The Global Forum on Productivity at 10

Past and Future Perspectives on Reviving Productivity Growth

Report

15 September 2025

2. The best versus the rest: The global productivity slowdown, divergence across firms and the role of public policy

Copy link to 2. The best versus the rest: The global productivity slowdown, divergence across firms and the role of public policy

Abstract

This chapter shifts the global productivity slowdown debate from the traditionally macro perspective to a more micro one. We show that a striking feature of the productivity slowdown is not lower productivity growth at the global frontier, but rather rising labour productivity at the global frontier coupled with increasing labour productivity divergence between frontier and laggard firms. This divergence remains after controlling for differences in capital deepening and mark-up behaviour, suggesting that divergence in measured multi-factor productivity (MFP) may broadly reflect technological divergence. The divergence likely reflects how structural changes – namely digitalisation, globalisation and the rising importance of tacit knowledge – can fuel rapid productivity gains at the global frontier. Yet, aggregate MFP performance was significantly weaker in industries where MFP divergence was more pronounced, suggesting that divergence is not solely driven by frontier firms pushing the boundary outward. We contend that growing MFP divergence and the global productivity slowdown could reflect a slower diffusion process, as laggard firms face higher costs transitioning from production to idea-based economies, indicating increased entry barriers and reduced market contestability. The rise in MFP divergence was much more extreme in sectors where pro-competitive product market reforms were least extensive, suggesting policy weaknesses may be stifling diffusion.

Introduction

Copy link to Introduction

Aggregate productivity growth slowed in many OECD countries, even before the crisis, igniting a spirited debate on the future of productivity (e.g. Gordon, 2012 vs Brynjolfsson and McAfee, 2012)1. Following on the steps of previous analysis (Andrews, Criscuolo and Gal, 2015), this chapter goes further in exploiting firm level evidence to shed new light on the factors behind the global slowdown in productivity growth – a debate which has by and large been conducted from a macroeconomic perspective. While this debate often concerns the prospects for innovation at the global productivity frontier, little is known about the productivity growth performance of global frontier firms over time both in absolute terms and relative to laggard (i.e. non-frontier) firms.2 Even less is known about the policies that might help laggard firms close their productivity growth gap with the frontier. Yet, cross-country differences in aggregate-level productivity are increasingly being linked to the widespread heterogeneity in firm performance within sectors (Bartelsman, Haltiwanger and Scarpetta, 2013; Hsieh and Klenow, 2009).

To fill this gap, we highlight a number of policy-relevant issues related to the performance of frontier firms and laggards, with a view to also shed light on aggregate productivity developments in OECD countries. Using a harmonised cross-country firm-level database for 24 countries, we define global frontier firms as the top 5% of firms in terms of labour productivity or MFP levels within each two-digit sector in each year across all countries since the early 2000s. Our analysis suggests that a striking feature of the productivity slowdown is not so much a slowing in productivity growth at the global frontier, but rather rising productivity at the global frontier coupled with an increasing productivity gap between the global frontier and laggard firms. In fact, slow productivity growth of the “average” firm masks the fact that a small cadre of firms are experiencing robust gains.

We show that the rising labour productivity gap between global frontier and laggard firms largely reflects divergence in revenue-based MFP (MFPR), as opposed to capital deepening. Moreover, we explore the role of market power and conclude that divergence in MFPR does not simply reflect the increasing ability of frontier firms to charge higher mark-ups. While there is evidence that market power of frontier firms has increased in services, this amounts to less than one-third of the total divergence in MFPR. This leads us to the conclusion that the rising MFPR gap between global frontier and laggard firms may in fact reflect divergence in physical productivity or technology, broadly defined. Importantly, contemporary innovation relates not only to the capacity of firms to introduce technologically more advanced goods and services but also to their success in tacitly combining various intangibles – e.g. computerised information; innovative property and economic competencies (see Corrado, Hulten and Sichel, 2009) – in production processes.

This pattern of MFP divergence might seem surprising for at least two reasons. First, neo-Schumpeterian growth theory (Aghion and Howitt, 2006; Acemoglu, Aghion and Zilibotti, 2006) and models of competitive diffusion (Jovanovic and MacDonald, 1994) imply productivity convergence: that is, firms further behind the global frontier should grow faster, given the larger stock of unexploited technologies and knowledge that they can readily implement. Second, the extent of productivity divergence observed in the data is difficult to reconcile with models of creative destruction and a world where the process of market selection is productivity-enhancing (Aghion and Howitt, 1992; Caballero and Hammour, 1994; Campbell, 1998), raising questions about the competitiveness of markets. However, our results suggest that both the rate of convergence and growth-enhancing reallocation have slowed down during the last decade leading to the divergence evident in the data.

The chapter then explores a set of structural factors underlying MFP divergence, links with aggregate productivity performance and public policy implications. Structural changes in the global economy – namely digitalisation, globalisation and the rising importance of tacit knowledge – could underpin MFP divergence through two interrelated channels introduced below. While it is difficult to pinpoint their relative importance, symptomatic evidence emerges to suggest that each may be important in explaining MFP divergence.

First, the increasing potential for digital technologies to unleash winner takes all dynamics in the global market (Brynjolfsson and McAfee, 2012) has enabled technological leaders to increase their performance gap with laggard firms. In support of this hypothesis, we find three distinct patterns in information and communication technology (ICT) intensive services (e.g. computer programming, software engineering, data storage and other types of information service activities) – where winner take all patterns should be more relevant – that are less evident elsewhere: i) global frontier firms have increased their market share; ii) MFP divergence is more pronounced; and iii) within the global frontier, the productivity of the most elite firms (top 2%) has risen relative to that of other frontier firms (top 5%).

All else equal, these patterns are not necessarily a policy concern and could imply higher aggregate productivity growth via stronger innovation intensity and more efficient resource allocation. Yet, we find the opposite: aggregate MFP performance was significantly weaker in industries where MFP divergence was more pronounced. This suggests that the obstacles to the productivity growth of laggards increased, weighing on aggregate productivity growth. This leads us to explore a second source of MFP divergence and the aggregate productivity slowdown: stalling technological diffusion. One possible explanation is that the growing importance of tacit knowledge and complexity of technologies has increased the sophistication of complementary investments required for the successful adoption of new technologies. It also possible that we are entering into a new technological wave where know-how is tacitly held by few, for example, the early adopters – who are learning – while everyone else is still lagging. In either case, this creates barriers to the catch-up of laggard firms. At the same time, the concomitant decline in market dynamism and rising market power of frontier firms suggests that the stagnation in the MFP growth of laggard firms may be connected to rising barriers to entry and a decline in the contestability or competitiveness of markets.

This latter raises the prospect that while rising MFP divergence was somewhat inevitable due to structural changes in the global economy, there was scope for public policy to lean against these headwinds and to better align the regulatory environment with structural changes in the global economy. In fact, we find MFP divergence to be much more extreme in sectors where pro-competitive product market reforms or deregulation were least extensive. Given the link between product market competition and incentives for technological adoption (see Aghion and Howitt, 2006 and references therein), part of the observed rise in MFP divergence may be traced to policy failure to encourage the diffusion of best production practices in OECD economies. A simple counterfactual exercise suggests that had the pace of product market reforms in retail trade and professional services – where market regulations remained relatively stringent in OECD countries – been equivalent to that experienced by telecommunications – where reforms were most extensive – then the average increase in the MFP gap may have been up to 50% lower than observed. As most of the outputs produced by these heavily regulated sectors are used as inputs in production elsewhere in the economy (Bourlès et al., 2013), this may in fact provide a lower bound of the total impact of excessively stringent service regulation on MFP divergence.

The next section places our research in the context of the existing literature on the productivity slowdown. The following section discusses the firm level data set and productivity measurement issues, before identifying and describing the characteristics of firms at the global productivity frontier. We then present new evidence on labour productivity divergence between global frontier and laggard firms in OECD countries and then explores the relative roles of capital, MFP, market power, winner takes all dynamics and technology diffusion. Subsequently, we explore aggregate implications and the link between product market reforms and the MFP gap, with a particular focus on diffusion in the services sector. The final section provides a qualitative discussion of other factors that may potentially explain MFP divergence and identifies some areas for future research.

The productivity slowdown

Copy link to The productivity slowdown

The productivity slowdown has sparked a lively debate on its underlying causes and the future of productivity more generally. It also underpins the collapse in potential output growth – a key metric of societies’ ability to make good on promises to current and future generations (OECD, 2016). Indeed, potential output growth has slowed by about one percentage point per annum across the OECD since the late 1990s, which is entirely accounted for by a pre-crisis slowing in MFP growth and to a more recent slowdown in capital deepening (Figure 2.1). Against this background, this section reviews some of the competing explanations for the productivity slowdown and places our research in the context of the existing literature.

Figure 2.1. Weak labour productivity underpinned the decline in potential output in OECD countries
Copy link to Figure 2.1. Weak labour productivity underpinned the decline in potential output in OECD countries

Contribution to potential per capita growth (% points unless otherwise noted)

Techno-pessimists and techno-optimists

The debate on the productivity slowdown has focused on expectations about the pace and economic potential of innovations at the technological frontier. Indeed, there are strongly contrasting views on the potential of ICT to continue to propel productivity growth.

Techno-pessimists argue that the slowdown is just a reflection of a “return to normal” effect after nearly a decade of exceptional gains fuelled by information technology (IT), given that the slowdown is driven by industries that produce IT or use IT intensively (Fernald, 2014). This view holds that the most recent slowdown is a permanent phenomenon and that the types of innovations that took place in the first half of the 20th century (e.g. electrification) are far more significant than anything that has taken place since then (e.g. ICT), or indeed, likely to transpire in the future (Gordon, 2012; Cowen, 2011).3 These problems are likely to be compounded if it becomes more costly for researchers to innovate the further technology advances and ideas cumulate (Jones, 2012). Such arguments are reinforced by the slowdown in business dynamism observed in frontier economies such as the United States (Decker et al., 2014; Decker et al., 2016b; Criscuolo, Gal and Menon, 2014).

Against this, techno-optimists argue that while the slowdown might reflect a difficult transition from an economy based on tangible production to one based on ideas, the underlying rate of technological progress has not slowed, and that the IT revolution will continue to dramatically transform frontier economies. According to Brynjolfsson and McAfee (2012), the increasing digitalisation of economic activities has unleashed four main innovative trends: i) improved real-time measurement of business activities; ii) faster and cheaper business experimentation; iii) more widespread and easier sharing of ideas; and iv) the ability to replicate innovations with greater speed and fidelity (scaling-up). While each of these trends are important in isolation, their impacts are amplified when applied in unison.4 Similarly, Mokyr (2014) argues that advances in computing power and ICT have the potential to fuel future productivity growth by making advances in basic science more likely and reducing access costs and thus igniting a virtuous circle between technology and science. However, he also warns of the potential for bad institutions and policies to act as obstacles to this virtuous cycle (Mokyr, 2014).5

One interesting angle in the techno-optimist argument is that we might not have seen the full benefits of the “digital economy” because we are still in a transition phase characterised by staggered adoption of the new technology and transition costs. Such transition dynamics align closely with the idea that ICT is a general-purpose technology (GPT) whose adoption and diffusion is characterised by an S-curve (Griliches, 1957; David, 1991; Jovanovic and Rousseau, 2005). In particular, GPT adoption and diffusion is complicated by a high cost of learning on how to use it effectively; large adjustment costs and slow introduction of complementary inputs, especially knowledge-based capital (KBC). Indeed, the productivity slowdown may reflect the time-consuming process of making these complementary investments that are necessary for digital technologies to deliver large productivity benefits in the long run (Basu, Fernald and Kimball, 2006).

Macroeconomic factors

Although aggregate productivity slowed before the crisis in many economies, the debate has also focused on the role of non-technology macroeconomic factors, namely demand, savings and monetary policy. Accounts linking demand to the slowdown tend to emphasise “secular stagnation”, whereby there is an imbalance between savings and investment caused by an increased propensity to save and a decreasing propensity to invest which in turn leads to excessive savings dragging down demand, lower real interest rates and a reduction in growth and inflation (Summers, 2016). Significant growth, such as that characterising the 2003-2007 boom, was achieved thanks to excessive levels of borrowing and unsustainable investment levels.

Christiano, Eichenbaum and Trabandt (2015) analysed the role of macro shocks and financial frictions during the crisis as triggers of the slowdown, but such models take the slowdown in MFP as exogenous. Of more interest for our purposes is a paper by Anzoategui et al. (2016), which proposes a theoretical model whereby the increase in demand for liquidity, as observed during the crisis, increases the spread between the cost of capital and the risk-free rate of liquid assets. This leads to a decline in investment in R&D and technological adoption, which in turn yields lower output and lower MFP. According to the model, the spread of technology adoption varies over the business cycle, with the cyclicality mainly driven by fluctuation in the adoption rate, which depends also on fiscal and monetary policies. The model, however, must rely on exogenous medium-term factors to explain the pre-recession slowdown.

