RDP 2018-10: Wage Growth Puzzles and Technology Appendix A: RBA Research on Structural Factors That May Be Holding Down Nominal Wages Growth

Most of the RBA research looking to explain surprisingly low nominal wages growth has been done within the framework of a conventional wage Phillips curve relationship:

where w is nominal wages growth, ue the unemployment rate, ue* the NAIRU, z a vector of other exogenous cyclical factors that may affect wages growth and ε the error term. In the RBA's preferred formulation, the vector z includes inflation expectations, actual inflation, trend growth in labour productivity and the change in the unemployment rate (Arsov and Evans 2018, Appendix A).

It is worth noting that β, the coefficient on the unemployment gap, is not only a measure of the strength of the relationship between nominal wages growth and the economic cycle: it is also a measure of real wage flexibility and its inverse, real wage rigidity (1/β). To quote Coe (1985):^[38]

[real wage rigidity is] the degree of non-accommodation, measured in terms of unemployment, which would be necessary to maintain inflation constant in the face of an adverse shock. Thus real wage rigidity will be higher the less responsive are nominal wages to the unemployment rate. (p 115)

Returning to Equation (A1), explanations of the nominal wage growth puzzle that are focused on a falling β – that is, a flattening of the Phillips curve – are likely to be of limited use in examining structural declines in labour bargaining power. This is because, if nominal wages growth is less sensitive to labour market conditions over the course of economic cycles, wages growth will be higher than expected when unemployment is high and rising and lower than expected when unemployment is low and falling, with the ‘surprises’ to some extent at least cancelling out over the course of an economic cycle.

Ongoing falls in β over the course of a cycle – that is, ongoing flattening of the Phillips curve – have been ‘observed’ both in Australia and overseas in certain periods.^[39] However, they are not relevant to wage behaviour in Australia over the course of the most recent cycle, during which nominal wages were initially falling faster than expected during a period of high and rising unemployment, implying that the Phillips curve was steepening or the intercept coefficient was rising (Jacobs and Rush 2015). More broadly, ongoing flattening of the Phillips curve over the course of economic cycles would imply a trend decline in real wage flexibility. This seems unlikely in most western countries, where the trend has been in the direction of labour market deregulation. In Australia's case, the decentralisation of wage bargaining that has occurred since the early 1990s is widely perceived as having significantly increased real wage flexibility and contributed to stabilising both activity and unemployment (e.g. Borland 2011).

Recent RBA research looking at possible structural factors dampening wages growth is discussed below.

A.1 Job Insecurity

If employee objectives were to change in a way that placed less emphasis on securing real wage increases and more on maintaining jobs, this would show up as a fall in α in Equation (A1): that is, a downward shift in the Phillips curve. While much of the public discussion on the potential adverse effects of automation on employment is undoubtedly exaggerated, the evidence does suggest that it is having a hollowing-out effect on certain types of jobs and workers, forcing many of them to look for work in less well paid and lower-skilled occupations (e.g. Autor 2015; OECD 2017). Similarly, as the use of digital technology enables and encourages the outsourcing of work to lower wage economies, it would be surprising if this were not impacting on job security in the relevant industries (for some references, see Goldschmidt and Schmieder (2017)).

This is, of course, not to imply that the various effects of ICT take-up are the only factors affecting job security. The recent global recession has likely had a longer-lasting impact on job security than earlier recessions given its severity and the slow recovery from it. However, to the extent that ICT is having a significant and separate impact, it is likely to both predate the Great Recession and outlast it, if not increase over time as more and more industries and firms are affected by ICT.

Recent work by the RBA provides some partial support for the above hypotheses. Anecdotal evidence from the RBA's business liaison program suggests that employees are increasingly concerned about losing their jobs to automation or offshoring. Foster and Guttmann (2018) use data from the HILDA Survey to examine the impact of higher trade exposure (globalisation) and automation risk, inter alia, on perceived job security, and find a statistically significant relationship. They also find that automation risk and trade exposure contribute significantly to lower wages growth. However, they go on to note that these explanatory variables cannot account for low aggregate wages growth in recent years, because employment growth has been moving away from highly trade-exposed industries and from occupations that are more easily automated. They also note that the fall in perceived job security in recent years has been very broad based across industries and occupations, and is most likely reflecting generalised concerns about job security in the wake of the GFC. While this broad-based deterioration in perceived job security has a statistically significant effect on wages growth over the full period examined, it cannot explain much of the recent decline in average wages growth.

A.2 Falling Unionisation

While job security and bargaining strength are closely intertwined, with lower levels of perceived job security reducing the bargaining power of employees, other institutional factors may also affect bargaining power. The clearest example is falling levels of unionisation. Empirical work in Australia and elsewhere finds a wage premium for trade union membership, including for workers with the same skills in the same sectors.^[40] It seems unlikely that this primarily reflects different perceptions of job security by individual employees: more likely, it mainly reflects the greater bargaining power of a union as against an individual.

