Firms and inequality

Abstract

We review the existing literature on falling business dynamism and present a new analysis using comprehensive UK firm-level panel data. Since the mid-1990s, there has been a large increase in UK firm-level inequality (especially in the upper tails) of productivity, wages, markups and labour shares, similarly to the USA. We suggest a simple theoretical framework for understanding some of these trends and quantitatively analyse why, despite increasing markups, the UK labour share has not fallen as sharply as that in the USA. Finally, we suggest some policy options in response to these worrying trends, including modernizing competition rules to deal with the growth of superstar firms and strengthening worker bargaining power.

1. Introduction

There have been some dramatic changes in the business landscape in American firms over the last four decades that have been extensively documented by researchers and are the subject of much public debate (e.g. Van Reenen 2018; Akcigit and Ates 2019). In broad terms, one could summarize many of these trends by saying that firm inequality has increased—companies are looking increasingly different in terms of their productivity, wages, markups and size. These changes have been accompanied by several worrying trends, such as declining productivity growth, stagnating real wages (especially for low-educated workers), rising markups, a falling share of labour in gross domestic product (GDP) and declining business dynamism (e.g. the share of workers in young firms has declined).

There are a large number of possible explanations for some (or all) of these trends. Broadly, some authors stress factors that are common across countries, such as technological change and globalization (e.g. Aghion et al., 2019; Hsieh and Rossi-Hansberg 2019; Peters and Walsh 2019; Baqaee and Farhi 2020; De Ridder 2020). Others point to US-specific factors such as weakening anti-trust enforcement and the erosion of labour market institutions such as the minimum wage and unions (e.g. Philippon 2019 and Stansbury and Summers 2020, respectively). Of course, it is very likely that these forces interact with each other. Take the presence of technological change in a rapidly globalizing world, where firms source from highly complex supply chains across the globe while at the same time potentially selling to a much larger potential pool of consumers. Changes in technology have enabled companies to trade in larger markets and fragment production in global value chains. This has differential implications for local labour and product markets. Certain countries might find themselves as the beneficiaries of this process, while others (at least relatively) are mostly absorbing the negative shocks. Even within a country, the ‘China shock’, for example, displaces manufacturing jobs in local US labour markets (see Autor et al. 2016), but there are other places where firms and people benefit from having access to cheaper final and intermediate goods (e.g. Jaravel and Sager 2019). Recent empirical work has shown the benefits of increased globalization in lowering the cost of production (a standard ‘gains from trade’ argument). However, firms with market power will capture part of this cost saving, and this will give rise to increasing dispersion in firm performance and can dynamically further widen the gap between firms, ultimately giving rise to some of the facts mentioned above (see World Bank (2020) for a discussion in the context of global value chains).¹

Studying whether these trends have also occurred in Britain is important for two reasons. First, such developments can have important implications for overall UK inequality. To give some examples: (i) rising dispersion in firms’ productivity translates into increasing wage inequality between those employed in high- versus low-productivity companies; (ii) a lower labour share means falling pay or jobs for a given level of GDP, as well as higher household inequality (as capital income is more unequally distributed than labour income) and (iii) rising markups imply higher prices for a given level of marginal costs (or that less of any cost falls gets passed on to consumers), reducing living standards.

A second reason for studying UK business trends is to help assess the explanations for such economic shifts. If the trends are just US-specific, then this suggests that American institutions are more likely explanations than global factors such as technology or trade, which broadly affect all countries. While several papers take an international approach to these trends or have studied a particular economy, there have been few ‘deep dives’ into what the UK experience has been like and, to our knowledge, that cover a wide range of business sector outcomes. The primary purpose of this article is to summarize what we know about these trends in the UK. We draw on several data sources, in particular Historical Orbis (HO), a relatively new panel of the population of incorporated firms since the mid-1990s. We also consider more traditional administrative data sources from the Office for National Statistics (ONS).

We find that almost all of the US trends are present in the British data, although the UK information is generally not as rich as that in the USA. In particular, we document at the micro level increasing between-firm differences in productivity, wages, size and markups, similar to those in the USA. Generally, the increase in inequality is greater for firms in the top half of the distribution, with high performing‘superstar’ firms pulling away from the rest of the pack. Similarly, at the macro level, the UK has experienced a rise in aggregate markups (like the USA), slower growth in productivity and wages since the Global Financial Crisis (more dramatically than in the USA) and a fall in the wage share of GDP since 1980 (much less stark than in the USA).² UK entrepreneurship trends in the UK are more ambiguous.

The structure of the article is as follows. In Section 2, we ask why we should care about inequality between firms before discussing data sources in Section 3. Section 4 is our main contribution, where we document the stylized facts of business trends in the UK and implement a rough framework for quantification. We examine possible explanations for these changes in Section 5. In Section 6, we offer some tentative policy implications before concluding in Section 7. Online appendices cover the data in more detail, offer some further analysis and give more mathematical details of the calculations for markups and labour shares (C).

2. Why should we care about inequality between firms?

There are many good reasons to care about inequality between people, households and communities. However, why should we care about inequality between firms? We first document some facts about cross-sectional firm inequality and then discuss this fundamental question.

There is certainly a great deal of cross-sectional firm inequality; similar to other countries, the UK firm size distribution is very skewed (see Table 1). There were just fewer than 6 million firms in the UK at the start of 2020, with 3.3 million of these being unregistered for employee withholding tax [Pay-As-You-Earn (PAYE)] or value added tax (VAT), so employing no workers—the employment figure is for the owners of these firms.³Table 1 shows that 95.7% of businesses have nine employees or fewer (55.6% in unregistered firms and 40.1% in registered firms). Although numerous, these micro firms account for only a third of employment and 21.5% of turnover,⁴ implying that they have lower labour productivity than larger firms as measured by turnover per worker. By contrast, there are only 8000 firms (0.1% of businesses) with at least 250 workers, but they employ almost 11 million people and produce just over £2 trillion in turnover. Consequently, these big firms accounted for almost two in five of all jobs (39.3%) and just under half of aggregate turnover (47.8%).

We can compare these firm demographics with those in the USA using Census Bureau Business Dynamics Statistics data.⁵ The USA is a larger economy, with 5.17 million employer firms (i.e. with at least one worker in addition to the owner), compared with 1.41 million in the UK. Of the US firms, 41 772 have 250 employees or more, and these accounted for 57.7% of all employment in employer firms. This compares with 47.9% in the UK. This is consistent with the idea that there are fewer barriers to firms growing to greater scale in the USA (e.g. Hsieh and Klenow 2009; Bartelsman et al., 2013).

Many politicians talk about helping small firms, as if this is an end in itself. The logic for this is unclear, however. Given the large numbers of workers in big firms, shown in Table 1, focusing solely on small firms is a poorly targeted way of helping individuals. Furthermore, smaller firms are on average less productive, pay lower wages and offer fewer fringe benefits. None of these facts is an argument for subsidizing large firms. However, these business demographics suggest that the multitude of policies explicitly or implicitly subsidizing small firms needs a more solid economic justification. One argument is that small firms suffer from more financial constraints. However, this is likely to be more due to the age of the firm than their size (Decker et al., 2014). Studies of firm growth suggest that young firms create many jobs, not small firms per se. Consequently, distorting taxes and regulations to favour small firms can be a drag on growth (e.g. Garicano et al., 2016). Moreover, if financial constraints are the friction affecting smaller firms, then there may be more direct ways to correct financial market failures than offering blanket subsidies to small- and medium-sized enterprises (SMEs).

In summary, it is important to realize that the quite justifiable desire to support households with low income is quite different from the arguments to support SMEs.

With these policy caveats established, there are several good reasons for studying inequalities between firms. One primary reason is simply a descriptive interest in business demographics, which has fascinated industrial organization scholars such as Gibrat (1931). But beyond this, there are many reasons that we might care about firm inequality. Understanding how business structures have evolved may help us understand broader questions of what is happening to the economy and how it affects people.

First, there is the issue of whether the changes in the inequality between firms cause (or signal) a fall in aggregate productivity growth, which is important as productivity growth is the critical determinant of long-run income growth. For example, several OECD papers (e.g. Andrews et al., 2015) have argued that increasing productivity differences between leading firms and followers (‘the best versus the rest’) indicate a slowdown of the diffusion of new technologies and/or management practices. Since diffusion is a key component of productivity growth, this could have first-order welfare effects (see also Akcigit and Ates 2019). Another example is from the voluminous literature on misallocation. In a wide class of models (e.g. Hsieh and Klenow 2009), an increase in the dispersion of (revenue-based) firm productivity indicates greater misallocation and therefore lower aggregate output.⁶ In UK policy circles, this is often termed the ‘long tail of unproductive firms’ and reducing this is part of the current government’s ‘levelling up’ agenda. It is important to realize that this difference is not simply an issue of inequality in the lower part of the productivity distribution. For example, if median-productivity firms find it increasingly hard to catch up with leading firms (say those in the top 5%), this could be an even larger drag on productivity.