Rising resource misallocation

Gopinath et al. (2015) explore the implications for sectoral MFP of the decline in real interest rate, observed in Southern Europe during the euro-convergence process. They find that the associated capital inflow was increasingly misallocated towards firms that had higher net worth but were not necessarily more productive, which could explain why MFP slowed in Southern Europe – especially Spain – even before the crisis. This misallocation-driven slowdown was further exacerbated by the additional uncertainty generated by the crisis and more generally is likely to be related to weakening market selection, declining business dynamism and deteriorating business investment.

Measurement issues

Finally, the debate had also raised the possibility that the productivity slowdown might be a reflection of increasing mismeasurement of the gains from innovation in IT-related goods and services.6 However analysis for the United States (Byrne, Fernald and Reinsdorf 2016; and Syverson, 2016) suggests that this is highly unlikely (see also Derviş and Qureshi (2016) for an overview). Given that IT producing sectors have seen rising import penetration and most of the IT production is now done outside the United States, the effect (either way) would be small and in no way large enough to explain the slowdown observed in the United States. In fact, “improving” measurement would, if anything, make the slowdown more pronounced to the extent that US domestic production of these products has fallen over the 1995-2004 period. Furthermore, mismeasurement of IT hardware was also an issue prior to the slowdown. Finally, the largest benefits of recent innovations in ICT go to consumers in non-market production activities which again would not show up in GDP measures. In fact, Syverson (2016) shows that the slowdown is not correlated with IT production or use.

Our contribution

Following on the steps of our 2015 work (Andrews, Criscuolo and Gal, 2015), this chapter aims to bring some new firm level evidence to shed new light on the possible mechanisms underlying on the global productivity slowdown, including the rise of winner takes all dynamics and the slowdown of knowledge diffusion. While the Gordon-Brynjolfsson controversy (see endnote 1) is essentially a debate about prospects at the global productivity frontier, it is remarkable how little is actually known about the characteristics and performance of firms that operate at the global frontier. In this regard, we provide new evidence that highlights the importance of separately considering what happens to innovation at the frontier as well as the diffusion of new and unexploited existing technologies to laggard firms. This micro evidence is both key to motivating new theoretical work and to shifting the debate to areas where there may be more traction for policy reforms to revive productivity performance in OECD countries.

We show that a particularly striking feature of the productivity slowdown is not so much a lower productivity growth at the global frontier, but rather rising labour productivity at the global frontier coupled with an increasing labour productivity divergence between the global frontier and laggard firms.7 This productivity divergence remains after controlling for differences in capital deepening and mark-up behaviour although there is evidence that market power of frontier firms has increased in services. This leads us to suspect that the rising MFPR gap between global frontier and laggard firms may in fact reflect technological divergence.

MFP divergence could plausibly reflect the potential for structural changes in the global economy – namely digitalisation, globalisation and the rising importance of tacit knowledge – to fuel rapid productivity gains at the global frontier. Yet, aggregate MFP performance was significantly weaker in industries where MFP divergence was more pronounced, suggesting that the divergence observed is not solely driven by frontier firm pushing the boundary outward. In this regard, we contend that increasing MFP divergence – and the global productivity slowdown more generally – could reflect a slowdown in the technological diffusion process. This stagnation could reflect increasing costs for laggards firms of moving from an economy based on production to one based on ideas. But it could also be symptomatic of rising entry barriers and a decline in the contestability of markets. In both cases, public policy can play an important role in “alleviating” the productivity slowdown. Consistent with this, we find the rise in MFP divergence to be much more extreme in sectors where pro-competitive product market reforms were least extensive, suggesting that the observed rise in MFP divergence might be at least partly due to policy weakness stifling diffusion and adoption in OECD economies.

Evidence of technological divergence at the firm level is significant in light of aggregate level analysis suggesting that while adoption lags for new technologies across countries have fallen over time, long-run penetration rates once technologies are adopted diverge across countries, with important implications for cross-country income differences (Comin and Mestieri, 2013). More specifically, new technologies developed at the global frontier are spreading more and more rapidly across countries but their diffusion to all firms within any economy is slower and slower, with many available technologies remaining unexploited by a non-trivial share of firms. A key implication is that weak productivity performance in OECD countries may persist unless a new wave of structural reforms can revive a broken diffusion machine.8

Data and identification of the global productivity frontier

Copy link to Data and identification of the global productivity frontier

This chapter uses a harmonised firm-level productivity database, based on underlying data from the OECD-Orbis database (see Gal, 2013). The database contains several productivity measures (variants of labour productivity and multi-factor productivity, MFP) and covers to 24 OECD countries9 over the period 1997 to 2014 for the non-farm, non-financial business sector.10

As discussed in Gal (2013), these data are sourced from annual balance sheet and income statements, collected by Bureau van Dijk – an electronic publishing firm – using a variety of underlying sources ranging from credit rating agencies (Cerved in Italy) to national banks (National Bank of Belgium for Belgium) as well as financial information providers (Thomson Reuters for the United States).11 It is the largest available cross-country company-level database for economic and financial research. However, since the information is primarily collected for use in the private sector typically with the aim of financial benchmarking, a number of steps need to be undertaken before the data can be used for economic analysis. The steps we apply closely follow suggestions by Kalemli-Ozcan et al. (2015) and previous OECD experience (Gal, 2013; Ribeiro, Menghinello and Backer, 2010).12 Three broad steps are: i) ensuring comparability of monetary variables across countries and over time (industry-level purchasing power parity [PPP] conversion and deflation); ii) deriving new variables that will be used in the analysis (capital stock, productivity); and iii) keeping company accounts with valid and relevant information for our present purposes (filtering or cleaning). Finally, Orbis is a subsample of the universe of companies for most countries, retaining the larger and hence probably more productive firms. To mitigate problems arising from this – particularly the under-representation of small firms – we restrict our sample to firms with more than 20 employees on average over their observed lifespan.

Further, several issues that commonly affect productivity measurement should be kept in mind. First, differences in the quality and utilisation of capital and labour inputs cannot be accounted for as the capital stock is measured in book values and labour input by the number of employees.13 Secondly, measuring outputs and inputs in internationally comparable price levels remains an important challenge.14 Finally, similar to most firm-level datasets, Orbis contains variables on outputs and inputs in nominal values and no additional separate information on firm-specific prices and quantities (i.e. we observe total sales of steel bars, but no information on tonnes of steel bars sold and price per ton), thus output is proxied by total revenues or total value added. Even though we deflate these output measures by country-industry-year level deflators (at the two-digit detail), differences in measured (revenue) productivity across firms within a given industry may still reflect both differences in technology as well as differences in market power.15 As described below, we attempt to correct our productivity measures for differences in market power by deriving firm- and time- specific mark-ups following De Loecker and Warzynski (2012).

Productivity measurement

As a starting point, we focus on labour productivity, which is calculated by dividing real value added (in United States 2005 PPP that vary by industry) by the number of employees. Using labour productivity has the advantage that it retains the largest set of observations, as it does not require the availability of measures for fixed assets or intermediate inputs (proxied by materials) potentially used for deriving multi-factor productivity (MFP). Our baseline MFP relies on a value-added based production function estimation with the number of employees and real capital as inputs. We employ the one-step generalised method of moments (GMM) estimation proposed by Wooldridge (2009), which mitigates the endogeneity problem of input choices by using material inputs as proxy variables for productivity and (twice) lagged values of labour as instruments. The production function is estimated separately for each two-digit industry but pooled across all countries, controlling for country and year fixed effects. This allows for inherent technological differences across industries, while at the same time ensures comparability of MFP levels across countries and over time by having a uniform labour and capital coefficient along these dimensions.

Correcting for mark-ups

To mitigate the limitations from not observing firm-level prices, we correct our revenue-based MFP measure by firm- and time-varying mark-ups. To do that, we apply the mark-up estimation methodology of De Loecker and Warzynski (2012). We introduce a notation for the “standard” MFP estimates as MFPR (denoting revenue-based productivity) and the mark-up corrected MFP estimates as ${M F P R}_{c}$ , and we define it for each firm i and year t as follows:

Equation 2.1

${M F P R}_{i t}^{c} = {M F P R}_{i t} - \log (μ_{i t})$

where the MFP values are measured in logs and $μ$ denotes the estimated mark-up. The ${M F P R}_{c}$ measure provides an estimate for productivity that is purged from mark-up variations and hence is not influenced by market power changes under the assumption that at least one input of production is fully flexible (e.g. labour or materials).16

The mark-up is derived from the supply-side approach originally proposed by Hall (1986) and more recently re-explored by De Loecker and Warzynski (2012). The approach computes mark-ups without needing assumptions about the demand function but only relying on available information on output and inputs, under the assumptions that at least one input is fully flexible and that firms minimise costs. Thus, the mark-up – defined as the ratio of the output price P over marginal cost MC – is derived from the first order condition of the plant’s cost minimisation problem with respect to the flexible input k as:

Equation 2.2

$μ_{i t} = \frac{P_{i t}}{{M C}_{i t}} = {O u t p u t E l a s t i c i t y}_{i k t} / {O u t p u t S h a r e}_{i k t}$

That is, the mark-up of firm i at time t can be computed as the ratio between the elasticity of output17 with respect to the flexible input k (estimated in a first step via the Wooldridge approach) and flexible input k shares in output (observed in the data). In our baseline specification, we use labour (as opposed to materials) as flexible input to ensure the largest coverage of countries in our baseline specification. Thus, mark-ups are calculated as the ratio between the estimated production function parameter for labour ${\hat{β}}_{L}^{j}$ in industry j where firm i operates and the “corrected” wage share $w s_{i t} = \frac{{W L}_{i t}}{\tilde{{V A}_{i t}}}$ :

Equation 2.3

$μ_{i t} = \frac{{\hat{β}}_{L}^{j}}{w s_{i t}}$

The labour coefficient ${\hat{β}}_{L}^{j}$ in the numerator is estimated using the GMM estimation method by Wooldridge (2009). The denominator $w s_{i t}$ is obtained by using a prediction of firm-level value added $\tilde{{V A}_{i t}}$ by a rich polynomial function of observable inputs in order to retain only the anticipated part of output developments.18 The rationale for using this correction is the assumption that firms do not observe unanticipated shocks to production when making optimal input decisions.

Given potential criticisms that labour input may not be fully flexible – especially in countries with rigid labour markets – we also calculated mark-ups using materials as the fully flexible input for a subset of 18 countries for which data are available. In that case, a gross output-based production function is estimated to obtain a coefficient for materials, again following Wooldridge (2009). The main result of a strong divergence in MFP is robust to these different choices.

As De Loecker and van Biesebroeck (2016: 25) note, the intuition behind this mark-up measure is as follows:

“Holding other inputs constant, a competitive firm will expand its use of [the flexible input, i.e. labour] until the revenue share equals the output elasticity [hence the mark-up measure would be 1]. […] If a firm does not increase [its flexible input use] all the way until equality holds but prefers to produce a lower quantity and raise the output price instead, it indicates the firm is able to exercise market power and charge a price above marginal cost.”

As noted in De Loecker and Warzynski (2012), the low demand in terms of additional assumptions of their approach and the lack of information on firm level prices bear some costs. Given that we do not observe firms’ physical output, the approach is only informative on the way mark-ups change over time (not their level) and in relative terms, i.e. on the correlation with firm characteristics (e.g. productivity, size, export status) rather than in absolute levels. Therefore, we will look at relative trends in mark-ups for frontier and laggard firms in what follows.

Measuring the productivity frontier

In line with standard practice in an admittedly thus far limited literature (see Andrews, Criscuolo and Gal, 2016) we define the global productivity frontier as the top 5% of firms in terms of productivity levels, within each industry and year. As there is a tendency in Orbis for the number of firms (with available data to calculate productivity) to expand over time, we slightly deviate from this practice in our preferred definition of frontier firms.19 One implication of the increasing coverage of Orbis over time is that smaller – and presumably less productive – firms get included in the frontier in the latter years of the sample. Thus, the evolution of the top 5% on the expanding Orbis sample could artificially underestimate average productivity at the frontier over time just as a reflection of the expanding sample. To avoid this, we calculate the 5% of firms per industry using a fixed (absolute) number of firms across time. This circumvents the expanding coverage problem but still allows for differences across industries in terms of their firm population, which is important given the heterogeneity of average firm size across industries.20 More specifically, frontier firms are identified using the top 5% of the median number of firms (across years), separately by each industry. This approach aims to capture as close as possible the top 5% of the typical population of firms. Using a MFPR-based productivity frontier definition, for example, results in a global frontier size of about 80 companies for the typical two-digit industry.21

Importantly, however, while the number of frontier firms is fixed, the set of frontier firms is allowed to change over time. This choice is necessary to ensure that when assessing the evolution of the frontier, we account for the phenomenon of turbulence at the top: some firms can become highly productive and push the frontier, while other, previously productive, businesses can lose their advantages and fall out of the frontier. This will not necessarily lead to a bias where the frontier becomes relatively more productive over time, however, since the composition of the laggard grouping is also allowed to change, the average productivity of this grouping could also in principle be boosted by the entry (exit) of more (less) productive firms.22 As discussed in the following, while there is churning at the frontier, this is largely concentrated amongst the top quintile of the productivity distribution.

