Actuarial Property Casualty Data Thoughts¶
Raw Data¶
- Claims Lossrun
- Latest Exposures at consistent granularity
- Latest Policy Data and Details
- Manual Adjustments, Mapping, and Reference Tables
Raw Data Ingestion¶
- Validation
- Reconciliation
- Reasonability Checks
- High Level Comparison to Prior (i.e. dropped claims, missing exposures, etc.)
- Cataloging and Metadata
- Data Dictionary
Summary Data¶
- Loss data summarized by policy/accident year for paid, reported, counts, etc. with filtering based on any dimensions such as segment, limit, etc.
- Exposure data summarized by policy/accident year and aggregated to the necessary granularity for matching to loss data
- Indicated Rates:
- Frequency = Claim Counts / Exposures
- Severity = Losses / Counts
- Loss Cost = Losses / Exposures
- Indicated Trends
- Development and Diagnostic Triangles
Analysis Specific Data¶
- Selected Rates, Trends, and Loss Development Factors
- Selected Ultimates and methods utilized
- Manual Adjustments and Assumptions related to data manipulations
- Notes and Comments
- Explanation of anomalies
Flow¶
- Raw Data -> Working Data + Prior, Reference, and Industry Data
- Summary Tables
- Triangles
- Selection of Development Factors
- Interpolation and Extrapolation of Factors (i.e. Persistencies, decay factors, ILFs, interpolation methods, seasonality, priors, industry, weighted averages)
- Derivation of Discount factors based off paid loss development pattern
- Derivation of Actual vs. Expected comparing to Prior and using prior’s interpolated Development Factors (with prior interpolation logic/persistencies)
- Note: Common to adjust latest periods for updates to exposures
- Initial Development Ultimates
- Selection of Rates, Trends, Experience Factors, Credibility Weighting Factors, etc.
- Selected Ultimates
- Derivation of Unpaid Reserves (discounted and undiscounted)
- Simulation and confidence intervals for unpaid reserves
- Allocation of unpaid reserves/future premiums to entities
Related¶
Backlinks:
list from [[Actuarial Property Casualty Data Thoughts]] AND -"Changelog"