FAMILY_BELGISCH_MARKT_KOPPELING: Experiment Log

Overview

Testing whether Belgian-Dutch coupling strength variations create predictable Dutch potato price movements through crisis coupling amplification, seasonal coupling patterns, and processing demand coupling using corrected methodology from FAMILY_CROSS_MARKET_COUPLING with Belgian-specific focus.

Hypothesis Origins

• FAMILY_CROSS_MARKET_COUPLING (CONDITIONALLY SUPPORTED - 86.8% improvement): Corrected weekly alignment methodology with 312 overlapping Belgian weeks provides proven foundation; honest attribution shows 16.1% from cross-market effects + 82.4% from domestic seasonality
• FAMILY_APRIL_STOCK_TIGHTNESS (CONDITIONALLY SUPPORTED - 82.5% improvement): Belgian TIGHT markets (<25% free ratio) show 74.5% higher prices; 2024 Belgian 24.82% free ratio classified as TIGHT creating procurement pressure
• FAMILY_WEATHER_ACCUMULATION (SUPPORTED - 95.5% improvement): Proven cumulative methodology for stress accumulation over time provides framework for seasonal coupling analysis
• FAMILY_BELGIAN_PRICE_SHOCK_TRANSMISSION (REFUTED): Despite confirming June 2021 Belgian shock (523.0) transmission model failed; innovation: focus on coupling STRENGTH variations vs shock transmission
• Industry catalyst: Belgian processors source 2.1M tons annually from Netherlands (45% of needs); June 2021 Belgian shock (523.0) demonstrated crisis coupling; 2024 TIGHT market (24.82% free ratio) created systematic procurement pressure
• Academic basis: Threshold cointegration theory supports regime-dependent coupling strength; processing integration creates systematic cross-border procurement patterns

Key Innovation - Belgian Specificity

Unlike FAMILY_CROSS_MARKET_COUPLING's multi-country approach, this focuses on Belgian-Dutch coupling with superior data density: - Belgian data richness: 312 overlapping weeks (57.8% coverage) vs 96 German + 124 French combined
- Processing integration: 45% Belgian sourcing from Netherlands creates systematic coupling - Geographic advantage: 150km NL-BE enables faster arbitrage vs 300km NL-DE, 500km NL-FR - Crisis amplification: June 2021 Belgian shock provided single-country crisis experiment

Experiment Design

• Method: Rolling-origin cross-validation using corrected weekly alignment methodology
• Initial window: 104 weeks (2 years minimum for seasonal patterns)
• Step size: 4 weeks (monthly refit)
• Test windows: Varies by horizon (30-day, 60-day)
• Refit frequency: Every 8 weeks for coupling adaptation
• Baselines: ALL 4 mandatory standard baselines (persistent, seasonal_naive, ar2, historical_mean)

Data Sources (REAL DATA ONLY)

• Dutch Prices: BoerderijApi NL.157.2086 (target variable)
• Belgian Prices: BoerderijApi BE.157.2086 (312 overlapping weeks, legacy=True) - VALIDATED from FAMILY_CROSS_MARKET_COUPLING
• Belgian Stocks: StockAPI FIWAP surveys (2010-2025) for TIGHT/NORMAL/LOOSE classification
• Transport Costs: CBS API 80416NED (diesel prices for €12/ton thresholds)
• Weather Data: OpenMeteoApi NL/BE locations for coupling modifiers
• Version control: git:exp/FAMILY_SEASONAL_PLANTING/variants_abc
• CRITICAL: Uses corrected weekly alignment (NL Monday with BE Friday in same ISO week) NOT fraudulent exact-date matching

Experiment Runs

Variant A: Crisis Coupling Amplification

Status: Ready for implementation - Model: RandomForest, XGBoost, Ridge regression - Features: be_nl_price_spread (REAL 312 weeks), crisis_coupling_strength, market_stress_amplifier, be_free_market_ratio (StockAPI), stress_duration, transport_activation (€12/ton), be_nl_correlation_dynamic, processing_demand_pressure, arbitrage_volume_proxy, crisis_momentum - Horizons: 30-day, 60-day - Target: Test if Belgian stress periods amplify NL-BE coupling strength - Expected improvement: 15-25% based on crisis amplification mechanisms - Mechanism: Market stress periods strengthen coupling as Belgian crises create urgent procurement from Dutch markets through crisis arbitrage channels