Characteristics of frontier firms

Table 2.1 reports cross-sectional differences in average characteristics for global frontier firms relative to non-frontier firms along several measurable dimensions, focusing on the last year of our sample, 2013. Panel A reports these differences based on a labour productivity measure while Panel B does likewise use MFPR and Panel C using mark-up corrected MFPR. A few interesting facts emerge from the tables.

First, firms at the global productivity frontier are on average three to four times more productive than non-frontier firms.23 At first glance, these differences appear large but are to be expected given the widespread heterogeneity in firm productivity that is typically observed within narrowly defined sectors (Syverson, 2004).24A host of literature has focused on how such large differences in productivity can be sustained in equilibrium, given the expectation that market selection and the reallocation of resources would necessarily equalise them over the longer run. Supply-side explanations have typically emphasised factors related to technology shocks, management skill, R&D, or investment patterns (Bartelsman and Doms, 2000). The demand side also appear relevant, given evidence that imperfect product substitutability – due to geographical segmentation (i.e. transport costs), product differentiation (i.e. consumer preferences, branding/advertising) and intangible factors (customer-producer relationships) – can prevent industry customers from easily shifting purchases between industry producers (Syverson, 2004). The combination of demand and supply side imperfections can indeed lead to large and persistent differences in productivity levels across firms (Syverson, 2011). Note that most studies focus on within-country productivity dispersion, while our analysis pools together different countries, potentially further widening the productivity distribution.
Second, on average, global frontier firms have greater sales and are more capital intensive – as expected, more so for labour productivity.
Third, global frontier firms pay higher average wages, which ranges between $20,000 and $26,000 (in 2005 USD terms) depending on the measure. These differences might reflect the sorting of better workers into frontier firms (Card, Heining and Kline, 2013; Song et al., 2015) and the potential sharing of higher rents by frontier companies with their workers.
Fourth, in manufacturing, firms at the frontier in terms of MFP (MFPR and its mark-up corrected variant) have significantly higher employment size than laggards, in line with existing evidence that productivity is positively correlated with size of manufacturing firms. However, frontier firms do not employ a significantly larger number of employees in services for any of the productivity measures analysed.
Fifth, frontier firms are also shown to charge higher mark-ups in the case of labour productivity and MFPR, particularly in services. This could reflect weaker competition in the less tradable and more regulated services sector, which allows for larger market power differences across firms. However, when the frontier is defined based on mark-up corrected MFPR, frontier firms are found to charge lower mark-ups. This is consistent with the idea that the most productive firms can afford to charge lower prices and thus attract more demand. In particular, this is in line with the findings of Foster, Haltiwanger and Syverson (2008) using US firm level data on prices and quantities, who show that there is a strong negative relationship between measures of MFP based on physical output rather than revenues (and thus purged from markups) and firm level prices.25
Finally global frontier firms are also more likely to belong to a multinational group/conglomerate and patent more intensively than other firms (Andrews, Criscuolo and Gal, 2015).26

Table 2.1. Mean firm characteristics: frontier firms vs laggards
Copy link to Table 2.1. Mean firm characteristics: frontier firms vs laggards

Productivity divergence between the global frontier and laggard firms

Copy link to Productivity divergence between the global frontier and laggard firms

Labour productivity divergence

Figure 2.2 describes the evolution of labour productivity for firms at the global productivity frontier and non-frontier firms for the broadest possible sample of firms and years for which comparable data are available. In this exercise, the global frontier is defined as the top 5% of firms in terms of labour productivity levels within each two-digit industry and year (see the preceding discussion), while laggard firms refer to all other firms. In turn, the chart then shows how the unweighted average of log labour productivity across firms in these two groupings evolved over time, with the initial year – 2001 – indexed to 0 and separately for two broad sectors: manufacturing and business services.27

Between 2001 and 2013, firms at the global frontier have become relatively more productive, with their labour productivity increasing at an average annual rate of 2.8% in the manufacturing sector, compared to productivity gains of just 0.6% per annum for non-frontier firms (henceforth laggards). This pattern of divergence is even more pronounced in the market services sector, with the labour productivity at the global frontier growing at an average annual rate of 3.6%, compared to an average of just 0.4% for the group of laggards.28 These statistics will tend to understate the true gap between frontier and laggard firms as frontier firms will inflate the average industry level productivity, particularly when their weight in industry activity is large. Reassuringly industry level trends look very much in line with the picture obtained with information from Orbis.29

Digging deeper, two distinct time periods emerge, which are essentially punctuated by the global financial crisis. Between 2001 and 2007, labour productivity at the global frontier grew at a rapid rate of 4-5% per annum, significantly eclipsing the growth of non-frontier productivity which averaged roughly 1% per annum.30 From 2008 onwards, labour productivity growth at the global frontier slowed to around 1% per annum, while the growth of non-frontier labour productivity ground to a halt. Reflecting these patterns, around three-quarters of the labour productivity gap between frontier and other firms that had accumulated by 2013 had been realised by 2007.

These broad patterns are robust to: i) using turnover-based labour productivity; ii) defining the global frontier in terms of top 100 firms or top 10% of firms; iii) taking median labour productivity in the frontier and non-frontier firms groupings as opposed to average productivity; iv) excluding from the sample firms that are part of a multi-national group (i.e. headquarters or subsidiaries) where profit-shifting activity may be relevant and v) using a smoothed definition of productivity by taking 3-year moving averages. The patterns in Figure 2.2 and subsequent Figures in this section are robust to using more narrowly-defined industries (i.e. 3 and 4-digit industry classifications) to better ensure that firms are competing in the same market and producing similar products.31 In some instances, however, this leads to a non-trivial reduction in the number of firms within each sector – raising difficulties for production function estimation and increasing the prevalence of idiosyncratic and noisy patterns. This leads us to conduct our baseline analysis at the two-digit level.32

Figure 2.2. A widening labour productivity gap between global frontier firms and other firms
Copy link to Figure 2.2. A widening labour productivity gap between global frontier firms and other firms

Labour productivity: value added per worker (2001-2013)

Labour productivity divergence: capital deepening or MFP?

Since gains in labour productivity at the firm level can be achieved through either higher capital deepening or multi-factor productivity (MFP), Figure 2.3 plots the evolution of these two components for global frontier and non-frontier firms, using the same definition of the global frontier as in Figure 2.2. Given that the sample of firms for which reliable capital stocks are available is smaller than the baseline sample used for the previous analysis, Figure 2.3, Panel A reproduces the evolution of labour productivity for global frontier and non-frontier firms in this smaller sample of firms, which broadly confirms the labour productivity divergence illustrated in Figure 2.2.

Figure 2.3. The widening labour productivity gap is mainly driven by MFP divergence
Copy link to Figure 2.3. The widening labour productivity gap is mainly driven by MFP divergence

From a comparison of Panels B and C in Figure 2.3, it is evident that the rising labour productivity gap between global frontier and non-frontier firms in the manufacturing sector entirely reflects divergence in revenue-based MFP (MFPR), while capital deepening of non-frontier firms slightly outpaces that of global frontier firms over the sample period.33 For the market services sector, there is evidence of divergence of both MFPR and capital deepening between global frontier and non-frontier firms, although labour productivity divergence in the pre-crisis period appears to be more strongly related to MFPR than capital deepening. Even so, vintage capital models imply that weak capital deepening amongst laggards in the post-crisis period could exacerbate MFP divergence if new technology is embodied in new capital which often requires a retooling process in existing plants (see Cooper, Haltiwanger and Power, 1997).

Since our capital measure is based on balance sheet information, it misses some important elements of intangible investments such as brand-building, worker training, the development of organisational practices and some types of R&D spending (Corrado, Hulten and Sichel, 2009). To the extent that the most productive businesses implement more and more of these type of investments, and at a faster pace than other firms, this may contribute to a widening gap in measured MFP.34 Accordingly, our subsequent discussion on the likely drivers of MFP divergence explicitly acknowledges that measured MFP reflects these and other factors beyond narrowly defined technology or technical efficiency – such as management practices, the qualities of employees or tacit knowledge more generally (Hulten, 2001).

MFP divergence: mark-ups or technology?

While evidence of divergence in MFP points towards a technological explanation of the rising labour productivity gap between global frontier and other firms, it might also reflect the increasing market power of frontier firms, given that our measure of multifactor productivity MFPR is based on information on revenues. If the increasing differences in MFPR between frontier and laggards reflect unobserved differences in firm level prices, the rising gap between global frontier and other firms in MFPR may simply reflect the increasing ability of frontier firms to charge higher mark-ups, and thus profitability as opposed to differences in technical efficiency. Accordingly, we attempt to assess the contribution of mark-up behaviour to MFPR divergence, following the methodology described in the section on data and identification of the productivity frontier.

Given the focus on MFPR, the global frontier in Figure 2.4 is redefined in terms of the top 5% of firms in terms of MFPR levels within each two-digit industry and year. Using such a definition, the divergence of MFPR in Figure 2.4, Panel A is very similar to that in Figure 2.3, Panel B, which defines the global frontier in terms of labour productivity. These patterns are robust to using alternative definitions of MFPR, based on a Solow residual or the Wooldridge gross-output estimation approach and to using materials (a proxy for intermediate inputs) as the fully flexible input in De Loecker and Warzynski (2012) methodology for a subset of 18 countries for which data are available. Similarly to labour productivity, MFPR divergence is also robust to using a smoothed variant by taking three-year moving averages.

Figure 2.4. The widening MFP gap remains after controlling for mark-ups
Copy link to Figure 2.4. The widening MFP gap remains after controlling for mark-ups

Figure 2.4, Panel B plots the evolution of the unweighted average of the estimated mark-ups for global frontier and non-frontier firms. While estimates are quite volatile, the divergence in MFPR in the manufacturing sector does not appear to be driven by frontier firms charging increasingly higher mark-ups, relative to non-frontier firms. Turning to the services sector, there is evidence that frontier firms increased their mark-ups relative to non-frontier firms in the pre-crisis period, in particular after 2005, but this divergence in mark-up behaviour is significantly unwound in the post-crisis period. Still, their mark-up levels are significantly higher than those of non-frontier firms (Table 2.1, Panel B). Once we correct MFPR for these patterns in mark-ups, the divergence in mark-up corrected MFPR between frontier and non-frontier firms in the pre-crisis is reduced by a factor of about one-third, while the divergence becomes somewhat larger in more recent years (Figure 2.4, Panel C).

For completeness, Figure 2.5 presents the evolution of mark-up corrected MFP where the global frontier is now defined as the top 5% of firms in terms of corrected MFPR levels within each two-digit industry and year. Taken together, the evidence in Figure 2.4 and Figure 2.5 implies that even though rising mark-ups for frontier firms play a non-trivial role in services, divergence seems mainly unrelated to the evolution of mark-ups. This suggests that the divergence is likely to be related to growing differences in the capacity of frontier firms vs laggards to invest in and successfully combine technological and non-technological innovations (intangibles), and the concomitant increasing importance of tacit knowledge (e.g. organisational know-how) for succeeding in the market.

Figure 2.5. A widening gap in mark-up corrected MFPR
Copy link to Figure 2.5. A widening gap in mark-up corrected MFPR

Global frontier defined in terms of mark-up corrected MFPR

MFP divergence: contributing factors

The evidence presented so far suggests that the MFP gap between the global frontier and other firms has risen significantly over time and that this pattern has emerged even before the crisis. This might be surprising as it is at odds with neo-Schumpeterian growth theory (Aghion and Howitt, 2006; Acemoglu, Aghion and Zilibotti, 2006), models of competitive diffusion (Jovanovic and MacDonald, 1994) and of growth enhancing creative destruction (Aghion and Howitt, 1992; Caballero and Hammour, 1994; Campbell, 1998), but as shown below, might be the result of structural changes in the global economy. To the extent that these developments reflect digitalisation, globalisation and the rising importance of tacit knowledge, MFP divergence could either reflect “winner takes all” dynamics propelling productivity growth at the global frontier or stalling technological diffusion and market dynamism inhibiting productivity gains amongst laggard firms. In practice, it is difficult to distinguish with the data at hand the relative importance and causality of these factors, but a number of smoking guns emerge to suggest that each may be relevant.

