MacroLens

MacroLens A benchmarking corpus for **contextual financial reasoning under macroeconomic scenarios** across **4,416 U.S. small- and micro-cap equities (2021-01-04 — 2026-03-31)**. MacroLens unifies seven tasks over a single point-in-time panel: contextual time-series forecasting, public valuation, financial-statement generation, scenario-conditioned return forecasting, private-company valuation, generator evaluation from natural-language descriptions, and real-estate valuation. | Task | Type | Output | |---|---|---| | **T1** Contextual Forecasting | Time-series | Horizon-length close trajectory | | **T2** Public Valuation | Tabular regression | Equity market cap | | **T3** Financial Statement Generation | Structured generation | 11 canonical XBRL fields per (ticker, fiscal year) | | **T4** Scenario-Conditioned Return | Event forecasting | 63-day post-event return percentage | | **T5** Private-Company Valuation | Tabular regression (price-stripped) | Equity value w/o market data | | **T6** Generator Evaluation | NL→ structured | Same 11 fields from a natural-language company description | | **T7** Real-Estate Valuation | Cross-domain regression | Rent + price per RentCast address | Every instance carries a 131-numeric / 141-column point-in-time panel (prices, 46.8M XBRL accounting facts, 53 macroeconomic series, filing recency, derived ratios), an optional macroeconomic scenario object (1,130 events across 49 types), and optional SEC filings + financial-news context. Temporal alignment is strictly point-in-time: every observation visible at prediction timestamp $t$ was publicly available by $t$. Quickstart 10 lines per task; swap the model class to compare methods. Dataset structure **Note on `data/fundamentals/`**: Raw per-ticker fundamentals CSVs (~12,000 files: `{TICKER}_balance.csv`, `{TICKER}_income.csv`, `{TICKER}_cashflow.csv`) are **not shipped directly** here because HF enforces a hard cap of 10,000 files per directory. They are reproducible end-to-end via t