1According to economists such as Robert Gordon, the low GDP and productivity growth observed in all major geographical regions since the start of the 21st century could be a lasting phenomenon (see Gordon, 2012, 2013, 2014, 2015). Gordon posits that the slowdown in productivity is linked to the smaller gains in productive performance derived from today’s innovations. Innovations, he suggests, now deliver less growth than the previous technological revolutions, which profoundly changed modes of production and consumption. As a consequence, in addition to the risk of a secular stagnation caused by insufficient demand, as discussed by Summers (2014, 2015) or Eichengreen (2015), [2] there is also the risk of a supply-side stagnation, caused by subdued productivity growth.

2This pessimistic vision of future productivity growth has been countered by several economists, including Mokyr, Vickers and Ziebarth (2015), Brynjolfsson and McAfee (2014), van Ark (2016), and Branstetter and Sichel (2017). In their view, the current slowdown is a temporary lull ahead of a sharp pick-up fuelled by the digital economy. Moreover, the acceleration could prove to be particularly strong as it will affect all segments of the economy simultaneously.

3This article aims to revisit the debate over secular stagnation. In section 1 we describe empirically the long-term slowdown in GDP and productivity growth in advanced countries; in section 2 we examine the situation in a sample of emerging countries; and in section 3 we offer various explanations for these long-run trends. Section 4 then discusses the outlook for the future and section 5 concludes.

1 – The Decline in Growth is Attributable to a Slowdown in Productivity

4Chart 1 provides an accounting breakdown of average annual growth in nominal GDP for the period 1890-2016 in the main developed economies. [3] Five components are identified: population growth, the employment rate (here the number of people in employment as a share of the total population), the number of hours worked, total factor productivity (or TFP for short) and capital intensity. The sum of TFP and capital intensity corresponds to the contribution of labour productivity.

Chart 1

Accounting breakdown of average annual GDP growth from 1890 to 2016

Accounting breakdown of average annual GDP growth from 1890 to 2016

Lecture note: On average, from 1890 to 1913, US GDP grew by 3.6% per year. The contributions to this growth were 1.0 percentage point for TFP, 0.5 percentage point for capital intensity, 1.8 percentage points for population growth, 0.4 percentage point for the employment rate and -0.1 percentage point for hours worked.

5Over the entire period and in all geographical areas studied, the strongest contribution to GDP growth comes from hourly labour productivity. Moreover, within hourly labour productivity, the TFP component makes a much larger contribution than capital intensity. It should be noted, however, that the breakdown of hourly labour productivity into TFP and capital intensity is statistically fragile. In particular: (i) the weighting applied to the two main factors of production, capital and labour, which is necessary to calculate TFP, relies on major assumptions, notably that this weighting remains stable over time and space; (ii) the volume-price breakdown of investment and therefore capital is based on investment price indices that do not accurately capture gains in product performance and quality, especially in the case of information and communication technologies (ICTs for short); [4] (iii) in order to construct capital stock figures from investment data, assumptions need to be made about mortality rates for different investment components. These assumptions and how they evolve over time are based on incomplete information.

6Chart 1 also reveals that TFP and labour productivity have not grown steadily over the period. Several studies have shown that they have in fact increased in a wave-like pattern, and that different countries have emerged as leaders at different times. Moreover, not all countries have succeeded in catching up with the leaders (see, for example, Crafts and O’Rourke, 2013, or Bergeaud, Cette and Lecat, 2016), and the success or failure of this catch-up process depends on interactions between innovation, education levels, and economic and political institutions (see notably Aghion and Howitt, 1998, 2009).

7In the United States, three main stylised facts can be singled out concerning the contributions of TFP and labour productivity to GDP growth:

