This article appeared in Contingencies Magazine, November/December 1997. Reprinted with permission of the American Academy of Actuaries.

Actuaries look at the future as part of their everyday work, but a new approach called Dynamic Financial Analysis may set traditional forecasting methods on their ear. Rather than looking only at certain aspects of a balance sheet, this new methodology considers the broad spectrum of a company's financial condition, and analyzes its health in an uncertain and changing world.

Recently the term Dynamic Financial Analysis (DFA) has become a buzzword of sorts within the actuarial profession. This term seemed practically to spring into existence, bringing with it an assortment of papers, models, and seminars devoted to the topic. This term is so new that most actuaries did not learn about it when they took their actuarial exams, and many have had little or no exposure to DFA—yet. Thus, many actuaries and others in the insurance field are not quite sure what this term means, what DFA can accomplish, or how to develop a model.

Dynamic usually means "active, energetic, forceful"—terms not often associated with actuarial work. In this context, a more appropriate definition is "stochastic or variable, as opposed to fixed or static." The term is used to recognize that future outcomes involve uncertainty, and this uncertainty needs to be reflected in the process. Financial in this context, given the historic focus of actuarial talents on the liability side of the balance sheet, reflects the integration of an insurer's assets and liabilities. Analysis is "an examination of a complex, its elements, and their relations," with complex further defined as "a whole made up of complicated or interrelated parts." This meaning is very appropriate to this approach.

Putting the pieces together provides a working definition of DFA that would be "the process of examining the entire financial position of an insurance company over time, considering both the interrelations among the various parts and the stochastic nature of the factors that can affect the results." Not quite as good as Fred Kilbourne's definition of an actuary as "someone who determines the current financial impact of future contingent events," but it's a start.

What caused DFA to become such an important issue for insurers recently? Understanding this can help put the entire concept in better perspective and indicate what the goals of this process should be.

The factors that led to the development of DFA were interest rate volatility and the effect it had on life insurance companies starting in the late 1970s. Figure 1 illustrates the annual change in yield on long-term government bonds, a typical investment of insurers, from 1926 to 1995. Before 1977, changes in interest rates were rarely more than one percentage point, and any change during one year was generally offset by an opposing change the following year. Thus, an institution that ignored the current interest rate when determining the statutory value of bonds for financial statements was well insulated from the relatively minor interest rate swings, which tended to simply seesaw. These interest rate movements really did not significantly affect the financial results of the company. However, the repeated large interest rate increases of the 1977-81 era, and large changes in following years, completely altered this pattern.

Life insurers are highly sensitive to interest rate changes, being highly leveraged with long-term liabilities and even longer-term assets and subject to disintermediation risk. As shown in Figure 2, the number of financially impaired life insurers started to increase in the late 1970s and early 1980s.

These problems sensitized the entire industry to financial risk. Rating agencies began considering the effect that interest rate or equity changes would have on surplus. Regulators began questioning how a company would be affected by specific economic conditions. The initial focus was on the life insurance industry, although in Great Britain and Canada property-liability insurers also became subject to these considerations. By the early 1990s, DFA developed as an issue for property-liability insurers in the United States, as well.

Approaches to DFA

There are two general approaches to DFA—scenario testing and stochastic simulation. Under scenario testing, a few specific potential situations are preselected, and the resulting financial position of an insurer under those circumstances is determined. This approach addresses such questions as, "What happens if interest rates increase 200 basis points?" or "What effect would a 20-percent increase in the lapse rate have on this insurer?"

Scenario testing has long been done by actuaries, even before the interest rate volatility increase of the 1970s. The Social Security system uses this approach, projecting the financial status of the system under three scenarios. (See Figure 3. ) One benefit of this approach for actuaries is that it avoids criticism associated with incorrect point estimates, as long as the actual outcome is somewhere in the range provided. Nevertheless, this approach is not very useful for policy makers, since no indication of the likelihood of the different outcomes is provided. Although the uncertainty of the future is reflected, the range is so wide that making decisions based on these data is fruitless.

Stochastic simulation is grounded on mathematical models that reflect uncertainty in such factors as interest rates, equity values, mortality rates, or loss frequency and severity. Based on the distributions associated with the models, values are randomly selected and used to calculate a large number of potential outcomes. The entire distribution of these outcomes is then available for analysis. One common use of this approach is to determine the proportion of outcomes that are unacceptable (e.g., surplus less than zero). If this proportion is considered too high, then changes in operations or current financial position can be made to reduce this value to an acceptable level.

Thus, stochastic simulation provides much more information than scenario testing. Under scenario testing, the result indicates only whether an insurer is, or is not, in a viable position if a certain event or series of events were to occur. Stochastic simulation provides information about the likelihood of an entire range of outcomes. Thus, DFA is generally based on stochastic simulation.

Classifying Insurance Risk

The risks that insurers face can be categorized into two components—balance sheet and operations. Balance sheet risks affect the value of assets and liabilities. Increases in interest rates, for instance, reduce the value of bonds. Adverse loss development increases the value of loss reserves. Similarly, there are two components of operating risk—underwriting and investments. Thus, there are four risk elements that need to be considered in DFA. Only two, however, have traditionally been considered by actuaries—liabilities and underwriting. (See Figure 4.)

The four components of risk are interrelated. An increase in interest rates lowers the value of bonds, but at the same time increases investment income. Increases in interest rates are generally accompanied by increases in inflation. In this case, losses are more expensive to settle (for health and property-liability insurers), affecting loss reserves and underwriting results. Similarly, a recession could depress the value of currently held equities, reduce expected investment income, and reduce some types of loss frequency (e.g., automobile accidents). Consideration of these complicated interrelationships is a key element in the analysis aspect of DFA.

