42
Predictive Modeling in Reserving Predictive Modeling with Claim Analytics CLRS Presentation ST3 Chris Gross 1

Predictive Modeling in Reserving - Gross Consulting Modeling with Claim Analytics.pdf · Predictive Modeling in Reserve Analysis • It’s all predictive modeling isn’t it? •

  • Upload
    others

  • View
    9

  • Download
    0

Embed Size (px)

Citation preview

Page 1: Predictive Modeling in Reserving - Gross Consulting Modeling with Claim Analytics.pdf · Predictive Modeling in Reserve Analysis • It’s all predictive modeling isn’t it? •

Predictive Modeling in Reserving

Predictive Modeling with Claim AnalyticsCLRS Presentation ST‐3

Chris Gross

1

Page 2: Predictive Modeling in Reserving - Gross Consulting Modeling with Claim Analytics.pdf · Predictive Modeling in Reserve Analysis • It’s all predictive modeling isn’t it? •

Predictive Modeling in Reserve Analysis

• It’s all predictive modeling isn’t it?• This discussion refers to the what is commonly termed ‘predictive modeling’‐multivariate models, statistical rigor, etc.

• Emphasis in the past on pricing• Reserving getting attention

2

Page 3: Predictive Modeling in Reserving - Gross Consulting Modeling with Claim Analytics.pdf · Predictive Modeling in Reserve Analysis • It’s all predictive modeling isn’t it? •

Ways to Incorporate Predictive Modeling Into Reserve Analysis

• Analysis of specific loss development data/processes, for example:– Case reserve adequacy– Closure rates

• Modification of triangles• Reserve segmentation• Full description of the entire process, with resulting estimate of reserves

3

Page 4: Predictive Modeling in Reserving - Gross Consulting Modeling with Claim Analytics.pdf · Predictive Modeling in Reserve Analysis • It’s all predictive modeling isn’t it? •

Why do it?

• Use more of the information contained in your data

• Improve predictive accuracy• Quicker recognition of changing environment • Better reserve allocations• Layering of losses• Improved operational or strategic business decisions

4

Page 5: Predictive Modeling in Reserving - Gross Consulting Modeling with Claim Analytics.pdf · Predictive Modeling in Reserve Analysis • It’s all predictive modeling isn’t it? •

Challenges

• Same as with P&C reserving in general– Loss development occurs over time, mature periods are old

– Immature claims contain information

• Many facets of loss development• Helpful to concentrate on a single time‐step (e.g. beginning of quarter to end of quarter)

5

Page 6: Predictive Modeling in Reserving - Gross Consulting Modeling with Claim Analytics.pdf · Predictive Modeling in Reserve Analysis • It’s all predictive modeling isn’t it? •

A Fairly Simple Example

• Loss development on reported claims• Ignore for now questions of emergence• Also ignore re‐opened claims, salvage and subrogation.

• Individual claim detail for 20 calendar quarters• Only need data on claims that were open at the beginning of each quarter

6

Page 7: Predictive Modeling in Reserving - Gross Consulting Modeling with Claim Analytics.pdf · Predictive Modeling in Reserve Analysis • It’s all predictive modeling isn’t it? •

DataFinancial Data Exposure Characteristics

Beginning Case Reserve TypeEnding Case Reserve ProductPayment in Period ZIP Code

Timing Data Claim CharacteristicsAccident Quarter Loss CauseReport Quarter Loss Cause ‐ DetailValuation Quarter

7

Page 8: Predictive Modeling in Reserving - Gross Consulting Modeling with Claim Analytics.pdf · Predictive Modeling in Reserve Analysis • It’s all predictive modeling isn’t it? •

Claim activity from the beginning of the quarter to the end of the quarter

Did the Claim Close?

Does the Claim Have a New Value?

Is there a Payment?

What is the New Value?

How much is the Payment?

Arrows indicate dependency on other results

A number of available claim or exposure characteristics may have predictive value for any of these questions.

8

Page 9: Predictive Modeling in Reserving - Gross Consulting Modeling with Claim Analytics.pdf · Predictive Modeling in Reserve Analysis • It’s all predictive modeling isn’t it? •

Probability of a Claim Closing

• Base probability of 71%

• Modification of this probability by various claim characteristic values that were found to have predictive value

9

Page 10: Predictive Modeling in Reserving - Gross Consulting Modeling with Claim Analytics.pdf · Predictive Modeling in Reserve Analysis • It’s all predictive modeling isn’t it? •

Close Probability – Claim Age

10

Page 11: Predictive Modeling in Reserving - Gross Consulting Modeling with Claim Analytics.pdf · Predictive Modeling in Reserve Analysis • It’s all predictive modeling isn’t it? •

Close Probability – Loss Cause (detailed)

11

Page 12: Predictive Modeling in Reserving - Gross Consulting Modeling with Claim Analytics.pdf · Predictive Modeling in Reserve Analysis • It’s all predictive modeling isn’t it? •

