Regression Modeling with Actuarial and Financial Applications

International Series on Actuarial Science

by Edward W. Frees

Statistical techniques can be used to address new situations. This is important in a rapidly evolving risk management and financial world. Analysts with a strong statistical background understand that a large data set can represent a treasure trove of information to be mined and can yield a strong competitive advantage. This book provides budding actuaries and financial analysts with a foundation in multiple regression and time series. Readers will learn about these statistical techniques using data on the demand for insurance, lottery sales, foreign exchange rates, and other applications. Although no specific knowledge of risk management or finance is presumed, the approach introduces applications in which statistical techniques can be used to analyze real data of interest. In addition to the fundamentals, this book describes several advanced statistical topics that are particularly relevant to actuarial and financial practice, including the analysis of longitudinal, two-part (frequency/severity), and fat-tailed data. Datasets with detailed descriptions, sample statistical software scripts in "R" and "SAS," and tips on writing a statistical report, including sample projects, can be found on the book’s Web

584 pages, Paperback

First published January 1, 2009

Reviews

[Sam Wendt · Rating 4 out of 5]

But when do they kiss?

[Hailey Atkins]

When you have to study for your exam but you’re trying to hit your reading goal :/

[Chris · Rating 4 out of 5]

Professionally, I'm not in finance, nor am I an actuary. I'm a software engineer. I consider myself a practical and simple person. I don't do a lot of theory. I do something and then I observe what effect it had. Call it intuitive hypothesis testing. Call it eyeball statistics.

I did take some basic statistics at University, and I have read a bit of Shewhart and Deming, but that's the extent of my statistical training. Control charts are the most advanced tool I use in my line of work, and even then only rarely, because (a) they answer only a small set of questions, and (b) I haven't figured out a way to make them work with the fat tailed reality of software development.

I was under the impression that if you were going to do anything meaningful with statistics, you had to bring out the big guns and do very advanced analysis.

I don't know where I got the idea, but I was convinced the most simple statistical tools are no better than intuition and eyeballing plots.

Then I started stumbling over a growing mountain of evidence that even the most naïve of statistical techniques outperform, well, intuition and eyeballing plots.

This is a complete 180 in my world view. Highly trained professionals perform much worse than plain old ordinary least squares linear regression. This is a result that has been replicated again and again, in all sorts of fields. It's really bad. For a brief summary, see the classic paper Clinical Versus Actuarial Judgment. For more modern review, maybe Superforecasters or Expert Political Judgment is a start.

So here we are. I've probably been performing worse than linear regression my whole life. Linear regression, for God's sake! The stupidest and dumbest of models. There are, of course, several reasons for this. An obvious one is that humans are biased and noisy and make mistakes.

A much more important reason, I think, is that phrasing your decision in the form of a statistical model means you have to propose your hypothesis up front (!), and it will be painfully obvious when it is falsified, and give you reason to re-think your position. When you eyeball statistics you unconsciously and continuously adapt your hypothesis to match what you see, all the while both overfitting it, and coming up with causal links that aren't there.

Okay, so why care? It's not immediately obvious to me that it is worth investing effort in getting better at explaining and predicting things. Sure, it sounds good, but at what cost? Is something any engineer should ask themselves.

But prediction is literally about seeing into the future. It's a slice of clairvoyance. It is insanely valuable in almost any line of work.

In light of this, the least I can do is brush up on these basic statistical techniques, and use them as often as I can. I don't know which book is best for that, but this one is good. It goes through the theory in sufficient detail for it to stick in my memory, it has plenty of practical examples and exercises to aid retention.

It surveys the most important simple tools of the trade, with references both to more advanced techniques and more elaborate works on the simple tools.

This book is exactly what someone in my position needs.