Published on Fortnightly (http://www.fortnightly.com)

Home > Printer-friendly > Business & Money

Business & Money

Obtaining a position measurement in energy markets has become more complex and has increased financial risks for integrated utilities.

"What's your position?" The answer to that simple question in today's energy markets is anything but simple. In fact, answering this question may be the single most difficult challenge faced by a fully integrated energy firm in its efforts to manage risk. Position measurement, and therefore risk management, in today's deregulated energy market is complicated by the fact that weather, fuel costs, outages, transmission availability, embedded optionality, and a host of other interrelated factors all dramatically affect position on a real-time basis.

Without an accurate, granular, and unbiased estimate of position, risk management and generation optimization are impossible. Moreover, without consistent and timely position discovery, hedging programs cannot safely be implemented-and, of course, the efficiency and performance of the trading and marketing team cannot be evaluated.

Discovering, estimating, and valuing one's position requires a rigorous methodology and a computational architecture to back it up.

In late 2000, Cinergy Corp., the Cincinnati-based utility and one of the largest electricity trading companies in the United States, undertook a comprehensive review of its risk management methodology and systems. Cinergy was fully aware of the volumetric risks and embedded optionality in its merchant energy portfolio. However, its internally developed client-server system was not sufficiently up to the task of handling all of the complex data capture, valuation, storage, and presentation nuances of its complex deregulated portfolio.

After conducting an exhaustive review of trading systems in late 2000, Cinergy concluded that no silver bullet existed. Available systems were incapable of managing and pricing positions with hourly granularity. The best systems still relied upon valuation models derived for financial markets that lacked the unique volatility characteristics of the power markets.

Worse, the products required non-vanilla deals to be valued outside the core system and plugged back into the core portfolio, creating the classic apples-and-oranges dilemma. Finally, the dated technical architecture of all available alternatives precluded the ability to scale to meet Cinergy's requirement to calculate and report portfolio results on a real-time basis. In short, the perfect system didn't exist and would need to be built.

Real-Time Variables

On the surface, meeting such an objective for any deregulated utility seems relatively straightforward. Deal terms and conditions are clearly spelled out in contractual documents. The vast majority of deals are for standard blocks. Utilities have detailed historic records of loads, and pricing in forward markets is becoming increasingly transparent. Finally, advances in computing power have enabled more rigorous analytical modeling to handle complex deals with variable load- and path-dependent characteristics.

However, a more thorough review of the dynamics of the power market reveals an additional layer of complexity that cannot be ignored if one is to get a true estimate of position and risk: Position measurement is complicated by the fact that weather, fuel costs, outages, transmission availability, embedded optionality, and a host of other interrelated factors all dramatically affect position on a real-time basis. Moreover, any position estimation and valuation model also needs to transition between various states of the market without having any discontinuities or inconsistencies. Finally, enormous computational and data storage hurdles must be overcome to deliver hourly risk metrics in real time. A viable solution needs to meet the following minimum requirements:

Other Factors

Wall Street typically has used one-factor, short-term interest rate models to drive its fixed-income valuation models. In the power market, the most important factor to consider is weather-the single strongest determinant of price, load, generation utilization, and transmission availability. Higher temperatures increase load, which in turn is served by higher marginal cost generation. As the load/available generating capacity ratio approaches 1.0, price-stack-based economics give way to scarcity-pricing, gaming, and other factors that are possible in an environment characterized by inelastic supply and demand curves. If modeled properly, a position estimation and valuation model will transition between various states of the market without having any discontinuities or inconsistencies.

Calibration Is the Key

Models must be continually recalibrated to existing market conditions. The goal of econometric modeling has always been to design models that use as few observable parameters as possible to accurately describe real-world conditions. A properly calibrated energy pricing model generates option prices that are identical to observed market prices of all options, regardless of whether they are in or out of the money. (Contrast this to the Black-Scholes model, which uses a "smile" function to imply that price distribution varies at different price levels.) The basic premise of a weather/option-calibrated model is that weather is the driving force defining the general shape of price distribution, but that the market continually reveals its estimate of the market level, amplitude, and skewness of price distributions based on changing forward markets and option quotes.

