We begin by revisiting some past predictions and the assumptions underlying them.
This exercise requires an understanding of forecasting in the energy sector: why,
how, and to what end predictions are made, and how we evaluate their successes or
failures. Forecasts serve a range of purposes. They provide a context for organizing
and making sense of large data sets, help define possible future scenarios,
and make clear current assumptions. Many forecasts, however, exist to further political
or social agendas, and these motivations are not necessarily explicit in their construction.
For example, projections of rapidly growing demand may be used to push for the construction
of new power plants, while worst-case climate change scenarios may be used to spur
action on emissions reduction.
The intended use of a forecast, whether explicit or hidden, shapes the process by
which it is made. Many forecasting techniques have been applied to the energy sector
with varying degrees of success. A simple but problematic way to predict the future
is to extrapolate from present conditions; this method can yield some insight but
obscures the motivating forces behind current trends. More sophisticated models
may incorporate several variables, but these are sensitive to parameter choice and
to assumptions about how variables interact with the wider environment. There is
always, of course, the option of taking stated targets at face value. Governments
and industries often make public pronouncements regarding their future plans, but
these often obscure intent as well as reality. Still, these targets can help provide
important insight into the motivations of policy-makers and the factors they consider
Considering these forecasting methods leads to an uncomfortable truth: many past
predictions have been not merely inaccurate, but spectacularly wrong. Perhaps most
notorious is the 1954 prediction by Lewis Strauss, then head of the Atomic Energy
Commission, that “our children will enjoy in their homes electrical energy
too cheap to meter.” Such energy was to be provided by a fleet of nuclear
fission reactors that were presumably efficient, safe, and secure as
well as inexpensive to license, site, and build.7
Strauss, however, was not the only one who proved overconfident
about the potential of nuclear power. The Energy Information Administration of the
Department of Energy (DOE) anticipated that the United States would have 1,200 GWe8 of installed nuclear capacity by 2000;
the actual capacity was 98 GWe. The forecasts failed not only to predict the magnitude
of nuclear energy but also to capture the prevailing trend. While the DOE anticipated
a growth in nuclear capacity of almost 700 GWe between 1990 and 2000, in reality
the industry saw a slight decline as reactors were taken out of commission.
Such problems are not limited to the nuclear industry but are found in many long-range
energy models. Figure 1 shows the total U.S. energy demand in the year 2000 as predicted
by several models developed in the early 1970s.9 Notably, all the models drastically overestimate the actual 2000 figure,
having failed to take into account the oil price shocks of the late 1970s and subsequent
efficiency measures. They extrapolate trends from the relatively profligate
late 1960s and early 1970s, when readily available cheap oil made efficiency
and conservation unnecessary. Paul Craig, Ashok Gadgil, and Jonathan Koomey note
that only one forecast,10 designed to show
the possibility of a future powered by renewables (rather
than attempt a reasonable forecast from contemporary trends), comes close to approximating
the actual energy consumption.
Figure 1: Predicted versus Actual U.S. Primary Energy Use, 1975
The figure suppresses the zero baseline. Each line represents
a different model used to make a prediction. Source: Paul P. Craig, Ashok Gadgil,
and Jonathan G. Koomey, “What Can History Teach Us? A Retrospective Examination
of Long-Term Energy Forecasts for the United States,” Annual Review of Energy
and the Environment 27 (November 2002): 83–118. Figure reprinted
here with permission.
These problems remain endemic to energy forecasts. Long-term energy models that
aim to track greenhouse gas emissions similarly failed to anticipate the success
of shale gas drilling technologies, which have helped increase known U.S. natural
gas reserves by 35 percent.11 Because
gas-fired power plants produce during combustion roughly half the greenhouse
gas emissions of traditional coal-fired generation, many estimates of U.S.
emissions growth have had to be revised downward. Further discoveries may lead to
the widespread use of natural gas as a transition fuel, altering the picture for
international climate agreements and domestic policy.
It may seem that these failures are insignificant; after all, the inability
of energy planners to foresee the oil shocks of the 1970s did not lead
to catastrophic energy shortages, nor did the United States’ underestimation
of its natural gas reserves significantly affect national security. In both
cases, the market was able to handle the unforeseen changes and devise solutions
that did not lead to economic or socio-political catastrophe. The system, it seems,
has proven to be relatively resilient against prediction failure. It does not follow,
however, that the solutions devised were the best of all possible solutions, nor
that mistakes were consequence-free. The fortunes of individual companies rise and
fall with changing market conditions, and reliable predictions are important for
them to increase their competitiveness under changed conditions.
Predictions are important at the state and international levels as well. Particularly
when prediction mistakes involve a common good, such as environmental protection,
market solutions may be inadequate to address the failure. Regulation or some form
of externality pricing may be required and can be imposed only by governments or
outside bodies. Trusting market solutions to materialize in the absence of reliable
predictions means governments may be ill-prepared to provide for the common good
of their citizens. Additionally, preparation may be a question of scale: terrorist
attacks, accidents, or proliferation concerns are best dealt with at the state or
international level, not at the vendor or utility level. Accurate predictions, or
at least a thorough understanding of their limitations, are therefore crucially
important, for both the players that must compete in a changing market and the governments
and international institutions that must prepare for a changing world.
The question remains, why do forecasts so often fail to anticipate future events?
It may be that they are simply wrong; their initial assumptions may give an inaccurate
or incomplete picture of the present. Physical or economic processes may be poorly
understood or modeled so that even correct inputs lead to incorrect conclusions.
More often than not, however, models fail because they do not anticipate events
outside these initial assumptions: they do not consider game changers. In this paper,
we focus on this latter category of events as applied to future projections for
nuclear power. This requires us both to explain what we believe to be the current
state of forecasting for nuclear energy over the next fifty years and to make
clear the scope of the assumptions and initial conditions that enter into these
predictions. We call this assumed reference case the “no-surprise scenario.”