Casualty ILS Reserve Trajectory Forecasting with LangChain and LangGraph
Casualty Insurance-Linked Securities involve long-tail liability dynamics requiring stateful trajectory modeling. You will build a stateful trajectory forecasting workflow using LangChain and LangGraph to estimate ultimate loss development and reserve adequacy. The workflow must predict 10-year loss development trajectories with a Root Mean Squared Error (RMSE) below 0.05.
What you are building
The core problem, expected build, and operating context for this challenge.
Build a stateful casualty loss trajectory development agent using LangChain and LangGraph.
How work is evaluated
Evaluates RMSE of predicted loss development factors against actuarial ground truth curves.
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.
ten_year_curve_length
Validates predicted_factors contains exactly 10 elements.
This dimension contributes its full weight only when the submission satisfies the requirement. Partial credit is not awarded.
trajectory_rmse
Root Mean Squared Error of predicted factor curve • target: 0.05 • range: 0-0.2
This dimension contributes its full weight only when the submission satisfies the requirement. Partial credit is not awarded.
What you should walk away with
Construct multi-step stateful graph nodes using LangGraph for loss development cycles.
Integrate LangChain tools for actuarial chain-ladder model calculations.
Predict long-tail casualty claim escalation probabilities over multi-year horizons.
Reference links and supporting material
Dataset of 200 historical casualty loss development triangles spanning 10 development years with ground-truth ultimate loss outcomes.
Challenge setup
Evaluates RMSE of predicted loss development factors against actuarial ground truth curves.
Challenge input
JSON object with portfolio_id, historical_development_years (array), initial_paid_losses
LangChain
Assigned framework for chain and tool orchestration.
LangGraph
Required extension for stateful node-based graph execution.
Evaluated output
JSON object with predicted_ultimate_loss, predicted_factors (array of 10 floats), reserve_adequacy_ratio
- Validates predicted_factors contains exactly 10 elements.
- Root Mean Squared Error of predicted factor curve • target: 0.05 • range: 0-0.2
- Benchmark: Actuarial Time-Series Trajectory Benchmark
- Trajectory Rmse target: 0.05
- 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
- LangChain · Required
- Langchain · Optional
- LocalAI · Optional
- LangChain · Required
- Langchain · Optional
Evaluation contract
- ten_year_curve_length · Weight 1
- trajectory_rmse · Weight 1
Metrics
- trajectory_rmse · Target 0.05
Benchmarks
- Actuarial Time-Series Trajectory Benchmark
- general
Reference cases
- 1 public reference case
Workflow tasks
- Task 1
Input: JSON object with portfolio_id, historical_development_years (array), initial_paid_losses
Output: JSON object with predicted_ultimate_loss, predicted_factors (array of 10 floats), reserve_adequacy_ratio
Execution harness: Python
Evaluation tools
- LangChain · Agent Framework
- LangGraph · Stateful Workflow Orchestrator
Recipe freeze marker
No recipe freeze time is recorded.