The point of all this is that the LT forecasts, high and low aren't worth much. Beyond 2-3 years they are historically horrible at predicting energy fuel types, prices and levels. I'm in the energy business and every year we do 10-year forecasts and beyond 3 or 4 years we don't put much faith in what they show. Things are changing so fast in the energy sector and technolgy has a lot to do with that. On any given day one can find bullish LNG prices out to 2030 and you can also find bearish. When oil stopped being absolute king of everything the energy world has changed for the overall better. But it will be in flux for the next 10-15 years at least. Too many variables that can change things substantially.
This artice is from 2014 and is part 3 of 3 and shows how bad LT forecasting has been since the '70s. What will be will be and I've yet to see a crystal ball that comes with a performance guarantee.
THE TRACK RECORD OF LONG-TERM ENERGY FORECASTING
THE TRACK RECORD OF LONG-TERM ENERGY FORECASTING
Long-term energy forecasts have a poor track record. They have failed to accurately predict total energy demand, sector demand and energy prices. Attempts to improve accuracy by adding more factors to explain consumption and pricing has complicated the models without necessarily improving their results. This last of 3 articles on long-term energy forecasting focuses on why long-term energy forecasting is so difficult, ideas to improve it and suggested alternatives.
The Track Record is Poor for Predicting Long-Term Energy Consumption
Long-term energy forecasting only became common after oil prices spiked in the 1973-74. Afterwards, there were dozens of forecasts of US energy demand in the year 2000. The graph below summarizes the range of predictions. Actual demand in 2000 was 105 exajoules. The curve numbers refer to the particular studies as described in the DOE report, Energy Demands 1972 to 2000, published in 1979.
Predictions of US Energy Demand in 1990 and 2000
Source: Craig, Gadgil and Koomey (2002).
Energy demand had always been well-correlated with population and economic growth and population and GDP growth were well-behaved for the US. So, why did so many forecasts greatly overestimate energy demand in 2000?
Because forecasters did not anticipate higher energy prices would lead to a sustained demand for more energy-efficient vehicles and appliances resulting in a marked long-term decrease in energy consumption.
This example illustrates one of the biggest problems facing forecasters—predicting the unexpected: changes in behavior and technology, and global events like the rise of OPEC and terrorism. Consequently, maybe it is not surprising that forecasting accuracy has not improved much in the last 30-40 years.
An article written late last year looked back at the US EIA’s (Energy Information Administration) 2005 Annual Energy Outlook’s predictions for 2013. It forecasted $25-30/barrel oil and gasoline prices of $1.50/gal. Actual prices were $100/barrel and $3.50/gal. Other nations’ forecasts were not much better.
Failures to Accurately Predict Other Aspects of Energy Use
In addition to failing to predict primary energy demand and prices, long-term energy forecasts have also failed to accurately predict other aspects of energy use:
-
Sector demand (electricity generation, transportation, etc.)
-
Exhaustion of energy supply sources
-
Changes in energy sources (oil for coal)
-
The contributions of various energy sources in meeting overall demand and,
-
The impact of technology.
Just as total US energy consumption in 2000 was overestimated, so was electricity demand.
In 1977, 70 energy experts from government, business and academia met for the Workshop on Alternative Energy Strategies. They concluded that global oil demand would outstrip supply before 2000. Oil production would most likely peak between 1994 and 1997 and be just 40% of that peak value today.
In the 1950’s it was widely thought that nuclear energy would replace all fossil fuels for electricity generation. By the 1970’s interest in nuclear power was declining and today, no developed nation has plans to install more nuclear power and some are decommissioning existing nuclear plants.
The time for photovoltaics to become economical and widely adopted has been consistently underestimated.
Why Have Most Long-Term Energy Forecasts Been Inaccurate?
Bezdek and Wendling analyzed 49 energy studies done between 1952 and 2001 which attempted to forecast long-term energy developments. They concluded that many shared the same problems:
-
Discounting the impact of energy prices on consumption and the adaptability of markets.
-
Not allowing for improvements in existing technologies.
-
Overestimating the rate of adoption of new energy sources.
-
Assuming the major barriers to adopting renewables were politics and policy and not economics.
Moreover, the accuracy of both government and private forecasts are adversely affected by bias—changing assumptions so the forecast supports a particular policy or desired future outcome.
Are Long-Term Energy Models Too Complicated?
Yes. These models need to be simplified.
The previous article about forecasting methods discussed efforts to make forecasts more accurate by combining econometric methods with “what-is-possible” methods, like end-use and scenario analysis. But, attempts to make long-term energy forecasts more realistic by incorporating the impact of discontinuous changes have led to increasingly complicated models without significantly improving their accuracy (see above).
V. Smil argues that the large number of linked assumptions in these models defeats the quest for realism. The greater the number of variables to be estimated, the greater the overall error since each estimate has error limits associated with it. If the confidence limits are too wide, the spread of outcomes may be too large to be useful for decision-making. Smil’s concerns are shared by others.
In his book, Forecasting: An Appraisal for Policymakers and Planners, Ascher states that forecast accuracy is affected less by the choice of methodology than by the forecaster’s core assumptions.
What Other Changes Have Been Proposed to Improve Long-Term Energy Forecasts?
Those who believe that long-term forecasting is necessary but flawed argue for 3 similar approaches:
-
Replace quantitative forecasts with scenario analysis so that decisions can be based on a range of outcomes (Smil).
-
Consider forecasting to be a form of risk analysis and aim for forecast-proof society (Cobb).
-
Introduce uncertainty into forecasts (Morgan).
Each of these approaches is based on the fact that the future is uncertain. Rather than pursuing a point forecast, we acknowledge the inherent uncertainty in long-term forecasts and present a range of outcomes.
Smil emphasizes a contingency planning approach based on exploring a range of scenarios. The analysis methods can range from narrative and normative scenarios to Monte Carlo simulations and stochastic programming.
While normative scenarios (what should happen instead of what is likely to happen) can be useful, their analysis must be probing and critical. They must not be allowed to become tools for those advocating particular views.
Cobb writes that since forecast accuracy declines rapidly with time, the range of outcomes is more important than the middle value. He compares preparing for an uncertain energy future to buying homeowner’s insurance. Since energy supply risks cannot be quantified, he recommends analyzing the types of risk we face and insuring ourselves as warranted.
Morgan accepts that quantitative forecasts are unlikely to disappear. To make them more realistic, he advocates replacing definitive forecasts with probabilistic ones.
While I endorse Morgan’s proposal, error estimates can be very subjective. This is especially true for those variables which are the most uncertain. A forecaster faced with wide error bars is likely to reduce them to make the forecast more credible.
Conclusions:
-
Long-term energy forecasting has a poor track record for predicting primary and sector energy demand, energy prices, exhaustion of existing energy sources and the rate of adoption of new energy sources.
-
Attempts to improve long-term forecasts have made models too complex, requiring the estimate of many critical variables, introducing more error with each variable added.
-
Decisions about long-term energy demand are best based on a range of alternative outcomes, acknowledging the uncertainty in trying to predict the future and the long history of failure trying to do so.
Log in to post comments
