Investment in new coal-fired power plants after the COVID-19 pandemic: experts expect 170–270 GW of new coal

The focus of this study is on the countries with the largest coal pipeline as of July 2021: China, Bangladesh, India, Indonesia, Laos, Mongolia, Pakistan, Turkey, Vietnam, and Zimbabwe. 89.5% of the coal-fired power capacity globally in the pipeline is concentrated in these countries as of July 2021 (Global Energy Monitor 2021). Between September 2021 and November 2021, we conducted 29 interviews (see supplementary table 2 for a list of the interviewees and supplementary table 3 for a detailed overview of the study timeline). We selected knowledgeable experts by using a snowballing process (Cohen and Arieli 2011). The snowballing process develops in two stages: first by selecting and contacting an initial pool of experts, and then asking those experts further knowledgeable experts in their network. This is an approach used in other expert elicitations (Jenni et al 2013), particularly when the population of people experts on the questions being asked is limited. The initial pool of experts was identified from two main sources: we reviewed the academic and grey literature for recent studies on coal power capacity and COVID-19 and we contacted employees and collaborators working for the Global Coal Plant Tracker project (Global Energy Monitor 2021) which have been monitoring the coal pipeline since 2015. After interviewing those initial contacts, we asked them to provide further experts whom they deemed suitable to carry out the same elicitation procedure and who could provide a different perspective. Such a snowballing procedure ensured that we only selected respondents whose domain of expertise falls clearly in the questions we are asking of them.

Because the snowballing process relies on recommendations from existing experts, it can lead to a biased sample that over-represents or excludes important perspectives, as any small bias in selecting the initial pool could compound as the selection process continues (Salganik and Heckathorn 2004). However, in the case of coal-fired power plants experts, this bias might be mitigated by the small nature of the total pool of potential experts. Furthermore, relying on two sources of initial contact increases the possibility of contacting most of the pool.

Most of the interviewees in the final pool came from think tanks or research institutions (23 out of 29), two of them were from international organizations, two of them from universities, and the other two from national NGOs.

The interview was based on a structured script to ensure consistency and consisted of three parts (see supplementary table 1 for details). We followed best practices in the literature to mitigate the risk of common biases such as anchoring, overconfidence, and availability bias (Aspinall 2010, Morgan 2014). To mitigate the effect of anchoring and overconfidence, the interviewees were introduced to the concept of such cognitive biases (Morgan 2014). To avoid that the interviewee would focus only on recent developments or specific factors (availability bias) we asked the interviewees to identify all the most relevant factors contributing to the realization or cancellation of the coal pipeline (O'Hagan 2019). After such initial set of questions, we conducted the formal elicitation process. As a further measure to mitigate overconfidence and anchoring, the interviewees were asked to first provide an upper and lower bound of their estimates. We explicitly asked for extreme case scenarios (with cumulative probabilities of 5% and 95%, respectively). Having defined these boundaries, we performed a roulette elicitation (Daneshkhah et al 2006) following guidelines from Oakley and O'Hagan (2019).

In the third part, we asked the interviewee to imagine a scenario without the COVID-19 pandemic, with economic growth following a business as usual pathway since 2020 and with no changes in electricity demand patterns and repeated the elicitation procedure. We are less confident that the counterfactual exercise can be interpreted as causal. While one of the strengths of the expert elicitation approach is its flexibility in asking about specific comparisons of counterfactual outcomes (Verdolini et al 2018b), we are still limited by the scope and extent of each respondent's knowledge and imagination. When asked to imagine a counterfactual scenario without COVID-19, experts may have incomplete information, and their responses may be influenced by factors such as their personal beliefs and more recent experiences. Yet, a real counterfactual world without a global pandemic does not exist. Thus, the experts informed counterfactual can provide a valuable complement to other counterfactual exercises based on quantitative data extrapolation (Morgan 2014). Furthermore, this is the only method that can inform policymakers and researchers about the long-term effects of COVID-19.