Frontier firms and winner takes all dynamics

The productivity divergence patterns unveiled so far may partly reflect the increasing potential for digital technologies to unleash winner takes all dynamics (Brynjolfsson and McAfee, 2012), which allows the technological leaders to increase their MFP gap with laggard firms. More specifically, by making the replication of informational goods and business processes at near zero marginal cost possible, digital technologies enable the top-quality provider to capture most, or all, of the market, while only small market shares accrue to the next-best providers (even if they are almost as good as the best provider). These patterns are reinforced by network externalities that favour the emergence of a single dominant player (e.g. providing a specific network; platform or standard) vis-à-vis other firms, even though their products are not necessarily inferior. At the same time, given the global nature of frontier firms, these patterns are likely to be reinforced by globalisation, which increases the returns to investing in non-rival technologies via expanded market size (Acemoglu and Linn, 2004).35

While it is hard to think of a single statistic that could capture winner takes all dynamics with the data at hand, a number of findings may support the existence of such dynamics:36

Divergence in MFPR is accompanied by divergence in sales between frontier and laggard firms, particularly in ICT-intensive services such as computer programming, data storage or cloud services.
- The divergence in sales has been growing over time: global frontier firms have gained significant market share relative to laggards in manufacturing and to a larger extent in services.37
Divergence in sales is particularly stark in ICT intensive services (Figure 2.6, Panel A), compared to non-ICT-intensive services (Figure 2.6, Panel B). This divergence is also apparent within the global frontier grouping: sales of firms in the top 2% of the global MFPR distribution grew by 14% on average in ICT intensive services over the sample period, compared to 7% in non-ICT-intensive services. In comparison, sales on average across firms in the top 5% of the global MFPR distribution grew by 6% and 3.5% in ICT-intensive services and non-ICT-intensive services respectively.
- One concern is that firms that make it to the top might gain too large of a market share, making entry into the frontier more difficult and more generally leading to lower competition in the market.38 This may entail a cost to aggregate productivity performance, above and beyond any gains associated with higher allocative efficiency.
A more pronounced MFP divergence in ICT-intensive services between frontier and laggard firms as well as within the global frontier grouping.
- Within the global frontier grouping, we see that a small cadre of the most elite firms (top 2%) become more productive relative to other frontier firms in ICT-intensive services (Figure 2.66, Panel C), while this pattern is not particularly evident within non-ICT-intensive services (Figure 2.66, Panel D).39
A divergence in mark-up corrected MFPR, despite developments in the mark-up.

These patterns are most easily detected in the sectors that are the most intensive users of ICT and in fact, are themselves, also providers of ICT services. However, the intensive use of ICT by some firms in other sectors could also contribute to a growing productivity gap between them and other firms. Moreover, the increased management and communications capabilities that ICT bring can potentially enable them to outsource a large share of their inputs. In extreme cases, outsourcing could even be related to the core activity and the core inputs of the firm, be it labour (e.g. Uber with its contracted taxi drivers) or capital (e.g. Airbnb with its rooms on offer).

Figure 2.6. Evidence on winner takes all dynamics
Copy link to Figure 2.6. Evidence on winner takes all dynamics

Laggard firms, stalling technological diffusion and market dynamism

The rising gap in MFPR between frontier and laggard firms might also signal stalling technological diffusion and market dynamism amongst laggards. This stagnation could reflect the increasing costs for laggard firms of moving from an economy based on production to one based on ideas. This would be the case if the strength of global frontier firms not only reflects their capacity to technologically innovate but also to optimally combine intangibles, i.e. technological, organisational, and human capital, in production processes. Indeed, the importance of tacit knowledge as a source of competitive advantage for frontier firms may have risen if increasingly complex technologies were to increase the amount and sophistication of complementary investments required for technological adoption.40 But it could also be symptomatic of rising entry barriers and a decline in the contestability of markets, which could reflect the inability of the policy environment to adapt to structural changes in the global economy and the rising market power of frontier firms, particularly in services. Both factors could act as a barrier to the catch-up of laggard firms and cause the technological diffusion machine – which sustained productivity growth in the OECD between 1950-1995 (OECD, 2015) – to break down.

To more robustly test whether the pace of technological convergence has slowed over time, we estimate a neo-Schumpeterian model where firm level MFP growth depends principally on a firm’s lagged MFP gap with the global frontier controlling for a battery of fixed effects (i.e. industry and country*year), firm size and age. The results suggest that on average across time, firms further behind the technological frontier have higher MFP growth, reflecting their ability to catch-up based on the adoption of a larger stock of unexploited technologies. As Figure 2.7 demonstrates, however, the pace of productivity convergence via this mechanism has declined significantly over time. For example, the estimated coefficient on the lagged MFPR gap term declined by almost 30% from the late 1990s to the most recent period, with most of this decline realised by 2007 (Panel A).41 This pattern holds within firm size and age classes and is even more pronounced when the model is estimated using mark-up corrected MFPR (Panel B), providing further evidence that the pace of technological convergence has slowed.

Figure 2.7. The pace of convergence slowed, even before the crisis
Copy link to Figure 2.7. The pace of convergence slowed, even before the crisis

Estimated convergence parameter from neo-Schumpeterian model and 95% confidence intervals

One symptom of stalling diffusion could be the increasing persistence of incumbents at the frontier or churning that increasingly comes from firms close to the frontier (i.e. within the top decile or top quintile of the MFP distribution). We might also expect these patterns to be especially evident in the services sector where intangibles and tacit knowledge are becoming ever more important and where the increase in market power at the frontier is most apparent.

Figure 2.8 provides supporting evidence for these conjectures. On average over 2001-2003, 50% of firms at the global frontier in terms of MFPR in the services sector (Figure 2.88, Panel A) were either classified two years earlier as frontier firms (i.e. 33% of firms where in the top 5%) or resided outside the frontier grouping but were in the top decile (10% of firms) or top quintile (7% of firms). By 2011-2013, however, this figure had risen to 63%, driven by a significant increase in the proportion of incumbent firms retaining their position in the frontier (43%) and a more modest increase in entry to the frontier from firms residing just outside the frontier but in the top decile (13%) some two years early. These patterns – which are also evident for corrected MFPR (Figure 2.88, Panel B) – suggest that it has become more difficult for laggard firms outside the top quintile of the MFP distribution to enter the global productivity frontier over time.

Figure 2.8. Entry into the global frontier has become more entrenched amongst top quintile firms
Copy link to Figure 2.8. Entry into the global frontier has become more entrenched amongst top quintile firms

Proportion of frontier firms in time t according to their frontier status in *t-2*

Rising entrenchment at the frontier is consistent with the broader decline in business dynamism observed across OECD countries (Decker et al., 2014 for the United States and Criscuolo, Gal and Menon, 2014 for 18 countries), which in turn raises the prospect that the degree of competitive pressure – a key driver of technological diffusion (see upcoming results) – may have declined. To explore the role of market dynamism among laggard firms, Figure 2.9 distinguishes between four groups of firms: i) young firms (aged zero-five years) to proxy for recent entrants; ii) mature firms (aged six to ten years); iii) firms teetering on the brink of exit in a competitive market, proxied by firms older than ten years that record negative profits over at least two consecutive years (non-viable old firms); and iv) all other firms (i.e. viable old firms; the excluded category). Two key patterns emerge. First, the data suggest that firm turnover has fallen, as reflected by a decline in the share of young firms and a higher survival probability of marginal firms that would typically exit in a competitive market (Panel A).42 Second, the average productivity of recent entrants relative to viable incumbent firms has risen, while the average productivity of firms on the margin of exit has fallen over time (Panel B).

These patterns are consistent with a decline in the contestability of markets, which implies less indirect pressure on incumbent firms to improve their productivity via the adoption of superior technologies and business practices (Bartelsman, Haltiwanger and Scarpetta, 2004).43 The corollary is that it has become relatively easier for weak firms that do not adopt the latest technologies to remain in the market. Moreover, the decline in firm turnover coupled with an increase in the implied productivity gap between entering and exiting businesses is what one would typically observe if barriers to entry had risen (Bartelsman, Haltiwanger and Scarpetta, 2013). This leads us to suspect that there may be more to the stagnation of laggard firm productivity than just the rising importance of tacit knowledge, thus motivating an analysis of the link between product market regulations and MFP divergence in the next section. Taken together, the estimated decline in convergence, the entrenchment of the frontier and the decline in business dynamism, provide indications on the potential drivers behind the divergence observed in the data.

Figure 2.9. Indicators of declining market dynamism amongst laggard firms
Copy link to Figure 2.9. Indicators of declining market dynamism amongst laggard firms

Frequency and productivity of firms by age and financial viability

Productivity divergence: aggregate implications and the role of policy

Copy link to Productivity divergence: aggregate implications and the role of policy

Before proceeding, it is important to recognise that some degree of MFP divergence across firms is organic to the working of a market economy – especially during periods of rapid technological change where experimentation looms large. Indeed, while MFP divergence may have contributed to the slowdown in aggregate productivity if technological adoption became increasingly difficult for laggard firms, the aggregate consequences are less clear if MFP divergence reflects higher innovation intensity at the frontier and more efficient resource allocation associated with the rapid gains in market shares of the globally most productive firms – a potentially positive aggregate consequence of winner takes all dynamics. Accordingly, this section begins by presenting evidence which suggests that aggregate MFP performance was weaker in industries where MFP divergence was more pronounced, suggesting that MFP divergence is a relevant policy concern for growth.

As we have seen, the increase in the MFP gap is not uniform across sectors. Productivity divergence is particularly apparent in-service sectors that are typically more sheltered from competitive pressures due to lower exposure to international competition and more stringent product market regulations. This generates two sets of potentially relevant policy issues: To what extent is divergence itself creating barriers to technology diffusion by stifling competitive forces? And to what extent is the failure of policy to encourage competition in service sectors contributing to maintain such barriers? We present evidence in the remainder of this section which suggests that the rise in the MFP gap was less pronounced in sectors where the pace of product market reform was more intense, suggesting scope for public policy to “lean against the wind” of rising MFP divergence.

Aggregate implications

To provide suggestive evidence on a link between MFP divergence and weakness in aggregate productivity performance, Figure 2.10 relates aggregate industry-level MFP (sourced from the EU KLEMS sectoral-level database) to the MFPR gap between frontier and laggard firms. The data are collapsed to the industry-year level by taking an unweighted average of both variables across countries. Moreover, each variable is purged of industry fixed effects and year fixed effects to facilitate a within-industry interpretation and to abstract from time-varying global shocks. A robust negative relationship emerges, whereby above-average MFP divergence between frontier and laggard firms within industries is associated with below-average aggregate MFP performance. Moreover, the coefficient estimates imply that a 0.3 rise in logs (around 35%) in the MFPR gap – roughly equivalent to that observed between 2001 and 2007 (Figure 2.4) – is associated with a 3.5% decline in the level of aggregate MFP across industries. This is economically significant given that the cumulative loss in MFP due to the productivity slowdown in OECD countries over this period amounts to about 6%.44

Figure 2.10. Aggregate MFP performance was weaker in industries where MFPR divergence was greater
Copy link to Figure 2.10. Aggregate MFP performance was weaker in industries where MFPR divergence was greater

Residual aggregate MFP and the MFPR gap at the industry level; 1998-2007

This negative relationship between aggregate MFP and MFPR divergence is statistically significant and holds within both manufacturing and services.45 While this analysis is restricted to the pre-crisis period due to the lack of more recent industry-level MFP data, a negative relationship is also evident between aggregate labour productivity and labour productivity divergence over 1998-2013, although this relationship is mainly driven by the manufacturing sector.46 In any case, even with a growing market share accrued to frontier firms, aggregate productivity is not benefiting positively from a widening gap between the frontier and laggards. This illustrates that winner takes all dynamics discussed in the previous section do not necessarily translate into aggregate gains, because they tend to imply poor productivity performance of laggard firms.

