Industry Loss Warranty (ILW) Index Trigger Calculation & Estimation Engine
Industry Loss Warranties (ILWs) pay out when total industry losses exceed pre-defined index thresholds. Build an XGBoost and Monte Carlo simulation pipeline that forecasts industry catastrophe loss totals and calculates ILW trigger probability during active storm seasons.
What you are building
Develop an XGBoost and Monte Carlo prediction pipeline to estimate industry catastrophe loss triggers for ILW contracts.
How work is evaluated
Evaluates accuracy of final industry loss estimations and ILW trigger probability calibration.
Datasets
Public datasets, plus the private datasets you have access to.
How submissions are scored
prob_validity
Verifies probability value is between 0 and 1
This dimension contributes its full weight only when the submission satisfies the requirement. Partial credit is not awarded.
loss_estimate_mape
Mean Absolute Percentage Error of industry loss prediction • target: 0.12 • range: 0-0.5
This dimension contributes its full weight only when the submission satisfies the requirement. Partial credit is not awarded.
Learning goals
Train XGBoost regressors on earlyPCS storm reporting indices
Execute Monte Carlo simulations for industry aggregate losses across Atlantic hurricane seasons
Determine contract payout probability for specific ILW trigger bounds (e.g. $50B PCS index)
Evaluate forecast uncertainty and confidence intervals
Resources
Historical catastrophe storm bulletins and final industry loss reporting index dataset.
Challenge setup
Evaluates accuracy of final industry loss estimations and ILW trigger probability calibration.
Challenge input
JSON containing storm_central_pressure_mb, landfall_state, and initial_pcs_bulletin.
XGBoost
Loss development regression modeling
Evaluated output
JSON with predicted_final_industry_loss, trigger_50b_probability, and recommended_hedge.
- Verifies probability value is between 0 and 1
- Mean Absolute Percentage Error of industry loss prediction • target: 0.12 • range: 0-0.5
- Benchmark: ILWForecastBench
- Loss Estimate Mape target: 0.12
- 1 public reference case
- Python execution harness
- Python sandbox (unavailable on Versalist)
Shows the first task, up to three tools, and up to three evidence items.
View technical recipe
Configured tools
- llama.cpp · Optional
- Replicate · Optional
- Mistral Saba · Required
Evaluation contract
- prob_validity · Weight 1
- loss_estimate_mape · Weight 1
Metrics
- loss_estimate_mape · Target 0.12
Benchmarks
- ILWForecastBench
- general
Reference cases
- 1 public reference case
Workflow tasks
- Task 1
Input: JSON containing storm_central_pressure_mb, landfall_state, and initial_pcs_bulletin.
Output: JSON with predicted_final_industry_loss, trigger_50b_probability, and recommended_hedge.
Execution harness: Python
Evaluation tools
- XGBoost · Machine Learning Framework
Recipe freeze marker
No recipe freeze time is recorded.