Steps in Building a DFA Model

A workable model must be a simplified version of reality. Not all risks can, or should, be modeled. One consideration is whether a particular factor can be quantified. For example, a leading cause of insurer insolvency is management fraud. However, quantifying this risk is impossible. A company either is, or is not, being defrauded by its managers. Any probability of fraud included in a model would overstate this risk for a well-managed insurer and grossly understate it for an insurer that is being defrauded. Since it is not the function of the actuary to determine whether adequate financial controls to prevent fraud are in place, it is better to treat fraud as an unquantifiable risk and leave it out of the model.

One of the first steps to take when building a DFA model is to decide how the model will be used. DFA models can accomplish a number of goals, but generally only if the goals are considered during the design phase. If a DFA model is designed solely for solvency testing, then the only question that can be answered is how frequently an insurer gets into financial distress. If properly designed, a DFA model can also provide information about what events led to the company's financial difficulties, the financial condition ofthe insurer when it is not in financial distress, and the distribution of key financial variables. Figure 5, for example, illustrates the ending surplus resulting from 1,000 runs of a stochastic simulation. If the results were unacceptable in 10 of these 1,000 cases, the specific factors that led to these problems could be determined. Perhaps, in eight of the 10 cases, the culprit was rapidly rising interest rates. In that case, the insurer would know that further hedging of interest-rate risk would be the best way to reduce the chances of financial impairment.

The next step in building a DFA model is to decide which factors to allow to vary and which to project deterministically. Remembering that a model is a simplified version of reality, one must resist attempting to recognize every risk or forecast every possibility. Since the DFA model is, in essence, going to project the balance sheet and operating statement of the insurer over the planning horizon, all risks that affect assets, liabilities, underwriting, or investment income need to be considered. However, the model should incorporate only the most relevant factors.

Insurer assets generally consist of bonds, equities, mortgages or mortgage-backed securities, and premium balances, as well as assorted other categories. Bond value is affected by interest rates and default rates. Equities change with market movements. Mortgages, in addition to being affected by interest rates and default, are also affected by prepayment patterns. Although many other factors could affect asset value, the most important factors to model are interest rates, default risk, mortgage prepayment patterns, and stock market movements.

Liabilities differ by type of insurer. For property-liability insurers, the primary liabilities are loss reserves, loss adjustments expense reserves, and unearned premium reserves. Thus, forecasting the value of future loss payments is important. Although the statutory valuation of loss reserves is generally undiscounted, the economic value of these reserves would be the discounted value. Thus, the interest rate is also an important consideration for the economic value of these reserves.

The underwriting results of an insurer depend on premium income and losses incurred. Although premiums, in general, are a function of expected losses and expenses, in some jurisdictions premiums may not be allowed to be adjusted to the appropriate level, due to statutory, regulatory, or judicial requirements. Losses are a function of both frequency and severity. Loss frequency is affected by catastrophes and by trends in society, for example safety, health management, and economic conditions. Severity is affected by inflation and construction issues.

Investment income is affected by many of the same factors that affect assets—interest rates, default levels, and market returns. An additional consideration is whether the investment portfolio will change over time. If the insurer increases investments in lower-rated bonds, interest rates would be expected to increase, but the volatility of results would increase as well.

Many of the important factors cited above are related. The relationship between interest rates and inflation was first noted in 1896 by Irving Fisher. Inflation is also related to equity returns, as well as the likelihood of adverse economic conditions that could also affect the frequency of losses. Catastrophes affect future premium levels.

Sources of Information

The Casualty Actuarial Society [Dynamic Financial Analysis] web site has links to the historical values of many important variables, including interest rates. The CAS has also sponsored several seminars on DFA, and papers on this topic are available in print or on the Web site. Many financial texts describe sophisticated interest-rate models that have been developed and used to price complex financial instruments.

Miller, Rapp, Herbers & Terry, Inc., an actuarial consulting firm in Bloomington, Illinois, has made a basic DFA model for property-liability insurers accessible on its Web site. This model, which is run on Excel and Risk software, is available at no charge. Some of the stochastic processes included in this model are shown on Figure 6. Benchmark values for the interrelationships among variables are included in the program, but these can be changed by users to reflect the specific characteristics of each insurer. Input is derived from the insurer's financial statement. The output includes the distribution of key financial variables. The simplicity of the model facilitates its use as a learning tool for DFA.

The Future of DFA

DFA represents a new area of study for actuaries, an area requiring new tools and new expertise. The need for DFA developed as the financial world became a riskier place, beginning in the 1970s. Previously, actuaries could safely confine their attention to the liability side of the balance sheet and the underwriting side of operations. Work on DFA will cause actuaries to consider both the asset and liability sides of insurance operations, and both underwriting and investment income.

DFA can be a powerful business planning tool that will allow insurers to assess the financial risk of different business strategies and select the plan that provides the best returns for the risks that are undertaken. It can highlight the conditions that generate unfavorable outcomes, so that managers can deal with these effects appropriately. It can provide a calibrated tool to replace rough intuition.

However, we have a long way to go in developing DFA into a useful technique. In addition to learning how to build and use realistic DFA models, actuaries have to communicate the value of this approach to those in the company who can use this information. Thus, we are in the position of trying to perfect a tool while simultaneously justifying its need. As this process continues, the companies that use DFA are those that are most likely to thrive in the next decade.

Stephen P. D'Arcy and Richard W. Gorvett are in the Department of Finance at the University of Illinois at Urbana-Champaign. Josef A. Herloers and Thomas E. Hettinger are consulting actuaries at Miller, Rapp, Herbers & Terry, Inc., in Bloomington, Illinois.