There’s a lively debate going on about who pays for compulsory superannuation. Even though employers write the cheque, it has long been conventional wisdom in Australia that workers ultimately bear the cost of super through lower wages. But this conventional wisdom has been challenged recently by some people who claim that workers won’t bear the cost of increases in super now, and perhaps never did bear the cost of super in the past.
Grattan Institute’s latest paper No Free Lunch: higher super means lower wages sought to shed light on the debate. In this paper, we examine a detailed administrative database of enterprise bargaining agreements (EBAs), going all the way back to 1991. We find that wages growth in EBAs that cover increases in compulsory super is lower than wages growth at other times, after taking account of other factors that affect wages growth. Our analysis shows that 80 per cent of the cost of increased compulsory super is passed on to workers via lower wages over the course of an enterprise agreement, typically 2-to-3 years. The long-term impact is likely to be even higher.
Yet a new report by Per Capita seems to argue that not only does higher super not come from workers’ wages, but that the postponing of previously-scheduled increases in the Super Guarantee (SG) is also why real wages have fallen since 2014. As the authors argue:
Therefore, as a result of the freeze on the SG rate in 2014, the average worker has lost $4332.99 in super over the intervening five years, and their take-home pay has declined by $1092.00 in real terms, giving them a net loss of $5424.99.
The argument seems to be that because your employer didn’t have to pay you more super, they also didn’t give you a pay rise. But that ignores the myriad other drivers of low wages growth, from slower growth in productivity, technological change, globalisation, an under-performing economy, or weaker bargaining power among workers. And this is on top of the assumption – and it is just an assumption – that super is a free lunch for workers.
The Per Capita report follows an analysis by Jim Stanford at the Australia Institute in November 2019 which found no significant trade-off between higher super and lower wages, and in some cases found a positive association – that is, higher super being associated with higher wages growth. While the econometric approach Stanford used suffers from serious flaws1, his approach at least sought to measure the impact of higher super on wages after taking account of some of the other potential drivers of low wages growth. The researchers from Per Capita simply assume there is no trade-off.
The Per Capita authors and Stanford have also made a number of unsubstantiated claims about our work. These myths are worth correcting.
Myth: Grattan claims that super is the main – even only – factor that affects wages growth
The first and biggest misunderstanding of our work is the suggestion that we claim super is the main – or even only – thing that determines wages growth. This is not true. Our results imply that when super goes up, wages will end up being lower than they would have been if super hadn’t gone up. This does not mean that wages growth should be fast during periods when super isn’t rising, and slow when it is going up. Wages growth is affected by a broad range of economic and social factors – super is just one of them.
Myth: Grattan assumes an unchanged super-wages relationship since 1991
Emma Dawson and Shirley Jackson from Per Capita claim that:
The most problematic element of the [Grattan] analysis is that they have applied the same methodology and assumptions to EAs throughout the 27-year period (1991 – 2018) for which the [workplace agreements] data is available, apparently ignoring the different economic circumstances and policy settings that shaped the macroeconomic context in which wage movements occurred over this time period.
Jim Stanford of The Australia Institute similarly asserted that we ‘assume stability over the entire 1991-2018 period. This conflates periods with very different institutional & policy contexts.’
This too is not true. The econometric models we use in No Free Lunch account for different time periods, precisely to allow for the possibility that differences in labour institutions and regulation have affected wages growth. We use a variable – called labour_law – that takes on a different value for each of four periods: pre-1996 (the beginnings of enterprise bargaining), 1996-2006 (the Workplace Relations Act), 2006-2009 (Work Choices), and the period since mid-2009 (the Fair Work Act).
Our models also – crucially and conventionally for work of this sort 2 – include lagged average wages growth as an explanatory variable. This means that if average wages growth is 3 per cent, the model will ‘expect’ lower wages growth in an agreement than if average wages growth is 4 per cent, other things being equal. Changes in policy that affect overall wages growth are reflected in our model through this variable.
Stanford also stated that he ‘would like to see separate post-2008 results’. These are included in our paper. Our main result – 80 per cent of super rises are passed through to workers via lower wage rises during the life of an EBA – comes from data covering the period 1991-2018. But we also estimate the model using data from the period 1997-2018 and the period 2008-2018.3 When we fit the model to these periods, we find that the degree of pass-through from higher super to lower wages is broadly consistent with what we find when we look at the full period since 1991.4
Myth: Grattan’s analysis excludes the terms of trade
Jim Stanford claims that our models do not include the terms of trade, the price of Australia’s exports relative to our imports.
Again this is not true. We agree that the terms of trade is an important factor affecting wages growth, particularly in the past decade or so, which is why it is in every model we use in the paper.5
Myth: Grattan’s argument is ‘highly theoretical’
The researchers from Per Capita claim that our argument in No Free Lunch is ‘highly theoretical’ and ‘does not provide compelling evidence’ on the super-wages trade-off.