Second, a growing literature in labour economics has documented how workers’ wages and well-being depend on the firms that they work for over and above their individual skills and characteristics (see the survey in Card et al. (2018)). Larger and more productive firms pay higher wages (see Van Reenen (1996) for early causal evidence from the UK). Additionally, employer–employee matched panel data show that the same worker gets very different wages depending on the firm where she is employed (Abowd et al., 1999). Given these stylized facts, increasing differences between firms will translate into increasing pay dispersion between workers and therefore higher overall income inequality. This could be one of the reasons behind the substantial increase in UK wage inequality.⁷

Third, there is evidence of an increase in the aggregate markup (the wedge between prices and variable cost) since the early 1980s in US firms (De Loecker et al., 2020a) and in other countries (De Loecker and Eeckhout 2018). This seems to be related to increasing differences between firms, as the markup of the median firm has not risen. The well-documented fall in the share of US labour in national income could be the flip side of this (as Autor et al. (2020) argue) because labour is a major component of variable costs. A rise in aggregate markups could have several negative welfare effects. For example, if it reflects rises in monopoly distortions, this means higher prices and lower real wages. Merely documenting rising markups, rising profit margins and rising concentration is insufficient to evaluate the welfare effects without understanding the underlying causes, which we discuss in Section 5.⁸

Fourth, there is the issue of whether growing inequality between firms translates into greater corporate power to lobby and skew the ‘rules of the game’ in favour of business over consumers and workers. The textbook view of economics considers a firm as an organization that delivers a product or service using a bundle of inputs, knowledge and expertise. It does so while operating under a set of rules governing labour and product markets, and adhering to tax law and financial regulations. However, in practice, firms do influence these rules, and try to create a more advantageous position for themselves. Firms engage in a host of lobbying activities from funding interest groups and campaigns to direct financial relationships with politicians, and the ability to engage in these activities is very likely to be a function of firm profitability. This would suggest a connection between the state of competition and lobbying, or, in other words, we expect that in a highly competitive market place, firms do not have the extra cash to try to influence the rules of the game. However, as markets get less competitive, and economic activity is more concentrated, we could expect lobbying to become prevalent, and of course further fuel the increasing market power. This hypothesis is complicated to reliably test, not least because measuring lobbying activities (broadly defined) is challenging.

Recently, a few attempts have been made to relate lobbying activities to firm characteristics and outcomes. Almost all studies have been done on the USA, both because there has been a sharp increase in corporate spending geared towards influencing policy and because there are, in general, better data to measure the objects of interest. Lobbying expenditures in the USA rose from $1.5 billion in the late 1990s to over $3 billion today, the vast majority of this by businesses (Center for Responsive Politics and Federal Lobbying Disclosure Act—see Philippon (2019)). Huneeus and Kim (2018) study the effect of lobbying activities in the USA. They match political connections to firm-level financial variables, and find that lobbying activities are positively correlated with firm size (measured by sales) and firm profits. They use these facts to calibrate a macro model and find sizeable productivity losses from lobbying activities, through a misallocation of resources. Ferracuti et al. (2020) find that politically active firms are better insulated from volatility, and perform better during periods of policy uncertainty. It is also worth bearing in mind that lobbying also occurs through SMEs, many of which are very effective at organizing through trade and professional associations (see Atkinson and Lind 2018). For example, US physicians are a very effective lobbying organization through, for example, the American Medical Association.

The evidence is much less clear for the UK and the European Union. The levels of spending certainly seem much lower as total lobbying expenditure in the EU in 2016 was about $1.5 billion—about half of that in the USA (see LobbyFacts.eu and the EU Transparency Register). There are no longer-term databases to track the trends, however. It may be that the EU will follow in the footsteps of the Americans, and certainly Big Tech has increased its lobbying activities in Brussels. There is a risk that the UK will be more vulnerable to lobbying activities (especially by large British firms) having left the powerful and independent umbrella of DG-COMP, the EU’s competition enforcer.

There are many other interconnections between firm dispersion and society’s well-being, but we believe these four certainly justify some investigation of UK trends.

3. Data sources

We give full details on all the data sets used in the data appendix (Appendix A) and the relevant parts of the article. This section gives a briefer overview. We draw on many data sources in our study. We use several publicly available data sets from the UK Office for National Statistics (ONS), which are aggregations of administrative micro data sets to the country or broad industry level. We also draw on the Business Structure Database (BSD), which contains the population of all UK registered firms since 1997.⁹ The BSD contains no information on value added or wages. For this, we use the Annual Business Survey (ABS; formerly known as the Annual Business Inquiry, ABI), which has the population of firms with 250 or more employees, but only a stratified random sample of smaller firms.

The main micro data set we use in this study is Historical Orbis (HO). HO is a relatively new data set from Bureau Van Dijk (BVD), a company that provides digitized versions of the available accounts of essentially all incorporated firms in the world. We focus on firms that are incorporated in the UK, and are therefore listed in Companies House. Table A1 shows that there are about 2.9 million firms in the latest full year (2016). Many previous versions of these data (e.g. BVD’s AMADEUS or ORBIS products) dropped inactive firms after five or ten years, so were subject to serious selection bias. To deal with this, researchers had to append overlapping cohorts of the data (e.g. Kalemli-Ozcan et al., 2015; Bloom et al., 2016). By contrast, HO keeps in principle all inactive firms where available. In brief, HO has the advantage that the data are public and not anonymized, so companies and business groups can be analysed and merged with other data sources. Further, the accounts are audited, which provides some degree of data reliability. Moreover, HO is clearly more comprehensive than a survey.

Nevertheless, there are limitations of the HO data. First, although data from the UK are available from 1982 onwards, the coverage is only comprehensive for more recent years. Our analysis suggests that the UK data coverage is broadly reliable from 1996 but, prior to this, coverage seems to be increasingly selected, missing some inactive firms. The most recent years also have incomplete coverage, as there is a lag between the information being gathered by a firm, the publication of its accounts, the accounts being lodged in Companies House and the data then being digitized by BVD into HO. Consequently, our main analysis sample uses the two-decade period 1996–2016. A second drawback of HO is that the availability of data items depends on accounting rules. For example, in the UK (as in other countries), almost no firms report some useful data items such as intermediate inputs, as it is not an item investors pay much attention to. By contrast, the ABI/ABS administrative data have purchases of intermediate inputs. Broadly, large publicly listed firms have to report many accounting items, whereas small unlisted firms only have to report very basic information. These reporting requirements also change over time. Some firms voluntarily choose to report more than what is legally required, but obviously this raises concerns over sample selection. This means, for example, we have information on the capital of just about every firm as even the smallest firms have to report an abbreviated balance sheet with assets and liabilities. However, we only have employment for a subsample of firms. Incomplete coverage of data items for the very small firms is not a problem for some types of analysis (aggregate productivity), but is more of an issue when we want to examine the lower quantiles of distributions.

In general, in our main HO analysis, we restrict our analysis to the ‘market economy’, dropping difficult-to-measure industries such as those with connections to the public sector (e.g. health, education), agriculture, energy extraction (including oil), finance and real estate.¹⁰ We also drop duplicate firms, use data at the highest level of the business group within the UK and drop firms with under 10 employees.¹¹ We develop two main ‘analysis samples’ from HO that differ by which variable we insist on having no missing values. ‘Sample A’ drops missing values on employment, the wage bill and profits (as measured by EBITDA, earnings before interest, tax, depreciation and amortization). These are the items we need to construct labour productivity (value added per worker) and labour shares. We measure value added by the wage bill plus profits. We measure the labour share as the wage bill divided by value added. ‘Sample B’ has non-missing values on sales, employment and cost of goods sold (COGS). These are the crucial items needed to construct our proxies for gross markups (price divided by marginal cost)—see Appendix C.

We do several other cuts of the data, in particular distinguishing between UK publicly listed firms and unlisted firms (this includes both private UK firms and the subsidiaries of foreign multinationals). We also compare the listed firms in HO with Worldscope, another database of (only) publicly listed firms around the world, which is available for earlier years than HO.

4. Changes in the UK business landscape

Overview

We seek to document some simple facts. Table 2 presents a series of changes to the level and distribution of several important economic outcomes: productivity, wages, markups, firm size, labour shares and entrepreneurship. Column 1 summarises the US trends and column 2 our analysis of the UK trends (presented in more detail below). Our broad take is that the similarities between Britain and the USA are quite striking.

