predictive modeling developments: us market86ac6afc-d4d4-44f8-ab3f...process in si business •there...

Predictive modeling developments: US MarketDr. Brian IvanovicInsurance Medicine Summit 2017

Insurance Medicine Summit 2017 2

Agenda

• Origins of predictive models in L&H business

• Approaches to risk scoring

• State of the evidence on mortality experience and risk scores

• Market uptake trends

• Best practices in implementing scores

• New developments in scores/AU processes

Insurance Medicine Summit 2017

Background

3


The DI incidence model provided us key insights on where costing required minor or more significant tweaks or a fundamental reconsideration of the nature of the risk we were taking on.

Early history on predictive models: (>20 yrs. ago)

4

0.0

0.5

1.0

1.5

2.0

2.5

3.0

3.5

4.0

Ris

k s

pre

ad

Modeled risk spread for DI pricing factors

Magnitude of risk (spread ofrisk) estimated frompredictive model

Magnitude of risk (spread) asoriginally costed

>500x


Early history: 2011 Swiss Re US Client Research interests survey Predictive modeling: Individual respondent comments

• 22 US client companies provided input

• 3rd most important “current” pricing research topic (PLT lapse and mortality experience #1)

• 2nd most important “future” pricing research topic (Final expense mortality #1)

• What people meant by PM, PA, etc. varied considerably

Definition for this session:Predictive modeling uses statistical tools to separate systematic patterns from random noise, and turns this information into business rules, which should lead to better decision making.


• There is wide variability in the approaches used to integrate PM into the underwriting process and the products it can be applied to.

– Expand non-med underwriting limits

– Cross-sell to other insured groups

– Add to the risk assessment process in SI business

• There is “hope” that use of PM will either help maintain price on life products sold as traditional requirements are dropped or lower the price when the tool is added to an existing process

Common predictive modeling themes in underwriting

6

Alcohol abuse

Previously Declined/Rated

High risk drugs

Impairment Major

# High risk drugs within 1 year

Business channel

Moving Violation

# Medium risk drugs

Age

Actively At Work

Decline model


Mortality implications

Traditional approach to fluidless underwriting meant skipping the paramed. Depending on their implementation new approaches may bring mortality closer to fully underwritten

7

Fully Underwritten

•Paramed with blood, urine

•Rx, MVR, MIB

Non-Med Underwriting

•No blood/urine

•Rx, MVR, MIB

Fluidless with Predictive

Model

•No blood/urine

•Rx, MVR, MIB

•Predictive model used along with triage process

-(25-35)%

+40%


Mortality forecasting at the producer level

8

We have worked with over 60 US clients in better understanding / implementing / assuming risk on business where predictive models are utilized.

Questions that can be answered sooner rather than later when a predictive model is used to estimate future mortality:

1. What is the quality of business individual producers are placing?

2. Trends in the quality of business booked at the individual or group producer level over time.

LMS predicted risk at producer level


Approaches to risk scoring

9


• Clinical

– Prescription histories

– Fluid testing results (non AU models)

– Facial analytics

• Non-clinical FCRA1 data

– Consumer credit data

– Public records (selected examples)

– property ownership

– criminal records

– licenses and permits

– bankruptcies

• Non-FCRA data

– Census level data (geo-spatial)

– Marketing data

Elements being evaluated/scored in predictive models

10

Vendors include different “features” (predictors) in their scores.

1. FCRA stands for Fair Credit Reporting Act. It is a US federal law that regulates how consumer reporting agencies use individual consumers' information. In order to make life insurance underwriting decisions on data, it must be (i) disclosable, (ii) disputable, and (iii) correctable.


