P. Boeing
ZEW – Leibniz Centre for European Economic Research, Germany
P. Hünermund
Copenhagen Business School, Denmark
Artificial Intelligence in Science
Challenges, Opportunities and the Future of Research
This essay provides evidence for a decrease in research productivity in the last decades for the People’s Republic of China (hereafter “China”) and Germany. Estimates imply that research productivity falls, on average, by 5.2% per year in Germany and by 23.8% per year in China, which corresponds to a reduction by half in 13 years and in 3 years, respectively. Results indicate that policy measures to increase the productivity of research and development (R&D) are important for curbing the ongoing global productivity slowdown.
In contrast to assumptions invoked by endogenous growth theory (Romer, 1990), the productivity of R&D might not be constant. Breakthrough innovations may be getting more difficult to achieve over time. This would result in higher levels of R&D spending being needed to maintain constant rates of economic growth.
Figure 1 plots aggregate R&D spending and gross domestic product (GDP) growth rates in China and Germany for the last three decades. It shows that while aggregate spending on R&D was steadily increasing, growth rates of GDP remained constant or even decreased. Similar trends can be observed for the United States (Bloom et al., 2020).
This finding is inconsistent with assumptions invoked in standard endogenous growth models, which usually posit a one-to-one link between R&D spending and growth rates. Several economists have proposed that a decline in R&D productivity over time could explain this decoupling.
Cowen (2011) argues the US economy has benefited from low-hanging fruit in science and technology for the last two centuries; this started to run out in the second half of the 20th century. Gordon (2016) makes a similar argument, stating that new technologies such as electrification, indoor plumbing, home appliances and the rise of motor vehicles created exceptional drivers of economic growth unlikely to be replicable in the coming decades. Jones (2009) describes a “burden of knowledge”; as science is generally cumulative, researchers at the scientific frontier must keep up with an increasing body of knowledge. This, in turn, prolongs training times and renders scientific breakthroughs harder to achieve.
Notes: Gross domestic spending on R&D measured in USD billion at constant prices, using 2010 as a base year. Growth domestic product growth rates expressed in percentages.
Source: OECD data, https://data.oecd.org.
Bloom et al. (2020) provide ample empirical evidence, derived from industry case studies and firm-level analyses, that research productivity has fallen over time in the United States. Consequently, and if correct, R&D activities today create a much smaller growth impulse for a given level of spending than 40 years ago.
It is also important methodologically to examine micro-level evidence next to the aggregate data in Figure 1. If new industries are added in an expanding economy, aggregate R&D spending could in principle be increasing while the level of spending per industry is kept constant. Macro trends such as the ones presented in Figure 1 would then suggest a decline in research productivity, although it remains stable at the micro level. To avoid such a misleading picture, researchers need to zoom in and estimate research productivity trends over time for individual sectors and firms. Bloom et al. (2020) analyse the semiconductor industry, agriculture, health care and the US manufacturing sector as a whole. They find evidence for a substantial decline in R&D productivity in all these domains.
Boeing and Hünermund (2020) replicate the findings of Bloom et al. (2020) for the two largest R&D spending economies in Asia and Europe: China and Germany. They focus on firm-level data and analyses since these provide the most generalisable evidence across different sectors. In other words, because the microdata cover firms in most sectors, the results are likely to be representative of the business sector as a whole rather than just a specific industry or technological field.
As a starting point for measuring the productivity of R&D over time, the analysis uses the following idea production function, which is standard in the endogenous growth literature (Romer, 1990; Aghion and Howitt, 1992):
Economic growth = Research productivity × Number of researchers
In line with the approach of Bloom et al. (2020), this allows computation of research productivity at the firm level by dividing growth rates by the number of R&D employees. They compute growth rates, in the numerator, based on several commonly used output measures: sales revenue, employment level, market capitalisation and revenue labour productivity (i.e. sales revenue per worker). The number of researchers, in the denominator, is proxied by R&D spending divided by the average wage of high-skilled workers in the economy. This operationalisation has the advantage of also accounting for complementary capital expenditures within the R&D process.
