Through its enterprise risk management (ERM) reviews of North American insurers over the years, Standard & Poor's Ratings Services has observed rated insurers undertake various capital modeling frameworks. In our discussions with the insurers, we've gained insight into their analytical modeling tools, methodologies, and risks metrics and measures, as well as the lessons learned from the recent financial turmoil. In our view, insurers are using a variety of capital modeling practices, and we've seen some emerging trends in capital adequacy analyses in the insurance industry.

An insurer's capital needs arise from its risk profile, which comprises the risks the insurer has assumed and is likely to assume in the future, as well as the risk controls in place to mitigate these risks. The macroeconomic environment (such as a protracted period of low interest rates) further affects insurers' capital needs. The ability to continue to operate under extreme circumstances is a key factor that we assess in determining whether an insurer maintains adequate capital. Insurers typically make capital management decisions based on their liquidity and financial flexibility, as well as on regulatory requirements and rating agencies' expectations. We also factor our opinion of an insurer's risk profile into our risk-based capital model, which we use to form an opinion of an insurer's capital adequacy. (Our capital model sets higher capital targets for higher ratings.) But because our capital model may not fully capture all of the nuances of a firm's risk profile, for those insurers that have economic capital models (ECM), that use them extensively in decision making, and that provide sufficient information, we expand our ERM analysis to include a review of their internal capital models. (We define these reviews as ERM Level III reviews. See "A New Level Of Enterprise Risk Management Analysis: Methodology For Assessing Insurers' Economic Capital Models," published Jan. 24, 2011.) We incorporate the results from an insurer's own internal capital model into our capital adequacy assessment if, based on our ECM review, we find that model achieves a minimum level of credibility.

Through our ERM Level I and II assessments, we've gained insight into the ways many of the leading rated North American insurers approach risk capital modeling, the sources and implications of modeled risks, as well as the analytical lessons learned from the financial crisis and recent developments. We plan to publish a separate article of the findings from our detailed reviews of insurers' ECM (ERM Level III reviews) that we've so far completed worldwide.

The most common analytical practices that insurers use to evaluate their capital needs include:

• Stress testing: The use of deterministic, extreme scenarios to evaluate risk-bearing capacity and identify capital and funding needs, as well as to establish risk appetite, test risk controls, explore risk contagions, and test liquidity.
• Stochastically calibrated stress scenario models, which use scenarios taken from specific confidence levels from the scenario distribution: A structured scenario approach to evaluating economic capital (EC)--the amount of capital needed in addition to future income flows and assets backing reserves and liabilities to meet all future benefits and expenses in all but the most extreme events--separately for each risk ("silo" capital calculations). It combines aspects of both stress testing and fully integrated stochastic capital modeling.
• Integrated stochastic capital models: Dynamic financial models that simulate the evolution of an insurer's risk profile as an interplay of the in-force business, new business, macroeconomic risks, amount and structure of the firm's capital, financial flexibility, existing risk controls, and prospective risk appetite.

Stress Testing

Stress scenarios have long been a part of insurers' decision-making, and over the past two years they have become an indispensable risk management tool for many of the insurers that Standard & Poor's rates. Companies use stress scenario testing to analyze their capital adequacy and liquidity, to test the effectiveness of their risk-mitigation tools (such as reinsurance and hedging), to evaluate potential risk correlation in the tails (i.e., the extremes of the scenario distribution), and even to validate complex stochastic models. Moreover, Standard & Poor's, other rating agencies, and regulators are increasingly incorporating stress testing into their reviews to evaluate insurers' ability to withstand hypothetical shocks (such as sudden increases or decreases in interest rates or 1-in-250-year catastrophic events) over given time periods.