Before starting our statistical tour, a few caveats are in order. First, even in the USA, where research work has focused, there is still controversy over some of the stylized facts as we lay them out, qualitatively as well as (inevitably) in magnitude and timing. For some of the findings, the controversy is less about the facts and more about the interpretation. We mention these in the relevant subsections below. Second, space constraints prevent us from a deeper investigation into countries other than the UK and the USA, but we mention some of the similarities and differences especially with regard to other European countries. Third, much of what we present on the UK is new work on the Historical Orbis (HO) database, so it is more preliminary and we discuss the source in more detail. Where possible, we compare this with administrative data.

With these caveats in mind, Table 2 gives a brief and crude summary of our findings for the UK compared with the USA. There are many issues that we will discuss, but broadly there is much similarity. One apparent difference, however, is that the labour share of GDP does not appear to have fallen as dramatically in the UK as it has in the USA, and we will discuss this in more detail below.

To summarize:

• There has been a slowdown in aggregate labour productivity growth (GDP per hour) after the 2008–09 Global Financial Crisis in the macro data. This is also evident in the HO and other micro databases when aggregated. The US and other OECD countries also experienced a slowdown, although less dramatically than the UK. The HO data also show an increase in productivity dispersion between firms, which is consistent with other UK sources (especially for the above-median upper tail of the distribution). This is similar to the US and other OECD countries.

• UK aggregate real wage growth also stagnated after the Global Financial Crisis, at a similar rate for the mean and the median worker. Wage inequality has also risen between firms (and individuals) since the late 1970s. In the USA, wage stagnation for median workers began earlier, at about the same time as wage inequality took off (from the mid-1970s).

• Aggregate markups appear to have risen in the UK and the USA.¹² The dispersion of markups has increased (e.g. markups for the median firms have stayed broadly stable in both countries).

• As in the USA, the share of turnover in large firms has risen and industrial concentration has increased.

• The labour share of GDP has fallen substantially in the USA. The share has been more stable in the UK since 1981, but falls to some degree when we consider measurement issues relating to self-employment and non-wage benefits.

• Entrepreneurship indicators such as the share of activity in young firms have fallen in the USA. In the UK, these are more ambiguous: the share of turnover in young firms has fallen, but their share of jobs has not.

Trends in indicators: detailed analysis

For each business indicator, we start with looking at aggregate patterns from (usually) public administrative data and then compare these with suitably aggregated data from our micro sources. We look at the private sector market economy first and then (when available) very broad industrial sectors. We then look at dispersion across firms using the micro data.

Productivity

Aggregate patterns

The UK has a well-known productivity problem in levels, with lower GDP per hour than its peer nations such as the USA, France and Germany (see Valero and Van Reenen 2019). Furthermore, although most countries experienced a slowdown in productivity growth after the Global Financial Crisis, this slowdown was particularly severe in the UK. Figure 1 shows labour productivity as measured by real GDP per hour (normalized to 100 in 1981).¹³ UK productivity growth averaged about 2.4% per year in the period until the Global Financial Crisis, but fell by an order of magnitude afterwards, being almost flat. A similar picture holds for output per worker (rather than per hour) growth.

How does this compare with aggregated micro data from Historical Orbis? Figure 2 shows the HO trends from 1996 in labour productivity (the sum of all value added divided by the sum of all employees). This is taken from our ‘Sample A’. The macro trends are in line with the aggregate data, with strong increases in productivity until the Global Financial Crisis. Productivity then drops sharply, recovers somewhat and then stagnates after 2010. The broad similarity is reassuring, since there are a large number of reasons why the ONS numbers could differ from HO.¹⁴

A further way to compare the productivity trends across data sources is by industry. Appendix Figure A1 breaks down value added per worker in HO by the five broad sectors in the market economy and compares it with administrative data. The patterns seem similar across the two data sources.

Productivity dispersion

In order to examine firm inequality, we must use the full micro data. Having validated the aggregate productivity trends in HO, Fig. 3 uses it to show one measure of the dispersion of productivity across firms. Following the OECD (e.g. Andrews et al., 2015), we compare average productivity for the ‘frontier’ (defined as the employment-weighted average productivity of firms in the top 5% of the productivity distribution in each year) compared with ‘followers’ (the rest of the economy). We normalize the series to 0 in 1996 and use logarithms, so cumulative productivity growth can be read off the figure. Hence, the value of 0.03 in 2005 for followers indicates a growth of 3 log points (or 3%) between 1996 and 2005. As in the USA and other OECD countries, dispersion in productivity appears to have risen since 1996 on this measure. Leading firms’ productivity grew by 67 log points between 1996 and 2016 whereas follower productivity grew by 14 log points. Since the Global Financial Crisis, both groups of firms have seen their productivity stall.

Figure 4 presents a more conventional quantile plot comparing the 10th, 50th, 90th and 95th percentiles. The left-hand panel is unweighted (by firm) and the right-hand panel is weighted by firm employment. The unweighted panel again shows a fanning out of the distribution, especially after the financial crisis: there was some mild slowdown from firms in the top decile, but more dramatic stagnation—and even falls—in productivity for the median and the bottom decile of firms. The figure shows upper-tail productivity rising much more than dispersion in the lower tail, below the median.

A broadly similar picture emerges in the right-hand panel of Fig. 4, which weights firms by their employment. Here, the median line reflects the productivity of the firm where the median employee works (rather than simply the unweighted median across all firms regardless of size as in the left-hand panel), so it is arguably more policy relevant. The fanning out of the distribution is also evident, although here the change is more dramatic, starting in the late 1990s. Median productivity has not increased over this period and the lower tail (the 50:10, the ratio of the median to the 10th percentile) widens more than in the unweighted quantiles. Firms at the top of the distribution (which are the larger firms) have enjoyed steady productivity growth.¹⁵ As noted above, these figures are all in log points. So in panel b the value of almost 0.4 in 2016 indicates that the 90th percentile had almost a 50% ((e^0.4–1) × 100 = 49.2%) increase in productivity over the 1996–2016 period.

Other papers

Our finding of widening UK productivity dispersion is broadly consistent with other sources using administrative data (e.g. Bahaj et al., 2017; Office for National Statistics 2019). Oliveira-Cunha et al. (2021) use the public release version of the ABI/ABS administrative data in Office for National Statistics (2020) to examine productivity dispersion. We give a detailed comparison of the trends in the two data sets in Appendix B. Broadly, the ONS data also find substantial increases in productivity dispersion in the upper tail. For example, productivity at the 95th percentile of the (employment-weighted) firm distribution has risen a lot, whereas it has been stagnant at the median, consistent with our analysis of HO in panel b of Fig. 4. The magnitude of the increase is lower in the ONS data, however, and there is no sign of increasing dispersion in the lower tail. We think this is likely to be because coverage in both HO and ONS is comprehensive for the upper tail, but less reliable in the lower tail.

Summary on productivity

UK aggregate productivity growth has dramatically slowed since 2007 in the macro data and in most broad industrial sectors. We see similar trends when aggregating our firm-level data from Historical Orbis, which is a useful validation. We also document a large increase in productivity dispersion between firms, and even more so when we take into account the weight of a firm in terms of employment, indicating that growth itself was unequal across the firm size distribution. Comparing this with ONS firm-level data suggests that this increase is strongest in the upper-tail (above-median) part of the productivity distribution.

Wages

Aggregate mean and median wages

Figure 5 plots real wages for the mean and median employee since 1980 from administrative sources (ASHE) over roughly the same period as the labour productivity series in Fig. 1. As can be seen, the qualitative features of average wages follow those of productivity trends. There was a healthy growth up to the Global Financial Crisis, then a sharp fall thereafter. Real wages began rising from about 2013, but by 2020 had still not fully recovered to where they were pre-crisis.

A second important fact regarding Fig. 5 is that median real wages have grown much more slowly than mean wages since 1980, reflecting the well-known growth of UK wage inequality. As can be observed in the graph, there has not been an increase in this measure of wage inequality from the mid-2000s onwards. Although wage inequality has also grown dramatically in the USA since the mid-1970s, a major difference is that median real wages have been quite stagnant in the USA since the 1970s, whereas Fig. 5 shows that they grew at a healthy clip in Britain until the financial crisis. Indeed, for the three decades preceding the financial crisis (1979–2007), national income per person grew faster in the UK than in the USA, France and Germany, reversing Britain’s relative decline over the previous century. Corry et al. (2011) argue that there were many UK policy-related factors underlying this, including reforms to labour and product markets, educational attainment and joining the European Union. Since the financial crisis, however, median wage growth has broadly tracked mean wage growth and productivity. Again, this suggests that for the last decade, the main UK problem has not been inequality (at least as measured by the difference between the mean and median wage), but rather slow productivity growth. The fact that policy helped improve the UK’s productivity position after the 1970s holds up some hope that appropriate institutional and policy reforms could help this happen again (Besley and Van Reenen 2013).

