FAMILY_NETWORK_EFFECT_PROPAGATION: Experiment Log

Overview

Testing network effect propagation models for Dutch potato price forecasting through market shock transmission patterns, cross-border contagion networks, and multi-layer network dynamics. This hypothesis builds on FAMILY_CROSS_MARKET_COUPLING (CONDITIONALLY SUPPORTED with 86.8% improvement) by advancing beyond simple transmission to sophisticated network propagation algorithms and FAMILY_BELGIAN_PRICE_SHOCK_TRANSMISSION (REFUTED with -24% performance) by using network topology modeling instead of simple shock transmission.

Hypothesis Origins

• FAMILY_CROSS_MARKET_COUPLING: CONDITIONALLY SUPPORTED with 86.8% improvement using cross-border transmission, validating international linkages but limited to simple correlation without network topology
• FAMILY_BELGIAN_PRICE_SHOCK_TRANSMISSION: REFUTED with -24% performance despite confirming June 2021 shock (523.0), demonstrating need for network propagation vs simple transmission
• FAMILY_CROSS_BORDER_STOCK_ARBITRAGE: Active testing of cross-border arbitrage with systematic Belgian-Dutch processing linkages (2.1M tons annually)
• FAMILY_EUROPEAN_STORAGE_CASCADE: Active testing of CASCADE patterns providing foundation for multi-layer network modeling
• Industry catalyst: June 2021 Belgian spike (523.0) revealed network propagation patterns requiring sophisticated modeling
• Academic basis: Network propagation theory, epidemiological SIR models, multi-layer network dynamics, market contagion modeling

Experiment Design

• Method: Rolling-origin cross-validation
• Initial window: 365 days (1 year minimum)
• Step size: 7 days (weekly)
• Test windows: 60 days maximum
• Baselines: MANDATORY STANDARD BASELINES - persistent, seasonal_naive, ar2, historical_mean (from get_standard_baselines())
• REAL DATA ONLY: Boerderij.nl API international prices, StockAPI processing demand, network topology from real market relationships

Data Sources (REAL DATA ONLY)

• International prices: Boerderij.nl API (NL/BE/DE/FR.157.2086/2083) - multi-market potato prices for network nodes - git:current
• Processing demand: StockAPI processing demand (NL/DE/BE) - network flow drivers - git:current
• Prices: Boerderij.nl API (NL.157.2086) - Dutch consumption potatoes - git:current
• Transport networks: Calculated from price spreads and distance models - network edge weights - git:current

Experiment Runs

Variant A: Market Shock Propagation Networks

Status: INCONCLUSIVE - INSUFFICIENT_DATA (2025-08-19) - Model: Network propagation using SIR epidemiological model adapted for price shocks - Features: be_price_shock_indicator, nl_shock_reception_signal, de_shock_propagation, fr_shock_propagation, network_centrality_nl, shock_propagation_velocity, network_resistance_index, cascade_amplification_factor, shock_decay_rate, price_lag_1w - Horizons: 1-month, 2-month - Mechanism: Market shocks propagate through network topology creating predictable transmission patterns from Belgium → Netherlands → Germany/France - Expected: >14% improvement through network shock propagation modeling

Variant B: Cross-Border Contagion Networks

Status: INCONCLUSIVE - INSUFFICIENT_DATA (2025-08-19) - Model: Multi-layer contagion network with threshold activation across processing, transport, storage layers - Features: multi_market_contagion_index, processing_layer_contagion, transport_layer_contagion, storage_layer_contagion, cross_border_activation_signals, contagion_velocity_index, network_fragmentation_signals, contagion_clustering_coefficient, network_diameter_effects, price_lag_1w - Horizons: 1-month, 2-month - Mechanism: Cross-border contagion spreads through multi-layer networks (processing, transport, storage) creating compound transmission effects - Expected: >16% improvement through multi-layer contagion modeling

Variant C: Multi-Layer Network Dynamics

Status: INCONCLUSIVE - INSUFFICIENT_DATA (2025-08-19) - Model: Temporal multi-layer network with inter-layer coupling and network evolution over time - Features: processing_network_centrality, transport_network_efficiency, storage_network_capacity, inter_layer_coupling_strength, network_evolution_velocity, layer_specific_propagation, cross_layer_amplification, network_adaptation_signals, structural_resilience_index, temporal_network_persistence, price_lag_1w - Horizons: 1-month, 2-month - Mechanism: Multi-layer network dynamics capture complex interactions between processing, transport, and storage layers creating comprehensive network propagation effects - Expected: >18% improvement through comprehensive multi-layer network modeling

Statistical Tests

• Diebold-Mariano test with Harvey-Leybourne-Newbold correction vs ALL 4 standard baselines
• TOST equivalence test with SESOI = 14% improvement (network effects show strong propagation)
• Bai-Perron structural break test for network topology vs normal periods
• FDR correction for multiple comparisons across variants
• Directional accuracy threshold = 60%

Network Analysis Framework

• Network Topology: Correlation-based adjacency matrices from REAL price data
• Node Dynamics: SIR propagation model (Susceptible-Infected-Recovered) adapted for price shocks
• Edge Weights: Transport-cost adjusted correlation + processing flow volumes
• Shock Detection: Price changes >2 standard deviations trigger network propagation
• Multi-layer Modeling: Processing + transport + storage layers with inter-layer coupling
• Temporal Evolution: Network adaptation and memory effects over 8-12 week windows

Network Validation Events

• June 2021 Belgian Spike: 523.0 vs base 100 (5.2x normal) → network propagation pattern validation
• 2022 Energy Crisis: Network stress testing during energy cost volatility
• 2024 Storage Crisis: Multi-layer network effects with storage constraints propagating through processing and transport

Expected Network Effects

• Shock Propagation: Belgian shocks propagate to Dutch markets with network-determined velocity and amplification
• Contagion Spreading: Multi-layer contagion through processing (demand flows), transport (cost thresholds), storage (capacity constraints)
• Network Evolution: Adaptive network topology based on changing market relationships and processing patterns
• Inter-layer Coupling: Processing demand drives transport optimization which affects storage release timing

Verdicts

Execution Results: 2025-08-19

Overall Verdict: INCONCLUSIVE - INSUFFICIENT_DATA

Data Successfully Loaded: - ✅ 84 weeks of REAL multi-market price data (2018-2021) - ✅ Network topology: 4 nodes (NL, BE, DE, FR) - ✅ Price range: €4.12-€32.50/100kg (REAL volatility)

Critical Issue: 84 weeks insufficient for network propagation analysis - Network effects require 200+ observations minimum - Need multiple shock events for propagation pattern detection

Data Quality: 100% REAL DATA - No synthetic data used

Recommendation: Re-run with 2015-2025 data for robust network analysis

HE Notes

• Created 2025-08-19 advancing beyond FAMILY_CROSS_MARKET_COUPLING (86.8% success) and learning from FAMILY_BELGIAN_PRICE_SHOCK_TRANSMISSION (-24% failure)
• Superior to simple transmission: uses sophisticated network propagation algorithms vs basic cross-market correlation
• SESOI set to 14% due to strong network propagation literature evidence
• All variants use ONLY REAL DATA from verified Boerderij.nl API and StockAPI interfaces
• Focus on 30/60-day horizons where network propagation effects strongest
• Validates using June 2021 Belgian spike, 2022 energy crisis, 2024 storage crisis natural experiments

Decision Log

(To be updated after experiment completion)