S&P: A Vision Of The Future

While he said it would be premature to say that ERM is the main factor driving ratings today, "you'll see risk management traces in everything we comment on if we upgrade or downgrade a company." A past downgrade statement may have said the company's profits weren't commensurate with the rating they had. "We'll still say that," but S&P will also say "we believe the company has not demonstrated an understanding of risks in a particular line–or across the board–and has not positioned itself to maximize risk-adjusted returns."

It's more than a different way of saying something, Mr. Dreyer said. "It actually does reflect a deeper understanding of what can go wrong with companies. I'm not going to sit here and say we'll never get a rating wrong, but I do feel a lot more confident about our ratings with ERM analysis being part of the process," he added.

Why has S&P decided not to build its own economic capital model?

"When we started talking about ERM, we discussed building an S&P stochastic economic capital model. But we concluded that a capital model is the end of the process.

"If a company came to us today and said, 'Here's our economic model. Raise our rating or lower our capital requirement,'" Mr. Dreyer said, "we're not going to accept that."

In some cases, however, "they're actually correct. They do have a lower capital requirement than our more simplistic model indicates." They've understood their risks and managed or hedged them in ways not evident with S&P's existing analysis.

Mr. Dreyer said ERM criteria was devised, in part, "to give us a language to get to the point where we could entertain that discussion with companies." He noted that this also helps S&P sort out "which ones are really telling the truth and which ones are really based on more hope than actual science."

Ms. Osborne noted the enormous level of data and detail that would have had to go into building an S&P economic capital model in order "to understand all the products, the interrelationships and the impacts" that could occur as a result of macroeconomic shift or product design shift, for example. "A company has the best view of that," she said. "An outsider trying to replicate it can only replicate it at a summary level." She agreed that it's preferable for S&P to be in a position allowing it to instead draw conclusions from insurers' tailored views of their own risks.

Why is S&P changing its capital model?

Ms. Osborne said that S&P's capital model, first developed in the mid-1990s, has proven to be a useful tool. But "there has been a lot of activity [and] volatility in the intervening years," she said, noting that the revisions are being made to capture that volatility and refresh model factors.

Speaking at S&P's annual insurance conference in June, Ms. Osborne noted, for example, that S&P will revise its non-life pricing factors, which were previously based on the worst industry accident-year loss ratio from 1994-2003 (the same method used by the regulators), to consider the most recent 10 years. "We wanted to have a more robust view, taking in both a hard and a soft underwriting cycle," she said.

For reserve risk factors, S&P will be more ambitious. "Recognizing that quite a lot of the reserve deficiencies have been reported over the past couple of years, we needed to see if we could identify a risk factor and methodology that could get at the volatility," she said. She explained a complex calculation that begins with building data histories of "loss-development metrics," or ratios of current views of accident-year ultimate losses, compared to the views existing one and two years earlier.

How else is the capital model changing?

Ms. Osborne explained that beyond refreshing risk-based factors in the model, the entire framework of the model is being redesigned. "Now we will be able to provide to all companies what we believe are their target capital levels for a "triple-B" rating, an "A," "double-A" or "triple-A" based on confidence levels, she said. "We'll have a monetary view" of how much more capital is needed to get to the next rating level.

(S&P gives a hypothetical example in its recent literature indicating a $1.0 billion capital requirement for an insurer to get an "A" rating–a rating that corresponds to a 98 percent confidence level. In other words, with $1.0 billion of capital, there's a 98 percent probability that the capital will be sufficient to cover all the insurer's risks. Actual capital requirements will vary by insurer risk profile.)

Mr. Dreyer noted that with these probabilistic-type analyses, model results will not be linear as they used to be. Explaining by example, he said that under the existing model, if a "triple-B" level of capital is $100, the "A" level would be $125, and the "double-A" would be $150. With the new model, the distance from "triple-B" to "A" may not be the same as the distance from "A" to "double-A."

Where is S&P in terms of rolling out its new capital model?

"We are still reviewing the tested results, so we are not yet rolling it out to the public," Ms. Osborne said, adding that it will be released shortly, "certainly within this year."

