Print this
Adapting Value At Risk

Corporates and pension funds are trying to transplant value-at-risk techniques from the banking world. Will the graft take, or is VAR too rarefied a thing for the real world?

By Karen Spinner

In recent months, multinational corporations and pension funds have been depicted as innocent lambs and placid cows wandering through a lush and dangerous jungle straight out of central casting. Predators lurk everywhere, from the serpentine forces of market risk to wily broker-dealers more interested in bagging big game than guiding lost little lambs to safety.

How can corporates and institutional investors find shelter in this fierce Darwinian world? For many consultants and systems vendors, the magic answer is value-at-risk (VAR).

Wildly successful in the banking world, in no small part due to JP Morgan's tireless efforts, VAR has begun to pique the interest of top managers in the corporate sphere. The hope is that VAR, de-bugged through its implementation at countless banks and dealers, could be a user-friendly way of satisfying boards of directors' appetites for concise reporting as well as impending regulatory requirements. Late last year the SEC itself included VAR as one of three methods by which companies may be required to provide a quantitative disclosure of derivatives-related market risk.

That's all well and good, but corporate risk managers who are eager to integrate VAR measures into their risk management operations are bound to be disappointed. Here's the sad but undeniable truth: VAR as practiced in most banks is a woefully inadequate way to calculate the intrinsic risk of derivatives purchased as part of a corporate hedging program.

"Corporate hedging is a very different activity from the sort of trading that most banks practice," says one New Jersey-based assistant treasurer. "The purpose of a derivatives portfolio at a bank is usually to speculate according to a particular set of market views. But the purpose of derivatives held as hedges is to offset exposures and reduce earnings volatility."

Therein lies the problem: in order to get a complete picture of a company's portfolio-wide risk profile, it is necessary to somehow include forecasted sales, balance sheet and economic exposures into the VAR equation. Boiling these complex business elements down into a series of cash flows or basic financial instruments is just one of many problems treasury managers will encounter on the road to implementing VAR. Similarly, pension funds, which are engaging in financial transactions to make a profit, do not resemble banks in their operations or their outlook. Where banks are interested in P&L volatility over the relative short term, pension funds are much more concerned with their investments' long term performance relative to a benchmark.

The first challenge, then, for both corporations and pension funds interested in adding VAR to their quiver of risk management arrows, is to take a hard-headed look at VAR and to decide whether or not it really is suitable for their needs. If the answer is "Yes, but...," then the second, and probably greater, challenge is figuring out how to adapt VAR for use in a new environment. Because VAR is largely untried outside of banks and brokerage houses, this process of adaptation may take more work than anyone now imagines.

Big Misunderstanding

Certainly there are enough starry-eyed risk management gurus around today to constitute an official trend. The consultants are both reacting to demand and feeding the VAR frenzy. "Over the past six to nine months, we've had a lot of clients asking for value-at-risk," says Stephen Wiehe, vice president of New Jersey-based Multinational Computer Models, which offers a VAR module as part of a larger package of risk management software for corporate treasuries. "VAR is more likely to take hold first among corporates than among plan sponsors, but give it time," says a New York-based risk management consultant. "Pretty soon everyone with a derivatives exposure will use some sort of VAR."

But VAR is a sophisticated technique that requires a learning curve to master. Before VAR makes a place for itself among corporates and institutions, it will have to be understood. Thus far, however, VAR's ascendancy within corporate and institutional circles has been accompanied by widespread confusion, ignorance and misunderstanding. Although boards seeking a concise one-number risk report may encourage their financial officers to explore VAR, consultants say that board members are often the first to dismiss it as too complicated when they try to read the fine print.

Corporate financial officers are often no better informed. Indeed, many of the twenty or so corporates interviewed for this article misunderstood basic VAR concepts. Several, for example, explained that they planned to use VAR to measure their maximum losses in a "worst case" scenario. While VAR can be used to measure the potential variance of P&L over a particular period of time with 95-99 percent certainty, worst cases typically fall outside those certainty levels. Most risk managers use stress testing to measure risk under extreme scenarios.

