Photo by CHUTTERSNAP on Unsplash https://bit.ly/3tm8CnZ

Innovation is the cornerstone of prosperity. With better technology, societies can do more and live better; think, for example, of what the world was like before the invention of the wheel, writing, or computers.

How do we get more innovation, then? One key policy lever is taxation: lower taxes for firms which perform research and development (R&D) should get us more innovation.

A quasi-experimental approach

In a recent paper, we study a policy change in Australia which resulted in lower taxes for R&D performing firms. In 2012, the Australian Government replaced the R&D Tax Concession with the R&D Tax Incentive. We calculated that the new policy reduced the after-tax cost of R&D by 45 per cent for a company with no tax liability, and by 15 per cent for a company with a positive tax liability. This is a sizable change – but did it work?

To find out, we use an econometric method called difference-in-differences. We compare changes in R&D spending between 2011 and 2012, across two groups of firms: (1) treated firms, who went from claiming the Concession in fiscal year 2011 to claiming the Incentive in fiscal year 2012; and (2) control firms, who claimed neither in either year. The change in policy generosity only affects treated firms, not control firms.

The difference-in-differences method assumes that, in the absence of a policy change, treated firms would have been on the same time path as control firms, and thus estimates the effect of the policy change as the difference between the outcomes for treated and control firms. The method is quasi-experimental in the sense that it mirrors what a true experiment (in other words, a randomised controlled trial) would have done, if running one had been possible. Our paper is part of a so far small literature using quasi-experimental approaches to study R&D tax support schemes: other papers have used difference-in-differences and regression discontinuity, another quasi-experimental design, to study the effectiveness of a 2008 reform in the United Kingdom.

Data

We draw on a newly released dataset from the Australian Bureau of Statistics (ABS): the Business Longitudinal Analytic Data Environment (BLADE). BLADE is a confidential, anonymised firm-level dataset, which combines data from across several government sources and ABS surveys to facilitate comprehensive analysis of Australian businesses. Our main variable of interest, R&D expenditure, is drawn from the ABS Business Expenditure on R&D (BERD) survey, which uses a standardised and time invariant definition of R&D activity. Although we cannot perform an analysis using the entire population of Australian firms, the BERD sample is representative, since it is collected for a stratified random sample of firms. 2,671 R&D-performing firms are included in our analysis.

Findings

We find not only that lower taxes do work, but also that the effect is large: each dollar of tax revenue forgone by the Australian Government induces R&D-performing firms to spend $1.90 on R&D.

Earlier, non-quasi-experimental results placed this figure at about $1 for OECD countries, which raises the question of whether our estimate is too large. What do other quasi-experiments find?

Guceri and Liu (2019) and Dechezlepretre et al. (2016) respectively report figures of $1.60 and $1.70, both for the UK, which are very close to our estimate. Considering that quasi-experimental estimates are generally more credible than strictly correlational results, we think our result is sensible and of interest to policymakers in Australia and elsewhere, as long as one does not think that the UK and Australia are outliers among OECD countries. We do not believe they are.

Potential concerns

One key concern with our approach is that control firms do not claim tax benefits. Who are these firms, and why are they not claiming benefits?

It turns out the majority of control firms in our study do not claim tax benefits because they are not eligible to do so. R&D financed by government grants is not eligible: 71 per cent of non-claiming firms do have government grants, so they cannot claim tax benefits. Most of the other 29 per cent of non-claiming firms likely perform other ineligible activities, such as software development or contract R&D (that is, R&D financed by other Australian firms). Among other things, we thus account for whether firms are grant recipients in our paper. Our results remain unchanged when we do.

A common problem with the type of approach we use in our work is that we do not know for sure whether the growth rate of R&D expenditure would have been the same for the two groups of firms in the absence of the policy. Ultimately, the difference-in-differences  assumption, which provides the statistical basis for our result, should be scrutinised.

A convenient way to assess whether R&D expenditure would have been on a similar time path across the treatment and control groups is to perform a placebo exercise: we take the two groups of firms and see how they tracked between the fiscal years 2010 and 2011, when there was no change in the generosity of R&D tax benefits for either group. If the outcomes for the two groups moved in tandem before the reform, we can be more confident that they would have continued to do so if a reform had not been implemented. This turns out to be true: we find no difference between groups in the pre-reform period – in fact the placebo difference is almost exactly equal to zero.

Why is it good for Australia?

Clearly, our estimate that $1 of tax revenue forgone by the government causes a $1.90 of R&D spending by firm is good news. But firms’ increased spending on R&D will likely make them more profitable in the future, meaning their future taxable income rises, which means more tax revenue for the government. So, there is good reason to believe our $1 / $1.90 result is an understatement of the true benefits Australia reaps from the R&D Tax Incentive.

 

Further reading

Holt, J., Skali, A., & Thomson, R. (2021). The additionality of R&D tax policy: Quasi-experimental evidence. Technovation107, 102293, https://doi.org/10.1016/j.technovation.2021.102293.

 

Leave a comment

Your email address will not be published. Required fields are marked *

*