When conducting our reviews, we expect the stress scenarios to be severe, yet plausible. We expect highly rated firms to be able to withstand more severe stresses without defaulting. In fact, we have articulated a set of economic stress scenarios associated with particular rating categories that reflect the levels of stress that issuers rated in that category should, in our view, be able to withstand without defaulting (see Appendix IV of "Understanding Standard & Poor's Rating Definitions," published June 3, 2009). Even more challenging scenarios combine multiple unfavorable factors, such as macroeconomic shocks coupled with industry- and/or company-specific events. For example, higher-than-normal hurricane-related losses experienced during a major financial downturn could exacerbate the economic stress for a property/casualty insurer. Another example is the significant challenges that many life insurers encountered during the recent financial crisis, when both the equity market and interest rates declined. Industrywide antiselective policyholder behavior--policyholders using the provisions of their policies in their own self interest and to the detriment of the company--further magnified the adverse impact on insurers' earnings and capital as policyholders took advantage of in-the-money benefit options such as guaranteed living benefits embedded in their contracts.

Many insurers also use stress scenarios to describe their risk tolerance. For instance, an insurer may conclude that under normal business conditions that it anticipates for the next year, it would be able to absorb $200 million of unexpected losses from its insurance operations, while under severely depressed business conditions, the company would only be able to withstand a $100 million unexpected loss. With this analysis, the company may decide that it should try to prudently confine such losses to $80 million. In fact, in our assessments, many companies with solid ERM practices can readily demonstrate that their risk tolerances are set quite prudently and are frequently stress tested.

In addition to testing for solvency and liquidity, stress scenarios may consider various other implications. For example, a stress event could undermine a company's competitiveness, erode its market share, reduce the availability and increase the cost of borrowing, trigger contingent obligations such as recapture provisions in reinsurance treaties (which would force direct writing insurance companies to take back risks they've ceded to reinsurers), prompt the departure of key personnel, or force the company to forgo strategically important plans. Economic capital models focus on an insurer's ability to remain solvent and usually do not include these additional effects. As such, some insurers try to incorporate these possibilities through scenario analyses. As a result, they are able to better assess the true financial impact of stress scenarios and identify the maximum adversity they can withstand without falling into financial distress.

Integrated Stochastic Capital Models

Integrated stochastic capital models attempt to re-create insurers' risk profiles by modeling the dynamic interplay of all material risks and risk controls in the context of the available (existing and accessible) capital and projected balance sheet valuations. Rather than developing a handful of scenarios for each risk (for example, sudden increases or decreases in the yield curve) as in the stochastically calibrated stress scenario approach, these models use random-sampling simulation techniques to generate large numbers of future scenarios, often in hundreds of thousands, including extreme events and contagion effects that may hurt an insurer's financial health--its profitability, liquidity, and even solvency. Rather than relying on diversification or correlation assumptions, whenever possible, these models try to reproduce the relationships between sources of risk and the risks themselves and recognize ripple effects and the heightened interconnectedness between risks in extreme circumstances. These models do not try to calculate the capital levels for the risk categories and then aggregate them to arrive at overall capital, but rather they focus on the overall capital needs caused by the complex risk interdependencies and then allocate the capital to specific risk sources based on their contribution to the aggregate risk profile.

These models' starting point is defining the capital adequacy targets, both for the overall group and for individual operating companies. Most of the models we have seen in our rating reviews focus on capital targets of specific rating categories. For example, an insurer may be aiming to maintain capital at a level consistent with a 'A' rating, which is generally interpreted as the level that ensures ongoing solvency in 99.92% of the modeled one-year outcomes.

Stochastically Calibrated Stress Scenarios

The use of stochastically calibrated stress scenarios is another common approach for estimating EC. Under this approach, a company will generate a distribution of stress scenarios, such as for equity, interest rate, credit, insurance, or operational risks. The insurer then uses these scenarios to analyze adverse deviations from the expected cash flows. A common practice is to calibrate each of these "shocks" to a level of confidence consistent with the rating or regulatory solvency (for example, 99.5%) targets, and then compare the impact of these shocks with the insurer's existing capital. Furthermore, insurers typically aggregate capitals from the risk-specific shocks--for example, sudden increases and decreases in interest rates--to estimate their EC needs using a correlation matrix of factors that imputes the relationships between individual risks to consider the benefit of diversification.

