What happened
BIS Working Paper No 1395 (by Marco Lombardi, Taejin Park and Fernando Perez-Cruz) introduces MaRCo@BIS — the MAcroeconometric Research Companion — an agentic-AI framework to support macroeconometric research in central banks. Unlike conventional econometric software that estimates researcher-specified models, the agent helps formulate the research design itself by selecting variables, identification strategies, samples and model assumptions while keeping "consequential decisions" with the human researcher. It benchmarks the agent's replications of three widely-used structural VARs against the originals and illustrates the framework across five exercises. The paper finds agentic AI can substantially reduce the mechanical effort of econometric research and surface economic/data problems early, but concludes "human review remains essential because of the risks of cherry-picking specifications, automation bias, data limitations and economically inappropriate interpretations." The framework is released as a public good for the central banking community.
Why it matters
This is a Tier-1 central bank's first systematic design for governing agentic AI inside consequential analytical work — a concrete reference on human-oversight placement, specification-search and automation-bias risk as agentic AI moves into institutional decision-making.
Action needed
For analytics and financial-stability leads, map the MaRCo@BIS human-review and audit-log design onto your own quantitative environments as a template for governing agentic AI.