RDP 2018-10: Wage Growth Puzzles and Technology 3. Technology and Labour's Declining Share

As noted earlier, labour's share of national income has been falling in most developed countries since around the mid 1980s. Not surprisingly given its significance both economically and politically, considerable research work has been undertaken trying to explain this observed phenomenon. Some of that research has focused on measurement issues, suggesting that the decline can be partly or largely explained by the treatment of capital depreciation, income from housing, income of the self-employed, or intangible capital.[12] However, while such ‘explanations’ may have some validity for some countries at certain historical periods, the continuation of the declining trend across so many countries has led to a general consensus that, data queries aside, the fall in labour share is real and significant and requires more fundamental explanations (Autor et al 2017, p 3). In addition, such ‘measurement issue’ explanations do not fit easily with the now widely accepted fact, further discussed below, that within-industry rather than between-industry changes account for the vast bulk of the shift in labour's share.

As was the case in relation to work on structural explanations of low nominal wages growth in recent years, many – indeed, most – of the causal factors examined in explaining the decline in labour's share relate, directly or indirectly, to technology and its effects. Indeed, there is a close overlap in the structural factors examined with respect to short- and long-run wage puzzles, which is not surprising given the importance in both cases of shifts in bargaining power. However, a notable exception is the most recent work linking labour's declining share to the uneven take-up of technology and the resultant growing productivity gaps between firms in the same industry. Little if any work appears to have been done examining whether this framework can also help explain the more recent puzzle of very low nominal wages growth: the focus of Section 4 below.

In examining and evaluating the many and varied explanations for the trend decline in labour's share, a number of useful filters can be used. The primary ones are as follows.

Firstly, a growing number of shift/share studies have shown that the decline in labour shares in the United States and in most other countries is primarily a within-industry phenomenon, not a between-industry one (e.g. OECD 2012; Elsby et al 2013; Karabarbounis and Neiman 2013; IMF 2017b). More recently, Autor et al (2017) have taken this analysis one step further and found that in the United States and (at least within the manufacturing sector) other countries examined, the fall in labour's share is driven primarily by between-firm reallocation within an industry, rather than by the within-firm component.[13] Taking these two sets of findings together, explanations of declining labour shares that focus on shifts in activity between labour and capital intensive industries, or on factors such as trade exposure or outsourcing that should impact equally across all or most firms within an industry, are unlikely to identify the key drivers, at least for most countries.[14]

Secondly, most empirical studies suggest that the elasticity of substitution of capital for labour is less than one.[15] Consequently, explanations of factor share shifts that focus on the decline in the cost of capital relative to labour at an aggregate level are also questionable, although it is worth noting that automation and artificial intelligence may over time raise the elasticity of substitution in at least some industries and for some workers.

Thirdly, given that the declining labour share has been observed across many countries, developed and developing, typically since around the 1980s, causal factors that are specific to only a few countries, or cover a shorter or quite different time period, are unlikely to provide a convincing explanation.

Fourthly, and as noted earlier, explanations for significant changes in factor shares must ultimately relate, directly or indirectly, to factors influencing the relative bargaining power of labour and capital over the distribution of productivity gains. Given that a declining labour share has been observed in many countries since around the 1980s, it seems reasonable to conclude that a key part of the explanation revolves around ongoing structural changes in labour and product markets that are impacting on the bargaining power relationship between management and labour within many firms and industries over roughly this same period.

The following literature survey of overseas studies linking labour's declining share to technology-related factors is far from comprehensive, aiming instead to briefly evaluate the most widely cited technology-related studies in order to put the more recent productivity divergence explanations that are of primary interest into a broader context.[16]

3.1 Globalisation: Supply of Cheap Labour, Increased Import Competition and Outsourcing

While the digital revolution has greatly facilitated globalisation and the increased integration of both labour and product markets, it is but one of many factors that have facilitated globalisation more broadly. The integration of China into the global economy since the fundamental reforms that began in the late 1970s and the dismantling of many barriers to both trade and capital flows in recent decades have, by way of example, also been major factors behind increased globalisation.

