QC2025 - Overview
| ← Write-Ups | Execution | Competition ↗ |
Part 1: Market Prediction
Dataset
| Set | Variables | Time Range |
|---|---|---|
| Training | 14 hidden vars (A-N, Y1, Y2) | 1-80000 |
| Test | Same vars excluding Y1/Y2 | 80005-96000 |
| Late Addition | O-P columns (~80% N/A) | 1 day before deadline |
Objective
Achieve highest R-squared score in out-of-sample forecasts.
Evaluation: Weighted average of Y1 and Y2 R-squared scores
| Details: Algorithm | Infrastructure |
Part 2: Basketball Trading
Materials Provided
- Variable documentation explaining factors influencing home team win probability
- JSON file with example game event sequences
- Base code for trading strategy
Objective
Achieve highest P&L based on trading strategy.
Trading Mechanics
| Action | Home Wins | Home Loses |
|---|---|---|
| Buy at 50 | +50 profit | -50 loss |
| Short at 50 | -50 loss | +50 profit |
Contracts pay 100 if home team wins.
Strategy: Possession-Based Model
| Component | Description |
|---|---|
| Outcomes | 4 per possession: [0, 1, 2, 3] points |
| Scoring | Score per second using team averages |
| Weights | Public basketball statistics |
| Adjustments | Individual player game fitness |
Concepts Explored: Bayesian updating, reinforcement learning, time-remaining probability
Details: Algorithm