Actuarial Review Return to Main Page

Brainstorms


The Art of Algorithms

by Stephen W. Philbrick

It's an election year, so many subjects that would barely merit media coverage suddenly get front-page treatment because some politician decides that banning something (or perhaps mandating it) will make the world a better place. Outsourcing is garnering many of these headlines this year. Offshoring —the "hot potato" aspect—is dubbed statistically insignificant by some economists. Nevertheless, it doesn't seem so insignificant if it happens to you. Many in the tech sector and call center business have reason to be concerned.

There's less concern among actuaries. I wondered why, which led me to think about the nature of actuarial work versus other occupations.

A couple of decades ago one could make a long list of reasons why an actuarial function couldn't be outsourced to, say, Bangalore. It's a long way to travel for a rate hearing. Good reserving actuaries need to get their hands "dirty" reviewing claim files. Even triangle "manipulation" required an understanding of the underlying data, access to related data, and discussions with many people cognizant of the insurance process.

Over time, some of the reasons for being local have disappeared. The advent of PC's, standardization of software, electronic capture of claim files, e-mail, the Internet, and decreases in long distance rates have gradually whittled away at some of the rationale for the actuary to be physically local. Presence at rate hearings is still on the list, but maybe Web cams will change that.

Many of these changes have transformed the tech sector and call center industry, yet these innovations haven't prompted management to relocate the actuarial function to other countries. So what is it that is different about the actuarial function?

One possibility is the nature of a call center operation can be reduced to a well-defined algorithm. The steps required to figure out why your PC stopped working properly after you installed the latest version of Doom may be numerous and complex, but they are algorithmic. They can be logically mapped out as trees, with scores of nested branches. Once mapped out, the process itself may not be much fun for either the caller or the callee, but the odds are high that the process will reach a unique node on a tree. It is important to distinguish the skill level required to maneuver the algorithm from the skill level required to create the algorithm. For any given problem, we might expect an inverse relationship between the skill level used to create the algorithm and the skill level required to navigate the resulting algorithm.

Actuarial problems aren't as easily adapted to the algorithmic model. One can think of specific examples—if the paid indication significantly exceeds the incurred indication for all years, look to see if the closing rate has accelerated. But the attempt to create a few examples illustrates the difficulty of reducing all of the reserving or pricing process to an algorithm.

I doubt this will surprise anyone. The phrase, "more art than science" is ubiquitous in actuarial circles, and supports the notion that converting actuarial work into an algorithm is close to impossible. I've often wondered if this phrase was more of a crutch than an insight.

I decided to use one of my favorite tools, a Google search for a phrase, to quantify my expectations. The answers did not turn out as expected. A search on the phrase "more art than science" turned up more hits than the phrase "more science than art," when each were coupled with the word "actuarial." However, simply searching on the two phrases without the term "actuarial," turned up more hits for "more science than art" than "more art than science." This surprised me.

Maybe we've been relying on this crutch too long.

Some of the hits for "more science than art" involved pricing. The reference wasn't specific to insurance pricing, but if the professionals in other areas are now contending that pricing can be more science than art, is there any good reason insurance pricing should be less amenable to scientific rigor?

I started this discussion in the context of outsourcing. A naysayer might object that making our work more rigorous might well make it more susceptible to description by algorithm, and thus more likely to be shipped to another country. I don't share this concern for several reasons, one of which is even if true, ignoring it won't stop it from occurring.

My major area of concern is that we may have been too quick to assume that actuarial work is inherently too subjective to ever allow it to be formalized. For this reason, I suspect few have even attempted to formalize it. While I am certain that we are still many years from creating anything that would rival a call center algorithm, I think the effort would be worthwhile. Like creating a mission statement, the value may not be the end product, but the insights gained by those involved in the process.

Click here to write a Letter to the Editors