Mobile Edition
Print
Get the Mag
Weekly Updates1. INTRODUCTION
In its September 2007 Report 23, Prices and Trade in a Globalizing Natural Gas Market, the Stanford Energy Modeling Forum (EMF) has once again pursued a challenging agenda seeking to combine leading edge thinking on economic theory with practical application in order to provide insights that actually improve both energy policy and corporate strategic decisions. As the executive representing Suez Energy North America,1 Inc. in the study group, the emphasis on getting from theory to practical insight was paramount. GDF Suez, after all, is a company with large positions within the LNG value chain for whom such insights are critical to strategic success. At the same time, industry practitioners must recognize that sound theory is an essential pre-requisite for practical success.
The purpose of this paper is to advance some principles for practical application of energy models that were used during EMF 23. These principles, which are arguably more philosophical than technical in their nature, include an unbending decision-focused perspective, recognizing and embracing uncertainty, fact-based rather than subjective, able to navigate the complexity of systems, and unequivocally recognizing the important distinction between insight and unattainable precision. Models grounded in these principles have in fact been applied in the energy industry and the best of them have been systematically successful in helping companies make better decisions. The principles offered here are designed to foster the assistance of better decision making by models and modelers.
It is said that economics is the science of negative feedback. (2) Prices always adjust to equate supply and demand. Quantities and capacity additions adjust in the face of prices. Proper market modeling and interpretation must always represent the negative feedback in markets. Although there may be positive feedback and amplification for some period of time, and protracted positive feedback often catalyzes overconfidence in models and results, over the medium and long-term there is significant negative feedback in economic systems. Markets tend to undermine the apparent stability of extrapolated trends. How many analysts in the month of July 2008 said that oil and gas prices were never going to drop from $140/BBL and $13/MMBtu because the forwards were up in the stratosphere and avowed energy experts were singing dirges about the limits of resource availability? Many--and while a few may have dissented from the conventional wisdom, the majority did not. Less than three months later in October 2008 oil and gas prices have declined dramatically to levels predicted by models with negative feedback.
The most recent experience on Wall Street, characterized by the severe impairment in the value of securitized debt portfolios far exceeding anything that the models for these securities anticipated, has sparked further commentary on the failure of models. (3) However, the more thoughtful commentary on the subject seems to have recognized that the present difficulties have not so much been a failure of models but rather a failure of thought. Models provide useful insights but do not provide a substitute for thought and must always be applied with careful attention to their limitations. The illusion of precision that from a practical, fact-based perspective was not realistic was arguably a significant cause of model failure. The amplification of model risk through errors that are compounding rather than diversifying--always an issue in navigating the complexity of systems--was another significant cause.
There are perhaps useful lessons that energy modelers can learn from having observed the Wall Street experience even though the models that "failed" were not energy models per se. Brief summaries of each of the five aforementioned modeling principles and how they apply to prices and trade in a globalizing natural gas market follow. (4)
2. DECISION FOCUS
Several market trends relevant to future company and government infrastructure investment decisions were highlighted by the study, which consequently turned it into much more than just an "academic" exercise. Of most interest from a practitioner's perspective were those trends on which the several models included in the study came to a consensus. One important example of this is the conclusion that by 2020, given the geographical configuration of the world natural gas resource base, LNG imports will capture a greater percentage market share in North America than in Europe. (5) Along with this comes the related conclusion that North American delivered gas prices will be higher than in Europe. (6)
The evolution of this trend would not be apparent at all from observations of today's market conditions, which are characterized by natural gas prices higher in Europe and Asia than in North America with higher LNG imports into Europe and Asia. The changing dynamic of the Atlantic Basin gas market has significant implications for where investments in LNG re-gasification terminals need to be made and it appears that EMF 23 has played a useful role in better informing such investment decisions.
