Forecasting global climate change: A scientific approach
Kesten C. Green
University of South Australia and Ehrenberg-Bass Institute, Australia
J. Scott Armstrong
University of Pennsylvania, U.S.A, and Ehrenberg-Bass Institute, Australia
Working Paper – June 2014
Subsequently published as “Forecasting Global Climate Change” in Climate Change: The Facts 2014.
Alan Moran (Editor). Published by the Institute of Public Affairs, Melbourne, Victoria 3000, Australia
The Golden Rule of Forecasting requires that forecasters be conservative by making proper use of
cumulative knowledge and by not going beyond that knowledge. The procedures that have been used
to forecast dangerous manmade global warming violate the Golden Rule. Following the scientific
method, we investigated competing hypotheses on climate change in an objective way. To do this, we
tested the predictive validity of the global warming hypothesis (+0.03°C per year with increasing CO2)
against a relatively conservative global cooling hypothesis of -0.01°C per year, and against the even
more conservative simple no-change or persistence hypothesis (0.0°C per year). The errors of forecasts
from the global warming hypothesis for horizons 11 to 100 years ahead over the period 1851 to 1975
were nearly four times larger than those from the global cooling hypothesis and about eight times larger
than those from the persistence hypothesis. Findings from our tests using the latest data and other data
covering a period of nearly 2,000 years support the predictive validity of the persistence hypothesis for
horizons from one year to centuries ahead. To investigate whether the current alarm over global
temperatures is exceptional, we employed the method of structured analogies. Our search for analogies
found that environmental alarms are a common social phenomenon, with 26 similar situations over a
period of two hundred years. None were the product of scientific forecasting procedures, and in all
cases the alarming forecasts were wrong. Twenty-three of the alarms led to government actions. The
government actions were harmful in 20 cases, and of no benefit in any.
We are grateful to Steve Goreham, David Legates, Craig Loehle, Alan Moran, and Willie Soon for
their helpful suggestions on this chapter. Hester Green, Jen Kwok, Lynn Selhat, and Angela Sun
provided useful suggestions on the writing. The responsibility for any errors or omissions remains with
“Warming by 2070, compared to 1980 to 1999, is projected to be… 2.2 to 5.0°C.”1
“By 2100, the average U.S. temperature is projected to increase by about 4°F to 11°F.”2
“If we do not cut emissions, we face even more devastating consequences, as unchecked they could raise
global average temperature to 4°C or more above pre-industrial levels by the end of the century. The
shift to such a world could cause mass migrations of hundreds of millions of people away from the
worst-affected areas. That would lead to conflict and war.”3
Forecasts such as these are made by scientists and repeated by the political leaders they advise.
The principal source of the forecasts is the United Nation’s Intergovernmental Panel on Climate
Change (the IPCC). The IPCC’s forecasts are the product of a collaboration of scientists and computer
modellers working for lobbyists, bureaucrats, and politicians (as documented by Laframboise 2011,
and Ball 2014). The forecasts of dangerous manmade global warming and its consequences are made
with great confidence, as are recommendations of actions to counter the forecasted danger.
History is replete with experts making confident forecasts. The record also shows that the
accuracy of such forecasts has been poor. Consider, for example, Professor Kenneth Watt’s forecast of
a new Ice Age in his 1970 Earth Day speech at Swarthmore College:
“The world has been chilling sharply for about twenty years. If present trends continue, the world will be
about four degrees colder for the global mean temperature in 1990, but eleven degrees colder in the year
2000. This is about twice what it would take to put us into an ice age.”
Professor Watt is not unusual among experts in making confident forecasts that turn out to be
wrong. Evidence from research on forecasting shows that an expert’s confidence in making forecasts
about complex uncertain situations is unrelated to the accuracy of the forecast (Armstrong 1985, pp.
138-144, summarizes research). Those who believe that we can learn to avoid poor forecasts from
history may wish to consult the diverse examples in Cerf and Navasky’s book The Experts Speak
We suggest that government policy makers and business managers consider whether the
IPCC’s forecasting methods are valid before they consider making decisions on the basis of the
1 CSIRO, State of the climate – 2014, “Future climate scenarios for Australia”,
Australia.aspx, viewed on 29 April 2014.
2 United States Environmental Protection Agency web page on “Future Climate Change”,!
http://www.epa.gov/climatechange/science/future.html#Temperature, viewed on 29 April 2014.
3 Lord Stern, quoted in The Guardian on 14 February 2014 in an article titled “Flooding and storms in UK are clear
signs of climate change, says Lord Stern”, http://www.theguardian.com/environment/2014/feb/13/flooding-storms-
uk-climate-change-lord-stern, viewed on 29 April 2014.
forecasts. To that end, we examine whether or not the IPCC’s forecasts of dangerous manmade global
warming are the product of scientific methods.
We then investigate whether alternative hypotheses of climate change provide more accurate
forecasts than the dangerous manmade global warming hypothesis. Specifically, we test forecasts from
the hypothesis of global cooling and from the hypothesis of climate persistence. We then make
forecasts of global average temperatures for the remaining years of the 21st Century and beyond using
an evidence-based forecasting method.
Finally, we ask whether the IPCC forecast of dangerous manmade global warming is a new
phenomenon. To answer this question, we use the method of structured analogies to seek out and
analyse similar situations.
Are the alarming forecasts the product of scientific forecasting methods?
The IPCC forecasts are derived from the judgments of the scientists that the IPCC engages.
Computer modellers write code to represent the scientists’ judgments that, in turn, provides long-term
forecasts of global mean temperatures. Is this use of expert judgment a valid approach to climate
The science of forecasting
For nearly a century, researchers have been studying how best to make accurate and useful
forecasts. Knowledge on forecasting has accumulated by testing multiple reasonable hypotheses about
which method will provide the best forecasts in given conditions. This scientific approach contrasts
with the folklore that experts in a domain will be able to make good forecasts about complex uncertain
situations using their unaided judgment, or using un-validated forecasting methods (Armstrong 1980;
Scientific forecasting knowledge has been summarized in the form of principles by 40 leading
forecasting researchers and 123 expert reviewers. The principles summarise the evidence on
forecasting from 545 studies that in turn drew on many prior studies. Some of the forecasting
principles, such as “provide full disclosure” and “avoid biased data sources,” are common to all
scientific fields. The principles are readily available in the Principles of Forecasting handbook
4 In addition, the ForPrin.com web site provides a checklist of the forecasting principles and software that help users
to determine which methods to use in a given situation.
