Guest posting by Richard Tol
The Fourth Assessment Report (AR4) of Working Group II (WG2) of the Intergovernmental Panel on Climate Change (IPCC) has been discussed extensively in recent months. A number of errors were discovered. Few documents are without fault. What is surprising, however, is that the IPCC has denied obvious mistakes; and that the errors all point towards alarmism about the impacts of climate change.
The WG3 report did not attract the same scrutiny. This could create the impression that WG3 wrote a sound report. That impression would be false. Just as WG2 appears to have systematically overstated the negative impacts of climate change, WG3 appears to have systematically understated the negative impacts of greenhouse gas emission reduction.
1. Climate policy could stimulate economic growth and create jobs
I will first focus on Chapter 11: Mitigation from a cross-sectoral perspective. This chapter assesses the economic costs of greenhouse gas emission reduction.
The first and second order draft (FOD, SOD) of the chapter and the review comments can be found here:
The first order draft omits two crucial tables, both on the cost estimates of emission abatement. One of the key results of Chapter 11 therefore has not gone through the double review that is a hallmark of the IPCC procedures.
Section 11.4.4 has the following paragraph:
The Stern Review (2006), which was commissioned by the UK Treasury, also considers a range of modelling results. Drawing on estimates from two studies, it reports the costs of an emissions trajectory leading to stabilization at around 500-550ppm CO2-eq. One of the two studies (Anderson, 2006) calculates estimates of annual abatement costs (i.e. not the macro-economic costs) of 0.3% of GDP for 2015, 0.7% for 2025 and 1% for 2050 from an engineering analysis based on several underlying reports of future technology costs. His uncertainty analysis, exploring baseline uncertainties about technology costs and fuel prices, shows a 95% prediction range of costs from -0.5% to +4% of GDP for 2050. The other study is a meta-analysis by Barker et al. (2006a) and looks at the macro-economic costs in terms of GDP effects. The study aims to explain the different estimates of costs for given reductions in global CO2 in terms of the model characteristics and policy assumptions adopted in the studies. With favourable assumptions about international flexibility mechanisms, the responsiveness and cost of low-carbon technological change, and tax reform recycling revenues to reduce burdensome taxes, costs are lowered, and in some cases become negative (i.e. GDP is higher than baseline).
Three papers are referred to: Anderson, Barker, and Stern. None of the three was peer-reviewed. Anderson and Stern were omitted from the Second Order Draft. Barker et al. (2006) was referred in the SOD, as follows:
Research has continued to focus on differences in various cost estimates across models (Weyant, 2000; Weyant, 2001; Lasky, 2003; Weyant, 2003; Fischer and Morgenstern, 2005; Barker et al., 2006).
These prices and costs are largely determined by the approaches and assumptions adopted by the modellers, with GDP outcomes being strongly affected by assumptions about 35 technology costs and change processes (see 11.5 above), the use of revenues from permits and taxes (see 11.4 above), and capital stock and inertia (considered below) (Fischer and Morgenstern, 2006, Barker et al., 2006).
That is, in the SOD, the gray publication Barker et al. (2006) was used to support peer-reviewed material; and Anderson (2006) was omitted. In the chapter, Anderson (2006) and Barker et al. (2006) are used to support the notion that "costs are lowered, and in some cases become negative".
Section 11.8.2 reads as follows:
A number of studies point out that investments in greenhouse gas mitigation could have a greater impact on employment than investments in conventional technologies. The net impact on employment in Europe in the manufacturing and construction industries of a 1% annual improvement in energy efficiency has been shown to induce a positive effect on total employment (Jeeninga et al., 1999). The effect has been shown to be substantially positive, even after taking into account all direct and indirect macro-economic factors such as the reduced consumption of energy, impact on energy prices, reduced VAT, etc. (European Commission, 2003) The strongest effects are seen in the area of semi-skilled labour in the building trades, which also accounts for the strongest regional policy effects. Furthermore, the European Commission (2005) estimates that a 20% saving on present energy consumption in the European Union by 2020 has the potential to create, directly or indirectly, up to one million new jobs in Europe.