This distinction, while subtle, indicates that VAR is moving from its humbler identity as a particular matrix-based model to a buzzword synonymous with risk management in general. That's why it is critical for corporates to define their own terms rather than blindly using RiskMetrics or any other generic VAR-flavored model. "VAR is fast becoming an umbrella term," admits Jacques Longerstaey, vice president of JP Morgan and patron saint of RiskMetrics. "As value-at-risk moves into mainstream corporations as well as banks, the question for risk managers becomes: 'How do you define value?' and 'How do you define risk?' Naturally, firms that have different definitions of value and of risk will see value-at-risk as wholly different concepts."

Corporate Conundrums

Adapting VAR to corporate risk exposures is fraught with difficulties. VAR works best with financial instruments. While it is possible to represent a financial institution's P&L as the sum of a global, firm-wide portfolio-which can then be analyzed via value-at-risk-a corporate's portfolio of hedge transactions does not begin to describe its total risk exposures. Most hedges can be matched to an offsetting exposure, represented by business contracts, cash flow forecasts or assets. The problem, of course, is that these complex uncertain exposures can't be neatly plugged into a VAR equation.

Consider the following types of exposures and the implications they may have for VAR:

Contractual exposures. Contractual exposures are simply future business-related contracts that the company has entered into. An American company that has signed an agreement with a German company to buy ten tons of widgets in one month's time for Dm2 million will have a short deutsche mark position. The treasurer may wish to offset the exposure by buying a forward contract for Dm2 million, thus locking in the Dm/dollar exchange rate it will pay for the widgets.

How would you plug this exposure into a VAR calculation? The contract is booked and thus very likely to occur. Entering the financial hedge without some representation of the offsetting exposure would generate a VAR number that would distort the true risk of the position. To better assess that risk, it's necessary to create a pseudo-trade representing the underlying exposure being hedged.

Anticipated exposures. Anticipated exposures are transactions that a company can forecast with reasonable accuracy. If a firm has sold 2,000 gallons of cranberry juice to the same Mexican restaurant chain every April for the past five years, and the Mexican firm has paid in pesos every year, then it is reasonable to assume that the same transaction will occur again this year. Many firms choose to hedge anticipated exposures. Often, however, they will not hedge these anticipated exposures 100 percent, because there is a chance that "something will suddenly come up" and a few of these anticipated deals will fall through.

For the purposes of VAR calculations, the challenge here is again to create pseudo-trades that represent anticipated trades. Because these trades are not absolutely certain, there is considerable debate over whether the notional values of these pseudo-trades should be weighted according to their probability of occurring. For example, if a forecast sale has, say, a 60 percent chance of occurring, then the notional amount of this sale could be multiplied by 0.6 when assigning it a cash flow value, or the sale's face value could be multiplied by a delta figure. Another approach is to assigned an elevated volatility figure to account for the exposure's "theoretical" status. The forecast cash flows could be identified as a special asset class which would, in the VAR matrix, be assigned higher volatilities.

Another issue with regard to forecast cash flows is that assigning probabilities to these forecasts is more an art then a science. "We are also concerned because we do not have enough data right now to be completely comfortable with the probabilities we have assigned to our forecast cash flows," says one Illinois-based FX manager. Thus it is difficult to determine how much to "count" these exposures in a firm-wide VAR portfolio.

Balance sheet exposures. Some companies go beyond hedging contractual exposures and anticipated cash flows to hedge the value of the assets on their balance sheet. For example, a highly sophisticated company that owns a large manufacturing plant in Indonesia may want to use a relatively complex derivative to hedge the risk that political instability or a local market crash might render the plant drastically less valuable. Of course, to include this sort of hedging activity in a VAR calculation, it would be necessary to find a meaningful representation of the plant itself as either a complex "trade" or a bundle of cash flows.

There are other features of corporate hedging that can bring standard variance/covariance VAR to its knees. Many corporates favor options on contractual exposures because they usually can receive hedge accounting status. Plain vanilla VAR, however, is built in part around the assumption that instruments' payoff profiles are symmetrical. The assumption of symmetry is a shortcut designed to make VAR calculations faster; under this scenario, for example, purchased options can be assigned a negative P&L minus the premium. Some very large institutions with many options on the books rely on the "portfolio effect," meaning that false-negative P&Ls will balance out false positives.

