Posit Team
Cutting down iteration cycles for credit rating models that drive better and more efficient decisions
Company
KBRA is a full-service global rating agency providing trusted credit ratings and research through an innovative approach.
Size:
600 employees
Industry:
Finance
Technology used:
Posit Team (Posit Connect, Posit Workbench, Posit Package Manager) Quarto, Vetiver, Shiny
Apps/products/solutions delivered:
Credit rating models, data-driven documents, APIs, Shiny applications, and many others
Matt McDonald's quantitative modeling team at KBRA leverages Posit to rapidly develop and deploy credit models, providing analysts with critical insights for informed rating decisions.
Established in 2010 after the financial crisis, KBRA is dedicated to the restoration of trust in credit ratings by creating new standards for assessing risk and offering timely and transparent ratings.
To make informed credit rating decisions, analysts need to understand the probability of default across various factors, such as cash flow in a company's balance sheet. As a regulated industry, KBRA’s quantitative modeling team must ensure their work is repeatable, transparent, inspectable, and well-documented to ensure it can be thoroughly explained.
Using Posit and a code-first workflow, KBRA’s quantitative modeling team significantly reduces the time to deliver transparent and explainable credit rating models to rating analysts.
“In the past, we would build a model, create a PowerPoint presentation and then schedule a meeting with our stakeholders to get feedback. Now, we can build our model and put out a working version on Posit Connect either as an API or as a Shiny app to get feedback in real time from stakeholders”
KBRA's journey to iterative model development with Posit Team
Prior to onboarding Posit Team, KBRA’s quantitative modeling team utilized Jupyter and RStudio locally to perform their work. While effective in many ways, this approach posed challenges for reproducing work and distributing models efficiently.
Recognizing the need for faster feedback loops, Matt aimed to streamline the model development cycle using Posit Team, ensuring that data science insights directly informed and improved credit rating decisions.
Building trust with stakeholders started with developing interactive applications using Shiny. Matt effectively leveraged Shiny in model validation, enabling clearer communication with subject matter experts on:
What is driving the model
What is driving the sensitivities
How does this actually work
Interactive applications empowered users to explore model behavior by adjusting sliders and clicking buttons, instantly generating the desired graphs. Users were impressed by how quickly they could achieve this level of interactivity.
Matt added: ”Those moments are not just useful for having a [way] to look at the sensitivities, but also from a more personal perspective. Suddenly, people were saying ‘Oh, Matt seems to know how to get things done. Maybe we should trust him’. “It’s looking for those types of opportunities, and there’s no playbook there. You have to keep your head up and know what you’re capable of and deliver.”
Since those initial Shiny applications, the team has standardized on Posit Team as a common environment, ensuring their work is consistent and repeatable.
“I’ve been hiring many people who use Python and the ability to run Jupyter Notebooks or VS Code in Workbench or to deploy Flask APIs or Streamlit apps has been really positive. Posit Connect streamlines the whole process of distributing whatever we are working on (a visualization, a data-driven document, an API hosting one of our models, a Shiny app, etc.)”
Quarto enables the team to seamlessly create internal research reports that integrate directly with underlying data, eliminating the need to share multiple versions of spreadsheets and Word documents. By publishing these reports to Posit Connect, they can easily update the data while maintaining a single, consistent link for business stakeholders. Further, stakeholders can also download PDF versions using the same link.
As a small team bridging the gap between a technology organization developing software and credit analysts with different toolkits, collaboration can be challenging. They must balance speaking the language of both groups while retaining ownership of their work.
KBRA’s quantitative modeling team is exploring the vetiver framework, developed by Posit, to enhance R and Python model workflows. Their goal is to make their models more accessible, particularly to users without programming expertise. For example, when deploying a model as an API to Posit Connect, they can also provide credit analysts with VBA code, enabling them to call the API directly within Excel spreadsheets.
Posit also facilitates collaboration with the technology team by enabling data scientists to prototype advanced simulation models using Shiny. While Git adoption presents a learning curve for many teams, Matt emphasized its crucial role in fostering teamwork. Git version control integrates seamlessly with Posit, ensuring reproducibility and smooth collaboration with the technology team.