Energy consumption remains Algeria’s most stable predictor of carbon emissions, study finds

Primary energy consumption is the most persistent long-run correlate of Algeria’s carbon dioxide emissions, according to a new econometric study of annual data from 1990 to 2024, while the independent roles of income and real oil prices prove far more fragile across statistical methods. The research, published in Discover Environment, applies an autoregressive distributed lag bounds-testing framework to 35 annual observations and subjects its central result to an unusually demanding battery of robustness checks, including scale normalization, alternative lag-selection criteria, structural-stability diagnostics, a nonlinear oil-price decomposition, and an alternative long-run estimator.

The baseline model, selected by the Schwarz information criterion with a maximum of one lag, relates total carbon dioxide emissions to real GDP per capita, the real WTI crude-oil price, and primary energy consumption. The bounds test yields an F-statistic of 5.5611, which exceeds the finite-sample upper bounds at the 10% and 5% levels and the asymptotic 5% upper bound, though it remains below the finite-sample upper bound at 1%. The evidence therefore supports a long-run level relationship at conventional significance levels without implying rejection at the most stringent threshold. The error-correction coefficient is negative and highly significant, indicating that roughly 47% of the previous year’s disequilibrium is corrected within one year, with most of a disturbance absorbed over about two years.

Within that baseline, the long-run energy-consumption elasticity is 0.7977: holding the other regressors constant, a 1% increase in primary energy consumption is associated with an approximately 0.80% increase in total carbon dioxide emissions. The authors describe this as the strongest coefficient in the model, both statistically and economically significant. By contrast, the coefficient on real GDP per capita is positive but statistically insignificant, and the real oil-price coefficient is close to zero and also insignificant. The study’s authors stress that these estimates are interpreted as long-run associations rather than structural causal effects, and that the independent effects of income and oil prices are less precisely identified once energy consumption is included.

The study’s central contribution lies in testing whether that energy coefficient survives progressively more demanding assumptions. When the maximum lag length is raised to two and the Akaike information criterion is compared with the Schwarz criterion, the long-run energy elasticity remains within a narrow range of 0.798 to 0.827 and is highly significant in every specification. A dummy variable representing the exceptional disruption of 2020 is statistically insignificant under both information criteria, suggesting the pandemic disturbance does not generate a separate level effect once the selected dynamics are taken into account. The GDP coefficient, in contrast, becomes significant under the Akaike specifications but remains weaker under the Schwarz criterion, while the oil-price coefficient changes sign across models.

Structural stability is examined with procedures that allow the break date to emerge from the data rather than being imposed. An unknown-date all-coefficient test reaches its maximum in 2012, but the bootstrap probability of 0.876, based on 4,999 recursive residual-bootstrap replications, is far above conventional significance levels. A separate intercept-shift test also identifies 2012 as the largest candidate split, yet its search-adjusted probability remains insignificant. A leave-one-year-out exercise re-estimates the fixed baseline equation 34 times, and the long-run energy elasticity ranges only from 0.764 to 0.830, remaining highly significant in every re-estimation with the largest probability value below 0.001. No single annual observation accounts for the sign or significance of the central coefficient.

The most conceptually important check addresses aggregate scale. Because total emissions and total energy consumption both reflect the size of the population and the economy, their aggregate relationship could partly be an artifact of common scale. The study therefore re-estimates the model with emissions and energy consumption expressed per capita and relative to real output. All four normalized equations retain a positive and highly significant energy coefficient, but the elasticity declines from approximately 0.80 in the aggregate model to around 0.68 to 0.71 in the per-capita and intensity specifications. This reduction indicates that part of the aggregate relationship reflects common population or output scale, yet its persistence after both normalizations shows that scale alone does not explain the association. The income coefficient remains sensitive to the specification across these models, and the real oil-price coefficient remains insignificant throughout.

