MacroLens: New benchmark tests AI financial reasoning across 7 tasks

MacroLens, created by researchers including Patara Trirat, Jin Myung Kwak, Jay Heo, Heejun Lee, and Sung Ju Hwang, addresses a critical gap in financial AI benchmarks: the lack of a unified platform that jointly tests four key signals—price history, accounting fundamentals, macroeconomic regime, and contemporaneous text. The benchmark covers 4,416 U.S. small- and micro-cap equities over 2021-2026, providing a point-in-time panel of 46.8M XBRL accounting facts, 53 macroeconomic time series, 295,860 SEC filings, 215,882 news articles, and a novel scenario layer of 1,130 macroeconomic events across 49 types, automatically detected and rendered as natural language. This design respects real-world data challenges like publication-date gating, lagged reporting of fundamentals, redundancy in filings, and non-independent calendar splits.

Seven tasks test diverse financial reasoning: contextual forecasting, public and private valuation, statement generation from fundamentals and descriptions, scenario-conditioned returns, and real-estate valuation. The authors evaluate 19 methods across six families, from naive heuristics and gradient-boosted baselines to fine-tuned LLM-based time-series models and zero-shot large language models (LLMs). A five-step feature-context ablation on two frontier LLMs and a gradient-boosted baseline reveals that current models struggle with integrating macroeconomic context and handling scenario-based queries. The dataset and code are publicly released to spur further research. MacroLens aims to drive progress in AI that can reason like a financial analyst, combining quantitative data with qualitative context.