changes does profiling bring in assessing the consumers’ credit worthiness?

Technology is driving new forms of innovation that simply were not viable before. A key part of technological innovation is the use of new types of data which are currently being used and processed for the provision of financial services. For example, transactional data is particularly powerful. Data processing tools are also becoming more sophisticated: algorithms make it possible to analyze new ‘big data’ sets and the connections between different sets of data, including those supplied by the Internet of Things technology. Advances in technology and the capabilities of big data analytics and other tools have made it easier to create profiles.

The methods used for credit scoring and credit assessment are constantly evolving. In the past, traditional statistical techniques were the rule (e.g. linear regression models, discriminant analysis). Today, there are more innovative methods such as artificial intelligence, including machine learning such as random forests, gradient boosting and deep neural networks. Perhaps of more relevance is the availability of non-traditional data sources e.g. real-time transactional data, mobile and other devices data, social media, utilities data and data from applications. You name it.

The question is to what extent these data sources enhance the predictability of traditional (credit) related data. For example, transactional data can be used to identify whether consumers pay utilities, mobile phone bills and student loans on schedule. This behavior is associated to consumers that are most likely to make loan repayments on time. Hence, certain transactional data provides another measure that can determine the risk of lending to an individual.

Are there any new tools/methods for performing a reliable assessment of the consumers’ credit worthiness apart from the “traditional information,” such as income, living assets or alternative data?

Although there is not that much solid independent scientific evidence, we think that non-traditional data sources are a complement (i.e. not a replacement) to the predictability power of credit data and that those sources are particularly relevant for thin-file borrowers (i.e. borrower with little or no credit history).

The establishment of increasingly data-rich and accurate profiles facilitates the automatization of (credit-related) decision-making processes on credit applications. This is a separate issue. Broadly speaking, automatization is good for consumers as it cuts down costs, time, etc.

One has to be vigilant – and our industry is vigilant – that the development process by which profiles are built is documented. Models should be subject to an effective model governance framework that considers its conceptual soundness and regard for protected characteristics e.g. race, gender, religion, etc. However, as with any process, it needs to be explainable and operate within equal opportunity or discrimination laws.

In any case the consumer should have the choice of a non-automated processing in order to cover extreme cases in which the algorithms miss.

Online