An analysis of how carbon regulation affects Global South countries, the role of energy and infrastructure in decarbonization, and opportunities to use digital models to strengthen economic sovereignty.
11 min read
Share:
Behind the widely accepted concept of energy transition that emerged after the Paris Agreement lies the influence of Western capital. Instruments such as cross-border carbon adjustment mechanisms (CBAM) and unified global carbon intensity metrics are typically presented as neutral environmental policy measures. In practice, however, they can function as a hidden form of protectionism affecting the interests of Global South countries. Developing nations, Indonesia in particular, are effectively being turned into suppliers of carbon credits, while control over analytical methodologies and global supply chains remains concentrated in countries with high capital concentration.
Reducing the carbon intensity of economies has become one of the central tasks of climate policy. For developing countries, this metric is especially important since they must simultaneously develop industry, ensure energy security, and reduce emissions. The ability to quickly and accurately forecast carbon intensity therefore becomes essential for decision-making. It allows assessment of how effective measures are for transitioning to a sustainable low-carbon economy amid global changes.
To fully assess carbon intensity dynamics requires analyzing data across many countries over extended periods. A forecasting system must work with large international datasets and account for complex nonlinear relationships between economic and environmental indicators. This is particularly important as climate policy becomes increasingly complex, incorporating carbon pricing, tax measures, and various approaches to improving energy efficiency.
Traditional economic analysis methods have serious limitations. For example, vector autoregressive models and fixed-effects models primarily assume linear relationships. They therefore poorly account for complex connections between macroeconomic indicators. Attempts to manually specify nonlinear dependencies also yield limited results: they depend heavily on the chosen model form and perform worse when comparing countries that differ significantly from one another.
Behind the rhetoric about energy transition, according to the study's authors, Western capital's influence is once again visible. One of its instruments is protectionism in the form of cross-border carbon regulation. Environmental crises are increasingly being turned into objects of financial mechanisms, while Global South countries risk remaining in the role of quota suppliers. The technology gap and dependence on external scientific approaches simultaneously make developing countries consumers of analytical standards from the Global North. This limits their ability to independently manage data and determine the direction of national industrial development.
The global climate paradox is examined using a hybrid bidirectional LSTM neural network (HAL), whose parameters are tuned using Bayesian optimization.
Permutation Feature Importance analysis shows that physical efficiency indicators and energy characteristics—particularly the share of renewable energy sources and per capita electricity consumption—have the greatest significance for forecasts. This points to the importance of actual energy infrastructure for decarbonization compared to short-term pricing measures. Direct carbon pricing and strict regulation make small contributions to forecasts: their indicators do not reach the statistically significant thresholds built into the model. Such instruments are more likely to replenish local budgets than become independent drivers of global change.
At the same time, GDP per capita remains one of the key factors. This confirms that reducing carbon intensity is largely connected to the economy's transition toward knowledge and services, not just early introduction of carbon taxes.
This approach makes it possible to show how "green" metrics can be transformed into instruments of global economic control, and to outline for Global South countries a possible path toward greater autonomy in managing their economies, data, and industrial development.
The trap of Western metrics and Indonesia's strategic interests
For Indonesia and other developing countries, these findings have great significance for developing national strategy. Introducing strict carbon emission taxes without developing energy systems and modernizing electrical grids could weaken industrial competitiveness. When international emissions requirements are imposed without considering local infrastructure capabilities, developing countries find themselves trapped by global protectionism that limits their opportunities for full economic development.
To escape this trap, Global South countries need to strengthen economic sovereignty. According to the authors, this requires developing "green" markets based on data analysis and using algorithms for management decisions. Such sovereignty in the future will depend not on mechanically following Western standards, but on combining reliable physical infrastructure, proprietary carbon emissions forecasting systems, and consistent national industrial policy amid growing geopolitical fragmentation.
Real Decarbonization and Structural Realities
The study proposes a Hybrid Bidirectional LSTM (HAL) method for assessing the nonlinear dynamics of carbon intensity based on data from 191 countries over the period 1990–2019. The authors combined Bayesian optimization with time-series cross-validation, which prevents future information from leaking into the calculations. This approach overcomes the limitations of traditional linear models and earlier neural networks. The results show that processing data in both directions is essential for capturing complex temporal relationships and cross-country differences.
