RDP 2018-10: Wage Growth Puzzles and Technology 4. Technology Diffusion and Low Nominal Wages Growth: Some Research Proposals

This section examines how a firm-level framework that allows for heterogeneous productivity growth may also be applied to examining why nominal wages growth has been so low. It outlines a hypothesis linking uneven productivity growth to the nominal wage growth puzzle, and sets out some empirical research proposals for testing the validity of the hypothesis.

4.1 The Heterogeneous Productivity Growth Model

While the short time period for which data are available is a clear limitation, the relatively new BLADE database provides an opportunity for researchers to examine the relevance of a firm-level framework incorporating heterogeneous productivity growth in explaining surprisingly low nominal wages growth in Australia.[35]

The Autor et al (2017) study uses an illustrative firm-level model and a production function Y i = A i V i 1α K i α where Yi is value added, Vi is variable labour, Ki is capital and Ai is Hicks-neutral efficiency in firm i and is heterogeneous across firms. Given the way the model is specified, the level of total factor productivity (Ai) is in effect exogenous to the firm: it is randomly allocated when a firm starts up. Firms with higher productivity will have higher levels of factor inputs and greater sales. The authors assume a fixed amount of overhead labour (F) needed for production, so total labour is given by L = V + F. The model allows for imperfect competition in the product market. With respect to labour markets, all firms are assumed to pay the same, exogenously set wage. Firms cannot respond to increased competitive pressure by lowering wages.

From their model and the static first-order condition for labour, they derive the share of labour costs (wLi) in nominal value added (PiYi) as:

where ui = (Pi /ci) is the mark-up, the ratio of product price Pi to marginal cost ci. The firm i subscripts indicate that for given economy-wide values of (α, w, F), a firm will have a lower labour share if its mark-up is higher and/or its share of fixed labour costs in total value added is lower.

Superstar firms will be larger as they produce more efficiently and capture a higher share of industry output. The model suggests they will also tend to have lower labour shares.

Elements of the above framework – in particular, allowing for heterogeneous productivity performance across firms – are relevant to examining various hypotheses linking uneven technology take-up to wages growth. However, the assumptions of exogenous productivity and wages obviously need to be relaxed. Furthermore, while the Autor et al (2017) model is specified in terms of heterogeneous levels of productivity, allowing also for heterogeneous growth rates is both more consistent with the literature (e.g. Andrews, Criscuolo and Gal 2015) and potentially more useful in terms of explaining recent nominal wage growth surprises.

If the assumption of exogenous wages is relaxed, the nominal labour share of value added for firm i is:

If we are comparing higher and lower productivity firms, there are three ways in which higher productivity firms could lower their nominal labour share/increase their profit share over time compared to lower productivity firms:

  • for a given wages bill, by increasing their nominal value added faster than the less productive firms;
  • for a given level of value added, by increasing their wages bill slower than lower productivity firms via restrained average wage increases and/or slower growth in employment; or
  • by some combination of the above two.

The more higher productivity firms are lowering their labour share of value added by constraining nominal wages growth and employment growth, the stronger the potential relevance to explaining low nominal wages growth. Testing this potential relevance thus requires examining relative growth rates for value added, employment and nominal wages in low-, mid- and high-productivity growth firms.

4.1.1 The hypothesis

With perfectly competitive labour and product markets, each employee would be paid their marginal product. The faster the rate of productivity growth, other things equal, the faster the rate of wages growth; and the more employment is skewed towards higher productivity firms, other things equal, the greater the contribution to average wages growth of the more productive firms.

If the assumption of perfect competition in factor and product markets is relaxed, the possibilities broaden. They include the central hypothesis of this paper, which is based on partial evidence available both overseas and in Australia that is outlined below:

The uneven take-up of new technology is resulting in large and growing gaps in productivity performance across firms in the same industry. High-productivity firms that are adapting new technology most successfully are using their higher productivity levels and growth rates to increase profit mark-ups and/or hold down their output prices so as to further increase their market share, rather than primarily passing them on to their workforce via higher wages. They are also likely to be leading the way with respect to domestic outsourcing and casualisation of the workforce, in the process reducing the bargaining power of their employees. Low productivity growth laggard firms are under increasing competitive pressure and have limited capacity to raise wages. These factors are holding down average wages growth compared to a situation where all employees are paid their marginal product. The extent of their impact will depend on the distribution of employment across high- and low-productivity firms and how it is changing over time. Because high-productivity firms are typically not increasing their employment share in line with their value added, this is further reducing their contribution to average wages growth.