It’s true that our paper includes a description of relevant economic theory. But the core of the paper – the bulk of its contents – consists of empirical evidence, not theory. Our paper does not conclude that 80 per cent of the cost of higher super is passed to workers on the basis of theoretical reasoning. Our conclusion is instead a statistical finding arrived at using detailed administrative microdata – actual data about enterprise agreements in Australia.
By contrast to the Per Capita researchers, who think our argument is highly theoretical, Jim Stanford claims that we ‘do not present a theoretical model of why wages and SG might mostly offset each other’ and that we assert that ‘theory “suggests” a large trade-off, without explaining how and why.’ This is also inaccurate.6
Myth: Grattan’s analysis relies on a large number of dummy variables
Stanford claims that our results rely on a ‘huge number of dummy variables’.
This is not true. The first model we present, on page 32 of No Free Lunch, does not include any dummy variables. We present the results of this model to show that our results are broadly consistent in simple models as well as more complex models that more fully reflect the range of factors affecting wage setting. The results are indeed broadly consistent – using this overly simple model, we find 70 per cent of the cost of super is passed through to wages over the course of an EBA (see page 34). The explanatory power of our model rises as we add more variables – as would be expected – but the core result is broadly consistent across the various models.
It is a virtue of working with a large sample size (80,000 agreements, rather than 32 years of average wages growth) that we are able to control for a lot of the factors that affect wages growth, but our results do not rely on this sophistication – the result is consistent in the simple model.
Myth: the past tells us nothing about the future
The Per Capita researchers say:
there is no evidence whatsoever to show that a freeze in the SG rate will deliver bigger wage increases. It is one thing to show that previous increases in superannuation have come partly from wages; it is quite another to prove that holding down the rate of superannuation will result in bigger pay packets in future. (p. 10)
In our paper, we provide evidence that super has come out of wages in the past. Wage rises within EBAs that spanned increases in compulsory super were lower than in EBAs where super didn’t rise. We show that this is true in the private and public sectors, is true in the post-Global Financial Crisis period as well as the pre-GFC period, and is true using a range of different modelling assumptions (see Appendix B in No Free Lunch).
Given all this evidence, anyone claiming that the future will be different from the past should provide good reasons. Per Capita does not. Wages growth is slow now, but it was also slow in 2013 and 2014 when compulsory super most recently went up, and we find that the super-wages trade-off has not weakened over time.
Myth: Grattan’s analysis doesn’t take into account bargaining power
Per Capita says that our model:
fails to fully account for the impact that relative density of union membership has on wage setting… the Grattan authors incorporate a variable for union influence that is a vector of dummy variables for each of the 82 unions mentioned in agreements. The dummy variable is a method of accounting for the presence or absence of unions and does not adequately account for the relative density and corresponding bargaining power of trade unions in each of the negotiations.
If workers’ bargaining power has declined in a particular industry, you would expect to see average wages grow more slowly in that industry. Average wages growth at the industry level is included in our model, and the effect of this on wages growth in agreements is allowed to vary by industry. It is not true that bargaining power is not taken into account in our models – it is taken into account indirectly, via the effect on overall and industry-level wages growth, as well as through a variable that reflects which union(s) were involved in negotiating the agreement.
Stanford notes that:
Aussie labour compensation has lagged far behind productivity for years (in spite of SG, not because of it), due to weaker bargaining power and pro-employer policies. If we change those policies, there’s clearly room for both higher wages and stronger retirement security.
It’s plausible that if bargaining power were to increase, workers may be more able to resist efforts to pass the cost of super on to them in the form of lower wages. But, for a given level of bargaining power, higher compulsory super is likely to mean even lower wages growth.
Our report did not answer the question, ‘If workers obtained significantly increased bargaining power, would super come out of wages?’ Rather, we answered the question, ‘In Australia, over the past quarter-century, under the actually existing level of workers’ bargaining power, did super come out of wages?’ This is likely to be a better guide to what will happen if the SG goes up next year than imagining a world in which both compulsory super and workers’ bargaining power rise.
Myth: Grattan’s results understate the uncertainty around the super-wages trade-off
Jim Stanford suggests we might be overstating the precision of our results, saying ‘econometricians love huge micro data sets like the federal EA database, because it’s easier to find statistical significance in any results (esp w dummies)’.
We do not love huge micro data sets because they make it easier to find statistical significance. We love them because they contain a richer range of information that allows us to estimate models that more closely reflect the complexity of real-world phenomena, such as wage-setting.
In No Free Lunch (page 37), we acknowledge that working with large datasets can lead researchers to overstate the precision of statistical relationships. To address this concern, we do two things. These are technical in nature, but important. First, we estimate our models using ‘robust standard errors’, a common technique that widens the confidence intervals around our results. Second, we check these standard errors by also using the ‘bootstrap’ – a technique for running our model several thousand times on subsets of the data, to see if the results still hold. We find that our results and the confidence intervals around the result are broadly similar when we use the bootstrap.
