Predictive Modeling for Commercial Risks

1

• Predictive Modeling for Commercial Risks
• Cheng-Sheng Peter Wu, FCAS, ASA, MAAA
• CAS 2005 Special Interest Seminar
• Chicago
• September 19-20, 2005

2
Why Now for Commercial Lines PM?

• Fast technology development
• Fast computation and data processing capability
• Powerful statistical modeling tools
• Availability of wide range of internal and
  external information
• PM applications very successful for personal
  lines. So if personal lines can, why not
  commercial lines?

3
Why Now for Commercial Lines PM?

• Intensified competition for commercial lines
• Hard market/good profit over the past years
• Competition for growth intensified
• Grow profitably the mainstream small and mid
  size business
• Better segmentation how to go beyond class base
  underwriting?
• Lower expense ratio more automation, less touch,
  and less paper work for UW process
• Ready for the next cycle?
• PM provides a tool for better segmentation of
  risks and lower expense ratios

4
Predictive Modeling for Commercial Risks

• Lessons from PM for personal lines
• Credit scores
• GLM
• Large scale multivariate analysis
• Wide range of internal and external data
• Strong segmentation/lifts
• First movers advantages
• About 25 of small commercial risks are being
  modeled currently

5
First Movers Advantages Personal Auto
6
Segmentation - Loss Ratio Lift Curve
Loss Ratio
Credit Score Decile
7
Credit Score Power Personal Auto
8
Traditional Segmentation by Class
Roofers
140
90
135
87
Overall Loss Ratio of 75
125
82
115
78
110
75
Florists
100
72
90
68
80
65
70
63
60
Actual loss ratio
Below average
Internal data
Average
Above average
9
PM Scoring Go Beyond Class Based Rating and UW
Internal / External Data Predicted Loss Ratio
120
Joes Flower Shop Score 821
90
Lindas Flower Shop Score 324
82
Overall L.R. 75
78
Predicted Loss Ratio
74
70
66
62
58
50
10
Challenges for Commercial Lines PM

• Less uniform and less homogeneous for exposures
• Diverse lines of business
• Diverse industry classes/groups
• A wide range of policy size
• Market driven pricing
• Class driven pricing
• Rating bureaus driven pricing

11
Challenges for Commercial Line PM

• Large loss impact
• Longer tail loss development
• Data challenges
• Less cleaner (more changes, more missing, etc)
• Less data available
• Less standardized
• Not detailed enough
• Some information not captured
• Information not on exposure level
• IT support challenges lacking experience

12
Challenges for Commercial Line PM

• Implementation challenges
• IT challenges
• Business implementation challenges
• How to use model results?
• How to monitor the results?
• Management buy-in
• Just a modeling project or a strategic
  initiative?

13
Upsides for Commercial Line PM

• Less regulation
• Low hanging fruits
• Significant pricing and UW inadequacy for
  segmentation
• Significant expense saving
• Significant operation improvement
• First movers advantages still exist!

14
Keys for Successful Commercial Lines PM

• Model design critical
• Less likely for pure premium (freq/severity)
  modeling more likely for loss ratio modeling
• Less likely on exposure level more likely on
  policy level
• Actuarial design issues premium on-leveling,
  loss development, large loss impact, etc.
• Implementation considerations data availability,
  model design vs. implementation design, etc.

15
Keys for Successful Commercial Line PM

• Search for powerful predictive variables
• Fully utilize internal and external variables
  policy, agent, billing, drivers, vehicles,
  location, building, company financial and
  operation information, demographic information,
  etc.
• Be creative
• Garbage in garbage out, comprehensive
  diagnostics on model results
• Size
• Industry
• Geography
• Programs
• etc.

16
Keys for Successful Commercial Lines PM

• Strong project management
• Company management buy-in
• Well designed implementation plan
• IT implementation
• Business implementation, significant impact on
  operations
• UW process
• Business flow
• Agency management
• etc

17
Our Experience for Commercial Lines PM

• It can be done for all the major commercial
  lines WC, Auto, GL, Property, and BOP
• If carefully designed and executed, lifts curves
  comparable to personal lines can be achieved
• Some insights by line of business
• BOP typically has the cleanest data
• WC results is most sensitive to size of risks
• Autos results are most stable and can be
  extrapolated to mid and large size market
• GL results very experienced driven
• Subjective/market driven pricing typically random
• External financial data useful, but may have low
  hit rates
• Personal lines experience can be equally applied
  to small commercial risks, driver, credit, MVR,
  etc
• Significantly inadequate experienced rating and
  UW across all lines

18
Conclusions

• PM can work for commercial lines
• Similar segmentation power as personal lines
• Significant paradigm shift from traditional
  class based UW mind set
• Keys for successful commercial line PM
• Strong project management skill
• IT cooperation
• Careful model design
• Creative for predictive variables
• Implementation plan
• First movers advantages still exist!