12.4 Internal Systemic Risk
Definition 12.3 Internal systemic risk = Uncertainty arising from the liability valuation process/actuarial valuation models
- From source anywhere along the chain of the valuation
Examples:
Data record/collection/organization (e.g. not collecting the right data; not using the right data)
Analysis, actuarial judgement
Reserve selection (e.g. management overrides actuary’s opinion)
Definition 12.4 3 components of internal systemic risk:
Specification Error:
From not perfectly modeling the insurance process because it’s too complicated or just don’t have the data
Parameter Selection Error:
Difficulty in measuring all predictors and the trend in these predictors are particularly difficult to measure
Data Error:
Lack of data, lack of knowledge of the underlying pricing, u/w, and claim management process, inadequate knowledge of portfolio
12.4.1 CoV for Internal Systemic Risk
Benchmarking technique:
Need to define a list of risk indicators, score them against best practice, map the scores to a CoV
12.4.1.1 Score Against Best Practice
- For each valuation group, assign a score (1-5) for each risk indicator
| Risk Indicators | Best Practice |
|---|---|
| # of independent models used | Each model should add value by considering a different dimension of claims experience |
| Range of results produced by models | Low variation in model results |
| # and importance of subjective adjustments to factors | Few subjective adjustments; adjustments regularly monitored and reviewed |
| Ability to detect trends in key claim cost indicators | Models have performed well at detecting trends in the past |
| Risk Indicators | Best Practice |
|---|---|
| Best predictors have been identified (but not necessarily used) | Best predictors have been analyzed and identified (int or ext), and show a strong correlation with claims experience |
| Best predictors are stable over time, or change due to process changes | Predictors stable over time, stabilize quickly and respond well to process changes |
| Value of predictors used | Predictors are close to Best Predictors; lead (rather then lag) claims cost outcomes, modeled, rather than subjectively selected |
| Risk Indicators | Best Practice |
|---|---|
| Knowledge of past processes affecting predictors | Actuary has good and credible knowledge of past processes and change to processes |
| Extent, timeliness, consistency and reliability of information from the business | Regular, pro-active communication between the actuary and claims staff and the business |
| Data is subject to reconciliations and quality control | Reconciliation against financials, and prior studies; difference are well understood |
| Frequency and severity of past misestimation due to revision of data | No past instances of data revision |
Assign weight for each risk indicator (weight can vary by valuation group)
Calculate weighted average the scores using the selected weights
We score the 3 components for each valuation group (OCL and PL) and then roll up the score and assign the average grade for each valuation group to a CoV
12.4.1.2 Calibrate Score to CoV
Significant amount of judgement supplement by quantitative analysis
CoV \(\in [5\%, 25\%]\)
Analysis of past model performance should aid in estimating the potential variability
Hindsight Analysis:
Compare valuation of liabilities at prior point in time to the current view, to gain insight into how a better model can reduce volatility
Mechanical Hindsight:
Mechanically do various ex post analysis, and see how prediction error can be reduced; e.g. do a detailed vs crude and see the difference
Additional comments:
Improvement from poor to fair is greater than from fair to good
Longer tail line will have higher CoV due to difficulty in estimating the parameters
Larger liability will have smaller CoV when all else being equal
OCL and PL might not necessarily be on the same scale
PL may have additional uncertainty as it’s for future business
For short tail lines ELR might be sufficient for PL but might not be the best practice for OCL
Can always just add a load on top if justified