To smooth out short-term business cycle fluctuations, numbers are averaged over a ten-year period and the change in research productivity is computed over two consecutive decades. This methodology requires detailed firm-level panel data over a long period. For Germany, the analysis relies on data for 64 902 firms from the Community Innovation Survey (Peters and Rammer, 2013; OECD/Eurostat, 2018) for 1992-2017. For China, it takes the universe of 3 947 Chinese firms listed on the Shanghai and Shenzhen stock exchanges – China’s so-called A-share market – in 2001-19.
Since firms need to be observed over two consecutive decades and figures are averaged per decade, the sample size drops considerably (to 1 121 in Germany and 516 in China). This is consistent with the original analysis in Bloom et al. (2020). Moreover, as with this essay’s study of China, Bloom and co-authors also analysed research productivity trends in US publicly listed firms based on Compustat data. Compared to firms listed on a stock exchange, the Community Innovation Survey contains a larger number of privately held small and medium-sized enterprises. This must be considered when comparing results across countries.
For Germany, R&D expenditures, measured in head counts, increased by an average of 3.3% per year during the period of investigation. At the same time, this expansion of research activities is not accompanied by a similar increase in growth at the firm level. Averaged over all the firm-level outcome measures previously discussed, research productivity declines by 5.2% per year. This is remarkably similar to numbers that Bloom et al. (2020) report for the aggregated US economy. These negative compound average growth rates imply that research productivity reduces by half roughly every 13 years. In other words, research efforts must be doubled every 13 years to support constant economic growth rates.
For China, research productivity has declined even more drastically. This implies the initial growth of significant R&D activities in China that generated high returns in the 2000s subsequently diminished. Here, the effective number of researchers employed by publicly listed firms in the sample increases by, on average, 21.9% per year between 2001 and 2019. Meanwhile, economic growth rates again do not increase proportionally. Arithmetically, this entails a drop in research productivity of 23.8% per year.
The implied half-life of research productivity of around three years constitutes a swift decline. However, if analysis is restricted to the last decade (when China began large-scale R&D activities), and growth rates are compared in five-year intervals (2010‑14 and 2015‑19), research productivity declines by only 7.3%. These numbers are closer to the ones found for Germany and the United States. They may reflect China’s progression from when it was aiming to catch up to developed economies to one where it has been operating closer to the research frontier in many fields.
Figure 2 plots the histograms of changes in number of researchers (light blue bars) and research productivity (dark blue bars) across the two samples. A constant research productivity and a constant number of researchers over time would correspond to a factor change equal to one. As the blue bars in the histogram illustrate, most firms in the sample are located to the left of one, which implies declining research productivity over time. Many firms experienced positive growth rates in research productivity during the last three decades, especially in Germany. This substantial degree of heterogeneity is in line with what Bloom et al. (2020) find for the United States.
Note: Aside from the light blue and the dark blue bars, the third colour denotes where the distributions overlap.
Source: Peters and Rammer, 2013; OECD/Eurostat, 2018; authors’ calculations based on data from Shanghai and Shenzhen stock exchanges.
Discussion
According to this analysis and that of Bloom et al. (2020), the leading R&D-performing countries in North America, Asia and Europe have all been experiencing a decline in average research productivity over the last two decades. One implication is that further increases in global R&D inputs will be required to avoid contracting GDP growth. For example, according to China’s 14th Five-Year Plan for 2021‑25, gross R&D spending is expected to increase by at least 7% annually. This is well above the projected 5% annual GDP growth and implies that the current R&D-to-GDP ratio of about 2.2% will increase further. China already accounted for 24.4% of global spending on R&D in 2018, while the United States accounted for 25.6% (in purchasing power parity terms).1
However, there may be limits to the increase in actual R&D input. This is because an inelastic supply of researchers tends to increase the cost of R&D (e.g. through higher wages for scientists) but not the amount of R&D activity (Goolsbee, 1998). In the past, education reforms in China led to a steady growth in the number of college graduates. This resulted in a notable increase in the supply of students and researchers in China and abroad. The United States also benefited greatly from its attraction of foreign scientific talent from around the world.