Although insurers have been using the stochastically calibrated stress scenario approach for a number of years, they continue to improve and refine their existing methodologies, especially after the recent financial crisis. The market turmoil demonstrated that the distributions of risks affecting insurers tend to be "fat-tailed." That is, the probability of extreme events is much higher than the "normal" distribution. Moreover, the correlation between the risks tends to be much more pronounced in extreme tail events--stressful scenarios with the lowest probabilities of occurrence. As a result, some insurers have enhanced their models with regime-switching mechanisms, which dynamically change the assumed relationships between the risk factors as the environment is assumed to change, in an attempt to capture the correlations between risks in the tail. Some others have migrated toward integrated stochastic models.

Economic Scenario Generator

Most of the models we have seen use outputs from economic scenario generators (ESGs). In our opinion, an ESG is a critical component in a capital model's dependence structure (its implicit or explicit relationship between risks) because each simulated macroeconomic scenario naturally correlates many of the risks to which an insurer is exposed. The idea behind an ESG is to create realistic and consistent sets of economic scenarios, including future interest rates, bond spreads, inflation, stock-market indices, GDP growth, and unemployment, and then link them to asset and liability projections. Some models, especially multiyear ones, link the macroeconomic projections to future business volume, claim cost trends, asset purchases and sales, as well as management actions such as raising capital and scaling down unprofitable business. With the explicit link to the underlying cause (an economic environment), ESG-based models are likely better able to re-create risk contagions at an insurance enterprise, such as those associated with a drop in business volume, escalation of credit/counterparty risk, a drop in asset valuations, or an increase in liability valuations.

Insurers' use of ESGs varies widely in the industry, and one of the biggest distinctions is between the use of market consistent and real world scenarios. Many companies favor market-consistent ESGs, arguing that market consistency emphasizes alignment with the current valuation of the economic balance sheet. Proponents of the real-world ESGs believe that the EC should be sufficient to absorb the impact of potential heightened market volatilities that current observed market valuations may not fully reflect.

Model Risk

Models are simplified versions of reality, and there is inherently a risk that models will fail to accurately reflect and quantify all the risk exposures. This is referred to as model risk. Model risk is unavoidable and is one of the key issues in insurer's risk capital modeling. Modeled extreme events and the correlation of risks in the tails, both of which are inherently imprecise, tend to drive insurers' capital needs. In fact, even when substantial data underlie a model's projection, theoretical generalizations and simplified assumptions can heavily influence the projected results. Moreover, complex stochastic models are prone to various parameter-estimation, approximation, extrapolation, and sampling errors. Human error and poor execution also contribute to model risk.

However, insurers can reduce the risks associated with capital models by improving their data and methodologies through model validation and controls over the modeling process and execution. Additionally, increased computer processing power and a more efficient model design may help minimize the random sampling risk.

A capital model's robustness depends on its granularity and completeness, meaning its ability to capture the vast majority of risks that could affect an insurer's solvency. Most insurers note that some risks, such as emerging risks, man-made catastrophes, pandemics, and operational risks may defy reliable quantification. As a result, the models tend to simplify or even entirely omit these risks and their relationships. In our experience, few insurers attempt to evaluate the impact that such simplifications and omissions have on their capital. Nevertheless, most insurers try to, at least partially, compensate for these issues by scaling down their risk tolerances, by using conservative assumptions in extreme-event modeling, or by maintaining a level of capital in excess of the required amount of EC a model indicates.

Depending on the size and complexity of an insurer, a capital model can be quite large and complex, which often poses substantial computational challenges. This is especially true for higher-rated insurers because they tend to measure their capital adequacy against more-demanding targets (for example, the ability to remain solvent in one year with 99.95% confidence, rather than 99.5%) and therefore may need very large number of simulations--hundreds of thousands (or more)--to credibly evaluate potential insolvency events. For many models, this may involve days or even weeks of distributed computing, or it may be impractical, causing companies to take projection shortcuts. Runtime concerns might force users to reduce the number of simulations and, thus, increase the sampling error. Some users turned to other solutions such as replicating portfolios--combinations of a limited number of financial instruments used to model assets or liabilities--as a result of time and resource constraints. The size and complexity of models also make it more difficult to maintain, govern, and interpret the models. These risks are especially high in multiyear models.

