2017 Predictive Analytics Symposium

Transcription

1 2017 Predictive Analytics Symposium Session 29, Predictive Analytics for Inforce Management Moderator: Rohan Noel Alahakone, ASA, MAAA Presenters: Jenny Jin, FSA, MAAA Assaf Mizan Martin Snow, FSA, MAAA SOA Antitrust Compliance Guidelines SOA Presentation Disclaimer

2 SOA Predictive Analytics Seminar Session 29: Predictive Modeling for Inforce Management 15, Sept Jenny Jin, FSA, MAAA

3 Agenda Motivation for Predictive Models in Inforce Management Examples Society of Actuaries Predictive Analytics Seminar 2

4 My team I work with My role is Data scientists Actuaries Statisticians Technology specialists Data managers Business strategists Society of Actuaries Predictive Analytics Seminar 3

5 Industry Recognition of the Problem Moody s Investors Service Unpredictable Policyholder Behavior Challenges US Life Insurers Variable Annuity Business Though equity-market declines are generally seen as the biggest risk in VA contracts, most insurers effectively hedge that risk via derivatives. That leaves the less-easily hedged and more unpredictable policyholder behavior, and particularly lapses, as a key driver of the profitability of these popular products. Companies selling VAs misestimated and underpriced lapse risk. Retention by policyholders of these guaranteed products was much greater than expected, causing insurers to take significant, unexpected earnings charges and write-downs over the past year and a half. Recent experience for these guarantees provides [the takeaway that] Companies tend to retain customers that cost them the most and lose those that cost them the least. Society of Actuaries Predictive Analytics Seminar 4

6 What does tabular analysis really tell us? Descriptive analysis takes data and summarizes them using an average metric for the cohort but sometimes can be misleading if there are confounding variables Actuaries have the lowest divorce rate at 17%! Source: FlowingData.com, Wikipedia Society of Actuaries Predictive Analytics Seminar 5

7 Milliman VALUES Industry Study Data Set VALUES Industry Lapse Study covered 117 million quarterly observations, $500bn AV VALUES GLWB Utilization Study, 2 million policies, $200bn AV 70% training set 30% holdout set Society of Actuaries Predictive Analytics Seminar 6

8 Lapse Models: baseline and alternative implementations Baseline model Lapse Base Rate f(q) ITM Factor f(itm) Tabular approach Baseline predictive model Log Odds k 1 q k 2 ITM Factor f(itm) Milliman VALUES predictive model GLM regression model Log Odds k' 1 q k' 2 ITM Factor f(itm) Society of Actuaries Predictive Analytics Seminar 7

9 Milliman VALUES Lapse Study Post surrender charge lapse experience generally lower than current assumption Irrational Closest to actual experience More rational Less guess work on the effect of base vs dynamic lapse Predictive model provides a single framework for analyzing and attributing the impact to both duration and moneyness jointly. Society of Actuaries Predictive Analytics Seminar 8

10 Algorithms can help accelerate variable selection Relative importance of new predictors Category Behavior Demographic Distribution Policy size Policy state When faced with hundreds of potential variables, computers algorithms are much faster at selecting important variables based on their influence on the behavior Multiplicative impact to lapse rates Society of Actuaries Predictive Analytics Seminar 9

11 Predictive model improves predictions GWB GWB 4 3 Models VALUES predictive model Baseline predictive model Act 0.5 Models Baseline model Baseline predictive model Act Rank of relative probabilities Rank of relative probabilities Comparison of baseline tabular model to baseline predictive model Comparison of full predictive model to baseline predictive model Society of Actuaries Predictive Analytics Seminar 10

12 WB Withdrawals: Takeaways Policyholders who are older at issue tend to utilize their policies sooner Qualified policyholders will start their withdrawals sooner after age 70 Less than half of all policyholders currently taking GLWB withdrawals utilize their GLWB benefit with 100% efficiency Utilization inefficiency is a driver of lapse Proprietary and Confidential Society of Actuaries Predictive Analytics Seminar 11

13 What else can we learn about the customers? Insurance Company Data - Policy values - Product features - Policy behavior Outputs Vendor Data $ Customer Segmentation Consumer Data Enhanced Dataset Actuarial Assumptions by Segment Credit Data Analytics R x $ Policy Level Customer Value Vendor Data Mortgage Data Census Data Health Score Society of Actuaries Predictive Analytics Seminar 12

14 Annuity behavior modeling: progression of states High Companies are evolving on the analytics spectrum from descriptive and diagnostic analysis to predictive and prescriptive analysis. Complexity Prescriptive What can be done? Predictive What may happen? Foresight Insight Next-generation experience studies will use much wider set of explanatory variables and more sophisticated analysis techniques to find non-linear, multivariate effects, complex interactions Diagnostic Why did it happen? Descriptive What happened in the past Hindsight Insights from experience studies can be used to develop individual policyholder profiles and to drive product development and create positive engagement with customers Low Business Value High Society of Actuaries Predictive Analytics Seminar 13

