1 Using Predictive Analytics in Experience StudiesTeam: The Game of Life Charlie Andres, Long Du, Taylor Gallegan, Jessica Santos, Christopher Werner Long - “Our group “The Game of Life” was given the task to research predictive analytics in experience studies by the Goldenson Center, a case study presented to us by Prudential. 4/7/17 Uconn Goldenson Center Case Study; Case Study Courtesy of Prudential
2 Project Goals Using Predictive Analytics in Experience StudiesIdentify new, innovative variables that can be used in a predictive model for lapse and mortality rates Give a brief overview of how variables would be implemented Outline risks and challenges for this approach Long - The goal of our project was find new and innovative variables that could be implemented in a predictive model for lapse and mortality rates, as well as the risks associated with implementing them.
3 Why Predictive Analytics?Streamline underwriting process Cut costs Less invasive Higher volume and increased processing speed for sales of policies Prevents human error Model not prone to bias No fatigue Avoid “lack of experience” problem Allows for more precise marketing Identify individuals more likely to purchase a policy Long - “Predictive Analytics can be extremely beneficial for a company if implemented correctly. First and foremost, it allows for a streamlined underwriting process, which is cheaper for the company and also less invasive and quicker for the consumer to purchase. The model is also not subject to normal human behaviors that can be issues in the underwriting process such as bias, fatigue, or lack of experience. It can also be applied in other fields of the company such as to help the marketing team identify who is more likely to buy a policy and target these people to increase their sales.” -quicker purchase of policy can increase sales of policies -longer underwriting processes can deter consumers
4 Known Variables Useful in Predictive ModelingRiskiness of Profession Face Value Attained Age Gender Policy Limits Taylor
5 Innovative Variables Variables were selected based onPotential predictive power Accessability Efficiency Cost effectiveness Taylor
6 Medical Underwriting Not optimal to use medical underwriting LengthyInvasive Not cost effective Pharmaceutical Records Quicker Shows ailments that the patient has Most ongoing problems go with pharmaceutical data Jess medical- forms and blood (hard to obtain) Sometimes companies spend all the time and money to get these records but clients will lapse before their policy even starts so everything goes out the window goal: find a way to speed up underwriting process
7 Renting vs. Owning a HomeHomeowners are more likely to purchase life insurance Renters have more of a “temporary” mindset than homeowners Can be applied to lapse rates Homeowners less likely to make any changes in their life or lifestyle Leads to less lapse rates for homeowners than renters Jess
8 Credit Score Credit ScoresLower credit score shows riskier individuals Study shows higher mortality and lapse go with lower credit scores Easy and cheap to obtain Charlie
9 Motor Vehicle Records Motor Vehicle RecordsShows insight on risky behaviors driving Could mean risky life choices Readily available Charlie
10 Casino Gambling Propensity ScorePredicts the likelihood that an individual will have a gambling addiction based on online and in person visits to casinos and gambling hubs Available through third party sources Milliman study showed that this variable has high predictive power Those who have higher gambling propensity score are riskier individuals, leading to increased mortality and lapse rates Charlie
11 Autopay vs. Direct Pay Most insurers offer options to pay directly or automatic electronic payments Less likely to lapse automatic payments due to less thought about each payment Directly billed plans have more opportunities to lapse Charlie VS
12 Market Competition The rate of lapse will be affected whether or not a particular geographic location is more or less concentrated with alternative options for insurance If there are other insurers in the area, a policyholder may be lured away by a more appealing offer from the competition Taylor
13 Product Uniqueness and FlexibilityHow innovative and flexible a company's product is could also have an effect on a policyholder’s lapse rate The flexibility would allow them to change their policy as their needs change as opposed to lapsing Taylor
14 Geographical MortalityGrouping or classifying people from the same region in terms of high or low risk of mortality People living in the same area will likely have similar mortality rates because they face similar conditions Taylor
15 Wearable Technology Wearable fitness technology is growing in popularity More marketable Already used in the life insurance industry Policyholders can lower their premiums by living a healthier lifestyle Activities, gym, doctors Targets healthier population and deters riskier people Lower mortality rates with healthier policyholders Long -incentives to stay
16 Churn Scores A “Churn Score” is a score developed by third party data companies that looks at how often one cancel a phone/bank/TV contract Identifies consumer loyalty, which is correlated with lapsation Previous lapse information for the policyholder should be included as well Previous lapses indicate future lapse Chris
17 Family Member ReferralMillennials are likely to be loyal to a specific insurance company if a family member uses and values the company Less likely to depend on policy cost and company reputation than earlier generations By implementing some type of referral system to identify policyholders referred by a trusted family member, those identified would likely have lower lapse rates Chris
