Arbitrage Arena 2026 is a national-level quantitative finance competition where participants develop fully functional algorithmic trading models and portfolio optimization systems using real historical market data. The competition tests participants on model design, risk management, and robustness under stress conditions such as flash crashes and volatility spikes. Participants must choose one of two problem statements in the prelims and submit: A Jupyter Notebook (.ipynb) containing the full model implementation, A Technical Report (.pptx) summarizing approach, logic, and performance. Top-performing teams qualify for the Final Hidden Challenge, where a new dataset and modeling problem are released. The finalists will have time until 31 January 2026 to submit their final notebook + report. Final results will be announced during Pravega 2026 (Feb 7–8). What Participants Have to Do: Step 1 — Choose ONE Problem Statement Participants select either: Problem 1: Surviving the Crypto Flash Crash: Build a trading/risk model that survives flash crashes, minimizes drawdowns, and recovers quickly. Problem 2: Cross-Asset Portfolio Optimization: Construct a diversified systematic portfolio across equities, indices, and commodities with high Sharpe and low drawdowns. Step 2 — Use ONLY the provided datasets You will receive historical OHLCV data for: Cryptocurrencies (BTC, ETH), Equities (Tesla, Apple, Amazon, Microsoft, Nvidia, Meta), NASDAQ Index, Commodities (Gold, Silver, Crude Oil). Participants must load, clean, process, and analyze this data inside the notebook. Step 3 — Build Your Model Your model must include: For Problem 1 (Crypto Flash Crash): Volatility estimation, Crash detection logic, Position sizing / hedging strategy, Risk controls (stop-loss, exposure limits, drawdown limits), Return analysis before, during, after crash. For Problem 2 (Cross-Asset Portfolio): Return calculation, Risk estimation (covariance matrix, volatility, correlations), Portfolio optimization logic (mean-variance, risk parity, regime switching, etc.), Backtesting framework, Performance comparison across regimes. Step 4 — Submit Two Items: 1. Jupyter Notebook (.ipynb) Must contain complete code, explanation, outputs, and plots. 2. Technical Report (.pptx) Max 12–15 slides summarizing: Problem choice, Data processing, Model approach, Key formulas / logic, Backtest results, Metrics. REQUIRED NOTEBOOK FORMAT Your notebook must follow this structure: 1. Introduction State the problem you selected, Outline your model idea. 2. Data Import & Preprocessing Load provided datasets, Handle missing values, Convert timestamps, Compute returns / log returns, Align multiple assets (if needed). 3. Feature Engineering Examples (participants choose relevant ones): Moving averages, RSI, volatility, Correlation matrices, Crash indicators, Regime classification. 4. Strategy / Model Design Explain: Algorithms used (with formula), Risk management rules, Relevant constraints. 5. Backtesting Framework Must include: Start & end dates, Position rules, Portfolio value calculation, Transaction cost assumption (default 0 unless implemented). 6. Evaluation Metrics (Mandatory) For Problem 1 (Crypto Crash): Crash Survivability Index (CSI), Max Drawdown, Post-Crash Recovery Time, Sharpe Ratio. For Problem 2 (Portfolio): Annualized Sharpe, Sortino, Max Drawdown, Volatility, Correlation Heatmap, Stress-test performance. 7. Results & Plots Must include: Equity curve, Drawdown curve, KPI table, Rolling Sharpe. 8. Conclusion What worked, What failed, Possible improvements. EVALUATION & SCORING MATRIX: Prelims Scoring (100 points total) Component Weight Model Logic & Innovation 25, Data Handling & Preprocessing 10, Risk Management 15, Backtesting Framework 20, Evaluation Metrics Quality 20, Clarity of Explanation (Notebook + PPT) 10. Top teams qualify for finals based purely on these scores. Final Round Scoring (100 points total) Based on the hidden problem: Component Weight Model robustness 30, Performance under stress scenarios 30, Risk-adjusted returns 20, Reliability & reproducibility 20. only your final submissions are judged. IMPORTANT DATES: Registration 5–21 Dec 2025, Prelims Submission 5–25 Dec 2025, Prelims Result Announced Early Jan 2026, Hidden Final Problem Released Early Jan 2026, Final Submission Deadline 31 Jan 2026, Winner Announcement 7–8 Feb 2026 (Pravega). Rules: Use only provided datasets, Submissions must be original, Notebook must run end-to-end, Late submissions are not evaluated, Judges’ decision is final.