15 Key Takeaways Predictive models are well suited to investigating policyholder experience data Reduced cost of storage and computing has largely lifted constraints around predictive modelling methods and applications Actuarial judgement is still required, in particular to avoid creating models that are hard to interpret or implement. A multidisciplinary team is necessary to successfully advance in this new area: Subject matter expertise in the products, policyholder use of products, and the financial implications to insurers. Data managers Data scientists Technology developers IT infrastructure Start with the low hanging fruits such as maintaining high quality data and collecting supplementary data sources and grow from there. Society of Actuaries Predictive Analytics Seminar 14

16 Thank You! Jenny Jin Life/FRM Consulting Actuary Chicago office

17 Predictive Analytics Symposium Predictive Analytics for In Force Management Session 29 Some Practical Examples September 15, 2017 Assaf Mizan, FIA, FILAA 1

18 Why predictive analytics Predictive Analytics in insurance: Use of advance modelling and data mining tools to predict behavior in the insurance space Contrasted to traditional actuarial methods, these tools have important advantages in the insurance field, in particular: Few constraints on the volume of data and features used in the analysis Lower requirement for clean data Capture complex interactions Require significant processing power and data storage, but these are now much cheaper 2

19 Overall flow Historical / Snapshot data Policy Admin System Premium Collection Customer Rel-ship Management External data enhancement One view of the client Additional Tables Validation Feature engineering Business question Actions Results & Insights The modeling Arena 3

20 Business Question and Modelling Example 1 - Persistency High level business question: Why are policyholders lapsing? Target model: Fit a model to the historical persistency of each policy. Several models can be fitted depending on product complexity, variation by policy duration, data availability etc. The model can deliver insights as to the influencing features and rank current policies based on their propensity to lapse (now and in the future) Example 2 - Underinsurance High level business question: Is there an underinsurance opportunity? Target model: Cluster policyholders based on similar characteristics policy type, policyholder characteristics, external data. Find outliers which have significantly low sum assured compared to peers. Important to set clear validation goal to ensure model is adding value 4

21 Feature engineering Feature engineering is the process of using domain knowledge of the data to create features that make machine learning algorithms work. Feature engineering is fundamental to the application of machine learning, and is both difficult and expensive. (Wikipedia) 5

22 Feature engineering - examples How would we incorporate the sale date 24/11/2010? Feature Date (number) November Meaning Models long term trends Month-specific sale features, year end targets 24 th day of the month Month end targets? Monday Proximity to public holidays Proximity to financial / political events.. Significance of week days Policyholder behavior Policyholder behavior Premium / Contribution information: Missing / Skipped premium - How many times did a policyholder miss a premium? - How recent to the current date? - How to combine the information for example: Missing_Premium = (number of times missed in last 24 months) * (1 / distance of last one) 6

23 Feature engineering - examples Free-Text professions: - Thousands of different occupations, not useful for analysis - Used advanced clustering techniques to map to 10 groups - Result: Occupation class is significant to lapsation - Another enhancement connect to underwriting risk classification 7

24 External Data Most common Demographic data based on address, for example: Median earnings,average household, homeowner vacancy,median age Lifestyle type data (subscriptions etc) Depending on product, financial data can also be useful Lifetime events, from external sources, can be useful also triggers, for example: house purchase Job change New family member Company data which is external to the book analyzed for example from other operations (health, P&C?) All data and (engineered) features can be useful and may lead to powerful insights Time spent here is usually well rewarded! 8

25 Validation (lapse example) How do we know the model is working? Back-testing against historical data and against company assumptions: Define training period (eg ) Defined back testing period (eg 2016) Check actual and own company assumptions against model results for 2016 Continuous monitoring throughout 9

26 Insights Example - Persistency by features Once model is validated, we get three useful outcomes: Feature importance Predictions on a per policy level Ability to predict target based on simulated input Features can be grouped to three categories: External Indirect Pensioner Clerk Housewife Teacher Manager Direct 10

27 Insights Example - Persistency by features 11

28 Insights into actions and simulations For a lapse / underinsurance model, results are probabilities per policy. The model can assist in prioritizing policies for conservation or upsale: Either when the client approaches the company Or as a selective pro-active approach By adding an additional layer of per-policy profitability, insights turn from inforce projections to value projections In addition, model can be used to simulate business under various scenarios. Optimization techniques can be applied to select the preferred outcome 12