• Throughout most of the 20th century, productivity made a significant and incremental contribution to growth, a phenomenon referred to by Gordon (1999) as “the one big wave”. This wave corresponds to the Second Industrial Revolution which saw numerous innovations, the most notable of which, according to Gordon, were the increasing use of electricity in lighting and motors, the use of the internal combustion engine in industry and transport, the invention of chemicals and notably petrochemicals and pharmaceuticals, and the transformation of information and communication with the dissemination of the telephone, radio, cinema, etc. These new technologies translated into major productivity gains, thanks to an increasingly educated population.
• The decade 1995-2005 saw a sharp increase in the contribution of productivity to growth. This period corresponds to the Third Industrial Revolution or Digital Revolution, characterised by the diffusion of ICTs. There is ample literature on this phenomenon in the United States, notably Jorgenson (2001), and Jorgenson, Ho and Stiroh (2006, 2008).
• With the exception of the decade from 1995 to 2005, the contribution of productivity to growth has declined steadily since 1950, which explains the slowdown in GDP growth. Various studies have shown that the slowdown observed at the end of the recent period in fact began before the Great Recession (see, for example, Byrne, Oliner and Sichel, 2013; Fernald, 2015; Bergeaud, Cette and Lecat, 2016, 2017).

8In the other main economic regions studied here, the wave of labour productivity growth corresponding to the Second Industrial Revolution occurred several decades later than in the United States (although the lag was slightly smaller in the case of the United Kingdom). Moreover, the wave of productivity growth corresponding to the Third Industrial Revolution never actually materialised in the euro area or Japan, and was only felt to a limited extent in the United Kingdom. In these three economic areas, as in the United States, the contribution of productivity has declined steadily but, in contrast with the United States, the decline began after the first oil shock and not after the Second World War. Moreover, the United Kingdom saw a very slight rise in the contribution of productivity to growth in the decade from 1995 to 2005.

9These stylised facts have already been commented on (see for example Crafts and O’Rourke, 2013; Bergeaud, Cette and Lecat, 2016 and 2017) and are now widely accepted. The point we need to underline for the purposes of this study is the historically low level of productivity growth reached since the start of this century.

2 – Convergence Trends Differ Across Emerging Countries

10The very low rates of productivity growth observed recently in advanced countries have not been replicated in all emerging countries. In the latter, productivity growth tends to be driven by the process of convergence towards the productivity frontier in developed countries. And this convergence is in turn influenced by institutional factors, such as the educational attainment of the working age population and the quality of existing institutions (for a summary of the literature on this subject, see Aghion and Howitt, 1998, 2009).

11A study currently underway [5] has put together comparable productivity series for a number of emerging countries, in particular in South America, using a similar logic as for the developed countries discussed above. Chart 2 shows the level of hourly labour productivity [6] for five emerging countries (Argentina, Brazil, Chile, South Korea and Mexico) relative to that of the United States for the period 1890-2016. As can be seen, the speed and degree of convergence with the United States varies markedly across countries. The following main trends can be observed in relative productivity (i.e. expressed as a percentage of that of the United States): (i) an almost continuous decline over the entire period for Argentina; (ii) a relative stability in Brazil and Mexico over the entire period and, for South Korea, in the period prior to the war at the start of the 1950s; and (iii) a fairly rapid rate of convergence in Chile since the 1980s and in South Korea since the mid-1950s. These differences in trajectories confirm that convergence in productivity levels is not automatic and that the speed and success of the process depend on various factors. Argentina is a particularly interesting case as, at the start of the period, it was one of the leaders and the only country with a comparable level of productivity to the United States. Despite this, it failed to adapt its institutions sufficiently to profit from the growth delivered by innovation: due to strong demographic growth, it had insufficient domestic savings to finance its development when the international financial markets collapsed in the interwar period. As a result, from the First World War onwards, its productivity declined steadily relative to developed countries (see in particular Taylor, 1992, on Argentina, and Acemoglu, Aghion and Zilibotti, 2006, for a demonstration of the importance for growth in frontier countries of having institutions that are adapted to innovation).

Chart 2

Hourly labour productivity relative to US

Hourly labour productivity relative to US

3 – Growth in Productivity and Real Interest Rates: A Circular Relationship?

12One comment frequently made is that GDP growth (and therefore productivity growth) fails to accurately measure, or even ignores, several aspects of effective growth over the recent period, which is being increasingly driven by the digital economy and by new technologies. A number of studies have focused on this issue in recent years, and all seem to concur that the size of this underestimation has remained fairly stable for several decades and cannot therefore explain the recent slowdown (see, for example, Byrne, Fernald and Reinsdorf, 2016, Syverson, 2016, Aghion et al., 2017, or, on France, Bellégo and Mahieu, 2016). Moreover, this measurement bias is only one of the many difficulties with GDP – traditional measures of economic output also ignore other elements that have become increasingly important in recent decades, such as non-market home production.

