Revisiting energy consumption and GDP causality: Importance of a priori hypothesis testing, disaggregated data, and heterogeneous panels

•Panel long-run Granger-causality performed on end-use disaggregated energy and GDP.•Panels vary by geography, income, and relative energy intensity.•Results of reduced-form supply and demand models are highly consistent across panels.•Income causes per capita residential electricity consumption in all panels.•Income causes per capita motor gasoline consumption in all panels.

This paper disaggregates energy consumption and GDP data according to end-use to analyze a broad number of developed and developing countries grouped in panels by similar characteristics. Panel long-run causality is assessed with a relatively under-utilized approach recommend by Canning and Pedroni (2008) [1]. We examine (i) reduced form production function models for both the industry and service/commercial sectors, where aggregate energy consumption is expected to cause aggregate output; and (ii) reduced form demand models, where income is expected to cause (separately) per capita residential electricity consumption and per capita gasoline consumption. We uncover for 12 different panels a set of super-consistent causality findings across two demand models that income “Granger-causes” per capita consumption. By contrast, the results from the production function models suggest that a different modeling framework is required to glean new, useful insights.