We used that knowledge to assess whether the IPCC procedures described in Randall et al.
(2007) amounted to scientific forecasting. To do so, we first examined that IPCC chapter’s references
to determine whether the authors had relied on validated forecasting procedures. We found no
references to validation. We then sent emails to all of the Randall et al. (2007) authors for whom we
were able to obtain email addresses5, asking for references for credible forecasts of global average
temperatures and the methods used to derive them. The few useful responses we received referred us to
the Randall et al. chapter or to works that were cited in it.
Evaluating the IPCC procedures against forecasting principles
We then audited the IPCC forecasting procedures using the Forecasting Audit Software
available on ForPrin.com. Our audit found that the IPCC followed only 17 of the 89 relevant principles
that we were able to code using the information provided in the 74-page IPCC chapter. Thus, the IPCC
forecasting procedures violated 81% of relevant forecasting principles (Green and Armstrong 2007)
Appendix 1 of this chapter lists the principles that we found had been clearly violated by the IPCC
It is hard to think of an occupation for which it would be acceptable for practitioners to violate
evidence-based procedures to this extent. Consider what would happen if an engineer or medical
practitioner, for example, failed to properly follow even a single evidence-based procedure.
Evaluating the IPCC procedures compliance with the Golden Rule of Forecasting
We analysed the IPCC’s forecasting procedures to assess whether they followed the Golden
Rule of Forecasting. The Golden Rule of Forecasting requires that forecasters be conservative. This
means that they should use procedures that are consistent with knowledge about the situation and about
forecasting methods. The Golden Rule is the antithesis of the common antiscientific attitude that “this
situation is different,” which leads forecasters to ignore cumulative knowledge.
The Golden Rule is a unifying theory of how best to forecast. The theory has been tested for
consistency with the evidence in a review of the literature from all areas of forecasting that found 150
studies relevant to the Golden Rule. The studies provided findings from experiments on the effect of
conservative procedures compared to unconservative ones on forecast accuracy. All of the evidence
was consistent with the Golden Rule.
To assist forecasters, the evidence on the Golden Rule is summarised in the form of 28
guidelines, including “avoid bias by specifying multiple hypotheses and methods” and “select
5 The IPCC refused to provide the authors’ email addresses.
evidence-based methods validated for the situation” (Armstrong, Green, and Graefe 2014).6 The
median reduction in forecast error from following a Golden Rule guideline, rather than common
practice, is 25%. That is, error was reduced by one quarter.
We found that the IPCC procedures violated all 19 of the Golden Rule guidelines that are
relevant to long-term climate forecasting, including “be conservative when forecasting trends if the
series is variable or unstable” and “be conservative when forecasting trends if the short and long-term
trend directions are inconsistent.” As a consequence of the Golden Rule violations, the IPCC forecasts
are a product of biased forecasting methods.
Are forecasts of dangerous global warming nevertheless valid?
Having established that the IPCC forecasting procedures are unvalidated and are inconsistent
with scientific forecasting knowledge, we investigated whether it would be possible to test the validity
of the forecasts. The most recent global warming scare started around 1976, so testing the validity of
short-term forecasts against the few years since then is possible. Such a test is limited, however, given
that it is not unusual for temperatures to trend up or down, on average, for several years. Also, policy
makers and investors who consider large expenditures that are costly to reverse are concerned with
long-term trends. We therefore devised tests of the validity of the IPCC model’s short- and long-term
forecasts that made extensive use of available data.
Informal short-term tests
In 1999, Pat Michaels explained that short-term events were responsible for recent elevated
temperatures and offered an early test of the IPCC’s short-term forecasts in the form of a bet that
temperatures would go down in the next ten years (Michaels 1999). No one took the bet… and
temperatures went down.
Over the past nearly two-decades, atmospheric carbon dioxide concentrations have risen while
global temperatures have remained flat. Despite the disconfirming evidence, the IPCC claims to have
become even more confident about the manmade global warming hypothesis and they continue to
forecast dangerous warming. The IPCC’s response is typical of how people tend to react when their
forecasts are wrong: by having an even stronger belief that they will be proven correct (e.g., Festinger
et al. 1956, and Batson 1975). Moreover, scientists, like other human beings, tend to reject evidence
that contradicts their beliefs (see, for example, the experiments by Mahoney 1977).
6 A copy of the Golden Rule of Forecasting checklist of guidelines can be found at goldenruleofforecasting.com, as
can a draft copy of the paper.
By 2007, there still had been no proper validation of the IPCC’s forecasts. To generate interest
in the importance of validation, one of us (Armstrong) proposed a bet to former U.S. Vice President Al
Gore that a “no-change” forecast of global average temperature would be more accurate than any
model or forecast that Mr. Gore would support. Gore, advised by Professor James Hansen (see, e.g.,
Hansen 2004), was at the time warning that a “tipping point” in global temperatures was imminent. In
contrast to Gore’s expectation of supporting evidence soon, Armstrong expected that a much longer
period would be needed to obtain a clear result due to natural variations. Armstrong nevertheless
proposed a 10-year bet on the assumption that a shorter term would generate more interest, despite
estimating that he had a 1
chance of losing.
In order to have an objective standard against which to compare forecasts from the alternative
hypotheses, the bet uses the University of Alabama at Huntsville (UAH) lower troposphere series
(Christy et al. 2010). As of May, 2014, the errors from the IPCC’s business-as-usual forecast of
+0.03ºC per year!standing in for Mr. Gore’s tipping point due to his unwillingness to take the
bet!were more than 27% larger than the errors from Armstrong’s bet on the no-change forecasts.
Approach to long-term testing
The models that the IPCC uses for forecasting are based on the beliefs of some scientists that
exponentially increasing levels of carbon dioxide in the atmosphere will cause global mean temperature
to increase at a rate of at least 0.03ºC per year. That figure has been the central forecast of the IPCC
since 1990 (IPCC 1990, p. xi; IPCC, 1992, p. 17). Because carbon dioxide levels have been increasing
exponentially since the beginning of the Industrial Revolution, the IPCC model would seem to apply
over this whole period.
