Session 20 PD, Senior Management's Wander Through the Model Efficiency Countryside Moderator: Anthony Dardis, FSA, CERA, FIA, MAAA Presenters: Mark A. Davis, FSA, MAAA Nazir Valani, FSA, FCIA, MAAA SOA Antitrust Disclaimer SOA Presentation Disclaimer
Senior Management s Wander Through The Model Efficiency Countryside 2016 SOA Annual Meeting Session 20 PD Mark A. Davis, FSA, MAAA October 24, 2016 1
SOA Competency Framework COMPETENCY Communication Professional Values External Forces & Industry Knowledge Leadership Relationship Management & Interpersonal Collaboration Technical Skills & Analytical Problem Solving Strategic Insight & Integration Results-Oriented Solutions DEFINITION Demonstrating the listening, writing and speaking skills required to effectively address diverse technical and nontechnical audiences in both formal and informal settings Adhering to standards of professional conduct and practice where all business interactions are based on a foundation of integrity, honesty and impartiality Identifying and incorporating the implications of economic, social, regulatory, geo-political and business changes into the design and delivery of actuarial solutions Initiating, innovating, inspiring, creating or otherwise acting to influence others regardless of level or role toward a common goal Creating mutually beneficial relationships and work processes toward a common goal Applying the actuarial knowledge, skills and judgment required to provide value-added services Anticipating trends and strategically aligning actuarial practice with broader organizational business goals Providing effective problem solving that addresses relevant interests and needs 2
Model Efficiency Review of Literature & Presentations Early Focus on Technical Topics impacting run time Scenario Reduction techniques Model Cell Reduction such as Clustering Some focus more recently on Technology Grids, Cloud More recent attention to Process Efficiency and Model Governance But very little concerning Communications My Topic Today Communications to Senior Management and the Board 3
Sr. Mgmt. Interest Low Low Low/Med Model Efficiency Countryside Technology Increase Speed Machines/Grids Clouds/GPUs(?) Actuarial Practice Reduce # Calculations Scenario/Cell Reduction Curve Fitting, Sampling Techniques, Replicating Portfolios High Organizational & Structural Efficiency High Actuarial Transformation or Modernization 4
Model Efficiency Countryside Why do Sr. Mgmt. and Board have high interest in Organizational/Structural Efficiency and Modernization? For most, it s more in their sweet spot They can relate familiar territory & language they understand Stronger bottom line connection For many Board Members and Sr. Mgmt., Actuarial remains an unknown destination a mystery o Some would rather keep it that way Our challenge is to take the mystery out of Actuarial when communicating upwards and change the landscape! 5
Communicating Upwards My Qualifications Good fortune to have worked closely with top actuaries Worked 8 years with actuary who been in C-Suite positions since 2002 Worked on several assignments with a leading consultant in one of the major firms who rose to the very top management Another mentor is Chief Actuary at a very large company Presentations made to Boards of Directors Served on multiple Boards of Directors Current role (8 years) largely devoted to explaining complicated financial matters to Sr. Mgmt. Taught Calculus in college to Business Majors Also tutored football team Coached girls basketball 5-8 th grade 6
VA Hedging Presentation to Colleagues Scenarios reduced from 10,000 to 100 (99% reduction) Applied Dardis/Valani Algorithm using a tolerance of ε 2 and µ = log e (2.95) 90% likelihood that results are within 0.5% of the 10,000 scenarios Model Cells reduced by 66.67% via Reynolds Cluster Theorem and employing a slight refinement to the Howes Corollary Resulting increase to ê of 1.89% deemed acceptable Overall runtime has been reduced by 99.9% Overnight model run has been reduced to 43 seconds We have freed up 8,000 cores on the Grid Rebalancing now possible several times per day Bottom Line permits hedging in near-real time Note: Some poetic license and liberties have been taken above with the names and parameters of the techniques utilized and the improvements obtained. The point is that the work performed is supposed to represent the best thing that has happened since sliced bread (ok, since the iphone). 7
