Many thanks to Jim Guszcza (Predictive Analytics lead at Deloitte Consulting and Assistant Professor at UW-Madison) who gave a great webinar presentation yesterday on actuarial analysis with R.
Jim's demo (starting at the 20 minute mark in the video replay below) is a great way to get a sense of how R is used for exploratory data analysis and modeling, with a live examples of fitting a mixute distribution to bimodal claims data, and calculating loss reserves using Poisson regression.
Many actuaries use Excel to make calculations like these, but Jim makes a great point point about the benefits of programming with data instead of spreadsheet cells at the 32:00 mark:
Just one simple line of [R] code that would work just as well for a 100-by-100 loss triangle as it would for a 10-by-10 triangle. No hidden cells in the spreadsheet, no risk of spreadsheet error. It's a little bit of code you could look at in one screen, it's replicable … and this does all the work that a spreadsheet would do.
Also, check Jim's final case study at the 47-minute mark for a sneak preview of the next version of Revolution R Enterprise — big-data Generalized Linear Models. He uses the Allstate Claim Prediction Challenge data (from a recent Kaggle competition) to fit a Tweedie model to 13 million records of claim data. (The Tweedie distribution is often used to model insurance claims, where many claims are exactly zero, and non-zero claims follow a continuous Gamma-like distribution.) Using the forthcoming rxGLM function, he fit the model to this large data set in just over two minutes (140.22 seconds) using a single quad-core PC.
You can download the slides from Jim's presentation and an a replay for offline viewing at the link below.
Revolution Analytics Webinars: Actuarial Analytics in R