Considerations in Estimating Loss Cost Trends Kurt Dickmann & Jim Merz 2001 CAS Seminar on Ratemaking - Las Vegas, Nevada Overview of the Paper Importance of Trends in Ratemaking Ordinary Least Squares Regression Characteristics of Insured Losses Identifying Outliers Adjusting for Outliers Comparison of Results Simulation Results 2001 CAS Seminar on Ratemaking - Las Vegas, Nevada Importance of Trends in Ratemaking High Leverage on Estimated Losses Requires Judgment Scarcity of Actuarial Literature on Estimation 2001 CAS Seminar on Ratemaking - Las Vegas, Nevada Regression Understood by Management Commonly Used Widely Understood Accepted by Regulators Included in Spreadsheet Software 2001 CAS Seminar on Ratemaking - Las Vegas, Nevada Regression Basics General Form Yi f ( X i , i ) i Linear Yi 0 1 X i i Exponential Yi 0 1e 2 X i Exponential to Linear Transformation Ln(Yi ) 0 1 X 2001 CAS Seminar on Ratemaking - Las Vegas, Nevada Properties of Least Squares Estimators Unbiased Efficient Maximum Likelihood Estimators when error term is normal IID 2001 CAS Seminar on Ratemaking - Las Vegas, Nevada Regression Statistics R2 – Coefficient of Determination “Explained Variance” Durbin - Watson Serial Correlation 2001 CAS Seminar on Ratemaking - Las Vegas, Nevada Insured Loss Data Trend Seasonal Cyclical Random 2001 CAS Seminar on Ratemaking - Las Vegas, Nevada Insured Loss Data High Frequency Loss Events High Severity Losses Positively Skewed Error 2001 CAS Seminar on Ratemaking - Las Vegas, Nevada Insured Loss Data Seasonal Nature of Losses New York Collision Qtrly Paid Frequency 11.0 10.0 9.0 8.0 7.0 Year/Qtr 2001 CAS Seminar on Ratemaking - Las Vegas, Nevada 99/1 98/1 97/1 96/1 95/1 94/1 6.0 Treatment of Seasonality 12-Month-Moving Data (creates serial correlation) Residual Plot 0.06 0.00 -0.06 Seasonal Regression Model Ln(Yi ) 0 1 X i 2 D2 3 D3 4 D4 t 2001 CAS Seminar on Ratemaking - Las Vegas, Nevada Kurt’s Favorite Slide Simulation Results Underlying Trend = 0% 1.00 R-Squared 0.80 0.60 0.40 0.20 0.00 -3.5% -2.5% -1.5% -0.5% 0.5% 1.5% Estimated Trend 2001 CAS Seminar on Ratemaking - Las Vegas, Nevada 2.5% 3.5% Comments on R2 “by itself, is a poor measure of goodnessof-fit” – Lee Barclay Low Trend implies Low R2 High Error Variance implies Low R2 2001 CAS Seminar on Ratemaking - Las Vegas, Nevada Identifying Outliers Visual Method Studentized Residual Studentized Deleted Residual DFITS Cooks D 2001 CAS Seminar on Ratemaking - Las Vegas, Nevada Visual Method Oregon Homeowners 4QE Paid Frequency 13.0 11.0 9.0 7.0 Year/Qtr 2001 CAS Seminar on Ratemaking - Las Vegas, Nevada 99/1 98/1 97/1 96/1 95/1 94/1 5.0 QPV for Seasonality Indicator Variables for each “Season” Ln(Yi ) 0 1 X i 2 D2 3 D3 4 D4 t Single Trend Parameter Parameters for Seasonal Adjustments 2001 CAS Seminar on Ratemaking - Las Vegas, Nevada QPV for Seasonality First Quarter Yt [e 0 ] e 1 X t Second Quarter Yt e 2 [e 0 ] e 1 X t Third Quarter Yt e 3 [e 0 ] e 1 X t Fourth Quarter Yt e 4 [e 0 ] e 1 X t 2001 CAS Seminar on Ratemaking - Las Vegas, Nevada QPV verses Multiple Regression QPV Single Trend Multiple Regression Multiple Trends 2001 CAS Seminar on Ratemaking - Las Vegas, Nevada Comparison of Results Results using five year data Oregon Homeowner Frequency New York PPA Collision Frequency Nevada PPA Bodily Injury Severity 12 MM Trend -6.9% R2 .17 Trend -1.9% R2 .58 Trend 3.1% R2 .78 Quarterly -3.9% .03 -1.7% .10 2.7% .25 Annual -10.1% .34 -1.2% .37 3.7% .85 Manual Adjustment -2.6% .20 -0.8% .84 1.4% .41 Indicator Variables -2.6% .27 -1.2% .74 2.7% .27 2001 CAS Seminar on Ratemaking - Las Vegas, Nevada Simulation – Personal Auto BI Severity Trend Quarterly Seasonal Adjustment Random Error (Normally Distributed) Shock Variable Ln(Yi ) [ 0 1 X i 2 D2 3 D3 4 D4 ] Wˆi ˆ 2001 CAS Seminar on Ratemaking - Las Vegas, Nevada Simulation in Excel Using RAND, NORMSDIST, LN, EXP Macro Loop to generate simulated data remove outliers calculate regressions record results 2001 CAS Seminar on Ratemaking - Las Vegas, Nevada Simulation Results Shocked Early All Methods except Manual Adjustment underestimate the trend Shocked Late All Methods except Manual Adjustment overestimate the trend 2001 CAS Seminar on Ratemaking - Las Vegas, Nevada Conclusion Familiar Ideas Additional Tools Areas for Further Research 2001 CAS Seminar on Ratemaking - Las Vegas, Nevada