We tested the validity of the IPCC model for forecasting horizons up to 100 years using the
data on global mean temperatures that the IPCC use: the U.K. Met Office Hadley Centre’s HadCRUT3
series7. The Hadley temperature series are derived from selected weather stations and sea surface
records that are adjusted and aggregated to provide proxy average global temperatures. We derived
rolling IPCC-model forecasts of the HadCRUT3 series starting from the year 1851, and ending in the
year 1975, before the most recent global warming trend commenced. The forecasting procedure was
simple, and is consistent with the IPCC’s published business-as-usual forecasts: we added 0.03°C to the
previous year’s actual temperature to derive a one-year-ahead forecast, and then added the same figure
to the forecast for the previous year for each subsequent forecast horizon out to 100 years. By repeating
this procedure for each subsequent year, we obtained 125 one-year-ahead forecasts, 124 two-year-
7 Available from http://hadobs.metoffice.com/hadcrut3/
ahead forecasts, and so on up to and including 26 forecasts for 100 years ahead (Green, Armstrong, and
Given that the HadCRUT3 temperature series trends broadly upwards,8 one would expect the
IPCC-model forecasts that we generated to track the HadCRUT3 series quite well. To determine
whether the dangerous global warming hypothesis is a credible one, however, it is necessary to test the
forecasts against forecasts from alternative hypotheses, and to do so using scientific forecasting
Short and long-term testing using competing hypotheses
In the 1960s and early 1970s, scientists warned of a new ice age (see, e.g., Kukla and Matthews
1972 and Professor Kenneth Watt’s 1970 Earth Day speech quoted above). The scientists provided
hypotheses to support their belief that this time the climate really had changed. Some scientists still
advance the cooling hypothesis.9
Yet despite these forecasts of cooling, starting in the mid-to-late1970s there was actually a
warming trend, and warming alarmists began to inform us that virtually all scientists now subscribed to
the dangerous manmade global warming hypothesis. The claim of near unanimity of scientific opinion
has been discredited by Legates et al. (2014)10, however, and stands in contrast to the 31,487 U.S.
scientists who have publicly signed a statement that they consider the dangerous manmade global
warming hypothesis inconsistent with the evidence.11
While scientists who predict warming and those who predict cooling provide reasons for their
hypotheses, their reasons have been indecisive. In any event, science does not advance by asking
scientists to vote on hypotheses, but by testing them in competition with alternative reasonable
hypotheses (Chamberlin 1890).
We tested a cooling hypothesis of 1ºC cooling per century against the HadCRUT3 global
temperature data. The forecast of cooling is consistent with the various alarms over impending new ice
ages that have occurred over the last 100 years and longer, including those mentioned above.12 And the
8 Note that there is evidence that the series tends to substantially overstate any warming trend due to weather station
locations becoming increasingly surrounded by buildings, asphalt, and heat sources, and the deployment of more
sensitive measurement instruments, together with unexplained adjustments to the temperature readings (McKitrick and
Michaels 2007; McKitrick and Nierenberg 2010; McKitrick 2013; Watts et al. 2012).
9 For examples, search for “Earth undergoing global COOLING since 2002” on climatedepot.com, and George Kukla
interview in Krueger (2007).
10 See also Joe Bast and Roy Spencer's “The Myth of the Climate Change '97%’,” Wall Street Journal, May 26,
2014, available from http://hockeyschtick.blogspot.com.au/2014/05/wsj-myth-of-climate-change-97-what-is.html
11 See the Global Warming Petition Project at petitionproject.org
12 See, e.g., Dan Gainor’s summary of The New York Times cooling, and warming, alarms under the title of “Fire and
Ice,” here http://www.mrc.org/special-reports/fire-and-ice-0
rate is arguably consistent with the understanding of scientists who consider that the Earth is still
experiencing a cooling period, albeit with fluctuations, that commenced around 4,000 years ago (e.g.
For horizons from 1 to 100 years from the year 1851 to the year 1975, 7,550 forecasts in total,
the average absolute errors of the 0.03ºC per year warming forecasts and of the 0.01ºC per year cooling
forecasts increase as the forecast horizon increases (see Exhibit A). Because our tests use historical data
known to exhibit a warming trend, the warming model has an unfair advantage in this test. Despite that
advantage, across all forecast horizons, the average errors of the warming forecasts are more than twice
as large as the errors from the relatively more conservative cooling hypothesis. Remarkably, the natural
cooling forecasts are more accurate than the dangerous warming forecasts for all forecast horizons.
The global warming and cooling hypotheses were developed without the aid of scientific
forecasting. To develop a credible forecasting method against which to benchmark the warming and
cooling hypotheses, we needed a model that was both consistent with evidence-based forecasting
principles and with evidence on climate change. With that in mind, we asked climate expert and
astrophysicist Willie Soon to collaborate with us to develop a model and validation tests (Green,
Armstrong, and Soon 2009).
With Willie Soon, we established that the state of knowledge about the causes of climate
change was such that it would be inappropriate to develop a causal model. The strength and even
direction of proposed causal relationships, including with carbon dioxide, are much disputed among
leading climate scientists (Lindzen 2012; Soon et al. 2001). For example, Kukla and Matthews (1972)
reported from a meeting of climate scientists that “one conclusion reached at the session was that there
is no qualitative difference between the climatic fluctuations in the 20th Century and the climatic
oscillations that occurred before the industrial era. The present climatic trends appear to have entirely
natural causes, and no firm evidence supports the opposite view” (p. 190). A more recent analysis of
two 3,000-year temperature proxy series comes to the same conclusion (Loehle 2004).
We concluded from forecasting principles that because knowledge about climate change is so
poor, forecasts from a no-change forecasting model would be more accurate than forecasts from
methods that attempt to incorporate knowledge that is tentative at best. Depending on the situation, the
appropriate no-change model might be one that forecasts that the level (e.g. current temperature) will
not change, that the trend will not change, or even that the rate of change will not change. For
forecasting long-term global temperatures, we determined that the benchmark model that is most
consistent with the state of knowledge is one that forecasts no change in the level; in other words, no
We compared the forecasts from the no-trend model with the forecasts from the cooling and
warming hypotheses. We found that the average error of the no-trend forecasts was smaller than the
average errors of both the warming and the cooling forecasts for all forecast horizons (Exhibit A, 7,550
forecasts from each method). The average errors of the warming forecasts (dashed line) and the cooling
forecasts (faint line) over the short-term (1 to 10 years) were 45% and 10% larger, respectively, than
the average errors of the no-trend forecasts (solid line). The average error of the no-trend forecasts for
the longer-term horizons, from 11 to 100 years, was roughly one-quarter of the average cooling forecast
error, and one-eighth of the warming forecast error. In absolute terms, the average errors of the no-trend
forecasts were less than 0.20ºC for all horizons out to 75 years; beyond that, the average errors did not
exceed 0.24ºC. The small and steady forecast errors from the persistence model suggest that the Earth’s
climate is remarkably stable over human-relevant timescales. This is particularly remarkable given the
claims by warming alarmists that we have been experiencing “unprecedented” changes in the climate
over the period of the test (see, e.g. IPCC 2013, p. 4).
