## Dataset Highlights

### Climate Risk Data
- **Daily assessments** across global corn-growing regions
- **Risk classifications**: Low, Medium, High based on crop-specific thresholds
- **Four risk categories**: Heat stress, cold stress, drought, excess precipitation
- **Regional aggregation**: Location counts by risk level for each region/day
- **Economic context**: Production share data for weighting regional importance

### Futures Market Data
- **Comprehensive pricing**: Corn (ZC), wheat (ZW), soybean (ZS) futures
- **Technical indicators**: Returns, volatility, moving averages
- **Market structure**: Term spreads and cross-commodity relationships
- **Daily frequency**: Aligned with climate risk assessments

## Evaluation Methodology

Your submissions will be ranked on the **Climate-Futures Correlation Score (CFCS)** - a single composite metric that measures how well your engineered features correlate with futures market variables.

### 1. Correlation Computation
- **Granularity**: Monthly correlations by commodity, country, and month
- **Scope**: All climate risk variables vs. all futures variables
- **Filtering**: Only correlations with sufficient variance are computed
- **Precision**: Results rounded to 5 decimal places

### 2. Composite Score Calculation

Your leaderboard position will be determined by a single **Climate-Futures Correlation Score (CFCS)** that combines multiple correlation metrics:

**CFCS = (0.5 × Avg_Sig_Corr_Score) + (0.3 × Max_Corr_Score) + (0.2 × Sig_Count_Score)**

Where:
- **Avg_Sig_Corr_Score**: Average of ONLY significant correlations (|corr| ≥ 0.5), normalized to 0-100 scale
- **Max_Corr_Score**: Normalized maximum absolute correlation (0-100 scale)  
- **Sig_Count_Score**: Percentage of correlations that are significant (≥ |0.5|)

### 3. Score Components Explained

**Average Significant Correlation Score (50% weight)**
- Measures quality of meaningful relationships (only correlations ≥ |0.5|)
- Rewards consistently strong signals, ignoring weak noise
- Formula: `min(100, |mean_significant_correlation| × 100)`
- Caps at 100 when average significant correlation reaches 1.0
- Returns 0 if no significant correlations found

**Maximum Correlation Score (30% weight)**
- Rewards discovery of exceptionally strong climate-market relationships
- Incentivizes breakthrough insights and novel feature engineering
- Formula: `min(100, |max_correlation| × 100)`
- Caps at 100 when maximum absolute correlation reaches 1.0

**Significant Count Score (20% weight)**
- Measures breadth of meaningful relationships discovered
- Rewards approaches that find multiple strong signals
- Formula: `(count_significant_correlations / total_correlations) × 100`
- Percentage of correlations with absolute value ≥ 0.5

### 4. Example Scoring
**Baseline Performance:**
- Average significant |correlation|: 0.55 → Avg_Sig_Score = 55
- Maximum |correlation|: 0.60 → Max_Score = 60  
- Significant correlations: 8% → Count_Score = 8
- **CFCS = (0.5×55) + (0.3×60) + (0.2×8) = 46.1**

**Target Performance:**
- Average significant |correlation|: 0.65 → Avg_Sig_Score = 65
- Maximum |correlation|: 0.80 → Max_Score = 80
- Significant correlations: 20% → Count_Score = 20
- **CFCS = (0.5×65) + (0.3×80) + (0.2×20) = 60.5**

## Potential Approaches

While we encourage creativity, here are some directions to consider:

### Feature Engineering
- **Production-weighted risk scores**: Combine regional risks with market share data
- **Temporal aggregations**: Weekly, monthly, or seasonal risk summaries
- **Cross-regional patterns**: Correlations between geographically distant regions
- **Risk momentum**: Changes in risk levels over time
- **Composite indices**: Multi-factor climate stress indicators

### Advanced Techniques
- **Non-linear transformations**: Capture threshold effects in climate-market relationships
- **Lag analysis**: Account for delayed market responses to weather events
- **Seasonal adjustments**: Normalize for typical seasonal patterns
- **Volatility modeling**: Link climate uncertainty to market volatility
- **Regime detection**: Identify different climate-market relationship periods

### Domain Knowledge Integration
- **Growing season alignment**: Match climate risks to crop development stages
- **Supply chain modeling**: Consider storage, transportation, and processing impacts
- **Market psychology**: Model how weather news affects trader behavior
- **Cross-commodity effects**: Leverage substitution relationships between crops

## Submission Requirements

Your submission should include:

1. **Engineered Dataset**: Enhanced version of the provided data with your novel features
2. **Feature Documentation**: Clear explanation of your feature engineering approach
3. **Code**: Reproducible pipeline for generating your features
4. **Analysis**: Demonstration of improved correlations vs. baseline methods

### **Critical Naming Requirements**

**⚠️ IMPORTANT**: Your submission must follow these naming conventions:

- **Climate Features**: All engineered climate features must start with `climate_risk_`
  - ✅ Good: `climate_risk_heat_stress_weighted`, `climate_risk_drought_ma_30d`
  - ❌ Bad: `heat_stress_risk`, `my_climate_feature`, `weather_index`

- **Required Columns**: Your submission must include:
  - `date_on`: Date column (YYYY-MM-DD format)
  - `country_name`: Country name matching the original data
  - `region_name`: Region name (optional but recommended)

- **Futures Data**: The evaluation system will provide `futures_*` columns - do not modify these names

**The evaluation metric automatically detects features by their prefixes. Incorrect naming will result in zero score.**

## Evaluation Timeline

- **Development Phase**: Use historical data to develop and validate approaches
- **Final Evaluation**: Submissions tested on held-out time periods
- **Correlation Analysis**: Comprehensive correlation computation across all dimensions

## Why This Matters

Success in this challenge has real-world applications:
- **Risk Management**: Better weather-based hedging strategies
- **Trading Alpha**: Novel signals for commodity trading
- **Agricultural Finance**: Improved crop insurance and lending models
- **Supply Chain**: Enhanced forecasting for food companies
- **Climate Adaptation**: Better understanding of weather-market linkages

## Getting Started

1. **Explore the Data**: Understand the climate risk classifications and regional patterns
2. **Baseline Analysis**: Compute initial correlations using raw climate data
3. **Feature Engineering**: Develop novel transformations and aggregations
4. **Validation**: Test your approaches across different time periods and regions
5. **Optimization**: Iterate to maximize correlation strength and significance

## Resources

- `submission_sample.ipynb`: Example evaluation methodology
- `corn_climate_risk_futures_daily_master.csv`: Main dataset
- `corn_regional_market_share.csv`: Production weighting data

---

**Ready to turn weather data into market insights? Let's see how creative you can get with climate intelligence!**

*Good luck, and may the correlations be with you!* 🌽📈