Standard VAR includes a second shortcut: the assumption that market factors are normally distributed. But reams of research studies indicate that extreme market movements are more common than normal distribution would indicate. As a result you could end up losing much more in an extreme market move than your VAR number predicted with a 95 percent certainty.

Another problem: many corporates look at risk on a currency-by-currency basis. This means that they would tend to calculate VAR separately for each currency group. In this case, portfolio-wide market correlations could not be considered and so the firm's total additive VAR would likely be overstated.

Finally, VAR can give corporates a single "risk number," but the question remains: what can you do about it once you get it? A company with a VAR of $50 million may have no idea whether this is "good" figure or a "bad" figure, and-if it's bad-how it can be improved. "If we run VAR as the SEC suggests, simply looking at all our derivatives without somehow including their matching exposures, we get a number, but what in the world can we learn from it?" asks one New York City-based treasurer. "We are looking for some way of using VAR to get information to help us decide how much to hedge and what hedges to use, etc."

The Short List of Solutions

Despite the obvious difficulties of adapting VAR to a corporate environment, consultants and systems vendors are coming up with a number of innovative solutions.

1) New types of assets. One of the first steps required to make VAR useful in the corporate sphere is to create a whole new framework of new asset classes that can be used to represent the various exposures described above. A number of software firms have already tackled the problem. Gabriel Bousbib, vice president of Reuters' new risk management arm, adds that corporate exposures of any kind can now be handled by some robust middle-office system that let users custom-configure new asset classes. He cites New York-based Sailfish, a Reuters company, as one firm that can provide this sort of representation for corporate exposure on a large scale. Wiehe of Multinational Computer Models adds that his firm's systems also allow users to create custom asset classes.

2) Delta VAR. Financial Engineering Associates (FEA) has also recently released a VAR module that allows corporate users to create a "native trade" to represent a physical asset, such as oil, that a corporate is holding and hedging. FEA has also developed a new technique called Delta VAR (DelVAR) that promises to allow corporates to look at various hedging options and consider whether a particular strategy will increase or decrease their entire portfolio of hedges and exposures. A trade's VAR-improving potential can be expressed in a number of different ways, including VAR improvement in dollars per dollar of capital employed for each of several trades. (See the column by Mark Garman on page 52.)

3) Historical curves. Some users have also tacked two of VAR's most compromising shortcuts: that market factors and instrument payoff profiles are normally distributed. To get more accurate VAR numbers, some firms are using actual historical market factors from a particular period of time to create hundreds of scenarios. These scenarios are then used to price each instrument in the portfolio over and over again. "This is a brute force method that requires a lot of computing power," says Andrew Aziz, senior financial engineer at Toronto-based Algorithmics. "It does eliminate two of the most onerous assumptions of VAR for corporates, the normality of market factors and the linearity of payoff, but it can be time-consuming if the portfolio is particularly large."

4) Monte Carlo adaptations. Aziz describes another sort of VAR he has seen used by corporates, which he calls "an intermediate approach between standard covariance VAR and the historical approach." Instead of assuming that all payoffs are linear, the technique involves using Monte Carlo analysis to sample certain portions of the covariance matrix in order to generate scenarios. These scenarios are then used to price individual instruments. "While using the covariance matrix still assumes that market factors are normally distributed, the repricing of all instruments under each Monte Carlo scenario effectively neutralizes the symmetrical payoff assumption of standard covariance VAR," he explains. "As a result, intermediate VAR is safe for use on options."

The upside is that options and other assymetrical instruments are priced without the shortcuts in variance/covariance VAR. The downside, of course, is that Monte Carlo simulations must be run many times to generate an accurate figure, particularly when a whole portfolio is under consideration. Therefore the Monte Carlo method still requires considerable computing power.

5) CVAR. Emcor Risk Management Consulting Inc., an Irvington, New York-based consultancy, has developed its own four-part methodology to translate the complexity of the corporate environment into a VAR format.