We elicited experts' opinions about the probability of the current coal pipeline to be built using a roulette elicitation (for more details see supplementary figure 1). With such method, we obtain a discrete probability density function describing the likelihood of specific amounts of capacity to be built. For each expert, we transform these probability density functions into discrete cumulative probabilities. We then linearly smooth those probabilities and aggregate the resulting probability density functions by country using a simple average. We aggregate those experts' probability density functions for each country as the average of multiple experts can provide an accurate prediction if their biases go in different directions. Generally, obtaining information from multiple experts and aggregating the results increases the chances to cover a broad information base (Clemen and Winkler 1999). Some studies suggest to weight the interviewees based on expertise (Aspinall 2010); however, this practice is disputed in the literature (Bolger and Rowe 2015). As in our case, it would be difficult and subjective to judge the relative expertise of each expert, we decided on an unweighted simple average. We summarize the resulting average cumulative distribution function into a boxplot showing the relevant centiles of the cumulative distribution function (see supplementary figure 3 for an example). Sections 3.1 and 3.2 present the aggregate results by country; disaggregated results for each interviewee is available in supplementary figures 2 and 4.

To identify the underlying reasons behind the cancellation or realization of the current pipeline we qualitatively coded the interviewee's responses to the second part of the interview (results are presented in section 3.3); 27 out of 29 interviews were recorded and transcribed. We followed an open coding approach to code the determinants which were mentioned by the interviewee as important. The code structure was iterated among the co-authors, and codes were clustered around main categories. Aggregated categories and sub-codes are reported in supplementary table 5 together with their frequency. Finally, each determinant was also coded as 'Very important', 'Somehow important', or 'Not important' depending on the phrasing of the interviewee.

To assess the compatibility between the elicited capacity additions and Paris compatible mitigation pathways we compared our results with projections from the integrated assessment models (results are presented in section 3.4). We used vetted mitigation pathways underpinning the Sixth Assessment Report (AR6) by the Intergovernmental Panel on Climate Change (IPCC) (Byers et al 2022). To compare our results with the results from the IAMs, we first had to extrapolate our results until 2050. The total coal-fired power capacity expected to be online by 2050 is based on: data from Global Energy Monitor (2021), the experts' elicited expectations about new plants coming online, and a set of simple assumptions. We used the elicited values of the interviewees to assess how many of the plants planned and under construction will likely come online. We assume that the plants under construction as of July 2021 will be built before 2023, and the plants in the pipeline before 2032. We use the planned retirement year when available for a given plant or a standard assumption of a lifetime of 20, 30, or 40 years. Plants older than the assumed lifetime are assumed to be retired within the next 5 years from the oldest to the youngest.

We summarize our results by aggregating the experts' opinions by country and by presenting the most relevant percentiles of the probability distribution (figure 1).

Zoom In Zoom Out Reset image size

In the base case scenario, experts consider it likely that no new capacity will be built except for the ones already announced, permitted, or under construction (for ease of exposition, we refer to our median estimates as 'likely' realizations, and report in square brackets the 25th–75th percentiles). In total across countries, only 50% (40%–63%) of the coal pipeline is expected to be built. In absolute terms, 215 GW (169 GW–270 GW) are likely to be built. Aggregation masks large heterogeneity between countries.

In China, experts predict that 50.6% (38.9%–66.1%) of what is currently planned will likely be built. Such realization rate is above the historical rate of 45% (calculated for each country using the historical data between 2015 and 2022 from Global Energy Monitor (2021)). This corresponds to 132 GW (101 GW–172 GW) being built (of which 96.7 GW are already under construction). The uncertainty around those estimates is large compared to other countries: the interquartile range in China is 27%, meaning that estimates from the experts vary by about one-third between the 25th and 75th percentile.