Productivity divergence and product market regulation

At the margin, there are several channels through which pro-competitive product market reforms can strengthen the incentives for laggard firms to adopt frontier technologies, thereby moderating – but not necessarily reversing – the pressures towards higher MFP divergence induced by technological change. Indeed, a range of firm-level evidence generally supports the idea that competitive pressures are a driver of productivity-enhancing innovation and adoption.47 More specifically, pro-competition reforms in product markets could be expected to promote the catch-up of laggard firms to the global frontier for a number of reasons:

First, higher competition underpins within-firm productivity gains by weeding out inefficient firms and sharpening the incentives for incumbent laggard firms to adopt better technologies and business practices (Bloom, Draca and Van Reenen, 2015; Perla, Tonetti and Waugh, 2015; Steinwender, 2015; Baily, 1993; Baily et al., 2005).
Second, stronger product market competition can improve managerial quality (Bloom and Van Reenen, 2010), which is complementary to technological adoption (Bloom et al., 2012).
Third, reductions in administrative entry barriers can spur entry, which promotes technological diffusion to the extent that young firms possess a comparative advantage in commercialising cutting-edge technologies (Henderson, 1993; Baumol, 2002).
Fourth, pro-competitive reforms to market regulations in upstream services sectors may increase the returns expected by firms in downstream manufacturing sectors from adopting best-practice techniques (Bourlès et al., 2013).
Finally, product market reforms can promote productivity-enhancing reallocation (Andrews and Cingano, 2014), thereby enhancing the ability of firms to attract inputs complementary to technological adoption and achieve sufficient scale to enter global markets and learn from global frontier firms.

Product market reforms in OECD countries

To measure market reforms, we utilise the OECD database on product market regulations, which is based on a highly detailed questionnaire sent out to governments every five years (Nicoletti and Scarpetta, 2003).48 These indicators measure the extent of “anti-competitive” regulations; that is, regulations “that inhibit competition in markets where competition is viable”.49 A key strength of these indicators is their de jure nature – i.e. they focus on rules and regulations – which facilitates meaningful cross-country comparisons, but it is important to note that they do not account for differences in implementation and enforcement across countries. The indicators range from 0 to 6 and are increasing in the restrictiveness of product market regulations.

We exploit information on sector-specific provisions in ten separate industries (seven in network industries, one in retail, and two in professional services; see Conway and Nicoletti, 2006), as opposed to the economy-wide indicators.50 As discussed in Gal and Hijzen (2016), regulation in: i) network industries is largely about the organisation of network access to potential service providers; ii) retail trade typically takes the form of entry barriers, specific restrictions for large firms and the flexibility of shops in terms of opening hours and prices; iii) professional services concerns to barriers to entry and the way services are delivered and includes, amongst others, rules governing the recognition of qualifications and the determination of fees and prices.

According to the OECD indicators, there is considerable scope for further product market reform in many OECD countries, particularly in market services where the increase in MFP divergence has been most striking. Within non-manufacturing industries, most reform activity over the past 15 years has been concentrated in network industries (i.e. energy, transport and communication), and this is reflected in both a decline in the median level and dispersion of market regulation across countries (Figure 2.11, Panel A). While there remains some scope for further reform action in specific network industries (particularly road and rail transportation) and countries, the need for reforms in retail trade and professional services is clear. Between 1998 and 2013, the median restrictiveness of product market regulations only fell modestly in retail (Panel B) and was little changed in professional services (Panel C), while the dispersion in the restrictiveness of market regulations across countries in these sectors remains high. If product market regulations affect the incentives for laggard firms to adopt leading technologies and best practices, to what extent is the rising MFP gap between frontier and laggard firms in services a product of the slow pace of market reforms in these industries?

Figure 2.11. The restrictiveness of product market regulations over time, 1998-2013
Copy link to Figure 2.11. The restrictiveness of product market regulations over time, 1998-2013

By way of introduction and purely for illustrative purposes, Figure 2.12 provides some preliminary evidence on the link between the pace of market reforms and the evolution of the MFPR gap between global frontier and laggard firms in three selected services industries. As it turns out, the MFP gap increased more quickly in professional service industries such as legal, accounting and technical services (engineering and architecture) where the pace of reform lagged, compared to network industries such as telecommunications where the pace of market reform has been much more intensive. While these patterns are consistent with the idea that pro-competitive market reforms in services can sharpen the incentives for technological adoption, it is important to control for a number of potentially omitted country, industry and global factors to establish a more robust link between regulations and the MFP gap. Herein lies the aim of the next section.

Figure 2.12. Slower product market reform, a larger increase in the MFP gap
Copy link to Figure 2.12. Slower product market reform, a larger increase in the MFP gap

Selected industries; annual average change over time and across countries

Empirical strategy

Baseline model

To more rigorously explore the link between product markets regulations and the MFP gap between global frontier firms and other firms over time, we estimate two complementary econometric specifications.51

First, for ten market services sectors for which regulatory indicators are available over the period 1998-2013, we estimate the following long difference specification:

Equation 2.4

where: Δ^ld denotes the long difference operator, corresponding to five years in the baseline specification;52 MFPgap_s,c,t refers to the difference between the (unweighted) average MFP (MFPR or mark-up corrected MFPR) of global frontier firms and the (unweighted) average MFP of laggard firms in country c, industry s and year t; PMR_s,c,t refers to the overall restrictiveness of product market regulation in key service industries (expressed in log terms)53, which is increasing in the degree of regulation. If β₁>0, then it implies that a slowdown in the pace of pro-competitive product market reforms (i.e. a less negative oecd/components/embed/v1/embed/iframeoecd/components/descriptive-embed-composite/v1/descriptive-embed-composite/embednt:unstructuredembeddable500images/ca7295d9/embeddings/oleObject3.bin term) is associated with a rising MFP gap between the global frontier and non-frontier firms.

The regression also includes the growth of sectoral employment (E) to control for time-varying shocks within country*industry pairings and for changes in the coverage of the dataset over time.54 The baseline model includes separate country, industry and year fixed effects to control for omitted time-invariant country (δ_c) or industry (δ_s) factors and global shocks (δ_t), but as an extension, we include interacted country-year fixed effects (δ_ct) to control for time-varying country-specific shocks. To maximise the use of the data, we rely on overlapping five-year differences (e.g. 2013-2008, 2012-2007 etc.) but given that we cluster at the country-industry pair level this is innocuous (Bloom, Draca and Van Reenen, 2015)2015). Finally, country-industry-year cells that contain less than 10 firms are excluded in order to reduce the influence of highly idiosyncratic firm-level developments.

Second, we estimate a dynamic ordinary least square (see Stock and Watson, 1993; DOLS) equation in Equation 2.5, on the same sample of ten market services sectors over the period 1998-2013. This specification relates the level of the MFP gap to the (log) level of PMR and lags and leads changes (i.e. first differences) of the explanatory variables (ΔX) – the latter are included to control for serial correlation and endogeneity.55 It also includes a rich battery of fixed effects to control for unobserved time-varying country-specific shocks (δ_ct), time-varying global industry-specific shocks (δ_st) and time invariant industry-specific factors within countries (δ_cs).

Equation 2.5

Addressing identification concerns

One identification concern is that rigid services regulation might be a consequence, not a cause of the MFP gap (e.g. reverse causality). This would be the case if in service sectors with many firms lagging behind the global productivity frontier, there was a greater incentive for firms to exert political pressures for raising anticompetitive regulations. While such lobby activity by inefficient firms would upwardly bias the estimate of β₁, the tendency for product market reforms to be conducted when economic conditions are weak (Bouis, Duval and Eugster, 2016) would bias the estimate of β₁ in the opposite direction.

We adopt two identification strategies to confront the potential endogeneity of market regulation to economic conditions and at the industry level. First, we employ an instrumental variables (IV) approach in the context of our long difference framework, which exploits the existence of liberalisation waves across countries and the role of external pressure in driving them (see Bouis, Duval and Eugster, 2016).56 Here, we utilise two instruments which are unlikely to be affected by sector-level economic outcomes in the country considered, and should not have any effect on MFPgap_s_._c_._t other than through pressure on domestic authorities to undertake reform:

The lagged level of market regulation, based on the idea that the scope for reform as well as the push to implement reform is larger in country-sector pairs where the initial stance of product market regulation is stricter.
Average reform activity in other countries – as measured by the five-year change in product market regulation in the given sector – to capture peer pressure from reforms in other countries.

Second, if these baseline estimates are robust, then one might expect there to be a relationship between the MFP gap in manufacturing sectors and the extent of regulation in upstream service sectors in those instances where input-output connections are more intense. Accordingly, we estimate a variant of equation 2.5 for twenty-two industries in the manufacturing sector over the period 1998-2013 using the OECD regulatory impact indicator, which captures the knock-on or indirect effect of product market regulations in upstream services sectors (Bourlès et al., 2013; Conway and Nicoletti, 2006) on the MFP gap in downstream manufacturing industries. This is done by crossing the upstream regulatory indicators with the intensity of use of intermediate inputs calculated from input-output matrices since the impact of upstream regulations on downstream productivity is an increasing function of the intensity of use of intermediate inputs from the regulated upstream industries. While political economy factors may again be a source of bias – for example, if firms in downstream industries that use regulated (upstream) intermediate inputs, and whose productivity growth is low as a result, were to lobby for and obtain upstream deregulation – if anything, this would make it more difficult to find a positive relationship between upstream regulation and the MFP gap in downstream sectors (Bourlès et al., 2013).

Empirical results

Baseline results

Table 2.2, Panel A shows the baseline estimates for the five-year long difference specification (Equation 2.4) for the MFP gap based on MFPR and mark-up corrected MFP for the services sector. The odd numbered columns include separate country, industry and year fixed effects, while the even numbered columns include country-year fixed effects – which control for time varying country-specific shocks – and separate industry fixed effects. In each case, the change in the MFP gap is positively related to the change in PMR and the coefficient is statistically significant at the 1% level. Similarly, the DOLS estimates (Table 2.2, Panel B) – which provide an alternative estimate of the long-run relationship – suggest that higher PMR is associated with a larger MFP gap, with levels of statistical significance varying between 5 and 10%. The coefficient estimates in Column 1 of Table 2.2, Panel A imply that a one standard deviation increase in PMR is associated with about one-third of a standard deviation increase in the MFP gap.57

Table 2.2. MFP divergence and product market regulation in services
Copy link to Table 2.2. MFP divergence and product market regulation in services

One possible interpretation is that the rising MFP gap between frontier and non-frontier firms in market services could be related to a slowdown in the pace of pro-competitive reforms in product markets. That is, while the MFP gap between the global frontier and laggard firms increased over the sample period due to technological factors, this pattern of MFP divergence was much less pronounced in services sectors where the pro-competitive product market reform was more intensive.

To better illustrate the economic magnitude, Figure 2.13 performs a counterfactual simulation to estimate how much the MFP gap would have risen if market reforms in five key services sectors had proceeded at the same pace of that observed in telecommunications, where reform was most extensive. For example, the MFP gap increased at an annual average rate of 3.8% in legal and accounting services over the sample period, but our estimates imply that 1.7% of this increase might have been avoided if market liberalisation in this sector had proceeded more rapidly. On average across the sectors analysed, our estimates imply that up to 50% of the increase in MFP divergence might have been avoided if countries had engaged in extensive market liberalisation in services.

Figure 2.13. MFP divergence and market reforms in services
Copy link to Figure 2.13. MFP divergence and market reforms in services

Contribution to the annual change in the MFP gap of the slower pace of reform relative to the best practice industry

Sensitivity analysis

The baseline results are robust to using: i) an MFP gap that corrects for mark-ups (Table 2.2, columns 3-4); ii) alternative lengths of the long difference operator; iii) the median (rather than mean) productivity of laggards to construct the MFP gap; and iv) alternative lag and lead lengths in the DOLS estimator (Table 2.2, Panel B).58

Additional analysis suggests that it is unlikely that the baseline estimates are upwardly biased by the potential endogeneity of market reforms to industry-specific economic conditions:

A positive relationship between the change in the MFPR gap and the change in PMR is confirmed when the latter is instrumented using two plausibly exogenous measures of external reform pressure. The IV estimates are larger in magnitude than the baseline estimates although less precisely estimated (Table 2.2), suggesting that weak sectoral performance may trigger market reforms, as opposed to lobbying for anticompetitive regulation.
The IV regressions show a positive relationship between the mark-up corrected MFPR gap and PMR, although the imprecise estimation means that the coefficients are not always statistically significant. That PMR is more strongly related to MFPR gap than the corrected MFPR gap provides tentative evidence that market reforms may also operate through decreasing mark-up gaps between frontier and laggards.
We also checked the existence of a relationship between the MFP gap in manufacturing sectors and the extent of regulation in upstream service sectors in those instances where input-output connections are more intense. Accordingly, the DOLS estimates imply that higher PMR in upstream sectors is associated with a larger MFP gap in manufacturing, and this effect is statistically significant at the 1% level. These results raise the prospect that aggregate impact of services regulation of the MFP gap is likely to be somewhat larger than the direct estimates reported in Figure 2.13.

