Download - Past performance is no guide to future returns: Why we can't accurately forecast the future
Past performance is no guide to future returns: Why we can’t accurately
forecast the future Jonathan Koomey, Ph.D.
Research Fellow, Stanford University http://www.koomey.com
Presented on a webinar for US EPA and US DOE May 18, 2016
1 Copyright Jonathan G. Koomey 2016
My background
• Founded LBNL’s End-Use Forecasting group and led that group for more than 11 years.
• Peer reviewed articles and books on – Forecasting methodology – Economics of greenhouse gas mitigation – Critical thinking skills – Information technology and resource use
2 Copyright Jonathan G. Koomey 2016
3
Cost-benefit analysis: the standard approach
Copyright Jonathan G. Koomey 2016
True or False?: If only we had enough…
• Time • Money • Graduate Students • Coffee
we could accurately predict the cost of energy technologies in
2050 4 Copyright Jonathan G. Koomey 2016
Widespread modeling practice implies that the answer is “True”
5 Copyright Jonathan G. Koomey 2016
Based on my experience and reviews of historical
retrospectives on forecasting, I say “No way”
6 Copyright Jonathan G. Koomey 2016
Aside: Many of the best modelers acknowledge the difficulties in the pursuit of accurate forecasts, but in their heart of hearts they still
believe they can predict accurately with greater effort
7 Copyright Jonathan G. Koomey 2016
Uncertainty affects even physical systems
Es=mates of Planck’s constant "h" over =me. In this physical system researchers repeatedly underes=mated the error in their determina=ons. At each stage uncertain=es existed of which the researchers were unaware. The problem of error es=ma=on is far greater in long-‐range energy forecas=ng. Taken from Koomey et al. 2003.
8 Copyright Jonathan G. Koomey 2016
Forecasting Accuracy: The Models Have Done Badly
• Energy forecasting models have little or no ability to accurately predict future energy prices and demand (Craig et al. 2002)
• Even the sign of the impacts of proposed policies is a function of key assumptions (Repetto and Austin 1997)
• The dismal accuracy and inherent limitations of these models should make modelers modest in the conclusions they draw (Decanio 2003) Craig, P., A. Gadgil, and J. Koomey (2002). “What Can History Teach Us? A Retrospective Analysis of Long-term Energy Forecasts for the U.S.” Annual Review of Energy and the Environment 2002. R. H. Socolow, D. Anderson and J. Harte. Palo Alto, CA, Annual Reviews, Inc. (also LBNL-50498). 27: 83-118.
Repetto, R. and D. Austin (1997). The Costs of Climate Protection: A Guide for the Perplexed. Washington, DC, World Resources Institute. DeCanio, S. J. (2003). Economic Models of Climate Change: A Critique. Basingstoke, UK, Palgrave-Macmillan.
9 Copyright Jonathan G. Koomey 2016
One example: 1970s projections of year 2000 U.S. primary energy
Source: Craig, Paul, Ashok Gadgil, and Jonathan Koomey. 2002. "What Can History Teach Us?: A Retrospective Analysis of Long-term Energy Forecasts for the U.S." In Annual Review of Energy and the Environment 2002. Edited by R. H. Socolow, D. Anderson and J. Harte. Palo Alto, CA: Annual Reviews, Inc. (also LBNL-50498). pp. 83-118.
10 Copyright Jonathan G. Koomey 2016
What drove errors in US primary energy forecasts?
Copyright Jonathan G. Koomey 2016 11
Graph from Hirsh and Koomey 2015
Another example: Oil
price projec3ons by U.S. DOE, AEO 1982
through AEO 2000
12 Copyright Jonathan G. Koomey 2016
Not any beEer aFer 2000: Oil price
projec3ons by U.S.
DOE, AEO 2000
through AEO 2007
13 Copyright Jonathan G. Koomey 2016
Yet another example: NERC fan
Copyright Jonathan G. Koomey 2016 14
US electricity genera=on BkWh/year
Why Are Long-term Energy Forecasts Almost Always Wrong?
• Core data and assumptions, which drive results, are based on historical experience, which can be misleading if structural conditions change
• The exact timing and character of pivotal events and technology changes cannot be predicted
Laitner, J.A., S.J. DeCanio, J.G. Koomey, A.H. Sanstad. (2003) “Room for Improvement: Increasing the Value of Energy Modeling for Policy Analysis.” Utilities Policy, 11, pp. 87-94.
15 Copyright Jonathan G. Koomey 2016
Conditions for Model Accuracy • Hodges and Dewar: models can be
accurate when they describe systems that – are observable and permit collection of
ample and accurate data – exhibit constancy of structure over time – exhibit constancy across variations in
conditions not specified in the model
Source: Hodges, James S., and James A. Dewar. 1992. Is it you or your model talking? A framework for model validation. Santa Monica, CA: RAND. ISBN 0-8330-1223-1.