Variant B: Seasonal Coupling Patterns

Status: Ready for implementation - Model: ThresholdRegression, RandomForest, MarkovSwitching - Features: storage_season_coupling (Oct-May), growing_season_coupling (Jun-Sep), seasonal_regime_indicator, be_storage_pressure, seasonal_arbitrage_pattern, harvest_coupling_shift, quality_degradation_sync, seasonal_transport_costs, processing_seasonal_demand, coupling_regime_duration - Horizons: 30-day, 60-day - Target: Test if storage vs growing season creates distinct coupling regimes - Expected improvement: 15-25% based on seasonal regime switching - Mechanism: Storage season creates tight coupling through inventory synchronization while growing season shows looser coupling through independent production risks

Variant C: Processing Demand Coupling

Status: Ready for implementation - Model: GradientBoosting, RandomForest, Ridge - Features: be_fries_nl_consumption (grade arbitrage), processing_coupling_index, belgapom_export_intensity, dutch_procurement_signals, processing_margin_squeeze, be_de_competition_effect, quality_grade_arbitrage, processing_capacity_util, export_coupling_multiplier, cross_grade_transmission - Horizons: 30-day, 60-day
- Target: Test if Belgian fries industry demand creates coupling variations - Expected improvement: 15-25% based on processing demand transmission - Mechanism: Belgian fries industry creates systematic Dutch demand (2.1M tons annually = 45% of Belgian needs) with processing cycles driving coupling strength variations

Statistical Tests

• Diebold-Mariano test with Harvey-Leybourne-Newbold correction vs STRONGEST of 4 standard baselines (persistent, seasonal_naive, ar2, historical_mean)
• TOST equivalence test with SESOI = 8% improvement (standard threshold, not inflated)
• Directional accuracy threshold = 60%
• Regime detection: ThresholdRegression (B), MarkovSwitching (B)
• FDR correction for multiple comparisons across variants

Decision Criteria

• SESOI: 8% improvement threshold (standard methodology, not revolutionary claims)
• Statistical significance: p < 0.05 after FDR correction
• Practical significance: Improvement exceeds SESOI bounds
• Directional accuracy: ≥60% correct direction predictions
• Baseline requirement: Compare vs ALL 4 standard baselines, report vs strongest

Methodological Foundation

• Corrected methodology: Weekly alignment (NL Monday with BE Friday in same ISO week) from FAMILY_CROSS_MARKET_COUPLING
• Real data coverage: 312 Belgian overlapping weeks (57.8% temporal coverage)
• Honest expectations: 15-25% improvement based on Belgian specificity, not revolutionary methodology claims
• Transport thresholds: €12/ton using REAL BE-NL price differentials and CBS diesel costs
• Stock classification: TIGHT/NORMAL/LOOSE using REAL FIWAP Belgian stock surveys
• Processing quantification: 2.1M tons annual Belgian sourcing from Netherlands (45% of needs)

Expected Outcomes

Based on validated foundations: - Crisis periods: Amplified coupling during Belgian TIGHT markets (<25% free ratio) - Seasonal patterns: Tight coupling Oct-May storage season, looser Jun-Sep growing season - Processing cycles: Coupling strength varies with Belgian fries industry demand fluctuations - Transport activation: €12/ton threshold crossings trigger coupling regime changes - Overall improvement: 15-25% through Belgian-specific coupling analysis vs multi-country dilution

Next Steps for EX

1. Implement corrected weekly alignment methodology using FAMILY_CROSS_MARKET_COUPLING framework
2. Create Belgian-specific feature engineering with 312 overlapping weeks
3. Run rolling-origin CV with MANDATORY 4 standard baseline comparisons
4. Apply statistical tests with standard 8% SESOI (no inflated claims)
5. Focus on Belgian coupling specificity vs multi-country approaches

HE Notes

Creation Summary - 2025-08-19

• INNOVATION: Belgian-Dutch coupling specificity using corrected FAMILY_CROSS_MARKET_COUPLING methodology
• DATA FOUNDATION: 312 overlapping Belgian weeks (57.8% coverage) provides superior density vs multi-country
• METHODOLOGICAL BASIS: Proven weekly alignment methodology, NOT revolutionary claims
• REALISTIC EXPECTATIONS: 15-25% improvement based on Belgian focus vs multi-country dilution
• PROCESSING INTEGRATION: 2.1M tons Belgian sourcing (45% of needs) quantified systematic coupling
• CRISIS VALIDATION: June 2021 Belgian shock (523.0) and 2024 TIGHT market (24.82% free ratio) provide natural experiments
• HONEST FRAMEWORK: Standard 8% SESOI, mandatory 4 baseline comparisons, no inflated methodology claims