A number of studies have examined the impact of increased supply of low-wage labour, increased international competition, trade exposure and outsourcing on labour's share. Phelps (cited in Ellis and Smith (2007)) suggested that the entry of China and eastern Europe into the global economy has increased the supply of low-wage, low-skilled labour without an equivalent increase in the capital stock, thereby reducing the relative price of labour. If the elasticity of substitution is less than one – which most of the empirical literature suggests is the case – this would lower the labour share of national income. Closely related to this argument are studies suggesting that increased import competition (e.g. Harrison 2002; IMF 2007; OECD 2012, p 130) and offshoring (e.g. Jaumotte and Tytell 2007; OECD 2012; Elsby et al 2013; Autor et al 2017, p 2) can explain some or most of the fall in labour's share. By way of example, Abdih and Danninger (2017), in following up on an earlier IMF (2017b) study and using industry-level data for the United States for 2001 to 2014, found that, across a number of different model specifications and different proxy measures of the explanatory variables, import competition and the use of foreign inputs jointly explained between 41 and 51 per cent of the decline in labour's share.[17]

Looked at in a bargaining context, outsourcing, supply chains and rising import competition have all had a significant impact on both product and labour markets. However, while there is certainly evidence that outsourcing, supply chains and import competition have lowered both perceived job security and union bargaining power and raised the elasticity of labour demand with respect to labour costs (e.g. Rodrik 1997; Dumont, Rayp and Willemé 2006; Abraham, Konings and Vanormelingen 2009; Bloom, Draca and Van Reenen 2011; Boulhol, Dobbelaere and Maioli 2011; Goldschmidt and Schmieder 2017; OECD 2017), it should also have affected profit margins. Whether on balance this is likely to have raised or lowered labour's share of national income is thus unclear a priori and is an empirical matter.

At a conceptual level, it is difficult to reconcile explanations centred on trade exposure or outsourcing with the finding referred to earlier that most of the ‘action’ on falling labour shares is happening between firms in a given industry, rather than across all or most firms in an industry. It is not at all obvious why different firms in the same industry should face quite different levels of import competition or have very different opportunities to outsource. Thus it is perhaps not surprising that more detailed studies using micro-level data sources have failed to find a significant relationship between industries or firms with greater exposure to trade shocks, import competition and outsourcing on the one hand and the extent of the decline in labour's share on the other.[18]

In addition, most of these ‘globalisation’ explanations are not easy to square with the observation that labour-abundant countries, including China, India and Mexico, have also experienced significant declines in their labour shares (Karabarbounis and Neiman 2013, fn 1).

Finally, Ellis and Smith (2007) have pointed out that, if the explanation for the declining labour share centred on increased global competition and expansion of the global, low-wage labour supply, one might have expected the relative price of investment goods to have risen, whereas in fact the opposite has been the case.

3.2 Bargaining Power: Union Membership and Market Deregulation

Links between globalisation and increased global competition and reductions in workers' bargaining power are likely to exist regardless of the particular type or level of union coverage, bargaining arrangements or institutions in place. This section looks more specifically at research on whether changes in institutional arrangements in labour markets, and differences across countries, may be an important independent factor behind the declining labour share. Once again, while technological change has undoubtedly been an important factor behind some of these institutional changes, it is but one of many factors involved.

Studies in this area have often been more qualitative than quantitative, reflecting the difficulty of finding adequate proxy variables for measuring differences in bargaining arrangements and their impact on employee bargaining power.

An early study by Blanchard and Giavazzi (2003) focused on deregulation in product and labour markets and associated shifts in bargaining power in Europe in the 1980s. Similarly, Giammarioli et al (2002) linked the decline in labour's share in Europe to labour market deregulation and declining union membership. However, these studies fail to explain the pervasiveness across economies, geographical regions and time of declining labour shares. They also sit uncomfortably with more recent findings that most of the action with respect to the falling labour share is occurring between firms in a given industry.

Declining union membership and labour market deregulation are common features across many western economies over recent decades. The OECD (2012) provides a useful summary of some of the main factors that have been driving this trend, including the increasing use of ICT and associated ‘casualisation’ of the labour force discussed earlier. While some studies have found evidence of some relationship between union density and factor income shares (e.g. Charpe 2011; Stockhammer 2017), the evidence is at best mixed. This is not surprising: as the OECD 2012 study notes, there is no close relationship between the evolution of union coverage across countries and the extent of collective bargaining coverage, because many factors beyond level of trade union membership can affect collective bargaining coverage.[19] Nor does shifting attention to collective bargaining coverage itself help much. Noting the highly divergent trends in collective bargaining coverage across countries, the OECD finds no clear correlation between bargaining coverage and shifts in labour's share in the business sector. Indeed, they note that:

… some of the largest decreases in the business-sector labour share occurred in countries where collective bargaining coverage increased or decreased only slightly, such as Finland, Sweden and Italy. (OECD 2012, p 138)

With respect to Australia, the study found that, while institutional changes over the period examined (1990 to 2007) resulted in by far the largest drop in collective bargaining coverage of the countries examined, the decline in labour's share was around average.

