FAMILY_FERTILIZER_LAG_TRANSMISSION: Experiment Log

Overview

Testing fertilizer cost LAG TRANSMISSION effects where 2022 European fertilizer crisis cost advantages transmit to Dutch potato prices through 6-18 month production cycle delays rather than immediate cost pass-through. This hypothesis uses REAL DATA ONLY from repository interfaces to test temporal lag mechanisms.

Hypothesis Origins

• FAMILY_REGIONAL_INPUT_DIVERGENCE: STRONGLY REFUTED (-71.9% vs baseline) testing IMMEDIATE transmission
• Critical Learning: Agricultural markets respond through production cycles, not immediate cost pass-through
• Natural Experiment: 2022 fertilizer crisis created verified regional cost divergences
• Innovation: LAG TRANSMISSION approach (6-18 months) vs failed immediate transmission
• Data Foundation: Destatis API (German crisis peak 216.8), INSEE API (French +40.1% impact)

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: Destatis + INSEE + Eurostat + BoerderijApi

Data Sources (REAL DATA ONLY)

• German Fertilizer: DestatisPublicAPI - agricultural producer prices - git:current
• French Fertilizer: INSEEPublicAPI - input_prices.fertilizer_index (001762871) - git:current
• EU Harmonized: EurostatAPI - APRI_PI15_INQ quarterly fertilizer indices - git:current
• Potato Prices: BoerderijApi - NL.157.2086 consumption potatoes - git:current

Natural Experiment: 2022 Fertilizer Crisis

Crisis Documentation (VERIFIED REAL DATA): - German Impact: Fertilizer index peaked at 216.8 (Destatis API, 1,142 observations) - French Impact: +40.1% fertilizer increase (INSEE API, 1,534 observations) - Regional Divergence: 5.4x difference in crisis response between countries - Timing: Q1 2022 crisis → production advantages → 6-18 month later market effects

LAG TRANSMISSION Theory: - Feb-Apr 2022: Fertilizer cost advantages locked in during purchasing - Mar-May 2022: Production planning with cost advantages - Aug-Oct 2022: Harvest outcomes reflect cost advantages - Nov 2022-Apr 2023: Market supply effects from superior production

Experiment Runs

Variant A: 6-Month Lag Transmission (Planting Season Effect)

Status: Pending - Model: RandomForest with 24-week lag fertilizer advantage signals - Features: REAL German/French fertilizer indices at 6-month lags, planting season interactions, regional cost advantages - Horizons: 30-day, 60-day - Mechanism: Planting season cost lock-in → harvest supply effects (6 months) - Expected: 12-18% improvement over strongest baseline - CRITICAL: MUST use REAL Destatis/INSEE/Eurostat data only

Variant B: 12-Month Lag Transmission (Full Production Cycle)

Status: Pending - Model: GradientBoosting with 52-week full production cycle lag - Features: REAL fertilizer cost advantages at 12-month lags, competitive position changes, carry-over effects - Horizons: 30-day, 60-day - Mechanism: Complete production cycle - fertilizer → planting → harvest → market supply (12 months) - Expected: 15-22% improvement (highest due to full cycle capture) - CRITICAL: ALL data from REAL government APIs only

Variant C: Variable Lag Transmission (Competitive Response Timing)

Status: Pending - Model: XGBoost with adaptive lag structure (16-72 weeks, learned) - Features: REAL fertilizer differential matrix at multiple lags, optimal lag weights, competitive response timing - Horizons: 30-day, 60-day - Mechanism: Market-dependent transmission timing based on competitive pressures - Expected: 18-25% improvement through adaptive lag optimization - CRITICAL: Variable lag learning from REAL data patterns only

Statistical Tests

• Diebold-Mariano test with Harvey-Leybourne-Newbold correction
• TOST equivalence test with SESOI = 15% improvement
• Granger causality test for lag validation (6/12/variable month lags)
• Bai-Perron structural break test for crisis period regime changes
• FDR correction for multiple comparisons
• ALL 4 standard baselines (persistent, seasonal_naive, ar2, historical_mean) included

Key Innovation: LAG vs IMMEDIATE TRANSMISSION

Failed Approach (FAMILY_REGIONAL_INPUT_DIVERGENCE): - Tested immediate fertilizer cost → price transmission - Despite massive crisis (216.8 peak), NO immediate effects detected - STRONGLY REFUTED (-71.9% vs baseline)

LAG TRANSMISSION Innovation: - Recognizes agricultural production cycle delays (6-18 months) - Cost advantages manifest through production planning → harvest → supply - Natural experiment timing: 2022 crisis → 2022-2023 production → market effects

Verdicts

(To be updated after experiment completion)

HE Notes

• Created 2025-08-19 learning from FAMILY_REGIONAL_INPUT_DIVERGENCE failure (immediate transmission)
• Innovation: LAG TRANSMISSION approach through production cycle delays
• Natural experiment: 2022 fertilizer crisis with verified regional divergences
• All variants use ONLY REAL DATA from verified government APIs
• SESOI = 15% based on production cycle transmission theory
• Addresses fundamental timing mismatch in agricultural cost transmission

Decision Log

(To be updated after valid experiment completion)