Magazine article Risk Management

Risk Management Frontiers: The Quest for More Reserve Information

Magazine article Risk Management

Risk Management Frontiers: The Quest for More Reserve Information

Article excerpt

Risk managers and other corporate decision makers are about to discover a new frontier of reserving information. This frontier marks the increasingly dynamic boundary between risk and reserves, and those who learn to master it will find new ways to improve upon yesterday's results.

In recent years, risk managers and other corporate decision makers responsible for loss reserving have been working in an information environment that is in a state of flux, but the actuarial components of this environment are only beginning to change. The actuarial world has been dominated by algorithmic methods that each project a single-point estimate of future liabilities from a set of existing data. Generally labeled "deterministic" (because results are determined by analyzing data in defined steps) these methods can be used with various combinations of assumptions to produce a "range of reasonable estimates" from which the actuary would judgmentally select the "best" estimate. Such an approach offers a spectrum of fixed points--any one of which can serve as a target number for booking a loss reserve.

There is something beguiling about a range of estimates, especially one purported to be reasonable. But as every actuary knows, determining "reasonability" is ultimately a subjective process. It is influenced by a host of nonscientific but significant factors that, in their totality, define the specific business culture of any particular loss reserving process. Unfortunately, no range of estimates can account for every possible outcome. Indeed, these point estimates are each designed to reflect the average of the possible outcomes--they constitute the search for the pattern that leads to the best estimate of the mean.

While based on sound reasoning, the use of a range to define future uncertainties can he misleading. For example, a range, however determined, gives the impression that any number within it is equally likely to be the "right" answer, and therefore is "equally reasonable." Indeed, the use of ranges in loss estimation may impart a false sense of security since it implies that as long as the reported reserve falls within the range, it is reasonable if it can be justified by other means. Finally, while all methods and models assume that historic performance is a guide to future behavior, traditional deterministic methods often obscure assumptions about the future by focusing too exclusively on the past.

In order to create a clearer statistical picture of what actually goes on, actuaries have been turning to mathematical models to capture the organic nature of real-world loss information. Called "stochastic" (from the Greek for "conjecture," because they recognize the often random, probabilistic qualities of an uncertain future) these models seek to represent how losses could emerge and develop in the future. As an organization begins to operate in the stochastic universe that lies beyond the deterministic algorithms of traditional methods, corporate decision makers will discover that they have access to significantly more information about the liability estimates. For example, a traditional, deterministic point estimate provides no information as to the risk that the ultimate result might eventually exceed the estimate. On the other hand, a stochastic estimate would provide a wealth of statistical information, or risk profile, about the liabilities (e.g., the 75th percentile, which is a reserve level at which there is a 25% chance that future payments might ultimately exceed the reserve).

To be sure, the range of reasonable estimates produced by deterministic methods is still important. But the overriding task in the information-rich stochastic environment is to develop a model that captures the statistical features of the data under analysis--to search for the model that leads to the best estimate of all possible outcomes.

One kind of stochastic model, the bootstrap, demonstrates the strengths, adaptability and utility of this approach. …

Search by... Author
Show... All Results Primary Sources Peer-reviewed

Oops!

An unknown error has occurred. Please click the button below to reload the page. If the problem persists, please try again in a little while.