Increasingly, across many OECD countries, collective bargaining has become more decentralised, occurring more at the firm/establishment level rather than the industry level (OECD 2012, p 139). While a number of studies have found that this has increased the link between wages and local performance, there is little if any evidence of a significant link between decentralisation of bargaining and shifts in the labour share (OECD 2012, p 142).

Finally, these institutional explanations are also difficult to reconcile with the observation that changes between firms, rather than across whole industries or resulting from intra-industry shifts in activity, are at the heart of the decline in labour's share.

In summary, the common experience of a declining labour share across a wide range of countries with differing bargaining institutions and degrees of collective bargaining and union membership suggest this is not fertile ground for further exploration. Rather, the factors ultimately driving shifts in bargaining power would appear to be pervasive and operating regardless of differences or changes in institutional bargaining arrangements across time and countries.

3.3 Capital Accumulation, Technical Change, Relative Factor Prices and ICT: Earlier Studies

From early in the piece, a number of studies looking at reasons for labour's declining share focused on issues relating to capital accumulation, technical change and the relative price of capital and labour. Acemoglu (2003) and Bentolila and Saint-Paul (2003) examine factors which could lead to deviations from standard growth model stable paths in which labour is paid its marginal product and either technical change is labour augmenting and does not impact on factor shares, or the production function is Cobb-Douglas (elasticity of substitution of one) and neither capital nor labour-augmenting technical change affects factor shares. Relaxing these conditions to allow for deviations from stable growth paths, an increase in capital intensity results in an increase in both the capital-to-labour ratio and the marginal product of labour. If capital and labour are complementary – that is, the elasticity of substitution is less than one – the increase in real wages will more than offset the decline in labour intensity, and labour's share will rise. But if labour and capital are substitutes – the elasticity of substitution is greater than one – labour's share will fall.

Within this framework, the impact of capital-augmenting technical change is the same as rising capital intensity. Using a measure of total factor productivity (TFP) as a proxy for technical change – a rough proxy at best – Bentolila and Saint-Paul (2003) find that growth in capital intensity and TFP more than account for the fall in labour's share in OECD countries between 1972 and 1993. Follow-up analysis by the OECD (2012) using a similar framework reached roughly similar results, but importantly they used firm-level data rather than macro data to confirm the earlier finding holds on a within-industry basis.

Similarly, in a more recent and widely cited article, Karabarbounis and Neiman (2013) argue that the rapid decline in quality-adjusted equipment prices, in particular due to the spread of ICT, has driven a decline in the cost of capital relative to labour and a rise in the capital-to-labour ratio.[20] If the elasticity of substitution is greater than one, as their cross-country estimations suggest is the case, this would lead to a decline in the labour share.

How do these explanations fit with the view expressed earlier that, ultimately, significant and lasting changes in factor shares must reflect, directly or indirectly, changes in bargaining power? Many of the earlier technology studies do not even mention bargaining power, and seem to view technology and technical change as somehow independent of it. Ultimately, however, any impact must reflect the fact that labour has not benefited as much as capital from the particular nature of technological change in the production process, due to declining employee bargaining power. In short: ‘Technology must be interpreted as a factor influencing bargaining positions rather than a mechanical process determining distribution outcomes’ (Guschanski and Onaran 2018, p 6).[21]

On the surface at least, a fundamental problem with many of the studies focused on the declining cost of capital and rising capital-to-labour ratios is that, as noted above, the bulk of the empirical evidence at the macro level suggests that the elasticity of substitution is less than one. In addition, for a decline in the relative price of investment goods to lead to an ongoing trend decline in labour's share, an elasticity of substitution greater than one is a necessary but not sufficient condition: it is also necessary for the capital-to-labour ratio to continue trending upwards over the period observed, which in the case of the declining labour share is the period since the 1980s. Such a requirement does not a priori seem consistent with the substantial slowdown in labour productivity growth since the 1990s observed across many economies.

It is possible that more detailed firm-level examination of the role played by ICT may help explain the apparent inconsistency between the above empirical results and the filters outlined earlier. The ongoing take-up of ICT may have resulted in technical change in recent decades being capital augmenting, as some studies have indeed found (e.g. Arpaia, Pérez and Pichelmann 2009; Tyers and Zhou 2017). In addition, ongoing technological progress with respect to robotics and artificial intelligence may over time raise the substitutability of capital for labour in some firms and industries and for some employees, in particular lower-skilled non-manual workers.[22], [23] If, as more recent work examined below suggests is the case, these changes in technology are occurring very unevenly across firms, with the ICT take-up leading to increased capital/labour substitutability in the high productivity firms but not in the majority of technologically laggard firms, it may be possible to reconcile the results of these studies with the observation that, at a macro level, most estimates find the elasticity of substitution is less than one; and also with the observation that, despite the increased take-up of ICT, productivity growth at the macro level has slowed in recent decades across many countries.

