Open-Source Tools in Credit Risk Modelling: A Leap of Faith?

Like a herd of wildebeest, perched tentatively on the banks of a river’s edge, it takes a leap of faith to plunge into the unknown depths, and start an epic crossing to more fertile territory. Financial institutions are finding themselves on a similar precipice, pushed by cost pressures, archaic processes and a tempestuous regulatory environment, where the opportunities for smarter, faster, and more adaptable modelling lies on the other side.

In this blog series, we discuss why firms are taking the leap towards open source systems, including use of novel software tools and new ways of working, for deployment and validation of their risk models. In our first blog, we survey the current landscape, and look at what is pushing firms to make these technological migrations into new territory.

What is driving the migration?

Much as in the natural world, financial institutions must evolve and adapt to the environment around them to survive. And this environment has become a harsh one. Unprecedented uncertainty in the financial world has forced firms to challenge the adaptability of their models, against an environment of emerging risks. This is leading to most firms overhauling legacy models, in favour of new methodologies and techniques. Periodic model redevelopment is no longer just good practice, but also a regulatory necessity; with banking authorities frequently introducing new and updated model requirements, firms are now playing catch-up in a bid to retain model compliance. This is all happening against the background of a relentless commercial climate, where there is pressure to minimize the cost and time for deployment of risk models, in order to maintain a competitive advantage.

In the midst of these environmental perils, we are seeing an increased onus for modelled assumptions to be able to evolve efficiently to requirements. And since a Model’s DNA consists of a myriad of coding files, analytical procedures, statistical routines, judgements and conclusions, these aspects too must be sufficiently flexible to adaptation.

However, for most firms, changing the model DNA is a long and costly process. The current evolutionary path has led firms to adopt “run-it-once” workflows for their model deployment.  This means that any adaptation of the model DNA becomes a laborious effort, requiring analysts to trawl through a jungle of cluttered folders and questionable file names (who made this “Final_Final” file?). Prevailing operating models also fail to encourage code collaboration and shared effort, stretching development timelines, and increasing the risk of duplicating work. Finally, coding practices are often rigid and lack readability, further compromising effective code sharing.

What lies on the other side?

It is no surprise then, that eyes are turning towards greener pastures. Firms have already realised the benefits of open source programming languages (such as Python and R), which offer flexible ways of coding, as well as a release from the cost pressures of licensed software. Firms can take this leap further, through the use of version control software, wikis, and code inspection tools, which are open source, and can enable effective code collaboration.  Coupled with an agile project management structure, this allows for a more centralised and automated approach to developing and adapting the model DNA.

In many ways, these interactions draw parallels with a number of techniques that already exist in software engineering, which use a “DevOps” framework for agile software development. “DevOps” can be broadly seen as a set of principles in which developers can consistently build, test and deploy their code. By taking this framework into the model ecosystem, the result can lead to more streamlined, collaborative and efficient development of a firms risk models.

Taking the Leap

So, ready to take the jump? In the next blog, we take a leap into this new world, and explore the novel open source tools, coding practices, and project management structures, that are transforming model development practices across our industry, and changing what it means to “build” a model. 

However, caution must be taken, as this new territory is not without its dangers - we also look at what pitfalls will need to be navigated along the way.