Demo — Waterbridge Challenge

Methodological Approach

This project takes a rigorous, multi-model approach to probabilistic forecasting:

Data Acquisition: Collect 20+ years of global trade data, macro indicators, policy records, and bilateral relationships
Feature Engineering: Construct domain-specific indicators (tariff levels, supply chain metrics, tech adoption rates)
Model Selection: For each forecast, fit the appropriate econometric model (ARIMA, VAR, logistic regression, survival analysis, etc.)
Cross-Validation: K-fold validation, structural break tests, sensitivity analysis
Probabilistic Calibration: Convert model outputs to calibrated probability forecasts with 95% confidence intervals
Bayesian Integration: Combine forecast probabilities using Bayesian network to capture interdependencies

Forecast Categories (F1-F25)

Trade Restructuring (F1-F7): 70% Baseline Probability

F1: Tariff Escalation MFN tariffs rise ≥2pp above 2022 baseline	`70%`
F2: Supply Chain Diversification China's US import share falls below 12%	`75%`
F3: Vietnam Trade Surge US-Vietnam bilateral trade volume doubles	`72%`
F4: Services/Goods Divergence Trade composition shifts in developed economies	`60%`
F5: Tech Standards Bifurcation Distinct US- vs China-led standards emerge	`85%`
F6: Carbon Tariff Adoption ≥7 G-20 economies implement by 2029	`64%`
F7: USD Reserve Resilience USD maintains >55.5% of global reserves	`66%`

Institutional Changes (F8-F20): 65% Baseline Probability

Forecasts covering China trade restrictions, regional bloc formation, technology transfer controls, digital trade barriers, cross-border investment, and institutional coordination.

Range: F8 (90%) to F25 (27%) representing policy likelihood and systemic tipping points.

Systemic Risks (F21-F25): 55% Baseline Probability

Macroeconomic tail risks including global recession, currency volatility, commodity shocks, financial market integration, and central bank coordination failures.

Econometric Models Applied

ARIMA (F1)

Time series forecasting of tariff escalation using historical MFN tariff data.

Linear Regression (F2-F4)

Supply chain metrics on bilateral trade data with economic controls.

Binary Classification (F5)

Logistic regression on technology adoption and standards formation data.

VAR Models (F7, F22-F24)

Multivariate time series for reserve composition, currency volatility, market integration.

GARCH Models (F22)

Currency volatility forecasting with heteroskedastic variance.

Bayesian Networks

Interdependency analysis: tariff policy → supply chain changes → tech bifurcation.

Validation & Robustness

All models undergo rigorous validation to ensure calibrated, reproducible forecasts:

Cross-Validation

K-Fold Validation: 5-fold CV on training data, hold-out test period
Out-of-Sample Testing: 73% accuracy on held-out validation data
Structural Break Tests: Chow test, CUSUM for regime changes

Confidence Bounds

95% Intervals: All probability estimates include confidence bounds
Monte Carlo Simulation: 10,000 iterations for uncertainty propagation
Sensitivity Analysis: Key parameters tested across ±20% variation

Performance Metrics

Average R² Across Models:

0.68

System Fragility Index:

0.39 (LOW)

Bayesian Interdependency Analysis

The 25 forecasts are not independent. Tariff escalation affects supply chain decisions, which influence tech standards bifurcation, which impacts investment flows.

Network Properties

Interdependency Strength: 0.73 (HIGH) – Strong correlations between forecast groups
Nodes: 25 forecast scenarios
Edges: Causal relationships (trade policy → supply chain → macro effects)
Monte Carlo Paths: 10,000 simulations of correlated outcomes

This enables tail-risk analysis: joint probability of multiple adverse scenarios occurring together.

Competition Deliverable

The project submission includes a comprehensive PDF report with:

Exact 25 binary probability forecasts (submitted August 1-5, 2025)
Detailed methodology for each forecast
Data provenance and source documentation
Cross-validation results and confidence intervals
Bayesian network interdependency analysis
Sensitivity analysis results
Fully reproducible Python analysis code

Repository: github.com/Minoneshan/Waterbridge_MM
Submission: August 1-5, 2025 (LaTeX formatting update August 5)

Reproducing the Analysis

All results are fully reproducible using the Python analysis pipeline:

# Activate environment
conda activate waterbridge_mm

# Run complete analysis
python code/analysis.py

# Run statistical validation
python code/statistical_tests.py

# Generate sensitivity analysis
python code/generate_sensitivity.py

# Compile LaTeX report
make pdf

Ready to Dive Deeper?

Read the full technical documentation for complete methodology, data sources, and validation details.

Read Full Documentation →

Methodology & Key Results