4.1.2 Some supporting evidence

The above hypothesis incorporates the following testable assumptions:

  1. The uneven take-up of new technology is resulting in very uneven patterns of productivity growth across firms, with the gap between high and low productivity growth rates increasing.
  2. For high productivity firms, there is a large gap between productivity growth and wages growth; while the increasing competitive pressures on lower productivity firms significantly constrain their capacity to pay higher wages.
  3. The employment share of high-productivity firms is increasing much more slowly than their value added share.
  4. High-productivity firms are using their greater application of new technology to lead the way with respect to such practices as outsourcing and casualisation of their workforces.

A number of overseas and Australian studies contain empirical support for one or more of these assumptions. Andrews, Criscuolo, Gal and Menon (2015) confirmed (a) for a range of OECD countries, namely that the uneven take-up of new technology is driving very uneven patterns of productivity growth across firms, and the gap is increasing. In addition, Andrews et al found that frontier firms are significantly larger in terms of sales but not in terms of employment, lending support to (c). Faggio, Salvanes and Van Reenen (2010) found that industries that experienced the greatest increase in productivity dispersion also experienced the largest rise in IT capital intensity, lending support to (a).

Kehrig and Vincent (2017) use plant-level data for US manufacturing to examine what is driving the fall in labour's share. They begin by noting that, since the mid 1980s, labour compensation in manufacturing has fallen from 67 to 47 per cent of value added. At the same time however, the labour share of most US manufacturing plants actually rose. Their analysis finds that the explanation for this apparent paradox has been a major reallocation of production towards ‘hyperproductive’ (HP) plants, and a downward adjustment of the labour share of these HP plants over time.

Their paper then examines what is driving the dramatic difference in labour share trends between HP and non-HP plants, and finds that:

… almost all the divergence in the labor share trends between two types of plants is due to value added. According to the coefficient estimates, the typical HP plant was able to grow its value added by a staggering 250% more than typical non-HP establishment, or about 3.7% per year. This, amazingly, has come with very little divergence in the growth of the wage bill across the two types of plants, either through wages or employment. (Kehrig and Vincent 2017, pp 24–25)

This finding is consistent with (b) and (c).

Further support for (c), the proposition that the employment share of superstar firms is increasing much more slowly than their share of value added, can be found in Ganapati (2018). His study suggests that industry concentration increases in the United States are positively correlated to productivity and real output growth, negatively correlated with labour's share of value added and uncorrelated with employment, with the most productive industries typically maintaining or reducing their workforces.

The Autor et al (2017) study of factor income shares is consistent with (b) in that its findings imply that most of the productivity gains for higher productivity firms are being absorbed by a combination of lower prices (encouraging higher sales) and higher profit mark-ups, as against primarily being passed on to their workforce in the form of higher wages. It also provides some support for (c): the authors used sales as their measure of industry concentration, but they noted that, if they replaced this measure with employment, the relationship between concentration and labour share switched signs to become positive though generally statistically insignificant.

A recent OECD (2018) study looking at the decoupling of wages from productivity at the firm level found that wage divergence across the countries examined was much less pronounced than productivity divergence. Of particular interest in the context of the above hypothesis was their finding that not only was there a large divergence between productivity growth and real wages growth for high productivity firms, but for lower productivity firms real wages growth was also below their (lower) labour productivity growth rate. This is consistent with (b) and with the central tenet of the above hypothesis, namely that the competitive pressures on lower productivity firms are severely constraining their capacity to raise the wages of their employees.

Similarly, in an interesting study looking at key factors driving increasing wage inequality, the OECD (Berlingieri et al 2017) found that it was largely a result of increasing wage dispersion between firms in the same sector, rather than differences across sectors, and that the between-firm dispersion in turn was linked to differences in productivity across firms. Furthermore, the observed growing wage divergence was driven primarily by firms towards the bottom of the wage distribution in their industry paying increasingly less relative to the median firm, as against firms towards the top paying increasingly higher wages. Indeed, wages and productivity were less correlated at both tails of the productivity distribution. These findings are also consistent with (b).