The key to whether these observations can be reconciled thus lies in further studies at the firm level into the uneven impact of ICT on capital intensity, technical change, productivity growth and elasticities of substitution.

An interesting early study that did not suffer from these limitations and that placed technological change directly in the context of shifts in bargaining power was Ellis and Smith (2007). Hornstein, Krusell and Violante (2002), using a putty/clay model of capital accumulation made famous by Solow (1962), argue that higher innovation rates will result not only in faster rates of capital obsolescence and turnover but also faster churn in the labour market. This shifts bargaining power towards capital and away from labour, with the shift further exacerbated by any labour market rigidities that increase frictional unemployment and by any product market regulation that reduces competition. Building on this analysis, Ellis and Smith suggest that the increased use of ICT, and the incorporation of ICT into a wider range of capital goods, may have been a primary and ongoing factor driving the above process. While the data available to them precluded a direct test of their hypothesis, they note that it fits well with cross-country variations in factor shares: the decline in labour's share, they find, was greatest in countries with greater labour market rigidity and more product market regulation. They also show that, after controlling for other common explanations of labour's declining share – in particular, expansion of the global labour supply, changes in relative factor prices and exogenous shifts in bargaining power – there is still a significant trend decline in labour's share to be explained.

While not directly tested at either a micro or macro level, the Ellis and Smith (2007) hypothesis is consistent with the findings of more recent micro-level studies on what is driving factor shares, which are examined below.

3.4 ICT and Micro-level Evidence: Recent Studies

A number of promising studies of what is driving factor share shifts have been sparked by some seminal OECD research that looked at the diffusion of new technology across firms in 23 OECD countries (Andrews, Criscuolo and Gal 2015).[24]

Using a harmonised, cross-country firm-level database for 23 countries, the study concentrates on the relative productivity performance of global frontier firms (i.e. the most productive globally in terms of both labour and multifactor productivity), national frontier firms and laggards. Their main findings were:

  • despite the aggregate level slowdown in productivity growth in the 2000s, productivity growth at the global frontier level remained robust;
  • the gap between the high productivity firms and the rest has been rising over time;[25] and
  • global frontier firms are not only more productive than laggard firms but also more capital intensive, more patent intensive, have larger sales and are more profitable. However – and importantly both in terms of explaining the aggregate productivity growth slowdown and thinking about implications for average wage growth – they are not significantly larger than the laggards in terms of employment. Consistent with this last point, in their examination of two potentially key sources of productivity gaps across countries – namely, gaps originating from differences in the productivity levels of global versus national frontier firms and those originating from differences in size measured in terms of employment – the first was found to be generally much more significant.

These findings are significant on many fronts. In the context of this paper, two in particular stand out. First, the growing gap between frontier firms and laggards in terms of productivity may help resolve the debate between the technology optimists and pessimists with respect to the ongoing importance or otherwise of the ICT ‘revolution’. New technology is still having a positive impact on productivity growth, but primarily amongst frontier firms; while the growing gap between them and the laggards, along with the fact that the frontier firms are not significantly larger in terms of employment, may explain why aggregate productivity growth has been slowing down (Haldane 2017a).

The second potentially interesting implication is that frontier firms are not only more productive but also more profitable and have larger sales than laggards. This suggests that much of the additional productivity gain is being absorbed by a combination of lower prices (encouraging higher sales) and higher profits, as against higher real wages. If frontier firms are not increasing real wages in line with labour productivity growth and laggard firms simply don't have the capacity to do so as they are under increasing pressure to maintain their market share and survive, this may help explain the short-run low nominal wage growth puzzle in a low inflation environment. This implication is returned to below.

The findings of this path-breaking OECD research have, not surprisingly, spawned further research work looking at whether the uneven technology take-up can help explain the decline in labour's share across many countries. The most important and widely quoted study is that of Autor et al (2017). They develop a ‘superstar firm’ model based on the idea that ‘… industries are increasingly characterized by a “winner takes most” feature where a small number of firms gain a very large share of the market’ (p 2). Their model, which is set out in more detail in Section 4, assumes competitive factor markets but allows for imperfect competition in product markets. Specifically, and consistent with growing evidence, including the OECD paper just discussed, the model allows for heterogeneous productivity performance in which more productive firms have a higher level of factor inputs and sales. They also assume that some labour input is fixed and some is variable, such that firms will have a lower labour share of value added if they are larger and/or if they have a higher mark-up. For both these reasons, superstar firms are seen as likely to have lower labour shares of value added.

