Skip to Main Content

The Search for Effective Fair Lending Models 

In the complex space of fair lending, the quest to ensure a level playing field can be a challenge. Banks don’t want to end up with models that are accurate but not fair. They also don’t want models that are fair but not accurate. With this in mind, banks are tapping machine learning capabilities to develop and optimize the “less discriminatory alternative” lending models required by regulators. But what exactly are LDAs, and why do they matter?  

Understanding LDAs. In a new RMA Journal article, Tobias Schaefer and Dmitry Lesnik of Stratyfy explain that LDAs are alternative decisioning models that maintain accuracy while reducing biases in lending practices. In a landscape governed by regulations that mandate transparency and non-discrimination in lending decisions, LDAs have emerged as a critical component of fair lending. 

The role of machine learning. With their high predictive power, advanced machine learning algorithms are at the forefront of this quest for fairness, Schaefer and Lesnik write. But regulators demand more than just accuracy. They require lenders to search for LDAs that minimize bias and provide transparent explanations for adverse actions. Model quality can be evaluated by analyzing disparities in lending decision rates among protected and reference groups. 

Innovative approaches. To achieve fairness, lenders have explored approaches such as tuning parameters in decisioning models and dropping variables, but each has drawbacks. One promising avenue is probabilistic rules models, which leverage probabilistic graphical networks to define decision rules that balance accuracy and fairness. These models offer flexibility, transparency, and the potential to uncover LDAs that align with regulatory expectations. 

Moving forward. As lenders navigate the complexities of fair lending, the search for LDAs continues to evolve. Many lenders might try to choose from third-party models. Wherever models are sourced, being able to ask the right questions about them is paramount. After all, Schaefer and Lesnik conclude, “making the right choice ... might not only ensure fairness but also increase revenue.”