29 Some closing thoughts Predictive analytics can be a powerful tool in managing inforce business Results depend on availability of data. Frequency of client contact impact results. Use of external event data can substitute lacking internal data, but is more difficult / expensive to obtain. Feature engineering and augmentation are critical. We have not discussed feature correlation and masking but these are important issues which are tricky to handle Additional layers (eg profit) can be incorporated for simulations to help with strategic decisions Best results are achieved when predictive analytics are integrated to the business process 13

30 Contact details Assaf Mizan FIA, FILAA VP Product and Actuary Atidot Ltd 14

31 Predictive Analytics Symposium Predictive Analytics for In Force Management Session 29 September 15, 2017 Martin Snow, FSA, MAAA 1

32 SOCIETY OF ACTUARIES Antitrust Notice for Meetings Active participation in the Society of Actuaries is an important aspect of membership. However, any Society activity that arguably could be perceived as a restraint of trade exposes the SOA and its members to antitrust risk. Accordingly, meeting participants should refrain from any discussion which may provide the basis for an inference that they agreed to take any action relating to prices, services, production, allocation of markets or any other matter having a market effect. These discussions should be avoided both at official SOA meetings and informal gatherings and activities. In addition, meeting participants should be sensitive to other matters that may raise particular antitrust concern: membership restrictions, codes of ethics or other forms of self-regulation, product standardization or certification. The following are guidelines that should be followed at all SOA meetings, informal gatherings and activities: DON T discuss your own, your firm s, or others prices or fees for service, or anything that might affect prices or fees, such as costs, discounts, terms of sale, or profit margins. DON T stay at a meeting where any such price talk occurs. DON T make public announcements or statements about your own or your firm s prices or fees, or those of competitors, at any SOA meeting or activity. DON T talk about what other entities or their members or employees plan to do in particular geographic or product markets or with particular customers. DON T speak or act on behalf of the SOA or any of its committees unless specifically authorized to do so. DO alert SOA staff or legal counsel about any concerns regarding proposed statements to be made by the association on behalf of a committee or section. DO consult with your own legal counsel or the SOA before raising any matter or making any statement that you think may involve competitively sensitive information. DO be alert to improper activities, and don t participate if you think something is improper. If you have specific questions, seek guidance from your own legal counsel or from the SOA s Executive Director or legal counsel. 2

33 Presentation Disclaimer Presentations are intended for educational purposes only and do not replace independent professional judgment. Statements of fact and opinions expressed are those of the participants individually and, unless expressly stated to the contrary, are not the opinion or position of the Society of Actuaries, its cosponsors or its committees. The Society of Actuaries does not endorse or approve, and assumes no responsibility for, the content, accuracy or completeness of the information presented. Attendees should note that the sessions are audio-recorded and may be published in various media, including print, audio and video formats without further notice. 3

34 Predictive Analytics for In Force Management - Introduction Interest in Predictive Analytics (PA) for In Force Management is increasing Yet, significantly more can be done: More companies can become engaged Engaged companies can do more use cases Projects can move from the pilot phase to implementation Types of in force management use cases today Envisioning future projects Who is doing the work? How can actuaries do more? How is success defined for a PA project? How to integrate multiple PA initiatives one with another What are the top obstacles to success and what can be done about them? 4

35 Predictive Analytics for In Force Management How Insurers Use PA Today Increase market penetration Expand customer relationships Develop accelerated underwriting programs Set mortality and policyholder behavior assumptions In force management 5

36 Predictive Analytics for In Force Management Prevalence of Insurer Use of PA for In Force Management 6

37 Predictive Analytics for In Force Management In Force Management Use Cases Today Targeted conversion Post level premium term conservation management In force management pre lapse In force management post lapse In force management other customer interaction Agent monitoring/management Up-sell Buy-outs on non-profitable customers Other 7

38 Predictive Analytics for In Force Management Who does PA Work for Inforce Management? Who leads the work? Internal vs. External Marketing, Data Science / Statistics, Actuary 8

39 Predictive Analytics for In Force Management How is Success Defined? SOA Report does not define success What criteria can be used to define success? What criteria are companies using? 9

40 Predictive Analytics for In Force Management Integration of Multiple PA Initiatives Rationalize Obtain Identify Vision Core Competency Milestones 10

41 Predictive Analytics for In Force Management Top Obstacles to Success Human behavior Legacy Infrastructure Buy-in Competing Priorities Cost-benefit Data sources, implementation, designing and building the model 11

42 Predictive Analytics for In Force Management Overcoming the Obstacles to Success Compelling Business Case Expert Opinions Carefully designed policyholder surveys Learn from products with similar characteristics Bayesian models 12

43 Presenter Contact Information Martin Snow, FSA, MAAA (732)