Fitch: Shining A New Light

Prism, Fitch's new dynamic global economic capital model, is "putting the industry in a whole new light," according to the rating agency's marketing materials.

The model, Fitch said, generates results from "thousands of scenarios, hundreds of risk curves [and] billions of calculations."

Keith Buckley, managing director, and senior directors Jeff Mohrenweiser and Peter Patrino described the process of developing Prism and the specifics of the new model to NU in June.

Given your obvious enthusiasm for this project, and the man-hours that went into it, should we conclude that capital is now the key determinant of Fitch ratings?

"No. The weighting of capital will be what it has been. Management quality, liquidity, earnings, franchise--all those are critical," Mr. Buckley said.

"We hope [Prism] is the best model out there at this stage, but it's still one aspect of the ratings process. It's certainly not the case that what Prism says drives the ratings."

Why has Fitch developed an economic capital model?

Mr. Buckley noted that new technologies and theories made it possible to make a leap forward, which he compares with the giant step regulators took with risk-based capital requirement in the 1990s.

He also suggested that building Prism gives Fitch a leg-up in understanding economic capital models that insurers are putting together, and that competitive considerations were another driver.

In building Prism, "we've really come to appreciate what makes a good, robust economic stochastic model. We're in a really good spot now having gone through all this work. [And] where we see differences" with company models, "we have a common platform to discuss them."

When Fitch announced the Prism launch, it simultaneously published a report explaining how it will assess insurer in-house models and weigh those models' results with Prism results. "We wanted to put a process in place to marry the two in a harmonized way, rather than in ways that butted into each other," Mr. Buckley said. "We recognized that as good as our model would be, a company can probably be more granular in analyzing their exposures than any third-party observer," he said.

Fitch analysts also believe their efforts to create a dynamic model will prompt insurers to follow their lead. And looking ahead to a next step for Fitch--assessing insurers' enterprise risk management capabilities--they noted that Prism results will help guide those assessments.

"Unless you have the economic tools to measure how management is doing, it's very hard. It's little more than lip service," he said, referring to the need to rely on management descriptions in the absence of a model.

Mr. Patrino noted that he heard participants at a recent actuarial meeting comment that an economic capital model is not an important part of ERM. "That's a statement we fully disagree with," he said. "If you're talking about risk management, what is your main cushion against all perceived risks? It's capital," he said.

Going on to discuss competitive considerations that prompted Fitch to build Prism, Mr. Buckley noted, "We think and hope this will distinguish Fitch in the insurance space relative to the other rating agencies, especially S&P in Europe and A.M. Best in the United States."

By doing "something very big and bold," he said Fitch will demonstrate that a rating agency "can take the lead [on] a new way of risk analysis."

This has been a huge investment, he noted. "We had several people working almost full time on this for a year-and-a-half." Fitch also hired consulting firm Ernst & Young and has invested "several million dollars." All the effort and expenditure, he said, will enhance Fitch's perception, demonstrating the firm's commitment to the industry.

Beyond that, Fitch analysts, who now pepper their conversation with references to statistical measures, VaR and T-VaR, are convinced that such measures provide them with better answers about insurer's ability to withstand various stresses.

Fitch's move toward the Tail-Value-at-Risk approach that's embedded in Prism first came to light in November 2005, when Fitch published a paper on its new method for assessing catastrophe losses, giving examples to explain what T-VaR is about.

To assess catastrophe risk, Fitch cast aside popular "return-time analyses," which assess an insurer's ability to withstand 1-in-100-year or 1-in-250-year events. Given today's environment, "we view a single loss point on a loss distribution as simply not adequate," Mr. Patrino said two months after Katrina.

The cat-risk report explained that two insurers can both have simulated 100-year losses of $100 million, but have exposures to extreme events that vary greatly. One company might have a simulated 500-year-event loss of $200 million, while another's might be $1 billion. T-VaR captures this by averaging all extreme losses above a selected probability threshold. (T-VaR(99), for example, captures events occurring with a 1 percent probability or less. The insurer with the riskier profile in the example would average the $100 million loss, the $1 billion, and all points between.)

Hinting at things to come with Prism back then, Mr. Patrino said the new approach would not be applied to cat risk in isolation in Fitch's overall capital analysis. "The [catastrophe] risk will be simulated with other risks, such as asset risks or reserving risks. Then the aggregate of all risks will be looked at together," he said.

Competitor rating agencies are sticking with static capital models. Do you expect, then, that your ratings are going to be very different from theirs going forward?

"On average, I don't think the introduction of Prism is going to radically change the level of ratings," Mr. Buckley said. But the "hope is that our stochastic model should do a better job of differentiating" lower and higher risk companies. He added that Prism should indicate lower capital requirements than static models for companies that manage risks very well. "And where a company is pushing the envelope with intense product features, we hope to capture that risk and have higher capital requirements," he said.

Still, differences in capital requirements may not translate into rating differences. He noted, for example, that high-risk companies may react by adding capital or reducing risk.

Mr. Buckley also said Fitch is trying to be sensitive to companies it rates, and to the users of ratings, trying to avoid unnecessary ratings volatility. This will be accomplished with a long-term phase-in period, he said.

If Fitch were to simply announce one day, "All ratings will now be adjusted to reflect results of the new model," then downgraded companies might take action to get their ratings back up within six months or so. But in the interim, ratings would swing back and forth, which doesn't help users or companies, he said.

