Coal consumption is a major source of carbon emissions, which massively contribute to climate change. Therefore, investigating possibilities to substitute for cleaner energy sources is of great importance in estimating the costs of mitigating the unwanted effects of coal consumption.

In economics, the measure used to capture substitution possibilities is called elasticity. Specifically, my focus lies on the estimates of fuel elasticities, which are the key parameters of many environmental models assessing the impact of various policies trying to tackle climate change. The goal is to present a quantitative systematic review of the literature dealing with the topic of coal substitution. Based on 893 collected estimates of the own-price elasticity of demand for coal and pairwise cross-price elasticities of substitution between coal and electricity, gas, and oil, I investigate publication bias in the literature and the drivers of heterogeneity in the reported estimates. Publication bias is present if researchers systematically underreport estimates with certain characteristics. In such cases, the existing evidence becomes distorted and exaggerated. Using several rigorous methods, I find that the own-price elasticity for coal and the cross-price elasticity between other fuels and coal is significantly affected by publication bias.

On the contrary, I find that the cross-price elasticity between coal and other fuels is not likely to suffer from substantial publication bias. Based on this, I demonstrate that the substitution of other fuels for coal is feasible, although limited. Apart from publication bias, I examine other sources of potential heterogeneity in the estimates. Using up-to-date techniques and robustness checks, I find that the estimates are systematically impacted by the underlying study settings and estimation techniques used by the researchers. However, I do not find any systematic relationship between the estimates of the elasticity and the level of development of the country. Similarly, the estimates are not significantly different at the state, industrial and sub-industrial levels. This suggests that the substitution possibilities are not greater for industries or sub-industries, compared to the state-level substitution possibilities. Nonetheless, a few possible drawbacks of this analysis should be kept in mind. The tests used to detect publication bias might suffer from endogeneity, even though I demonstrate my attempt to control for it. Moreover, the results are clearly conditional on the quality of reviewed studies. If the studies share common misspecification, I am unable to control for it and the analysis could be biased. Thus, my judgment is purely based on the reported empirical research.