Their model generates a number of testable hypotheses:

  1. If technological change advantages the most productive firms in each industry, product market concentration will rise as industries become increasingly dominated by superstar firms.
  2. Industries where concentration rises the most will have the largest decline in labour's share in value added.
  3. The fall in labour's share will be driven largely by between-firm reallocation rather than by a fall in the mean unweighted labour share within firms.
  4. The between-firm reallocation component of the fall in labour's share will be greatest in sectors with the largest increase in market concentration.
  5. The above patterns should be observed not just in the United States but also internationally.

Using firm- and industry-level data, they find statistical support for each of these hypotheses for the United States. Due to data limitations only the second and fourth hypotheses were tested for a number of European countries. They were both confirmed for most of the countries examined.

Whilst the authors see the ‘superstar firm’ model and associated rising industry concentration as most obviously applicable to high-tech industries, it could be a reflection of other factors, such as barriers to entry. In order to test whether rising concentration is most prevalent in dynamic industries exhibiting rapid technological change, they map it against two proxy measures of technical change, namely patent intensity and TFP. They find significant positive correlations in both cases.

In contrast to the explanations considered, the ‘superstar’ explanation of labour's declining share meets the requirements of all the filters outlined earlier. It is consistent with the shift/share analyses showing that the action on declining shares is concentrated within as against between industries, and at the industry level is occurring mainly between firms rather than evenly across all firms. It does not rely on an aggregate level elasticity of substitution greater than one, although the model does tend to imply that, for the technological frontier firms, more successful adoption of new technology is likely to have increased the substitutability of capital for labour – at the extreme, by way of automation of some jobs and tasks.

The analysis also fits the time profile for a declining wage share well. Most studies on the so-called ICT revolution see it as commencing around the 1980s, and as having a significant and ongoing impact on both labour and product markets in many developing as well as developed economies.

Finally, while the superstar firm model is not couched directly in terms of relative bargaining power of labour versus capital, it has a number of important links to that framework. While there is some evidence that employee/union bargaining power is greater in larger firms[26], the Autor et al (2017) study suggests that the increased market power of leading technology firms allows them to use most of their higher productivity gains partly to lower prices and partly to increase their profit margins, rather than passing most of it through to their employees.[27] If at the same time the technological laggards are struggling to maintain their market share, they will simply not have the capacity to pay higher wages; while perceived lower job security in such firms is likely to reduce union/employee bargaining power.

The Autor et al analysis is consistent with some earlier but less comprehensive US studies. Peltzman (2014) found that US industries that experienced the largest increase in concentration also experienced the strongest increases in output prices, suggesting increased mark-ups and a lower within-firm labour share. Barkai (2016), noting increasing levels of industry concentration in the United States, found a relationship between the decline in labour's share and increased profit mark-ups, but does not examine whether the latter is occurring across most firms or primarily among larger superstar firms.

Since the Autor et al (2017) paper, a number of follow-up papers have been written that, for the most part, confirm and build on their work. A few, however, have questioned their results.

De Loecker and Eckhout (2017) document trends in firm-level mark-ups in the United States for the period since 1950, and find that, after a period of relative stability, they started to trend upwards from around 1980: roughly when the labour share started to fall. They also find a close negative correlation between the aggregate labour share in the United States and an aggregated measure of firm mark-ups, although they make no attempt to control for other variables.

Benmelech, Bergman and Kim (2018), noting that US labour market mobility has declined significantly in recent decades, examine the impact of local-level employer concentration on wages in the US manufacturing sector. Using employment as their measure of local-level industry concentration, they find a significant negative relationship between it and wages that is more pronounced at high levels of concentration and has increased over the period examined (1977 to 2009). Their results are robust after controlling for other factors such as establishment-level labour productivity and local labour market size. They also find that this negative relationship is stronger when unionisation rates are low; and that the link between productivity growth and wages growth is stronger when labour markets are less concentrated.

A recent OECD (2018) study uses a firm-level database to confirm the ‘superstar’ hypothesis for the eight countries they examine in which labour's share fell over the period 2001–13. Interestingly, their results suggest that the decline in labour's share at the technological frontier is largely driven by the entry of new firms with higher mark-ups and/or higher capital intensity, rather than increasing mark-ups or capital intensity in firms remaining at the technological frontier.

A weakness of the superstar firm model is that it assumes an elasticity of substitution of one, thereby excluding any possible role for capital intensity in their explanation of shifts in labour's share. Adrjan (2018), in a longitudinal study using UK firm-level data, uses a model of the firm in which the labour market is perfectly competitive and product markets are not but, unlike Autor et al (2017), the elasticity of substitution is not constrained to one. Market power can affect labour's share because it allows a firm to set a higher profit mark-up over cost. But capital intensity can affect labour's share in a firm as well, depending on the elasticity of substitution. His empirical analysis finds that both market share of a firm and its capital intensity are highly significant as explanators of labour's share, with market share being statistically more significant than capital intensity. These results are found to be highly robust to alternative econometric methods, data definitions and sample periods.