Conclusion

Copy link to Conclusion

In this chapter, we aim to contribute to the debate on the global productivity slowdown – which has by and large been conducted from a macroeconomic perspective – bringing it towards a more micro level. We provide new firm level evidence that highlights the importance of separately considering what happens to innovation at the frontier as well as the diffusion of technologies to laggards firms. This micro evidence is both key to motivating new theoretical work and to identifying areas where there may be more traction for policy reforms to revive productivity growth in OECD countries.

The most striking feature of the productivity slowdown is not so much a slowing in the rate of productivity growth at the global frontier, but rather rising productivity at the global frontier coupled with an increasing productivity divergence between the global frontier and laggard firms. This productivity divergence remains after controlling for differences in capital deepening and mark-up behaviour although there is evidence that market power of frontier firms has increased in services. This leads us to suspect that the rising MFPR gap between global frontier and laggard firms may in fact reflect technological divergence.

We show that this pattern of MFP divergence, which is at odds with existing models of new-Schumpeterian growth and of creative destruction, could plausibly reflect the potential for structural changes in the global economy – namely digitalisation, globalisation and the rising importance of tacit knowledge – to fuel rapid productivity gains at the global frontier. Yet, aggregate MFP performance was significantly weaker in industries where MFP divergence was more pronounced, suggesting that the divergence observed is not solely driven by frontier firm pushing the boundary outward. In this regard, we contend that increasing MFP divergence – and the global productivity slowdown more generally – could reflect a slowdown in the technological diffusion process. This stagnation could reflect increasing costs for laggards firms of moving from an economy based on production to one based on ideas. But it could also be symptomatic of rising entry barriers and a decline in the contestability of markets. Crucially, in both cases, there is scope for policy to alleviate the productivity slowdown.

Indeed, we find the rise in MFP divergence to be much more extreme in sectors where pro-competitive product market reforms were least extensive, suggesting that the observed rise in MFP divergence might be at least partly due to policy weakness stifling diffusion in OECD economies. A simple counterfactual exercise suggests that had the pace of product market reforms in retail trade and professional services been equivalent to that observed in the best practice service sector (i.e. telecommunications), then the extent of MFP divergence may have been up to 50% less than what was actually observed. Put differently, structural changes in the global economy meant that technological catch-up to the global productivity frontier became more difficult for the typical firm over the 2000s, but these difficulties were compounded by policy weakness. From this perspective, the opportunity cost of poorly designed product market regulations may have risen over time.

This analysis raises a number of issues for future research. First, it would be interesting to explore the impact of the crisis and macroeconomic policies on global frontier and laggard firms, via the channels identified in Anzoategui et al. (2016) and Gopinath et al. (2015). Second, a logical next step is to connect these findings with other cross-country productivity-related research at the OECD, which reveals declining business dynamism (Criscuolo, Gal and Menon, 2014), rising resource misallocation (Berlingieri, Blanchenay and Criscuolo, 2017) and weakening market selection, particularly the distorting effects on resource reallocation of the rise of “zombie” firms (Adalet McGowan, Andrews and Millot, 2016). Third, policies that can contain MFP divergence may carry a double-dividend for inclusiveness to the extent that the observed rise in wage inequality is closely related to the rising dispersion in average wages paid across firms (Card, Heining and Kline, 2013; Song et al., 2015).

The results of this analysis also suggest scope for policy relevant research in other areas. For example, to the extent that technological adoption is complementary to investments in organisational capital (Bloom, Sadun and Van Reenen, 2012), it would be interesting to explore the link between managerial quality and technological divergence. The same is true for intellectual property rights (IPR) regimes given their intuitive link with winner take all dynamics and opportunities for technological diffusion, although nuanced cross-country policy indicators of IPRs are currently lacking. Finally, given the increasing potential for entrenchment at the frontier, concerns about non-technological barriers to entry – such as lobbying activity by incumbents to prevent the proliferation of new business models and regulatory incumbency more generally– represent a fruitful area for future research.

References

Acemoglu, D., P. Aghion and F. Zilibotti (2006), “Distance to the Frontier, Selection and Economic Growth”, Journal of the European Economic Association, 4(1), pp. 37-74.

Acemoglu, D. and J. Linn (2004), “Market Size in Innovation: Theory and Evidence from the Pharmaceutical Industry”, The Quarterly Journal of Economics, 119, issue 3, p. 1049-1090.

Aghion, P., and P. Howitt (1992), “A Model of Growth Through Creative Destruction", Econometrica 60, No. 2: 323-351.

Aghion, P., R. Blundell, R. Griffith, P. Howitt, and S. Prantl (2004), ‘‘Entry and Productivity Growth: Evidence from Micro level Panel Data’’, Journal of the European Economic Association 2 (2004), 265–276.

Adalet McGowan, M., D. Andrews and V. Millot (2016), “The Walking Dead? Zombie Firms and Productivity Performance in OECD countries”, OECD Economics Department Working Papers, forthcoming.

Aghion, P. and P. Howitt (1998), Endogenous Growth Theory, MIT Press.

Aghion, P. and P. Howitt (2006), “Appropriate Growth Policy: A Unifying Framework”, Journal of the European Economic Association, 4(2-3), pp. 269-314.

Ahmad, N. and P. Schreyer (2016), "Measuring GDP in a Digitalised Economy", OECD Statistics Working Papers, No. 2016/07, OECD Publishing, Paris, http://dx.doi.org/10.1787/5jlwqd81d09r-en.

Andrews, D. and F. Cingano (2014), “Public Policy and Resource Allocation: Evidence from firms in OECD countries”, Economic Policy, Issue 74 April.

Andrews, D., C. Criscuolo and P. Gal (2015), “Frontier Firms, Technology Diffusion and Public Policy: Micro Evidence from OECD Countries”, OECD Productivity Working Papers, No. 2, OECD Publishing, Paris, https://doi.org/10.1787/5jrql2q2jj7b-en.

Anzoategui, D., D. Comin, M. Gertler and J. Martinez (2015), "Endogenous Technology Adoption and R&D as Sources of Business Cycle Persistence", NBER Working Paper, No. 22005.

Baily, M. N. (1993), “Competition, Regulation, and Efficiency in Service Industries”, Brookings Papers on Economic Activity, No. 2.

Baily, M. N., D. Farrell, E. Greenberg, J-D. Henrich, N. Jinjo, M. Jolles and J. Remes (2005), “Increasing Global Competition and Labor Productivity: Lessons from the US Automotive Industry”, McKinsey Global Institute, McKinsey & Company.

Bartelsman, E.J. and M. Doms (2000), “Understanding Productivity: Lessons from Longitudinal Microdata”, Journal of Economic Literature, 38(3), pp. 569-594.

Bartelsman, E., J. Haltiwanger and S. Scarpetta (2013), "Cross-Country Differences in Productivity: The Role of Allocation and Selection," American Economic Review, 103(1): 305-34.

Bartelsman, E., J. Haltiwanger and S. Scarpetta (2004), “Microeconomic Evidence of Creative Destruction in Industrial and Developing Countries”, SSRN, https://ssrn.com/abstract=612230.

Basu, Susanto, John G. Fernald, and Miles S. Kimball (2006), "Are Technology Improvements Contractionary?", American Economic Review, 96 (5): 1418–1448, http://dx.doi.org/10.1257/aer.96.5.1418.

Baumol, W.J. (2002), The Free-Market Innovation Machine: Analyzing the Growth Miracle of Capitalism, Princeton University Press.

Berlingieri, G., P. Blanchenay and C. Criscuolo (2017), “The great divergence(s)”, OECD Science, Technology and Industry Policy Papers, No. 39, OECD Publishing, Paris, https://doi.org/10.1787/953f3853-en.

Bloom, N., and J. Van Reenen (2010), "Why Do Management Practices Differ across Firms and Countries?", Journal of Economic Perspectives, 24 (1): 203–24, https://doi.org/10.1257/jep.24.1.203.

Bloom, N. and J. Van Reenen (2007), “Measuring and Explaining Management Practices Across Firms and Countries”, Quarterly Journal of Economics, 122 (4), pp. 1351-1408.

Bloom, N., M. Draca and J. Van Reenen (2015), “Trade Induced Technical Change? The Impact of Chinese Imports on Innovation, IT and Productivity”, Review of Economic Studies, 83(1), pages 87-117.

Bloom, N., C. Genakos, R. Sadun and J. Van Reenen (2012), “Management Practices Across Firms and Countries”, Academy of Management Perspectives, Vol. 26, No. 1, https://doi.org/10.5465/amp.2011.0077.

Blundell, Richard, Rachel Griffith, and John Van Reenen (1999), “Market Share, Market Value and Innovation in a Panel of British Manufacturing Firms”, Review of Economic Studies, 66, pp. 529–554.

Bouis, R., R. Duval and J. Eugster (2016), “Product Market Deregulation and Growth: New Country-Industry-Level Evidence”, IMF Working Paper, WP/16/114.

Bourlès, R., G. Cette, J. Lopez, J. Mairesse and G. Nicoletti (2013), "Do Product Market Regulations in Upstream Sectors Curb Productivity Growth? Panel Data Evidence for OECD Countries", The Review of Economics and Statistics, MIT Press, vol. 95(5), pages 1750-1768, December.

Brynjolfsson, E, and A. McAfee (2012), Race Against The Machine: How the Digital Revolution is Accelerating Innovation, Driving Productivity, and Irreversibly Transforming Employment and the Economy, Digital Frontier Press.

Byrne, D., J. Fernald and M. Reinsdorf (2016), “Does the United States have a productivity slowdown or a measurement problem?”, Brookings Papers on Economic Activity Conference Draft, March 10-11.

Caballero, R. J and M. L. Hammour (1994), "The Cleansing Effect of Recessions," American Economic Review, vol. 84(5), pages 1350-68, December.

Campbell, J. (1998), “Entry, Exit, Embodied Technology, and Business Cycles”, Review of Economic Dynamics, 1, (2), 371-408.

Card, D., J. Heining and P. Kline (2013), “Workplace Heterogeneity and the Rise of West German Wage Inequality”, Quarterly Journal of Economics, 128(3), pp. 967-1015.

Christiano, L.J., M.S. Eichenbaum and M. Trabandt (2015), “Understanding the Great Recession”, American Economic Journal: Macroeconomics, 7(1): 110–167.

Comin, D and M. Mestieri (2013), “If Technology Has Arrived Everywhere, Why Has Income Diverged?”, NBER Working Paper Series, No. 19010.

Conway, P. and G. Nicoletti (2006), “Product Market Regulation in the Non-Manufacturing Sectors of OECD Countries: Measurement and Highlights”, OECD Economics Department Working Papers, No. 530, OECD Publishing, Paris, https://doi.org/10.1787/362886816127.

Cooper, R., J. Haltiwanger and L. Power (1997). “Machine replacement and the business cycle: lumps and bumps,” NBER Working Paper, No. 5260.

Corrado, C., C. Hulten and D. Sichel (2009), "Intangible Capital And U.S. Economic Growth," Review of Income and Wealth, International Association for Research in Income and Wealth, vol. 55(3), pages 661-685, 09.

Cowen, T. (2011), The Great Stagnation: How America Ate All the Low-Hanging Fruit of Modern History, Got Sick, and Will (Eventually) Feel Better.

Criscuolo, C., P. Gal and C. Menon (2014), “The Dynamics of Employment Growth: New Evidence from 18 Countries”, OECD Science, Technology and Industry Policy Papers, No. 14, OECD Publishing, Paris, https://doi.org/10.1787/5jz417hj6hg6-en.

David, P. A. (1991), "Computer and Dynamo: The Modern Productivity Paradox in a Not‐Too‐Distant Mirror," Technology and Productivity: The Challenge for Economic Policy, Organization for Economic Co‐operation and Development, Paris, pp. 315‐48.

Decker, R., J. Haltiwanger, R.S. Jarmin, and J. Miranda (2016a), “Declining Business Dynamism: Implications for Productivity?”, Brookings Institution, Hutchins Center Working Paper.

Decker, R., J. Haltiwanger, R.S. Jarmin, and J. Miranda (2016b), “Where Has All the Skewness Gone? The Decline in High-Growth (Young) Firms in the U.S”, European Economic Review, 86 (July): 4-23.

Decker, R., J. Haltiwanger, R.S. Jarmin, and J. Miranda (2014), “The role of entrepreneurship in U.S. job creation and economic dynamism”, Journal of Economic Perspectives (28), pp. 3-24.

De Loecker, J. and J. v. Biesebroeck, 2016, “Effect of International Competition on Firm Productivity and Market Power”, NBER Working Paper, No. 21994.