Meyer and Lutz (2002) use the COMPASS model to study the carbon taxes for the G7 countries. They find that recycling revenues via social security contributions increases employment by nearly 1% by 2010 in France and Germany, but much less in US and Japan. Bach et al. (2002), using the models PANTHA RHEI and LEAN, find that the modest ecological tax reform enacted in Germany in 1999-2003 increased employment by 0.1 to 0.6% by 2010. This is as much as 250,000 additional jobs. There is also a 2-2.5% reduction in CO2 emissions and a negligible effect on GDP. The labour intensity of renewable energy sources has been estimated to be approximately 10 times higher in Poland than that of traditional coal power (0.1-0.9 jobs/GWh compared to 0.01-0.1 jobs/GWh). Given this assumption, government targets for renewable energy would create 30,000 new jobs by 2010 (Jeeninga et al., 1999).
In a study of climate policies for California, Hanemann et al. (2006) report small increases in employment for a package of measures focusing on the tightening of regulations affecting emissions.
Six studies are cited to support the notion that emission reduction creates jobs. Only one of the six is peer-reviewed: Bach et al. (2002). That paper adds:
Another important assumption shaping the results relates to the way in which wage formation is modeled. In our core simulations we assume that the induced increase in employment does not trigger higher wage claims. If, instead, it is assumed that the trade unions react to employment growth by increasing their wage demands, this could significantly dampen economic growth and neutralize the positive employment effects.
That is, the positive impact of climate policy on employment is fragile.
This is indeed the conclusion of Patuelli, Nijkamp and Pels (2005, Ecological Economics, 55 (4), 564-583). Their meta-analysis was published before the AR4 deadline, but was overlooked by the authors. They assess 94 estimates of the impact of ecological tax reform and find an average increase of employment of 0.64% but with a standard deviation of 1.33%. The 6 studies cited in Chapter 11 only have positive effects - Bach et al. after censoring - and are thus not representative of the literature.
Section 11.8.2 does not alert the reader to the fact that climate policy would only have a positive impact on employment if the revenues of a carbon tax or an auction of emission permits are used to reduce taxes on labour. There is no positive impact on employment if emission reduction is achieved by subsidies on renewables or if emission permits are given away for free - as is common.
Similarly, it is well-accepted in the literature that emission abatement would stimulate economic growth if policy reform is smart and well-designed. By the same token, a badly designed policy could greatly enhance the costs. Boehringer et al. (2009) estimate, for instance, that the EU 20/20/2020 package is more than twice as expensive as needed.
Chapter 11 of AR4 WG3 suggests that climate policy could stimulate economic growth and would create jobs. These claims are supported by gray literature only, and they are biased.
2. Technological change
To a first approximation, the costs of emission reduction are driven by the difference in the costs of fossil energy and its carbon-neutral alternatives. It costs about 4 cents per kilowatthour to make electricity with coal, about 8 cents with wind, and about 24 cents with solar. That is today. The bulk of emission reduction will take place in the future. Estimates of the costs of emission reduction are therefore largely driven the assumed evolution of the prices of carbon-neutral energy sources, relative to the prices of fossil fuels.
It is hard to predict future price changes. This implies that the estimates of the costs of emission reduction are very uncertain. If you assume rapid technological progress in renewable energy and scarce oil and gas, emission reduction will be cheap. If you assume slow technological progress in renewables and rapid progress in unconventional oil and gas, emission reduction will be expensive.
Models used to assume that technological change in energy is independent of climate policy. This assumption has been challenged, and rightly so. There is ample evidence than inventors and innovators respond to policy and price signals. There are now a number of models in which technological progress is partly driven by climate policy. The Summary for Policy Makers (SPM) of AR4 WG3 states
In the models that adopt these approaches, projected costs for a given stabilization level are reduced; the reductions are greater at lower stabilisation levels.
Studies that assume the possibility that climate change policy induces enhanced technological change also give lower costs.
Although most models show GDP losses, some show GDP gains because they assume [...] that more technological change may be induced by mitigation policies.