Adrjan also examines variations in elasticities of substitution across firms. Using average firm wage levels as a proxy for average skill levels, he finds that the negative impact of capital intensity on labour share is driven by low-wage firms, which he interprets as implying that low-skilled labour is more substitutable with capital. Alvarez-Cuadrado, Van Long and Poschke (2018) show that a model allowing for cross-sectoral differences in productivity growth and in the degree of capital/labour substitution can explain the movement in labour's share in the United States.

These studies help to integrate the ‘superstar’ analysis with some of the earlier studies on the impact of capital-augmenting technical change and capital-to-labour ratios on labour's share.

A few papers that have followed on from Autor et al (2017) have questioned some of their findings. A key testable hypothesis flowing from the superstar firm model is that, over the periods of declining labour share, industry concentration measures should be rising. Bessen (2017) confirms the Autor et al finding that this is indeed the case in the United States: on his estimates, both industry concentration and firm operating margins have been rising since 1980.[28] However, the Autor et al paper, while finding that the decline in the labour share is closely correlated with the change in concentration ratios across 12 of the 14 non-US countries examined, did not directly examine changes in industry concentration ratios for any countries other than the United States. Some other studies have questioned whether European industry concentration has in fact been on a clear upward trend over the relevant period. Weche and Wambach (2018), using estimates of firm-level mark-up, find that ‘market power’ of European firms showed a sharp decline during the global financial crisis (GFC) followed by a post-crisis increase that has not yet reached pre-GFC levels, and contrasts this with findings for the United States where concentration ratios are already back above pre-GFC highs. Guschanski and Onaran (2018) go further: using a measure of concentration which – unlike that used by Autor et al (2017) – includes a firm's international sales but only covers publicly listed firms, they find that concentration in the EU-15 group of countries declined between 1995 and 2015. In addition, in contrast to the findings of Autor et al, they find little relationship in their European data between concentration and the extent of decline in labour shares; and that declines in labour shares are mainly driven by declines within firms, rather than between firms. This study is very recent and at this stage it is difficult to know whether it is the more limited dataset used, methodological differences or other factors that are driving the differences in results.

As has been mentioned, these micro-level studies of what is driving factor income share changes may provide an important framework for analysing what is driving the short-run, low nominal wage growth puzzle, both in Australia and in many other countries. This issue is examined below in Section 4.

3.5 Relevance of Productivity Dispersion Model to Australia

How relevant, if at all, is the superstar firm model for explaining factor share shifts in Australia? One of the data challenges in measuring factor income shares is how to divide up the income accruing to proprietors of unincorporated businesses (gross mixed income or GMI), which is a mixture of labour income (reflecting the value of the time the owner puts into the business) and capital income (reflecting a return on capital and entrepreneurialism). The Australian System of National Accounts selected analytical series reports the wage and profit shares of total factor income excluding the GMI component, which means these shares do not add up to one. In addition, for the general government sector of the national accounts – which is primarily in the non-market sector – the net operating surplus is by convention assumed to be zero. For these reasons, factor income shares derived directly from economy-wide national accounts data need to be interpreted with caution.

The ABS productivity statistics (ABS 2018) attempt to overcome these limitations by focusing on just the market sector of the economy and by allocating GMI to the labour and capital income shares. For the labour share of GMI, the labour income of proprietors of unincorporated businesses is imputed based on the hours they work and assuming the average hourly wage of wage and salary employees in that industry. For the capital share of GMI, the rental price of unincorporated productive capital stock is imputed assuming the rental prices of the incorporated businesses and the incorporated rate of return on capital. The sum of these two imputations equals imputed GMI, with the difference between actual and imputed GMI allocated proportionally.[29]

The longest current data series using the above approach commences in 1973/74 and covers 12 selected market sector industry aggregates.[30] The trend is clear from Figure 5, where the following pattern can be seen:[31]

  • labour's share rose sharply in the 1970s and early 1980s to a peak of just under 65 per cent in 1982/83;
  • it then fell rapidly through to the late 1980s, before rising significantly in 1989/90 and 1990/91;
  • it continued to trend lower, albeit gradually, up to 2010/11, rose for a number of years after that, then fell sharply in 2016/17. It is now back close to its lowest point over the data series.
Figure 5: Labour Share of National Income – Australia
Aggregate of 12 selected industries
Figure 5: Labour Share of National Income – Australia

Source: ABS

The extent of the trend decline in labour's share shown in Figure 5 is exaggerated by the starting point for the current data series. An earlier market sector series is available for the period 1964/65 to 1995/96 (ABS 1997), but it is based on an older industry classification and excludes improvements to the estimates in subsequent national accounts releases over the last two decades. The earlier series suggests that labour's share rose by 5 percentage points from 1964/65 to 1973/74. However, it then trended down from the mid 1970s to the mid 1990s when the series ended, to a level 2 percentage points lower than at the beginning of the series. The current series shown above suggests a further trend decline since then, by a little over 5 percentage points.

