Our price forecasting model is based on key inputs and specific AI techniques.
Key Inputs:
Weather data and forecasts: ensemble modeling approach to track and forecast weather combines 20 years of historical climate risk trends, seasonality factors, and forecasted weather data (including forecasted climate risk trends)
Crop-specific machine learning models: based on our proprietary classification system composed of 81 distinct climate archetypes
Historical price trends: Our forecasting models integrate extensive historical commodity pricing data to identify and analyze recurring market patterns and correlations with climate-related disruptions
Supply and demand information: We continuously gather and analyze detailed global agricultural data, including production volumes, inventory statuses, trade flows, and consumer demand patterns.
Geopolitical risk monitoring: We monitor geopolitical developments by analyzing data scraped from 250,000 global news sources every 15 minutes, enabling us to identify and factor geopolitical risks into our forecasts.
AI Techniques utilized:
Time series analysis: a statistical technique used to analyze data points collected over time to identify patterns
Deep learning models: deep learning to analyze climate variables, soil conditions, and historical yield patterns, identifying correlations that traditional statistical models might miss
Anomaly detection models: these identify deviations from expected patterns in climate, trade, and supply chain data.
Ensemble forecasting techniques: Ensemble forecasting combines multiple models to improve prediction accuracy and robustness.
Pricing Backtesting & Validation
We evaluate forecast accuracy using standardized error metrics, primarily focusing on Mean Absolute Error (MAE) adjusted for commodity price levels. Each forecast is benchmarked against conventional prediction methods to verify performance advantages. We conduct error analysis across different market conditions to ensure consistent reliability.