SOA Predictive Analytics Seminar – Malaysia 27 Aug. 2018 | Kuala Lumpur, Malaysia Session 2 Predictive Analytics in Policyholder Behavior Eileen Burns, FSA, MAAA David Wang, FSA, FIA, MAAA
Predictive Analytics in Policyholder Behavior Eileen Burns, FSA, MAAA Principal & Consulting Actuary Milliman Inc. David Wang, FIA, FSA, MAAA Principal & Consulting Actuary Milliman Inc. 27 th August 2018
Agen enda Ei Eileen een B Burns, F FSA, MA MAAA David W Wang, F , FIA, F , FSA, M , MAAA Principal & Consulting Actuary Principal & Consulting Actuary Seattle Seattle Current state in life and Eileen.Burns@milliman.com David.Wang@milliman.com annuity Education ion a and Qu Qual alif ific ication ions Education ion a and Qu Qual alif ific ication ions Examples of where University of Washington, University of California at Berkeley, Quantitative Ecology and Resource HAAS School of Business (2005 - predictive analytics helps Management (2008 - 2011) 2006) Masters MFE, Financial Engineering Implication on assumption setting process Lawrence University (1998 - 2002) Nanyang Technological University BA, Mathematics (1994 - 1998) Interesting applications B. Business Current r t responsibilities es Principal on Milliman’s data Current r t responsibilities es analytics team Co-leads Milliman’s team specializing in applying data Product manager for Recon, a analytics to assist the life and Milliman predictive analytics and annuity industry in the United data product targeted at States. enhancing experience analysis Co-leads Milliman life consulting Vice-chair of SOA Predictive practice in Seattle Analytics and Futurism section 2
Current State in Life and Annuity
What is Predictive Analytics and Predictive Modeling Predictive analytics uses many techniques from data mining, statistics, modeling, machine learning, and artificial intelligence to analyze current data to make predictions about future. Predictive modeling is a process used in predictive analytics to create a statistical model of future behavior.( Google Search ) 4
Policyholder Behavior Modeling: Progression of States Traditional one-way actuarial techniques to estimate Traditional S State behavior by age/duration and limited number of other characteristics using experience where it exists Primarily macro-oriented… little use of detailed information on policyholder characteristics Judgment and guesswork where experience does not exist Next-generation experience studies using policyholder Pred edictive M e Model eling longitudinal data. State Use much wider set of explanatory variables readily available to company – Internal data (Product features, distribution channel, policyholder and contract characteristics) – Macro data (Economic data, financial market conditions) More sophisticated analysis techniques to find non- linear, multivariate effects, complex interactions Employ external consumer/financial/health and Big Da Data S State big/unstructured data sources in a full Predictive Analytics framework. Develop individual policyholder profiles 5
Applications of Predictive Analytics in Life and Annuity Actuarial Data Analytics 6
Examples of Where Predictive Analytics Helps
Improve Predictions Overall Improvement in Predictions Relative impact from predictors 8
Test Hypothesis and Answer Question • Is there a difference in sensitivity to crediting spread among distribution channels? • Does the MVA effectively eliminate sensitivity to crediting spread? 9
Identify drivers ¢ Previous behavior – e.g. withdrawal al b behavior People – demographics and distribution channel Product design – MVA, surrender charge structure, guaranteed minimum Macroeconomics – market rates, unem employm ymen ent 10
Confidence Intervals Model predictions and confidence bands versus actual experience Baseline model Full model 1.40% 1.40% 1.20% 1.20% 1.00% 1.00% 0.80% 0.80% 0.60% 0.60% Qu Quarterly lapse rates 0.40% 0.40% Actual lapse rate 0.20% 0.20% Predicted lapse rate 95% Confidence interval 0.00% 0.00% 2007 Q2 2007 Q4 2008 Q2 2008 Q4 2009 Q2 2009 Q4 2010 Q2 2010 Q4 2011 Q2 2011 Q4 2012 Q2 2012 Q4 2013 Q2 2007 2007 Q4 2008 Q2 2008 Q4 2009 Q2 2009 Q4 2010 Q2 2010 Q4 2011 Q2 2011 Q4 2012 Q2 2012 Q4 2013 Q2 11
Implication on Assumption Setting Process
Typical predictive modeling process 13
Era of Big Data has come, but Life Insurers Need to Catch Up! Little systematic collection and storage of data Legacy system inadequate for new data Challenges the life analytics insurance industry faces Limited data to differentiate customer Silos still exist 14
Data visualization is more than just better pictures More data, more information, more dimensions, calls for better visualization Makes traditional date reporting inefficient Provides guidance and tips on how predictive models should be built 15
Bring predictive model in assumption setting process Impl plem emen entation Com ommunic icatio ion Can we model all the predictive How do actuaries convince drivers in the actuarial cash themselves and management that flow projection? PM is needed? If not, how do we make How do actuaries communicate compromise and recognize the model results to senior loss of accuracy. management? Assumption Validatio ion Control & & Governa nanc nce Setting How is the goodness of fit over Predictive modeling requires new different dimensions? controls & governance. How do we develop appropriate standards? How are we comfortable with confidence intervals? Who is qualified to review and sign off? Domain knowledge is essential to What type of documentation should be make sense of results. retained? 16
Some Interesting Applications
Evaluation of behavioral tail risk Types of lapse tail risk Diffusion Drift Dr Extr trem eme E Even vent Risk that some Risk that best Risk that unprecedented estimate lapse estimates of the events may rates vary under entire lapse impact lapse in an different market function are off extreme way conditions Captured by Resort to some Captured by a simulation of manner of dynamic lapse lapse behaviour judgement call component using predictive model 18
Lapse behavior simulation 19
Lapse behavior simulation – Determine best estimate ITM p 225% 1.8% 175% 4.7% 125% 11.9% 75% 26.9% 25% 50.0% 20
Lapse behavior simulation – Simulating the risk of model misestimation Best Estimate ε = {-0.2, -0.1} ε = {0.2, 0.1} ε( i) ε ε ITM p ITM P ITM p 0 225% 1.8% -0.2, -0.1 225% 1.2% 0.2, 0.1 225% 2.8% 0 175% 4.7% -0.2, -0.1 175% 3.3% 0.2, 0.1 175% 6.8% 0 125% 11.9% -0.2, -0.1 125% 8.9% 0.2, 0.1 125% 15.8% 0 75% 26.9% -0.2, -0.1 75% 21.8% 0.2, 0.1 75% 32.6% -0.2, -0.1 25% 44.4% 0 25% 50.0% 0.2, 0.1 25% 55.6% 21
Customer segmentation Identify segments of policyholders Segment specific behavior modeling reveals how people use insurance differently Data-driven segments identify policyholders likely to behave in similar Unsegmented ways Number and defining characteristics of segments will be specific to the particular dataset Likely defining values for segments include credit score, income, home value, home mortgage loan-to-value, etc. 22
Differentiation between policyholder behavior and corresponding profitability Plots show profitability differences driven purely by behavioral difference 1 due to belonging to 2 different segments. 3 Segment 4 Help identify groups of people whose needs are 5 not served properly by 6 current product offerings and identify need for new 7 products Profitability relative to expectation 23
A new perspective of product profitability Show profitability at state level, or at county or zip code view 24
Final thoughts
Goal and elements of predictive analytics in policyholder behavior To predict (individual) policyholder behavior by applying rigorous statistical techniques to large amounts of data under the guided framework designed by subject experts Statistics Data Subject Business expertise application Individual behavior 26
Recommend
More recommend