Simulations of price, load, and generation are not independent and should share a common weather pattern within each simulation. If temperatures are high, we generally expect load, prices, and generation to be high as well. If one treated these variables independently, the true cost of unforced outages and transmission constraints would be grossly understated.

Just as importantly, all commodities and regions should have simulations that are derived from fully integrated weather data. Common sense tells us that it's unlikely for temperatures to simultaneously be 10 degrees above normal in the Midwest and 10 degrees below normal in the East. This underscores the need for our estimates of weather, prices, and loads to be plausibly related on a geographical basis. If position and valuation models are not driven by common weather patterns across regions, then the value of spread options and transmission products may be overestimated. Similarly, the viability of interregional hedges may be either understated or overstated if other stochastic pricing assumptions are used.

The Valuation Model

Once you have the necessary weather, generation, and load data in place, you need software that can process that data in a rigorous, consistent, and market-relevant fashion. First and foremost, you need to ensure that all transactions are valued using the same underlying assumptions regarding price distributions. Transactions valued outside the core system ignore the inherent correlation effects with the remainder of the portfolio. This severely limits the quality of risk analysis, particularly Value-at-Risk and sensitivity analysis.

Power traders know that traditional parametric-based position estimation and valuation methods are inappropriate for electricity markets, particularly when risk is at its highest. Valuation and position estimation methodologies must be able to smoothly transition from a marginal cost pricing paradigm into a more volatile paradigm. When sufficient excess generation capability exists, price-stack-based economics work fairly well because the supply curve is fairly elastic. However, once load begins to approach total generation capacity, the potential for much more dramatic increases in price exist for small increases in demand. Valuation methodologies must accommodate the greater likelihood of price spikes to occur simultaneously with load spikes.

When implementing a valuation methodology, it is also important to ensure serial or chronological integrity. Stochastic processes that simulate pricing paradigm shifts are inferior to prices derived from actual weather patterns because they are incapable of modeling path-dependent characteristics of the market.

For a real-world example, consider the many position- and price-related effects of a sustained period of high temperatures. Chronologically based algorithms are capable of handling the progressive pressure on prices, loads, generation efficiency, and transmission availability as a heat wave persists. Conversely, stochastic methods fail to do so, or must be modified by less precise and arbitrary factors in an attempt to represent these phenomena.

Computational Efficiencies

Possibly the most important consideration in valuation computation is keeping up with the incredibly fast pace of power trading. Valuation algorithms should be designed to generate accurate results with fewer computations. The need for a scalable solution-one that can process the enormous volumes of data required to accurately value positions with hourly granularity and serve up results in a real-time trading environment-is obvious. And, of course, throwing hardware and memory at a problem is no substitute for robust, efficient algorithms that provide far greater increases in performance. Using smart algorithms enables a dramatic reduction in processing needs-far less than 1 percent of a naïve brute-force approach. Moreover, the final crucial step in implementing an optimal position and risk reporting system is to design a reporting structure capable of delivering real-time summarized and detailed risk metrics to the user.

There are sound business reasons why traders and risk managers will want to see their data aggregated/disaggregated in multiple views depending upon the particular task at hand. Hourly traders are rarely concerned with financial or forward positions, while daily traders look only a few days ahead. It is common for term traders to view their positions on a rolled-up monthly basis, but they want the ability to drill down instantly to see details of their positions at the daily or hourly level. Developing a system that can meet these disparate needs in real time without becoming overwhelmed by data storage or processing requirements can be achieved only by following a rigorous design standard and applying advanced supporting technology.

First, all transactions should be valued at an hourly level for the entire term of the transaction. Regardless of the contractual nature of transactions, each deal must be valued at a granular level, preferably hourly, or in some cases sub-hourly. The more granular your positions, the more flexibility you will have to roll up and summarize the position for varied types of analysis. Taking a shortcut and marking all hours of a block product to a block price makes computation easier; however, one's ability to disaggregate the position and reaggregate along another dimension at a later date will be impossible.