Average absolute errors of 0.03ºC warming, 0.01ºC cooling, and persistence forecasts
(Forecasts for 1851 to 1975 by forecast horizon. Errors in ºC)
Very long-term testing of predictive validity
In order to assess the validity of the hypotheses over very long horizons, we tested the accuracy
of forecasts from warming, cooling, and no-trend model hypotheses against the Loehle series of proxy
annual temperatures (Loehle and McCulloch 2008). Proxy temperature data are obtained from naturally
occurring records of biological and physical processes that vary with temperature. The Loehle series
was constructed from 18 series obtained and calibrated by other researchers who used such proxy
records as boreholes and pollen counts that each covered most of the Common Era and, between them,
covered much of the globe. The resulting Loehle series extends from AD 16 to AD 1935, allowing us
to test forecasts from variations of the hypotheses for horizons of up to nearly 2000 years. The series
includes the Medieval Warm Period and the Little Ice Age. Evidence suggests that the current climate
is not as warm as that of the Medieval Warm Period when cows grazed and willows grew in Greenland
and seals basked on the shores of Antarctica (see, e.g., Soon et al. 2003, and ongoing research reported
by the Medieval Warm Period Project at CO2science.org).
A forecaster living 100 years after the beginning of the Loehle series in AD 115 might
reasonably have forecast that the average temperature trend that had prevailed over the previous 100
years, an increasing one of roughly 0.003ºC per year (0.3ºC per century), would prevail indefinitely.
Indeed, some researchers have suggested that the Earth has been warmed by human activity for at least
5,000 years (Vavrus, Ruddiman, and Kutzbach 2008). The errors of the warming forecasts increased as
the forecast horizon lengthened as the dashed (topmost) line in Exhibit B shows.
A competing forecaster in AD 115 might well have reasoned from the knowledge of the time
that the Sun is like a large fire that must slowly burn down. Given that temperatures over the previous
century had been trending more upwards than downwards, she might have proposed that while the
Sun’s fire may splutter and flare up from time to time, there would be a long slow decline in the energy
emitted. With these observations in mind, she might have forecast that the average temperatures would
trend downwards at the relatively more conservative rate of 1ºC per millennium or 0.001ºC per year on
average, a much more conservative forecast than those of the 1st Millennium warmer and of the 20th
Century warmers and coolers described above. While the errors of her cooling forecasts increased only
slowly out to the year AD 750, beyond that year the errors of her forecasts tended to increase as the
forecast horizon lengthened (middle, dotted, line in Exhibit B).
We compared the records of the warming and cooling hypotheses forecasters with the record of
our benchmark no-trend hypothesis in the form of a forecaster who predicted that the global average
temperature for the 1,820 years from AD 116 to 1935 would be the same as the AD 115 average. The
solid line in Exhibit B shows the errors of the no-trend forecasts by year, one forecast error per year.
The modest size of the errors and the lack of even a very small persistent trend in them suggest that
there have been no changes in the climate system. In other words, the claim that “things are different
now,” although often made in relation to forecasting in many fields, is once again unsupported. Over
longer policy-relevant periods, annual global mean temperatures are highly stable.
Even with a much more conservative forecast warming rate (one-tenth that of our previous
tests), when applied to this series the warming hypothesis again performed relatively poorly. The
average error of the 1,820 0.3ºC-per-century warming forecasts was more than nine times the average
error of the no-trend forecasts. Again, the errors increased with the forecast horizon. For example, the
errors of warming forecasts for the 4th Century made in AD 115 were nearly three times larger than the
errors of the no-trend forecasts. The equivalent figures for the 8th, 12th, 16th, and 18th Centuries were 4,
14, 23, and 27 times larger. The findings are consistent with those of Green, Armstrong, and Soon
Absolute errors of warming, cooling, and no-trend forecasts made in AD 115
By year from AD 116 to AD 1935, in degrees Celsius
These findings from a long period of varied climate, then, are consistent with those of our
analysis for the 1851 to 1975 warming period above: the more conservative hypothesis and forecasting
method provides the more accurate forecasts. In particular, the most conservative model, the no-trend
model, has greater predictive validity than long-term trend models under diverse conditions. No matter
when one starts forecasting and no matter how global average temperature is estimated, the evidence-
based persistence model produces by far the most accurate forecasts. The findings on the accuracy of
forecasts from long- and short-term tests of the alternative climate change hypotheses are summarised
in Exhibit C in the form of Relative Absolute Errors (RAEs). The reported RAEs are the absolute error
of the forecasts from the hypothesis relative to the corresponding absolute error of the forecasts from
the persistence (no-change) model over the forecast horizon. Thus, a figure of 0.5 means the error was
only half as big as that from the persistence model forecast, and 2.0 means it was twice as big.
0" 100" 200" 300" 400" 500" 600" 700" 800" 900" 1000" 1100" 1200" 1300" 1400" 1500" 1600" 1700" 1800" 1900" 2000"
Yea r &o f &f o re ca s t&
Relative accuracy of forecasts from alternative climate change hypotheses
Warming, and Cooling, versus Persistence
(Rate, ºC p.a.)
(v. Persistence = 1)
1 – 6 1
1 – 10
11 – 100
1 – 1,820
* Monthly forecasts.