Autor et al (2017) noted that the rise of superstar firms appears to be related to ‘changes in the boundaries of large dominant employers’ (p 26), with such firms increasingly using domestic outsourcing to contracting firms, casualisation of their workforce and automation, reducing both the bargaining power of their employees and any wage premia typically paid by larger firms. Goldschmidt and Schmieder (2017), in a study of German firms, made a similar observation, noting that:

Large firms are increasingly relying on nontraditional employment arrangements such as outsourcing, temporary or contingent work, offshoring, and subcontracting. (p 1166)

These findings are consistent with (d).

The evidence to date for Australia is more limited. As mentioned earlier, the Australian Treasury (2017) used the BLADE database to classify firms into high-, mid- and low-productivity groups. Consistent with (b), they found that, although high-productivity businesses pay higher average real wages than lower productivity firms, the dispersion of real wages is substantially less than the dispersion of productivity. However, as noted this analysis was based on averages across all firms examined rather than on higher versus lower productivity firms within an industry. The higher average wages for higher productivity firms could thus simply reflect a greater concentration of skilled job classifications in higher productivity firms, as against such firms paying more for the same job classification.

Extending the Treasury analysis to look at these factors within each industry and to include employment distribution across high-, mid- and low-productivity firms within each industry could provide a potentially important link between the uneven pattern of productivity growth and the short-run wage puzzle in Australia.

As noted earlier, preliminary explorations at the RBA regarding the proposition that the growing dominance of superstar firms has weighed on labour's share in Australia suggested that, to the extent that the effect is relevant to Australia, it may be concentrated in a few sectors of the economy, in particular retail and wholesale trade. If confirmed by detailed analysis, this may reflect lower levels of competition in many sectors of the Australian economy compared to the US economy. A key aspect of the superstar firm model for explaining factor income share movements is that higher productivity firms are using a fair amount of their higher productivity to hold down their output prices and increase their market share: hence the rise in industry concentration ratios. But with respect to using a similar framework that allows for heterogeneous productivity levels and growth rates across firms to look at why nominal wages growth is surprising on the downside, the key is not whether higher productivity firms are increasing their market share, but rather the pressure they are putting on lower productivity firms. Regardless of whether higher productivity firms are reducing output prices or increasing profit margins, their actions put pressure on lower productivity firms to keep their costs down, including wage costs, because they are either less profitable or losing market share, or both. These pressures may be greater for listed than for unlisted companies, but they are relevant for both.

If Australia is a less competitive economy than the United States and – at least in some sectors of the economy – that is being reflected in higher productivity firms raising their profit margins more than reducing their output prices, this may help explain why Hambur and La Cava (forthcoming) found that industry concentration in Australia had risen substantially in only a limited number of industries. The research proposals set out below should help shed light on whether this possible explanation for differences in results for Australia as against the United States has some validity or not.

In short, a firm-level framework that allows for divergent productivity performance may help shed light on slow nominal wages growth regardless of whether high-productivity firms are primarily reducing their output prices to increase their market share or are raising their profit margins. Either way, if the application of digital technology that is driving much of the growing divergence in productivity performance across firms is, as suggested in Section 2, spreading to more and more sectors and being embedded in more areas of investment, the pressure on lower productivity firms across the economy to reduce their wage costs is likely to increase.

4.1.3 Timing and sustainability

One of the filters used earlier for examining research on what is driving the trend decline in labour's share across many countries was timing: it was suggested that causal factors should have been in place for most or all of the period of declining labour shares going back to around the early 1980s. One apparent weakness in suggesting that some of the same factors driving the trend decline in labour shares may help explain recent nominal wage growth surprises is that the latter have only been evident for a much shorter period of time.

In explaining such timing differences, it is necessary to clearly delineate those elements of the superstar factor share model that are central to the hypothesis above regarding nominal wage growth surprises. The Autor et al factor share hypothesis assumes disparate levels of productivity across firms, not disparate growth rates. Firms with a higher level of productivity attract more resources and capture a higher share of industry output over time, by means of the usual allocative efficiency mechanisms (see Andrews, Criscuolo, Gal and Menon (2015)). As they also have lower labour shares, this growth in their market share, it is hypothesised, accounts for the trend decline in labour's share. However, the hypothesis above regarding nominal wage growth surprises relies more on the dispersion of productivity growth rates across firms – that is, a growing divergence in terms of productivity levels – putting increasing pressure on lower productivity firms to reduce cost increases, including wage costs.

