Are there any game-changing emission mitigation policies? Are the proposed climate policies by a country sufficient to deliver its stated climate goal? As a growing number of countries are implementing climate-related policies (such as sectoral standards, carbon taxes, and strategic planning), these questions are crucial because to reach the net-zero goal, the level of carbon emissions has to fall substantially at a speed rarely seen in history, highlighting the need to identify structural breaks in carbon emission patterns and understand forces that could bring about such breaks. Our user-friendly toolbox will help country teams identify and analyze these structural breaks and shed light on potential game-changing policies for mitigating emissions.
Addressing these questions faces at least two challenges: the non-stationarity of climate data, and the difficulty of disentangling contributing factors to emissions. As widely documented in the literature, climate data are highly non-stationary: for example, the CO2 emissions for a given country may accelerate as the country transitions from an agricultural economy to an industrialized economy, posing a significant challenge for using traditional approaches (e.g., panel regressions) to identify the driving forces of emissions. To address this, we apply a well-established machine learning-based method developed by Oxford University (Hendry, 2020). The method has been increasingly applied by researchers, and a training course was offered to the IMF in May 2021. In addition, because numerous factors could contribute to climate change, it is challenging to disentangle the contributions from different factors and offer targeted policy advice. We address this second challenge using a simulation tool based on an intuitive accounting identity.
Specifically, we propose a user-friendly, machine learning-based toolbox built on the analytical framework in Yao and Zhao (2022). The tool consists of four pillars, as indicated in the flow chart:
Pillar I: Country-specific structural breaks of emission patterns identified using machine learning. This pillar itself contains an unconditional analysis (i.e., analyzing the emission pattern without controlling for any other factor) and a conditional analysis (i.e., analyzing the emission pattern after controlling for GDP). In both cases, we identify two types of structural breaks (trend shifts and level shifts) for each of the top 50 carbon emitters (accounting for 92 percent of the worldwide cumulative CO2 emissions as of 2019). We then report the timing of these breaks so that users can place them in the context of significant economic, political, or environmental changes.
Below is an example for China, extracted from Yao and Zhao (2022). After controlling for GDP, our machine-learning tool identifies only three trend shifts in China's carbon emissions, in 1978, 1992, and 2012, respectively. Interestingly, based on our tool, major global climate initiatives such as the Kyoto Protocol (adopted in 1997 and in force in 2005) were not followed by significant downward trend shifts in China's carbon emissions.
Users can then dig deeper into the major developments of the country around the years of trend shifts. For example, 1978 was the year when the country started implementing its influential reform and opening-up strategy that reshaped its economic structure away from heavy industries and reduced emissions (see Pillar II below). And the trend shift in 2012 is likely driven by a dramatic tone shift by the central government in 2013. As documented by Finamore (2018), China experienced a severe air pollution crisis in January 2013, which was widely regarded as a "turning point". In response, in September 2013, the central government announced a four-year plan to significantly improve the air quality of the entire country by end-2017. And in December 2013, the central government even changed the long-standing GDP-oriented evaluation system that may have incentivized local officials to pursue high local GDPs at the cost of the environment. Such a top-down tone shift, accompanied by drastic strategic planning and government-oriented policies, may have driven the (rare) downward shift in China's CO2 emission trend around 2013. These findings could help the country team identify game-changing factors that are either directly or indirectly linked to carbon emissions and thus offer customized policy advice.
Pillar II: Comprehensive driving forces identified using the intuitive Kaya Identity. The Kaya Identity provides a powerful conceptual framework to analyze the drivers of CO2 emissions (Kaya and Yokoburi, 1997; Wang, Tukker, and Rodrigues, 2019). It is widely used by policymakers, scholars, and civil societies (e.g., Bill Gates used a form of the Identity during a TED Talk in 2010). It is an accounting identity expressing the total CO2 emission as the product of four factors: population, GDP per capita, energy intensity (per unit of GDP), and carbon intensity (per unit of energy consumed):
The energy intensity term, Energy/GDP, reflects the economic structure: If a country relies more on heavy industries, then its energy intensity tends to be higher. And the carbon intensity term, CO2/Energy, reflects the energy structure: If a country uses more "dirty" energy sources (e.g., coal), then its carbon intensity tends to be higher. With these decompositions, users can shed light on the underlying driving forces for any structural changes in the carbon emissions identified in Pillar II in a comprehensive way. For example, for countries where population growth is found to be a major driver to carbon emissions, country teams could further analyze issues like gender inequality (and education), given that higher gender inequality is associated with higher fertility (McDonald, 2000).
Our tool will report the above Kaya decompositions for each of the top 50 carbon emitters, both graphically and numerically. Below is the example of China obtained from Yao and Zhao (2022). One clear feature is that since 1978, the contribution of China's energy intensity (blue) to its CO2 emissions has been increasingly negative over time, reflecting the shift of its economic structure away from energy-intensive industries to less intensive industries. This "structural break" in China's economic structure is likely to be the main contributing factor to the structural break in its carbon emissions in the same year identified in Pillar I of our toolbox. Similarly, the chart makes it clear that since 2012, both the energy intensity (blue) and the carbon intensity (red) have been contributing significantly less to China's emissions, consistent with the structural breakpoint in carbon emissions identified in Pillar I (whereas the contribution of population growth since the late 1970s does not display clear structural breaks). These analyses will help country teams dive deeper into specific factors that drive carbon emissions, including demographic, economic, and political factors. Moreover, a unified toolbox such as ours would also ensure that IMF country teams have access to a consistent set of data.
