Indonesia’s growth-emissions link flips depending on how the model is built

Whether Indonesia’s economic growth travels with rising carbon emissions or appears to loosen its grip depends, strikingly, on a single modeling choice, according to a new open-access study published in Discover Energy. Researchers led by Muhammad Hasan Imaduddin, Soumya Basu, and Hideyuki Okumura built two versions of a vector error correction model, or VECM, using quarterly data for Indonesia from the first quarter of 2000 through the final quarter of 2022. One version included energy exports as its structural variable; the other included the Human Development Index. In the export-inclusive model, the GDP-related long-run term in the emissions equation came out positive, at 0.015. In the HDI-inclusive model, the same kind of term came out negative, at minus 0.019. Both terms were statistically different from zero within their own specifications, yet the study is emphatic that this sign flip does not prove Indonesia has achieved decoupling or that swapping one development strategy for another would causally cut emissions.

The authors frame their finding as evidence of specification sensitivity rather than a discovery about Indonesia’s actual development path. Because the two models were estimated separately and carry different cointegration ranks, one fewer for the HDI model, the study cannot establish that the two coefficients differ from each other statistically. The GDP-related term is a within-specification level-feedback parameter from a fitted multivariate system, not a conventional decoupling elasticity, not a normalized cointegrating coefficient, and not a direct change-ratio index. A negative sign is described only as directionally consistent with decoupling within the fitted specification. The authors also note the coefficients are very small in magnitude, and robustness checks soften the contrast further: under cubic spline interpolation the HDI-inclusive term becomes roughly neutral, and an annual-frequency sensitivity check attenuates the negative result, while the export-inclusive term stays weakly positive throughout.

The data assembly behind the comparison is itself a technical contribution. The team built an integrated quarterly dataset spanning macroeconomic, trade, human-development, energy, and emissions indicators for Indonesia over nearly a quarter century. Variables observed only annually, including electricity consumption, non-electric energy use, territorial emissions, energy exports, the Human Development Index, the Gini index, and the Palma ratio, were converted to quarterly frequency using the Chow-Lin method, with real GDP as the auxiliary indicator. GDP, capital formation, consumer prices, and trade openness were already quarterly. All variables were log-transformed except the dimensionless ratios and indices. Uncertainty from the interpolation was quantified with GLS-based 95 percent prediction intervals: for the core energy and emissions variables the average half-width was below about 2 percent of the mean, and around 1 percent for HDI, suggesting the disaggregation is unlikely to overturn the qualitative contrast between the two specifications.

A central design feature is the separation of electricity from non-electric energy use. Electricity consumption stands in for grid-based demand, while non-electric energy captures direct fossil-fuel combustion in transport, industry, households, and other final-use sectors. This split matters because the two channels behaved differently across the models. In the short-run block of the emissions equation, lagged electricity change entered negatively in both specifications, at minus 0.698 in the export model and minus 0.580 in the HDI model, while lagged non-electric energy change entered positively in both, at plus 0.311 and plus 0.201. Energy-export change was positive in the export model at plus 0.247, and HDI change was negative in the HDI model at minus 0.420. The authors stress these are conditional short-run relations within different fitted systems, not evidence that substituting HDI for export exposure would causally change emissions.

The methodological scaffolding is careful to a degree that shapes how much the headline contrast can carry. Unit-root testing with the Augmented Dickey-Fuller and KPSS tests produced mixed evidence, with consumer prices showing conflicting level results, so the cointegration rank is interpreted as the number of stationary directions in the system rather than a count of distinct economic equilibrium mechanisms. Lag order was set to one by the Akaike and Bayesian information criteria and the log-likelihood statistic for both retained models. Jarque-Bera tests rejected residual normality across specifications, a limitation the authors treat as constraining exact finite-sample inference, and they place greater weight on residual independence, which the Ljung-Box test supported for the two retained models but not for inequality-based variants using Gini or the Palma ratio. Those inequality models were therefore relegated to robustness status.