13The mismeasurement of GDP does not therefore appear to be the cause of the observed slowdown, and various other explanations have been put forward. Analyses conducted by the OECD on firm-level data, for example, indicate that the global productivity slowdown since the start of the 2000s has not affected frontier firms, and could therefore be explained in part by stalling technological diffusion between these firms and the laggards (see Andrews, Criscuolo and Gal, 2015). This decrease in diffusion could in turn be attributable to various factors, some of which relate to the digital economy: difficulties appropriating certain forms of intangible capital, “winner-takes-all” dynamics in many sectors of activity, etc. However, the study says nothing of the causes of these phenomena, and why they appeared simultaneously in all developed economies, despite marked differences in their respective productivity levels, technological progress, education levels and institutions. Moreover, these phenomena only apply to certain sectors of activity, whereas the observed slowdown extends beyond those sectors that are ICT-intensive.

14Recent analyses by Cette, Corde and Lecat (2017) on a vast sample of French firms confirm that the slowdown in productivity in the 2000s does not stem from a loss of momentum at the technology frontier. There has been no visible slowdown in productivity at frontier firms which, for France at least, appears to refute the theory that we have exhausted the potential gains from technological progress. However, the same data also show that there was no slowdown in the convergence of followers towards the technology frontier in the 2000s, which contradicts the theory that there was a decline in the diffusion of innovations between frontier firms and laggards. At the same time, the dispersion of productivity levels appears to have increased, which could point to a less efficient allocation of factors of production towards frontier firms. This problem could stem from the fact that various shocks have made it necessary to reallocate resources (globalisation, emergence of ICTs, financial crisis) but that this reallocation process has been made difficult by existing rigidities.

15One explanation for the increase in productivity dispersion could be the steady fall in real interest rates to ultra-low levels. These enable the least productive firms to survive but also make less efficient investment projects more profitable. Chart 3 shows that real interest rates did indeed start to decline in the main advanced countries from the mid-1980s onwards.

16The fall in real interest rates from the mid-1980s could indeed have slowed mortality rates for less productive firms (decline in the “cleansing effect”), thereby hampering the reallocation of factors of production to firms at the frontier. Lower rates could also have made it easier to finance less efficient projects, and this combination of factors could in turn have reduced productivity gains. Several studies have provided support for this explanation (see, for example, Reis, 2013, Gopinath et al., 2017, Gorton-Ordonez, 2015, and Cette, Fernald and Mojon, 2016). It is interesting to note that the majority of these studies, in particular those of Reis (2013) and Gopinath et al. (2017), have focused on southern European countries (notably Spain, Italy and Portugal) and on the recent period. For the same period (i.e. since the start of the 2000s), the studies find no such relationship between financing and productivity in other countries such as Norway, Germany or France. Moreover, the decline in productivity gains and hence in potential growth is itself a contributing factor behind the fall in real interest rates (for an empirical analysis of this relationship and a summary of the existing literature, see Teulings and Baldwin, 2014, Bean, 2016, or Marx, Mojon and Velde, 2017).

Chart 3

Real long-term interest rates – 10-yr government bond yields

Real long-term interest rates – 10-yr government bond yields

17Low interest rates thus appear to lead to a fall in productivity which in turn leads to a decline in rates, creating a circular relationship between TFP growth and real interest rates. Only a technology shock could disrupt this downward spiral, but for an economy to reap the full benefits of such a shock, it needs to have the right institutions in place. Not all countries would derive the same TFP gains from a technology shock. Yet, due to capital mobility, all would experience a simultaneous rise in real interest rates caused by the increase in potential growth in those countries that have benefited fully from the shock because they have adequate institutions. Countries with poorly adapted institutions would thus be dually penalised: real interest rates would rise, but they would not profit fully from the acceleration in productivity growth stemming from the technology shock.

18In this study, we carry out a model estimation based on this circular relationship, using both macroeconomic data and individual firm-level data. The results of our estimations using macroeconomic data for 17 developed countries over the period 1950-2016 are described in the appendix. These results provide an initial confirmation that a circular relationship exists between TFP growth and real interest rates.

4 – What is the Outlook for the Long Term?

19The literature generally cites two potential sources of future productivity growth. The first is an acceleration in ICT performance gains and the second the extension of the use of existing ICT performances to other segments of the economy.