The distinction may be important with respect to the timing issue. As noted earlier, recent studies using a firm-level database that commences around 2000 found that the firm-level dispersion of labour productivity has increased significantly since then.[36] Data on firm-level productivity growth rates for Australia are similarly only available for the period since 2002, and also show both high and growing levels of dispersion. Longer-run studies on productivity dispersion across firms are relatively sparse due largely to data constraints (see Faggio et al (2010)), but those available show mixed results for periods prior to this century. By way of example, Bartelsman, Haltiwanger and Scarpetta (2009) found that, while the dispersion of firm-level labour productivity was substantial during the 1990s across the eight countries and industries examined, it was also reasonably constant over the decade. However, Faggio et al (2010) found that within-industry productivity dispersion in the United Kingdom started trending upwards around the mid 1980s. They attribute growing productivity dispersion primarily to the uneven take-up of new technology: after controlling for a range of other factors, they find a strong positive correlation between ICT capital intensity across industries and the extent of the rise in productivity dispersion.

Uneven take-up of new technology is hardly a new phenomenon. However, among the defining features of digital technology are its pervasiveness and the ongoing and rapid technological advances being made in this area. While speculative, this may help explain the increasing divergence in productivity performance across firms in recent decades, which this paper suggests has resulted in a gradual but rising pressure on the productivity laggards to control costs.

A related factor of relevance to the issue of timing is that the period of wages growth overestimation appears to be longer than just the past few years. Analysis of OECD wage growth forecasts versus outcomes shows that, in addition to the downward wage growth surprises of the last two years discussed in, for example, Arsov and Evans (2018), wages growth was consistently overestimated in the six years leading up to the GFC (Figure 7). By contrast, during the period of the GFC and its immediate aftermath, the OECD repeatedly underestimated inflation, including wage inflation, despite the fact that it consistently overestimated GDP growth (Pain et al 2014). Given the severity of the GFC-related recession in most OECD countries, it is not surprising that historical relationships between labour market slack and wage/price inflation broke down, so this period is probably best put to one side in thinking about timing issues.

Figure 7: OECD Wage Growth Forecasts and Outcomes
Compensation per employee, total OECD
Figure 7: OECD Wage Growth Forecasts and Outcomes

Source: OECD, Economic Outlook

In the case of Australia, the GFC-related downturn in GDP growth and rise in unemployment were much less severe than in other OECD countries and, as noted earlier, wages growth in the aftermath of the GFC was lower than expected and has remained so (Figure 1). There is also some evidence of downward wage growth surprises in Australia in the 2001–06 period shown in Figure 7, at least for some forecasters: OECD wage growth forecasts for Australia exceeded outcomes in four of those six years.

Finally, it is possible that, while the impact of the uneven take-up of technology on the relationship between average real wages growth and productivity growth started to occur early in the period of rising ICT investment, the link to nominal wages growth strengthened in the wake of the more recent coincidence of both lower-than-expected productivity growth and very low and stable inflation expectations (see IMF (2013), Jacobs and Rush (2015), and Bishop and Cassidy (2017)).

A closely related issue concerns the sustainability of large and – more recently – growing gaps in productivity performance across firms in the same industry. Economic theory suggests that competitive pressures should, over time, see productivity dispersion between firms in the same industry narrow, by way of less-competitive firms adapting the new technology and improving their productivity performance; or else failing to do so and exiting the industry as resources are directed away from them and towards more productive firms.

The persistence of high and growing levels of productivity divergence has led to a number of studies focused on why static and dynamic allocative efficiency mechanisms are not working as might be expected in many countries (see Bartelsman et al (2009) and Andrews, Criscuolo, Gal and Menon (2015), including references therein). Factors examined include product market regulations that inadvertently act as barriers to new entry, employment protection legislation that raises the cost of hiring and firing labour, and access to risk capital.