De Loecker, J. and F. Warzynski, 2012. “Mark-ups and Firm-Level Export Status”, American Economic Review, Vol. 102, No. 6, October 2012 (pp. 2437-71).

Derviş K. and Z. Qureshi, 2016, “The Productivity Slump – fact or fiction: the measurement debate”, The Brookings Institution, Washington D.C.

Duval, R., D. Furceri, J. Jalles, and H. Nguyen, 2016, “A New Narrative Database of Product and Labour Market Reforms in Advanced Economies”, IMF Working Paper.

Fernald, J. (2014), “Productivity and Potential Output Before, During, and After the Great Recession”, NBER Macroeconomics Annual, Volume 29.

Forlani, E., R. Martin, G. Mion and M. Muûls (2016), “Unraveling Firms: Demand, Productivity and Mark-ups Heterogeneity”, EP Discussion Paper, No 1402.

Foster, L., J. Haltiwanger and C. J. Krizan (2001), “Aggregate Productivity Growth: Lessons from Microeconomic Evidence.” In New Developments in Productivity Analysis, Edward Dean, Michael Harper, and Charles Hulten, eds. Chicago: University of Chicago Press, 303-372.

Foster, L., J. Haltiwanger and C. Syverson (2008), “Reallocation, Firm Turnover, and Efficiency: Selection on Productivity or Profitability?”, American Economic Review, 98(1), pages 394-42.

Gabaix X. and A. Landier (2008), “Why has CEO Pay Increased So Much?”, The Quarterly Journal of Economics, Volume 123, Issue 1, February, Pages 49–100, https://doi.org/10.1162/qjec.2008.123.1.49.

Gal, P. (2013), “Measuring Total Factor Productivity at the Firm Level using OECD-ORBIS”, OECD Economics Department Working Papers, No. 1049, OECD Publishing, Paris, https://doi.org/10.1787/5k46dsb25ls6-en.

Gal, P. and A. Hijzen (2016), “The short-term impact of product market reforms: A cross-country firm-level analysis”, OECD Economics Department Working Papers, No. 1311, OECD Publishing, Paris, https://doi.org/10.1787/5jlv2jm07djl-en.

Gamberoni, E., C. Giordano and P. Lopez-Garcia (2016), “Capital and labour (mis)allocation in the euro area: some stylized facts and determinants”, ECB Working Paper, No. 1981, November.

Geroski, P (1995a), ‘‘What Do We Know about Entry?’’, International Journal of Industrial Organization 13, pp. 421–440.

Geroski, P (1995b), Market Structure, Corporate Performance and Innovative Activity, Oxford University Press.

Gopinath, G., S. Kalemli-Ozcan, L. Karabarbounis and C. Villegas-Sanchez (2015), “Capital Allocation and Productivity in South Europe”, NBER Working Paper, No. 21453.

Gordon, R. (2012), “Is U.S. Economic Growth Over? Faltering Innovation Confronts the Six Headwinds”, NBER Working Papers, No. 18315.

Griffith, R., S. Redding, and H. Simpson (2002), ‘‘Productivity Convergence and Foreign Ownership at the Establishment level,’’ CEPR discussion paper 3765 (2002).

Griliches, Z. (1957), “Hybrid Corn: An Exploration in the Economics of Technological Change”, Econometrica vol. 25 (October).

Hall, R. E. (1986), “Market structure and macroeconomic fluctuations”, Brookings Papers on Economic Activity, 2:285–338.

Haltiwanger, J. (2016), “Firm Dynamics and Productivity: TFPQ, TFPR and Demand Side Factors”, Economia, http://dx.doi.org/10.31389/eco.43.

Haskel, J., S. Pereira, and M. Slaughter (2007), ‘‘Does Inward Foreign Direct Investment Boost the Productivity of Domestic Firms?’’, The Review of Economics and Statistics, 89:3, 482–496.

Henderson, R. (1993), “Underinvestment and Incompetence as Responses to Radical Innovation: Evidence from the Photolithographic Alignment Equipment Industry”, RAND Journal of Economics, 24(2), pp. 248–270.

Hsieh, C-T. and P. Klenow (2009), "Misallocation and Manufacturing Productivity in China and India," Quarterly Journal of Economics, 124(4), pp. 1403-1448.

Hulten, C. (2001), "Total Factor Productivity: A Short Biography," New Developments in Productivity Analysis, Charles R. Hulten, Edwin R. Dean, and Michael J. Harper, eds., Studies in Income and Wealth, vol. 63, The University of Chicago Press for the National Bureau of Economic Research, Chicago, 2001, 1-47.

Jones, B. F. (2012), "The Burden of Knowledge and the 'Death of the Renaissance Man': Is Innovation Getting Harder?", Review of Economic Studies, Vol. 76, pp. 283-317.

Jovanovic, B and G. MacDonald (1994), “Competitive Diffusion”, Journal of Political Economy, 102(1), 24-52.

Jovanovic, B. and P.L. Rousseau (2005), “General Purpose Technologies”, P. Aghion and S. Durlauf (ed.), Handbook of Economic Growth, edition 1, volume 1, chapter 18, pages 1181-1224, Elsevier.

Kalemli-Ozcan, S., B. Sorensen, C. Villegas-Sanchez, V. Volosovych and S. Yesiltas, 2015, "How to Construct Nationally Representative Firm Level data from the Orbis Global Database," NBER Working Papers, 21558, National Bureau of Economic Research, Inc.

Koske, I., I. Wanner, R. Bitetti and O. Barbiero (2015), “The 2013 update of the OECD's database on product market regulation: Policy insights for OECD and non-OECD countries”, OECD Economics Department Working Papers, No. 1200, OECD Publishing, Paris, https://doi.org/10.1787/5js3f5d3n2vl-en.

Levinsohn, James and Amil Petrin (2003) “Estimating Production Functions Using Inputs to Control for Unobservables,” Review of Economic Studies, Vol. 70, No. 2, pp. 317–341.

Melitz, M. (2003), “The Impact of Trade on Intra-Industry Reallocations and Aggregate Industry Productivity”, Econometrica, Vol. 71, pp. 1695-1725.

Mokyr, J (2014), “The Next Age of Invention: Technology’s future is brighter than pessimists allow”, Brian C. Anderson ed., City Journal, Manhattan Institute, Winter 2014, pp. 14-20.

Nickell, Stephen, ‘‘Competition and Corporate Performance,’’ Journal of Political Economy, 104 (1996), 724–746.

Nickell, S., D. Nicolitsas, and N. Dryden (1997), “What makes firms perform well?”, European Economic Review, Volume 41, Issues 3–5, 783-796, https://doi.org/10.1016/S0014-2921(97)00037-8.

Nicoletti, G. and S. Scarpetta (2003), "Regulation, Productivity and Growth: OECD Evidence", OECD Economics Department Working Papers, No. 347, OECD Publishing, Paris, http://dx.doi.org/10.1787/078677503357.

Nicoletti, G., S. Scarpetta and O. Boylaud (2000). “Summary indicators of product market regulation with an extension to employment protection legislation”, OECD Economics Department Working Papers, No. 226, Paris.

OECD (2016), Economic Outlook, Volume 2016, Issue 1, OECD Publishing, Paris.

Perla, J., C. Tonetti and M. Waugh (2015), “Equilibrium Technology Diffusion, Trade and Growth”, NBER Working Papers, No. 20881.

Pinto Ribeiro, S., S. Menghinello and K. De Backer (2010), “The OECD ORBIS Database: Responding to the Need for Firm-Level Micro-Data in the OECD”, OECD Statistics Working Papers, No. 2010/01, OECD Publishing, Paris, https://doi.org/10.1787/5kmhds8mzj8w-en.

Song, J., D. J. Price, F. Guvenen, N. Bloom and T. von Wachter (2015), “Firming up Inequality”, NBER Working Papers, No. 21199.

Steinwender, C (2015), “The Roles of Import Competition and Export Opportunities for Technical Change”, CEP Discussion Paper, No 1334.

Stock, J., & Watson, M. (1993), “A Simple Estimator of Cointegrating Vectors in Higher Order Integrated Systems”, Econometrica, 61(4), 783-820.

Summers, L (2016), “The Age of Secular Stagnation: What It Is and What to Do About It,” Foreign Affairs March/April issue.

Syverson, C. (2016), “Challenges to Mismeasurement Explanations for the U.S. Productivity Slowdown”, NBER Working Paper, No. 21974.

Syverson, C. (2011), "What Determines Productivity?", Journal of Economic Literature, 49 (2): 326–65, https://doi.org/10.1257/jel.49.2.326.

Syverson, C. (2004), "Product Substitutability and Productivity Dispersion," Review of Economics and Statistics.

Wooldridge, J. M. (2009), “On estimating firm-level production functions using proxy variables to control for unobservables,” Economics Letters, Vol. 104.

Notes

Copy link to Notes

← 1. Some argue that the low-hanging fruit has already been picked: the IT revolution has run its course and other new technologies like biotech have yet to make a major impact on our lives (Gordon, 2012). Others see the IT revolution continuing apace, fuelling disruptive new business models and enabling a new wave of productivity growth across the economy (Brynjolfsson and McAfee, 2012; Mokyr, 2014).

← 2. Throughout the paper we use the term “laggard” and “non-frontier” interchangeably – they refer to the group of firms that are not at the frontier.

← 3. Gordon also argues that future growth in the United States will slow further due to several headwinds, including ageing population, plateauing of gains from education, growing inequality, decelerating globalization, environmental unsustainability and the overhang of consumer and government debt.

← 4. For example, measurement is far more useful when coupled with active experimentation and knowledge sharing, while the value of experimentation is proportionately greater if the benefits, in the event of success, can be leveraged through rapid scaling-up.

← 5. According to Mokyr, potential barriers could come from: i) outright resistance by entrenched interests which could lead to excess regulation and lack of entrepreneurial finance; ii) a poor institutional set up of research funding which favours incremental as opposed to radical innovation; and iii) new forms of crime and insecurity (e.g. cyber insecurity).

← 6. See also the discussion in Ahmad and Schreyer (2016) on measuring GDP in a digitalised economy.

← 7. Preliminary results from the OECD Multiprod project (http://oe.cd/multiprod) based on the micro-aggregation of official representative firm-level data for 15 countries over the last 20 years (Berlingieri, Blanchenay and Criscuolo, 2017) show that most countries have experienced growing labour, and multi factor productivity dispersion coupled with increased dispersion in marginal revenue product of capital (MRPK) allowing for non-constant returns to scale. Gamberoni, Giordano and Lopez-Garcia (2016) use micro-aggregated firm-level data sources mainly based on firm level data from Central Banks compiled in the European Central Bank’s CompNet project from 5 European countries and show increasing dispersion in MRPK and MRPL across firms in the 2000s up to the crisis. Under their assumptions of constant returns to scale, MRPK and MRPL are simply multiples of capital and labour productivity, hence their findings can also be interpreted as rising divergence in productivity levels. Finally, Decker et al. (2016a) show labour productivity divergence with data covering the near universe of businesses in the United States. These studies all find that dispersion is stronger amongst the services sectors. By focusing on the frontier, our paper provides a discussion specifically on the upper half of the productivity distribution. Moreover, it provides evidence on the global rather than the national dispersion and, in particular, frontier.

← 8. Countries could also adopt active diffusion policies as was done for technology in the agricultural sector (see for some examples http://www.oecd.org/tad/agricultural-policies/Analysing-policies-improve-agricultural-productivity-growth-sustainably-december-2014.pdf).

← 9. These countries are Austria, Belgium, Czech Republic, Denmark, Estonia, Finland, France, Germany, Great Britain, Greece, Hungary, Ireland, Italy, Japan, Korea, Netherlands, Norway, Poland, Portugal, Spain, Sweden, Slovenia, the Slovak Republic and the United States. The country coverage is somewhat smaller in the policy analysis.

← 10. This means retaining industries with 2-digit codes from 5 to 82, excluding 64-66 in the European classification system NACE Rev 2, which is equivalent to the international classification system ISIC Rev. 4 at the 2-digit level.

← 11. See the full list of information providers to Bureau van Dijk regarding financial information for the set of countries retained in the analysis in Appendix E in the original OECD Productivity Working Paper: https://doi.org/10.1787/f5a3a2df-en.

← 12. We are grateful to Sebnem Kalemli-Ozcan and Sevcan Yesiltas for helpful discussions about their experience and suggestions with the Orbis database.

← 13. The measurement of intangible fixed assets in the balance sheets follows accounting rules, hence the total fixed assets (sum of tangibles and intangibles) may understate the overall capital stock (Corrado, Hulten and Sichel, 2009). Moreover, different depreciation rates and investment price deflators cannot be applied, since an asset type breakdown is not available. The implications of these limitations will be discussed when we examine the contributions of capital deepening and MFP to labour productivity divergence.