Figure 6 presents HO data for productivity (value added per worker), wages (wage bill per worker) and profits per head (EBITDA per worker). Payroll is about two-thirds of our measure of value added.

Wage dispersion

Although much is known about wage inequality between individuals, much less is known about the distribution of wage inequality between firms. Figure 7 (from HO) shows that it has also increased since 1996. The patterns are similar to those for productivity in Fig. 4 with an increase in dispersion in both unweighted (left-hand) and employment-weighted (right-hand) panels.¹⁶

These patterns are broadly consistent with US work. For example, Song et al. (2019) use employer–employee matched data from the Social Security Administration and Barth et al. (2016) use data from the LEHD (state-specific Unemployment Insurance systems). These papers suggest the majority of inequality growth across workers is due to increasing average wage differences between firms. Unfortunately, there is not comprehensive employer–employee matched data in the UK over many decades to do a rigorous decomposition of the within- versus between-firm wage differentials as these papers can do. The similar firm-level patterns do suggest an important role for companies in wage inequality, however.

Summary on wages

UK wages have stagnated at the mean and median since 2008–09. Wage inequality between individuals has been increasing since the late 1970s, and firm-level wage dispersion has been on the increase.

Firm markups of prices over marginal costs

The markup of price over marginal cost is challenging to measure: although prices are in principle easy to record, marginal costs are not.¹⁷ First, one can use engineering estimates, but detailed cost data are rarely available. Second, one can estimate demand systems using quantity, price and product characteristic data and infer costs indirectly using an optimization assumption. However, detailed brand price data are unavailable across large ranges of goods. To look across the whole market economy, a more popular approach starts from the production function and a cost minimization assumption that generates the well-known relationship between a factor’s share of revenue and the output elasticity (see Appendix C). When prices are equal to marginal cost (a markup of one with zero profits) as under perfect competition, the share of a variable factor (such as materials) should equal the output elasticity with respect to that factor (a technological parameter). As markups increase, the share of the variable factor will fall relative to the elasticity. The cost of goods sold (COGS) is the most easily observable proxy for variable costs in accounting data. For stable technology, a fall in the COGS share indicates an increase in the markup.

We stress that the results presented here rely on a calibrated output elasticity (time and firm invariant).¹⁸ This was the starting point of De Loecker et al. (2020b) in their initial analysis of the US data.¹⁹ Note that profit rates will include measures of fixed costs, so that even when markups over variable costs rise, profit rates may not rise because of the increase in the importance of fixed costs. Although one should worry about the heterogeneity and time-varying aspect of the output elasticity, this markups calculation should be seen as a first step in characterizing the main UK patterns.²⁰

Aggregate markups

Many investigators have analysed markups in the US economy, although increasingly more work has centred on other regions of the world. For the US economy, most work has found an increase in aggregate markups, with a range depending on specific modelling assumptions and methodologies. De Loecker et al. (2020a) find that the increase in their estimated aggregate markup among publicly listed US firms is driven by a reallocation of activity towards high-markup firms, rather than an overall within-firm increase in markups. The increase in markups is shown to come predominantly from the falling share of costs of goods sold to total revenues.²¹

Figure 8 shows the change in the aggregate COGS share across all Historical Orbis firms.²² We believe this is the first time such an analysis has been performed across such a large range of firms for the UK. We observe a significant fall in the share of variable costs of almost 6 percentage points, particularly in the 1996–2000 period. Figure 9 converts this into a markup using an output to COGS elasticity of 0.85, as is standard,²³ unsurprisingly showing a large increase in the aggregate markup. Appendix C discusses markup calculations in more technical detail and Appendix B shows various robustness tests. For example, when we use input weights instead of output weights, we see similar trends, albeit with smaller magnitudes.

Figure 10 compares the markups for publicly listed firms and for unlisted firms. Unlisted firms have lower markups than their listed counterparts, which is unsurprising since they are smaller. Markups rise over time for both sets of firms, but the increase is more marked for listed firms and the timing of the changes differs (strongest 1996–2004 for listed and more continuous for unlisted). We can further decompose the unlisted firms between those that are foreign subsidiaries and those based in the UK. Interestingly, the unlisted UK firms have seen faster increases in their markups, whereas the foreign affiliates have been more stable. All this suggests that there may be some interesting differences between the ways multinationals structure themselves that lead to changes in estimated markups. We also look at markup changes by sector, finding that markups have risen broadly across industries (see Appendix Fig. A5).

Markup dispersion

Figure 11 looks at the quantiles of markups (unweighted in the left-hand panel and weighted by revenues on the right). As with firm productivity and wages, there has been an increase of dispersion across firms in terms of markups. The increase in upper-tail inequality is particularly impressive. There has been essentially zero change in markups for firms at the median or below for both panels. By contrast, firms at the 90th percentile have had a 40% increase in the markup and those at the 95th had one twice as large for the unweighted series. The sales-weighted quantiles in Fig. 11 show the same pattern, with zero change for the ‘median consumer’ but an even larger increase at the top of the distribution (note that the scale of the vertical axis is twice as large as for the left panel). The timing is slightly different however, being more focused in the late 1990s.

Profit rates

The markups we have discussed are gross markups, i.e. estimates of the wedge between prices and marginal costs. Rising markups do not necessarily mean an increase in economic profits. For example, there may have been an increase in the importance of fixed costs (intangible capital of software, research, marketing, etc.), so that net markups are actually stable. Measuring fixed costs and economic profits is extremely challenging, of course. One crude measure is simply the ratio of earnings before interest, taxes, depreciation and amortization (EBITDA) to sales. EBITDA is net of COGS and other expenses (SG&A—selling, general and administrative expenses) that may be more fixed. Figure 12 shows that this ratio has risen by about 2 percentage points from about 9.5% to 11.5% between 1996 and 2016, hinting at the idea that fixed costs account for some, but not all, of the increase in markups.

The analysis of US data in De Loecker et al. (2020b) highlights the increase in fixed cost (proxied by SG&A). Although this partly offsets the rise in gross markups (i.e. the higher margins are required to cover these fixed costs), there is still an increase in profit rates even net of these fixed costs. The latter result is consistent with the reported increase in the aggregate profit share in the overall US economy.

Other papers

The finding of increasing markups (and markup dispersion) is consistent with other work on other OECD countries and also on UK listed firms such as De Loecker and Eeckhout (2018), Díez et al. (2018) and Aquilante et al. (2019). The only other UK study we are aware of that includes unlisted firms is Competition and Markets Authority (2020), which focuses on Orbis data, but only includes firms with over 250 employees. It also finds increases in markups. If we condition on these larger firms, we also find similar results to those presented here, which is unsurprising as the patterns are driven by the larger firms, as we have shown.

Summary on markups

Aggregate markups appear to be rising in the UK since the mid-1990s using the HO data. The rise is stronger in listed firms, which are larger and more global than unlisted firms, and also appears when looking at profit rates (which should strip out fixed costs). The dispersion of markups is also increasing, with median markups (like productivity) suggesting an important role for reallocation, rather than a general rise in markups across all firms.

Labour share

Aggregate labour share

The evolution of the share of labour in output is interesting for many reasons. First, if the economic pie is growing strongly, people may be relaxed if workers are getting a smaller share of it. However, as we have seen, productivity growth has been weak in the UK since the financial crisis. Hence, a labour share fall implies downward pressure on wages or jobs or both. Second, the distribution of capital income is much more unequal than that of labour earnings. Therefore, a fall of the labour share will be a force for increasing inequality across households. Third, the macro stability of the labour share was one of Kaldor’s famous stylized facts, which was, as Keynes said, always ‘something of a miracle’. The fall in the US labour share since the early 1980s requires revision of this ‘stylized fact’. The labour share has also been falling (albeit at different rates and timings) in most other OECD countries (see Karabarbounis and Neiman (2014), Autor and Salomons (2018) and Autor et al. (2020)).²⁴

By contrast, the UK share of employee compensation in value added appears to have a different evolution at the macro level. Figure 13 taken from Dunn et al. (2018) shows that the UK share of employees’ compensation fell from 63% in 1957 to 50% in 1987 where it stabilized until the mid-1990s. It then rose before settling at around 57% in 2001 and has more or less remained there since.

