Research

This study examines the impact of housing infrastructure under the Reconstruction and Development Programme (RDP) on multidimensional energy poverty (MEP) in South Africa from 2009 to 2018. We use a doubly robust estimator with entropy balancing technique that relies on exact matching to construct an appropriate counterfactual and address endogeneity. The results showed that RDP beneficiaries experience 11.7% improvement in MEP compared to the baseline MEP among similar non-beneficiaries. The heterogeneous analysis revealed that the results are driven by the Black/African race. Also, we found that as the houses get older, the impacts on MEP decline. Our results are robust to alternative specifications and estimation techniques. We therefore suggest that while RDP housing is a highly effective structural measure, further targeted energy efficiency measures within the subsidized housing stock could still realize greater gains.

Informal enterprises employ most of the workforce in developing countries and face the least reliable electricity, yet how they absorb disruption is largely unknown. We provide within-firm and between-firm evidence from 12,283 informal businesses in twelve countries, exploiting a survey design in which each proprietor reports outcomes both for the month during the outage and for a typical month a year earlier. We find that, when the power fails, the consequences operate within-firms. Sales fall by about 12.4%, while labour does not meaningfully respond, leading to a fall in labour productivity (sales per labour) almost as large as the fall in sales (11.7%). We explore various mechanisms and find that the loss is concentrated among grid-connected businesses and absent among those off the grid. Rather than shedding labour or reducing hours, firms absorb the shortfall through working capital. They rely on supplier credit and run down inventory, while formal credit access does not respond. The sales result survives every alternative specification, sample, and inference check we apply. Conventional cross-sectional comparisons between firms reverse the employment finding, which may explain why this effect has gone undetected in this sector. Therefore, the cost of unreliable power is substantial, but it falls on output, proprietors' incomes, and trading partners, rather than on jobs.

Energy poverty remains widespread in Ghana, yet conventional regression studies identify numerous significant determinants while offering limited evidence on which household characteristics matter most for prediction and targeting. This study addresses this gap by identifying the key predictors of multidimensional energy poverty using machine-learning methods and data from 37,219 households across three rounds of the Ghana Living Standards Survey (GLSS 5, 6, 7). Using four tree-based ensemble models namely – Random Forest, Gradient Boosting (GBM), Extreme Gradient Boosting (XGBOOST) and Categorical Boosting (CatBoost) we evaluate the key determinants of household energy poverty. The results indicate that of the four models, CatBoost performs best, explaining about 71% of the variation in household energy-deprivation scores. Of 26 candidate predictors, predictive importance is concentrated in six: energy expenditure, welfare expenditure, housing quality, education of the household head, rural–urban residence and survey round, while demographic characteristics contribute relatively little. The findings also reveal distinct rural–urban patterns in the structure of energy poverty. Energy expenditure emerges as the strongest predictor, underscoring the distinction between access to modern energy and the ability to use it. These findings support a shift from access-centred policies towards affordability and sustained energy use, alongside context-specific rural and urban interventions and more data-driven targeting of energy-poor households.