Renewable Energy Infrastructure Underwriting Pricing Engine with CatBoost
China's massive clean energy deployment has created tens of thousands of solar farms and wind turbines operating in extreme weather zones. Renewable energy asset underwriters must accurately price risk premiums for wind farm physical damage and solar hail damage under shifting El Niño climate probabilities. Mispricing leads to either massive underwriting losses or uncompetitive quotes. In this challenge, you will implement a risk pricing engine using CatBoost. Your model will handle categorical asset attributes (turbine manufacturer, panel mounting specs, terrain type) alongside continuous climate risk variables without manual target encoding. Your solution will execute Tweedie loss regression to model pure premium, calculate risk-adjusted insurance pricing, and evaluate model calibration across asset categories.
What you are building
The core problem, expected build, and operating context for this challenge.
Build an actuarial risk pricing model using CatBoost to price insurance premiums for solar and wind assets under extreme weather shifts.
How work is evaluated
Evaluates Tweedie deviance loss and Gini coefficient on test asset portfolio.
Challenge datasets
Public datasets, plus the private datasets you have access to.
How submissions are scored
These dimensions define what the evaluator checks and which criteria separate a passable run from a strong one.
tweedie_loss_validation
Model converges cleanly under Tweedie loss function
This dimension contributes its full weight only when the submission satisfies the requirement. Partial credit is not awarded.
gini_coefficient
Gini metric measuring risk segmentation capacity • target: 0.42 • range: 0-1
This dimension contributes its full weight only when the submission satisfies the requirement. Partial credit is not awarded.
What you should walk away with
Train CatBoost Regressor models using Tweedie objective for zero-inflated claim loss distributions.
Incorporate native categorical feature handling for asset hardware specs and locations.
Calculate pure premium and technical risk price for renewable energy assets.
Assess pricing model calibration using Gini coefficient and loss ratio Lorenz curves.
Reference links and supporting material
Synthetic dataset of commercial solar and wind asset hardware specs, local climate hazard indices, and historical claim losses.
Challenge setup
Evaluates Tweedie deviance loss and Gini coefficient on test asset portfolio.
Challenge input
JSON with asset_type, sum_insured_usd, hail_index, max_wind_gust_ms, mounting_type
CatBoost
Leading library for categorical features and Tweedie loss regression.
Evaluated output
JSON with pure_premium_usd, technical_rate_pct, risk_tier
- Model converges cleanly under Tweedie loss function
- Gini metric measuring risk segmentation capacity • target: 0.42 • range: 0-1
- Benchmark: CleanEnergyPricingBench
- Gini Coefficient target: 0.42
- 1 public reference case
- Python execution harness
- Python sandbox (unavailable on Versalist)
Summary shows the first task, up to three tools, and up to three evidence items. Open the technical recipe to view all configured items.
View technical recipe
Configured tools
- LMDeploy · Required
- Google Vertex AI · Optional
- Gemini 2.5 Flash · Optional
Evaluation contract
- tweedie_loss_validation · Weight 1
- gini_coefficient · Weight 1
Metrics
- gini_coefficient · Target 0.42
Benchmarks
- CleanEnergyPricingBench
- energy
Reference cases
- 1 public reference case
Workflow tasks
- Task 1
Input: JSON with asset_type, sum_insured_usd, hail_index, max_wind_gust_ms, mounting_type
Output: JSON with pure_premium_usd, technical_rate_pct, risk_tier
Execution harness: Python
Evaluation tools
- CatBoost · Tabular ML Model
Recipe freeze marker
No recipe freeze time is recorded.