FAMILY_INPUT_COST_TRANSMISSION: Experiment Log

Overview

Testing agricultural input price transmission to Dutch potato prices using Eurostat data for fertilizer, diesel, and electricity costs. This hypothesis uses REAL DATA ONLY from repository interfaces to measure cost-push inflation dynamics with 4-8 week lags.

Hypothesis Origins

• FAMILY_DIESEL_CORRELATION: Showed energy cost importance but used proxies and lacked other inputs
• FAMILY_STORAGE_TEMPERATURE_GRADIENTS: Storage costs matter but missed electricity price component
• 2022 Fertilizer Crisis: 300% price spike transmitted to potato prices within 6 weeks
• Industry Evidence: Dutch farmers report 4-8 week lag for input cost adjustments
• Academic Basis: McCorriston et al. (2001) cost pass-through; Gardner (1975) farm-retail transmission

Experiment Design

• Method: Rolling-origin cross-validation
• Training Window: 365 days minimum
• Step Size: 7 days (weekly)
• Test Window: 60 days maximum
• Baselines: ALL mandatory standard baselines (persistent, seasonal_naive, ar2, historical_mean)
• REAL DATA ONLY: Eurostat API + BoerderijApi + Open-Meteo

Data Sources (REAL DATA ONLY)

• Input Prices: Eurostat API - APRI_PI15_INA/INQ (fertilizer, pesticides, feed) - git:current
• Diesel Prices: Eurostat API - NRG_PC_204 diesel price indices - git:current
• Electricity: Eurostat API - NRG_PC_205 industrial electricity prices - git:current
• Potato Prices: BoerderijApi - NL.157.2086 consumption potatoes - git:current
• Weather: Open-Meteo API - for stress interactions - git:current

Experiment Runs

Variant A: NPK Fertilizer Price Transmission

Status: Pending - Previous run INVALID (used synthetic data) - Model: GradientBoosting with 4-8 week NPK fertilizer lags - Features: REAL NPK quarterly indices from APRI_PI15_INQ at 4, 6, 8, 10 week lags - Horizons: 30-day, 60-day - Mechanism: NPK fertilizer represents 25-30% of production costs - Expected: 18-25% improvement over seasonal_naive - CRITICAL: MUST use REAL Eurostat data only - NO synthetic data

Variant B: Energy Cost Pass-Through Model

Status: Pending - Model: RandomForest with 1-4 week energy transmission - Features: REAL diesel (NRG_PC_204) + electricity (NRG_PC_205) from Eurostat - Horizons: 30-day, 60-day - Innovation: Rapid energy cost pass-through for transport and cold storage - Expected: 20-28% improvement over seasonal_naive - CRITICAL: MUST use REAL Eurostat energy price data

Variant C: Combined Input Cost Index Model

Status: Pending - Model: XGBoost with production-weighted input cost index - Features: REAL NPK + diesel + electricity weighted by cost shares (30/15/10%) - Horizons: 30-day, 60-day - Innovation: Comprehensive input cost pressure with cumulative effects - Expected: 22-30% improvement over seasonal_naive - CRITICAL: ALL inputs from REAL Eurostat APIs only

Statistical Tests

• Diebold-Mariano test with Harvey-Leybourne-Newbold correction
• TOST equivalence test with SESOI = 20% improvement
• Granger causality test for input→price relationships
• Threshold cointegration for asymmetric transmission
• FDR correction for multiple comparisons
• ALL 4 standard baselines (persistent, seasonal_naive, ar2, historical_mean) included

Transmission Analysis

• Lag structure testing: 1-8 weeks for each input type
• Pass-through rate estimation via error correction models
• Asymmetry testing: Different speeds for cost increases vs decreases
• Regime-dependent transmission during input shocks

Verdicts

Variant A: NPK Fertilizer Price Transmission - 2025-08-19

Verdict: REJECTED - INVALID DUE TO SYNTHETIC DATA USE Scope: Dutch potato prices with fertilizer cost transmission Data Versions: INVALID - Used synthetic data instead of REAL Eurostat data

Baseline Comparison: - Results INVALID due to synthetic data violation - Previous run used fabricated fertilizer data - MUST be re-run with REAL APRI_PI15_INQ data only

Stats: N/A - Results invalidated Notes: This run violated the CRITICAL requirement to use ONLY REAL DATA from repository interfaces. The experiment used synthetic/mock fertilizer data and is completely invalid. MUST be re-run with actual Eurostat APRI_PI15_INQ quarterly NPK indices. MLflow Run: INVALID - e8a45ea4e0744cd889909fb2e4ffab98 (rejected)

HE Notes

• Created 2025-08-18 leveraging newly available Eurostat input price data
• First comprehensive input cost transmission analysis in repository
• 2022 fertilizer crisis provides natural experiment for validation
• All variants use ONLY REAL DATA from verified APIs
• SESOI raised to 20% due to strong theoretical foundation
• Critical for understanding cost-driven price dynamics

Decision Log

2025-08-19: Hypothesis Reformulation (HE)

• Critical Issue: Previous Variant A used SYNTHETIC fertilizer data - COMPLETELY INVALID
• Action: Rejected all synthetic data results, reformulated all variants with REAL data only
• Data Sources: Confirmed Eurostat APRI_PI15_INQ (NPK), NRG_PC_204 (diesel), NRG_PC_205 (electricity)
• Next Steps: EX must re-run ALL variants with REAL Eurostat data only
• Risk: Quarterly fertilizer data may limit statistical power - monitor alignment

(To be updated after valid experiment completion)