The analysis reveals that energy efficiency and energy transition indicators account for 58.1% of the model's predictive power. Another 37.5% comes from macroeconomic factors, while carbon pricing-related indicators contribute only 4.4%. The foundation of decarbonization remains energy efficiency and technological readiness, rather than individual emissions pricing mechanisms. This approach enables data-driven assessment of climate policy and corporate green product strategies in emerging markets, including Indonesia, taking into account long-term economic shifts.
Research on carbon intensity is developing along several tracks. One involves identifying factors that influence emission levels, including the Environmental Kuznets Curve (EKC) hypothesis. Using semiparametric analysis of cross-country data, Verbič and other researchers found an inverted U-shaped relationship at the global and regional levels. However, no such relationship was found for consumption-based carbon footprints in OECD countries due to emissions offshoring. That said, their study does not focus on forecasting carbon intensity and does not employ recurrent neural networks to analyze its dynamics over time.
Despite the rapid growth of research using deep learning methods, most studies focus on individual countries or economic sectors, such as the work by Acheampong, Boateng, Fan, and other researchers. The authors of the study under review sought to expand this approach. For model hyperparameter tuning, which is typically done using Bayesian optimization—as in the work by Habtemariam and colleagues—they combined a Bidirectional LSTM network (BiLSTM) with Bayesian optimization and applied it to data from 191 countries.
Core economic and energy indicators are drawn from the World Bank's WDI database, while data on carbon pricing measures come from the emissions-weighted average carbon price indicator published on Our World in Data. Carbon intensity dynamics were calculated based on the World Bank's greenhouse gas emissions data. As a result, the researchers obtained four main groups of indicators for analysis. However, data for several parameters, including fiscal indicators, real interest rates, and carbon prices, are incomplete. This reflects challenges in data collection and reporting in developing countries.
The global dataset covers 191 countries and consolidates indicators across four areas: key outcomes, government policy measures, energy, and macroeconomics.
Macroeconomic indicators and the level of economic development play the primary role in structural economic changes, consistent with the Kuznets model. Energy efficiency and infrastructure conditions follow. Direct government policy measures, meanwhile, have considerably less influence.
Global carbon intensity dynamics depend primarily on the overall state of the economy, its level of development, and energy efficiency. Analysis of changes across countries over time shows that the HAL model tracks the decline in average global carbon intensity quite accurately and produces a small relative error. Deviations for individual countries collectively offset each other and do not lead to systematic calculation errors.
To assess the results and theoretical significance of the bidirectional LSTM neural network (HAL) from the perspective of carbon imperialism, the authors examine them across five dimensions: advantages of the model architecture, the role of energy and infrastructure, constraints related to development levels, weaknesses of Western carbon pricing mechanisms, and model parameter tuning through Bayesian optimization. The authors conclude by separately examining the method's own limitations.
The BiLSTM Advantage: Bidirectional Temporal Dynamics
The study's main finding is that bidirectional data processing markedly improves HAL's forecast accuracy. Testing individual model components showed that removing BiLSTM substantially reduces prediction accuracy on new data. The losses prove greater than all other architectural modifications combined, and roughly comparable to the difference between a standard unidirectional LSTM and an OLS model.
For forecasting global carbon intensity, this architecture accounts for varying energy transition dynamics. Moving forward, the model analyzes initial conditions, while moving backward it incorporates subsequent changes—such as rapid clean energy adoption or sharp economic deterioration. Through this bidirectional analysis, HAL better captures complex temporal changes, especially for countries with vastly different economies.
Energy Efficiency as the Primary Driver of Carbon Intensity
Permutation Feature Importance (PFI) analysis shows that three energy transition indicators—renewable energy share, energy intensity, and per capita electricity consumption—together determine most of the HAL model's predictive power. This confirms that energy efficiency and power generation technologies play a key role in carbon intensity differences between countries. Emissions reductions through cleaner energy transitions, improved consumption efficiency, and grid development depend on each country's specific conditions, yet can be captured by the model without explicitly specifying complex indicator relationships.
