Can the economy decouple GDP growth from carbon emissions? A standard way to measure this is by looking at aggregate emissions intensity – the amount of CO₂ emitted per unit of output. In countries like the UK, this has declined by a factor of two over the past three decades. What drives this decoupling? Some of it reflects a shift in domestic production, away from heavy industry and towards services, which tend to emit less per unit of value added. So, do aggregate emissions intensity decline because the composition of the economy change, or because each industry is reducing its emissions? Recent work (Agarwala and Martin, 2022) shows that most of the decline in aggregate emission intensity over the past few decades comes from within-sector changes: that is, individual industries becoming cleaner, rather than the economy simply shifting toward cleaner sectors. This composition effect, if the data is not granular enough to capture the shift away from coal within electricity generation, appears relatively minor compared to industry-specific technological progress.
This observation has shaped how climate policy is framed: there is a widespread and often implicit belief that each industry faces its own decarbonisation challenge. Electricity generation, construction, and steel, for example, have very different emissions profiles and technological paths. In response, climate policy frameworks have become increasingly granular, with net-zero commitments and transition plans now formulated not only at the national level, but also by sector and even by individual firms. This growing emphasis on sectoral targets raises a fundamental question: when we observe emissions reductions within a sector, to what extent do they reflect that sector’s specific choices and circumstances, and to what extent are they shaped by broader, economy-wide forces? After all, macro-level incentives, such as shared infrastructure, energy prices, regulatory pressure, or changing societal norms, can affect multiple industries at the same time, even if to varying degrees. Recognizing and quantifying this interdependence is crucial for both understanding emissions trajectories and designing fair and effective climate policy.
Do common shocks drive changes in aggregate emissions intensity? A bottom-up analysis
To answer this question, in the paper Ren, Marotta, Lafond (2025) we examine industry-level emissions intensity in the UK from 1990 to 2022, using emissions data constructed from activity-based environmental accounts. We find evidence of co-movement, particularly striking amongst service sectors. This steers us toward a block-level factor model that captures both aggregate and cluster-specific dynamics in emissions trends.
To explain such a pattern, we examine at how the data is constructed; emissions are first constructed at the level of emitting activities (e.g., driving a diesel car), and then allocated to industries based on their use of those activities. This process introduces mechanic co-movement – since nearly all industries rely on transport, so the emissions of all industries are correlated. This effect is particularly pronounced among industries with few direct emissions of their own, such as services by contrast to, say, steel. By contrast, blocks of industries, such as chemical-related industries, feature strong co-movement due their co-allocation of activities. To capture these shared structures, we use the bipartite network connecting industries to activities and use a network community detection algorithm to identify five blocks of industries that share similar activity profiles. These blocks form the basis for our block-level factor model, allowing us to separate global, block-level, and idiosyncratic variation in emissions intensity.
All told, we find that idiosyncratic shocks remain the dominant driver of changes in emissions intensity. However, both a global common factor and block-level factors explain a non-trivial share of the variance. The global factor exhibits significantly higher loadings in service industries, suggesting that these sectors are more responsive to broad, economy-wide influences, such as policy signals promoting decarbonization and shifts in the carbon intensity of universally utilized inputs like road transport. The block-level factors can explain a substantial amount of variance, particularly in the Agriculture, Energy, and Chemicals blocks, sectors where shared technologies or input dependencies are especially pronounced.
Sectoral versus aggregate shocks
Our approach is directly inspired by the work from Foerster et al (2011), who showed that in production network settings, a naively estimated factor model on industry growth rates often captures co-movement generated by input-output linkages, rather than a true common shock affecting all industries. With time series of emissions intensity, a different albeit similar logic applied – shared activities create “local” co-movement among related industries, so that a mis-specified factor model with only global co-movement may attribute local co-movement to a global factor. While we do not have the network data that would allow us to disentangle exactly co-movement due to shared inputs from the effects of a truly global latent dynamics, our block-level factor model provides a step in this direction.
Carbon pricing and the elasticity of CO2 emissions
Although many studies focus on aggregate drivers of decarbonization, there is limited quantitative evidence on how industry-level dynamics are driven by economy-wide mitigation efforts, and how these dynamics shape the aggregate outcome. A notable study is Rafaty et al (2025), who introduce interactive fixed effects (effectively, a factor model) as control in their study of the impact of carbon prices on emissions. While they do not provide additional discussions of the factors and loadings, their study makes it clear that careful econometric work should rule out common factors as explanatory variables, or risk an omitted variable bias.
Network-based allocation of responsibility for greenhouse gas emissions
Third, our study speaks to the questions raised by Van Den Ende et al. (2025) – in a system where all industries are interdependent, how can we attribute responsibility for emissions? Their setup considers input-output systems, and their answer follows from axiomatic principles. Our contribution is to note that sometimes, observed individual progress may be attributable to common trends, and do not necessarily reflect an individual industry’s idiosyncratic effort.