The best scenario

Financial institutions need to know how changes in economic scenarios may affect their risk and capital. In the wake of the subprime crisis, a growing number of banks are using economic scenario generators to devise new situations and measure the impact on capital. Clive Davidson reports

Banks have been found sorely wanting over the past year in their ability to model future events and understand the depths of the risks in their portfolios. For all the sophistication and complexity of pricing and risk models, many financial institutions severely underestimated their exposures, had little idea of how they would be affected by heightened market stress and failed to account for liquidity risk. Part of the problem is the narrow and short-term view of the world taken by many firms. More robust and longer-term scenarios of the future would give banks a better understanding of their exposures, especially under extreme conditions when normal rules no longer apply.

Insurance companies have long had to take this longer-term view, while regulation and best practice have forced them to adopt mark-to-market pricing and stochastic modelling for the valuation of assets and liabilities. In doing so, the insurance industry has developed powerful and sophisticated tools to generate realistic economic and financial scenarios, including potential stress conditions and how they might affect assets and liabilities. Bancassurers are also beginning to use common economic scenarios as a means to achieve consistent risk and capital calculations across both their insurance and banking arms, and stand-alone banks are starting to ask whether these tools might be useful for them too.

Belgian bancassurer Fortis, for example, uses economic capital as a uniform risk capital metric across its various business lines, which include retail banking, asset management, investment banking and insurance. "For Fortis, economic capital is based on consistent risk-type definitions across banking and insurance," says Fred Bos, the firm's chief risk officer. "It allows us to compare risks across all our activities on a like-for-like basis."

To model its economic capital, the bank uses a set of 14 macroeconomic scenarios, applied across all its activities, which it generates itself. "A number of the economic scenarios we apply are based on general formulaic assumptions, applying loss distributions generally observed and accepted for the various risks we face," explains Bos. As well as creating and regularly updating its own stress scenarios, the bank uses historical events, such as the 1987 stock market crash. "Thus, we can assess how set economic scenarios will affect the group and its sub-components, also allowing us to identify co-dependencies and differences," he adds. In addition, Fortis employs scenarios that are specific to certain lines of business, such as claims behaviour in property and casualty insurance.

Hanover-based insurance group Talanx, meanwhile, decided to look to outside specialists when it came to creating scenarios to model its risk exposures. "To evaluate investment and risk management strategies, you need to replicate the real markets as closely as possible," explains Gerhard Stahl, deputy chief risk officer at Talanx. "Models must capture extreme tail events. Using naive models can miss the true risk and result in expensive mistakes."

Talanx concluded it did not have the time or resources to develop scenarios to the level of sophistication it required, so it bought the Gems economic scenario generator (ESG) from Purchase, New York-based DFA Capital Management. "Insurance companies require an economic scenario generator that can cope with long forecasting horizons and can produce real-world and risk-neutral scenarios, reproducing observed market prices for financial instruments including options that are out-of-the-money," says Stahl.

Gems generates a distribution of possible global economic future scenarios, including derivatives and alternative asset prices. A crucial part of an ESG is that it has to reproduce the same mathematics the financial markets use for pricing, notes Richard Urbach, the head of quantitative finance at DFA. "If you don't reproduce the pricing mathematics, you can't address asset and liability questions, or hedging or risk management strategies, or asset allocation, because each of these requires buying and selling in the market. If your prices are inaccurate, your conclusions will be inaccurate," he says. The ESG also has to incorporate historical experience, especially during periods of volatility, as well as a set of unexpected but plausible outcomes, adds Urbach.

One major challenge in creating economic scenarios is "thinking the impossible in terms of possible threats, such as the breakdown of correlations that we have recently experienced and how this affects structured credit portfolios, or the significant liquidity dry-up in money markets", says Fortis's Bos. Urbach agrees, noting that DFA mixes up its scenario simulations by introducing a random factor representing new and unexpected information. "The human mind has a great deal of difficulty in imagining scenarios that are too far from its own experiences; a machine doesn't," he says.

It is difficult enough to create scenarios for a single economy, but it is considerably more challenging to create an ESG that can calculate multiple economies simultaneously and account for all the differences and dependencies between them. "Although every economy has the same sort of structure, its traded instruments are different - for example, while the US has mortgage-backed securities, Germany has the Pfandbriefe," explains Urbach. An ESG should be able to run individual economies on their own, but also in parallel, with all their interdependencies. "In which case, the economies must be coupled in the right way," he adds.

All in all, models don't come much more ambitious than ESGs. A team of mathematicians, economists and financial engineers took seven years to build Gems, says Urbach. Meanwhile, Edinburgh-based software and consultancy firm Barrie & Hibbert has spent 100 staff years building its ESG, and 25 economists and financial engineers maintain its calibration, claims Gavin Kretzschmar, the firm's global head of banking. The system is now used by nine member companies of the European insurance industry's Chief Risk Officer Forum, including Aegon, Aviva, Axa and Munich Re, he adds.

French reinsurer Scor has bucked the trend among insurers to buy systems from third-party vendors and has developed its own ESG. When Michel Dacorogna, Zurich-based head of group financial analysis and risk modelling for Scor Switzerland, joined the company in 2000, it was using an ESG provided by Morgan Stanley. But the ESGs created by banks at that time generally operated to far shorter time horizons than required by reinsurers. In addition, most of the ESGs at the start of the decade used the Cox-Ingersoll-Ross model for interest rates. "Cox-Ingersoll-Ross is okay if you want to price a three-month swaption, but not if you want to price life insurance over 30 years. If you run the model over 30 years, you will see that, two-thirds of the time, the yield curve will be inverted. The reality is different," Dacorogna notes.