Modelling studies [...] show carbon prices rising to 20 to 80 US$/tCO2-eq by 2030 and 30 to 155 US$/tCO2-eq by 2050. For the same stabilization level, studies [...] that take into account induced technological change lower these price ranges to 5 to 65 US$/tCO2-eq in 2030 and 15 to 130 US$/tCO2-eq in 2050
The SPM asserts three times that induced technical change reduces the costs of abatement, and once that it may even revert the sign. Chapter 11 is the source of these claims. What evidence does it offer?
The main source of information is the Innovation Modelling Comparison Project (IMCP), which was led by Barker, Edenhofer and Grubb who were all lead authors of Chapter 11. Most of the models surveyed indeed show a drop in emission reduction costs if innovation responds to policy. The extent to which costs fall depends, among other things, on the assumed "crowding-out" - that is, if economies invest more in research and development (R&D) of clean energy, do they then invest less in other R&D? Chapter 11 identifies Nordhaus (2002) as the study that assume the greatest crowding-out: Energy R&D comes at the expense of other R&D. Chapter 11 (p. 653) writes: While some models find a large reduction in mitigation costs (e.g. Popp, 2006a), some find small impacts (e.g. Nordhaus, 2002).
Nordhaus (2002) writes: the introduction of induced innovation increases the discounted value of world consumption by US$238 billion. This is about 40 percent of the welfare gain from substitution policies,
which is $585 billion
That is, Nordhaus reports a small gain in welfare if the model includes induced technological change; Nordhaus finds a welfare gain because the benefits of avoided climate change are larger than the costs of emission reduction. In Nordhaus' results, welfare falls by $585-$238=$248 billion. While Chapter 11 claims that Nordhaus finds a small but positive impact, Nordhaus in fact finds a negative impact.
Nordhaus explains: The primary reason for the small impact of induced innovation on the overall path of climate change is that the investments in inventive activity are too small to make a major difference unless the social returns to R&D are much larger than the already-supernormal returns. R&D is about 2 percent of output in the energy sector, while conventional investment is close to 30 percent of output. Even with supernormal returns, the small fraction devoted to research is unlikely to outweigh other investments.
That is, energy is a small factor in the economy; focusing R&D on energy has a large opportunity cost as energy R&D detracts from other R&D.
Nordhaus' result is well in line with the more theoretical work by Lans Bovenberg, Larry Goulder, Adriaan van Zon, Sjak Smulders and others. In fact, Smulders shows that an incomplete specification of R&D tends to lead to cost reductions, while a complete specification tends to lead to cost increases. This issue was raised by two referees of the FOD. The authors respond thus: A very few authors (e.g. Smulders) have found that allowing for ETC in top-down models increases costs, and many have found that it reduces them. This is not a consensus, but it does suggest that the balance of findings is that inclusion of ETC in the modelling reduces the cost estimates.
That is, the existence of Smulders' work is acknowledged, but its theoretical superiority is not.
The issue was again raised by a referee of the SOD. The authors respond thus: The text is describing the literature. ITC through LBD reduces the costs in the model applications reviewed.
That is, Smulders' work is no longer deemed relevant.
In the published version of the chapter, Smulders appears as follows: There have been many reviews (see Clarke and Weyant, 2002; Grubb et al., 2002b; Löschel, 2002; Jaffe et al., 2003; Goulder, 2004; Weyant, 2004; Smulders, 2005; Grübler et al. 2002; Vollebergh and Kemfert, 2005; Clarke et al., 2006; Edenhofer et al., 2006b; Köhler et al., 2006; Newell et al., 2006; Popp, 2006b; Sue Wing, 2006; Sue Wing and Popp, 2006).
A paper that was known to give a contradictory result in the FOD, was hidden in the chapter.
The Executive Summary of Chapter 11 reads: Using different approaches, modelling studies suggest that allowing for endogenous technological change reduces carbon prices as well as GDP costs, this in comparison with those studies that largely assumed that technological change was independent of mitigation policies and action.
