By Sam Gutterman
This week, reading Elphie: A Wicked Childhood, a prequel to Wicked (one of my favorite novels) by Gregory Maguire, raised several questions: Why are we so different? And for actuaries, what are the implications of these differences? In Maguire’s novels, the differences are clear and stark—Elphaba Thropp, the eventual Wicked Witch of the West who has green skin, and Nessarose, her sister, known as the Wicked Witch of the East, has no arms.
These types of differences are actuaries’ stock-in-trade. If we were all the same, there would be far fewer jobs for actuaries, as risk classification systems and a lot of modeling would be far easier. Fortunately, we aren’t.
The first step in an actuarial assessment of a risk-pooling facility, such as an insurer, a pension program, or a bank, is to identify the key drivers of expected outcomes, including death, illness, property damage, loan default, or investment return.
It is easiest if experience data containing these factors are available for conducting a proper analysis, identifying trends, and projecting expected changes in future conditions. If experience by these factors is not available, other means will be needed. Possibly, predictive analytics can be used. These factors may be pretty obvious. For example, age, pre-existing health conditions, educational attainment, location, and income are often key.
Even with all relevant data, questions of fairness can arise in applications such as the design of a risk classification system. Sensitive or boundary conditions may be involved, especially where data are lacking or are fuzzy. Indeed, a factor demonstrating a high association or correlation with loss may only be a symptom of or proxy for the underlying factor contributing to a claim. For example, although it is well known that married individuals or those who are Hispanic or Asian American tend to have better mortality outcomes than others, should those who are unmarried or are not part of these ethnic groups be charged more for life insurance or receive less payouts for lifetime payout annuities?
An algorithmic or artificial intelligence approach may not be able to select all ethically or legally sensitive factors. We may be challenged to apply actuarial fairness—differentiating by expected cost—a practice I have argued for in some cases, with its advantages and disadvantages, which I first encountered during my initial experience on an actuarial risk classification committee over 40 years ago. If we are not careful, correlations between variables can perpetuate disparities, sometimes across generations; causation is always the gold standard of relationships.
Greater granularity may lead to better conclusions. For example, age-sex-race-income may produce more helpful experience than just age-sex-race. Another example is that insured mortality experience has lower mortality than the general population, as insureds often have greater income, education, and access to health care. The mortality of those with higher income has improved significantly over the last several decades, while the mortality of those with lower income has stagnated.
Actuaries and society have sometimes struggled to identify acceptable risk classes, as removing sensitive variables does not eliminate their influence. Simple rules promoting equity are possible, but may also be difficult to apply. No single approach is satisfactory in all cases.
Reflecting certain attributes may be seen as unfair discrimination, posing an ongoing challenge for actuaries to best consider and reflect differences in risk. Selection bias and anti-selection may remain. I don’t pretend to be able to answer or even address all of these questions in this column—it remains a continuing public policy and actuarial issue.
Risk classification is not the only problem associated with these differences, evident in the current debate regarding diversity, equity, and inclusion (DEI). One side suggests that members of a disadvantaged group should be given an equal opportunity to have a fair and balanced chance in a competition, possibly to help overcome past disadvantages, such as limited exposure to educational opportunities, even though many in the group may not need them. In contrast, the other side wants all to be treated equally, as the perception of unwarranted help might overlook merit. The study of differences between groups can also be tricky because of disparities within any group.
I appreciate those with diverse views and backgrounds. But why are we different? I don’t have an answer; sometimes, we just have to do our best and move to the next problem. In some cases, those with fundamental differences should be allowed to excel within their groups, for example, having males and females competing separately in certain sports. Other examples abound.
Everyone is in favor of fairness, but what does fair mean? Once, many years ago, I was told never to use that word, as it can mean different things to different people, and what one person views as fair, another may not. Nevertheless, we need to continue striving to achieve enhanced equity among our diverse population. Let’s take advantage of these differences rather than fight over them. The struggle will continue. It is worth it. Viva la difference!
Sam Gutterman is chairperson of the Social Security Committee and member of the Retirement Practice Council.