The Cycle
By Stephen W. Philbrick

The cycle is well known in property-casualty insurance. At first blush, it is hard to understand why it should occur. If companies estimate the expected loss for a contract, then add expenses and required return on capital, the observed loss ratios conceptually should randomly depart from the expected mean. This is decidedly not the case, as industry loss ratios tend to remain above or below the long-term average for several years at a time.

Parameter risk can explain this partially. If we think in terms of coin tosses, we can analogize the pricing process as estimating the results of coin tosses for each contract, where the underlying probability of heads is not known. If the probability of heads is higher (or lower) than one-half for all coins, the aggregate estimates will have errors tending in the same direction, until there is enough experience to estimate the true probability. Even if the

While it is tempting to point out the inherent unpredictability of losses in advance, projecting aggregate losses at the industry level is remarkably accurate.

true underlying probability is a moving target (to mix metaphors), it either is moving in a random way, in which we would expect to see more randomness in the resulting loss ratios, or it is moving in a systematic way, in which case we would expect people to figure it out, and the resulting loss ratios would also become more random.

Shocks to the system, such as unanticipated trend or changes in the legal climate, can explain why results might cluster on one side or the other for a short period of time, but these reasons seem inadequate to explain the multiyear length of the cycle. While it is tempting to point out the inherent unpredictability of losses in advance, projecting aggregate losses at the industry level is remarkably accurate. Anomalies, such as asbestos and D&O (due to securities lawsuits), exist, but aggregate casualty estimates are fairly stable.

If the industry can estimate, within a reasonable range, the losses for an upcoming year, how is it that the industry, as a whole, manages to get the pricing wrong? One possibility is that each company can estimate the total, but convinces itself that it has a smaller share of the total than it actually has. This could be because they convince themselves that they have superior underwriting, or perhaps they simply don't estimate what share they should have. The first possibility is sometimes called the Lake Wobegon effect-where all the companies view themselves as better than average. I will refer to the second as the market share effect-when a company has a larger (or smaller) share of the total market than it realizes.

One's first reaction might be that calculating market share is straightforward: divide your premiums by the industry total. I don't know that anyone has tried to compile these estimates and compare to actual, but I would expect it would come close. What I suggest is that market share should be measured by underlying exposures, some measure correlated with loss expectations, rather than premium.

This isn't a trivial exercise. In some cases, there are decent surrogates for exposure, for example, millions of truck miles (possibly by class of vehicle). In other cases, such as D&O coverage, people are still scratching their heads trying to determine a reasonable surrogate. In most cases, the exposure base will be the same as that used to rate the policy, but this doesn't have to be the case. For example, suppose you use a proprietary credit score algorithm in your pricing but don't have any way of applying that approach at the industry level. You might use one exposure measure for pricing and another for the estimate of exposure market share.

The results might be enlightening. Suppose a company calculates that its share of the industry exposures is going up by five percent, and the industry expects a five percent increase in overall losses for the line of business. Management might be able to persuade itself that certain pieces of new business may be better than average (after reminding itself that every piece of new business is someone else's reunderwriting of the book), but management may find it difficult to believe that its entire book of business is somehow ten percent better than the year before. In theory, this is no different than observing that your premium estimates are flat and knowing that you have added more business than you've lost, but the quantification of an exposure measure might prompt you to confront the issue as well as giving you a way to start looking at it.

I don't wish to pretend that failure to measure exposures is the answer to the existence of the cycle. It has many causes, several of which have been documented. However, it is my impression that, as an industry, we haven't worked hard at compiling aggregate exposures, and this may be another contributor to the cycle.