18 Broker vs. Online PurchasingThose that purchase a policy through a broker they are less likely to lapse Broker keeps in contact with consumer Large portion of the broker’s commision relies on the policyholder not lapsing Chris
19 Model Building OverviewCharlie - Wawa
20 Compiling Data Collect data in a format so one can perform statistical analysis 12-18 month timeframe would remove statistical variation yet keep credibility Data is from multiple different sources Sources may not be in same format and difficult to match Need to append all sources and match to company’s current database Charlie - Traditional underwriting variables such as age, policy value,geographical location etc. are already available within the company Data from the DMV, Credit Scores, etc is available for companies to access and is easy to match Third Party data sources can be used to find Churn scores, casino gambling scores, etc Pharma records have to be collected for current consumers in order to identify what drugs relate to higher mortality Other sources can be harder to obtain/map
21 Building a Dataset Turn dataset into variables (synthetic and nonsynthetic) Watch out for extremely correlated variables Look into dataset and perform univariate analysis to confirm the relationships and data make sense May require you to talk to other business units Further data scrubbing to fix the dataset Chris synthetic created, make sure distributions and data sets makes sense compliling data from the innovative variables and apending them into the company’s exsisting database scrubbing data set is removing any blank cells and stuff like that
22 Implementing the Predictive ModelPartition the dataset into 3 sets (train, validation, test) Find the best method to produce a model using train and validation dataset ex. Stepwise Regression for a Linear/Logit/Probit model Assess performance of model using “test” dataset Implement the selected Model Chris
23 Monitoring Results Traditionally underwrite most people after the model is initially implemented Allows one to make sure the model doesn’t open the company up to adverse selection Can check assumptions on case-by-case basis It is important to monitor results after the model is implemented Important to fix issues in the model as quickly as possible Chris
24 Risks and Challenges Legality; uncertain what is legal under insurance regulations Ethics and social constraints Cost/difficulty of obtaining variables Jess ethics makes shit tough new legislation to allow web service providers to sell information - what is acceptable is always changing so it may be hard to tell what people are okay with
25 Conclusion Predictive Modeling could allow insurance companies to…Streamline underwriting process Prevent human error Improve Marketing Variables that we believe may be useful in a model… Age, gender, profession, policy limits, attained age Credit score, Motor Vehicle records, Pharmaceutical records Casino Gambling Propensity score How you pay (direct vs. auto) Market Competition, Product Flexibility and Uniqueness Geographical Mortality Wearable Technology Renting vs. Owning a Home Churn Scores Referred by Family Where you purchase (online vs. broker) Jess
26 Questions? Final ThoughtsFor any further questions, feel free to contact… Thank you to Prudential and the UConn Goldenson Center for this opportunity! Jess
27 References [A] Batty, Mike, Arun Tripathi, Alice Kroll, Chen-sheng Peter Wu, David Moore, Chris Stehno, Jim Guszcza, and Mitch Katcher. "Predictive Modeling for Life Insurance." Deloitte. Deloitte Consulting LLP, Apr Web. 10 Mar [B] "Climate Effects on Health." Centers for Disease Control and Prevention. Centers for Disease Control and Prevention, 26 July Web. 10 Mar [C] Dixon, Pam, and Robert Gellan. "The Scoring of America: How Secret Consumer Scores Threaten Your Privacy and Your Future." World Privacy Forum (2014): n. pag. 2 Apr Web. 10 Mar [D] Gallup, Inc. "Insurance Companies Have a Big Problem With Millennials." Gallup.com. N.p., 05 Mar Web. 10 Mar [E] Harker, Patrick T., and Stavros A. Zenios. "Performance of Financial Institutions." Google Books. Cambridge University Press, Web. 10 Mar [F] "How the Vitality Wellness Program Works." Vitality. N.p., n.d. Web. 11 Mar [G] Key Findings from the EY Global Consumer Insurance Survey 2014 (n.d.): n. pag. Web Mar [H] "National Climate Assessment." National Climate Assessment. N.p., n.d. Web. 10 Mar [I] Purushotham, Marianne. "U.S. INDIVIDUAL LIFE PERSISTENCY UPDATE." Soa.org. The Society of Actuaries, n.d. Web. 10 Mar [J] Schaber, Ron, Tim Hill, Derek Kueker, Jean-Marc Fix, and Chris Stehno. "Session 8: The Latest on Practical Uses of Big Data and Predictive Analytics." Soa.org. The Society of Actuaries, 4 Aug Web. 10 Mar [K] Sharps, Kevin, David Hitsky, Stacy Hodgins, and Chin Ma. "Life Insurance Consumer Purchase Behavior." Deloitte (n.d.): Deloitte. Deloitte. Web. 10 Mar [L] Sijbrands, Eric J. G., Erik Tornij, and Sietske J. Homsma. "Mortality Risk Prediction by an Insurance Company and Long-Term Follow-Up of 62,000 Men." PLoS ONE. Public Library of Science, 6 May Web. 10 Mar [M] "2016 Life Insurance Awareness Month." LIMRA (n.d.): n. pag. Web. 10 Mar
28 Bibliography https://www.the-digital-insurer.com/wp-content/uploads/2013/12/Predictive-Modeling.jpg https://www.erictyson.com/custom/rent-vs-buy7-436x242.jpg https://nebula.wsimg.com/3d16e a7f247abee067a45245?AccessKeyId=B0A9147ED0ABA5832F78&disposition=0&alloworigin=1 https://blogs-images.forbes.com/benkerschberg/files/2014/09/predictive-analytics-300x200.jpg?width=960 https://www.skipprichard.com/wp-content/uploads/2015/09/bigstock-D-Knob-Maximize-Efficiency x443.jpg https://mobilemarketingwatch.com/wp-content/uploads/2016/03/data-1.jpg https://www.cdc.gov/heartdisease/images/heart_disease_deaths.png https://taxistartup.com/wp-content/uploads/2013/06/referral-connection.jpg https://shawndriscoll.com/wp-content/uploads/2014/04/uniqueness.jpg