First, companies run a VAR calculation on their business exposures-which may include contractual exposures, forecasted cash flows, economic exposures, etc. This number represents their total "business risk," a calculation that may be used to help firms compare, say, the risk of building a plant in Europe versus building a plant in China. Next, companies run a VAR on all the financial transactions they have used to hedge their business exposures. This figure is useful, both from a reporting perspective and as a reality check, should top management ever decide to liquidate a derivatives portfolio en masse.

Third, companies run VAR on the entire portfolio, including both business exposures and financial hedge contracts. If a firm's hedging program is working as it should, then the portfolio-wide VAR figure should be less than either the business-VAR or the financial-VAR. "If this is not the case," observes Emcor managing director Robert Baldoni, "then there is probably some flaw in the firm's hedging strategy, such as inappropriate proxy hedging."

Finally, Baldoni recommends that firms run VAR on their credit exposures to various financial institutions. Considering potential positive P&L movements in, say, options, makes it possible for firms to gauge how their exposures to various counterparties are fluctuating and where they are relative to their limits.

All these solutions may help corporates adapt VAR to their specific needs. The goal of the value-at-risk approach is to give board members a simple number that defines risk. Ironically, however, all the proposed solutions to the corporate VAR problem require a more complex level of understanding-and may cause board members to give up on the approach altogether. "Because VAR, with its many theoretical assumptions, is trickier to explain than the more straightforward scenario analysis, it is particularly difficult to convince boards that they should accept a VAR number as a concrete basis for decision-making," says Jeffery Wallace, a managing director at Greenwich Treasury Advisors. "Therefore, corporate treasurers should make very sure that there is a clear connection between their VAR methodology and specific business objectives."

VAR is probably here to stay, but as a dynamic, evolving concept rather than as a fixed model. "Initially we were skeptical, but I now believe that VAR will be a mainstream risk management methodology for years to come," says Multinational Computer Model's Wiehe.

Even JP Morgan is tinkering with RiskMetrics' classic VAR so it can encompass more instruments and reflect a more realistic picture of any entity's risk. Clearly VAR is coming to the real world, where in each unique habitat-be it the corporate treasury, a trading floor or a pension plan-the forces of natural selection can weed out those models not grounded in practical decision-support concerns.

Institutional Investors Take A Sniff

By Karen Spinner

Although VAR has yet to take over the plan sponsor community, it's clearly beginning to make headway. "We have seen a great deal of interest from pension funds and money managers with regard to VAR, particularly VAR adjusted for return," says Katherine Condon, a managing director at Bankers Trust. Paul Kaplan, vice president and chief economist at Ibbotson, concurs: "We are seeing a 'trickle down' effect from the treasury side to the plan sponsor side. The treasurers are interested and now so are the plans."

Plan sponsors and fund managers have a different set of problems with VAR than their colleagues on the treasury side. The obstacles do not necessarily relate to what instruments are in their portfolios. Instead, they involve a reluctance to grapple with a complex and theoretical risk management methodology.

Many funds find the volatility and correlation matrix assumptions that go into a VAR calculation much too confusing. As a result, many consultants prefer to offer risk management solutions that do not outstrip clients' levels of expertise. For example, Frank Russell Co., the large Washington-based pension fund consultancy, does not actively encourage its clients to pursue VAR. "The many assumptions that go into the calculation of VAR are not always apparent to the prospective user," says George Oberhofer, director of fixed income research for the firm. "For this reason, we prefer that clients take a slightly less elegant, less aggregated but more intuitively understandable approach to risk analysis, namely looking at multiple scenario analysis, risk factor by risk factor. There is no reason VAR should not be looked at as well."

Capital Management Sciences, a software firm that specializes in analytical systems suitable for fixed income derivatives, is another firm that isn't pushing VAR as a single solution to risk management. Terri Geske, vice president of product development, explains that while the firm has developed a number of new products that give users the ability to monitor their portfolios' sensitivity to a wide variety of market factors, including volatility, they do not have a particular module named VAR. According to Geske, VAR is quite complex and has less immediacy for pension funds than, say, scenario analysis, which has fewer moving parts and enjoys greater acceptance among plan managers.

