Despite digitalisation in the 21st century, productivity growth rates in advanced economies seemed to stagnate in the 2010s. One hypothesis to explain this phenomenon is the ‘productivity J-curve’, which suggests that the high costs associated with digitalisation during an investment boom induce an underestimation of productivity growth. This column tests the hypothesis using data from five advanced economies. It finds evidence that conventional total factor productivity growth rates in advanced economies were negatively biased in the 2010s, particularly in the US. These results have important implications regarding government investment support policies.
Digitalisation in the 21st century has revolutionised our lives and businesses. We expected this digital transformation to lead to further productivity growth and help us achieve a better life. Among these digital innovations, two stand out: the rise of platform businesses such as Airbnb and Uber in the early 2010s, and the development of generative AI technologies such as ChatGPT, Gemini, and more recently DeepSeek. However, productivity growth rates in advanced countries seemed to stagnate in the 2010s. These contradictory facts have led to controversial discussions suggesting that the productivity slowdown in the 2010s may be due to official economic statistics not capturing real business trends adequately.
An approach using the productivity J-curve
Brynjolfsson et al. (2021) suggest that the introduction of a new general-purpose technology (GPT) into the economy often triggers an initial phase of investment during which many associated intangible assets may remain unaccounted for, causing productivity to be measured inaccurately. For example, the provision of new digital services requires not only investment in hardware and software, which are usually recorded as fixed assets, but also significant costs for recruiting and training talented personnel and for developing research systems. However, instead of recording these costs as fixed assets, they are treated as current expenses. In this case – and because of this – these large costs reduce standard measured productivity during periods of aggressive investment in technological innovation. However, if we consider that human resources and the organisational structures that use them contribute to investment and provide services for an extended period, it is then more appropriate to treat them as capital assets rather than as current costs.
When total factor productivity (TFP) is measured in the standard way, the growth rate is low during periods of high levels of investment in digitalisation, because increasing adjustment costs associated with these investments decrease GDP. Therefore, once the adjustment costs are recognised as intangible investment, the revised GDP is higher. Since the standard TFP growth rate recovers after the investment boom, the movements over time in the gap between the standard TFP growth rate and the revised TFP growth rate resemble the letter J. Brynjolfsson et al. (2021) called this locus the ‘productivity J-curve’. The arguments on the productivity J-curve imply that the large associated costs arising from the rapid digitalisation during the investment boom induce an underestimation of the TFP growth rate.
Figure 1 Conceptual diagram of the productivity J-curve (the gap between the standard TFP growth rate and the revised TFP growth rate)
Productivity J-curves in advanced countries
Following Brynjolfsson et al. (2021), in our recent study (Bounfour et al. 2024), we measure the productivity J-curve not only in the US but also in France, Germany, the UK, and Japan using data from the Orbis dataset (which covers listed firms in advanced countries), and industry-level databases such as the EUKLEMS/INTAN Prod data (the version released in 2023), and the Japanese Industrial Productivity (JIP) database (the version released in 2023).
First, we estimate associated costs of three types of intangible capital (R&D, software, and organisational capital) along with ordinary tangible capital assets (buildings and structures and machinery). As Brynjolfsson et al. (2021) demonstrated, we find large associated costs of investment in all assets in the US case. In particular, the associated costs of R&D and software investment are higher in the US and lower in the other advanced countries.
Using these empirical results, we construct measures of revised TFP using the estimated associated costs of investment in all three assets for the five countries. Figure 2 shows the gap between standard and revised TFP growth for the countries in our sample, revealing several potential productivity J-curves. This shows that the conventional TFP growth rates in advanced countries are negatively biased in the 2010s, indicating that the conventional TFP growth rates may have been underestimated. However, we find that the underestimation of the TFP growth rate is large in the US due to the larger investment and its associated costs in intangibles, while it is comparably small in the other countries. This implies that the productivity gap between the US and other advanced countries when corrected for the adjustment costs of investment is larger than that measured by the standard statistics.
Figure 2 International comparison of productivity J-curves aggregated for all assets in the five developed countries
Notes: The vertical axis represents the TFP mismeasurement (standard-revised TFP).
When we revise TFP to account for software only, we find that correlated intangible investments in the US are very large and the large investment in digitalisation may have led to an underestimation of the traditional TFP growth rate. On the other hand, in the case of France and Japan, no productivity J-curve is observed, as these countries have not been as active as the US in terms of digitalisation, further emphasising the unique behaviour of the US.
Policy implications
These results have important implications regarding government investment support policies. First, significant ancillary costs accompany capital investment involving advanced technological development, including the training and acquiring of human resources, especially in the US, where digital firms make significant investments in talented human resources. In the US, these enormous additional costs of investment in digitalisation are financed through large capital markets. Therefore, to support digitalisation, governments in other advanced countries have two possible policy tools: they can develop capital markets where advanced technology firms can obtain financing for large digital investments, and/or they can extend the range of assets to be supported (and eventually subsidised) to several types of intangibles such as software and training of specialised human resources.
Second, an investment boom may be accompanied by a short-term productivity slowdown. Governments should take this into account when assessing current and future productivity trends and be wary of overcompensating with excessive support policies based on an apparent productivity slowdown.
Source : VOXeu