In Pakistan and Zimbabwe, experts consider it likely that less than 50% of what is currently planned will be built ((46%–62%) and (37%–62%) respectively). The total capacity in GW in those two countries is smaller than in China: the likely cancellation rates correspond to 3.9 GW (3.4 GW–4.6 GW) and 2 GW (2.7 GW–3.4 GW), respectively. In Pakistan, almost no new capacity apart from what is already under construction will likely be realized, while in Zimbabwe large compared to other countries: the interquartile range in China is 27%, meaning that estimates from the experts vary by about one-third between the 25th and 75th percentile.

In India and Indonesia, cancellations are less pronounced compared to other countries. In India, around 72% (60%–84%) of currently planned plants are likely to be built. This corresponds to 40 GW (33 GW–46 GW), of which 34 GW are already under construction. Similarly, in Indonesia likely realizations (74% (65%–79%)) will correspond to 22.5 GW (20 GW–24 GW) (12 GW already under construction). For both countries, rates of realization historically have been lower. Indonesia, however, is the only country that has a probability higher than 5% to expand its pipeline compared to what is currently planned. The uncertainty around these estimates is low for Indonesia and higher for India.

In Vietnam, Bangladesh, Turkey, Laos, and Mongolia, experts predict substantial cancellations. Vietnam and Turkey are the only countries where experts considered it likely that less than what is already under construction might in the end be realized. In Vietnam, likely realizations are 26% (13%–35%), while 31% of the pipeline is already under construction. In Turkey, likely realizations are 8% (3%–16%), while 11% of the pipeline is already under construction. Such cancellation rates deviate consistently from historical trends: power plants under construction are almost always finalized, and in Vietnam, the average historical realization rate of the pipeline was 53% (54% in Turkey). It is considered likely that Bangladesh and Mongolia will cancel almost everything that is currently planned. In the case of Mongolia, this is mostly because a large power project of 5.3 GW, i.e. the Shivee Ovoo power station, makes up almost the entire pipeline. Considering that it has been delayed since 2008, experts consider it unlikely to be realized. In Bangladesh, our interviewee considered it likely that at most one or two plants on top of what is already under construction could eventually come online. Finally, Laos, which has no power plants under construction, will likely build 23% (8%–38%) of its pipeline, corresponding to 1.5 GW (0.5 GW–2.5 GW). Uncertainties around those estimates are the highest for Laos while Bangladesh, Mongolia, and Turkey have the lowest uncertainty of the sample.

The effect of COVID-19 on the likely realization of the current coal pipeline has been limited at most; 15 out of 29 interviewees reported that COVID-19 had no significant effect on their elicitation results. Overall, our elicitation finds that 4 GW (3.4 GW–8 GW) in total have been canceled due to COVID-19 and 10 GW (10 GW–30 GW) have been added (in China only) so the impact is just over 1% of the overall pipeline.

China is the only country in which experts reported that COVID-19 increased the likelihood of more capacity being built. Experts report that 3.9% points more coal will likely be built in response to COVID-19. This has mainly to do with the initial response of the Chinese government to the crisis: 41 GW of coal-fired power plants were granted permission to be built in July 2020 to boost regional GDP. Furthermore, given the quick recovery relative to the rest of the world, China picked up some manufacturing demand from the countries which were still under lockdown in 2021. This resulted in coal price hikes and electricity shortages, which brought to the forefront economic and energy supply issues as opposed to environmental protection in the policy arena.

On the other end of the scale, the greatest effects have been reported in Zimbabwe and Indonesia, with a decrease in the forecasted installed capacity of 9.1% and 2.3% points respectively. In Zimbabwe, the local expert reported that the COVID-19 pandemic redirected already scarce resources from large power infrastructure projects toward healthcare and poverty alleviation measures. Furthermore, financing opportunities for coal shrunk. In Indonesia, the main driver for this decrease was the role of international coal prices. The energy shortage experience in the aftermath of the pandemic reduced the incentive to build domestic power plants, as exporting coal would be much more profitable. In India and Vietnam COVID-19 had little impact on the expected realizations. In those countries, according to our elicitations, COVID will reduce the likely realizations by 3.2% and 3.1%, respectively. In Laos, Bangladesh, Turkey, and Pakistan, we see no effect of COVID-19.