Most Australian studies explain shifts in labour's share primarily in terms of two factors: changes to our wage-bargaining framework in previous decades; and large shifts in our terms of trade. On the first explanation, the sharp rise in labour's share in the 1970s and early 1980s occurred in the context of a highly centralised and heavily unionised wage bargaining process, which contributed to a series of large nominal wage breakouts; while the reversal of this trend from around the mid 1980s occurred in the context of the then Labor Government's series of wage/tax trade-offs negotiated with the trade unions.

These trade-offs were specifically designed to reduce the ‘real wage overhang’ (historically high labour share), which was seen as impeding investment and employment.[32] The subsequent move to a more decentralised wage-bargaining model is often cited as helping to maintain wage moderation and avoid a repeat of the 1970s real wage overhang.

Terms of trade shifts and the associated movement of resources into and out of Australia's heavily capital-intensive mining sector are often cited as the other critical factor explaining shifts in Australia's factor income shares. Parham (2013) found that the unprecedented rise in Australia's terms of trade and associated mining investment boom in the first decade of this century (Figure 6) accounted for all of the 4 percentage points or more fall in labour's share during that period. However, when looked at on a shift/share basis, the fall in labour shares within industries over the 2000s actually accounted for more of the drop in labour's share than between industry shifts.

Figure 6: Terms of Trade – Australia
2015/16 average = 100, log scale
Figure 6: Terms of Trade – Australia

Note: Annual data are used prior to 1960

Sources: ABS; RBA

Shift/share analysis can at times disguise some of the underlying factors at work, because it is done on a net basis. This means, for example, that a shift in resources from one industry to another may have a major impact on factor shares, but it is largely or entirely offset by a shift between two other industries. A recent ABS study (ABS 2018) looking at shifts in labour's share in Australia between the two decades 1997/98 to 2006/07 and 2007/08 to 2016/17, and using similar shift/share analysis, found that 83 per cent of the decline in labour's share was due to within-industry changes as against between-industry shifts. However, this was in large part because the substantial downward impact on labour's share from expansion of the mining sector was largely offset by growth in a number of labour-intensive service industries.[33]

It is quite possible that, while terms of trade cycles do indeed account for some of the major shifts in factor income shares in Australia, the growth of high productivity firms with low labour income shares may still help explain the underlying trend decline in labour's share. James (2017) found that, since the early 1990s, most of the shifts in the profit share can be explained by the economic cycle and the terms of trade, but there is still an unexplained upward trend in the profit share over a longer period. His panel regressions using industry data suggest technological progress may be a key driver.

Furthermore, there are a number of a priori reasons for thinking the superstar firm model may help explain the trend decline in labour's share in Australia. Firstly, the 2018 ABS study referred to above showed that within-industry effects dominated net between-industry effects. It also found that, of the 16 industry sectors examined, labour's share fell in 10 non-mining sectors.

Secondly, recent work by the RBA (Hambur and La Cava forthcoming) has found that the average Australian industry is quite concentrated, with a small number of large businesses selling most of the industry's output.[34] While over half of Australian industries have seen some increase in concentration over the last two decades, the overall rise in concentration is primarily due to the retail trade sector. These findings provide some support for the ‘superstar’ hypothesis, although they are also consistent with other explanations for high and rising industry concentration.

Thirdly, a recent analysis of wages growth by the Australian Treasury (2017) used an important new firm-level data source, the Business Longitudinal Analysis Data Environment (BLADE), to classify firms into high-, mid- and low-productivity groups. While this analysis was done across the whole sample rather than on an industry-by-industry basis, it is stilll instructive. In particular, they found that:

  • although high-productivity businesses pay higher average real wages, they do not pay real wages anywhere near in proportion to their high relative productivity: they were on average 7.1 times as productive as mid-productivity businesses, but only paid 1.6 times as much in average real wages;
  • more productive businesses have more capital per worker. For the top five deciles, capital per worker and average real wages increase with productivity, but capital per worker increases much faster than average real wages. This implies that the labour share decreases as productivity increases, at least for the top five deciles; and
  • the largest businesses, with more than $50 million turnover, were also the most productive.