Another set of challenges for VAR analysis of sponsor portfolios involves institutional investors' historic obsession with return-based benchmarks. Most plan sponsors and money managers evaluate their returns by comparing them to various benchmarks-an index, a 'model' portfolio or even, in the case of pension funds, a representation of their liabilities. The issue here, then, is how VAR can be adapted to tell pension funds and money managers how their risks may relate to their benchmarked return targets. Bankers Trust's Condon explains that many of her firm's fund manager clients are interested in forms of risk measurement that bring returns into the equation. "Money managers are not interested in hedging all their risk away," she explains. "They want to take risks. The important thing is to balance return expectations with risk taken."

One way in which VAR and return are linked is in BT's much-discussed RAROC (Risk Adjusted Return on Capital) product. RAROC, which has been adopted by some very large plan sponsors, money managers and fund managers, analyzes portfolio risk by market type (e.g. equity, interest rate, etc.) and by account. BT uses its own proprietary correlation and volatility matrices to run Monte Carlo simulations which, in turn, are used to value the different instruments in the portfolio. The approach is generally more sophisticated than JP Morgan's VAR methodology and allows for better valuation of the asymmetrical return profiles of options and other products.

The valuations RAROC provides are based on a one-standard-deviation move over a one-year holding period, which stands in marked contrast to RiskMetrics' one- and 90-day time buckets. The longer time horizon was deemed more appropriate for the investment horizons of RAROC's institutional clientele, and RAROC's success illustrates that VAR is under some circumstances very adaptable.

Andrew Aziz, senior financial engineer at Toronto-based Algorithmics, describes another methodology that could work for pension funds. "They may wish to look at the difference between the VAR of their portfolio and the VAR of a benchmark or with respect to their individual returns. The goal would be to outperform the benchmark's return while undercutting its risk." Aziz notes that Algorithmics offers a set of models known as "benchmark VAR" based on the concept of "regret," which is similar to classic VAR but instead helps users to determine the risk of a particular investment underperforming its benchmark.

Why One Corporate Treasury Doesn't Use VAR

By Karen Spinner

Merck, the New Jersey-based drug giant, has managed to build its own custom risk management program without resorting to variance/covariance VAR...and things are working out just fine. According to Stephen Propper, the firm's director of FX, when Merck first began looking at VAR, they were looking for a methodology that could measure risk within a business-specific context. "Merck has a long-term commitment to research and development, and so, to ensure that a strong dollar does not adversely impact Merck's ability to continue its investment in R&D, we generally layer-in hedges over a three-year period," says Propper. "However, Merck does not fully hedge its sales, but rather considers natural offsetting exposures and self-insures a portion of the exposure.

To arrive at a desired hedging strategy, Merck uses Monte Carlo analysis to model the dollar value of Merck's foreign cash flows and the performance of alternative hedging strategies. The model evaluates the effectiveness of hedging strategies under a range of possible exchange, rather than providing only the maximum loss at a particular confidence interval. The model also permits Merck to stress-test results by forcing devaluations or increasing volatility.

A classic VAR approach, he explains, did not match up with business needs for a variety of reasons. First, VAR is not compatible with a long-dated program where hedges are layered-in over time. Indeed, companies often do not hedge 100 percent of their exposures for many reasons. For example, a corporate treasurer might reasonably conclude that it doesn't make sense to hedge more that 50 percent of long- dated forecast cash flows simply because there is a chance that these cash flows do not pan out; in that case, hedging 100 percent of future cash flows, no matter how far off in the future, could actually increase risk.

And if one calculates VAR in a partially hedged corporate environment where the firm's "portfolio" includes all actual and forecast exposures plus existing hedges, there will inevitably be "leftover" exposures which will then translate into an oversized VAR. Says Propper, "In this case, VAR doesn't provide information that can be used to improve its hedging program."

Nor, he explains, would it be appropriate to simply run a VAR calculation on the financial hedge transactions alone: "Hedges are meant to stabilize cash flows; when considered with the underlying business exposures, they neither 'earn' nor 'lose' money. A VAR performed on hedge transactions without their matching exposures implies a speculative risk that really isn't there. It suggests that the institution will benefit or suffer from these deals' P&Ls when, in reality, any profit or loss will be offset by an equal and opposite gain or loss on the business side."