Myth: Our findings are invalid because enterprise agreements ‘now cover fewer than one in five workers’
There is some truth in the one-in-five figure, cited by the Per Capita researchers and Stanford, but it by no means invalidates our findings.
About 38 per cent of Australian workers are on enterprise agreements (about 30 per cent on federal agreements and about 8 per cent on state agreements). This means that 3.2 million workers are paid according to federal EBAs. However, only 2.19 million of these workers – 20.5 per cent of all Australian workers – are on agreements that have not expired. Expired agreements keep going – they still determine workers’ pay and conditions – until they are replaced or terminated.
In the short term, workers on expired agreements don’t get a pay rise – whether or not super goes up. But in the future, they’ll move off these expired agreements, whether because they negotiate a new agreement, fall back to the award, negotiate a one-on-one arrangement with their boss, or change jobs. When they do, they’re likely to find that the wage they can get in a world with compulsory super at 12 per cent is lower than the wage they could have got if compulsory super had stayed at 9.5 per cent.
Our empirical evidence relates to enterprise bargaining agreements. The critics of our work point out that these cover only a subset of the workforce. But the critics provide no reason to think people on EBAs have less bargaining power than other workers – and are therefore less able to resist employers shifting the cost of super onto them.
The Reserve Bank agrees that 80 per cent of higher super comes from wages
But don’t just take our word for it. The Reserve Bank, in its latest Statement on Monetary Policy, makes it clear that it agrees there is a trade-off, and that it is forecasting wages to grow more slowly over the coming years than it would’ve if super wasn’t set to rise. And in evidence to the House Economics Standing Committee on Economics, Deputy Governor Luci Ellis said:
Historically, about 80 per cent of the increase in the non-cash benefit tends to show up as somewhat slower wages growth than you would otherwise have seen…
Why all this matters
Under legislation supported by both sides of federal Parliament, compulsory super contributions are scheduled to increase incrementally from 9.5 per cent of wages now to 12 per cent by July 2025.
The key question we answered in No Free Lunch was: ‘What happens to wages growth when super goes up?’ Answering this question helps policy makers to assess the costs and benefits to workers of higher super. If super was indeed a free lunch for workers – if higher compulsory super came at no cost to them – then the case for much higher compulsory super might look stronger.7 But if higher super means lower wages, then policy makers need to weigh the benefits of higher super, namely higher incomes for people in their retirement years, against the costs, namely lower incomes for people during their working years. We found in No Free Lunch that there is indeed a trade-off between super and wages.
We think compulsory super is a good thing. But you can have too much of a good thing.
Grattan Institute’s 2018 report, Money in retirement, showed that the trade-off between more super in retirement and lower living standards while working isn’t worth it. It demonstrated that most Australians can already look forward to a comfortable retirement, and that raising compulsory super would force many Australians to save for a higher living standard in retirement than they enjoy when working. Where Australians are at risk of poverty in retirement — namely renters — higher compulsory super is unlikely to help much. Instead other policies, like boosting Commonwealth Rent Assistance, are needed.
If people have lower living standards while working, they are less able to be able to afford to buy a home, or invest in their children’s education, or start a business. And it leaves vulnerable Australians at greater risk of poverty or financial stress. And making Australians save more than they need (or are likely to spend) in retirement is a recipe for larger inheritances, which will ultimately exacerbate wealth inequality.
Footnotes
1. See No Free Lunch, pages 17-20.
2. For example, see the Reserve Bank’s ‘Phillip’s Curve’ model.
3. The results of these robustness tests are on page 62 of No Free Lunch.
4. Table B.6 of No Free Lunch shows that the coefficient on our super variable is 1.086 over the period 1997-2018 and 0.996 over the period 2008-2018. The confidence intervals get wider when we use smaller subsets of the data.
5. See pages 32 and 33 of No Free Lunch for a specification of the econometric models adopted, each of which include the terms of trade, denoted tot_change.
6. We do present a theoretical model of why wages and SG might mostly offset each other: see pages 10-12 of our report. The report notes that theory, by itself, does not have an unambiguous prediction about who bears the burden of mandated benefits like super. Instead, it depends on how much workers value super relative to wages; and the relative elasticities of supply and demand for labour. We set out the logic, which reflects classic papers in the economics literature on mandated benefits.
In brief, the theory goes like this. If a tax or compulsory payment – such as super – is imposed on employers, this adds to the cost of employing people. This means that, at a given pre-super wage rate, employers will be less willing to employ people. At the same time, if the payment is strongly tied to benefits that workers receive (as super is) then workers will be more willing to work at a given ex-super wage rate. This combination of reduced labour demand and increased labour supply leads to wages being lower than they would have been if the compulsory payment hadn’t been levied. Importantly, this logic does not rely on the assumption that the labour market is perfectly competitive.
7. Of course, if higher compulsory super was paid for by employers and that in turn led to higher unemployment, it might not be in all workers’ interests.