Oil-price asymmetry is tested within a nonlinear ARDL framework that separates positive and negative changes in the real oil price into distinct cumulative components. The energy-consumption coefficient remains positive and highly significant in all four nonlinear equations, ranging from 0.718 to 0.881. The oil-price results are far less stable: in the Schwarz specifications both components are small and insignificant, and under the Akaike criterion only the positive component in the per-capita equation reaches the 5% level, an isolated result not reproduced in the total-emissions model. Crucially, none of the Wald tests rejects equality between the positive and negative effects, with long-run symmetry probabilities ranging from 0.249 to 0.863 and short-run probabilities from 0.364 to 0.877. The data therefore do not provide robust evidence of asymmetric oil-price transmission, despite the theoretical possibility that rising and falling prices operate through different fiscal and demand channels in a hydrocarbon-exporting economy.

Dynamic ordinary least squares, estimated with one lead and one lag of first differences, provides an estimator-based check on the cointegrating relationship. The energy coefficient remains positive and highly significant for all three outcomes, with an aggregate elasticity of 0.801 and normalized elasticities of 0.703, values close to their ARDL counterparts. The income coefficient is positive and significant for total emissions but negative and insignificant in the intensity equation. Notably, the real oil-price coefficient becomes negative and statistically significant in all three DOLS equations, contrasting with its insignificance in the ARDL and NARDL models. The authors interpret this as estimator sensitivity rather than a uniformly stable negative effect. Toda-Yamamoto tests, used only as diagnostics of predictive direction, provide no significant direction in the selected one-lag systems, and some two-lag results suggest emissions predict energy consumption, with the reverse direction at most marginal. The authors caution that these tests do not establish a stable causal ordering and do not resolve simultaneity or omitted-variable concerns.

Supplementary specifications examine energy composition. The natural-gas share in combined oil and gas consumption carries a negative coefficient in all four equations, consistent with lower direct carbon emissions from natural gas relative to oil, but it becomes statistically significant only when the maximum lag length is increased to two, limiting the strength of that conclusion. A renewable-energy share added to the per-capita model is negative but statistically insignificant, with a coefficient of -0.029 and a probability value of 0.774 under the Schwarz criterion. The authors attribute this weak identification to the very small and slowly varying renewable share in Algeria’s primary energy mix over the sample, not to evidence that renewable energy lacks environmental benefit. They also note that fossil-fuel energy consumption, with a correlation of 0.999997 with primary energy consumption, is retained only as a consistency check.

The authors frame their policy implications cautiously, as directions consistent with robust long-run associations rather than causal effects estimated by the model. Because the energy relationship persists after normalization by population and output, they argue that Algeria’s emissions challenge cannot be addressed through income growth or favorable oil-price movements alone, pointing instead toward energy efficiency, gradual energy-pricing reform, renewable-electricity deployment, grid investment, and the productive use of hydrocarbon revenues. They also emphasize that the study does not test the Environmental Kuznets Curve, since no squared income term or turning point is estimated, and that the 35-observation sample limits the power of break tests and prevents the inclusion of many determinants suggested in the wider literature. The defensible conclusion, they write, is that the direct oil-price effect is weak and estimator-dependent, while the persistent link between domestic fossil-intensive energy consumption and emissions is the finding that survives every check the study could devise.

Subject of Research: Long-run associations between primary energy consumption, income, real oil prices, and carbon dioxide emissions in Algeria

Article Title: Energy consumption and carbon dioxide emissions in Algeria with scale normalization and real oil price asymmetry tests

Article References: Azzeddine, N., & Rebiai, B. (2026). Energy consumption and carbon dioxide emissions in Algeria with scale normalization and real oil price asymmetry tests. Discover Environment, 4(1), Article 600. https://doi.org/10.1007/s44274-026-01138-9

Image Credits: AI Generated

DOI: 10.1007/s44274-026-01138-9

Keywords: Algeria, carbon dioxide emissions, primary energy consumption, ARDL bounds testing, NARDL, oil prices, cointegration, energy intensity, Environmental Kuznets Curve, econometrics, energy policy, hydrocarbon economy