How Are Banks Using AI for Stress Testing and Scenario Analysis?

What Stress Testing Is Designed to Do

Stress testing is a risk management practice where banks model how their financial position, including capital levels, loan losses, and overall balance sheet health, would hold up under hypothetical adverse economic conditions, such as a severe recession, a sharp rise in unemployment, or a significant decline in asset values. Following the 2008 financial crisis, regulatory stress testing became a formal, required practice for large U.S. banks, with the Federal Reserve conducting supervisory stress tests to assess whether major institutions could continue lending and absorb significant losses during a severe hypothetical downturn without needing emergency intervention.

Where AI Adds Capability to This Process

Traditional stress testing has typically relied on a relatively small number of predefined scenarios, in part because thoroughly modeling each scenario’s impact across a bank’s full balance sheet and loan portfolio requires significant analytical effort. AI and machine learning can expand this in a couple of important ways. First, they can help banks generate and evaluate a substantially larger and more varied set of hypothetical scenarios, since automated modeling reduces the marginal effort required to analyze each additional scenario, allowing risk managers to explore a broader range of “what if” situations beyond the standard regulatory set.

Second, machine learning models are often better suited than traditional linear statistical models to capture non-obvious, non-linear relationships between different risk factors, such as how a decline in one sector might indirectly affect risk in a seemingly unrelated part of a bank’s portfolio through complex economic linkages. This can help risk managers identify vulnerabilities that a more simplified traditional model might not surface clearly.

How This Fits Alongside Required Regulatory Testing

It’s important to understand that AI-enhanced stress testing generally supplements rather than replaces the specific regulatory stress test process, which follows scenarios and methodologies set by regulators like the Federal Reserve. Banks use AI-driven internal stress testing and scenario analysis as part of their own broader risk management practices, often examining scenarios more tailored to their specific business mix and risk exposures than the standardized scenarios used in required regulatory tests. This internal analysis can help a bank identify and address vulnerabilities proactively, complementing rather than substituting for the formal regulatory process banks are required to undergo.

Bottom Line

Banks use AI to expand and enhance stress testing by generating and analyzing a wider range of hypothetical adverse scenarios than traditional methods typically allow, and by helping identify non-obvious risk relationships across their balance sheets, work that supplements the formal regulatory stress testing process that large U.S. banks are required to undergo through the Federal Reserve.