FAMILY_REGIONAL_ARBITRAGE: Experiment Log

Overview

Testing whether regional price differentials between Dutch production regions and neighboring markets create predictable arbitrage opportunities that influence spot prices through transport economics, storage distribution, and cross-border trade flows.

Hypothesis Origins

• FAMILY_NW_MARKET (CONDITIONALLY SUPPORTED): Regional dynamics proven with RMSE 1.55-1.76 EUR/100kg but limited to NL-only data
• FAMILY_IMPORT_FLOWS (INCONCLUSIVE): Transport costs identified as key driver despite data limitations
• FAMILY_SUPPLY_CHAIN_INTEGRATION Variant B (SUPPORTED): 64.8% improvement with regime-switching suggests arbitrage may exhibit regime behavior
• Industry catalyst: 2024 storage crisis with 45% increase in Belgian imports; traders report systematic NL-BE convergence when spreads > €12/ton
• Academic basis: Devaux et al. (2022) showing 96-99% correlation with 8-12% exploitable spreads; Cambridge (2024) confirming 10% differentials trigger flows within 14-21 days

Experiment Design

• Method: Rolling-origin cross-validation
• Initial window: 156 weeks (3 years)
• Step size: 4 weeks
• Test windows: Varies by horizon (1m, 2m)
• Refit frequency: Every 8-12 weeks
• Baselines: Naive seasonal, ARIMA, linear trend

Data Sources (REAL DATA ONLY)

• Boerderij.nl API: Products NL.157.2086 (consumption), NL.157.2083 (fries quality)
• CBS API: Tables 80416NED (diesel/transport costs), 85676NED (regional storage)
• Open-Meteo API: Weather data for NL (52.6°N, 5.7°E) and border regions
• Local CSV: Regional production and acreage data
• Version control: git:66ca8cd, CBS 2024-Q4
• CRITICAL: NO synthetic data - all features derived from REAL repository interfaces

Experiment Runs

Variant A: Transport Cost Arbitrage

Status: Not started - Model: RandomForest, XGBoost, Ridge regression - Features: price_differential_ma7, transport_cost_index, price_spread_ratio, convergence_signal - Horizons: 1-month, 2-month - Target: Test if price spreads exceeding transport thresholds (€12/ton) converge within 2-3 weeks - Expected improvement: >5% based on transport cost dynamics

Variant B: Storage Distribution Arbitrage

Status: Not started - Model: GradientBoosting, RandomForest, ElasticNet - Features: storage_pressure_index, regional_stock_ratio, storage_cost_gradient, border_distance_km - Horizons: 1-month, 2-month - Target: Test if regional storage imbalances predict 5-8% price premiums - Regime detection: Markov-switching with 3 storage pressure regimes

Variant C: Cross-Border Integration

Status: Not started - Model: Stacking ensemble (RF + XGBoost + GradientBoosting with Ridge meta-learner) - Features: Combined transport + storage + market integration signals - Horizons: 1-month, 2-month - Target: Test if integrated arbitrage model achieves >5% RMSE improvement - Regime detection: Bai-Perron structural break test

Statistical Tests

• Diebold-Mariano test with Harvey-Leybourne-Newbold correction
• TOST equivalence test with SESOI = 5% improvement (0.75 EUR/100kg)
• Directional accuracy threshold = 60%
• Regime detection: Threshold (A), Markov-switching (B), Bai-Perron (C)
• FDR correction for variant C (multiple comparisons)

Arbitrage Regime Definitions

• Active arbitrage: Price spread > transport cost threshold (€12/ton)
• Storage imbalance: Regional utilization difference > 20%
• High integration: 30-day correlation > 0.95
• Crisis regime: Import dependency > 30% (2024 benchmark: 33.2%)

Verdicts

(No runs completed yet)

HE Notes

• Created 2025-08-17 building on regional market findings from three prior families
• Key innovation: Explicit arbitrage mechanism modeling with transport thresholds
• All variants use ONLY REAL DATA from repository interfaces
• Transport costs proxied through diesel prices; cross-border prices inferred from patterns
• Storage distribution based on CBS regional statistics
• Consider enhanced data collection for BE/DE prices if initial results promising

Decision Log

(Awaiting experiment results)

Codex Validation — 2025-11-10

Files Reviewed

• hypothesis.yml
• hypothesis.md
• experiment.md

Findings

1. No code in repo. The family comprises documentation only; there is no runner or dataset builder for the arbitrage variants.
2. No real-data evidence. With no script, we cannot confirm that transport or cross-border price feeds were accessed.
3. Baseline comparison absent. The experiment log says “No runs completed yet,” so no metrics exist against the price-only baselines.

Verdict

NOT VALIDATED – The hypothesis remains untested until executable code ingests real data and demonstrates improvements over the required baselines.