Arsov and Evans (2018) note that unionisation rates have been declining across all advanced economies since the 1990s, but the fall has been particularly large in Australia and a few other countries in recent years (Figure A1). Their econometric analysis suggests that, on average across the countries examined – which included Australia – a 1 percentage point decline in unionisation rates reduced wage growth by around one-quarter of a percentage point in the following year.

Of course, the spread of new technology and its ramifications is only one of many factors that have led to declining levels of unionisation in Australia and elsewhere, but it is certainly a factor and is likely to have an ongoing effect.

A.3 Casualisation of the Workforce

In a recent paper examining reasons why wages growth in the United Kindom has surprised the Bank of England on the downside, Haldane (2017b) looked at the impact on wages growth of what he calls the increasing ‘divisibility’ of work, which as outlined earlier has been greatly enabled by digital technology. Haldane looked at the effect of not only declining union membership but also, closely associated, the growth in self-employment, in part-time and casual work, and in so-called zero hours contracts. While noting that, for some workers, these changing patterns of work and associated shifts in bargaining power have been forced upon them and may be associated with increased job and wage insecurity, for many others this may not be the case: he refers to survey data showing that for some at least the shift to part-time work and self-employment has been welcome, providing greater control and flexibility. He also refers to UK survey data showing a marked increase in the share of temporary workers reporting they do not want a permanent job; and an upward trend in the share of part-time workers reporting they do not want full-time work.

Whether welcome or not, Haldane suggests that these changes in working arrangements are reducing aggregate wages growth. He cites estimates of a union membership wage premium in the United Kingdom of between 10 and 15 per cent; a wage discount for self-employed of around 15 per cent, 5 to 6 per cent for temporary workers and 7 per cent for zero hours contracts. While his back-of-the-envelope estimates suggest that these numbers have probably only directly reduced average annual wages growth since 2000 by a small amount, he argues that their overall impact is likely larger due to spillover effects on other workers. In addition, as digitisation spreads across industries, the impact of these changing work arrangements seems likely to increase over time.

More broadly, an OECD (2015) study found that, across almost all countries for which data were available, there was a marked wage gap between workers on temporary or casual contracts as against standard or ‘permanent’ contracts.

Looking at labour market casualisation in Australia, Mooi-Reci and Wooden (2017) noted that casual (rather than fixed-term) employment contracts are the most common form of non-standard employment in Australia, accounting for between 22 and 24 per cent of employment at their time of writing. They also noted that industrial laws in Australia require a reasonably substantial hourly wage premium for casual workers to compensate for loss of other benefits such as paid leave. In examining whether casual employment has a lasting effect on wages, they found that it does for male workers – of the order of around 10 per cent – but for women the wage penalty is both smaller and less long-lasting. Lass and Wooden (2017) found that low-paid casual workers experience a significant wage penalty, whereas high-paid casual workers enjoy a wage premium compared to their permanent counterparts. Nonetheless, this premium is, with only a very few exceptions, less than the 20 per cent loading specified in most industrial awards that is supposed to compensate for the loss of other benefits.

If, as has been suggested, the increasing take-up of digital technology is resulting in a growing divisibility or casualisation of the workforce, this might be expected to show up in changing patterns of employment, as Haldane's (2017b) study suggested was the case in the United Kingdom. However, the relationship is often likely to be more complex and to be heavily influenced by differences in labour market regulations and practices across countries (OECD 2012). The UK labour market, for example, is generally regarded as much more flexible and less heavily regulated than the Australian labour market.

Limited analysis of this issue appears to have been done in Australia. Referring to Haldane's UK study, Cassidy and Parsons (2017) suggest there is little evidence in Australia of an increase in the share of workers on temporary contracts or working for labour hire firms. The Australian Treasury (2017) suggest that, while casualisation trends are hard to measure, the share of casual workers rose significantly in the 1990s but has remained reasonably steady since then at around 21 per cent of the workforce. On the other hand, the self-employed share has declined.

These findings may be in part due to inadequacies in the ABS survey questionnaires in terms of picking up changes in the nature of working arrangements. They may also reflect the fact that changes in working arrangements are heavily concentrated amongst the productivity leaders rather than being spread across most firms, and hence are not showing up strongly in the aggregate data.

Some RBA work has been done on the relationship between pay-setting arrangements and wages growth. Bishop and Cassidy (2017) note that wages growth has declined in Australia across all pay-setting arrangements, although it has declined most in industries with a high prevalence of individual agreements. Given there has been an increase in the share of workers on awards in recent years, this does not help explain lower-than-forecast average wages growth. More broadly, the Australian Treasury (2017) notes that reclassifications of groups of workers along with other factors mean that it is difficult to draw any firm links between methods of pay setting and wages growth. Furthermore, in terms of linking this work to the impact of technology, the relationship between casualisation of employment and pay-setting arrangements would not appear to be clear-cut, other than the fact that casual workers are less likely to be unionised.

