In a paper published in the Economic Analysis and Policy journal, Labor’s shadow assistant treasurer finds that companies paying an effective tax rate of less than 25% are shedding jobs while those paying 25% or more are growing their workforce at an annual rate of 2%.
Leigh has used the finding to reject the Turnbull government’s contention that lowering the company tax rate for companies earning more than $50m from 30% to 25% by 2026-27 will boost job creation.
So being a curious type I went and read the paper.
So a few little things occured to me that should also have occured to the referees of the paper and the journal editor but apparently did not.
Andrew Leigh collects data from IBIS World and the ATO’s Corporate Tax Transparency report. As far as I can work out that means he has three years worth of data for about 2000 entities liable for company tax. He doesn’t quite say so in the paper but I think he has used data for 2013/14, 2014/15, and 2015/6. He then averages the data over those years (why?) and then discards all those firms that made a loss – i.e. he truncates the data set (why?). He then estimates a regression where average annual employment growth (with three years of data he has two growth figures that are now averaged) is the dependent variable and effective tax rates is the independent variable. For robustness he adds in Log(Revenue) for a size effect (but didn’t he also use number of employees as a weight to control for size?). This is a mis-specified model for a start – there is a non-linear relationship between size and effective tax rates (ungated version here).
The biggest problem though is that he doesn’t seem to have controlled for profitability. It makes perfect sense that firms that are more profitable would both pay more tax and employ more workers. There is a correlation there. So his regressions should have included some or other lag structure between profit growth in the previous year (or investment) and current employment growth. That analysis is simply lacking. His averaging of the data specifically washes away any effect that we might expect to see. Look at his table 4.
First point: His analysis only provides a result for one of his effective tax rate proxies.
Second point: His adjusted R^2 are very low (except for one regression that I suspect is a typo).
Third point: He starts off with over 2000 observations and is down to 594 and 690 in those regressions with significant results.
Fourth point: He claims to include unprofitable firms in the analysis in equation 6 – but compare the sample size difference from equation 4. There are only 96 unprofitable firms? Compare the R^2 of those two models – down dramatically. So too the size of the coefficient – down from 0.253 to 0.188. I can’t tell if that is statistically significantly different, but it is a big drop and makes me wonder what would happen if he had controlled for lagged profitability?
Fifth point: He cites Tran 1997 on the Book-Tax Income Gap but makes no effort to control for factors that are likely to impact that gap.
So all up a highly incomplete and inadequate effort that would see an undergraduate econometrics student getting a very average mark were this a mid-term assignment. At the very least this analysis should be performed over a longer time period and probably over an entire business cycle. It is very likely that some firms were still carrying forward losses.
As always – this is not an invitation to slag my good friend Andrew in the comments.
Update: Another interesting observation – the paper was accepted for publication the same day it was received, and went online just 4 days later.
Gives a whole new meaning to speedy turnaround times. A tad suspicious – especially since the journal hasn’t updated the affiliations of its editorial board in several years.