FAMILY_STORAGE_DECAY: Experiment Log

Overview

Testing storage cost accumulation and quality decay impacts on Dutch potato prices through three mechanisms: simple monthly cost accumulation, temperature-accelerated deterioration, and stock depletion market power dynamics.

Hypothesis Origins

• Prior experiments: FAMILY_SPRING_VOL identified storage depletion as key volatility driver
• Industry catalyst: 2024 Dutch storage crisis with 650,000 tons lost and 33.2% import dependency
• Academic basis: Quantified decay models showing 2-5% monthly losses under optimal conditions

Experiment Design

• Method: Rolling-origin cross-validation
• Initial window: 156 weeks (3 years)
• Step size: 4 weeks
• Test windows: 52 weeks (1 year)
• Refit frequency: Varies by variant (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)
• CBS API: Table 85676NED (production/storage estimates)
• Open-Meteo API: Temperature and precipitation for storage modeling
• Version control: git:31ab258, CBS 2024-Q4

Experiment Runs

Variant A: Simple Decay Model

Status: Not started - Model: Linear regression with monthly storage cost accumulation - Features: months_since_harvest, cumulative_storage_cost (10 EUR/ton/month) - Horizons: 1-month, 2-month - Target: Test if linear cost accumulation explains price floors

Variant B: Temperature-Adjusted Decay

Status: Not started - Model: Multiple regression with temperature interactions, optional Arrhenius kinetics - Features: temp_excursion_8c, cumulative_degree_days, respiration_rate - Horizons: 1-month, 2-month - Target: Test if temperature stress accelerates price impacts

Variant C: Stock Depletion Dynamics

Status: Not started - Model: Threshold regression at 40% stock level, GARCH with depletion variance - Features: stock_depletion_pct, stock_below_40, market concentration proxy - Horizons: 1-month, 2-month, 9-month - Target: Test non-linear market power effects in low-stock regimes

Statistical Tests

• Diebold-Mariano test with Harvey-Leybourne-Newbold correction
• TOST equivalence test with SESOI = 5% improvement (0.075 EUR/100kg)
• Regime detection: Markov-switching (variant A), CUSUM (variant B), Bai-Perron (variant C)
• Threshold testing at 40% stock level for variant C

Verdicts

(No runs completed yet)

HE Notes

• Created 2025-01-16 based on RA literature review
• Builds directly on FAMILY_SPRING_VOL findings about storage depletion
• 2024 storage crisis provides natural experiment for validation
• All variants use ONLY REAL DATA from repository APIs
• Consider separate analysis for crisis period (2024) if patterns differ significantly

Decision Log

(To be added after experiments)

Verdict v1 — 2025-08-16

Label: INCONCLUSIVE Scope: Dutch consumption potatoes, 1-2 month horizons, weekly frequency Effect: ΔRMSE = -33.9% (storage: 7.330, baseline: 5.473 EUR/100kg) Stats: DM p=0.124 (HLN-corrected); TOST equivalent=False Data/Code: git=31ab258; Boerderij.nl API (NL.157.2086); CBS table 85676NED Notes: Insufficient evidence (p=0.124, improvement=-33.9%). Storage cost accumulation at 10 EUR/ton/month. MLflow Run: 1e37cc9b61d6446fb27197670138dfcc Artifacts: hypotheses/FAMILY_STORAGE_DECAY/artifacts/1e37cc9b61d6446fb27197670138dfcc/

Verdict v1 — 2025-08-16

Label: INCONCLUSIVE Scope: Dutch consumption potatoes, 1-2 month horizons, weekly frequency Effect: ΔRMSE = -33.9% (storage: 7.330, baseline: 5.473 EUR/100kg) Stats: DM p=0.124 (HLN-corrected); TOST equivalent=False Data/Code: git=31ab258; Boerderij.nl API (NL.157.2086); CBS table 85676NED Notes: Insufficient evidence (p=0.124, improvement=-33.9%). Storage cost accumulation at 10 EUR/ton/month. MLflow Run: 875c677190354501a8ffd204c93c0ef5 Artifacts: hypotheses/FAMILY_STORAGE_DECAY/artifacts/875c677190354501a8ffd204c93c0ef5/

Variant B Quick Test — 2025-08-16

Status: Preliminary test with 5 CV splits Temperature Model RMSE: 0.946 EUR/100kg Baseline RMSE: 13.091 EUR/100kg if baseline_rmse else "N/A" Improvement: 92.8% Preliminary Assessment: POTENTIALLY SUPPORTED (needs full CV) Note: This is a simplified test. Full rolling-CV with statistical tests required for final verdict. Data: Boerderij.nl prices + Open-Meteo weather (REAL DATA ONLY)

Variant C Quick Test — 2025-08-16

Status: Simplified test with 5 CV splits Stock Depletion Model RMSE: 1.473 EUR/100kg Baseline RMSE: 21.843 EUR/100kg Improvement: 81.3% P-value: 0.008 Preliminary Assessment: POTENTIALLY SUPPORTED Note: This is a simplified test. Full rolling-CV with statistical tests required for final verdict. Data: Boerderij.nl prices (REAL DATA ONLY), simplified stock estimation

Variant B Quick Test — 2025-11-11

Status: Preliminary test with 5 CV splits Temperature Model RMSE: 0.946 EUR/100kg Baseline RMSE: 13.091 EUR/100kg if baseline_rmse else "N/A" Improvement: 92.8% Preliminary Assessment: POTENTIALLY SUPPORTED (needs full CV) Note: This is a simplified test. Full rolling-CV with statistical tests required for final verdict. Data: Boerderij.nl prices + Open-Meteo weather (REAL DATA ONLY)