← 14. We use the country-industry level purchasing power parity database of Inklaar and Timmer (2014), see details therein for the trade-offs involved in deriving their PPP measures.

← 15. In the above example, it is unclear whether revenue-based productivity is higher because the firm is producing more steel bars, or whether the firm’s higher observed productivity is driven by higher prices reflecting high mark-ups, which the firm can charge because of a lack of competition, for example.

← 16. A further step would be a separation of market power and quality and/or demand. See Foster, Haltiwanger and Syverson (2008) and Forlani et al. (2016) on a related discussion about the role of different business strategies and their impact on measured productivity through the example of Nissan (high number of produced cars into the cheaper segment) and Mercedes (lower number of cars produced into the premium segment). Even if firm level prices were observed, complications would still arise – see Byrne and Corrado (2015) who demonstrate that official output prices of communication products are significantly under-estimated due to ignoring some quality improvements. Haltiwanger (2016) discusses in great detail the various types of MFP calculations and to what extent they are influenced by demand and market frictions.

← 17. Note that for simplicity we have assumed that the firm only produces one product. In the case of multiproduct firms, one should calculate mark-ups for each of the products sold by the firm.

← 18. The polynomial includes all possible interactions between labour, capital and materials containing first and second-degree terms, along with first- and second-degree base effects. This follows the Stata code provided by De Loecker and Warzynski (2012) with their online Appendix, with the difference that for computational reasons we omitted the third-degree terms.

← 19. In Andrews, Criscuolo and Gal (2015), we adopted a definition based on a fixed number of firms across time as well as across industries (top 100 or top 50). By allowing the frontier size to vary across industries as we do in this paper, we better tailor the frontier definition to each industry. The choice among these alternatives does not affect the main finding of a growing productivity gap between the frontier and the rest. Allowing for the number of firms to change over time also results in a diverging pattern (see Figure A1 in Andrews, Criscuolo and Gal, 2015).

← 20. Note that to ensure that our results are robust to the changing coverage of Orbis especially at the bottom of the distribution, we also compare our results to aggregate industry year trends based on official industry-country level data based on National Accounts. The results are robust of this and given that the industry level trends are going to be also affected by frontier firms, the gap between frontier and industry level trends are going to be a lower bound estimate of the actual gap.

← 21. The number of firms at the global frontier is 83 for the median industry (i.e. manufacture of basic metals). For the industries populated with a large number of businesses, the frontier represents about 400-500 companies (e.g. retail and wholesale trade, construction).

← 22. The empirical literature on productivity-enhancing reallocations indeed finds an important role for the entry-exit margin of firms (e.g. Foster, Haltiwanger and Krizan, 2001), and the theoretical literature also emphasises its potential role (Caballero and Hammour, 1994; Campbell, 1998).

← 23. Note that productivity is measured in logs, so relative to laggard firms, global frontier firms are exp1.3=3.6 times more productive.

← 24. For example, within 4-digit manufacturing industries in the United States, Syverson (2004) finds a 2-to-1 ratio in value added per worker between the 75th- and 25th-percentile plants in an industry’s productivity distribution. Including more of the tails of the distribution amplifies the dispersion, with the average 90–10 and 95–5 percentile labour productivity ratios within industries in excess of 4-to-1 and 7-to-1, respectively.

← 25. Note that we abstract from potential differences in input prices when making the link between mark-ups and output prices.

← 26. These results are based on our previous analysis for 2005, when information of firms’ patenting activity and multinational status was available. The OECD is currently in the process of updating this information for the new vintage of Orbis. Reassuringly, both in the current and previous vintage of Orbis, numerous well-known multinational companies make it to the frontier, such as Google, Apple, Amazon or Microsoft among the ICT services, Samsung, Nokia, Siemens among electronics manufacturing as well as BMW, Ford and Volkswagen within the car manufacturing sector.

← 27. We restrict the time horizon of the figures between 2001 and 2013 because the years before the 2000s and the latest year (2014) are less well captured in Orbis. In the regressions below we control for a rich set of fixed effects capturing potential changes in coverage, hence we are able to utilise a longer span of data (1997-2014).

← 28. These growth rates, expressed in percentages, are approximated by changes in log-points (i.e. log-differences multiplied by 100).

← 29. Since the detailed OECD National Accounts is an industry level database, its evolution over time reflects not only within-firm productivity developments but also changes in allocative efficiency. Further, the aggregate labour productivity measures from the industry data also reflect developments among the smallest companies (below 20 employees) as well as the self-employed. As such, it is not strictly comparable with the frontier and non-frontier firms but simply provides a benchmark against which the patterns obtained using the Orbis sample can be compared.

← 30. Note that “growth” here does not refer to the average growth of firms within the productivity frontier (laggard firms) but rather to the change over time of the average log productivity in the group of firms that are at the frontier (are laggards), with this group of firms allowed to vary over time.

← 31. To avoid estimating production functions with too few firms per industry (see next footnote), the production function parameters are still estimated at the 2-digit level and only the frontier definition is applied at the 3- or 4-digit level.

← 32. The median number of firms across 2-digit sectors and years is about 2000, but this figure falls to 210 and 130 for 3- and 4-digit sectors respectively. When looking across country*industry*year cells, these medians are 53, 8 and 6, respectively for 2-, 3- and 4-digit industries. Thus, we chose the 2-digit detail level as our benchmark, which is a compromise between avoiding too small cells and the appropriate differentiation across economic activities.

← 33. In fact, in manufacturing the divergence in MFP is larger than the divergence in labour productivity, given the faster capital deepening of non-frontier firms.

← 34. As a flipside to this issue, our value-added measure subtracts spending on these intangibles as costs. As discussed in Corrado, Hulten and Sichel (2009), the underestimation of capital and value added tends to lead to an upward bias on MFP.

← 35. The rise of “winner takes all” dynamics amongst firms could also have a knock-on effect on CEOs, for whom the rise of “superstars” with big salary premiums reflects differences in capital value of the firms they work for rather that in their talent (Gabaix and Landier, 2008).

← 36. See the original OECD Productivity Working Paper for more graphs showing these results: https://doi.org/10.1787/f5a3a2df-en.

← 37. In contrast, the average size of frontier firms and laggards in terms of employment show similar trends over time.

← 38. Exploring mark-up developments in ICT intensive sectors vs other services is outside the scope of this paper, given the challenges faced by current methodologies aimed at disentangling price-, quality- and quantity, which are particularly severe for these activities.

← 39. One limitation with the data at hand is that we do not have firm level information on ICT capital, so we are obliged to abstract from differences in ICT capital across firms within ICT intensive sectors.

← 40. We attribute this idea to Chad Syverson’s comments at the OECD-NBER Conference on Productivity and Innovation in the Long-Run.

← 41. The coefficient estimate (0.15 for the period 1997-2000 and 0.11 for the period 2010-2014) imply that the time it takes for the average laggard firm to catch-up half its initial MFPR gap with the global frontier has risen from about 4.3 years (log(1/2) / log(1-0.15) = 4.265) in the late 1990s to about 6 years by 2010-14.

← 42. We use these categories to have a more robust picture of market dynamism and selection instead of working directly with entry and exit rates. The latter tend to be more volatile and noisy, in particular because our sample contains only those firms which have at least 20 employees on average over their observed lifespan. Also, the incidence of non-viable firms is likely to be understated since we compute them for the sample where MFP is available, and this excludes cases with negative value added, i.e. firms that have larger negative profits (in absolute value) than labour costs.

← 43. For example, using cross-country microdata aggregated to the industry level, Bartelsman, Halitwanger and Scarpetta (2004) find that productivity growth within incumbent firms is positively correlated with the firm turnover rate.

← 44. This is relative to a counterfactual where MFP growth had not slowed from its 2001 rate of 0.9%: $C u m u l a t i v e L o s s_{2001 - 2007} = \sum_{t = 2001}^{2007} ({M F P_{2001} (1.009)}^{t - 2001} - M F P_{t}),$ where MFP is expressed as an index number with a base year of 2001. This means that $M F P_{2001} = 1$ and 1.009 represents the (gross) growth rates under the counterfactual of steady growth at the pace observed in 2001. $M F P_{t}$ is cumulated using the actual observed growth rates (see Figure 2.1).

← 45. These results are also significant in light of the fact that the aggregate data also reflect the MFP performance of firms that employ fewer than 20 employees and the changes in allocative efficiency (i.e. resource shifts across firms within sectors), which we do not account for in our analysis.

← 46. We also attempted to more directly assess the contribution of the MFP gap to the slowdown in aggregate productivity growth – i.e. by regressing the change in MFP growth on change in the MFP gap in a long difference specification – but these attempts were frustrated by the limited availability of reliable industry-level MFP data over a sufficiently long-time span.

← 47. See, for instance: Geroski (1995a, 1995b); Nickell (1996); Nickell, Nicolitsas and Dryden (1997); Blundell, Griffith and Van Reenen, (1999); Griffith, Redding and Simpson (2002); Aghion et al. (2004); Haskel, Pereira and Slaughter (2007).

← 48. For example, the 2013 questionnaire includes around 1400 questions on economy-wide and sector-specific provisions (see Koske et al., 2015).

← 49. As outlined in Nicoletti, Scarpetta and Boylaud (2000), the restrictions to competition captured by the OECD indicators of Product Market Regulation were defined either as barriers to access in markets that are inherently competitive or as government interferences with market mechanisms in areas in which there are no obvious reasons why such mechanisms should not be operating freely (e.g. price controls imposed in competitive industries as road freight or retail distribution). Given that the indicators cover a relatively homogenous set of countries, the underlying assumption is that regulatory patterns do not reflect cross-country differences in the level of public concern for the market failures that motivate regulations but rather reflect regulatory failure or policies adverse to competition.

← 50. Network industries include two energy sectors (electricity and gas), three transport sectors (road, rail and air) and two communication sectors (post and telecommunications). The two professional services industries refer to the business services sector (accounting and legal services) and the technical services sector (engineering and architecture services).

← 51. Throughout the analysis of PMR’s impact on the productivity gap, the coverage is restricted to cases where the annual PMR indicators are available and where at least 10 firms are present in Orbis. The included 14 OECD countries are Austria, Belgium, Denmark, Finland, France, Germany, Italy, Japan, Korea, Portugal Slovenia, Spain, Sweden and the United Kingdom.

← 52. As discussed below, the results are not particularly sensitive to the choice of the length of the long difference window.

← 53. For retail and professional services industries, where the indicators are updated only every 5 years (1998, 2003, 2008, 2013), additional information was used on the timing of reforms following the calculations of Gal and Hijzen (2016) and Duval et al. (2016). Taking the log of PMR is a useful transformation to the extent that it allows for reforms to be evaluated in relative terms, in relation to the pre-reform policy stance. This is particularly relevant as in many industries the level of regulation – as expressed by the PMR indicator – is already at low levels, and there is limited scope for further reforms that lead to a similar reduction in absolute terms in the indicator than in the past.

← 54. The employment variable is based on information from the Orbis database and is calculated as the average of log employment levels across firms.

← 55. DOLS estimates are presented based on both one and two lags and leads.

← 56. We employ the IV approach in the context of our baseline long-difference specification since it’s easier to find plausible instrument variables for reform – i.e. PMR change – then the level of PMR.

← 57. The standard deviation of PMR and the MFP Gap are 0.673 and 0.470 respectively.

← 58. See the original OECD Productivity Working Paper for more tables showing these results: https://doi.org/10.1787/f5a3a2df-en.

Featured topics

Agriculture and fisheries

Climate change

Development

Digital

Economy

Education and skills

Employment

Environment

Finance and investment

Governance

Health

Industry, business and entrepreneurship

Regional, rural and urban development

Science, technology and innovation

Society

Taxation

Trade

Energy

Nuclear energy

Transport

Featured topics

Agriculture and fisheries

Climate change

Development

Digital

Economy

Education and skills

Employment

Environment

Finance and investment

Governance

Health

Industry, business and entrepreneurship

Regional, rural and urban development

Science, technology and innovation

Society

Taxation

Trade

Energy

Nuclear energy

Transport

Countries A - C

Countries D - I

Countries J - M

Countries N - R

Countries S - T

Countries U - Z

Regional and global engagement

Countries

Countries A - C

Countries D - I

Countries J - M

Countries N - R

Countries S - T

Countries U - Z

Regional and global engagement

Publications

Publications

Featured publications

Data

Data

Featured data

News & events

News & events

Featured events

About OECD

About

Engage with us

Work with us

Featured topics

Agriculture and fisheries

Climate change

Development

Digital

Economy

Education and skills

Employment

Environment

Finance and investment