I would argue that the higher quality studies show the opposite of this conclusion. Others may disagree with me, but one cannot deny that the literature is ambiguous. Chapter 11 claims a certainty that does not exist.
Chapter 11 (p. 650) writes: The TAR [...] reported that endogenizing technological change could shift the optimal timing of mitigation forward or backward (8.4.5). The direction depends on whether technological change is driven by R&D investments (suggesting less mitigation now and more mitigation later, when costs decline) or by accumulation of experience induced by the policies (suggesting an acceleration in mitigation to gain that experience, and lower costs, earlier).
This is an accurate summary of the TAR and indeed the literature. However, on p. 651, we read: Learning-by-doing implies that larger (and more costly) efforts are justified earlier as a way to lower future costs.
That is a remarkable turnaround. An ambiguous finding (up or down) is turned into a clear result (up). What is more remarkable is that there is no discussion of this at all in Chapter 11: No new studies are cited that support the claim on p. 651. Chapter 11 could have cited Schwoon and Tol (2006, Energy Journal, 27 (4) 25-60; working paper available since 2004), who show that, if anything, the literature has shifted in the opposite direction.
3. Selective results in the Summary for Policy Makers
Table SPM.4 summarizes the costs of emission reduction in 2030. The title comes with a footnote: "GDP reduction would increase over time in most models after 2030". Deep cuts in emissions would come after 2030, and the real costs of emission reduction would therefore be felt later. The cost estimates in Table SPM.4 are low by construction, not because emission reduction is cheap.
Table SPM.6 shows the costs of emission reduction in 2050. This table does not warn that the bulk of emission reduction and its costs will be in the second half of the century. There is no table on the costs of emission reduction in 2100.
Tables SPM.4 and SPM.6 show the reduction in economic growth for three alternative targets, averaged over a number of studies. For 2050 (SPM.6), the results are a loss of economic growth of 0.05% per year if greenhouse concentrations are stabilized between 590-710 ppm CO2eq; and 0.10% if the target is between 535-590 ppm CO2eq. That is, costs double if the target becomes considerably more stringent. However, the economy slows down by 0.12% per year if the target is between 445 and 535 ppm CO2eq. Although the target becomes substantially more stringent, costs increase by only a little bit!
This is an amazing result. The models assessed by the IPCC all have that abatement costs grow and accelerate as targets become more stringent. Typically, doubling the rate of emission reduction would lead to a quadrupling of costs. The cost curve in SPM.6 (and SPM.4) bends the wrong way: Incremental costs fall as policy become stricter.
This was not picked up by the referees of the SPM because neither Table SPM.4 nor Table SPM.6 appeared in the drafts circulated for comment.
This travesty is partly explained in footnote g: "The number of studies that report GDP results is relatively small and they generally use low baselines."
Table SPM.5 specifies the numbers: 118 studies estimated the costs of stabilizing atmospheric concentrations between 590 and 710 ppm CO2eq; 21 between 535 and 590 ppm CO2eq; and 24 between 445 and 535 CO2eq.
There are a large number of models that estimate the costs of emission reduction. Some have high costs, and others have low costs. Modelers self-censor their results, or are censored by referees and editors. If a relatively lenient target implies already relatively high costs, then there is no reason to show the results for more stringent targets. More stringent targets would lead to unacceptably high costs. Why waste journal pages on unrealistic scenarios?
This implies that only the "cheap models" ran the most stringent scenarios. The "expensive models" did not report the results, did not try to run these scenarios, or tried and failed. Clarke et al. (2009) investigate this matter, as do Tavoni and Tol (2009).
Furthermore, footnote g reveals that even the "cheap models" could only meet the most stringent targets if the no-policy scenario has benignly low emissions to start with.
In other words, the numbers in Tables SPM.4 and SPM.6 cannot and should not be compared to one another. The results for the relatively lenient targets are representative for the literature. The results for the relatively stringent targets suffer from selection bias.
Tables SPM.4 and SPM.6 cherry-pick results. These tables are misleading.
Richard Tol, 28. February 2010