16 Copyright Jonathan G. Koomey 2016
∑: Accurate forecasts require structural constancy and no
surprises
17 Copyright Jonathan G. Koomey 2016
Market structure can change fast
Source: Scher and Koomey 2010.
18 Copyright Jonathan G. Koomey 2016
Fast changing markets #2: US electricity consumption
Copyright Jonathan G. Koomey 2016 19
Graph from Hirsh and Koomey 2015
Surprises can be big: U.S. nuclear busbar costs
Source: Koomey and Hultman 2007. Assumes 7% real discount rate.
Projected cost range from Tybout 1957
20 Copyright Jonathan G. Koomey 2016
Implications for long-term energy forecasting
• Forecasting models describing well-defined physical systems using correct parameters can be accurate because physical laws are geographically and temporally invariant (as long as there are no surprises)
• Economic, social, and technological systems do not exhibit the required structural constancy, so models forecasting the future of these systems are doomed to be inaccurate. Four big sources of inconstancy – Pivotal events (like Sept. 11th or the 1970s oil shocks) – Technological innovation – Institutional change – Policy choices
21 Copyright Jonathan G. Koomey 2016
∑: Economics ≠ Physics
22 Copyright Jonathan G. Koomey 2016
So no matter how many $, coffee cups, months, or graduate
students you have, accurate long-run forecasting of technology
costs is impossible
23 Copyright Jonathan G. Koomey 2016
Two senses of the word “impossible”:
Practically and
Theoretically
Either way, the net result is the same: inaccurate forecasts
24 Copyright Jonathan G. Koomey 2016
So what does this result imply for predictions of the costs of
energy technologies?
25 Copyright Jonathan G. Koomey 2016
Some lessons • The world is evolutionary and path dependent
– Increasing returns, transaction costs, information asymmetries, bounded rationality, prospect theory
– Our actions now affect our options later (so do surprises!)
• Experimentation is the order of the day • Use real data to prove results
– For nuclear power, we’re in the “show me” stage. Cost projections are no longer enough
• Prefer technologies that – are mass produced vs. site-built – have short lead times vs. longer lead times
26 Copyright Jonathan G. Koomey 2016
Nuke costs: here we go again?
Source: Koomey and Hultman 2007. 27 Copyright Jonathan G. Koomey 2016
“No battle plan survives contact with the enemy.” –Helmuth von Moltke the elder
Copyright Jonathan G. Koomey 2016 28
More lessons • Use physical and technological constraints to
define bounding cases. Examples: – 2 degrees Celsius warming limit implies a carbon
budget, which implies a certain rate of implementation of non-fossil energy sources to avoid worst effects of climate change.
– Certain technologies use materials that are in limited supply. Working backwards from a goal can help identify resource constraints.
– Lifetime of power generation technologies and buildings limits penetration of new technologies unless we scrap existing capital
29 Copyright Jonathan G. Koomey 2016
Reconsidering benefit-cost analysis for climate
• "A corollary is that it is fruitless to attempt to determine the "optimal" carbon tax. If neither the costs nor the benefits can be known with any precision, just about the only thing that can be said with certainty about the welfare maximizing price of carbon emissions is that it is greater than zero. Economists have a great deal to say about how to implement such a tax efficiently and effectively, about the similarities and differences between a tax and a system of tradable carbon emissions permits, about about the best way to recycle the revenue from such a tax or permit system. And, as we have seen above, the distributional consequences of such a tax or permit auction plan will affect other economic variables through system-wide feedbacks. However, any attempt to specify the exact level of the "optimal" tax is less an exercise in scientific calculation than a manifestation of the analyst’s willingness to step beyond the limits of established economic knowledge."
• –DeCanio, Stephen J. 2003. Economic Models of Climate Change: A Critique. Basingstoke, UK: Palgrave-Macmillan. p.157.
Copyright Jonathan G. Koomey 2016 30
Conclusions • It is impossible to accurately forecast energy
technology characteristics because of – structural inconstancy and – pivotal events
• Forecasting community has yet to absorb the implications of this insight
• To cope we need new ways to think about the future – Experimental approach to implementation (try many things,
fail fast, learn quickly, try again) – Rely on physical and technological constraints to create
bounding cases – Embrace path dependence (there is no optimal solution,
just lots of possible pathways of roughly similar costs)
31 Copyright Jonathan G. Koomey 2016
“The best way to predict the future is to invent it.” –Alan Kay
Copyright Jonathan G. Koomey 2016 32
Some Key References • Craig, Paul, Ashok Gadgil, and Jonathan Koomey. 2002. "What Can History Teach Us?: A
Retrospective Analysis of Long-term Energy Forecasts for the U.S." In Annual Review of Energy and the Environment 2002. Edited by R. H. Socolow, D. Anderson and J. Harte. Palo Alto, CA: Annual Reviews, Inc. pp. 83-118.
