Description: Internal modernization initiatives and external changes of the financial reporting regulations (e.g., LDTI, IFRS 17) increase the amount of validation activities for actuarial models. In addition, the use of predictive analytics in actuarial modeling requires a different set of validation skillsets/techniques in model validation. These internal and external forces call for a structured, streamlined and robust model validation practice.

Description: Model runtime is a common challenge for actuaries in the life and annuity space. In particular, applications such as pricing and forecasting may require the use of nested-stochastic calculations, which may require extensive computing resources or use of rough approximations. There are also other areas where runtime may be a challenge, such as projecting certain reserves or layering solving algorithms within actuarial calculations. Recently, runtime challenges have been exacerbated by the development of increasingly complex insurance products and changes in reporting frameworks, such as IFRS 17, LDTI, PBR, and Solvency II. Many approaches have been explored by actuaries in the past to address model runtime challenges. To further this research, the SOA has recently published the paper Predictive Analytics and Machine Learning - Practical Applications for Actuarial Modeling. The authors of the paper will share their insights on innovative techniques using the implementation of artificial intelligence and machine learning (AIML). Come examine an introduction into AIML, review how AIML can be used to circumvent nested stochastic and other runtime challenges, explore case studies, review how costs and benefits can be measured and discuss model risk management considerations. Key learning objectives include: - Understanding the fundamental concepts of AIM - Understanding applications of AIML for actuarial modeling - Learning about the costs and benefits - Learning how AIML can be applied in practice to improve model runtime

Description: Analyzing large quantities of raw complex data is often a challenge in the health care industry. During this webcast, there will be a demonstration of how unsupervised machine learning techniques can be applied to such data to help bring useful insights. This will include examples that use unsupervised learning techniques spanning dimensionality reduction, clustering and anomaly detection.

Description: "Risk adjustment (RA) is a key component for CMS Medicare Advantage program and California Medicaid Managed Care program. While both RA programs follow generally accepted basic principles at a high level, their underlying methodologies and assumptions are also quite different. Due to recent events such as COVID-19 Public Health Emergency and Medicaid redetermination delay and restart, it adds complexity to Risk Adjustment. This is a hot topic in the industry and has significant impact to health plans. Keeping current with risk adjustment program changes and understanding the commonalities and differences of these two government risk adjustment programs is important to our health actuaries. Compare these two risk adjustment programs so you can learn key elements of a risk adjustment program, similarities and differences in methodology and assumptions , as well as special considerations due to serving different populations."

Description: "Explore a general introduction to the opportunity of applying a relatively newer area in machine learning called uplift modeling, which has been successfully used in retail (Wayfair etc). The concepts are parallel to the intervention programs in health care, where we are interested in measuring the treatment effects and ultimately directing the right members to the most effective programs, driving the highest value for the programs both from a financial and health outcomes perspective. Uplift modeling can also be viewed in context of causal modeling where we are interested in assigning cause to a treatment relative to a member not receiving the treatment. Check out an introduction on the concept, seehow it relates and builds on existing work, and use a demo to showcase the potential of integrating this ML approach to current work."