Transcript Document

Renewing LTD
Using Data Mining Techniques
Canadian Institute of Actuaries
November 10, 2005
Barry Senensky FCIA
www.claimanalytics.com
Claim Analytics
Discovery Through Statistics
Agenda
• Data mining
• Claims scoring
• Using claim scoring to
develop LTD reserve
termination assumptions
Claim Analytics
Discovery Through Statistics
Data Mining
Defined
• Extraction of previously unknown
information from large data sets or
databases
• Finding and quantifying of hidden
patterns and trends in databases
Claim Analytics
Discovery Through Statistics
Data Mining
Applications
• Used extensively in industry:
•
Credit card and tax fraud detection
•
Credit scoring
•
Weather prediction
•
Handwriting to text conversion
•
Many, many other applications
Claim Analytics
Discovery Through Statistics
Data Mining Tools
1. CART
Filter.
Identifies factors with
greatest impact.
2. Neural Networks
Optimization tools
3. Genetic Algorithms
Claim Analytics
Discovery Through Statistics
Neural Networks / Genetic Algorithms
How they learn

Model is presented with data sample
with known outcomes

Model predicts result, then compares it
to actual outcome

Model parameters are changed to better
approximate the sample…

…Over and over again.
Claim Analytics
Discovery Through Statistics
Claims Scoring
Claims are scored from 1 to 10.
P. Chang Score: 8
# 451156
Scores show likelihood of return to
work within a given timeframe.
Scores are calibrated:
• score of 1 indicates 0 – 10%
J. Loe Score: 6
# 452009
chance of recovery within given
timeframe, score of 2 indicates 10 –
20% chance of recovery within
given timeframe, and so on.
J. Spratt Score: 4
# 452135
Claim Analytics
Discovery Through Statistics
Scoring Report
Claim #
Q.P.
Elim
Diagnosis
451156
119
Depression Reactive
(Prolonged)
M
42
1411
452009
364
Tear Medial
Meniscus (Knee)
M
47
2500
452135
180
Fibromyalgia
F
37
3899
452338
180
Major Depressive
Disorder
F
35
1773
452341
119
Lumbar Disc
Degen/Disease
M
42
1150
Herniated Disc Acute
F
452494
210
Sex Age
Benefit (Other )
…
…
…
6M
24M
7
10
4
7
6
6
6
8
2
5
2
2
…
59
3564.9
…
Claim Analytics
Discovery Through Statistics
Five steps to developing
LTD termination rates for Dave
using claim scoring
Dave
Claim Analytics
Discovery Through Statistics
Developing termination
rates for Dave
About Dave
Sex
Male
Age
44
QP
90 days
Diagnosis
Osteoarthritis
Claim Analytics
Discovery Through Statistics
Developing termination
rates for Dave
Dave’s claim scores
Likelihood of RTW (%)
3 months
6 months
12 months
24 months
5.9
14.7
27.5
34.5
Claim Analytics
Discovery Through Statistics
Developing termination
rates for Dave
Step One
Get Cumulative RTW Probabilities
•cumulative RTW Probabilities, 1-24 Months after EP
•expressed as %
1
2
3
4
5
5.9
13
14
15
6
7
8
9
10
11
14.7
16
17
18
12
27.5
19
20
21
22
23
24
34.5
Claim Analytics
Discovery Through Statistics
Developing termination
rates for Dave
Step Two
Interpolate between months
• choose uniform distribution, constant force or Balducci
• here, used uniform distribution
• expressed as %
1
2
3
4
5
6
7
8
9
10
11
12
2.0
3.9
5.9
8.8
13
14
15
16
17
18
19
20
21
22
23
24
28.1
28.7
29.3
29.8
30.4
31.0
31.6
32.2
32.8
33.3
33.9
34.5
11.8 14.7 16.8 19.0 21.1 23.2 25.4 27.5
Claim Analytics
Discovery Through Statistics
Developing termination
rates for Dave
Step Three
Get mortality rates
• Canadian Group LTD experience /1000 shown here
• alternative is company experience
• may want to make adjustments, e.g. improvement from midpoint of study
1
2
3
4
5
6
7
8
9
10
11
12
.27
.32
.40
.45
.49
.51
.52
.53
.52
.52
.50
.49
13
14
15
16
17
18
19
20
21
22
23
24
.47
.46
.44
.42
.40
.38
.37
.35
.34
.32
.31
.29
Claim Analytics
Discovery Through Statistics
Developing termination
rates for Dave
Step Four
Convert cumulative RTW probabilities to
month-to-month RTW rates
# of claimants who
will recover in
period.
TM cumulative RTW - LM cumulative RTW
1 - LM cumulative RTW - LM cumulative death rate
1
2
3
4
5
6
7
8
9
# of claimants
still on claim at
start of period.
10
11
12
1.97 2.00 1.99 2.96 2.98 2.97 2.15 2.12 2.11 2.10 2.10 2.09
13
14
15
16
17
18
19
20
21
22
23
24
.57
.56
.56
.56
.55
.55
.55
.55
.55
.55
.55
.55
Claim Analytics
Discovery Through Statistics
Developing termination
rates for Dave
Step Five
Calculate Termination Rates
• Termination rate = recovery rate + mortality rate
1
2
3
4
5
6
7
8
9
10
11
12
2.24 2.32 2.39 3.41 3.47 3.48 2.67 2.65 2.64 2.62 2.60 2.58
13
14
15
1.04 1.02 1.00
16
17
18
19
20
21
22
23
24
.98
.96
.94
.92
.90
.88
.87
.85
.84
Claim Analytics
Discovery Through Statistics
What to do after 24 months
• Produce scores for 36 months, then use
traditional methods thereafter
• Produce scores for all future terms
Claim Analytics
Discovery Through Statistics
Credibility
• Significant benefits over traditional methods:
• Rates are based on internal experience
• Data mining offers advantages over table of claims analysis
Table of Claims
Data Mining
Accuracy
Accurate in aggregate
Allows reserves to be accurately
allocated between claims:
important for renewal pricing,
experience-rated refunds etc.
Sensitivity
Sensitive to changes in
the age / elimination
period distribution of
claims.
Sensitive to many other factors
as well: diagnosis, gender,
income, province, occupation,
etc.
Claim Analytics
Discovery Through Statistics
Credibility
Testing the model
• Normally use back-testing to confirm fit of model
Claim Analytics
Discovery Through Statistics
Back-testing the Scoring Model
Recovery Rate
Recovery by Score - Validation Sample
0 - 24 Months
100%
16
80%
14
60%
12
40%
10
20%
8
0%
1
2
3
4
5
6
7
8
9
10
Recovery Rate
8%
20%
24%
37%
40%
54%
61%
79%
91%
96%
Pred Rec Rate
5%
15%
25%
35%
45%
55%
65%
75%
85%
95%
# of Claims
34
69
94
86
129
111
86
103
86
60
6
Claim Analytics
Discovery Through Statistics
Benefits
• More appropriate reserve for each
claim, avoid “averages of
averages”
• Aligned with claim management
practices
• Facilitates repricing / renewal
• Earlier recognition of changes in
trends and experience
Claim Analytics
Discovery Through Statistics
Summary
Claim scoring offers a new and
innovative way of setting LTD
termination rates that results in a
more appropriate reserve for
each claim.
Claim Analytics
Discovery Through Statistics