• Ghanadan, Rebecca, and Jonathan Koomey. 2005. "Using Energy Scenarios to Explore Alternative Energy Pathways in California." Energy Policy. vol. 33, no. 9. June. pp. 1117-1142.
• Hirsh, Richard F., and Jonathan G. Koomey. 2015. "Electricity Consumption and Economic Growth: A New Relationship with Significant Consequences?" The Electricity Journal. vol. 28, no. 9. November. pp. 72-84. [http://www.sciencedirect.com/science/article/pii/S1040619015002067]
• Koomey, Jonathan. 2008. Turning Numbers into Knowledge: Mastering the Art of Problem Solving. Oakland, CA: Analytics Press. 2nd edition. <http://www.analyticspress.com>
• Koomey, Jonathan. 2002. "From My Perspective: Avoiding "The Big Mistake" in Forecasting Technology Adoption." Technological Forecasting and Social Change. vol. 69, no. 5. June. pp. 511-518.
• Koomey, Jonathan G., Paul Craig, Ashok Gadgil, and David Lorenzetti. 2003. "Improving long-range energy modeling: A plea for historical retrospectives." The Energy Journal (also LBNL-52448). vol. 24, no. 4. October. pp. 75-92.
• Chapter 4: “Why we can’t accurately forecast the future”, in Koomey, Jonathan G. 2012. Cold Cash, Cool Climate: Science-Based Advice for Ecological Entrepreneurs. Burlingame, CA: Analytics Press. [http://www.analyticspress.com/cccc.html]
• Koomey, Jonathan. 2013. "Moving Beyond Benefit-Cost Analysis of Climate Change." Environmental Research Letters. vol. 8, no. 041005. December 2. [http://iopscience.iop.org/1748-9326/8/4/041005/]
• Laitner, J.A., S.J. DeCanio, J.G. Koomey, A.H. Sanstad. (2003) “Room for Improvement: Increasing the Value of Energy Modeling for Policy Analysis.” Utilities Policy, vol. 11, no. 2. June. pp. 87-94.
• Scher, Irene, and Jonathan G. Koomey. 2011. "Is Accurate Forecasting of Economic Systems Possible?" Climatic Change. Vol 104, No. 3-4, pp.473-479.
33 Copyright Jonathan G. Koomey 2016
More Key References • Armstrong, J. Scott, ed. 2001. Principles of Forecasting: A Handbook for Researchers and
Practitioners. Norwell, MA: Kluwer Academic Publishers. • Ascher, William. 1978. Forecasting: An Appraisal for Policy Makers and Planners. Baltimore,
MD: Johns Hopkins University Press. • Cohn, Steve. 1991. "Paradigm Debates in Nuclear Cost Forecasting." Technological
Forecasting and Social Change. vol. 40, no. 2. September. pp. 103-130. • Grubler, Arnulf, Nebojsa Nakicenovic, and David G. Victor. 1999. "Dynamics of energy
technologies and global change." Energy Policy. vol. 27, no. 5. May. pp. 247-280. • Hodges, James S., and James A. Dewar. 1992. Is it you or your model talking? A framework for
model validation. Santa Monica, CA: RAND. ISBN 0-8330-1223-1. • Huntington, Hillard G. 1994. "Oil Price Forecasting in the 1980s: What Went Wrong?" The
Energy Journal. vol. 15, no. 2. pp. 1-22. • Huss, William R. 1985. "Can Electric Utilities Improve Their Forecast Accuracy? The Historical
Perspective." In Public Utilities Fortnightly. December 26, 1985. pp. 3-8. • Landsberg, Hans H. 1985. "Energy in Transition: A View from 1960." The Energy Journal. vol. 6,
pp. 1-18. • O'Neill, Brian C., and Mausami Desai. 2005. "Accuracy of past projections of U.S. energy
consumption." Energy Policy. vol. 33, no. 8. May. pp. 979-993. • Tetlock, Philip E. 2005. Expert Political Judgment: How Good Is It? How Can We Know?
Princeton, NJ: Princeton University Press.• Tybout, Richard A. 1957. "The Economics of Nuclear Power." American Economic Review. vol.
47, no. 2. May. pp. 351-360.
34 Copyright Jonathan G. Koomey 2016