Replicating the WSJ's "Laffer curve" graph

Back on July 13, the Wall Street Journal editorial page published an editorial (sub. req.) claiming that "Lower corporate tax rates with fewer loopholes can lead to more, not less, tax revenue from business," a claim that it attempted to support with this graphic:

But as numerous bloggers pointed out at the time, the alleged "Laffer curve" drawn in the graph is absurd. It's fitted directly through the data point for Norway, an obvious outlier with significant oil revenue (and an omitted excise tax), and then plunges straight down toward zero (who knew that increasing your corporate tax rate from 28% to 32% was so destructive?).

Well, Kevin Hassett of the American Enterprise Institute, who provided some of the WSJ's data, released a defense of the Laffer curve yesterday (PDF) that he co-wrote with Alex Brill. By using the data sources he provided, I was able to reconstruct the Journal's data exactly. (Replication data and the Stata .do file for this analysis are here.)

Unsurprisingly, when I fit linear and quadratic models to the same data (29 OECD countries plus the United Arab Emirates in 2004) using the approach of Brill and Hassett (regressions with a linear and squared term), the predictions come nowhere close to WSJ's "Laffer curve":

When we exclude the UAE, which is not directly comparable to OECD countries and seems mostly to be included because it has a zero corporate tax rate, the results become still less favorable to the WSJ:

When we exclude three possible problem countries identified by Brill and Hassett: "Ireland (a noted tax haven), Norway (a country with unusual oil revenues), and Switzerland (a country with significant internal variation in taxation)," there is again virtually no difference between the predictions of the linear and the quadratic model and certainly no evidence of a "Laffer curve":

The difference from the original WSJ graphic is especially clear when we put these results on the same scale and place the graphs side-by-side:

In sum, let's just say that Rupert Murdoch shouldn't let these guys do his books...

(Postscript: As for the larger issues Hassett raises, I haven't replicated his analysis, but I remain skeptical that the negative sign on the squared term in his quadratic models is evidence of a substantial Laffer effect for a variety of reasons.)

Update 8/2 2:10 PM: As commenters have pointed out on the blogs of Matthew Yglesias and Kevin Drum, the dependent variable here -- corporate tax revenue as a percentage of GDP -- doesn't make a whole lot of sense. I agree -- that's part of what I was alluding to in the postscript about my skepticism. The purpose of this post, however, is simply to show that even if we grant that the WSJ's measures are appropriate, the data don't prove what the Journal claims. I make no claims about the value of my regression results, which are provided solely as a counterpoint to the original graph.

Comments

[GradStudent]

I love that you use Stata. Maybe Dook does some stuff right after all. HaHa.

You could use a Stata ado file to conduct a Grubbs test to see if any countries would be classified as official outliers. I'd bet Norway is a naturally occurring outlier, and not just a theoretical one.

[Marie]

I'm hopeless at economy, but there's two data points that seem a bit strange: France and the US seem to have nearly the same corporate tax rate. Income tax is supposed to be much higher in France than in the US, yet the corporate tax revenue is a higher percentage of the GDP in France than in the US.
Isn't that contradictory?
And does that mean that people who are always going on on the fact that businesses are taxed much higher in France than in the US don't know what they're talking about?

[eightnine2718281828mu5]

It's also been said that the Norway data point is an 'outliar' since it was constructed by including revenue from petroleum excise taxes while removing that tax from the marginal tax rate side of the ledger. If the excise tax is included in the marginal rate, it pushes norway's rate up to 40-50%.

And of course that chart would never see the pages of the WSJ.

[eightnine2718281828mu5]

Whoops; just saw you had a link to the outliar data. Haven't had my coffee yet. :-)

[DMoore]

The corrected curve indicates that it might be very beneficial to decrease corporate taxes from 35% to 25% in the US. You might succeed in decreasing taxes substantially, while receiving a very similar amount of revenue. I assume that's over the long term, as I would expect a short term (how many years is short term?) decrease in revenues before a hoped for increase in growth could close that gap.

However, the ideologues almost certainly prevent this from occurring. Rather than propose a change that might have real benefits, and rather than argue for it based on a rational analysis of the data, the WSJ makes an obviously bogus argument that damages the credibility of proposals to decrease corporate taxes.

The ironic part is that their bogus curve actually peaks around 25% also. I'm not sure if they're lacking in integrity or just don't know how to apply math. I'm continually depressed and disappointed at how frequently I see policy debates conducted with either fabricated or incompetent arguments, when honest arguments could easily be made.

[Glenn]

I understand the point about excluding the UAE, but shouldn't the curve be forced through the origin (which is pretty much where the UAE is)?

[Joe D]

Confidence intervals, anyone?

[spencer]

The study uses the legislated rate for the US, but the effective rate is actually about half of that -- federal corporate tax revenues as a share of pretax profits. On this basis the US would be located about the same point on the bottom axis as Iceland.

Does anyone know if the other countries have a similar spread between the headline rate and the effective rate?

[Brendan Nyhan]

left out confidence intervals for simplicity, but they're obviously highly overlapping for both models (and wide in general). if people want to see them i can put them up...

[TrishB]

What's the correlation coefficient on the model?

[Clueless]

If you take the Laffer assumptions: 0% tax rate = 0 revenue, and 100% tax rate = 0 revenue, and plot a quadratic, you will find that revenue is maximized at the 50% tax rate. So, there is a Laffer argument for increasing taxes in the US.

Clueless.

[Brendan Nyhan]

Trish, the bivariate correlations are .31 (all data), .14 (no UAE), and .24 (no problem countries).

[KAP]

As an engineer, when I look at this data I see no correlation, or at best a weak correlation. I would guess that the goodness of fit characteristic (r squared)of both of these lines (linear and quadratic) to be quite small.

As fare as I am concerned, the real story is that the data does not support a relationship between tax rates and coporate revenue as a percent of GDP.

[Seth]

Ah the WSJ. I can see it will be a great loss to sell this national treasure to a partisan hack like Murdoch.

Who fit that curve? I know first-year students who would have found that Laffer laughable.

[Questionable]

I might be missing something here, but they weren't trying to fit the data. They were illustrating where the Laffer curve should be, and where various countries are.

Apparently they did a poor job explaining that part.

I think the analysis indicating the curve is incorrectly drawn is right-on, since 33% tax will not give us zero taxes. The curve is slanted compared to where it should be, but it isn't a data fit.