20Regarding the first source, various recent analyses based on in-depth technological studies of semiconductor manufacturers indicate that there could be significant gains in the performance of these products at various stages in the future: first, in the nearer term, the widespread operational use of 3D chips; second, in the longer term, the harnessing of the potential offered by quantum computing (see summary by Cette, 2014 and 2015) and artificial intelligence (see Aghion, Jones and Jones, 2017).

21Regarding the second source, various analyses have stressed that it always takes a long time for the full impact of a technological revolution to be felt in productive activity (see, for example, Brynjolfsson and McAfee, 2014; van Ark, 2016; Branstetter and Sichel, 2017). As Robert Solow famously wrote in a 1987 article in the New York Times, [7] “You can see the computer age everywhere, but in the productivity statistics”. This impatience suggests we have forgotten what happened in previous technological revolutions: the profound changes were only diffused gradually, and their impact on productive performances was not felt until decades later. David (1990) has shown that between 50 and 60 years passed between the invention of a working electric dynamo in 1868 and its full exploitation in production (in the 1920s to 1930s). The widespread use of ICTs in the most developed countries has clearly had an impact on productivity, but the benefits have so far been limited and the best could yet be to come. All, or nearly all, sectors of the economy could be profoundly affected by the digital revolution. The huge improvements in ICT performance are making it possible to exploit massive databases almost instantaneously (big data) and at the same time are fuelling the development of artificial intelligence. In other words, as van Ark said (2016), the current pause in the productivity gains from the Third Industrial Revolution could in fact be a period of transition between the creation and installation of new technologies and their full deployment. As with previous technological revolutions, notably the invention of electricity, this deployment phase will take time and will require major changes to our institutions and to our methods of production and of management. However, it is already close at hand.

22It is still difficult to predict with any accuracy how the digital economy will change productive activity and, more broadly, our way of life. The historical analyses conducted by Mokyr, Vickers and Ziebarth (2015) remind us that forecasts of this type are frequently wrong. At best, we can probably predict what will happen in a few sectors where the changes are already partly visible or imminent. One example, of course, is in transport, where the emergence of driverless vehicles will lead to major gains in productivity, and will completely transform the production of transport equipment, such as cars. These changes will relate not just to the technological content of the equipment itself, but also to the quantities manufactured, as the same needs will be met more efficiently with smaller amounts of materials. In other areas such as banking and retail, similar radical changes are already starting to make themselves felt.

5 – Concluding Remarks

23There is no real consensus among economists as to the causes of the marked productivity slowdown in advanced economies. However, numerous studies suggest the phenomenon could be temporary and that productivity could in fact accelerate again, although it is still unclear when. According to this hypothesis, a secular stagnation could yet materialise if the conditions are not in place for an improvement in demand. In the euro area, these conditions are particularly difficult to achieve, as they imply genuine economic policy coordination between fiscally independent countries, in a context where weak demand, characterised by high unemployment, is concentrated in certain countries (mainly southern Europe), while the fiscal leeway and current account surpluses are concentrated in others (essentially northern Europe, mainly Germany and the Netherlands). Monetary policy has done a great deal to stimulate domestic demand in the euro area, with the implementation of so-called non-standard tools in the past few years, including the purchase of sovereign debt. But monetary policy is not the only game in town and it certainly cannot make up for a lack of coordination in domestic demand policies. The only way to alleviate this lack of coordination is to stimulate domestic demand in those countries where there is room for manœuvre, via stronger wage growth or more expansionary fiscal policies (cuts in taxes or hikes in public spending).

24With regard to productivity, the euro area undoubtedly suffers from ill-adapted institutions, which are preventing it from reaping the full benefits of new technologies and the associated productivity gains. However, as part of this debate over productivity, another important issue needs to be addressed: the outlook for the euro area as a whole. The bloc’s underperformance relative to the United States is not inevitable, but is the result of institutional choices and specific policies. Without important changes in these fields, the euro area will increasingly be left behind by other advanced economies, and will struggle to face the numerous challenges of the future. These challenges, which Gordon (refers to as headwinds 2012, 2013, 2014, 2015), are significant and include population ageing, growth sustainability and the reduction of public debt. Moreover, without sufficient productivity growth to oil the wheels of the economy, the political risks to European democracy would inevitably increase.