**Successive updating used.
Evidence-based climate forecasts for the 21st Century
Our testing used alternative data sources, different time periods, different starting points, and
different horizons. The findings were always the same. Forecasts from the more conservative cooling
hypotheses were more accurate than forecasts from the warming hypotheses. Forecasts from the most
conservative hypothesis, the no-trend model, were always much more accurate. The no-trend model is
consistent with evidence-based forecasting principles and with the state of knowledge about the
behaviour of the Earth’s climate. The IPCC’s alarming-warming model is not. Consistent with
knowledge about the proper model for this situation, the predictive validity tests finds no support for
the global warming hypothesis for forecasting global mean temperatures over this century and beyond.
Our forecasts for each year’s global average temperature for the 100 years to 2113 are that they
will be the same, more or less, as the 2013 global average temperature. We suggest that our forecasts
should be monitored against the University of Alabama at Huntsville’s (UAH) lower troposphere
temperature series because this satellite-based measure provides a better assessment of the global
average than the Hadley (HadCRUT3) series, and because it is fully and openly documented and is,
therefore, less likely to be biased.
Perhaps it is possible to improve on the already very accurate long-term temperature forecasts
from the no-trend model, for example by estimating the global average temperature level from a
weighted average of temperatures over recent years, rather than from only the latest year. We have not
attempted to improve upon our very simple no-change model, however, because the errors of the
forecast from the model are too small to be of concern to policy makers and business decision makers;
the no-trend model forecasts are more than good enough.
Are recent temperature trends so unusual that evidence on past climate is irrelevant?
The global warming alarm derives from the claim that recent increases in temperatures are
unprecedented. As we describe above, the claim is by no means generally accepted. Our long-term tests
of predictive validity provide no evidence that anything unusual has occurred.
In a further test of the claim that there are no precedents, we asked forecasters at the 2012
International Symposium on Forecasting in Boston to forecast the next 25 years for two 50-year
sequences of monthly global mean temperatures from the Hadley data series (Exhibit D). We told the
forecasters that both sequences occurred during the age of industrialism. In other words, both coincided
with exponential increases in atmospheric carbon dioxide.
Exhibit D: Forecasting task using Hadley (HadCRUt3) global mean temperatures
Test your forecasting skills:
Print this page and draw in your forecasts
Monthly global mean temperatures over two half centuries*
Draw in your forecasts for the next 25 years for both charts.
*Both during industrial era
Time 1 25
Of the 51 forecasters who responded, 23 made forecasts that were consistent with carbon-
dioxide causality, while the others forecast little or no trend. Exhibit E shows the data in context and a
trend line for the 25-year period that the forecasters were asked to forecast and for which data are
available. The similarity of the two sequences is inconsistent with the hypothesis that climate changes
are different now than in earlier times.13 In fact, after the decline in the 25 years following series A,
there was an effort to get the government to stop global cooling. However, the warming in the series
after that led to a switch to ask the government to stop global warming. Many of the leaders in the
global cooling movement were then involved in the global warming movement.
Exhibit E: Forecasts for the two similar graphs
What do previous environmental alarms tell us?
Having investigated the forecasting procedures behind the IPCC forecasts and found them to
lack validity, and having found the forecasts to be much less accurate than no-change forecasts, we
were concerned that governments are taking the dangerous manmade global warming alarm seriously,
to the extent that they have already implemented costly policies and regulations. We decided, therefore,
13 The idea behind this comparison came from a presentation by Meyer (2009).
Hadley global mean °C temperature anomalies showing selected half-centuries
to examine whether the global warming alarm is an unusual phenomenon. To do so, we used the
structured analogies approach (Green and Armstrong 2011).
The structured analogies method involves asking experts to think of similar situations to the
situation of interest. Information is then obtained about the outcome of each analogy. We had
previously tested the structured analogies method for forecasting complex situations involving
interactions between parties with conflicting interests, including a special interest group occupying a
public building and demanding taxpayer funding, and an international crisis over access to water. The
research found that a structured search for and analysis of analogous situations produces forecasts that
are much more accurate than the usual method of asking experts what they think will happen (Green
and Armstrong 2007). Other researchers have subsequently found the method useful for forecasting the
outcomes of policy initiatives (e.g., Nikolopoulos et al. 2014).
With the help of domain experts we have, to date, identified 26 analogous situations (Green and
Armstrong 2011). They all began with an allegedly portentous incident or with claims that an apparent
trend was ominous. Searches for evidence supporting each alarm followed, along with calls for
government action. In no case was there recourse to scientific forecasting. The fact that we were able to
identify as many environmental alarm analogies as we did, and the frequency with which they have
occurred in recent times, suggest that they are a common social phenomenon and that the global
warming alarm is not at all unusual. More generally, it is another example in a long history of
calamity forecasts similar to those described in Extraordinary Popular Delusions And The
Madness Of Crowds (MacKay 1841).
Short descriptions of the analogies are provided in Appendix 2 of this chapter.14 Evidence on
the nature and outcomes of all 26 analogies is provided in our online working paper (Green and
Armstrong 2011) at publicpolicyforecasting.com.15 We welcome further evidence on each of these
analogies, invite others to submit their ratings of the analogies for publication at
publicpolicyforecasting.com, and encourage others to propose other environmental alarms in case we
have missed important analogies.
What were the outcomes of the alarms? The forecasts of harmful outcomes all turned out to be
wrong. For the 23 alarms that resulted in government actions, the measures that were taken caused
harm in 20 cases. The alarms faded from public attention slowly over time, but harmful policies have
remained in many cases. We suggest using the Golden Rule of Forecasting to identify and to expose
such false alarms, and to thereby help to minimize the harm that they cause.
14 Some of the descriptions were previously published in Green (2011).
15 That site also includes a list of analogies that had been compiled by Julian Simon.
Climate has varied in the past and can be expected to do so in the future. Mankind has adapted
to both cool and warm periods, and trade and economic growth over the past 300 years has greatly
increased our ability to do so. In that context, forecasts of climate are of little value unless they are for a
strong and persistent trend, and are accurate.
The IPCC “forecasts” are for a strong and persistent trend, but they have been inaccurate in the
short term. Moreover, there is no reason to expect them to be accurate in the longer term. The IPCC’s
forecasting procedures violate all of the relevant Golden Rule of Forecasting guidelines. In particular,
their procedures are biased to advocate for the hypothesis of dangerous manmade global warming.
We found that there are no scientific forecasts that support the hypothesis that manmade global
warming will occur. Instead, the best forecasts of temperatures on Earth for the 21st Century and
beyond are derived from the hypothesis of persistence. Specifically, we forecast that global average
temperatures will trend neither up nor down, but will remain within half-a-degree Celsius (one-degree
Fahrenheit) of the 2013 average.