Empirical Evidence on the EKC: The Role of Per Capita GDP and Structural Constraints
Per capita GDP emerges as the single most important factor in this model. Its impact on carbon intensity is significantly stronger than that of indicators directly linked to government policy. This provides substantial support for the Environmental Kuznets Curve (EKC) concept, while simultaneously revealing its limitations. The identified relationship does not arise spontaneously through market forces, but under specific economic conditions. High-income countries that remain dependent on natural resources and achieved prosperity without building sustainable infrastructure maintain high carbon intensity levels.
The Carbon Imperialism Matrix: Limitations of Western Carbon Pricing Systems
Direct market regulation measures—primarily carbon pricing and stringent government requirements—have very limited predictive significance in this model. This is indicated by the limited global coverage of such measures, carbon prices that remain well below calculated thresholds, and these instruments' modest contribution to government revenues. The results show that pricing mechanisms alone are insufficient for global structural transformation. In developing countries, such measures work more effectively when supported by developed infrastructure, technological capabilities, and other conditions that enable the transition to a cleaner economy.
Digital Sovereignty, Algorithmic Governance, and Economic Resilience
Beyond traditional tax and fiscal measures, carbon intensity forecasting can give Global South countries additional tools to preserve economic autonomy amid shocks in energy and trade markets. For resource-dependent developing nations, combining real-time carbon intensity forecasting with digital platform-based governance enables data-driven decision-making and strengthens economic resilience.
By linking digital governance systems with national carbon accounting, countries can better shield local businesses—from microenterprises to small and medium-sized companies—from sharp shifts in external markets and stringent international requirements. Such measures include, for example, the EU's Carbon Border Adjustment Mechanism (CBAM).
Model Parameter Tuning and Overfitting Protection
Following Bayesian optimization, a compact network architecture was selected: one recurrent layer with an optimal number of neurons in each direction, moderate learning rate, L2 regularization, and high dropout levels. The network's limited complexity prevents it from simply memorizing training data. Instead, it identifies general patterns and therefore performs better on new data, including when there are significant differences between countries.
Methodological Boundaries and Limitations
These results have four main methodological limitations. First, the data ends in 2020, meaning it doesn't capture the acceleration of climate policy in the subsequent post-pandemic period. Second, some missing public finance data was imputed, which may have smoothed out certain differences. Third, the pooled time series analysis doesn't explicitly account for country-level differences. Finally, the model assumes countries don't directly influence each other, so it doesn't capture cross-border trade and technology transfer.
Evaluating the Hybrid Bidirectional LSTM (HAL) network on cross-country data yields four key findings. Bidirectional analysis better captures changes over time, energy efficiency emerges as the primary practical driver of carbon intensity reduction, economic scale forms the basis for the Environmental Kuznets Curve (EKC) relationship, and carbon pricing remains a supplementary tool limited by adoption in few countries and low price levels.
From a practical policy standpoint, this points to the need to strengthen economic autonomy. Governments should combine emissions pricing with investments in clean energy and infrastructure, while exporters should calculate their products' carbon footprint in advance to better manage external requirements. Embedding such forecasts into digital platforms allows computational resources to be allocated based on expected emissions while simultaneously reducing domestic digital infrastructure's vulnerability to external trade and energy shocks.
Under pressure on global trade, companies find it harder to rely on general environmental responsibility statements: this can lead to accusations of "greenwashing." Within the framework the authors describe, an accurate forecasting model allows companies to assess product carbon footprints accounting for changes in the energy sector. For exporters from developing countries, this provides a foundation for preparing for border restrictions, possible additional tariffs, and the risk of exclusion from international supply chains.
Indonesia, as a developing country heavily dependent on foreign trade, vividly illustrates the constraints described by the HAL model. Climate policy shouldn't constrain industrial growth through premature carbon taxes. Instead, the country should prioritize capital-intensive clean energy projects, including power grid modernization. Combining real-time carbon intensity forecasting with digital governance, the authors argue, will allow developing economies to enhance resilience, maintain control over their own data, and better adapt to the conditions of the global "green" market.