Error-correction dynamics add a second dimension to the comparison. In the export-inclusive model, one of three cointegrating relations showed statistically significant, restoring adjustment through the emissions equation after normalization, correcting roughly 72.7 percent of a deviation in the following quarter. In the HDI-inclusive model, one of two relations was restoring, correcting about 42.1 percent per quarter. The authors are explicit that faster adjustment does not mean better environmental performance; it means only that the emissions equation absorbs a larger share of the relevant disequilibrium in the next quarter. Nor do the tables support a claim of a larger causal policy impact, since the coefficients are within-specification level-feedback terms rather than directly comparable effect sizes.

Impulse response functions, reported as non-orthogonalized reduced-form responses without confidence bands, traced how model disturbances propagate over a 30-period horizon. In the export-inclusive system, a GDP disturbance initially produced a small negative emissions response before turning positive with damped oscillations. In the HDI-inclusive system, the response fell below zero almost immediately and stabilized at a persistent negative level. Energy-export disturbances were followed by persistent positive emissions responses, while HDI disturbances were followed by negative ones. Because several variables were temporally disaggregated, the short-horizon responses are read as model-based quarterly adjustment paths rather than directly observed within-quarter movements, and residual non-normality adds further caution.

Granger-predictive tests, interpreted strictly as predictive rather than causal, showed no short-run predictive content from GDP to emissions in either specification. In the export model, the strongest predictive links into emissions came from trade openness and electricity, with additional content from non-electric energy and energy exports. In the HDI model, predictive content was more strongly associated with HDI and non-electric energy, with trade and electricity weaker. Capital formation and consumer prices showed no meaningful short-run predictive content for emissions in either model. The authors suggest the growth linkage operates more through long-run equilibrium structure than through near-term quarterly predictability, and that the divergent predictive patterns help reconcile earlier Indonesian studies that reported either strong or weak GDP-emissions causality depending on their setup.

The policy discussion stays deliberately cautious. The authors argue the results are consistent with channeling capability gains, export revenues, and capital formation toward power-sector decarbonization, reduced direct fossil-fuel use, and cleaner productive investment, while accounting for Indonesia’s energy-sector governance, including the state-owned utility PLN’s single-buyer structure, electricity price regulation, and subsidy burdens that shape the electricity transition. They point out that Indonesia remains a net energy exporter, tying growth, trade, and emissions together through coal, gas, and oil export earnings, and that its power sector has historically remained strongly fossil based. Comparisons with developed economies that recorded sustained emissions declines, driven by reductions in the fossil share of energy and final energy use, and with ASEAN peers showing heterogeneous outcomes, frame Indonesia as undergoing an incomplete structural transition in which capability gains coexist with carbon lock-in.

The study’s limitations are enumerated rather than buried. The specifications include only a linear GDP term and no income turning point, so the conventional Environmental Kuznets Curve hypothesis is not formally tested. No conventional decoupling index or dedicated emissions-income elasticity is calculated. Short-run estimates involving annually observed variables are model-implied signals rather than precise intra-annual effects. The VECM is reduced-form and does not estimate mediation, a sequential development process, or a causal development trajectory, and the study does not formally test equality of the GDP coefficients across the two separate models. The findings, the authors conclude, should be understood as evidence on comparative conditional associations and adjustment patterns, with the sign and magnitude of the GDP-related terms sensitive to both specification and temporal treatment.

Subject of Research: Specification sensitivity of the long-run GDP-emissions relationship in Indonesia's energy-economy-environment system

Article Title: A comparative analysis of trade exposure and human capability in Indonesian growth and emissions dynamics

Article References: Imaduddin, M. H., Basu, S., & Okumura, H. (2026). A comparative analysis of trade exposure and human capability in Indonesian growth and emissions dynamics. Discover Energy, 6(1), Article 83. https://doi.org/10.1007/s43937-026-00208-6

Image Credits: AI Generated

DOI: 10.1007/s43937-026-00208-6

Keywords: Indonesia, VECM, carbon emissions, economic growth, energy exports, Human Development Index, electricity consumption, fossil fuels, decoupling, cointegration, Chow-Lin interpolation, 3E nexus