Project Management For Class Plan Projects CAS Special Interest Seminar on Predictive Modeling

advertisement
Project Management For Class
Plan Projects
CAS Special Interest Seminar on
Predictive Modeling
October 11-12, 2007
Jonathan White
Hope Is Not a Strategic Plan
2
Project Management Basics
Set research project goal
 Define project tasks
 Organize – work breakdown
structure (WBS)

–
–
–
–
Identify dependencies
Sequence of tasks
Resource requirements of tasks
Duration of tasks
Project schedule – Gant chart
 Project milestones

3
Planning

Clear understanding of the
business objectives
– Where is the plan to be used –
Channel/state/
– Marketing source
– Product/business expectations
– Current competitive position
– Implementation strategy
4
Planning

Clear research objectives
–
–
–
–
–
–


Statistical methodology
Software
New data
New variables
Scope of analysis
Required research infrastructure
Defines
Research
Scope
Research vision of end product
IT / Operations / U/W …
integration
5
Balancing Triangle
SCOPE
6
Planning

Environmental changes
– Business strategy
– Regulatory

Vulnerabilities
– Data
– Methodology
– Software

Research contingencies
– What if xxxx does not work?
7
Phases of a Class Plan Project
1.
2.
3.
4.
5.
6.
Planning
Data
Equity Modeling
Selection
Implementation
Monitoring
8
Data
 Internal
– Previously used
– New
 External
– Hit rate
 Understand
the data
 Data
quality
 Significant potential for delays
9
Equity Modeling
 Many
times the simplest phase
 Modeling options
– Regression
– Classification trees
– Neural nets
 Dependent
variable
– Loss ratio
– Claim frequency/claim severity
– Pure premium
10
Equity Modeling
 Holdout
sample
 Threshold for variable inclusion
 Statistical measures of overall
model predictiveness
 Deliverables
– Structure
– Indicated factors
 Peer
Review
11
Selection
 Recent
performance vs. Modeling
indications
 New plan vs. Old plan comparison
– Dislocation
– Conversations with the business
– Percentage change vs. Observed loss
ratio
 Cost
benefit conversations
12
Percentage Change Vs. Observed
LR
2.5
60%
50%
Rel LR
Avg change
40%
30%
1.5
20%
10%
1.0
0%
Average change
Relative loss ratio
2.0
-10%
0.5
-20%
-30%
0.0
-40%
Largest
decr
2
3
4
5
6
7
8
9
Biggest
incr
Decile by average change
13
Selection
 Benefits
analysis
 Price optimization
– Data requirements
– Analysis requirements
 Environmental
changes
– Internal
– External
14
Benefits analysis
 Loss
ratio/issue rate
 Requirements
– Equity model (indicated class plan)
– Quote data - Issue rate model
– Retention data - Retention rate model
(if applied to renewal business)
– Re-rating capabilities
– Competitor pricing data or proxy
15
Implementation

Off-balance
– Traditional – static environment
– Dynamic – reflect mix changes

Filing support
– What do you do when loss ratios don’t
mean very much?

State exceptions – personal auto
– Credit, driving record, prior carrier info
– Balance efficiency and accuracy
16
Monitoring
 Key
metrics
– Loss performance
– Issue rate
– Operational metrics
 Importance
of early reads
– Leading indicators
 Benchmarks
17
Loss Performance Monitoring
 Determine
key variables/niches
 Establish and calculate benchmarks
 Look for leading indicators
– Claim frequency
– Stabile, quick settling coverages
– Cap severity
– Credibility weight observed with
benchmarks
18
Issue Rate Monitoring
 Leading
indicator
 Problem of quotes
 Comparison points
– Prior plan
– Benchmark
 Do
you believe the market or your
equity modeling?
19
Hope Is Not a Strategic
Plan
20
Download