Some of the shortcuts insurers take can become particularly problematic if they underestimate an insurer's capital needs. To minimize these concerns, regulators and rating agencies are starting to look for evidence that insurers have robust model testing and validation, as well as thorough documentation. In our dialogue with insurers, we assess whether companies understand and, to the degree possible, mitigate the effects of these shortcuts on projected capital needs. In recent years, we have observed improvements in insurers' efforts to address these issues.

Time Horizon

Another issue in insurers' risk capital model development is their choice of time horizon. Many insurers use a one-year time horizon, though others use multiyear projections, typically many years into the future, until virtually all in-force liabilities from business previously written run off.

The insurance industry has long used stochastic multiyear projections of cash flows to serve a variety of purposes, including pricing and valuation. Take, for example, stochastic asset-liability-management (ALM) models. For a given and predictable liability profile, ALM models test the assets' ability to fully offset all in-force contractual obligations and related expenses projected many years into the future. This leads to a definition of EC as the amount of assets needed today over and above the current value of reserves and liabilities to fully run off a defined set of liabilities with only a minimum risk of insolvency to a predetermined level of confidence.

Some practitioners, typically from property/casualty companies, have been using multiyear models that try to reproduce the real-world dynamics between risk and capital, stochastically simulating the future balance sheets considering new business and the period-to-period runoff of the existing obligations. In each scenario, the model assumes a particular course of management action under the simulated circumstances. For example, if six quarters from now capital falls below a predefined threshold, the rules built into the model may dictate that the company raise debt or equity to replenish capital, reduce new business writings, or transfer more risk. Unfortunately, with each modeled time increment (a quarter or a year), the credibility of the balance sheet projections decreases because the layered assumptions cause the model risk to balloon.

These limitations of the multiyear models are some of the reasons capital models that focus on one-year solvency dominate the insurance industry. However, the one-year approach has its own limitations. Most of the implementations of one-year models do not perform intermediate (such as quarterly) solvency checks, and therefore may understate the modeled probability of insolvencies. Moreover, with their focus on one-year solvency, such models may only partially reflect the longer-term capital implications of most strategic decisions. Excluding new business from the analyses further reduces a capital model's ability to serve as strategic-guidance tool.

Analytical Lessons Learned

Economic capital models' complexity and focus on extreme events make them highly prone to model risk--which raises questions about whether it's even possible for these models to credibly assess insurer's capital needs. And if a given risk profile is too complex to be credibly quantified, is the company's business too complex to manage?

On one hand, some insurers operate under strict risk limits and controls and are of a size and complexity that, in our opinion, make their ERM frameworks more effective. This may explain why some of the insurers we rate highly (many of which have excess capital) opt for a less robust approach when assessing their capital needs. These insurers combine a set of extremely adverse scenarios--on the premise that in extreme circumstances most risks tend to correlate--and still are able to demonstrate that capital is sufficient to absorb even the most adverse outcomes.

On the other hand, many large financial institutions fail to adequately measure and manage their risks, as we saw during the financial crisis in 2008. The insurers facing the biggest challenges have risk profiles that are dominated by complex financial products such as variable annuities with living benefits and universal life policies with secondary guarantees. Some of these products have long risk horizons, meaning that they generally have many years of loss exposures or claims payments and often involve complex long-term embedded options. Many of the insurers with significant amounts of these products continue to struggle with finding appropriate methodologies and overcoming technological constraints when trying to quantify and mitigate the aggregate risks.