As noted earlier, uneven adaptation of new technology is hardly a new phenomenon, so another way of coming at the issue of why productivity dispersion has remained high and growing in recent decades is to consider whether there is something unique about digital technology that acts as a barrier to new entry. Two key aspects of ICT discussed in Section 2 are the gathering and processing of information to improve a firm's efficiency and expand its client base; and the use of automation to reduce costs. In both cases, successful adaptation of new technology to the particular needs of a firm can be expensive and disruptive. It requires high levels of managerial skill and imagination. It also requires staff with the skills, capacity and willingness to put the new technology in place, use it effectively and adapt to the many changes to workplace practices that it may bring. Because its introduction can be highly disruptive, it also requires specific change management skills on the part of senior managers and boards.

These required skills, imagination and attitudes are far from uniformly spread across countries or firms and, in many countries, may be in short supply. That in itself is a barrier to perfectly functioning markets, and has implications for the education and training of both employees and management.

In addition, there may be industries in which the technological leaders are able to use their leadership position to establish their own barriers to entry. Antitrust lawsuits in the United States and, more so, Europe against leading technology companies in recent decades provide examples.[37]

Looking forward, the pace of technological advance with respect to ICT, and the number of industries in which it has the capacity to be applied, are if anything accelerating, particularly with respect to big data, automation and artificial intelligence. Successful ongoing take-up of the opportunities such technological change provides, either by existing technological leaders or by new entrants, is also likely to operate in the direction of ensuring that the dispersion in the level and growth rates of labour productivity across a growing number of industries remains high for some time.

4.1.4 Testing the hypothesis

The hypothesis set out earlier generates a number of testable propositions for Australia using firm-level data.

(a) Uneven technology take-up and product markets

Testable proposition 1: The uneven take-up of technology is resulting in a large and growing divergence in productivity performance across firms. Technological leaders are appropriating most of their higher productivity gains into either higher profit margins or lower output prices.


Firm-level output price data are not available, but increases in market share provide a rough proxy for the extent to which high productivity firms are holding down their quality-adjusted output prices relative to their competitors. Use the BLADE database to classify firms in each industry into high-, mid- and low-productivity growth groups to examine whether productivity dispersion has been increasing. Look at changes over time in productivity, profit margins, nominal wages and value added for high-productivity firms within each industry, to test the proposition that they are primarily using their higher productivity growth to increase profit mark-ups and/or increase their market share.

Testable proposition 2: To the extent that higher productivity firms are using their higher productivity growth rates to increase profit mark-ups or increase their market share via lower output prices, this is putting increased competitive pressure on lower productivity firms to reduce costs, including by paying below industry-average wages and wage increases for the same occupation.


The BLADE database includes data on average firm wages, but not wages by job classification. The ABS are currently building a longitudinal employer/employee database which may be linked to BLADE and which would allow a detailed examination of the above proposition, but the full database is not yet available. In the interim, a very partial test would be to look at average wage levels and increases in each industry for low-, mid- and high-productivity firms and compare them to productivity levels and growth rates.

Testable proposition 3: Productivity leaders' share of total industry employment is rising more slowly than their share of industry sales or value added, which (in combination with propositions 1 and 2 above) is holding down average industry wages growth.


Use the BLADE database to look at changing shares of industry employment for each productivity group relative to shares of industry sales or value added.

Look at what would have happened to average industry wages growth if employment to sales or value added ratios for each productivity group had remained constant.

(b) Uneven technology take-up and labour markets

Reference was made earlier in this paper to the observation that the rise of superstar firms appears to be related to ‘changes in the boundaries of large dominant employers’ (Autor et al 2017).

The BLADE database may allow at least part of this observation to be tested for Australia.

Testable proposition 4: Higher productivity firms are leading the way with respect to domestic outsourcing and casualisation of the workforce, reducing employee bargaining power and any ‘large firm’ wage premia.


The BLADE database includes survey data on working arrangements. It may be possible to use these data sources to examine differences in the degree of labour force casualisation between high-, mid-and low-productivity firms across different industries.


BLADE is a firm-level statistical system maintained by the ABS that enables business datasets to be linked up via a common identifier, namely the firm's Australian business number (ABN). By enabling various datasets to be linked up, it increases the usefulness of the asset to researchers undertaking firm-level analysis. [35]

See, in particular, Andrews, Criscuolo and Gal (2015) and Berlingieri et al (2017). [36]

For some examples, see Couturier (2016). [37]