Reshaping macro investing

Recent advances in large language models have delivered measurable improvements in the extraction and interpretation of macroeconomic information, unlocking new and practical investment applications.¹

Identifying underappreciated views

Our work on LLM-based alpha engines shows the need for scalable feature generation pipelines and multi-agent workflows to improve accuracy in macro text. Our Macro Language Processing (MLP) platform uses semantic search and LLMs to extract nuanced macro sentiment from sell-side research without relying on rigid keyword dictionaries.

Rather than depending on rigid keyword dictionaries, our platform performs semantic search across a database containing millions of broker notes. LLMs allow us to filter down to what we believe are the most relevant excerpts relevant excerpts. A sequence of LLM scorers then validate relevance, identify forward-looking statements, and classify views along a long/short continuum. This framework is intended to give macro investors a scalable way to generate custom signals across a wide spectrum of macro impulses even where traditional macro data may be sparse or noisy.

Measuring market agreement to macro themes

Our Market Agreement to Themes (MATT) framework uses AI to quantify market consensus by decomposing proprietary themes into correlated directional hypotheses and analyzing broker report language to measure support or disagreement.

We aggregate agreement and disagreement across themes, dates, brokers, or other dimensions to build time series scores for each theme. Processing hundreds of relevant observations per day allows us to build a comprehensive and timely measure of market sentiment.

At the simplest level, MATT generates sentiment scores for all tracked themes—those we have traded historically, those currently active, and those being monitored for potential future impact. MATT offers us a structured way to test our macro views, assists in determining when to express them in trades, and effectively “debate” with the market about what’s priced in and what may be overlooked.

Extracting investor views with AI

We leverage LLMs to systematically harness BlackRock’s internal investor expertise. Using internal meeting summaries, these models convert unstructured narratives into actionable macro trades across rates, FX, commodities, and regional equities.

The mixture of speakers and breadth of topics acts as a natural crowdsourcing mechanism, where multiple trade recommendations with varying rationales inform the view on each asset at any moment in time. Take the below figure as an example, in early 2022, tight energy supply led to a long oil trade recommendation, while in mid-2025 rising U.S. fiscal risk prompted long gold and short dollar positions. These trades are combined into a composite signal traded systematically across more than 30 asset futures.

As technology evolves, so will the opportunities. With a robust infrastructure and rigorous validation frameworks, we believe LLMs will become foundational to macro alpha generation. The examples here represent early chapters in a much longer story and underscore our conviction that the fusion of advanced language models and macro research delivers a durable competitive edge.

Subject to change

Read the full report

Discover how macro teams are using LLMs to read the world’s economic narrative