Hourly positions must be maintained in forward months-but only for the attributes that matter from a position and risk-management perspective. For example, hourly representations of positions in forward months by location and pricing point are critical, while a monthly representation of position by counterparty is generally sufficient. A thoughtful review of which dimensions are important on an hourly versus a rolled-up view in forward months can reduce processing and storage requirements to a fraction (less than 10 percent) of a full hourly representation of all positions.

Second, to meet the demanding requirements of long-term market makers and traders, summarized positions should be maintained for all months. Most queries of positions beyond the spot month will be for monthly records. Keeping data in this rolled-up format allows data to be retrieved extremely quickly without incurring expensive roll-up operations with each query.

Finally, where possible, deals should be classified as specific delivery products that are generally consistent with products quoted in the market. This ap-proach has several benefits. Deal entry and position determination are streamlined, eliminating the most common cause of deal-entry error. A trader's position can be represented as a collection of different products that are denominated in the units where he can trade. For example, rather than classifying a series of trades as just off peak, a trader is aided by the classification of three separate off-peak products that may include 5x8, 2x16, and 2x8. When the trader sees the position represented in this manner, he or she quickly can do mega-watt/megawatt-hour conversions and monitor every position represented in the units in which trades typically will be executed in the market. Trading products can be matched with transmission products at a rolled-up level without having to worry about data that does not match up.

Technology has evolved to the point where enormous amounts of data can now be managed down to the smallest detail. The size and sophistication of today's power portfolios demand systems that can measure, monitor, and control position and risk management. Scarcely one year after embarking on the project, Cinergy has successfully implemented the methodology and technology needed to manage the sophistication of today's deregulated power markets. Only by applying next-generation technology and position management practices can today's energy market players have the confidence to answer the tough question: "What is your position?"

Business News Bytes

Progress Earnings Held Back by Weather

Progress Energy Inc. executives said they could not say when the synfuel tax credits/IRS audit issue would be complete. Progress reported third-quarter 2003 ongoing earnings of $306.2 million, or $1.28 per share, compared to $333.5 million, or $1.54 per share, at the same time last year. Consolidated's net income on a GAAP basis was $319 million, or $1.34 per share, compared with $151.9 million, or 71 cents per share, in the third quarter of 2002. The Thomson First Call consensus was $1.40 per share. Synfuel issues aside, major negative effects on earnings in the quarter were unfavorable weather, Hurricane Isabel costs, increased pension and benefit costs, share dilution and lower industrial sales. These factors were somewhat mitigated by increased sales of natural gas and customer growth.

FPL To Buy Enron Wind Assets

FPL Group Inc. agreed to purchase 130 MW of wind generation projects from bankrupt Enron Corp. for $80 million. The wind facilities, all of which are located in California, include the 40-MW Cabazon facility, the 16-MW Green Power projects near Palm Springs, the 18-MW ZWHC and the 7-MW Victory Garden Repower projects near Tehachapi. FPL has also agreed to buy out Enron's 50 percent ownership interest in the 77-MW Sky River project and the 22-MW Victory Garden Phase IV project. FPL Energy owns 50 percent of both Sky River and Victory Garden Phase IV projects, according to the release. The deal is subject to regulatory approvals and acceptance at a December bankruptcy auction. FPL will receive a breakup fee of $3 million in the event that it is not the confirmed buyer on all of the agreements. FPL Energy hopes to close the acquisition by early 2004. With the exception of Green Power Partners I LLC, all of these projects sell all of their output to Southern California Edison Co. under long-term contracts.

CMS To Delay Construction at LNG Terminal

CMS Energy Corp. received approval from FERC to delay expansion of its LNG plant in Lake Charles, La., according to an Oct. 22 Reuters report. In addition, FERC gave the nod to CMS's request to buy power from Entergy Corp. rather than expand on-site generation facilities.

 

Articles found on this page are available to subscribers only. For more information about obtaining a username and password, please call our Customer Service Department at 1-800-368-5001.

Source URL: http://www.fortnightly.com/fortnightly/2003/11-0/business-money