VA Hedging Example Communicating Upward to Sr. Mgmt. and/or Board Objective: Improve effectiveness of VA Hedge Program Background: Inefficient modeling currently permits only a mid-morning rebalance; effectiveness suffers Solution: Leading practice Model Efficiency techniques utilized to reduce model runtime dramatically Hedge positions can now be evaluated in near real-time Results: Hedge effectiveness projected to improve from 70% to over 90% Hedge transaction costs expected to increase 10% or about $200k per quarter More than offset by reduction in internal spend (headcount, Grid) projected to be $325k per quarter Conclusion: Big win-win for VA volatility reduced by two-thirds along with small reduction to overall costs 8
Actuarial Communications Challenges Subject matter can be highly technical even to an actuary! Subject matter becomes near and dear to one s heart! Easy to forgot how you struggled with it at the start Your audience likely to have same struggles you did Preparation time can be short in normal course of work Not a lot of opportunities to make formal presentations in many actuarial roles Some in Sr. Mgmt. or Board may have a natural skepticism to many things Actuarial Can be a tough crowd 9
Communicating Upwards to C-Suite and Board Know your audience if at all possible Some research may help, especially with Board Members Try to walk in their shoes Start at the beginning This is really the key actuaries like to start in the middle and typically it goes downhill from there Provide some background, give context, set stage for why you are here Check the actuarial jargon at the door may take some effort Try to leave the details out Can include if questions arise, or follow up later if requested Do NOT read the slides! Have a few talking points prepared for each slide not every bullet Be cognizant of allotted time and prepare slides/remarks accordingly Too many slides is often the culprit figure 2-3 min per slide 10
Concluding Thoughts The Model Efficiency countryside can be beautiful from the Boardroom But skilled communicators are required! 11
Nazir Valani, FSA, FCIA, MAAA Wandering Through the Model Efficiency Countryside October 24, 2016
Model Efficiency 2
The Big Picture Policy Admin Fields ETL Model Efficiency Modelling Reporting 3
Leading Practice Non-model cash flows General Ledger Policy Systems ETL Tool Models ETL Tool Data Warehouse Reporting Tool Assets Web based user interface Assumption Database Assumption data Model result data Subset of policy data Accounting Data 4
Cloud Computing key considerations Data Security Data Transfer Approvals from info risk PII Substantial volumes on critical timelines Architecture design and service levels Access & encryption Production Processes Opportunity to standardize and centralize model ops. and infrastructure management Modelling Platform on Cloud Reporting Faster user experience Long term storage Modelling Platform Central access control & internal domain names Central version control Standard operations management tools
Seriatim processing se ri a tim Adverb formal taking one subject after another in regular order; point by point.
Data Integrity IT Actuary
Model Internal Efficiency: Actuarial Modeling Techniques Clean data IT Implementing controls Object naming conventions Assigning assumptions Consistent coding Documentation 8
Reporting with drill down capability (e.g. RMA & SOEs) 9
Need for Faster Runs? Nested Stochastic runs USGAAP C3 Phase 2 USStat PBR CGAAP IFRS 10
Nested Stochastic vs. LSMC SOA Aug. 2014 session 45: Model Compression Techniques by Aatman Dattani 11
Model Compression Techniques SOA Model Efficiency Study Results NOVEMBER 2011 Scenario reduction Curve Fitting Sampling Clustering Replicating Portfolio 12
Model Compression Techniques 13
Scenario reduction Representative Scenarios Select a subset of scenarios that are representative of the full scenario set based on certain characteristics of the scenarios Transfer Scenario Order Determine the ranking of scenarios based on the ranking from running a subset of policies 14
Curve Fitting Determine an underlying distribution that fits well to the measured distribution of a variable, and report using the underlying distribution 15
Sampling Pre defined rules e.g. run every 100 th policy and gross up results Importance sampling Sample more data in parts of the distribution that are more critical to the overall result 16
Clustering Mathematically locate policies whose results are close and combine them to produce a reduced scaled subset of policies that will have similar characteristics to the full inforce 17
Replicating Portfolio Use optimization to determine a reduced scaled subset of policies that will have similar characteristics to the full inforce 18
Questions? 19