To avoid the volatility, Fitch will roll out Prism in three stages, starting with a period in which it solicits feedback on the modeling theory (just completed), and moving to a Beta-testing phase during which companies will have a model version they can put their data into to view their individual results and offer more feedback.

"If we're smart in accepting that feedback, and knowing who's giving us helpful feedback and who's trying to game the system, it will lead to an ultimately better model," Mr. Buckley said.

After those stages are complete, Prism will be finalized in early 2007, and upgrades will be given immediately. Insurers whose capital falls short of ratings' expectations will be granted a "reasonable cure period" before any downgrade is made.

What is Prism indicating for overall industry capital?

"We're purposely not saying," Mr. Buckley said. "We want to have a consultation period on the theory that is totally unbiased." If Fitch were to tell the industry there's a capital cushion or that required industry capital needs to be higher, "we would start getting very biased feedback. It would work against us," he said.

In fact, he said, going into the project, "we had no bias about whether we wanted it to raise capital requirements or lower them. Wherever the chips fall, that's fine."

In May, Fitch issued a report relating to the cat-risk analysis announced last year. Fitch said its new methodology, together with higher loss estimates now coming from cat-modeling firms, pointed to a 10 percent required capital increase, on average, for insurers with cat exposures.

What has the early feedback been from the industry?

"We've gotten favorable feedback," Mr. Mohrenweiser reported in June. "I think they really appreciate what we're doing, because it helps justify what they're doing," he said, referring to companies that are putting time and money into building in-house models.

He added that they appreciate the fact that "finally someone has taken a hard look at risk-adjusted capital--that not everyone's auto business should get a 2 percent charge, but that sometimes, if a company has good contract features and underwriting, it should be 1 percent." (Editor's note: NU asked the question before the first comment period was complete.)

Mr. Mohrenweiser and Mr. Patrino also pointed to Prism's use of local data as a core feature they believe "resonates well" with insurers. "Companies writing German insurance have German factors based on German data and product features. We're not trying to push U.S.-type factors down European throats," Mr. Mohrenweiser said.

Mr. Patrino noted that U.S. factors, for auto as an example, are inappropriate for U.K. motor. U.S. auto loss ratios typically fall between 75 and 85, while U.K. loss ratios might be 65-80. Only U.K. data would give the right volatility assumptions, he said, noting that volatility drives capital needs.

What will happen during Beta-testing?

The analysts said companies will be able to go to Fitch's Internet site and input data--filling out surveys to supply data beyond what's publicly available (as they will during the actual rating process). The system will run the model through and e-mail results to the lead analyst on the company, who in turn will review the data and give it to the company.

"We want to have a very interactive dialogue with companies," Mr. Buckley said, adding that he expects companies that do poorly will have conversations with the analysts so they can better understand what drives their individual results. Then they may want to run scenarios to test how changing policy limits or reinsurance programs, for example, will change their results.

Testing model sensitivities will also help companies determine what they agree and disagree with, perhaps allowing them to convince Fitch to think about their risks from a different perspective, he said.

Won't different users get different results? The documentation describes a process that starts with industry loss information, for example, but allows an analyst to choose low-risk or high-risk "dial knobs" to reflect their knowledge of the company.

"At the end of the day, 'dial-knob' selections will be made by the analyst," Mr. Buckley explained. But Fitch's analysts are encouraged to have as much discussion with the company as possible--"to get a meeting of the minds or agree to disagree" on whether a company's profile is low or high risk and other issues.

"Outputs are important, but equally as important are the inputs," Mr. Mohrenweiser said, noting that analysts may examine 10 years of historical loss ratios averaging 90, while the company may be targeting 80 next year with pricing or product changes.

Insurers will always have opportunities to provide Fitch with additional company-specific data, according to Prism's explanatory documents.

What inputs is the model most sensitive to for property-casualty insurers?

While Mr. Mohrenweiser said the dial-knob changes--from low to medium to high--are meaningful adjustments, Mr. Patrino noted that model sensitivity varies by rating threshold. He explained that outputs from Prism include capital requirements for "single-A," "double-A," and so on. "A company that just passes over the 'triple-A' capital threshold will be very sensitive to a lot of things really fast," he said.

What aspect of Prism are you least comfortable with? Are you comfortable with the operational risk component? And how do you define operational risk?

Operational risk is "the least sophisticated" component, Mr. Buckley agreed.

The documents explaining Prism say that evaluation of operational risk is based on a series of questions--covering issues like acquisition track record, shareholder lawsuits and regulatory environment--resulting in a 5-15 percent increase in required capital.

Operational risk, Mr. Mohrenweiser said, is essentially anything not covered in other components of the model (such as asset, credit, underwriting, reserving and cat risks). One description might be "the wrong people doing the wrong things at the wrong time," he said, noting that "there's no way our model can delve into" employee background checks.

He confirmed that terrorism risk would fall into the operational risk bucket.

While the measure is not sophisticated, he said it follows industry practice. And "we wanted to be careful not to exclude operational risk" and then come back to put it in next year, after going through the "Herculean effort" of calibrating and testing other parts of the model, he said.

Beyond that, the analysts said they are comfortable with the model but expect to learn more from interested parties.

"I think our brains are telling us we're probably pretty good, but there will probably be a few brains out there that will point out that there are areas where we can fine tune more," Mr. Buckley said.