Teichgraeber and Van Reenen (2021) analyse more recent data in greater detail (see Fig. 14), focusing on the period after 1980 when the US labour share started falling and before 2020 (to avoid the confounding impact of COVID-19). The blue line is the official ONS series, which adds an estimate of self-employed labour income to employee compensation in calculating the labour share. This measure of the labour share has fallen by 2 percentage points over the period. The share fell the most between 1980 and 1996 under the Thatcher–Major Conservative governments and rose between 1997 and 2010 under the Blair–Brown Labour governments, before falling slightly since then.

This disguises two important changes, however. First, there is the position of the self-employed, those on so-called ‘mixed income’.²⁵ This group has experienced much slower income growth (of about 44%²⁶) between 1981 and 2019 than the employed (whose hourly compensation rose by about 80%). Further, the fraction of workers who are self-employed has risen substantially (from about 10% to 16% of the workforce over the same period). Since the self-employed have lower incomes on average than the employed (primarily because of the solo self-employed, who do not employ other workers), this is an additional force depressing the labour share. The dashed red line in Fig. 14 shows what the labour share would have been had the self-employed share remained the same and had their compensation per hour grown at the same rate as employees’. It is clear that the share is essentially the same in 2019 as in 1981, implying that all the fall in the official series is due to the changing fortunes of the self-employed.

The second important factor that the earlier analysis abstracted from is that there has been a large increase in ‘non-wage compensation’ for employees, primarily from employer pension contributions. This could be seen as a bona fide element of deferred compensation, but a substantial fraction of this compensation turns out to be corporate payments to deal with the historical under-funding of defined benefit occupational pension schemes (see Bell 2015). For example, British Airways’ ‘compensation’ includes payments to pilots who have already retired and are obviously not contributing to current output.

The dotted purple line in Fig. 14 takes out non-wage compensation from the official series (blue line) and shows a fall of around 3.5 percentage points in this wage share of GDP. We discuss how the labour share trends relate to changes in the markups, technology and monopsony in more detail below.

Labour share: dispersion and further analysis

In Appendix B, we present some further analysis of the labour share in Orbis, the trends by industry and also between-firm dispersion. As in the US, the fall in the labour share is strongest in manufacturing. We also find increases in the dispersion of labour shares across firms, especially when we weight by size and more strikingly in the upper tail. This increased inequality between firms is consistent with our findings on productivity, wages and markups that we have already documented.

Summary on labour share

The aggregate labour share has drifted downwards in the UK over the long run. Since the early 1980s, there has been some fall after adjustments for non-wage compensation and the self-employed, but these are much smaller than in the US. Firm dispersion of labour shares has also risen.

Firm size and industrial concentration

We next turn to inequalities in firm size. In the UK administrative data, there is an increase in the share of turnover in the top 100 UK enterprises between 1998 and 2018. This ignores the firm’s industry, and recent years have seen an enormous amount of work measuring industrial concentration, partly in response to concerns of weakening market competition. It is well known, of course, that concentration may be a poor measure of market power, but it is still useful to look at it, in the round.²⁷ Concentration measures are built from market shares—a firm’s sales relative to those of the whole industry. This is non-trivial to estimate for large multi-industry firms, as the HO data do not break down turnover across industries. Appendix A discusses methodological issues, but they are sufficiently difficult that we choose to rely on work by others to summarize what has been happening.

Unfortunately, there is no long time series of concentration in the UK that we can draw on as the ONS stopped publishing these data. Competition and Markets Authority (2020) is the most recent work which tracks the sales share of top firms by four-digit SIC industry (so about 615 distinct sectors) between 1998 and 2018. It finds an increase in average concentration from about 46–50% (see Fig. 15) using the turnover share of the largest 10 firms (the ‘C10’). The increase was mainly in the 2006–11 period around the financial crisis, peaking at 54% in 2011. Similar patterns emerge from alternative concentration measures using the C4, C20 or HHI (Herfindahl–Hirschman Index). Concentration declined somewhat after 2014.

The increase in concentration around the Great Recession has also been observed in other UK studies on the BSD data. Bell and Tomlinson (2018) find a growth in concentration focused around the Global Financial Crisis using a subsample of (consistently defined) five-digit sectors over a shorter period of 2003–15. Bahaj et al. (2019) for the Bank of England also look at the C5 and C20 between 1998 and 2017 within detailed sectors and find a substantial increase (see Appendix Fig. A18). BEIS (2020b) also finds concentration increased during the 2006–10 period.

The best effort yet to use company accounts data, rather than just administrative data, to analyse concentration is probably that by Koltay et al. (2020) from the European Commission’s DG-COMP. They look at concentration in the UK (together with France, Germany, Italy and Spain) between 1998 and 2017. To construct a firm’s industry share, one needs a good measure of the denominator (industry sales). The partial coverage of Orbis (particularly of SMEs due to the reporting issues discussed in Section 3) means that it is likely to be better to use sales or gross output from administrative sources. ONS-type industry statistics on sales are publicly available, so these are used as the denominator of concentration in Koltay et al. (2020).²⁸ As noted above, an issue with the numerator for concentration is that company accounts do not generally break down turnover by industry or geography. For example, the consolidated accounts of a multinational will report sales regardless of where the good was produced or sold. To deal with this problem, Koltay et al. use the ‘Industrial Passport’ data from Euromonitor, a marketing firm that tries to disaggregate firm-level sales by geography, industry and year for major companies using a variety of information in the accounts and specialist trade knowledge. Dividing the private economy into 156 ISIC sectors and using this method, they find that concentration increased substantially in the UK by almost 10 percentage points between 1998 and 2018 (see Appendix Fig. A19).

Rising concentration in the UK is consistent with the patterns observed in other countries. For example, Autor et al. (2020) construct concentration measures at the four-digit SIC level (676 industries) from the US Economic Census 1982–2012 and find substantial increases, regardless of precise measurement of concentration (as do Grullon et al. (2019)). Berlingieri et al. (2017) using administrative micro data from a large number of OECD countries (Multiprod) also find a trend towards increased concentration. Koltay et al. (2020) find that concentration has risen in the other European countries as well (although less so than in the UK). The ECB’s COMPNET team using firm data from the Eurozone countries also find concentration increases (Melitz 2020).

We reiterate that an important distinction has to be made between concentration in well-defined product markets used in antitrust analysis, and firm concentration in SIC industries (which will typically stretch across different antitrust markets). Both are of potential interest for efficiency as concentration may be relevant for input markets (such as labour) as well as output markets.²⁹

Business dynamism and entrepreneurship

The entry rate and share of activity in young firms have declined in the US since the early 1980s (Decker et al., 2014). In terms of absolute numbers, there is no indication of a decline in the start-up rates in the UK, at least since 2000. For example using all 6 million firms in Table 1 earlier, entry rates began at 12% in 2000 and ended at 13% in 2019, with a fall and then a rise. This exceeded death rates of around 10%. Consequently, the total number of firms in the UK economy has risen from 3.5 million to 6 million. However, this is mainly due to the rise of non-employer firms—as noted above, there has been a big increase in the fraction of UK workers who are self-employed and these are mainly ‘solo self-employed’, employing zero workers. Since the vast majority of self-employed firms stay small or exit, a better measure of entrepreneurship is the share of activity in young firms (see Davies 2021). The fraction of economic activity (as measured by employees and turnover) in UK firms that are less than 5 years old declined between 2000 and 2010 quite significantly. For example, about 14% of non-financial firm turnover was in young firms in 2000 and this fell to 7% by 2010. Although for employment this had recovered completely by 2018, the share of turnover was still only about 10%.

Masson and Shafat (2020) describe a new ONS administrative data set that tracks job transitions at the quarterly level (the Longitudinal Business Database, LBD). The LBD is constructed from the business register population (IDBR). Comparing the pre- and post-crisis periods (1999–2007 versus 2011–19), they show that job creation has fallen from 5.12% to 4.82% and job destruction has fallen from 4.71% to 4.37%: there are drops, but not large ones. The fall in job destruction is all on the extensive margin of less exit, which raises the concern of increasing numbers of ‘zombie’ firms. Davies (2021) using the (annual) BSD also finds some evidence of a decline in job reallocation (job creation plus destruction) since 1998.³⁰ By contrast, Oliveira-Cunha et al. (2021) argue that there is no consistent evidence for falls in this measure of reallocation.

Overall, different UK data sources and analysis do not paint such a clear picture of what is happening to entrepreneurship and business dynamism as the US. Therefore, this is an area where there is more ambiguity in the UK than in the USA.

