Resources for the Future
Krautkraemer and Toman
like (4) through (6). For example, the model could incorporate a specification in which the reserves discovered per foot of new drilling depend on cumulative past drilling in that province. This specification could be used along with other information to determine the efficient level of reserve development as a function of the direct cost and user cost of drilling, plus the marginal (shadow) value of additional reserves. One prominent example of such a model in the United States is the oil and gas supply component of the National Energy Modeling System (NEMS) operated by the Energy Information Administration (EIA 2000); Deacon (1993) provides a smaller-scale example.
Numerical parameters in such models come from a variety of sources: statistical curve fitting, engineering experience, and informed judgment, among others. Because these models are not derived from statistically fitting equations to data, it is not possible to construct classical confidence intervals and statistics for assessing the reliability of the model outputs. They are useful principally for “what if” exercises—by supplying different sets of input variables (e.g., oil prices and technical recovery parameters for drilling), one can see the consequences of changes in these assumptions. Whether the model structure is properly specified empirically cannot be independently assessed. The simulation models also do not incorporate uncertainty and expectations formation. Implicitly in dynamic optimization models, individuals are assumed to accurately judge the future.
Econometrically estimated models also are rooted in the theoretical efficiency conditions. However, these models are parameterized using statistical methods that seek to define supply decisions as functions of causal variables. The most sophisticated of these frameworks explicitly model expectations formation using the rational expectations time series approaches pioneered in macroeconomics (Epple 1985, Hendricks and Novales 1987, Walls 1992).
In brief, these methods explicitly include representations of expectations formation based on past observations of price and cost influences; assuming that producers are not systematically biased in their expectations and fairly rapidly learn about a systemic change in market or technological conditions, statistical techniques can be used to estimate the parameters of the expectations process and the underlying parameters linking supply choices to the relevant economic influences. The point is to provide consistent estimates of the basic technological and geological parameters undergirding the supply process, so that the effects of policy intervention