A.4 Real Wage Overhang

A possible factor behind nominal wage growth surprises in recent years is that the level of nominal wages in earlier years was out of kilter with product prices and productivity, resulting in a real wage overhang that is taking some time to unwind. This possibility was raised by the then US Federal Reserve Chair Janet Yellen (2015), with the suggestion that this overhang may have built up during the GFC-related recession.

This explanation may be more relevant for countries such as the United States that experienced a very sharp recession as a result of the GFC. For many such countries, the surprise during the ensuing steep recession was that wage and price inflation did not fall nearly as far as expected despite the very severe rise in unemployment (e.g. IMF 2013; Blanchard et al 2015). For Australia, however, the surprise was, as noted earlier, in the opposite direction: namely, why wages growth fell as far as it did relative to the (much milder) rise in unemployment.

Nonetheless, this proposition has been put forward by both the RBA and Treasury as possibly helping explain recent weak nominal wages growth, and to a limited extent tested for Australia, primarily in the context of the mining boom and its aftermath (Jacobs and Rush 2015; Kent 2015; Davis, McCarthy and Bridges 2016; Australian Treasury 2017; Bishop and Cassidy 2017). The argument runs along the following lines. During the mining boom from 2003/04 to 2011/12, both nominal wages and real consumer wages growth picked up significantly and the exchange rate appreciated. Employers, on average, easily absorbed this rise in nominal wages due to the much faster growth in producer prices than consumer prices – in other words, the real product wage increased substantially less than the real consumer wage (Figure A2).

However, this period resulted in high Australian wage costs in the aftermath of the mining boom, once commodity prices started to fall. The subsequent period of very low nominal and real wages growth can be seen as part of a normal and desirable adjustment process to restoring competitiveness and encouraging employment. Due in part to downward wage rigidity in a period of very low inflation, it is taking a long time for this adjustment to occur (Jacobs and Rush 2015), which may explain ongoing low wages growth in the face of low and falling rates of unemployment.

Arsov and Evans (2018) attempted to test this hypothesis of a wage overhang in seven countries (including Australia) plus the core and peripheral bloc of euro area countries by estimating the deviation of the level of wages from their long-run relationship with productivity and output prices. Their results suggest that in most of the countries examined, including Australia, sluggish nominal wages growth in recent years partially reflects deviations of the level of wages from their ‘fundamental’ value in the aftermath of the GFC, but that this effect has largely ended. However, they suggest that ‘a renewed wage overhang may help explain the weakness in nominal wages in 2017 in the United States, Canada and Australia’. They attribute this real wage overhang effect to downward nominal wage rigidities in the context of low inflation.

One difficulty with the argument that slow nominal wages growth in recent years reflects, in part at least, correction of an earlier real wage overhang is that it is hard to reconcile with rapid growth in employment, falling unemployment and a high level of employment to output (low level of labour productivity) in recent years in many of the countries being examined, including Australia. If real wages were significantly out of line with productivity, profits would have been under pressure and so would have been employment, and the employment-to-output ratio would be expected to be falling. However, what we have witnessed is the reverse: which is hardly surprising given that, in fact, wage shares in many countries, including Australia, are at or close to historical lows over recent decades.

It is conceivable that, with the growing integration of countries such as China into the global economy and global supply chains, what we are now witnessing is the slow unwinding of a real wage overhang in more advanced western economies relative to the more successful emerging economies, in the sense that the gap in wage levels between the two groups of economies is much larger than the gap in labour productivity levels, putting competitive pressure on western economies that requires a ‘resetting’ of relative real unit labour costs. However, this too should have shown up in rising levels of unemployment in these western economies in recent decades, which – absent the impact of the GFC-related recession – has not been the case.

Footnotes

See also on this point Grubb, Jackman and Layard (1983) and Lim, Dixon and Tsiaplias (2009). [38]

Overseas research work earlier this decade on the observed flattening of price Phillips curves was in fact driven by the observation that price inflation had not fallen as much as expected in response to the sharp rise in unemployment rates to well above estimated NAIRUs coming out of the GFC. For international studies on this see, for example, IMF (2013), Matheson and Stavrev (2013) and Blanchard, Cerutti and Summers (2015). In Australia, Gillitzer and Simon (2015) observed the same flattening, although most of the changed behaviour came from greater anchoring of inflation expectations. For an earlier Australian study on Phillips curve flattening, see Kuttner and Robinson (2008). [39]

See, for example, Blanchflower and Bryson (2002), Bryson (2014) and Haldane (2017b). For Australian studies, see Cai and Liu (2008) and other references therein. [40]