Pillar III: Carbon emission forecasts and policy simulations. Our tool will also allow users to download the time-series data of each Kaya component for each of the top 50 emitters, which will enable users to conduct further analyses, such as building time series models and combining with new information to forecast, say, the Energy/GDP component. Our tool will offer built-in forecasts for the four Kaya components (Population, GDP/Population, Energy/GDP, and CO2/Energy) based on simple time-series models (e.g., ARIMA), where the parameters of the models will be calibrated based on the historical patterns of each component in each country.
These forecasts are intended to show the carbon emissions under the "business-as-usual scenario", i.e., based on historical patterns with no major policy changes. But given that these components highly depend on the country's policies, the tool will also offer an interactive platform to produce forecasts based on users' inputs accounting for new developments and policies. For example, users can specify the Energy/GDP forecast in 2030 based on a particular policy mix, and then revise the forecasts for other years based on the parameter estimates available from our tool. Doing so will also allow users to conduct simulations and assess which combinations of policies could achieve the intended carbon emission reduction goal. Depending on data availability, our tool could also decompose each of the four Kaya components further, e.g., by forecasting the Energy/GDP for each economic sector and then calculating the weighted average.
Pillar IV: Descriptive analyses of past and current climate policies using big data techniques. To position the policy combinations in Pillar III under historical contexts (especially for policies directly related to climate), the tool will also present a dashboard based on the analyses of past and current climate policies using the policy-level comprehensive "Climate Policy Database." Such analyses will serve as potentially useful references while designing the specific details of climate policies. Analyses will include: (1) Constructing a simple Climate Policy Intensity Index (CPII) for the top 50 carbon emitters since 1965, which measures the number and coverage of all the implemented climate change mitigation policies. Despite its simplicity, the CPII can serve as a preliminary overview of existing climate policies from both cross-country and country-specific perspectives. (2) Presenting the distribution of climate policies for each of the top 50 emitters based on categories that are clearly defined in the dataset (such as "policy objective"). (3) For categories that are "fuzzily defined" in the dataset (such as "policy type", which involves long text descriptions for many countries in many years), the tool will first apply natural language processing techniques to re-classify the categories, which are then used to present the distribution of climate policies. (4) Conducting some textual analysis of climate policy documents (available in the dataset) to better understand the directions and effects of policies.
Our proposed tool can contribute to Fund's operations in at least two ways:
First, our tool could help country teams conduct country-specific analysis and facilitate discussions of climate change mitigation policies (including policies about demographic development and economic structures). Our machine learning-based tool (Pillar I) identifies the years when the country's carbon emissions experienced trend shifts and level shifts, and the Kaya Identity decomposition (Pillar II) points to specific contributing (demographic, economic, and political) factors that drive these structural breaks in emissions. Moreover, our simulation tool (Pillar III) provides a forward-looking perspective and helps country teams quantitatively assess what policy combinations could achieve the intended carbon emission reduction goal; country teams could then conduct further analyses (beyond those in our toolbox) to determine the optimal policy combination. Finally, our big data-based climate policy dashboard (Pillar IV) provides country teams with details about all the mitigation policies implemented since 1965, which are potentially useful references for the teams' specific climate policy designs.
Second, our tool could help functional departments (such as SPR and RES) conduct cross-country analysis and facilitate peer learning among country teams and the authorities. For example, users can conduct cross-country panel analysis using some outputs of our tool (e.g., the CPII index and the panel data of the Energy/GDP series for all the top 50 emitters since 1965). In addition, users can conduct event studies using our re-classified climate change mitigation policies and the identified structural breakpoints in carbon emissions. Such analyses could shed further light on (climate or broader) policies that are effective in mitigating emissions, facilitating peer learning among country teams and the authorities.
1. Finamore, Barbara. 2018. Will China Save the Planet? (Hoboken, NJ: John Wiley & Sons).
2. Hendry, David. 2020. "First in, First out: Econometric Modelling of UK Annual CO2 Emissions, 1860–2017," Oxford University Working Paper.
3. Kaya, Yoichi; Yokoburi, Keiichi. 1997. Environment, Energy, and Economy: Strategies for Sustainability (Tokyo: United Nations University Press).
4. McDonald, Peter. 2000. "Gender Equity in Theories of Fertility Transition," Population and Development Review, 26(3): pp. 427–439.
5. Wang, J.; Hu, M.; Tukker, A.; Rodrigues, J.F.D. 2019. "The Impact of Regional Convergence in Energy-Intensive Industries on China's CO2 Emissions and Emission Goals," Energy Economics, 80: pp. 512–523.
6. Yao, Jiaxiong, and Yunhui Zhao. 2022. "Structural Breaks in Carbon Emissions: A Machine Learning Analysis," IMF Working Paper (Link).
1. Stephan Danninger, Assistant Director and Chief of SPR Macro Policy Division.
2. Aquiles Farias, Section Chief of ITD Econometric and Modeling Support Section.
3. Mark Flanagan, Assistant Director of EUR and Mission Chief to the U.K. team.
4. Pritha Mitra, Deputy Division Chief of AFRC2, Mission Chief to Republic of Congo, and lead of the AFR climate working group.
5. Catherine Pattillo, Deputy Director of AFR.