Casualty Actuarial Society Special Interest Seminar on Predictive Modeling October 5, 2006 The Canadian Loss Experience Automobile Rating Arthur Tabachneck, Ph.D., Manager Arthur Tabachneck, Ph.D., Manager Abdul Sattar Al-Khalidi, Ph.D., Senior Statistician Abdul Sattar Al-Khalidi, Ph.D., Senior Statistician Statistical Research and Development Statistical Research and Development CLEAR
14
Embed
Casualty Actuarial Society Special Interest Seminar on Predictive Modeling October 5, 2006 The Canadian Loss Experience Automobile Rating Arthur Tabachneck,
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
• Provide a forum for member companies to achieve the P&C insurance industry’s common vision
Casualty Actuarial Society
Special Interest Seminar on Predictive Modeling
October 5, 2006
Government Insurers
Non-Government Insurers
Government vs Non-Government Auto Insurance in Canada
BI Only
Man
dato
ry C
over
ages
Casualty Actuarial Society
Special Interest Seminar on Predictive Modeling
October 5, 2006
0.25 mil vehicles
0.076 mil vehicles
0.53 mil vehicles
0.45 mil vehicles
4.2 mil vehicles
6.8 mil vehicles
0.62 mil vehicles0.66 mil vehicles
2.2 mil vehicles
2.2 mil vehicles
0.024 mil vehicles
Total Private Passenger Vehicles: 18,123,885
0.003 mil vehicles
0.021 mil vehicles
Casualty Actuarial Society
Special Interest Seminar on Predictive Modeling
October 5, 2006
IBC’s Department of Statistical Research and Development
• Manage databases of the entire country’s auto premium and claims information
• Build and maintain databases of VIN, VIN groupings and vehicle characteristics
• Assist IBC’s investigative efforts’ with all needed analytical services
• Conduct analytical research to help identify any emerging trends
• Develop and apply statistical models to normalize data and estimate anticipated normalized claim data
• Provide advisory make/model/model year-specific Collision, Comprehensive, Property Damage and Accident Benefit ratings for all Canadian government and non-government insurers
Casualty Actuarial Society
Special Interest Seminar on Predictive Modeling
October 5, 2006
Canadian Loss Experience Automobile Rating (CLEAR)
•What
•Why
•How
Casualty Actuarial Society
Special Interest Seminar on Predictive Modeling
October 5, 2006
What it is
CLEAR
automobile insurance rating
developed in 1989 by a working group of actuaries, CIPs, IT professionals, statisticians and underwriters
equitable and defensible
based on each make/model’s pure vehicle experience
risk ≈ rate groups (symbols)
symbols are convertible to differentials (rating factors)
Lower risks=Lower rates
Casualty Actuarial Society
Special Interest Seminar on Predictive Modeling
October 5, 2006
CLEAR
Why it was developed
most claims result in repair not replacement
price is only one of many possible predictive factors
cost saving features shouldn’t increase insurance premiums
crash test results may not reflect overall experience
a vehicle’s experience may change over time
because there are over 200 insurers, and only 18 million vehicles, Canadian insurers have to share their data to enable credible predictions
Casualty Actuarial Society
Special Interest Seminar on Predictive Modeling
October 5, 2006
Low relationship between price and theft claim cost
CLEAR
Why it was developed
Honda CR-V 4 Dr AWD MSRP $31,529
Relative Loss Cost = 126
Subaru Forester 2.5 X Wagon AWD
MSRP $30,543
Relative Loss Cost = 388
Casualty Actuarial Society
Special Interest Seminar on Predictive Modeling
October 5, 2006
Low relationship between price and collision claim cost
CLEAR
Why it was developed
Subaru Impreza WRX 4 Dr AWD
MSRP $39,335
Volkswagen Passat GLS V6 4 Dr
MSRP $36,113
Relative Loss Cost = 295
Relative Loss Cost = 51
Casualty Actuarial Society
Special Interest Seminar on Predictive Modeling
October 5, 2006
CLEAR - How it works•Body style
•Drivetrain
•Wheelbase
•Weight
•Engine displacement
•Engine horsepower
•MSRP
•Indexed MSRP
•Type of brakes
•Theft deterrent system
•Track width
•Height
•Types of airbags
•Manufacturer
•Seating capacity
•Brake assistance
•Ground clearance
•Traction control
•Stability control
•Types of headrestraints
•Seatbelt pretensioners
•Lane departure warning
•Tracking system
•Parts marking
•Engine type
•Engine placement
•Age
•General model and model
Casualty Actuarial Society
Special Interest Seminar on Predictive Modeling
October 5, 2006
Ad ju st RLCs fo r Risk Lo ad ing an dco ntro l chan ge w ith p r io r RLC
Ad jRLC=(ip r ice >=65k)*[EstRLC((ip r ice -45k)/i5k)*10]
CLEAR - How it works
Balan ce Tab le
Ad ju st RLCsto ach ie v erate le v e ln e u trality
Acco mplishRe v e rsal Co n tro l
Assu re re v e rsalsare ju stifie d
Co n v e rt Ad jRLCs to Rate G ro u p s
Rate Gro u p =1*(Ad jRLC<34.5)+(Ad jRLC/10-1.95)*(34.5<=Ad jRLC<=304)+(Ad jRLC/20+13.275)*(Ad jRLC>304)
Calcu lateLC & Re l LC
(Ad jEstF *Ad jEstS)/
W t Av g LC
Ap prov alPro ce ss
Ens ure RateLe ve l Ne utralityand Acce ptable
Dis location
Adjust e stimate s tore fle ct actual e xpe rie nce
Ad jESTF=ESTF(1+M AFF)Ad jESTS=ESTS(1+M AFS)
Est Cla im s=Actua l # Cla im s le ss e ffe ctsdue to ta riffs &/or discounts
Est Loss=Actua l Loss le ss e ffe cts dueto due to ta riffs &/or discounts
Build /Incorpo rateData Normalization M od e ls
Pro je ct Pu b lica tio n Ye a r F le e t
Use link m ode l ba se d on thre em ost re ce nt a ccide nt ye a rs'
e x posure s (a s a t De ce m be r 31stof e a ch ye a r), curre nt ye a r
e x posure s (a s a t June 30th), a ndsa le s e stim a te s
Assu re re aso n ab ility o f in su ran ced ata (fo r a ll co v e rag e s)
Sta t P la ns V IN De codingError Che cking Re a sona bility Che ck
# Ex posure s P re m ium s # Cla im s Loss
De ve lop a nd m a inta inve hicle cha ra cte ristics
W he e lba se : 2718 TDS: NoW e ight: 1889 S tyle : S UVDrive tra in: 4 ABS: Ye sPrice : $43,356 Doors: 4VCODE: 6706 Ye a r: 2007Airba g: Ye s Pow e r: 190
De v e lo p /Re v ie w /Ap p lyStatistical M o d e ls
De v e lo p an d ap p ly fo rmu laeto e stimate n o rmalize d claim
fre q u e n cy an d se v e rityfro m v e h icle ch aracte r istics