Connecting labour and product markets: a simple framework

Overall, the evolution of the UK business landscape shows much similarity with the US (recall Table 2). One apparent difference, however, is that the employee wage bill share of GDP does not appear to have fallen in the UK as it has in the US. However, when we examined this in more detail, we did find some fall in the labour share between 1980 and 2019 when corrected for self-employment and non-wage compensation. Appendix C offers one simple imperfect competition framework for thinking of how to decompose the changes. There, we show that the change over time (⁠|$\Delta$|⁠) in the labour share of revenues (⁠|${S}_{it}$|⁠) of a single firm |$i$| at time |$t$| can be broken down as follows:

A fall in the labour share can come from: (i) changes in labour-saving technologies (automation) as indicated by a fall in |${\theta}_{it}$|⁠, the elasticity of output with respect to labour inputs; (ii) increases in firm product market power (markups) as indicated by a rise in |${\mu}_{it}$|⁠, the ratio of price to marginal cost; and/or (iii) increases in employer labour market power as indicated by a fall in |${\lambda}_{it}$| (marginal productivity of labour relative to the wage, the inverse ‘markdown’). As we discussed above, different authors have suggested all three as possible causes of the labour share fall, and of course these forces can interact. Moreover, at the macro level, we also have to aggregate across heterogeneous firms by size-weighting the equation, so a fall in the share can also come from a reallocation towards large firms which generally have low labour shares. This is the predominant force found by Autor et al. (2020).

Appendix C details how we can quantify the contribution of markups to the fall in the labour share. To summarize, we take equation 1 but assume the technology and monopsony terms are constant across time and firms and then aggregate across firms. We thus ask, holding technology and labour market fixed, how much the labour share would change as we let the markups change as observed. Recall from HO that there is a fall in the share of variable costs in revenues (as measured by the COGS/revenue ratio), which translates into an increase in estimated (COGS-weighted) markups of 0.44% per annum in the 1996–2016 period. Extrapolating this compound growth trend over the entire 1981–2019 period and calibrating to the 1996 labour share of GDP of 43.4% implies a 7.1 percentage point fall in the labour share according to equation 1.³¹ By contrast, the actual fall in the labour share was only 3.5 percentage points (see Fig. 14), about half as much.

Thus, the markup changes ‘over-explain’ the fall in the labour share in the sense that we would have expected the labour share to have fallen by an additional 3.6 (= 7.1–3.5) percentage points if product market power was the only factor. According to equation 1, the offsetting factors could have been monopsony power (markdowns) and/or technology. Markdowns of wages under marginal products may have become less pronounced due, for example, to the introduction in 1999 of the first UK National Minimum Wage (and its subsequent substantial increases). An alternative explanation is that technologies—which we assume are fixed and common for purposes of calibration—may have changed. This would require the output elasticity of labour to have risen, however, which is the opposite of the usual story of automation causing a substitution away from labour as in Acemoglu and Restrepo (2018, 2020).

There are at least four issues, however, with this calculation. First, extrapolating the markup trend backwards to the pre-1996 period where we do not have good Orbis data may be regarded as a stretch. We can, however, use the publicly listed firms in Worldscope, which has good coverage from about 1989 onwards. Although markups are higher amongst these larger firms (26% versus 19% on average), as we would expect, the markup increase is extremely similar to Orbis at 0.43% (instead of 0.44%) per year. Consequently, we get near-identical percentage point changes in implied labour shares using Worldscope even adding an additional 7 years. Second, the short-run time-series fluctuations of the implied and actual labour shares do not match up well. For example, the labour share was at a historical trough in 1996 according to ONS data and rose rapidly between 1996 and 2000, even though our estimated markups rose. Indeed, if we look at the whole 1996–2016 (or 1996–2019) period, the labour share does not fall on any of the measures regardless of corrections. Our view is that equation 1 should be regarded as a long-run trend, rather than something that can reliably match short-run fluctuations. Nevertheless, this issue is certainly worth further investigation. Third, recall that our markup estimates are based on incorporated firms in Orbis, in the market sector and for firms that report COGS and sales (Sample B). The aggregate GDP shares are based on a large number of different sources. In particular, the mixed income data are from non-incorporated firms, so we are implicitly assuming that markups are applicable to those in the mixed income sector. This may be inaccurate and it would be valuable to do more explicit work on the self-employed using administrative data. Finally, we have abstracted away from heterogeneity across sectors, which may be important.

In summary, despite these caveats, we believe that it is reasonable that the rise in markups helps account for a fall in the labour share since 1980. Our rough quantification suggests that there must be some offsetting factors to this pressure, reducing the magnitude of the implied fall in the labour share in the UK.

5. Explanations for the changes

There are many changes to understand and many theories to explain them. In this section, we draw some broad classes of explanations without attempting to be comprehensive. Most of the theories were developed to explain the US business trends. As we have seen, many (but not all) of these same trends are present in the UK data. We broadly group the theories of change into three classes—technology, globalization and institutions—but clearly these overlap, interact and are not mutually exclusive.

Technology-based theories

Network effects

Tirole (2024) has a detailed discussion on the economics of digital high-tech industries. A key feature of these sectors is the importance of network effects, which generate a tendency for these industries to be characterized by ‘winner takes all’ or ‘winner takes most’ competition. A classic example of network effects is internet search, currently dominated by Google. Consumers want to use a search engine that directs them to the website they want more quickly and accurately. The algorithm providing this will be more effective, all else equal, the larger the data set the company has on the history of consumer search. The more people use a particular platform for search, the more accurate will be the search engine’s algorithm (think of data-hungry machine learning techniques). As the quality of the search engine improves, even more people will tend to use it. Hence, a small initial difference in the quality of two search engines can lead the market to tip to one firm. Once a firm becomes dominant, it may be hard to dislodge if its endogenous advantage (data in this case) is not replicable/accessible by rivals. Other network effects in digital industries include social media platforms (e.g. Facebook), where the value to a consumer of the platform increases when other people use it, and operating systems, where developers will tend to create applications for the most popular platforms (e.g. Apple’s iOS on cell phones or Microsoft Windows on computers).

The growth of the GAFAM (Google, Apple, Facebook, Amazon and Microsoft) has often been linked to these network effects. In addition, the weight of the economy does seem to have shifted more towards these sectors. These are very large and profitable firms, tech titans which seem to have pulled away from both other firms in their industries and the economy as a whole. Whenever the issue of market power, and the likely remedies and implications, are discussed, these companies are usually assumed the targets. However, it is instructive to keep in mind that while the Big Tech have a major impact on the total market capitalization in the USA, high-tech industries employ only about 12% of the workforce (17 million workers in 2014) and contributed 23% of GDP (Wolf and Terrell 2016). The numbers are even smaller in the UK, more like 7.7% of GDP. While this share is non-negligible and moreover growing rapidly, this still leaves the vast bulk of the economy outside of these sectors. We have shown that many of the trends in productivity, wages, markups and concentration are at play in more traditional sectors of manufacturing, retail and transport. This means that, while there are surely factors specific to the high-tech industry worth studying, we cannot resort to these alone to explain the global phenomena.

Fixed costs: the importance of intangible capital

A rise in fixed costs will mean that low-productivity firms will struggle to survive, creating higher concentration and requiring a higher gross margin from survivors to cover the costs. This helps rationalize some of the trends, but what are the increasing fixed costs? Perhaps the most plausible story links fixed costs to intangible capital—e.g. software, R&D, marketing, training and management practices. Corrado and Hulten (2010) and Haskel and Westlake (2018) have shown that intangible investment exceeds tangible investment in the UK and the US today. In retail, for example, Walmart and other ‘Big Box’ grocers have made billions of dollars of investment in sophisticated inventory control systems to manage their ‘just-in-time’ global supply chains. These intangible investments do not typically show up in conventional measures of capital, and are largely fixed costs. Small chains, let alone Mom ‘n’ Pops, could not possibly make the same kind of investments. Intangibles may also make it easier for firms to expand into new geographical markets (Hsieh and Rossi-Hansberg 2019) or product markets (Aghion et al., 2019).

One of the main challenges in checking this hypothesis lies in measuring intangible capital. In an attempt to link variable profit margins to expenditures on intangibles, De Loecker et al. (2020a) correlate the markup with the share of SG&A over sales along the markup distribution over time. The results are quite stark: for firms in the lower part of the markup distribution, not much has happened to this ratio. However, for firms in upper percentiles of the markup distribution, not only did markups rise substantially, so did the ratio of SG&A to sales. Nevertheless, the rise in markups was still higher, suggesting that the return on these activities increased over time, as reflected in rising profit rates (as we showed in our EBITDA analysis in Section 4). At this point, however, the precise mechanism, and which elements are picked up in this accounting variable, are still very much understudied.