These findings, while not directly testing the superstar firm model, are certainly consistent with the hypothesis that higher productivity firms are using most of their higher levels of productivity to lower prices and gain market share and/or to increase profit margins, rather than to increase wages.

Some preliminary explorations at the RBA suggest that, while the superstar firm model may well explain some of the movements in factor shares in Australia, its relevance may be largely concentrated within a few sectors, in particular retail and wholesale trade. This is an area worthy of detailed study, and is returned to in Section 4.


See, for example, on the treatment of capital depreciation Bridgman (2015) or Cho, Hwang and Schreyer (2017); on housing income Rognlie (2015); and on intangible capital Elsby, Hobjin and ┼×ahin (2013). For a recent Australian study that attributes the apparent long-run decline in labour's share to the treatment of housing income and depreciation, see Trott and Vance (2018). [12]

In the Autor et al study, labour's share fell in the United States for four of six sectors over the period examined; namely manufacturing, services, utilities and transportation, and retail trade. It rose in finance; and rose sharply then fell almost as sharply in wholesale trade. In each sector, except ‘services’ where the firm entry effect dominated, the between-firm effect was the largest negative contributor to the change in labour's share. [13]

Australia may be a partial exception to this general point. This point is returned to below. [14]

See, for example, Oberfield and Raval (2014), Summers (2014) and Lawrence (2015). For reviews of the literature on estimating the elasticity of substitution, see Antràs (2004), Chirinko (2008) or León-Ledesma, McAdam and Willman (2010). [15]

One area of explanation not covered and which is not closely linked to technology is that of ‘financialisation’; that is, the increase in size and importance of a country's financial sector. For a literature survey and summary of the different channels whereby financialisation may impact on labour's share, see Guschanski and Onaran (2018). [16]

However, similar to the IMF (2017b) study on which this analysis is built, the study also found that technology – as measured by the intensity of ‘routinisation’ of job tasks – was a more powerful explanatory variable. [17]

For an international industry-level analysis and more references see OECD (2012), which found that offshoring plus increased import competition accounted for around 10 per cent of the decline in labour's share across the countries examined. For a US manufacturing sector study see Autor et al (2017). [18]

These include the importance of multi-employer bargaining and the role the state plays in a number of countries in extending collective bargaining arrangements to parties not directly affiliated with the bargaining parties (see OECD (2012, p 135)). [19]

See also Acemoglu (2003) and Bentolila and Saint-Paul (2003). [20]

See also on this point, Bhaduri (2006) and Hein (2014). [21]

In the case of tasks that can be fully automated, labour and capital become perfect substitutes. For studies focused specifically on the role of automation in explaining labour's declining share, see Acemoglu and Restrepo (2016, 2018) and Autor and Salomons (2018). [22]

On differences in elasticities of substitution between higher- versus lower-skilled labour, see Krusell et al (2000) and IMF (2017b). For analysis of the types of work most susceptible to automation, see Autor (2015) and OECD (2018). [23]

While a number of earlier studies such as Syverson (2004) and Hsieh and Klenow (2009) had also identified significant productivity dispersion within sectors, none of them were based on such a comprehensive firm-level data source. [24]

A more recent study suggests this increasing productivity divergence has continued (see Berlingieri, Blanchenay and Criscuolo (2017)). [25]

See, for example, Söderbom, Teal and Wambugu (2002). For an Australian study, see Peetz and Preston (2009). [26]

This is also the finding of Adrjan (2018), which is returned to below. [27]

See also Ganapati (2018). [28]

An alternative rule of thumb approach that has been used in some of the literature is to allocate two-thirds of unincorporated business profits to wages, but this implicitly assumes a constant capital-to-labour ratio over time. For a more detailed description of how mixed incomes are decomposed into labour and capital components, see ABS (2016). [29]

While the ABS also publishes 16 market sector industry aggregates, they only go back to 1993/94. [30]

The published ABS data are a smoothed series, whilst the data in Figure 5 are unsmoothed and were kindly provided to the author by the ABS. [31]

For a summary of the discussion and debate at this time on the relationship between the real wage overhang and employment, see Russell and Tease (1988) and Chapman (1998). [32]

Breaking the series at 2007/08 to reflect the impact of the GFC, the ABS analysis finds that the mining sector contributed around 1.5 percentage points of the 2.3 percentage point decline in labour's share between the earlier and subsequent decade (see ABS (2018)). [33]

While Minifie (2017) reached somewhat different conclusions, his dataset was much more limited. [34]