Presented By O’Reilly and Cloudera
Make Data Work
21–22 May 2018: Training
22–24 May 2018: Tutorials & Conference
London, UK

Risk sharing pools: winning zero-sum games through machine learning

Baiju Devani (Aviva Canada)
12:0512:45 Wednesday, 23 May 2018
Data science and machine learning
Location: Capital Suite 10/11 Level: Intermediate

Who is this presentation for?

data scientists, actuaries, business executives

Prerequisite knowledge

Since the problem and the solution will be introduced quite thoroughly we expect the audience to only have an intermediate level understanding of machine learning.

What you'll learn

* Understanding of limitation of classical models in insurance industry and generally the problems around risk-sharing pools * application of methods such as elastic nets, gradient boosted trees and other methods in an applied case * how businesses can realize bottom-line increases through use of data sciences

Description

Risk sharing pools allow insurers to get rid of risks they are forced to insure in highly regulated markets. Insurers thus cede both the risk and its premium. Since all participants in the system are trying to cede their worst possible risks, this becomes a zero-sum game i.e. more sophisticated models win at the expense of less sophisticated models. We divide this talk into three parts:

  • Problem description: we first describe what risk-sharing pools are and why these are zero-sum games. The objective is not for the audience to get a nuanced view of the problem we are trying to solve.
  • Traditional approaches and challenges: traditionally the insurance industry is heavily reliant on generalized linear models for such problems. We describe problems of overfitting and generalization common with use of such models. But especially so given the availability of a large number of attributes (dimensions) typical in such datasets and the variability in the modelling expertise within the industry.
  • Our approach: we present our approach of using an ensemble of methods including elastic nets and xgBoost to solve the problem and showcase results that not only outperform traditional methods but allow us to adds tens of millions of dollars to our bottom line. This is an example of winning through machine learning in the insurance industry.
Photo of Baiju Devani

Baiju Devani

Aviva Canada

BIO BAIJU – Vice President Enterprise Analytics

At Aviva Canada, Baiju leads a team of data scientists in application of analytics across all aspects of the insurance business: from core insurance activities such as pricing and risk selection to development of cutting edge robotic processes and application of machine learning algorithms to areas such as claims processing and optimizing client acquisition.

Prior to Aviva, Baiju led the analytics group at IIROC, an entity overseeing Canadian equity & fixed-income markets and ingesting 600M+ data points daily and in realtime. The analytics group at IIROC guided decision making in todays machine driven (algorithmic) markets and embedded machine learning and other algorithmic surveillance capabilities for market oversight. In addition Baiju spearheaded IIROC’s primary research into market-structure issues as well emerging technologies such as Blockchains.

Baiju was also part of the early team at OANDA, a FinTech startup that disrupted retail foreign-exchange markets, and that attracted one of the largest funding rounds in Canada of $100M from leading silicon valley venture firms. Baiju was responsible for building and leading the data engineering, data sciences and growth teams at OANDA. He is also a founder of Fstream, a SaaS provider for ingesting and analyzing high-frequency streaming data.

Baiju has a BSc and MSc in Computer Science from Queen’s university and developed his data chops working on large biological datasets as part of his graduate work and later at the Ontario Cancer Institute.

Leave a Comment or Question

Help us make this conference the best it can be for you. Have questions you'd like this speaker to address? Suggestions for issues that deserve extra attention? Feedback that you'd like to share with the speaker and other attendees?

Join the conversation here (requires login)