Confirming the finding that the COVID-19 pandemic did not have a strong effect on the realization nor cancellation of the pipeline in most countries, 14 interviewees explicitly mentioned that the COVID-19 pandemic only had a short-term effect on energy demand (due to the lockdowns and the reduction in economic activity and international trade) but that projections expect a quick rebound, leading to no permanent change.

Different factors can contribute to the realization or the cancellation of the current pipeline. A large literature exists that analyses the determinants of coal investments in those countries (Dorband et al 2020, Montrone et al 2021, Ordonez et al 2021, Jakob and Steckel 2022) figure 2 summarizes the most frequently mentioned determinants for each country and the level of agreement about their importance among the interviewees.

Zoom In Zoom Out Reset image size

Politics are more important to the realization of the coal pipeline than techno-economic considerations. Power sector planning in those countries is not solely informed by the least cost option. On the contrary, the power sector is instrumental for a variety of different goals: energy security, international relationships, use of domestic resources, and rent extraction from vested interests.

Implementation and enforcement of climate and energy policies are also critical for the cancellation of the pipeline. Generally, market-based instruments such as a carbon tax have been less relevant than government regulations. Coal moratoria, stopping the financing of coal projects, pollution limits, and net zero targets are the most frequently mentioned policies. Carbon border adjustment in the EU has been mentioned as an important determinant in Turkey and Laos.

The availability of finance will play a critical role in the realization of coal-fired power projects. When a large domestic financial sector is available, such as in China and India, this factor is not particularly relevant, but in all other countries of our sample, lack of finance is identified as a critical determinant. In Turkey, Pakistan, Laos, and Vietnam, the announcement of President Xi in 2021 that Chinese public banks will stop supporting new coal-fired power plants, has been identified by most interviewees as an important determinant of whether the current pipeline will be realized. A shortcoming of finance from China will imply that almost no other international financier is left, as the Republic of Korea and Japan had already withdrawn their financing for coal projects.

Average future GDP growth and costs of renewable energy technologies and batteries have generally been identified as important determinants, as they will respectively determine the level of electricity demand and the supply from the main alternative source. In some countries where prices of RE are already low and where the power sector is already in a situation of overcapacity, many interviewees reported that new coal power plants are not profitable investments any longer. These countries are India, Turkey, Pakistan, and Bangladesh.

Finally, some country-specific determinants emerge. For Laos and Mongolia, the possibility of becoming the 'battery of Asia' and exporting electricity has been identified as a potential determinant of coal power plants being realized. China is the only country where experts identified coal as being used as an economic stimulus.

Only 8 out of 29 interviewees mentioned COVID-19 specific factors when asked about the most important determinants of plants realization. Most of those mentions concerned the increase in coal prices which were registered in October 2021 once all economies started to reopen and supply chain shortages lead to high prices. High coal prices were mentioned as detrimental to future coal investments in China, Vietnam, and Indonesia.

By aggregating the elicited realizations for each country, we can forecast how much capacity will be online by 2050. Figure 3(A) shows that expansion plans are still large and experts predict that at least some of those plans will go ahead. If those plants are to be operated for their historically observed lifetime (37 years), coal-fired power capacity will total 1707 GW in 2030 (median estimates), and 742 GW will likely be online by 2050. All the plants that will be online by 2050 have not yet been built. While the majority of new plants are located in China, most of the countries in our sample will have some coal-fired power capacity operating in 2050 (figure 3(B)).

Zoom In Zoom Out Reset image size

The amount of coal that will be operating over the next decades in the median scenario following historically observed lifetimes would be in line with business as usual scenarios from IAMs (figure 3(C)), but not in line with cost-effective mitigation pathways towards 1.5 °C nor 2 °C (figure 3(C)). Mitigation pathways compatible with 2 °C require a reduction of coal capacity of 90% by 2040 (Riahi et al 2022), this will require limiting the maximum lifespans of power plants to 18 years in our median scenario.