In our opinion, one of the biggest lessons from the recent financial crisis is that sophisticated models may create a false sense of precision and that overreliance on complex models further magnifies the effect of model risk. As a result, many companies are moving toward simpler, more transparent risk profiles. Additionally, many insurers are now focusing more on stress tests that incorporate more extreme events and higher risk correlation in the tail risk scenarios as a means to verify whether the risks will likely be contained. Many insurers have been scaling down their exposures, imposing stricter risk limits, and embracing stress testing as a critical risk management tool. We view favorably the use of stress tests as a tool for capital adequacy analyses, especially those that rely on complex capital models.

With the shift toward mark-to-market and market-consistent valuations, the uncertainties that products with long-tail exposures create become visible on the balance sheet very quickly because the macroeconomic volatility and uncertainty tend to erode the asset valuations and amplify the volatility of the valuation of liabilities. Moreover, at least theoretically, the simulated future balance sheets will be affected even more, as the uncertainty of future projections tends to increase with every modeled time increment. The greater the uncertainty about the amounts and timing of cash flows, the higher the volatility of projected asset and liability valuations and the greater the EC that will be needed to ensure ongoing solvency. Accordingly, volatile longer-tailed risks will tend to consume more capital.

The silver lining to this is that the EC view imposes a more economic view on profitability, and hence pricing, of long-tailed risks. Consider, for instance, long-tailed casualty lines, such as excess workers' compensation in the U.S. Insurers are increasingly uneasy about these lines' high sensitivity to future claim cost inflation, among other uncertainties. For example, one large diversified insurer, after having quantified the capital implications from carrying such volatile and long-term liabilities, especially in light of the uncertainties about future medical costs, has questioned the viability of its current pricing and is drastically reducing new writings. This is an example of risk selection that better reflects the economic view of capital.

By itself, EC is just a capital adequacy indicator and is not immune to theoretical limitations or significant model risks. The benefit of economic capital modeling is not necessarily in its precise quantification, but rather in its ability to determine the sources of a firm's capital needs and its focus on the economic principles of decision-making.

Capital Modeling Has Become Even More Critical

Leading practitioners of ERM generally use advanced approaches to modeling and aggregation of risk, taking into consideration both available and targeted capital. At the same time, they have a good understanding of risk modeling limitations and offset them with their own judgment, drawing from their experience. They also emphasize forward-looking assessments of risk, with continual monitoring of trends, prompt assumption revisions, and frequent portfolio aggregations within the capital models. Deterministic stress testing supplements complex stochastic models.

In our opinion, the level of sophistication and the degree of adoption of these advanced practices have greatly increased since the financial turmoil of 2008-2009. In particular, many large insurers and reinsurers have embraced advanced stochastic capital models. According to many of these firms, increased attention from regulators and rating agencies may partially explain this trend. On the other hand, the industry as a whole is still in the early stages of adoption and development of such models. Although we see some promising trends in ERM development at medium-size companies, smaller insurers tend to lag far behind in terms of the depth of expertise, modeling tools, quality of data, resources, and management buy-in.

We believe that robust capital modeling is crucial to insurers in their attempt to quantify capital requirements given their risk profiles and is key to ensuring ongoing solvency and meeting other targets. We have observed widely divergent practices by insurers on some of the key issues of capital modeling, including their methodologies, approaches to stress testing, choice of time horizon, and use of economic scenario generator. Model risks will remain a key concern, not only because the modeling of extreme events is inherently imprecise, but also because of the many shortcuts insurers have to take for practical reasons. Some insurers have demonstrated thorough understanding of the issue and are proactively measuring the implication of model risks through thoughtful judgment. Although we have observed greater sophistication in capital modeling in the industry, we remain concerned about the robustness (either too much or not enough) of some of the adopted methodologies and the consistency of their implementation and integration into decision-making.

Related Criteria And Research

• A New Level Of Enterprise Risk Management Analysis: Methodology For Assessing Insurers' Economic Capital Models, Jan. 24, 2011
• Refined Methodology And Assumptions For Analyzing Insurer Capital Adequacy Using The Risk-Based Insurance Capital Model, June 7, 2010
• Appendix IV of Understanding Standard & Poor's Rating Definitions, June 3, 2009