Close Probability – Loss Cause

12

Page 13: Predictive Modeling in Reserving - Gross Consulting Modeling with Claim Analytics.pdf · Predictive Modeling in Reserve Analysis • It’s all predictive modeling isn’t it? •

Close Probability – Accident Quarter 

13

Page 14: Predictive Modeling in Reserving - Gross Consulting Modeling with Claim Analytics.pdf · Predictive Modeling in Reserve Analysis • It’s all predictive modeling isn’t it? •

Close Probability ‐ Product

14

Page 15: Predictive Modeling in Reserving - Gross Consulting Modeling with Claim Analytics.pdf · Predictive Modeling in Reserve Analysis • It’s all predictive modeling isn’t it? •

Close Probability ‐ Type 

15

Page 16: Predictive Modeling in Reserving - Gross Consulting Modeling with Claim Analytics.pdf · Predictive Modeling in Reserve Analysis • It’s all predictive modeling isn’t it? •

Probability of Change in Value (Given Not Closed)

• Base probability of 37%

• 4 characteristics found to be predictive

16

Page 17: Predictive Modeling in Reserving - Gross Consulting Modeling with Claim Analytics.pdf · Predictive Modeling in Reserve Analysis • It’s all predictive modeling isn’t it? •

Change Probability – Reported Quarter

17

Page 18: Predictive Modeling in Reserving - Gross Consulting Modeling with Claim Analytics.pdf · Predictive Modeling in Reserve Analysis • It’s all predictive modeling isn’t it? •

Change Probability – Claim Age 

18

Page 19: Predictive Modeling in Reserving - Gross Consulting Modeling with Claim Analytics.pdf · Predictive Modeling in Reserve Analysis • It’s all predictive modeling isn’t it? •

Change Probability – Loss Cause 

19

Page 20: Predictive Modeling in Reserving - Gross Consulting Modeling with Claim Analytics.pdf · Predictive Modeling in Reserve Analysis • It’s all predictive modeling isn’t it? •

New Claim Value (Given Changed but Not Closed)

• Base factor of 1.98 to beginning case reserve

• Modification to this linear relationship, as well as five additional predictive characteristics

20

Page 21: Predictive Modeling in Reserving - Gross Consulting Modeling with Claim Analytics.pdf · Predictive Modeling in Reserve Analysis • It’s all predictive modeling isn’t it? •

New Claim Value ‐ Case Reserve

21

Page 22: Predictive Modeling in Reserving - Gross Consulting Modeling with Claim Analytics.pdf · Predictive Modeling in Reserve Analysis • It’s all predictive modeling isn’t it? •

New Claim Value – Loss Cause

22

Page 23: Predictive Modeling in Reserving - Gross Consulting Modeling with Claim Analytics.pdf · Predictive Modeling in Reserve Analysis • It’s all predictive modeling isn’t it? •

New Claim Value – ZIP Code

0

10

20

30

40

50

60

70

80

90

100

0.6 ‐ 0

.7

0.7 ‐ 0

.8

0.8 ‐ 0

.9

0.9 ‐ 1

.0

1.0 ‐ 1

.1

1.1 ‐ 1

.2

1.2 ‐ 1

.3

1.3 ‐ 1

.4

1.4 ‐ 1

.5

1.5 ‐ 1

.6

1.6 ‐ 1

.7

Num

ber o

f ZIP Cod

es

Factor

23

Page 24: Predictive Modeling in Reserving - Gross Consulting Modeling with Claim Analytics.pdf · Predictive Modeling in Reserve Analysis • It’s all predictive modeling isn’t it? •

New Claim Value‐ Loss Cause (Detail)

24

Page 25: Predictive Modeling in Reserving - Gross Consulting Modeling with Claim Analytics.pdf · Predictive Modeling in Reserve Analysis • It’s all predictive modeling isn’t it? •

New Claim Value ‐ Product

25

Page 26: Predictive Modeling in Reserving - Gross Consulting Modeling with Claim Analytics.pdf · Predictive Modeling in Reserve Analysis • It’s all predictive modeling isn’t it? •

Probability of Payment

• Base probability of 51% for any given claim

• Additional predictive power for five claim characteristics

26

Page 27: Predictive Modeling in Reserving - Gross Consulting Modeling with Claim Analytics.pdf · Predictive Modeling in Reserve Analysis • It’s all predictive modeling isn’t it? •

Payment Probability – Case Reserve

27

Page 28: Predictive Modeling in Reserving - Gross Consulting Modeling with Claim Analytics.pdf · Predictive Modeling in Reserve Analysis • It’s all predictive modeling isn’t it? •

Payment Probability – Product

28

Page 29: Predictive Modeling in Reserving - Gross Consulting Modeling with Claim Analytics.pdf · Predictive Modeling in Reserve Analysis • It’s all predictive modeling isn’t it? •

Paid on a Claim Closed (Given Payment Occurs)

• Basic assumption of linearity with Case Reserves – base factor of 1.23

• Beyond that assumption, case reserve level and summarized cause of loss found to have predictive power

