Mind the gap: accounting for emissions in productivity measurement

Productivity is a key driver of economic growth and competitiveness. As such, internationally comparable indicators of productivity are central for assessing economic performance. One widely used productivity measure is multifactor productivity (MFP), which traditionally captures the efficiency with which an economy converts labour and produced capital, such as machines and buildings, into output. However, conventional MFP metrics disregard negative externalities of production, including greenhouse gas and other pollutant emissions, and neglect the use of natural capital, such as minerals.

When sustainability is omitted from productivity metrics, we risk treating pollution-heavy operations in the same way as environmentally responsible ones. Imagine two clothing factories operated by the same global brand, each producing an identical number of garments with the same number of workers and similar levels of investment in produced capital, such as machinery and equipment. One factory employs low-emission technologies and recycles water, while the other uses energy-inefficient machines and discharges toxic chemicals into local rivers. Traditional productivity metrics would classify both factories as equally productive, despite their vastly different environmental footprints. This gap in measurement paints an incomplete picture, making it harder for policymakers to align economic growth strategies with sustainability goals.

There is growing recognition that traditional MFP comes with an environmental blind spot. In response, efforts have emerged to incorporate emissions and natural capital into productivity measurement, leading to the development of a new metric known as environmentally adjusted MFP (EA-MFP). Figure 1 illustrates the augmented growth accounting framework used to derive EA-MFP, which explicitly incorporates two additional dimensions. First, it recognises emissions as undesirable byproducts generated alongside desirable outputs. Second, it extends the set of input factors to include natural capital, complementing the traditional inputs of labour and produced capital. EA-MFP is then defined as the residual growth in pollution-adjusted output, calculated as the difference between the growth of desirable and undesirable output, after accounting for the contributions of labour, produced capital, and natural capital.

Putting this framework into practice requires data on the two additions to traditional MFP measurement. While natural capital is increasingly reflected in official statics, emissions remain difficult to price – primarily because there is often no market to guide valuations.

To address this, researchers have proposed several methods to incorporate emissions. These approaches typically involve converting emissions into monetary costs, which can then be integrated into MFP calculations. However, in the absence of explicit emission markets, these monetary valuations, known as “shadow prices”, cannot be directly observed and must instead be estimated. Such estimates often carry significant uncertainty and can vary widely depending on the methodology used (Figure 2).

The literature generally relies on one of two main approaches: regression methods and efficient frontier methods. Both focus on the producer’s perspective, aiming to capture the implied cost that firms incur when reducing pollution by one additional unit. Regression methods are based on production functions and profit maximisation. They estimate the relationship between desirable and undesirable outputs and factor inputs, allowing for the derivation of pollution costs within a standard economic framework. Efficient frontier methods compare firms or economies to the best performers that produce the most with the least pollution, helping to assess how far others fall short. Two such methods are Data Envelopment Analysis (DEA) and Stochastic Frontier Analysis (SFA), which differ in how they model production efficiency in the presence of negative externalities.

DEA compares how efficiently firms or countries convert inputs into outputs relative to a reference frontier derived from observed data, without requiring assumptions about the shape of the production function. However, it attributes all deviations from the frontier to inefficiency, overlooking the role of noise or measurement errors. SFA, on the other hand, evaluates production efficiency based on a pre-specified production function. This method requires assumptions about the structure of the production technology but has the advantage of distinguishing inefficiency from random shocks or data imperfections. 

Many studies offer limited insight into why a particular method was chosen. Often, a method is selected based on a single consideration, without weighing the full range of advantages and drawbacks. For example, DEA is frequently preferred because it avoids the need to specify a functional form for production, while SFA and regression-based methods are typically chosen for their capacity to support hypothesis testing. This narrow basis for selection limits the ability to evaluate which tools are best suited to meet specific research objectives.

In this blog article, we identify five criteria to evaluate possible approaches to produce cross-country estimates that can inform policymaking:

• Compatibility: Is the method compatible with the OECD’s growth accounting framework for MFP estimation?
• Robustness: How sensitive are the results to small changes in the data?
• Replicability: Is the method transparent and reproducible?
• Data availability: Are the required data available across OECD countries and key emerging economies?
• Communicability: Are the results easy to convey to policymakers and other stakeholders?

Based on this framework, regression analysis offers several advantages. It is relatively easy to communicate, transparent in its assumptions, and highly replicable. However, it struggles with separating cause from effect (endogeneity problems), which can compromise the accuracy of shadow price estimates.

DEA often leads to implausible shadow prices and does not support hypothesis testing, raising concerns about the reliability and uncertainty of its findings.

On the contrary, SFA evaluates efficiency by specifying a production frontier, which requires assumptions about its functional form and statistical properties. However, certain features, such as translation properties, are difficult to fully account in the estimation process within a traditional SFA framework.  As a result, it is often necessary to verify ex post whether the estimated results satisfy these constraints.

Combining SFA with Bayesian estimation – an approach still novel in this field – offers a promising way to address these challenges. This hybrid method has the potential to produce robust and consistent shadow price estimates, to mitigate endogeneity problems faced by regression-based approaches, and to incorporate measurement error and random noise, thereby enabling hypothesis testing.

Despite its promise, Bayesian SFA has not yet been extensively tested across varied national and sectoral contexts. Compared to the classical SFA, its technical complexity poses challenges for both implementation and communication. For example, Bayesian SFA requires to specify prior assumptions and use computational methods to estimate results. These estimates could be sensitive, especially when working with small samples or limited data. Moreover, the method relies on Markov Chain Monte Carlo algorithms, which need careful tuning and checking to ensure they produce stable results. Looking ahead, testing it in diverse contexts will be essential to evaluate its real-world strengths and limitations.

1. The views expressed herein are those of the author and do not necessarily reflect those of the Joint Research Centre or the European Commission.