Credit Scoring, the process of building mathematical models to predict future credit behavior, is ubiquitous in developed economies and offers significant potential for increasing access to finance and financial inclusion in emerging markets. From its origins in small value credit decisioning in the UK and USA credit scoring has provided, through automation, a means to serve mass markets in a cost effective way. Today credit scoring is being used for a wide range purposes. From making nano loans (up to $1) to pre-paid cell phone users to central bank portfolio supervision, and everything in between.

Credit scoring has always had its critics, typically for the lack of transparency, but the development of Big Data techniques is raising more than a few eyebrows amongst regulators and data privacy advocates. And, in my opinion, quite rightly so.

To explore the issues the Finance and Markets Practice Group of the World Bank incorporated the subject in its recent Financial Infrastructure conference in Istanbul and invited subject matter experts from both sides of the equation. The private sector was represented by a variety of companies offering data analytics products and services in both the developed and emerging markets. The regulatory perspective was provided by representatives from a wide spectrum of central banks. Also present were several members of the International Committee on Credit Reporting, who published the General Principles for Credit Reporting in 2011.

The discussion and debate orbited around the following issues.

Firstly there is no doubt that data modelling solutions can help individuals and small businesses access credit for the first time, often times without needing a relationship with a formal financial institution. The use of non traditional data, such as utilities and telco data, for credit scoring has been proven beyond contestation and there are promising signs that telco metadata and trawling non structured data (social media) could deliver similar levels of predictiveness. There is also evidence that methodologies such as psychometric modeling can deliver results in environments where very little other data exists. So from a scientific perspective we seem to have a possible Game Changer. But...

The issue of transparency, how the models are built and what characteristics lie within, has recently become a whole lot more complicated. The machine learning techniques that are increasingly employed to wade through the Big Data ocean are even more wizard like than traditional structured data modeling techniques. In goes a few Terabytes of information and out pops a model, hopefully without a puff of smoke. 

So why is this a problem? Two reasons.

Firstly, responsible lending using traditional credit scoring (bureau scores and ‘in house’ scores) has over the years developed, in most parts of the World, policies and procedures to explain to customers why they have not reached the required cut-off score and how they can change their behavior to, hopefully, improve the score next time. This usually involves identifying the top 3 or so characteristics (too many credit enquiries, length of time in employment etc.) where the applicant scored low with mostly intuitive ways to improve. But if the models are built in a black box using social media data or psychometrics how do you tell someone why they have failed?

Secondly, and perhaps more importantly, structured data can be filtered to remove characteristics that are undesirable. Whilst gender, ethnicity and geographical location are almost always predictive of credit behavior they have for the most part been outlawed in the traditional credit scoring world. In an unstructured data environment it is almost impossible to screen out such characteristics as the machine may detect patterns by inference. The ratio of constants to vowels in a customer’s name, for example, may not be directly racist but it is predictive of ethnicity, and a machine would not know that. Telco metadata may not have details of the customers address but your location is tracked every time you pass a cell tower so it does not take Einstein, over a period of a few weeks, to pinpoint where you live work and play.

There is a third reason, data privacy, but I won’t dwell on that issue here as frankly there is no such thing as data privacy when people blindly click ‘Yes’ I agree to the terms and conditions that I did not read.

So the logical conclusion is that we need a balance. The world, and in particular the developing world, needs credit scoring and innovation. But, bearing in mind that even in developed markets the level of financial literacy is intolerably low, we need some form of oversight to make sure that good governance is applied to the credit scoring industry. 

The problem, however, is that the credit scoring industry does not fall into an obvious regulatory bucket in most countries and in the absence of an obvious overseer/regulator, responsibility is, de facto, falling into the path of the central bank. I mean no disrespect but frankly most central banks ‘don’t know what they don’t know’ and have insufficient resources to monitor the practices of lenders and scorecard developers at the micro level. It is not uncommon in these situations for a regulator to over regulate and effectively throw the baby out with the bath water.

So the title of this piece is whether or not it is time for the credit scoring industry to have an official set of global guiding principles? Not prescriptive, this is what you can and cannot do, but directional, these are the issues and this is how best practice commonly addresses them. This at the very least would provide guidance to potential overseers and my hope is that it would secure the future of this business critical infrastructure.

The International Committee on Credit Reporting (ICCR) is currently considering whether or not to take on the challenge of producing such principles but before we do we would like to solicit feedback from the broader audience of interested parties.