However, the long-term decline in population growth in industrialised countries, coupled with shocks to international mobility (e.g. during the COVID-19 pandemic) may take a toll on the number of deployable researchers. Thus, in addition to ensuring the quantitative supply of new researchers, policy makers must improve the quality of education and research and the optimal allocation of ideas to slow (or even reverse) the decline in research productivity.
As one limitation, the methodology in this paper cannot distinguish between a productivity effect and a business-stealing effect on firm growth. That is, a firm’s R&D and the associated process of creative destruction might influence not only its own productivity but also the market share of its competitors. In that case, output growth would not be the result of increased productivity but come at the cost of competitors. This, in turn, would lead to an overestimation of research productivity at the firm level. Even though both effects could occur at the same time, the productivity effect of R&D has been shown to dominate the business-stealing effect empirically (Bloom Schankerman and Van Reenen, 2013).
Likewise, the methodology does not explicitly consider technology spillovers. Since spillovers lead to productivity growth without direct R&D investments, they enter the measure of research productivity in the numerator. Thus, if technology spillovers had been slowing during the period of observation, it could partly explain the decline in research productivity found over time.
Ideas are not only getting harder to find in the United States but also in the European and Asian countries that spend the most on R&D. Although the estimates are difficult to compare due to different data sources, negative growth rates in Germany and the United States are remarkably similar. China has experienced a much higher decline in research productivity, but growth rates appear to be converging to those of the United States and Germany in recent years. This convergence coincides with a shift in innovation-driven growth in China. This growth has evolved from the imitation of inventions developed elsewhere to genuine innovation at the global technology frontier.
References
Aghion, P. and P. Howitt (1992), “A model of growth through creative destruction”, Econometrica, Vol. 60/2, pp. 323-351, https://doi.org/10.2307/2951599.
Bloom, N. et al. (2020), “Are ideas getting harder to find?”, American Economic Review, Vol. 110/4, pp. 1104-1144, https://doi.org/10.1257/aer.20180338.
Bloom, N., M. Schankerman and J. Van Reenen (2013), “Identifying technology spillovers
and product market rivalry”, Econometrica, Vol. 81/4, pp. 1347-1393, https://doi.org/10.3982/ECTA9466.
Boeing, P. and P. Hünermund (2020), “A global decline in research productivity? Evidence from China and Germany”, Economics Letters, Vol. 197/109646, https://doi.org/10.1016/j.econlet.2020.109646.
Cowen, T. (2011), The Great Stagnation: How America Ate All the Low-Hanging Fruit of Modern History, Got Sick, and Will (Eventually) Feel Better, Dutton, New York.
Goolsbee, A. (1998), “Does government R&D policy mainly benefit scientists and engineers?”, American Economic Review, Vol. 88, pp. 298-302, www.jstor.org/stable/116937.
Gordon, R.J. (2016), The Rise and Fall of American Growth: The US Standard of Living Since the Civil War, Princeton University Press.
Jones, B.F. (2009), “The burden of knowledge and the ‘death of the renaissance man’: Is innovation getting harder?”, The Review of Economic Studies, Vol. 76/1, pp. 283- 317, www.jstor.org/stable/20185091.
OECD/Eurostat (2018), Oslo Manual 2018: Guidelines for Collecting, Reporting and Using Data on Innovation, 4th. Edition, The Measurement of Scientific, Technological and Innovation Activities, OECD Publishing, Paris/Eurostat, Luxembourg, https://doi.org/10.1787/9789264304604-en.
Peters, B. and C. Rammer (2013), “Innovation panel surveys in Germany”, in Handbook of Innovation Indicators and Measurement, Gault, F. (ed.), Edward Elgar Publishing, Cheltenham.
Romer, P.M. (1990), “Endogenous technological change”, Journal of Political Economy, Vol. 98/5, pp. 71-102, www.jstor.org/stable/2937632.