Event Details
About This Event
In traditional GLM modelling, categorical features are often treated with one-hot encoding. However, this method of encoding removes any ordering information inherent in the data. With accurate GLM, we propose to utilize an alternative encoding approach which allows us to preserve the ordering information, without requiring that the variable be converted to numeric form. In this presentation, we will introduce the accurate GLM approach to modelling, discuss what motivated the approach, and illustrate the approach through simulated data modelling exercises and a case study.
Key Messages
GLM as a model structure is a known and often used tool for a pricing actuary. Despite this maturity, it is good to be reminded we can design algorithms and approaches to enrich the GLM modeling process. Accurate GLM, an approach conceived by Hirokazu Iwasawa and introduced in the 2021 Hachemeister Prize winning co-authored paper "AGLM: A Hybrid Modeling Method of GLM and Data Science Techniques", showcases this by offering a way to handle the shaping of the relationship between the ordered data and response. We hope the practical illustrations in this presentation will both introduce a new tool for the actuaries and spark a call to look at the modeling process with fresh eyes.
Learning Objectives
- Become familiar with the concepts and motivations behind the accurate GLM technique
- Learn how to set up data and carry out modelling using the accurate GLM technique
This webinar is complimentary to non-North American audiences who are part of the CAS and the CIA. The webinar will be conducted over GotoWebinar. The link to enter the webinar will be sent on successful registration. Please make sure to test your system.
Gary Wang, Willis Towers Watson
For over 15 years, Gary has provided actuarial consulting services to clients, with a heavy focus on predictive analytics. He has worked extensively on the application of advanced statistical modelling techniques to the insurance process. Gary has led, developed, and executed complex data processing and explorations taking client data to model ready extractions. These exercises included creating new variables from the raw company data as well as the joining, reviewing, and engineering of explanatory variables from external data sources. In addition to developing the models and analysing results, Gary has collaborated closely with clients to incorporate business considerations to develop implemented plans that properly balance the model indications with company and client domain knowledge and business constraints.
Hirokazu Iwasawa, Waseda University
Hirokazu Iwasawa (aka Iwahiro) is a central figure of education, dissemination, and research of data science in the actuarial field in Japan. He has been acting as a mentor of many data science related projects in IAJ's Data Science Related Basic Research Working Group and ASTIN Related Study Group. As an eminent educator, he gives regular lectures at IAJ as well as at several universities including Waseda University as a Guest Professor. He has published 20+ books among which eight books are single authored. They are on probability, statistics, math puzzles, non-life insurance math, predictive modeling, etc. He is the originator of the AGLM technique, a co-authored paper on which received the 2021 Hachemeister Prize.