29

Page 30: Predictive Modeling in Reserving - Gross Consulting Modeling with Claim Analytics.pdf · Predictive Modeling in Reserve Analysis • It’s all predictive modeling isn’t it? •

Payment Amount – Case Reserve

30

Page 31: Predictive Modeling in Reserving - Gross Consulting Modeling with Claim Analytics.pdf · Predictive Modeling in Reserve Analysis • It’s all predictive modeling isn’t it? •

Payment Amount – Loss Cause

31

Page 32: Predictive Modeling in Reserving - Gross Consulting Modeling with Claim Analytics.pdf · Predictive Modeling in Reserve Analysis • It’s all predictive modeling isn’t it? •

Fraction Paid on an Open Claim (Given Payment occurs)

• Base factor to new value (or previous if unchanged) of 56%

• None of the variables were found to add predictive value

32

Page 33: Predictive Modeling in Reserving - Gross Consulting Modeling with Claim Analytics.pdf · Predictive Modeling in Reserve Analysis • It’s all predictive modeling isn’t it? •

Bringing it together• Simulation can be used to project activity in the next quarter

• It is necessary to project not only the predictive relationships, but also the residual error term.

• Chain through quarters using information from the previous simulated quarter.

• Store results, preferably at the claim level.

33

Page 34: Predictive Modeling in Reserving - Gross Consulting Modeling with Claim Analytics.pdf · Predictive Modeling in Reserve Analysis • It’s all predictive modeling isn’t it? •

0  1  2 3  4  5 6  7  8 9  10  11 12  13  14 15  16  17 18  19  20Future Quarter

Simulated Future Development (Mean Path)

Case

Paid

34

Page 35: Predictive Modeling in Reserving - Gross Consulting Modeling with Claim Analytics.pdf · Predictive Modeling in Reserve Analysis • It’s all predictive modeling isn’t it? •

35

0 0.2 0.4 0.6 0.8 1

Grand Total

Probability distribution of total payments

Page 36: Predictive Modeling in Reserving - Gross Consulting Modeling with Claim Analytics.pdf · Predictive Modeling in Reserve Analysis • It’s all predictive modeling isn’t it? •

36

0 0.2 0.4 0.6 0.8 1

Grand Total

Mean of total payments

Page 37: Predictive Modeling in Reserving - Gross Consulting Modeling with Claim Analytics.pdf · Predictive Modeling in Reserve Analysis • It’s all predictive modeling isn’t it? •

37

0 0.2 0.4 0.6 0.8 1

Grand Total

Current case reserves

Page 38: Predictive Modeling in Reserving - Gross Consulting Modeling with Claim Analytics.pdf · Predictive Modeling in Reserve Analysis • It’s all predictive modeling isn’t it? •

0 0.2 0.4 0.6 0.8 1

Product 1

0 0.2 0.4 0.6 0.8 1

Product 2

0 0.2 0.4 0.6 0.8 1

Product 3

0 0.2 0.4 0.6 0.8 1

Product 4

38

Page 39: Predictive Modeling in Reserving - Gross Consulting Modeling with Claim Analytics.pdf · Predictive Modeling in Reserve Analysis • It’s all predictive modeling isn’t it? •

0 0.2 0.4 0.6 0.8 1

Type 1

0 0.2 0.4 0.6 0.8 1

Type 2

0 0.2 0.4 0.6 0.8 1

Type 3

39

Page 40: Predictive Modeling in Reserving - Gross Consulting Modeling with Claim Analytics.pdf · Predictive Modeling in Reserve Analysis • It’s all predictive modeling isn’t it? •

0 0.2 0.4 0.6 0.8 1

Loss Cause 1

0 0.2 0.4 0.6 0.8 1

Loss Cause 2

0 0.2 0.4 0.6 0.8 1

Loss Cause 3

0 0.2 0.4 0.6 0.8 1

Loss Cause 4

0 0.2 0.4 0.6 0.8 1

Loss Cause 5

0 0.2 0.4 0.6 0.8 1

Loss Cause 6

40

Page 41: Predictive Modeling in Reserving - Gross Consulting Modeling with Claim Analytics.pdf · Predictive Modeling in Reserve Analysis • It’s all predictive modeling isn’t it? •

Discussion of Additional Complexity

• Relationship between Loss and ALAE• Emergence

– Added problem of unknown claim characteristics– Simulation approach driven by need for detail (ex. path is important, or only ultimate)

– Valuable link to pricing• Re‐opened claims• Changing claim characteristics• Salvage & Subrogation

41

Page 42: Predictive Modeling in Reserving - Gross Consulting Modeling with Claim Analytics.pdf · Predictive Modeling in Reserve Analysis • It’s all predictive modeling isn’t it? •

Uses of Methods

• Opinion of reserve and uncertainty including interim projections and their uncertainty

• Investigating questions from a traditional triangle analysis

• Indicating potential problems with a triangle analysis

• More focused discussion with claim department

• More sophisticated allocation of reserves

42