Renewable Energy Adoption Slowed Short-Run Growth Across 7 Countries. The Policy Answer Isn't to Stop.
This is an uncomfortable result to publish, so let me state the conclusion before the caveats, and then be very careful about what it does and does not license.
Using World Bank World Development Indicators across seven countries from 1990 to 2023 — 238 country-year observations — I found that increases in the renewable share of energy consumption are consistently associated with slower short-run GDP per capita growth. Under two-stage least squares, a one-unit rise in the renewable share is associated with a 1.44 percentage point reduction in GDP per capita growth, significant at p = 0.029.
The sign did not flip. It held across panel OLS, robust OLS after a White test flagged heteroskedasticity, maximum likelihood estimation, fixed effects with year dummies, Newey–West standard errors, and random effects after a Hausman test. I instrumented renewable consumption with its own lag and checked the instrument with Cragg–Donald and underidentification tests.
I ran that many specifications specifically because I did not like the answer. It survived all of them.
What the other coefficients say
Gross capital formation is positive and highly significant across every specification. Tertiary enrolment, trade openness and population growth are not significant anywhere.
That pattern is the interpretive key, and it is easy to miss if you only read the headline coefficient. The variable that reliably drives growth in this panel is capital formation. Energy transitions in their early phase are capital-absorbing: they pull investment into replacing generating capacity that already exists and already works, rather than into capacity expansion. You are spending capital to stand still, in output terms, while buying something the output statistics do not measure.
What this result does not mean
I want to be blunt here, because a coefficient like this is easy to weaponise.
This is a short-run growth result. It is not a finding about long-run welfare, about climate outcomes, about avoided damages, or about the cost of inaction — none of which appear in this model, and the last of which is precisely what makes decarbonisation worth doing. GDP per capita growth captures none of the damage function. A transition that costs measured output today while avoiding uncosted catastrophe later will look exactly like this in the data, and that is a limitation of the dependent variable, not an argument against the policy.
Seven countries over 34 years is also a modest panel. The instrument is a lag, which handles simultaneity but not omitted time-varying confounders.
So: anyone citing this as evidence to slow decarbonisation is misreading it.
What it does argue for
It argues for sequencing, which is a genuinely different claim from slowing down.
If the short-run drag runs through capital formation, then the policy response is to attack the drag rather than the transition:
- Efficiency first, because efficiency gains reduce required generating capacity and therefore reduce the capital bill before you spend it.
- Grid integration ahead of capacity, because renewable capacity stranded behind inadequate transmission is capital that produces nothing.
- Workforce training in parallel, since the transition reallocates labour and reallocation frictions show up as lost output.
- Capital-formation incentives, targeted at the one channel this panel shows reliably drives growth.
Do the transition. Sequence it so the capital-absorption phase is shorter and shallower.
On publishing results you dislike
I was writing this as an econometrics paper in the Department of Economics at The American University in Cairo, supervised by Dr. Mina Ayad. The honest version of the work was the one where the coefficient stayed negative through every robustness check I could construct, and the paper said so, and then did the harder work of asking what a negative coefficient actually implies.
The alternative — quietly selecting the specification that produced the sign I wanted — would have been easy and undetectable. It is also the reason a lot of published empirical work does not replicate.
The full paper, the Stata do-file and the raw dataset are all published at zyadhusseini.com so the result can be replicated or refuted directly.
I'm Zyad Husseini — data analyst and economist, MSc Data Analytics at Kedge Business School, BA Econometrics and Quantitative Economics from The American University in Cairo. LinkedIn · GitHub