Who would have thought that a British cricket enthusiast would revolutionize fair lending?
Meet Lloyd Shapley, an unlikely hero of fairness analytics in the AI era.
The Inspiration
His inspiration?
A love for cricket and a keen interest in how to fairly attribute points in multi-player games.
His question: In a team sport, how do you fairly attribute credit (or blame) to individual players?
This cricket conundrum led Shapley to develop what we now call “Shapley values” — a concept that would revolutionize game theory and, decades later, transform fair lending analytics.
Shapley’s Method
Shapley’s curiosity led him to create a method that disaggregates each player’s contribution to a team’s overall success by considering every possible combination of players.
Fair lending compliance officers face a similar challenge: figuring out how different variables contribute to a loan being approved or denied.
While many fair lending pros still rely on regression analysis to grapple with this problem, the cutting-edge approach now embraces Shapley’s cricket-inspired innovation.
Breaking It Down
Let’s break it down:
Regression analysis aims to quantify how changes in independent variables (like income, credit score, race) affect a dependent variable (like loan approval), helping isolate factors contributing to disparities.
Shapley Values aim essentially to do the same thing, but work differently.
Imagine if we treated each factor in a lending decision like a player in Shapley’s beloved cricket match. (Or, let’s be real — few of us understand cricket so let’s use basketball).
Shapley values consider all possible “lineups” of players (variables), determining each one’s contribution to the final “score” (the lending decision).
Key Advantages of Shapley Values
- Handling complexity: They capture non-linear and interaction effects that regression might miss, much like how a basketball’s player’s performance might depend on who else is on the court.
- Accuracy in attribution: Just as Shapley ensured every cricket player got due credit, Shapley values consider all possible combinations of variables, enabling a more accurate understanding of the impact of each variable.
- Model-agnostic: Shapley values work with any model, from simple linear equations to complex machine learning algorithms – it’s like being able to analyze any sport, not just cricket!
- Interpretability: Shapley values offer clearer insights into how each factor contributes to the final decision, making it easier to explain to stakeholders.
The Shift to Shapley Values
The shift to Shapley values in fair lending is supported by recent research, including studies by FinRegLab and the public monitorship of Upstart Inc.
These studies show that Shapley’s cricket-inspired idea provides more accurate and nuanced insights into complex decisioning systems.
Conclusion
Are your fair lending analytics stuck in a pre-Shapley world?
FairPlay can help you score big at fair lending compliance – just don’t ask us to explain cricket!
