Physics

Quant Finance

Statistics

Computer Programming

Nerd

The Last Six Years…

• Lapse Modelling – Survival analysis
• Geospatial Modelling – PostgreSQL / PostGIS / Clustering
• Mortality Swaps – MonteCarlo simulation
• Underwriting Fraud – Bayesian networks
• Data Generation – Life/Motor/Captive/Employer/Corporate
• Car Insurance Pricing – GLMs / MonteCarlo simulation
• Email Analysis – NLP / LDA
• Property Insurance Loss Costing – Stan / Boosting / Dataviz
• Business Planning – Data engineering / MonteCarlo simulation
• Portfolio Profitability Analysis – Visualisation / MonteCarlo simulation
• Risk Evaluation of a Fixed-Income Portfolio – Data engineering

Dirty Little Secret

You (Probably) Do Not Have a Big Data / AI Problem

Sizes of Data

Hadley Wickham – Interview with a Data Scientist (Peadar Coyle)

Three Data Sizes

In-Memory

On-Disk

Multi-Disk

Three Problem Sizes

Sampleable Problems

Parallelisable Problems

Irreducible Problems

Sampleable Problems

\(\sim\) 90% of problems

Parallelisable Problems

\(\sim\) 9% of problems

Irreducible Problems

\(\sim\) 1% of problems

Various definitions

Generally bigger than one machine

Cynical Definition

“Big Data” is a marketing term used by salespeople to extract large sums of money from companies with a severe case of FOMO.

Circular Definition

A big data problem is one where big data solutions work best

Practical Definition

Software infrastructure to assist in the processing and use of large datasets

Most classic insurance problems are small data

Actuarial Problems

Pricing

Reserving

Capital Modelling

Loss-Costing DNF Business

5-10k Policies

SoV data can be large

Claim data poor quality

Estimating Claims Reserves

Estimate total claim amounts

Data sparse

Chain Ladder often ineffective

ICT Pricing

Exposure data aggregated

Partial view of claims (censored)

Cost of Guarantees

With-profit Funds

Quant finance

MonteCarlo simulation

Estimating Lapse Risk

Customer segmentation

Survival analysis

Other Problems

Less traditionally ‘actuarial’

Marketing Campaigns

Limited marketing budget

Optimise call list

Underwriting Fraud

Most data missing

Find fraudulent or misleading applications

Internal Operations

Communications data, emails

Identify bottlenecks

Improve customer turnaround

One Big Prediction

Things will not stay the same…

… and the industry will fight this tooth and nail!

(Two! Two big predictions)

Suggested Learning

• SQL
• Linear Algebra
• Statistics
• Computer Programming
• Source Control
• Linear Models
• Tree-based Models

Some Books

R for Data Science - Wickham & Grolemund

https://r4ds.had.co.nz/

Introduction to Probability - Grinstead & Snell

https://math.dartmouth.edu/~prob/prob/prob.pdf

Suggested Activities

Where to Learn

Help everyone you can!

Get in Touch

mcooney@describedata.com

GitHub: http://www.github.com/kaybenleroll