Predictive Modeling in Underwriting
6 likes3,102 views
This document discusses predictive modeling approaches for life insurance underwriting. It took a long time for predictive modeling to be applied to underwriting due to the conservative nature of life insurance and the time needed to see results. Now, more data and computing power are available. Approaches include replicating current underwriting decisions or directly modeling applicant mortality rates. Various data sources can be used, including internal, third party, and customer data. Issues in building the predictive model include how to develop and update the model over time. Companies must decide how to incorporate these approaches and start collecting relevant data.
Predictive Modeling in Underwriting
- 1. Predictive Modeling in Underwriting BARRY SENENSKY FSA, FCIA, MAAA 14 Oct 2015 1
- 2. Agenda 1. Why has it taken so long? 2. Predictive Modeling Approaches 3. Data Sources 4. Building the Predictive Model 5. Summary 2
- 3. Why has it taken so long?
- 4. A long time coming… Predictive Modeling has been used in Industry for 50+ years Predictive Modeling has been used by P&C Insurers for 20+ years Predictive Modeling has been used for scoring Disability Claims on the likelihood of recovery for over 10 years So why not Underwriting? 1. Life Insurance business is conservative/slow to change 2. Results take 5-‐10 years to become apparent 4
- 5. So why now? Availability of Data and CPU’s to process the data Fits well with Online Insurance Sales where companies are looking for less expensive, less intrusive and quicker ways to sell insurance policies Just makes too much sense 5
- 7. Predictive Modeling Approaches 1. Replicate Current Underwriting Decisions 2. Model mortality rates directly for unique applicants 7
- 8. Data Sources 8 1. Internal 2. Third Party 3. Customer’s own
- 9. Internal Data Sources 1. Data Collected from current underwriting practices 2. Application • Provides good underwriting information • May have material inaccuracies 3. Fluids and other medical tests/information • Provides good underwriting information • Slow and expensive to collect • Poor customer experience 9
- 10. Third Party Data Includes data about an individual obtained from a third party including data; • Purchased from data aggregator such as LexisNexis • Purchased from another company that has the individual as a customer such as a pharmacy or telecommunications provider • Scraped off the web such as from Linked in or Facebook 10
- 11. Third Party Data Advantages • Quick to obtain • Low cost • Physically Non-‐invasive Concerns • Reliability and completeness of data • Customer Privacy 11
- 12. Customers’ Own Data • Includes data collected from EHR’s, wearable devices and wellness programs • Early indications are positive for Auto Insurance • Skeptical of value in near future for Life Insurance Underwriting 12
- 13. Building the Predictive Model
- 14. Two Possible Approaches 1. Replicate current underwriting decisions 2. Model mortality rates directly for unique individuals 14
- 15. Replicate Current Underwriting Decisions Possible Objectives • Enhance consistency of decisions between underwriters • Identify predictive data fields • Replace existing process with one that is quicker cheaper Advantages • Historical experience is not required • Fairly straightforward to develop Issues • Maintains but does not improve underwriting decisions • Issue of how to keep current over time 15
- 16. Modeling Mortality Rates Directly Objectives • Identify predictive data fields • Replace existing process with one that is quicker and cheaper • Predict applicant specific mortality rates Advantages • Should improve underwriting decisions and profitability of business Issues • Need historical experience for all applicants • How do you get vital status? • Many modeling issues 16
- 17. Modeling Mortality Rates Directly-‐ Modeling Issues • Build a model from scratch or start with a standard table? • How many years from issue do we model? Then what? • How do we incorporate mortality improvement? 17
- 18. Smaller Company Issues • Accumulating large enough data sets to build credible models • Higher unit cost of building infrastructure 18
- 19. Ongoing Management • Need to periodically refresh models • Predictive models are good at assessing benefits of questions on applications and medical tests 19
- 20. Summary 20
- 21. Summary • Predictive Modeling in Underwriting has arrived • If you haven’t done so yet; ØNeed to decide how you want to incorporate into your underwriting process Øidentify and start collecting relevant data 21
- 22. 22