A guest post from Dr. Martin Seeger, Credit Risk Manager and Data Scientist, BeeEye’s Advisory Board member
Credit risk modeling poses different challenges to banks and Fintech companies. A solution can lay in a modeling platform that caters for both
In the context of credit risk modeling, traditional financial institutions (e.g., banks, insurers) and Fintech companies both have very specific risk profiles they need to maintain. For both, the cost of inaccuracy is incredibly high.
However, banks and fintech companies face two very distinct challenges when it comes to achieving this covetable accuracy.
Updating your platform once in a blue moon
If you’re a risk analyst or modeller in a traditional financial institution, you are likely all too familiar with regulations’ impact on internal processes. Specifically, how regulations stretch out these processes in a way that may hurt modeling accuracy. Usually banks perform rating model validations only once a year due to the complexity of the processes regulations require. When they do, validation data may be a year old (or more). In the current crisis, with drastic changes in default rates and shifts between sectors (think e.g. tourism vs. food retailing) this is too slow for your models to reflect reality.
IT and legacy tools weighing you down
When there are too many cooks in the kitchen, you are bound to spoil the broth. And when your sous-chef sits in the IT department, chances are the legacy, in-house tools are overly complex and inflexible, and rely heavily on slow IT processes. The combination of both makes for sub-optimal infrastructure in times where one would want to be able to test many model variants and their impact on credit portfolios quickly.
Relying on limited data sources
So you found a new database that you believe would infinitely improve your model’s accuracy. More often than not, modellers in financial institutions want to make new data sources available to their modeling landscape (e.g. bureau data, web data, centralized/pool data, market data). Any source that was not previously deployed will need to go through IT to be added, which is a time-consuming and expensive process. So you will likely find yourself questioning whether it is worth it at all, and give up the option to test it out. Unless, of course, you have a tool that allows you to accelerate the provision of raw data to the modellers and bypasses the in-house IT.
While fintech companies usually are not dealing with legacy IT tools, they certainly have their own challenges
Starting from scratch
The benefit of not having a legacy platform or tool can be outweighed by the complete lack thereof. Very often, the tools and processes these companies use are lacking in terms of capabilities and data sources. Analysts more often than not start their job with a blank jupyter notebook. They can be overwhelmed by the task of developing and deploying a credit risk modelling infrastructure from scratch.
In fintech companies, data is initially often obtained from third parties such as credit bureaus or data vendors. Data is also obtained from unconventional, unstructured sources such as social networks or internet platforms. In a situation like this, where many APIs need to be combined, it is welcome or sometimes even required to work with a tool that has the ability to integrate different data sources. This tool should be flexible enough to allow the analyst to deal with changing API definitions.
My advice to you, whether you are modeller in a traditional financial institution or a fast-paced fintech company, is to find a platform that solves for your key challenges. For traditional institutions, find an agile platform that enables shorter update cycles. This way you’ll be working with a fresh set of data. You might also prefer a platform that eliminates a lot of the red tape either via smart automations. A no-code approach will also eliminate some of your IT dependencies and let you easily test out features and datasets. For fintech modellers, find a platform or tool that enables easy integrations of internal and external data sources, as well as some model-enhancing capabilities like feature generators.
In these complex, uncertain times, finding a good platform to start you off is half the battle. Good luck!