(First read Stuart’s post immediately below.)
This is probably a “of course we thought of that” kind of question, but: Stuart, for those os us who haven’t read either version of the paper– do the authors show that the relationship is inverse even after countercyclical effects are taken into account?
That is, in a recession, spending rises and taxes fall automatically. The reverse happens under conditions of robust growth. Not all spending or all taxes change in these ways, but, e.g., income and capital gains taxes move rather a lot in response to the business cycle, as do unemployment benefits and– prior to welfare reform– welfare payments. It seems to me that this could account for a great deal of the inverse relationship they note, no? (I expect the answer to be, “Yes, it accounts for a lot, but they show that it doesn’t account for all”– just figured it would be worth getting the issue onto the table.)
Of course, one could also imagine a combination rational-expectations/ supply-side theory, as follows: when spending rises, regardless of whether it’s financed by current taxation or deficit spending, people respond as if they’re faced with a tax increase (because, eventually, they will be). Therefore there’s an economic slowdown, therefore current tax revenues fall. Conversely, a spending cut is interpreted as a cut in (the present discounted value of all future) taxes, and supply-side effects kick in, improving growth and increasing current tax revenues. I doubt that this could be very much of the explanation, but again, just thought I’d mention it.
Update:
Kevin Drum links to the paper itself (pdf). I think the answer to my first question above is complicated. For spending and taxation the authors use expenditures/ revenues as a share of GDP. That’s entirely reasonable, probably even standard. But it complicates the countercyclical issue by putting GDP into the denominator. If this year GDP is 1000 and tax revenues are 200 (for a ratio of .2), and next year GDP falls to 950, tax revenue is designed to fall to some point below 190, and to give a ratio below .2. Using the revenue-GDP ratio will show a countercyclical effect, but a small one; the ratio might fall from .2 to .19, a 5% drop. Actual revenue will fall from 200 to 180.5, almost a 10% drop.
Conversely, putting GDP into the numerator of the spending side exaggerates countercyclical spending increases. During boom times, it will exaggerate spending decreases and understate tax increases.
And then, to complicate matters further, unemployment is (as Kevin notes) part of the equation– but only on one side. That is, unemployment is allowed to affect changes in spending but not changes in tax revenues, which are taken as independent.
Figuring out how all of this adds up is the sort of thing that kept me in political theory and away from the kind of statistical social science that my wife does. I can’t wrap my head around whether all of this yields an adequate modelling of countercyclical effects or not. I’m happy to give the authors the benefit of the doubt– I sure wouldn’t want to go before the Public Choice Society with a basic modelling mistake in my equation, and Niskanen’s building on a paper that was published a while ago, so if the model were wrong others probably would have caught it by now. But it seems odd to me. I think there’s collinearity between unemployment and taxation/GDP– as unemployment rises, Tax/GDP falls, independent of the effect either has on spending/GDP. On its face, that’s a problem. Maybe it’s the problem that “autoregression” solves; my education ran out before we got to autoregression. But my impression is that autoregression is a way of dealing with time effects.
The semi-supply-side theory mentioned above seems unaddressed and as compatible with the results as the guess that the authors venture. Distinguishing between these would require some pretty complicated work.
A word about the paper’s other big idea, which Kevin also picks up on:
American participation in every war in which the ground combat lasted more than a few days – from the War of 1812 to the current war in Iraq – was initiated by a unified government. One general reason is that each party in a divided government has the opportunity to block the most divisive measures proposed by the other party.
Well, maybe. But maybe it’s that, in times of objective external security threat, there’s a tendency on the part of voters to rally ’round the President and reward his party, or otherwise to generate unified government. There is, after all, usually a period of security threat that precedes the outbreak of war; and during those periods voters might decide that divided government is a luxury they can’t afford. It’s not the case, as the authors suggest, that the only alternatives are sheer coincidence< and unified governments are unrestrained in their pursuit of ‘divisive measures.’
NB: I am not not not an economist, or even a quantitatively-trained public choice political scientist. I’m talking way out of my depth here, and Niskanen is a very prominent and respected public choice scholar. But the authors freely acknowledge that their explanations are guesswork, not uniquely dictated by anything in their results. I’m trying to think through some alternative guesswork explanations.
Update again:
I checked with the in-house statistician about autoregression, and she confirms that it’s the “this year is basically similar to last year” term. The autoregression soaking up all the explanatory power means that the model has no explanatory power for change over time. This is apparently a not-uncommon problem with time series data and makes working with them difficult.
I also asked her whether it was common to do as the authors did and report coefficients and standard errors but not p-scores; political scientists routinely either report the p-score or graphically represent them (* p<.05 ** p<.01 etc) and this paper did neither, leading me to feel like a baffled little monkey trying to figure out significance. She proceeded to make fun of political scientists who can't glance at a coefficient and standard error and divide by two in their heads, and confirmed that the authors' practice isn't particularly unusual.
Comments are closed.