Agentic Financial Reasoning, Hedging & Audit
IEEE Big Data Cup 2026 · Phoenix, Arizona · 14–17 December 2026
Official website · Submit Letter of Intent · IEEE Cup overview
FinReason Cup is Challenge 03 of the IEEE Big Data Cup 2026. It asks a central question for financial AI: can a system's reasoning be executed, checked, and reproduced—not merely presented as a plausible answer?
The competition brings together three complementary tasks spanning symbolic reasoning, sequential financial decisions, and structured verification.
Solve multi-step financial problems and return both a final answer and a step-by-step reasoning trace. Predictions are planned to be checked against gold traces generated from executable FinChain templates.
The exact problem subset, trace schema, ChainEval version, and numerical tolerance policy will be frozen with the starter kit.
Select an asset pair and manage a zero-net-dollar position over time using point-in-time prices, news, and corporate filings. The task is designed to reward relative-value reasoning rather than unhedged directional exposure.
The planned data are derived from HERCULEAN. Exact market windows, eligible assets, execution assumptions, transaction costs, and position-validity rules will be published with the dataset and scorer.
Perform targeted numeric-fact verification on organizer-packaged SEC EDGAR XBRL filing materials by comparing reported values with values derived from their calculation context. This task is not a full financial-statement audit.
The planned release combines public filing cases and separately constructed held-out cases subject to leakage review.
| Task | Provisional evaluation |
|---|---|
| Task 1 · Reason | Final-answer accuracy and step-level ChainEval |
| Task 2 · Hedge | Cumulative return, Sharpe ratio, and maximum drawdown |
| Task 3 · Verify | Accuracy and structural, extraction, and calculation error rates |
Final scoring formulas, tolerances, tie-break procedures, submission contracts, and validity rules will be published with the public scorers.
Teams planning to participate should submit one Letter of Intent per team:
Submit the FinReason Cup Letter of Intent →
The LOI supports challenge planning, organizer communication, and aggregate participation statistics. Technical participation and submission instructions will be released separately.
| Resource | Status |
|---|---|
| Official challenge website | Live |
| Letter of Intent | Open |
| Datasets and development splits | Coming soon |
| Starter kits, schemas, validators, and baselines | Coming soon |
| Participant support channel | Coming soon |
| Submission workflow | Under organizer validation |
The IEEE overview currently lists Kaggle with an organizer-run Docker evaluation path. Final verified platform and submission links will be posted on the official challenge website after workflow testing.
The confirmed conference dates are 14–17 December 2026 in Phoenix, Arizona, USA. Competition release dates, submission deadlines, and private evaluation milestones remain to be confirmed and will be published on the official challenge website.
This repository currently contains the source for the organizer-maintained FinReason Cup website. Verified participant resources—including datasets, starter kits, schemas, validators, baselines, and submission links—will be linked here as they are released.
Please use only links marked as verified on the official challenge website.
Requirements: Node.js 22.13 or newer.
npm ci
npm run devOpen http://localhost:3000.
Run the same checks used by the GitHub Pages workflow:
npm run lint
npm testnpm test builds the static GitHub Pages export and validates its rendered HTML
contract. Pushes to main deploy through
.github/workflows/deploy-pages.yml.
The Letter of Intent is hosted on Google Forms. Responses are available to the organizer team and are used for challenge operations, communication permitted by the form, and aggregate reporting. Do not include sensitive information. Participant support and the correction or deletion process will be published with the participant guidance.
The organizer team is led by The Fin AI, with contributors affiliated with MBZUAI, McGill University, Stevens Institute of Technology, Yale University, and the University of Manchester. Affiliations do not imply institutional sponsorship.
Competition dates, eligibility, platform rules, awards, and data terms remain subject to organizer and IEEE confirmation.
Last reviewed: 30 July 2026.