Slowing technological diffusion

In a dynamic economy, some firms are innovating and pushing the technological frontier forward and others are copying them and trying to catch up. New waves of firms build upon the innovations of current leaders and try to leapfrog them. This leads to productivity dispersion, but also with a pressure towards convergence through catch-up. If the process of diffusion has become harder, then leaders will pull away from laggards and we will observe increased inequality in size, productivity and markups. Akcigit and Ates (2019) argue that slowing technological diffusion accounts for the US trends better than any other single explanation, especially as it is consistent with slowing productivity growth.

One question, however, is: why should we think that diffusion has slowed? It seems surprising, as we live in a time where information spreads extremely quickly through electronic media and communication is much easier for geographically dispersed people. On the other hand, incumbency advantages (such as the proprietary data underlying internet search discussed above) may make challenging a leader harder than in the past. Further, firms may be using their intellectual property more effectively to prevent others using their innovations, due to changes in legal standards over patents (e.g. Kortum and Lerner 1999). Patents have become increasingly concentrated in top firms (Akcigit and Ates 2019), litigation has grown tremendously (Jaffe and Lerner 2006) and patent thickets can block potential start-ups. Leading firms increasingly buy up start-ups and acquire the talent, rather than these start-ups becoming larger firms to threaten dominant platforms. The incumbent firms may even kill off the technology post acquisition to prevent cannibalization of their existing products (see Cunningham et al. 2021, in pharmaceuticals).

Automation

Acemoglu and Restrepo (2018, 2020) argue for the importance of directed technological change in accounting for changes in the labour market. They stress the importance of the interaction between job displacement, which reduces labour demand, and job reinstatement, which increases it. The wage bill is the outcome of these two forces and the authors argue that the fall in the US labour share is largely due to the greater prevalence of ‘so-so’ technologies such as robots and automated checkouts that (they argue) reduce labour demand without much increasing productivity. Since the adoption of such technologies is most common in large firms, Acemoglu et al. (2020) argue that the reallocation effects generating greater dispersion are driven by these new technologies. There is controversy over whether recent technologies are as ‘so-so’ as Acemoglu and Restrepo argue. It took decades for organizations to learn how to effectively use information technology (and electricity), so it may be that many new technologies will have larger effects on productivity in the future (e.g. Bloom et al., 2012).

Globalization

Since the Second World War, countries have become increasingly economically integrated, with a fall in the costs of trade, communications and the mobility of people. This has been driven by technology, but also by international agreements to a rules-based trading system, WTO rounds of tariff and quota liberalization, and deep regional economic integration such as the European Union’s Single Market. China, in particular, became integrated into the world system following the change in policy after 1980 by Deng Xiaoping.

Although we may be witnessing some stalling or reversal of these trends in recent years with Brexit, Trump’s trade wars and the COVID-19 pandemic, globalization still could be a driver of the changes we have documented. One mechanism described in Autor et al. (2020) is that the reduction in trade costs will mean that firms that are more productive will expand their overall size at the expense of their less efficient rivals. This is generic in heterogeneous firm trade models and leads to increased concentration. Since high-productivity firms tend to have high markups,³² these reallocation effects push up aggregate markups and depress aggregate labour shares.

An alternative mechanism is that multinational firms can increase efficiency through global supply chains. This will lead to growing productivity differences with laggard firms as there may be a fixed cost to setting these up. This will enable them to grow larger and, if the lower costs are not fully passed through, will enable them to increase markups. There is some evidence on this mechanism in the literature. De Loecker et al. (2016) document rising markups for Indian manufacturing firms facing lower input tariffs on intermediate inputs. It is the incomplete pass-through of a cost reduction that is manifested in larger markups. These effects show up in rising aggregate markups, and rising returns to capital and stock market performance of Indian manufacturing firms.³³ Using US Bureau of Economic Analysis data, Keller and Yeaple (2020) find rising markups for US multinational affiliates and also an association with measures of domestic competition.

Institutions

There are a number of institutional changes that could explain the trends we see.

Antitrust policy

The most prominent story is that there has been a weakening of US competition policy (antitrust) enforcement since the 1970s. Wu (2018) argues that under the influence of the Chicago School of Law and Economics, judges and regulators moved towards a more laissez-faire attitude towards mergers and the activities of dominant firms. This enabled firms to gain higher markups and industries to become more concentrated. The increased common cross-ownership of large companies by institutional investors (such as Vanguard) may have also blunted competitive forces (e.g. Azar et al., 2018). Gutiérrez and Philippon (2018) have argued that weakening investment, productivity and wage growth are linked to the growth of US monopoly power. They contrast the US position with Europe. It is generally accepted that EU antitrust enforcement has not weakened over this period; if anything, the argument is often heard that it has got stronger with a single cross-EU enforcer (DG-COMP) and much political autonomy. Although Gutiérrez and Philippon concede that concentration and markups are not falling in Europe, they argue that the rise is smaller than that in the US. Our sense is that qualitatively, the trends are more similar than different in the UK, the EU and the US.

Regulations

A second institutional argument is that increased regulation may be a factor. An increase in regulatory burden may increase fixed costs as discussed above and help explain the trends. This is emphasized by Philippon (2019). It is unclear that regulations have increased so much over time. Although regulation increased after the Great Recession in finance (e.g. Dodd-Frank) and in healthcare (e.g. the Affordable Care Act), this was in the mid-2000s, whereas many of the trends of increased firm inequality and rising markups began in the early 1980s in the US.³⁴ Moreover, in the UK and the EU, there has been a downward trend in many kinds of regulation in an effort to increase market competition.

Labour market institutions

There has been a general decline in trade union power, which under some models could restrain markups through rent sharing. Since unions are located predominantly in large firms, their weakening could have helped big firms become more profitable. There is also a concern that monopsony power may have increased. The growth of firm size may have contributed to this—superstar firms could increase their markups over consumer prices and markdowns over the marginal product of labour. Although not an institution per se, there has been a big growth of the outsourcing of low-skilled occupations, such as cleaners, caterers, security guards and drivers. These outsourcing firms are very large employers and offer lower wages than direct hires.

One piece of evidence that does not sit well with a monopsonistic explanation is that Rinz (2018) and Crane and Decker (2019) find no systematic evidence of increases in local concentration of employment. It is possible that markdowns have increased despite the absence of concentration changes, however. Krueger and Ashenfelter (2018) point to the growth of no-poaching and non-compete arrangements, even for less skilled workers, and the increase in occupational licensing. Further, US job mobility has declined, as has the relative value of the minimum wage. These might add to employer power in the labour market, as Stansbury and Summers (2020) argue.

Although there is much evidence of some degree of monopsony power in US and UK labour markets (e.g. Manning 2011), there is not much robust direct evidence of increases in employer labour market power over time (i.e. markdowns of wages below the marginal product of labour). Lamadon et al. (2022) do not find evidence of trends over time in their analysis from the late 1990s onward in the US. Furthermore, with the exception of unions, these institutional trends are very different across countries. For example, the first national minimum wage was introduced in 1999 in the UK, and this has been strongly uprated over time, so now the UK minimum wage is high by OECD standards. Indeed, this may have reduced monopsony power and could be a reason why the labour share in the UK has not fallen by as much as we would expect given our estimated rise in markups (recall the discussion in Section 4 above).

Summary on institutions

In our view, institutional explanations do not seem the most compelling general explanation for the trends, which have broadly taken place across a number of OECD countries. This is not to say that differences in antitrust, and various forms of government interventions, do not play some role in explaining the differences, as they are likely to interact at a more subtle level.³⁵

Assessment

In terms of explaining the causes of increasing firm inequality, there are almost too many plausible stories. There is no one obvious ‘winner’ that stands out. We would instead make some broad points.

Industry heterogeneity

It is likely that there are different stories in different industries. For example, it does seem clear to us that one fundamental reason for the dominance of superstar firms (and their high margins) is the ‘winner takes most’ nature of many digital sectors due to network effects. Equally, the institutional explanation of weak antitrust enforcement allowing the creation of more market power seems true in sectors, such as US hospitals and communications. So seeking a ‘one size fits all’ explanation of the trends, although attractive in its parsimony, may not be credible.

Purely institutional explanations

As noted above, many of the trends are best documented in the US and explanations have naturally focused on this country. To the extent that these trends have similarities in other countries (which have a welter of different changes and levels of institutional types), one would expect something more fundamental is happening with respect to technologies or globalization, rather than American institutions. Although there are clearly differences in the rates, timing and magnitudes of change, our sense is that there are greater similarities between the USA and other developed nations. Overall, we are therefore sceptical about explanations that rely solely on changes in US institutions, such as declining antitrust enforcement. Add to this that in order to be definitive on the underlying sources, at the very least we should confront this leading candidate with the various stylized facts, and be able to jointly explain them.

