Bayesian Modelling in Real Life (Olga Semenova)

Bayesian Modelling in Real Life

Abstract: This presentation documents the real-world process of solving a problem via Bayesian modelling: ways of approaching a problem, failing, trying different things and finally learning a way that works better.

It was developed as an example for newbies taking their first steps into actually writing their own model to solve their own problem.

The presentation highlights some pitfalls, both in terms of real-life data and real-life modelling limitations as well as ways of possibly avoiding at least some of those pitfalls in future modelling endeavours.

The data in this presentation is real but abstracted so I don't breach confidentiality and the problem presented is modelling the expected numbers of Policies in Force (PIF) in a year.

About Olga: I've been an insurance professional for c.10 years. I realised what I really enjoyed about the job was data science after working at a start-up and having a great experience getting stuck into using tech and data analytics to solve all sorts of problems. I went on to work for Markel International as a data scientist after the start-up and that is where I first encountered Bayesian modelling. My job has mostly been implementing models that much more experienced people have developed rather than developing them myself, hence the presentation for newbies which I wrote after having a go at developing my own model to solve a business problem for the first time a couple of years ago.