When the testing in complete, a report will be published, and S&P will seek feedback from insurers and other interested parties. After the feedback is considered, some incorporated and the model retested, S&P will have its final capital model, which will be used as a parallel testing mechanism throughout 2007, she said. "We expect that as we start to get more comfortable with new model results, then we will start to rely more heavily on it, weaning ourselves off the old model."

Given that Fitch is developing a stochastic capital model, do you anticipate that ratings assigned by the two firms are now going to be very different?

"I can tell you from our side, we don't expect immediate rating changes based on our new capital model," Mr. Dreyer said. "Our expectation would be that for the industry overall, the capital required would be more or less the same. We didn't go into this process with an assumption that the industry is over- or undercapitalized."

Still, he said, "you could just change one little factor in our model and it will have ripple effects that will change the results for many companies" individually. That is why S&P will run the old and new models in tandem for a while. "Where the new model may be telling us something very different from the old model, it will not be the cause of an upgrade or downgrade," he said. "But it could be the cause of a conversation with the company to try and understand which of these measures is closer to the truth."

As to the direct question about difference with Fitch, "that would depend on how they play out the kind of scenario I just described," Mr. Dreyer said.

Your view of industry capital won't be changing then, correct?

"Yes. Although I should put a big asterisk on that and say, excluding catastrophe risk," Mr. Dreyer said.

In June 2005, S&P announced that cat-specific capital charges for reinsurers would reflect their ability to withstand losses from natural catastrophes that occur once every 250 years, replacing a previous 1-in-100-year event standard. The change was not just to the number of years. Rather, the new charge encompassed a company's losses from natural disasters in the aggregate, as opposed a single event.

In 2006, S&P introduced the same cat-risk charge to primary insurers.

A catastrophe modeler recently estimated the industry's capital to be deficient by $80 billion using S&P's model and loss estimates from his new model. Do you think that's accurate?

From a strictly quantitative perspective, someone could come up with the $80 billion figure. "But that's really misleading," Mr. Dreyer said, noting that when S&P analyzed companies like State Farm or Allstate in the past, even though the capital model did not include a specific charge for cat risk, the analysts did not simply ignore it. They asked how the company could handle catastrophe risks and looked at outputs from the modeling agencies. "What we're doing this year is simply being more direct about measuring that cat risk and incorporating it into the model," he said.

Do you anticipate that non-cat factors in the new capital model will be revised in the future?

Ms. Osborne said, "Where there is volatility in the market, we will want to reflect that in our factors." S&P won't do this every year, but perhaps at intervals of five or 10 years–allowing the firm to gather "informed conclusions about what the volatility truly is," she said.

Assessing the new model, which factor changes will have the most impact on overall capital levels?

While asset charges are changing, Mr. Dreyer and Ms. Osborne agreed that those will have much less impact on p-c insurers than pricing and reserving risk factors. Reiterating that the existing model was created in the 1990s, Mr. Dreyer noted that long-tailed lines, in particular, experienced a lot of adverse development later on.

A change in methodology for reinsurance recoverables–the introduction of a 20 percent surcharge on recoverables for asbestos and environmental losses–will also have a big impact on some insurers, they said, noting that that particular element went into effect earlier this year, even though the new model was not complete.

At this point, which factor updates are you least confident about–do you think need more work?

Mr. Dreyer said S&P is still struggling with pricing and reserve charges, attempting to unwind the cat risk from those charges. He explained that since S&P is now charging companies separately for cat risk, cat impacts need to be removed from the data used to develop the pricing and reserving charges–"otherwise, we'd be hitting them twice" for the same thing.

Any other concerns?

Mr. Dreyer said he worried about the unintended consequences of the introduction of new tools in a once overcapitalized industry. "There is a temptation for companies to get overconfident with their ability to measure risk, he said, adding that S&P expects that the industry as a whole will "decapitalize" as these tools are introduced.

"There's a danger in that," he said, noting that the tools may not be 100 percent–and aren't able to reflect everything that could happen in the future.