Predictive Modeling In Property Casualty Insurance
Gilbert Korthals
SVP, Chief Analytics Officer
GuideOne Insurance
Property Casualty Insurance
Personal Lines
• Personal Auto
• Homeowners
Commercial Lines
• Property
• General & Professional Liability
• Workers Compensation
• Commercial Auto
Specialty Lines
• Directors and Officers
• Employers Practice Liability
• Terrorism
• Warranty
• Medical Malpractice
• Crop
Not all lines listed
3
Distribution of Direct Premiums Written by Segment/Line, 2014
Sources: A.M. Best; Insurance Information Institute research.
� Personal/Commercial lines split has been about 50/50 for many years
� Pvt. Passenger Auto is by far the largest line of insurance and is currently the most important source of industry profits
� Billions of additional dollars in homeowners insurance premiums are written by state-run residual market plans
Distribution Facts
Commercial Lines$282.5B/51%
2014
Pvt. Pass Auto$190.3B/34%
Homeowners$86.1B/15%
Predictive analytics is a set of Business Intelligence (BI) technologies that uncover relationships and patterns within large volumes of data that can be used to predict behavior and events. But unlike other BI technologies, predictive analytics is forward-looking, using past events to anticipate the future
www.aicpcu.org
What Do We Mean By Predictive Modeling?
Use of predictive modeling for risk selection and pricing
Prevalence in P-C Insurance
Source: Towers Watson 2014 P&C Predictive Modeling Survey, 52 participating companies represent 17% of US personal lines carriers
and 22% of commercial lines carriers.
Prevalence in P-C Insurance
Source: Willis Towers Watson 2015 P&C Predictive Modeling and Big Data Survey, participating companies represent 11% of US
personal lines carriers and 17% of commercial lines carriers.
Applications in P-C Insurance
Pricing
• Rating plans
• Tiering
• Credit scoring
• Vehicle classification
• Territory analysis
• Telematics
• Price optimization
Underwriting
• Tier scoring
• Straight through processing
• Loss control inspections
• Property inspection
• MVR Ordering
• Premium audit
Claims
• Fraud detection
• Adjuster assignment
• Severity flag
• Litigation propensity
• Reserving
Sales and Marketing
• Retention analysis
• Campaign effectiveness
• Agent tiering
• Purchasing patterns
• Quote conversion
Variables in Personal Lines Pricing
∗ Credit history∗ Telematics∗ Advance purchasing∗ Prior BI limits or pay
plan
Variables that indicate
behavioral information
Behavioral Variables
Being Researched
∗ Purchase patterns
∗ Digital exhaust
∗ Nest devices
Other Variables Used
∗ Level of education, GPA
∗ Occupation
∗ Age, Gender, Marital status
∗ Household structure
∗ Distance from agent
∗ Crime Index of neighborhood
∗ # of mortgagees
∗ Smoker or dog present
∗ # of occupants
∗ Green home discount
Power of new variables teased out through predictive modeling.
Price GrowthBasic Analysis
Advanced
Judgment• No Testing
Scenario Testing• Model results • Metrics• Goals
OptimizationUse simulation to find changes that reach goals
∗ An example of Prescriptive Analytics (How can we make it happen?)
∗ May maximize insurer retention, profitability, premium, market share or combinations of such
∗ Create a link model that combines several other models such as Loss Cost Model, Retention Model, New Business Model, etc.
∗ Link model may incorporate the concept of price elasticity of demand, competitor rates into pricing factors or life time value considerations
∗ No single definition for P-C Insurance
Price Optimization
Price Optimization: What Is It?
11
5.9 Cents Per Ounce ($23.76 for 24 Bottles)
1.4 Cents Per Ounce ($5.49 for
24 Bottles)
320% Price Difference! Does It Cost $18.25 to Unpack the Bottles and Keep Them Cold?
U.S. Insurers Don’t Do This!!!
∗ Turns out to be highly controversial in insurance
∗ Prohibited in 17+ states, accepted in others
∗ Charges of unfair discrimination in auto insurance∗ Alleged low income drivers don’t price shop as much
∗ Does it violate any standards of practice?Actuarial Statement of Principles on Ratemaking
“A rate is reasonable and not excessive, inadequate, or unfairly discriminatory if it is an actuarially sound estimate of the expected value of all future costs associated with an individual risk transfer.”
Price Optimization
For more information on this topic, see NAAPA.org (National Association of Professional Allstate Agents) website and search for Price Optimization
∗ GLM (Generalized linear model) – Traditional predictive modeling technique for P-C insurance industry
∗ Decision trees
∗ Clustering
∗ PCA (Principal component analysis)
∗ Ensemble techniques such as random forests and gradient boosting machines
∗ Machine learning – neural networks, MARS (multiple adaptive regression splines) & SVM (support vector machines)
Techniques
∗ Special purpose GLM software such as Emblem is commonly used and is a typical entry level platform
∗ Research programming languages such as R, Python, Matlab & SAS are becoming more important as advanced techniques are more commonly used
∗ SQL is commonly used for database processing
∗ Big data tools such as Hadoop are becoming more important
Software
Big Data Applications
Source: Willis Towers Watson 2015 P&C Predictive Modeling and Big Data Survey, participating companies represent 11% of US personal lines
carriers and 17% of commercial lines carriers.
16
The Internet of Things and the Insurance Industry
� The “Internet of Things” will create trillions in economic value throughout the global economy by 2025
� What opportunities, challenges will this create for insurers?
� What are the impact on the insurance industry “value chain”?Sources: McKinsey Global Institute, The Internet of Things: Mapping the Value Beyond the Hype,
June 2015; Insurance Information Institute.