Technology

In addition to the clear importance of network effects in the digital sector, our view is that the fixed costs of adopting various forms of intangible capital are also important. There are multiple sources of evidence of the positive correlation between markups/productivity and higher intangible capital intensity (e.g. De Loecker et al., 2020a).

Globalization

The stronger markup growth for listed firms (which are mainly multinational) compared with unlisted firms in the UK hints that globalization may have played an important role. This should be an area for future study. The analysis so far seems to suggest that these groups have seen distinct markup trajectories.

6. Policy implications

Given the uncertainty of the causal drivers of the changes in firm inequality, it is hard to give definitive policy recommendations. However, the facts we have documented should help inform policy options and we make four basic points here.

First, we have argued that declining enforcement of competition policy is probably not the main explanation for the trends we have observed, as so many of them have occurred across different countries. Few observers claim that the EU and the UK have had a significant decline in antitrust enforcement (as some argue is the case in the USA). Nonetheless, since the UK, like the USA, has witnessed a growing dominance of large firms with high markups, this raises concerns about a general rise in market power. Such ‘superstar’ companies do have some capability to undermine the welfare of consumers and workers, even if they gained their powerful positions primarily through the forces of technological change and globalization.

So what are the tools for strengthening competition? Most directly, there are many ways to reform and modernize competition policy and we concur with many of the recommendations by Tirole (2024). For example, the standards of proof in merger and acquisition cases when dominant companies (especially in digital sectors) take over start-ups/smaller firms should be shifted towards the merging parties and away from the government agencies. Too often regulators have to provide beyond doubt that the acquisition will do harm, which is extremely hard to demonstrate empirically in high-tech fast-moving sectors. Instead, there should be more burden on the merging parties to reassure agencies that harm is unlikely. While this notion has long been discussed among antitrust scholars, under ‘structural presumptions’, given the bulk of evidence suggesting a fundamentally different landscape across industries, we believe it should be strengthened (e.g. see Hovenkamp and Shapiro 2017). Moreover, we also support strengthening the Competition and Markets Authority (CMA), with strong enforcement powers given to the new Digital Markets Unit, including an explicit reference to consider innovation and future competition in their decisions.³⁶

Another way to strengthen competition is to reduce regulatory barriers to entry and increase openness to trade and foreign direct investment. For example, Brexit has significantly increased barriers to trade with the EU by increasing regulatory red tape (non-tariff barriers). Since the forces of economic gravity inevitably imply that the EU will be Britain's largest trading partner for the foreseeable future, the UK should seek to rejoin the Single Market alongside non-EU members of the European Economic Area, such as Norway. This would be an effective spur to competition and productivity. If political constraints prevent this (and even better a rejoining of the EU), trade frictions (which are primarily regulatory differences with the EU) should be reduced to the absolute minimum.

Second, we have emphasized that the UK's biggest economic problem since the financial crisis has been the stagnation of productivity that has dragged down average wage growth. Strengthening competition is only one of many possible tools to improve productivity. The conclusions of the LSE Growth Commission (Aghion et al., 2013) still hold true: Britain has too little long-term investment in the critical areas of innovation, skills and infrastructure. Specific policies in these areas are covered in the studies of Aghion et al. (2013), Bloom et al. (2019) and Scur et al. (2021). However, a general point is that the UK suffers a kind of policy ‘attention deficit disorder’—a near-constant shifting, rebranding and elimination of policies as the political and media winds change. Building institutions (such as the Bank of England and the CMA) that have some independence and expertise is a way of reducing policy uncertainty and increasing investment. We welcome the new Infrastructure Bank, which joins the Infrastructure Commission, for example, as we need to create a new institutional architecture for long-term decisions over transport and energy (e.g. to tackle climate change). By contrast, the almost overnight abolition of the Industrial Strategy Council in 2021 was a case study of how to waste business effort and create greater unnecessary uncertainty.

Third, a theme of our analysis is that there has been an increase in inequality between firms (e.g. in productivity, wages and markups), especially at the top of the distribution. As noted at the start of this article, the welfare implications of increased firm inequality are murkier than those of income inequality between households, but increased firm dispersion is often a sign of increased market frictions. For example, if the gap between the ‘best and the rest’ is due to slowing diffusion of best practices from frontier firms to followers, this will likely depress aggregate productivity (and wages). Similarly, if widening performance gaps reflect increasing problems facing SMEs in scaling up due to inadequate access to financial, human or managerial capital, this misallocation reduces growth. Strengthening competition may again help reduce distortions (especially if dominant firms are keeping rivals back through deliberate actions) and foster catch-up, but there are more direct policy measures. For example, Scur et al. (2021) evaluate various policies to spread better management practices. They argue that consultancy/mentoring interventions have a better track record than the more classroom-based approach in the government’s new £0.5 billion ‘Help To Grow’ programme. Putting aside some resources—say 5–10% of the total—to pilot and evaluate these interventions alongside the main programme would be extremely valuable.

It is worth emphasizing that policies to help catch-up are valuable even if the UK does not have a particularly ‘long tail’ of low-productivity firms compared with other countries and even if firm dispersion had not grown larger. There is a huge dispersion of firm productivity levels, so cost-effective policies that address market failures, increase diffusion and reduce market frictions can be justified regardless of whether the UK is relatively ‘worse’ in any cross-sectional or time-series sense (Oliveira-Cunha et al., 2021).

Finally, as we have discussed, there has been a substantial increase in wage inequality since the late 1970s (see Fig. 5). One factor influencing wage inequality is institutions, so a possible countervailing force to the power of dominant firms is to strengthen labour. For example, reducing restrictions on the mobility of labour (e.g. non-compete and non-poaching clauses as discussed by Krueger and Ashenfelter (2018)), maintaining and expanding labour standards (e.g. to gig-economy workers as in the 2021 UK Supreme Court decision to give Uber drivers employee rights), providing greater protection for unions, and having decent minimum wages are all ways of trying to strengthen the power of workers. Ultimately, however, improving the human capital of workers (especially at the middle and bottom end of the distribution) is the surest way to raise wages and reduce inequality (Blundell et al., 2021).

7. Conclusions

Increasing inequality among people is much more concerning than high and growing inequality between firms. Nevertheless, there can be negative effects of higher firm dispersion if this is associated with greater market power, slower productivity (and therefore wage) growth and increased income inequality. Extensive US evidence in recent years has shown that there have been major changes in inequality across firms in a number of dimensions—productivity, wages, markups, firm size and labour shares. These changes have been accompanied by worrying macro trends, such as slowing productivity growth, stagnant median wage growth, rising markups and declining business dynamism.

In this article, we have looked into the current state of the evidence in the UK. We focus on a new analysis of a near-population panel of company accounts since 1996: Historical Orbis. Using HO and a range of other evidence from administrative sources at the micro and macro levels, as well as critically reviewing the existing literature, we find that the American trends are broadly also present in the UK.

We emphasize that there may be different things driving changes in different industries. We argue that institutional changes are less likely to be the explanation for these changes than more fundamental technological reasons, as we see common changes in the business landscape in the UK and the USA. A role for globalization is also likely, but it is also less explored. Policy-wise, it is important to modernize antitrust and improve the power of labour as a countervailing force.

Finally, it is worth highlighting the stagnation of median and average real wages in the UK since 2007. We believe that the major cause of this is near-stagnant productivity growth since the Global Financial Crisis—a phenomenon that has been much worse in the UK than in other advanced nations. Although there has been some fall in the labour share of GDP since 1980, this is less clear than in the USA. Slow aggregate productivity growth is perhaps the most severe long-term problem faced by the UK over the last 14 years. Policies to reignite productivity growth through stronger innovation and diffusion are crucial in building a sustainable and equitable economy.

ACKNOWLEDGEMENTS

Helpful comments on earlier drafts have been received from Philippe Aghion, Orazio Attanasio, Richard Blundell, Angus Deaton, Jan Eeckhout, Penny Goldberg, Rachel Griffith, Paul Johnson, Colin Mayer, Mariana Mazzucato, Juliana Oliveira-Cunha, Andreas Teichgraeber, Greg Thwaites, Jean Tirole and participants at presentations at the 2020 and 2022 Annual Meeting of the American Economic Association (AEA) and at the Institute for Fiscal Studies (IFS). Luis Moyano García provided excellent research assistance. The authors are grateful for support from the IFS Deaton Review of Inequalities and for financial support from the Economic and Social Research Council through POID. De Loecker gratefully acknowledges support from the ERC, Consolidator grant 816638 (MPOWER).

FUNDING

This article was written for the IFS Deaton Review of Inequalities, funded by the Nuffield Foundation (grant reference WEL/43603). The views expressed are those of the authors and not necessarily the Foundation.

CONFLICT OF INTEREST

None declared.

Footnotes

REFERENCES