This chapter provides good news. There is neither need to worry about climate change, nor
reason to take action.
Armstrong, J. S. (Ed.) (2001). Principles of Forecasting, Kluwer: Boston.
Armstrong, J. S. (1980). The Seer-Sucker Theory: The Value of Experts in Forecasting. Technology
Review, June/July, 16–24.
Armstrong, J. S. (1985). Long-range Forecasting. New York: John Wiley.
Armstrong, J. S., Green, K. C., & Graefe, A. (2014). Golden Rule of Forecasting: Be conservative. Journal of
Business Research, forthcoming.
Ball, T. (2014). The Deliberate Corruption of Climate Science. Seattle: Stairway Press.
Batson, C. D. (1975). Rational processing or rationalization? The effect of disconfirmation on a stated religious
belief. Journal of Personality and Social Psychology, 32. 222–227.
Cerf, C, & Navasky, V. (1984). The Experts Speak. New York: Pantheon.
Chamberlin, T. C. (1890). The method of multiple working hypotheses. Reprinted in Science, 148 (1965),
Christy, J. R., Herman, B., Pielke, R., Sr., Klotzbach, P., McNider, R. T., Hnilo, J. J., Spencer, R. W., Chase, T.,
& Douglass, D. (2010). What do observational datasets say about modelled tropospheric temperature
trends since 1979? Remote Sensing, 2, 2148–2169.
Festinger, L., Riecken, H. W., & Schachter, S. (1956). When Prophecy Fails: A social and psychological study of
a modern group that predicted the destruction of the World. Minneapolis, MN: University of Minnesota
Green, K. C. (2011). A history of scientific alarms. IPA Review, December, 13–15.
Green, K. C. & Armstrong, J. S. (2007). Global warming: forecasts by scientists versus scientific forecasts.
Energy and Environment, 18, No. 7+8, 995–1019.
Green, K. C., Armstrong, J. S. & Soon, W. (2009). Validity of Climate Change Forecasting for Public Policy
Decision Making. International Journal of Forecasting, 25, 826–832.
Green, K. C., & Armstrong, J. S. (2011). The global warming alarm: Forecasts from the structured analogies
method. Working Paper. [Available from http://www.kestencgreen.com/green&armstrong-agw-
Hansen, J. E. (2004). Defusing the global warming time bomb. Scientific American, March, 68–77.
Idso, C. D., Carter, R. M. and Singer & S. F. (2014). Climate Change Reconsidered II: Physical Science.
Chicago: heartland Institute.
IPCC (2013) Summary for Policymakers. In Stocker, T. F., D. Qin, G.-K. Plattner, M. Tignor, S. K. Allen, J.
Boschung, A. Nauels, Y. Xia, V. Bex and P. M. Midgley (Eds.), Climate Change 2013: The Physical
Science Basis. Contribution of Working Group I to the Fifth Assessment Report of the Intergovernmental
Panel on Climate Change. Cambridge, United Kingdom and New York, NY, USA: Cambridge
IPCC (1992). In Houghton, J. T. ,Callander, B. A. , & Varney, S. K. (Eds.), Climate change 1992: The
supplementary report to the IPCC scientific assessment. Cambridge, United Kingdom: Cambridge
IPCC (1990). In J. T. Houghton, G. J. Jenkins, & J. J. Ephraums (Eds.), Climate change: The IPCC scientific
assessment. Cambridge, United Kingdom: Cambridge University Press.
Krueger, M. (2007). An unrepentant prognosticator. Gelf Magazine, April 24. [Available from
Kukla, G. J. & Matthews, R. K. (1972). When will the present interglacial end? Science, 178, 190–191.
Laframboise, D. (2011). The delinquent teenager who was mistaken for the world’s top climate expert.
Legates, D. R., Soon, W., Briggs, W. M., & Monckton, C. (2014). Climate consensus and ‘misinformation’: A
rejoinder to Agnotology, scientific consensus, and the teaching and learning of climate change. Science
& Education, in press, doi:10.1007/s11191-013-9647-9.
Lindzen, R. S. (2012). Climate Science: Is it currently designed to answer questions? Euresis Journal, 2 (Winter),
Loehle, C. (2004). Climate change: detection and attribution of trends from long-term geologic data. Ecological
modelling, 171, 433–450
Loehle, C., & McCulloch, J. H. (2008). Correction to: A 2000-year global temperature reconstruction based on
non-tree ring proxies. Energy & Environment, 19, 93–100.
MacKay, C. (1841). Extraordinary Popular Delusions and the Madness of Crowds. New York: Three
Mahoney, M. (1977), “Publication prejudices: An experimental study of confirmatory bias in the peer
review system,” Cognitive Therapy and Research, Vol. 1, No. 2, pp. 161–175.
Matthes F. M. (1939). Report of committee on glaciers, April 1939. Transactions, American Geophysical Union,
McKitrick, R. R. (2013). Encompassing tests of socioeconomic signals in surface climate data. Climate Change,
McKitrick, R. R., & Michaels, P. J. (2007). Quantifying the influence of anthropogenic surface processes and
inhomogeneities on gridded global climate data. Journal Geophysical Research, vol. 112,
McKitrick, R. R. & Nierenberg, N. (2010). Socioeconomic Patterns in Climate Data. Journal of Economic and
Social Measurement, 35(3,4) pp. 149–175.
Michaels, P. J. (1999). Fighting fire with facts. CATO, January 18. [Available from
Nikolopoulos, K., Litsa, A., Petropoulos, F., Bougioukos, V., & Khanmash, M. (2014). Relative performance of
methods for forecasting special events. Journal of Business Research, forthcoming.
Randall, D. A., Wood, R. A., Bony, S., Colman, R., Fichefet, T., Fyfe, J., Kattsov, V., Pitman, A., Shukla, J.,
Srinivasan, J., Stouffer, R. J., Sumi, A. & Taylor, K. E. (2007). Climate Models and Their Evaluation, in
Solomon, S., Qin, D., Manning, M., Chen, Z., Marquis, M., Averyt, K.B., Tignor, M. and Miller, H.L.
eds., Climate Change 2007: The Physical Science Basis. Contribution of Working Group I to the Fourth
Assessment Report of the Intergovernmental Panel on Climate Change. Cambridge, UK and New York,
NY, USA: Cambridge University Press.