In the case of GDF Suez Energy North America, Inc., one of the firms providing financial support to EMF 23, these implications have translated into concrete actions in the form of investments estimated at more than one billion U.S. dollars to develop floating LNG terminals serving both the New England (Neptune) and southeast Florida (Calypso) natural gas markets. (7) During 2007, GDF Suez was, at 742 billion cubic feet, the largest importer of LNG volumes into both Europe and North America and with the completion of Neptune, Calypso, and the Fos-Cavaou terminal in France will have a total of 7 LNG terminals operating in Europe and North America. In addition, GDF Suez operates 15 LNG tankers with another 5 currently under construction. Given the size and significance of LNG to their strategy, a strong understanding LNG markets is critical.
3. UNCERTAINTY VS. PRECISION
Of course, even with a consensus view established it would be folly to treat baseline model results as a somehow infallible forecast of future market evolution. Decision focus also requires a thoughtful consideration of uncertainty and how market conditions might evolve differently. One method for addressing uncertainty that has gained especially wide acceptance in the energy industry is scenario analysis. (8)
Although EMF 23 explored only a handful of scenarios, the scenarios it did explore served both to illustrate how a scenario-oriented approach might be applied more broadly and proved to be of some interest per se in the sense of nontrivial impacts that would change the baseline conclusions.
An example of such a scenario is restriction of Russian gas supply, either because of political gaming (e.g. CPB Model) or the fact of higher estimated costs limiting competitiveness (e.g. RICE Model). In this scenario, the ratio of European to North American LNG imports increases significantly relative to the baseline along with a corresponding change in pricing. The consequences of this scenario for a strategy limited to investment in North American re-gasification would be quite dire. Industrial companies considering such investments, whether energy marketing intermediaries or end-users, would be wise to both undertake a comprehensive assessment of the risk that this scenario might occur as well as develop a more diversified strategy mitigating the risk.
Scenario analysis is not risk or uncertainty analysis, but it is nevertheless a technique oriented in that basic direction. Consider a deterministic model in which you run two scenarios characterized as "high supply" and "low supply". In the high supply (deterministic) scenario, every economic player in the model makes investment, operation, and retirement decisions assuming high supply. The decisions predicted by the model are such that every player in the model behaves under the deterministic supposition that there is high supply. All the decision makers execute a high supply-specific investment, operation, and retirement strategy. Consider the low supply case. In the low supply deterministic scenario, every economic player in the model makes investment, operation, and retirement decisions assuming low supply. The decisions predicted by the model are such that every player in the model behaves under the deterministic supposition that there is low supply, and the decisions every agent makes are completely different from the decisions made in the high supply case. Each scenario has an entirely different decision space for every agent. Yet consider what a real world market with true uncertainty is: There is a common set of investment decisions in the high and low supply case. Agents have to make one set of decisions in the face of forward high supply-low supply uncertainty. Agents today make their investment decisions, the coin is flipped, and the outcomes occur. One set of decisions--two prospective outcomes. That is the way uncertainty truly works. In that world, economic agents trade risk as well as commodity, and the price of each affects the price of the other. Framed with this distinction, it seems reasonable to conclude that scenario analysis is but a halfway house, indeed a valuable but approximate halfway house, on the way to truly probabilistic market models. There is a strong need for intrinsically probabilistic models, but EMF had none and we are not aware of any in the market.
Even with a "probabilistic model" such as a Monte Carlo model that samples from probability density functions and then plays out each sampled scenario, the foregoing problem is not alleviated. Indeed, every individual scenario has all the agents thinking deterministically within that sampled scenario. Furthermore, if there is a single probability distribution within a model, there is de facto an explicit assumption that every agent in the model sees and agrees on a common probability distribution over the uncertain lottery within the model. Agents in Monte Carlo models are not allowed to have different information sets or probability distributions. They are all forced by construction to have the same information set. This is a severe limitation when one thinks of the real world in which there are a plethora of information sets. The only way to overcome that would be for each and every individual agent to have a different probability distribution over uncertain outcomes. If there were 1000 agents in your model, you would have to sample individually for each agent from his or her private probability distribution. That simply is not conceptually or computationally tractable. However, it is the way uncertainty works.
FEED