Factors that impact mortality:

• Demographic of target population

• Risk score cut-off points

• Rules for pushing applicants to traditional underwriting

• Changes in APS practices

• Potential adjustments to preferred criteria

• How vendor score is used for applicants going through traditional underwriting (i.e. are applicants moved down a risk class for a poor score)

• Efforts made to improve disclosure (i.e. behavioral economics)

• Smoker detection model usage

• Post issue monitoring (i.e. random holdouts or post issue APS)

Decisions made will affect future mortality realized in product

11


State of the evidence behind risk scoring

12


• General population based studies

• General population subsets based on “insurance shoppers”

• Studies of aggregated previously issued life business (both inforce and lapsed)

• Individual client studies

Risk scoring and mortality results

13

What these studies provide:

• Distributional info on a particular score

• Mortality relationships within a particular scoring range compared to others

– Overall and by age and gender

• Additionally, for certain life studies

– Some indication of the relationship between traditional underwriter risk classification and scored business

– Some indications of the relationship of the risk score to groups of historic and ongoing underwriting and pricing interest (smokers vs non-smokers, preferred vs substandard, etc.)


• Early duration a/e experience in an open market environment for the subset of life business qualifying under a particular scoring threshold

• How future cause of death patterns are influenced for those in those within different risk score ranges

• The stability of risk scores over time in a particular individual and the mortality implications when scores materially change

• Performance differences by face amount band

• Differences in how scores work in persisting business only

What we would hope to learn from future studies?

14

Risk scores that include attributes linked to SES may modify cause of death patterns


Market uptake

15


• PA application areas

– Marketing

– Underwriting

– Post-Issue management

• Underwriting applications

– Choosing the right underwriting requirements

– Underwriting risk class

– Underwriting reclassification

SOA Predictive Analytics and Accelerated Underwriting Survey Report: May 2017

16

26 of 34 (76%) respondent companies have implemented in one or more of these areas.

Most programs have been implemented in past few years

Most impact <10% of total business underwritten (as of summer 2016)

Underwriting risk class was most commonly implemented underwriting application

Vendor data was the most common source used in implementing a program

Data limitations were the most common obstacle reported to developing a PA program.


Swiss Re’s support of clients interested in predictive analytics

17

We have worked with over 60 US clients in better understanding / implementing / assuming risk on business where predictive models are utilized.


Best practices in implementing PA models & new developments

18


• Before implementation perform retrospective studies of clients relevant business (% qualifying and mortality spread info)

• Reasonable AU qualification rates

• Random holdouts

• Review medical records post issue

– Can only rescind if the medical record finding was actually inquired on during the application process

Proactive AU program mortality management

19

Program auditing only has an impact on future mortality when the results of the investigations lead to changes in AU underwriting practices

0%

5%

10%

15%

Self reported BMI35kg/m2+

Paramed BMI35kg/m2+

Self reportedcurrent tobacco

usage

(+) Cotinine

Build disclosure Tobacco disclosure

Pre

vale

nce

Disclosure rate differentials between self reports and objective screens


• Non-fluid model based risk segmentation by smoking status

• Evolution of “scores” &/or release of new scores to include clinical attributes

– Prescription history based info

– Health history based info

– Medical claims info

• Integration of “wearables” data into risk assessment / scoring

– Need to factor in wearable “lapsation” (>50% stop using at 24 months)

What’s “in the news”

20

As first generation mortality experience emerges with predictive model underwriting its likely that many of the underlying models and approaches will have evolved.


Legal notice

22

©2017 Swiss Re. All rights reserved. You are not permitted to create any modifications or derivative works of this presentation or to use it for commercial or other public purposes without the prior written permission of Swiss Re.

The information and opinions contained in the presentation are provided as at the date of the presentation and are subject to change without notice. Although the information used was taken from reliable sources, Swiss Re does not accept any responsibility for the accuracy or comprehensiveness of the details given. All liability for the accuracy and completeness thereof or for any damage or loss resulting from the use of the information contained in this presentation is expressly excluded. Under no circumstances shall Swiss Re or its Group companies be liable for any financial or consequential loss relating to this presentation.

predictive modeling developments: us market86ac6afc-d4d4-44f8-ab3f...process in si business •there...

Documents