Soon, W., Baliunas, S., Idso, C., Idso, S., & Legates, D. R. (2003). Reconstructing climatic and environmental
changes of the past 1000 years: A reappraisal. Energy & Environment, 4, 233–296.
Soon, W., Baliunas, S., Idso, S. B., Kondratyev, K. Y., & Posmentier, E. S. (2001). Modeling climate effects of
anthropogenic carbon dioxide emissions: unknowns and uncertainties. Climate Research, 18, 259–275.
Tetlock, P. E. (2005). Expert political judgement. Princeton.
Vavrus, S., Ruddiman, W. F., & Kutzbach, J. E. (2008). Climate model tests of the anthropogenic influence on
greenhouse-induced climate change: the role of early human agriculture, industrialization, and vegetation
feedbacks. Quaternary Science Reviews, 27, 1410–1425.
Watts, A., Jones, E., McIntyre, S., and Christy, J.R. (2012) An area and distance weighted analysis of the impacts
of station exposure on the U.S. Historical Climatology Network temperatures and temperature trends.
PREPRINT DRAFT DISCUSSION PAPER [available for open peer review here:
• Describe decisions that might be affected by the
• Prior to forecasting, agree on actions to take
assuming different possible forecasts.
• Make sure forecasts are independent of politics.
• Consider whether the events or series can be
Identifying Data Points
• Avoid biased data sources.
• Use unbiased and systematic procedures to
• Ensure that information is reliable and that
measurement error is low.
• Ensure that the information is valid.
• List all important selection criteria before selecting
• Ask unbiased experts to rate potential methods.
• Select simple methods unless empirical evidence
calls for a more complex approach.
• Compare track records of various forecasting
• Assess acceptability and understandability of
methods to users.
• Examine the value of alternative forecasting
Implementing Methods: General
• Keep forecasting methods simple.
• Be conservative in situations of high uncertainty or
Implementing Quantitative Methods
• Tailor the forecasting model to the horizon.
• Do not use "fit" to develop the model.
Implementing Methods: Quantitative Models with
• Apply the same principles to forecasts of
• Shrink the forecasts of change if there is high
uncertainty for predictions of the explanatory
Integrating Judgmental and Quantitative Methods
• Use structured procedures to integrate judgmental
and quantitative methods.
• Use structured judgments as inputs of quantitative
• Use pre-specified domain knowledge in selecting,
weighing, and modifying quantitative models.
• Combine forecasts from approaches that differ.
• Use trimmed means, medians, or modes.
• Use tracked records to vary the weights on
• Compare reasonable methods.
• Tailor the analysis to the decision.
• Describe the potential biases of the forecasters.
• Assess the reliability and validity of the data.
• Provide easy access to the data.
• Provide full disclosure of methods.
• Test assumptions for validity.
• Test the client's understanding of the methods.
• Use direct replications of evaluations to identify
• Replicate forecast evaluations to assess their
• Compare forecasts generated by different
• Examine all important criteria.
• Specify criteria for evaluating methods prior to
• Assess face validity.
• Use error measures that adjust for scale in the
• Ensure error measures are valid.
• Use error measures that are not sensitive to the
degree of difficulty in forecasting.
• Avoid error measures that are highly sensitive to
• Use out-of-sample (ex-ante) error measures.
• Do not use tests of statistical significance.
• Do not use root mean square error (RMSE) to
make comparisons among forecasting methods.
• Base comparisons of methods on large samples
• Conduct explicit cost-benefit analysis.
• Use objective procedures to estimate explicit
• Develop prediction intervals by using empirical
estimates based on realistic representations of
• When assessing prediction intervals, list possible
outcomes and assess their likelihoods.
• Obtain good feedback about forecast accuracy
and the reasons why errors occurred.
• Combine prediction intervals from alternative
• Use safety factors to adjust for overconfidence in
• Present forecasts and supporting data in a simple
and understandable form.
• Provide complete, simple, and clear explanations
• Present prediction intervals.
Learning That Will Improve Forecasting Procedures
• Establish a formal review process for forecasting
• Establish a formal review process to ensure that
forecasts are used properly.!
Appendix 2: Analogies to dangerous manmade global warming alarm
Population growth and famine (Malthus)
Based on Benjamin Franklin’s observation
that animals and plants reproduce until they
exhaust resources then starve and die,
Malthus extrapolated that humans would
share this fate as a result of geometric
population growth and linear resources
growth. He later realized that foresight and
innovation prevent this fate in humans.
Timber famine economic threat – 1865
Forecasts that we will run out of wood for
construction and paper occur from time to
time around the world. Despite the alarms,
the world’s forested area has increased since
WWII, as has wood production. Planting
and efficiency have increased in response to
demand and competition.
Uncontrolled reproduction and
degeneration (Eugenics) – 1883
The idea of eliminating “undesirable”
people and encouraging elites to breed was
motivated by forecasts of being by
“inferior” people. Intellectuals in the 1920s
and 1930s were advocates. After the Nazis,
the policy is anathema to many, though
population control policies in different
guises are still advocated by some.
Lead in petrol and brain and organ
damage – 1928
Early observations of harm from heavy
exposure of industrial workers led to
speculation of wider community effects
from exposure to lead from petrol. A ban
was called for in the U.K. in 1928. It was
phased out in Australia in 2002. There is no
scientific evidence that lead from vehicle
exhaust was harmful in practice.
Soil erosion and agricultural – 1934
Despite periodic alarms from lobbyists and
politicians over soil being washed and
blown away, there has been a net gain in
soil on most U.S. cropland; and erosion
rates have been slowing. In Australia, too,
soils have improved with fertilization and
new plant species, and erosion has declined
as land management practices have
Asbestos and lung disease – 1939
People who worked with some kinds of
asbestos over extended periods developed
lung cancer more often. Researchers and
lobbyists, extrapolating to the general
population, hypothesized effects from
miniscule exposure to any kind of asbestos,
and improperly predicted millions of excess
deaths. Asbestos was banned in Australia in
Fluoride in drinking water health effects –
Fluoride is poisonous in quantity, but occurs
naturally in drinking water in low
concentrations. One part-per-million
reduces dental decay. Some scientists have
warned of potential ill effects and some
communities reject fluoridation of water
supplies. Claims of ill effects at 1ppm are
DDT and cancer – 1962
In Silent Spring, Rachel Carson forecast that
birds would die out and people would be
afflicted by cancer due to increasing
exposure to the insecticide DDT. There was
no plausible biological mechanism
identified and research failed to support the
claims. DDT was nevertheless banned.