In building its ESG, Scor decided to resample historical market behaviour, with some modifications, to create a range of plausible future behaviour scenarios. Working on a quarterly basis, the company randomly picks moments in economic and market history, and resamples the values of the variables that prevailed at that moment. A key advantage of this approach is that the sample includes all the relationships and dependencies between the variables, notes Dacorogna.

However, if Scor were to use the resampled moments just as they were, then the scenarios it generated would simply repeat history, without the potential for new developments. So Scor focuses on what it calls the innovations in the market - the unexpected changes and movements in the variables. It samples these innovations by comparing the actual values at the chosen moment in time against the values that were expected the previous quarter. For some variables, such as interest rates and foreign exchange levels, the expected values can be derived from forward rates, while the expected values of other variables, such as GDP, can be calculated as the average of historical changes in value. The difference between the actual and the expected values are the innovations, and these are applied to the current economic and market events to produce future scenarios - but not without some further modifications.

First, Scor applies a random modification to produce a small but realistic number of stress scenarios where there are extreme events. Then, it restores the Garch effect - the tendency of periods of volatility to cluster - which is destroyed by the random sampling of the historical record. (However, Scor does not apply Garch to the extreme events in the stress scenarios because that would not reflect reality and leads to absurd results, adds Dacorogna.) Other modifications include accounting for economic and business cycles.

"What we needed was an ESG not just for one economy, but for at least six different currencies," says Dacorogna. Scor's ESG generates values for around 700 economic factors, including GDP, inflation, interest rates and equity and real estate indexes. "There is no point in generating these separately - we need the dependencies. But I know how hard it is to calibrate these. So we started from the basis that we didn't know how to model the factors and their dependencies, but that we had seen it all before," he says. Hence, the historical resampling. A further advantage of this approach is its flexibility - new additions can be incorporated into the ESG simply by adding these to the historical samples. "If all of a sudden we write business in a new area, I want to be able to quickly adapt my model without having to do a huge calibration effort," Dacorogna adds.

Given the recent market turmoil, plus regulatory pressure to improve stress testing, banks are also beginning to look at the potential of the new generation of powerful ESGs to solve the problem of enterprise-wide economic capital modelling. DFA says both European and US banks have shown interest in its Gems product, and the company is currently talking to one major European bank about using the ESG for asset and liability management within its pension fund business. Meanwhile, Barrie & Hibbert is working with a major Wall Street firm and a leading UK high street bank on implementing its ESG. Such is the interest from banks that Barrie & Hibbert has created an ESG and simulation platform specifically tailored to the banking industry called Economic Capital Scenario Generator, and a new consultancy division, headed by Kretzschmar, to support it.

"The world is converging for us," notes Kretzschmar. "Bancassurers need to use the same set of economic scenarios to project economic capital across banking and insurance, otherwise they get inconsistencies in their diversification benefit and capital calculations. And stand-alone banks are asking us how they can apply calibrated best estimate economic scenarios to their banking books."

A major problem for banks is that Pillar I of Basel II does not allow them to hold less capital as a result of the diversification benefit between asset classes. This they must do under Pillar II, but the tools they have been using to calculate risk and capital - mostly variance/covariance matrices - are not up to the task, says Kretzschmar. "The variance/covariance approach has all sorts of problems when it comes to putting together multi-variate distributions for different market and credit risk processes, across multiple asset classes," he notes. "What banks have found recently is that there has been a lot more 'tail' correlation between asset classes than has been accounted for in the calibration of their variance/covariance matrices."

The growing interest by banks in ESGs suggests there is recognition that some of their risk and capital modelling problems can be solved by stochastic projections of asset values using consistent economic scenarios generated by these systems. Although they are generally behind the insurance industry in the use of ESGs and stochastic modelling, banks have one advantage in that Pillar I of Basel II has forced them to gather, clean and prepare all the data they need in order to project future asset values, including parameters such as probability of default, loss given default and exposure at default. The process of mapping this data to an ESG is relatively straightforward, says Kretzschmar: "Banks have got their data all there and waiting in their Pillar I data warehouses."

As sophisticated as ESGs have become, they still have their limitations. Current ESGs cannot run more than 10-15 currencies or economies simultaneously because of the complexity this entails. Nor do they produce perfect forecasts of the future. Nonetheless, ESG developers spend considerable effort validating the scenarios they generate - and in many cases, these scenarios have proved more prescient than those produced by more conventional means.

Dacorogna at Scor points to collateralised debt obligations (CDOs) as a measure of the value of its ESG. "Some CDO structures are similar to insurance contracts in terms of time horizons," he explains. In 2006, Dacorogna used the modelling abilities of Scor's ESG to price a CDO it was considering investing in. The AAA tranche was referenced to a portfolio of predominantly BB bonds. "I ran our model on this structure and reported that I did not believe it was possible to build the AAA tranche on such an underlying because our model generated very strong dependencies if there was a stress, which would affect the AAA tranche." On this basis, Scor declined to invest in the CDO and has had almost no exposure to the subprime market.

Meanwhile, companies already using more sophisticated ESGs believe these systems can have an impact on their bottom line. Talanx has a 25% stake in DFA, and as a result gets privileged access to the development of its ESG. "Due to the close relationship Talanx has with DFA, we have an in-depth insight into the ESG calculations," says Stahl. "This enables us to do better risk measurement and therefore better hedging, so we are able to make better profits."