Propper adds that while this sort of analysis is definitely appropriate for a portfolio of speculative trades, such as one might find at a bank or a profit-center-type treasury, it does not necessarily work in a hedging situation, the approach used in Merck's treasury. Merck, he explains, uses internally developed models that incorporate Monte Carlo analysis to determine the performance of alternative hedging strategies under a range of possible market conditions.

But what about the SEC's exposure draft, which suggests a variety of ways to express the risk of any firm's derivatives portfolio, including VAR? "The SEC's intention-to request that corporates provide more meaningful disclosure-is laudable, but the methodologies they suggest may not provide useful information from the potential investor's viewpoint," says Propper. "VAR disclosure is not relevant for a fully hedged position, since at expiration the company will be fully protected. Furthermore, comparisons of VAR across companies would be difficult, since alternative VAR models and different capital market assumptions can produce different results for the same exposure. Nor is the tabular listing method particularly useful, since it will require a high volume of detailed information that will be difficult to analyze. Furthermore, since the table categorizes outstanding derivative positions at year-end, it would not provide information on derivative activities over the course of the year."

Propper suggests that any "quantitative" disclosure can be augmented by a qualitative disclosure of the goals and methodology of a firm's hedging program. And he emphasizes that the SEC's draft is a step in the right direction that he hopes will be developed through further research and dialogue.

JP Morgan Responds To Its Critics

By Karen Spinner

When execs at JP Morgan developed the RiskMetrics VAR model and placed the data required to run it onto the Internet, they had no idea they were letting a powerful genie out of its bottle. Demand was fast and furious, first from banks and later from corporates. The concept was successful beyond their wildest dreams, but this success led to questions, particularly from treasurers looking for a quick, easy-and cheap-way to incorporate VAR into their risk management systems. As a result, JP Morgan is taking a striking new approach to RiskMetrics that will satisfy clients' demand for software as well as address some of VAR's limitations.

"Many corporates and non-financial firms are looking for software to implement VAR," says Jacques Longerstaey, a vice president at JP Morgan and well-known VAR expert. "In order to meet this considerable client demand, we are changing strategy. Although we have said before we do not want to be in the software business, we will soon offer a RiskMetrics software package to Morgan's clients. They need a simple application to use RiskMetrics effectively. That application is an Excel-based calculator which we hope will be a very effective entry-level tool."

Upcoming versions of RiskMetrics will also address another issue that has lately concerned options users: the assumption inherent in RiskMetrics-style VAR that instrument payouts are symmetrical. Some software vendors have begun addressing the drawbacks of this assumption by designing VAR models that crunch unheard-of quantities of historical market data. Others are using Monte Carlo analysis on covariance matrices to get around the problem. Both these solutions, however, require tremendous computing power and/or long periods of time.

Longerstaey explains that JP Morgan is in the process of testing a new model that effectively eliminates the assumption that instrument payouts are symmetrical without sacrificing the computational elegance and speed that RiskMetrics is known for. "Our goal is to have an effective model that can produce accurate results in a reasonable amount of time," he says. "While Monte Carlo analysis is one way of solving the symmetrical payoff problem, we want our system to work fast and be useful whether or not an end-user has minimal PC-type technology."

Longerstaey is also addressing corporate critics who have complained that RiskMetrics' use of cross-market correlations make the methodology fundamentally incompatible with strict currency-by-currency and market-by-market hedge accounting. VAR, strictly calculated currency by currency, then added to create a single portfolio-wide number, may be overstated. "There is a difference between managing risk and managing accountants," he says. Longerstaey adds that RiskMetrics was designed to look at economic risk, separate from risk as an accounting concept: "It is flexible enough, however, for those users who want to use an additive risk approach, without correlations, to do so."

Longerstaey is also responding to complaints about VAR's assumption that returns are normally distributed: "The point of VAR as expressed through RiskMetrics was to create a methodology that was simple and not computationally intensive. The assumptions about normally distributed market factors and instrument payouts were made so VAR could be calculated quickly. Of course everyone must consider trade-offs between accuracy and computational intensity."

So stay tuned. RiskMetrics, like VAR itself, will continue to evolve and reinvent itself.