Millions have died unnecessarily from
malaria as a result.
Population growth and famine (Ehrlich) –
Butterfly biologist Paul Ehrlich warmed up
Malthus, and also forecast global cooling
and, later, global warming disasters. In The
Population Bomb, Ehrlich wrote, “The
battle to feed humanity is over. In the
1970s, the world will undergo famines.
Hundreds of millions of people are going to
starve to death.”
Global cooling; through to 1975 – 1970
Temperatures had been declining since the
end of WWII, and some scientists forecast
an imminent ice age. Alarming forecasts
have alternated between ice ages and the
opposite several times since at least the 19th
Century. Media coverage of this most
recent cooling alarm ceased after
temperatures warmed again.
Supersonic airliners, the ozone hole, and
skin cancer, etc. – 1970
Forecasts that water vapour emitted by
planes would harm the ozone layer led to
calls to ban them. The forecast cause of
ozone decline changed to nitrogen oxides.
The forecast outcome then vacillated
between harm and benefit over the decade.
Then the alarm faded from public
Environmental tobacco smoke – 1971
Lobbyists extrapolated the evidence that
long-term smoking causes lung cancer and
heart disease to forecast the same effects
from “second hand” tobacco smoke. Proper
epidemiological studies failed to support the
forecasts of serious harm to third parties.
(Which is not to say that smoking around
others is not an unpleasant and
Population growth and famine (Meadows)
Computer modelling sponsored by the Club
of Rome predicted burgeoning population,
exhausted resources, and famine. With
minor and realistic changes in assumptions,
however, the model would produce
sanguine forecasts. The Club recanted the
original forecasts in 1976.
Industrial production, acid rain, and
forests – 1974
Sulphur dioxide from burning coal can
increase the acidity of rain. Scientists
ascribed fish deaths and predicted harm to
forests and people. The U.S. National Acid
Precipitation Assessment Program found
little environmental damage and no harm to
people. Acidity of rain varies naturally. The
costly Clean Air Act is still in effect.
Organophosphate poisoning – 1976
Insecticides that work by enzyme inhibition
rapidly degrade in the environment.
Forecasts of adverse health effects among
agricultural workers et al., and the general
population, are common. While there are
cases of deliberate poisoning and accounts
of nerve damage among workers, harm to
the general population is not evident.
Electrical wiring and cancer, etc. – 1979
A small epidemiological study reported an
association between hypothesized exposure
to electromagnetic fields and childhood
leukaemia. In the U.S., regulations intended
to reduce exposure cost $1 billion annually.
Thousands of studies have failed to
establish a link between actual exposure and
any health effect.
CFCs, ozone hole, and skin cancer – 1985
Speculation that the Earth’s ozone layer was
being depleted by chlorine from
chlorofluorocarbons, and forecasts that skin
cancer rates would increase, led to an
international ban. Knowledge about the
relationships is poor. Chlorine from the sea
is 400 times CFC peak production.
Replacement refrigerants are dangerous.
Listeria in cheese – 1985
Listeria monocytogenes occurs in soft
cheeses, but most strains do not cause
listeriosis. Listeriosis can be fatal for high-
risk people such as young children.
Detection is now easy, resulting in listeria
being more often identified in food and
deaths more often being attributed to it than
in the past, thus precipitating alarms.
Radon in homes and lung cancer – 1985
Radon historically caused lung cancer in
miners working in dusty uranium-rich
mines. A small survey found elevated levels
in some houses and the U.S. EPA estimated
8 million homes were affected and forecast
up to 30,000 lung cancer deaths per annum.
More rigorous studies have shown any
effect is small, or non-existent.
Salmonella in eggs – 1988
Careless investigations of food poisoning in
Britain attributed some to eggs. Minister
Currie asserted that “most” egg production
was infected with salmonella. Demand
plummeted. Costly flock testing was
imposed. There were calls to kill the entire
laying flock—one million were. Salmonella
has likely never been present inside eggs.
Environmental toxins and breast cancer –
Long Island breast cancer survivor and
lobbyist Barbara Balaban and some
scientists speculated, against understanding
of biological mechanisms, that toxins in the
environment, such as DDE and PCBs, were
causing cancer. Congress ordered studies
that cost $30 million. They found no link.
Mad cow disease (BSE) – 1996
Speculation that a variant Creutzfeldt-Jakob
disease might be contracted from eating
beef from cattle with BSE, and forecasts
that the disease would kill 10 million people
by 2010, led to the slaughter of 8 million
cattle in Britain at a cost to the taxpayer of
£3.5 billion. Suspected vCJD deaths never
exceeded 28 per year and any link to BSE
Dioxin in Belgian poultry – 1999
Dioxins occur naturally, and are produced
incidentally and deliberately by industry.
Some are toxic. When breeder chickens
became ill, the cause was traced to dioxin-
contaminated feed. Seven million chickens
and 60,000 pigs were destroyed. People
were consequently exposed to more dioxin
as a result of substituting fish for chicken in
Mercury in fish – 2004
Extrapolating from insupportably low
“safe” levels, a U.S. EPA employee
predicted that 630,000 babies would be
born with potential brain damage each year.
Women were warned to avoid fish.
Mercury occurs naturally in the
environment and most Japanese have higher
than EPA “safe” levels from eating a health-
promoting high-fish diet.
Mercury in childhood inoculations and
autism – 2005
Robert F. Kennedy, Jr. claimed on CBS
News that “The science connecting brain
damage with thimerosal is absolutely
overwhelming.” Thimerosal is a vaccine
preservative that contains mercury that the
industry claims is safe. When it was
eliminated, autism cases continued to climb.
Researchers found no link.
Cell phone towers and cancer, etc. – 2008
Periodically, community activists raise
alarms that the towers will cause cancer and
miscellaneous other health problems. The
towers transmit and receive weak
radiofrequency signals. The signals are
centimetres-long wavelength non-ionizing
radiation that, like heat and visible light,
cannot damage DNA. Scientific studies
have found no health effects.