RDP 2025-05: How Costly are Mark-ups in Australia? The Effect of Declining Competition on Misallocation and Productivity 5. Cost of Mark-ups in Australia

5.1 Aggregate productivity costs

We use the model to calculate how much lower productivity, specifically TFP, was relative to the economy where there was no dispersion in mark-ups and therefore misallocation. We show this both for the mid-2000s and late-2010s, and consider how these costs changed. These calculations reflect a ‘static’ assessment of the costs, as they do not consider flow-on effects in terms of investment choices and firm entry (the static social planner problem). They capture, all else equal, how much could we raise TFP by shifting resources and output across firms to offset the mark-up distortion.

We show three measures of these potential TFP improvements. The first is gross output misallocation cost. The second is value-added misallocation cost, which is more relevant in considering the impact on GDP. Finally, we also show a value-added measure excluding input distortions (which EMX refer to as ‘Value-added, $\mu$ = 1’). This latter measure is similar to the overall value-added measure, but is narrower and only includes the cost of misallocation across firms, whereas the main value-added measure also includes ‘input misallocation’. Input misallocation captures the fact that there is too much ‘churning’ of intermediate inputs, and so too much intermediate input use relative to final consumption. We calculate all three measures for both cost-weighted, and harmonic sales-weighted mark-ups.

Focusing on the cost-weighted mark-up we can see that the cost went up by 1.54 percentage points when focusing on the narrower value-added loss measure, and 3.67 percentage points for the broader measure (Table 4). Using the harmonic sales-weighted mark-ups the increases were slightly smaller, at around 1.12 and 2.52 percentage points, respectively. So had competition not declined, this suggests that productivity would be around 1–3 per cent higher due to a better allocation of resources across the economy.

The increase in the cost of mark-ups coming from misallocation reflects an increase in the dispersion of mark-ups across firms. In the cost-weighted case the gap between the 90th percentile and 25th percentile mark-up rose from 21 index points to 30 index points (Table 5). These are broadly in line with the change in dispersion seen in the actual data, where the cost-weighted gap went from 77 to 91 index points. Notably, the level of dispersion is much lower in the model than in the data. While we could use the observed dispersion in mark-ups to calibrate the model, EMX caution against this, noting that some of the dispersion in measured mark-ups may reflect noise and other factors.

As such, we should only incorporate the amount of dispersion that the model itself can justify based on theory.^[10]

It's important to keep in mind that these results suggest a level shift down in productivity due to misallocation. Whether or not these costs continue to rise will depend on whether or not mark-up and mark-up dispersion continue to rise. Existing work suggests that this was not the case over the two years to COVID-19 ( Andrews et al 2023; Champion et al 2023, Appendix A). Further work could consider post-COVID-19 outcomes once data become available.

5.1.1 Robustness

We consider a number of robustness tests to these main results. One small concern may be that in estimating the superelasticity we trim some firms, for example firms with only one year of mark-up estimates (as the regressions include firm fixed effects) and those with mark-ups below one (due to model assumptions). But our mark-up estimates are taken from the full sample in Hambur (2023). Focusing on the sub-sample of firms feeding into our superelasticity measure leads to higher markup estimates, and so higher productivity costs in levels terms. But the changes in costs across time are of a similar magnitude (Table B3). The results are also very similar if we use the sales-weighted average concentration measures, rather than the unweighted average.

A further robustness that can be considered is to allow the superelasticity to change over time, exploiting the time series dimension of the data. A key decision that needs to be made in doing so is how to split the sample when allowing the superelasticity to vary. There is no clear optimal window size to use, and any choice represents a trade-off between flexibility over time and noise coming from using a smaller sample. We consider the case of a three-year window for an, admittedly arbitrary, starting point.

Doing so, the superelasticity estimates have varied moderately over time. They increased moderately over the 2000s, peaked around 2010, then fell back slightly. Taking the start and end points (consistent with the rest of the paper), the superelasticity falls from around 0.11 in the mid-2000s to 0.09 in the mid-2010s. Incorporating this into the model leads to slightly smaller estimates of the change in the productivity cost of misallocation, but they are not substantially different and remain of the order of 1–3 per cent (Table B6).

5.2 Industry results

As discussed, one advantage of the Australian data is that we are able to estimate the key model parameters for a broad range of industries. This is valuable, as there may be significant heterogeneity in the key parameters, including mark-ups, superelasticity and concentration, across industries. Given the growing literature showing how input-output linkages matter, drawing out these differences can be informative. It can also help us understand more generally where these dynamics may account for more of the productivity slowdown.

Table 6 shows the key parameters, and their changes, across most 1-digit ANZSIC industry divisions.^[11] The first thing to note is that there is a very large range of superelasticity estimates across sectors, as noted in Champion et al (2023). In particular, those sectors that are downstream, and deal more directly with households, such as retail trade, tend to have superelasticities that are near zero (or even moderately negative). As such, the correlation between size and mark-ups is lower in these sectors. In accommodation & food services the estimated superelasticity is actually notably negative, indicating that smaller firms have higher mark-ups, rather than larger firm. Intuitively this makes sense. These downstream consumer-facing sectors are likely to have more small niche providers that have substantial market power due to quality and branding. That said, a full examination of the drivers of differences in superelasticities across sectors is beyond the scope of this paper.

In those sectors with near-zero superelasticities, taking the model at face value there will be no systematic misallocation across firms, and so no misallocation costs (though the welfare cost from having a positive aggregate mark-up still exists). As such we abstract from industries with superelasticity below 0.05 in absolute terms. We also abstract from the accommodation & food services. This is because the particular Klenow-Willis form of Kimball aggregator is not well behaved for a negative superelasticity.^[12] That said, its change in mark-ups and superelasticity are quite similiar to the aggregate economy (in absolute terms), so the costs in this sector may look broadly similar to the aggregate.

Table 7 shows the costs across the remaining divisions, and how they change. Again, the results suggest that the increase in cost has mainly been in upstream industries, as well as in professional services, which likely is an input into many other sectors.^[13] Recent papers have argued distortions in upstream input industries can be larger in terms of their impact on aggregate productivity, as the distortion gets amplified through creating further inefficiencies in the production network (e.g. Jones 2011; Liu 2019). As such, at face value it suggests that the earlier results may provide somewhat of a lower bound on the misallocation effects of decreasing competition. But alternative modelling frameworks would be needed to quantify this.

It is important to keep in mind that these results do not suggest that declining competition has not had a negative affect on activity in other sectors. For example, rising mark-ups would weigh on output in levels terms, not just due to misallocation. And they may have effects on incentives to improve. And we did see the average level of mark-ups increase in some of those sectors with near zero superelasticities, like retail trade. Here though we are only focusing on this one channel – misallocation across firms.

Footnotes

It's also worth noting that in the data, a non-negligible share of firms have mark-ups below one. The model does not allow for this. [10]

Note that value-added shares are also allowed to vary across sectors. These are taken from the ABS ‘Estimates of Industry Multifactor Productivity’ release. [11]

More precisely, quantities go very quickly towards infinity as mark-ups go towards zero. [12]

Some caution should be taken in interpreting the value-added cost estimates as, despite using different material shares, we have not allowed the elasticity of output with respect to value added to differ as estimates do not currently exist by sector. [13]