Citadel & Citadel Securities Competitions

Overview

Citadel is one of the world's most prestigious hedge funds. Their competitions are invite-based or application-based and directly feed into internship/full-time pipelines. Winning = near-guaranteed interview.

---

1. The Data Open (Datathon) — citadel.com/careers/the-data-open/

What it is: Teams analyze a large, complex real-world dataset and present findings to a Citadel judge panel. Format: 2–3 days, team of ~4, final presentation Prize: $15,000–$25,000 cash (varies by event) + recruiting pipeline How to enter: Apply via Correlation One (correlation-one.com) → 60-min assessment → team formation

Types of Datathons

Event	Eligibility	Prize
Women's Datathon 2026	Women in undergrad	$15,000 pool
Summer Invitational	Top undergrads (US/Canada)	$25,000+
Regional Datathons	University teams globally	Varies

How to Win

1. EDA first, model second: Judges care about insight, not just prediction accuracy
2. Narrative matters: Tell a story with your data. What did you discover? Why does it matter?
3. Presentation is 50% of the score: Practice your pitch. Every member should speak.
4. Use visuals: Clean matplotlib/seaborn/plotly charts > tables of numbers
5. Sanity-check everything: One wrong assumption in your analysis = credibility destroyed
6. Feature your unique insight: What did YOUR team find that others might miss?

---

2. Citadel Securities Quant Invitational — citadelsecurities.com

What it is: A hardcore quantitative challenge for top math/CS students Format: Invite-only, problems require deep probability + statistics + math Prize: Cash + recruiting fast-track How to qualify: Excel in the Data Open or apply directly

---

3. Terminal (Strategy Game) — citadel.com/careers/terminal/

What it is: Build an AI to play a tower-defense strategy game using Python Format: Online, open to all, year-round URL: terminal.c1games.com Prize: Recognition + recruiting signal Why it matters: Great entry point — no finance knowledge needed, pure algorithmic thinking

---

Preparation Resources

• Strong pandas, scikit-learn, matplotlib skills
• Practice on DrivenData and Kaggle (same "analyze + present" format)
• Read case studies: how hedge funds use data
• Practice presenting your analysis out loud (record yourself)

---

Confirmed Winning Formula (From 2021 PhD Datathon Champion + Team Documentation)

The datathon is an insight competition, not a modeling competition.

Research question (highest-impact decision):

• Choose impactful, human-relevant problems (not abstract optimization)
• Must be answerable with tools your team already knows
• Start with broad literature (IPCC reports, policy documents, domain papers), then narrow to a specific angle
• "Find a research question which is impactful and can be answered well using the tools you know." — 2021 winner

Data strategy:

• Supplement the provided dataset with higher-quality external sources
• Government databases, academic repositories, open data portals are fair game
• Document your sourcing — judges evaluate rigor
• Solve technical obstacles (file formats, APIs) — don't switch topics

Analysis:

• EDA before any modeling — always
• Feature engineering: remove highly correlated features, SHAP analysis to confirm feature importance
• Gradient boosting (XGBoost/LightGBM) + SHAP is the standard winning technical stack
• State assumptions explicitly — one wrong assumption challenged in Q&A = credibility destroyed
• Run at least 2 robustness checks and include them in the presentation

Presentation (worth ~50% of the outcome):

• Every chart must be self-contained — caption should convey the full point without reading surrounding text
• Every team member must present a named section
• Structure: context → key insight → evidence → implications → limitations → next steps
• Acknowledge limitations proactively — earns intellectual trust from judges
• Practice under time pressure, multiple times

Q&A preparation:

• "Why didn't you try X?" — pre-answer the obvious alternatives
• "How robust is this?" — have 2 robustness checks ready
• "What would change your conclusion?" — know your assumptions cold