Life & Annuity Symposium Webcast Series – Modeling Track Recordings
Practical Application of Least-Squares Monte Carlo to VA Pricing
July 13, 2020
VA pricing requires stochastic-on-stochastic modeling to predict future reserves and capital requirements based on value at risk or conditional tail expectation. Usually brute-force simulation approach cannot accommodate the time constraint in a real business world. Least-Squares Monte Carlo (LSMC) built on predictive models such as GLM, Random Forest, and Gradient Boosting Machine, when designed appropriately, can significantly reduce model run time with a highly satisfactory accuracy.
The presenters will use a case study to introduce LSMC based VA pricing, and more importantly, discuss the challenges and practical solutions in a real business case.
- How do we construct a meaningful training data set in a short window?
- How do we select explanatory variables? How do we interpret and compare different models?
- How do we balance between time and accuracy?
- How do we determine when we need to recalibrate the models?
With a focus on the implementation side, this webcast intends to help practitioners improve and accelerate their VA pricing process. It can also be used to project reserves and capital requirements using inforce data.
At the conclusion of the webcast, attendees will be able to:
- Design a practical LSMC process for VA pricing
- Analyze the pros and cons of different predictive models for LSMC
- Use feature engineering to improve LSMC accuracy
- Use sensitivity tests to assess the right time for model recalibration
From Experience Data to Assumptions
July 16, 2020
How do you take experience data and turn it into a full-fledged assumption? This webcast will include two assumption development case studies; one for life and another for annuities to show how data becomes an assumption, as well as how the same data can lead to multiple assumptions that are fit for different purposes.
At the conclusion of the webcast, attendees will be able to articulate how actuaries use data to develop assumptions.
Recent Development of High Efficiency Methods for Nested Monte Carlo Simulation
July 17, 2020
Today's computational power and technology make it possible for the life insurance industry to develop highly sophisticated models for stochastic projection, which were impossible just a decade ago. Nonetheless, as more industrial practices and regulations move towards dependence on stochastic models, the demand for computational power continues to grow. While the industry continues to rely heavily on hardware innovations, trying to make brute force methods faster and more palatable, we are approaching a crossroads about how to proceed.
This webcast is intended to give an overview of most recent research development on `software' methodologies for improving computational efficiency of nested Monte Carlo simulation. Most of these methodologies have been tested with wide applicability to interest/market risk sensitive insurance and annuity products.