RISK MARGINS FOR HEALTH INSURANCE. Adam Searle and Dimity Wall

Transcription

1 RISK MARGINS FOR HEALTH INSURANCE Adam Searle and Dimity Wall

2 Structure of presentation Why this paper? The nature of health insurance Our approach Benchmarks LAT Limitations 2

3 AASB1023 Why this paper? Potential future listings of health insurers IFRS and Solvency II 75% Probability of Adequacy benchmark 3

4 Nature of health insurance Very short tail No case estimates Service / Treatment date rather than Occurrence / Accident date Benefit types Predictability of ultimate benefits Reserving typically based on Paid Chain Ladder with reasonableness adjustment 4

5 Delay to settlement 120% 100% Proportion paid 80% 60% 40% 20% 0% Development month 5

6 Volatility in the claims payment pattern 120% Percentage of ultimate cost settled 100% 80% 60% 40% 20% 0% Fast Average Slow Extremely slow Development month 6

7 Stability of the ultimate liability Service m onth Paid to date Selected reserve Exponential trend Seasonality adjusted Burning cost ($)

8 General insurance risk margin methods Volatility in the payment pattern results in unadjusted chain ladder giving volatile results Mack, Stochastic Chain Ladder and Bootstrap give unreasonably high results Need to allow for the underlying predictability of ultimate benefits 8

9 Our approach to assessing risk margins Develop a mechanical reserving method Assess accuracy against actual payments Determine the risk margin using distribution free approach 9

10 Data used 19 Australian health insurers At least 4 years of data for each insurer Paid triangles (monthly / monthly) Claims data split by hospital / medical / ancillary Grouping of claim types for analysis Exposure data from PHIAC B Thanks to all the insurers who provided data 10

11 Available valuation points Valuation points Service month

12 Our approach 1. Example fund 2. For an individual insurer, project ultimate burning costs using paid chain ladder Monthly by monthly payments triangle Burning cost = ultimate claims cost exposure Development factors calculated from the insurer s own data with no smoothing 12

13 Standardise for working days Service month Work day standardised Actual data Daily burning cost ($)

14 Fit an exponential trend line Service month Work day standardised Actual data Exponential trend Daily burning cost ($)

15 5. Assess remaining seasonality Calculate average deviations for each calendar month Standardise these to ensure bias is not introduced 110% Percentage loading for seasonality 105% 100% 95% 90% 85% Service month 15

16 6. Calculate seasonality-adjusted trend line Apply loadings for working days and seasonality to the exponential trend line Burning cost ($) Service m onth Paid to date Selected reserve Exponential trend Seasonality adjusted 16

17 6. Accuracy of the seasonality-adjusted trend line 4% 3% 2% 1% 0% -1% -2% -3% -4% -5% Error percentage Service month Accuracy of trendline MAT error 17

18 Our approach (cont.) 7. Repeat all above steps at historic valuation dates 8. Determine best weighting between PCL and trend line Minimise least squares of difference between ultimate projection and actual result for each month Separately for each development month across all valuation periods 9. Assess goodness of fit Serial correlation Distribution of residuals Count of +ve / -ve strings Total +ve / -ve residuals 18

19 9. Goodness of fit Historic deviations of projected ultimate from actual Deviation from actual ultimate cost 4.0% 3.0% 2.0% 1.0% 0.0% -1.0% -2.0% -3.0% -4.0% -5.0% -6.0% Development month 19

20 Our approach (cont.) 10. Repeat all above steps for each insurer 19 insurers * 30 valuations 11. Assess the best weightings across all insurers Weighting between: PCL projection Seasonally-adjusted trend line Varies by size of central estimate reserve Logarithmic curve fit for each development month 20

21 11. Overall weightings adopted between PCL and trend line 100% Weighting to seasonally-adjusted trend 90% 80% 70% 60% 50% 40% 30% 20% 10% Central Estimate reserve $200m $100m $50m $20m $10m $5m $2m $1m BF 0% Development month 21

22 Our approach (cont.) 12. Calculate central estimate reserve: At all historic valuations For all insurers Using the standard weightings 13. Calculate the percentage errors Between projected reserve at each valuation for each insurer, and Eventual outcome of the actual reserve required 14. Assess the loadings required to give a 75% confidence level 22

23 14. Calculated OSC risk margins at 75% PoA 16% 14% 12% Risk margin loading 10% 8% 6% 4% 2% 0% Reserve ($m) 23

24 Benchmark OSC risk margins at 75% PoA Outstanding claims central estimate Benchmark risk margin Tillinghast paper - short tail < $1.5m 8% - 14% 20% - 30% $1.5m - $5m 5% - 8% 20% - 30% $5m - $50m 4% - 5.5% 8.0% % > $50m 2% - 3.5% 5.5% -13.2% 24

25 Liability Adequacy Test AASB1023 Industry interpretation Have only assessed for June balance dates Same data used as for OSC (minus 1 fund) Fewer valuation dates No benchmarks by size 25

26 Data available for premium liabilities risk margin assessment Valuation points for premium liabilities Service month

27 Percentage errors in the 9-month future projections for premium liabilities % PoA Count % to -14% -14% to -10% -10% to -6% -6% to -2% -2% to 2% 2% to 6% 6% to 10% 10% to 14% 14% to 18% 18% to 22% Percentage error in the projection 27

28 75% PoA for LAT Calculated across all funds of 2.8% This would vary significantly by size of central estimate (and other factors) Likely to be lower than the risk margin percentage for OSC Volatility due to size Trend accurate as %age of ultimate liability, but less accurate as a %age of the OSC component only 28

29 Indication of reserve proportions (for a valuation date) & prior Reserve ($m) Service month Outstanding claims Premium liabilities Risk margin (OSC) Risk margin (PL)

30 Limitations Probability of sufficiency Data limitations Not assessed by type of claim Confidentiality restrictions Using the benchmarks with other methods for assessing the central estimate LAT benchmarks need further analysis /data 30

31 Summary Ultimate liabilities are very predictable However, the monthly payment pattern is not Our approach takes account of the ultimate liability stability Risk margins are significantly lower than for shorttailed general insurance classes At a 75% PoA, our risk margins for outstanding claims vary from 2% - 14% of the central estimate by size of insurer As a percentage of the central estimate, risk margins for premium liabilities are lower than for outstanding claims 31