Christina Romer: Tax Cuts and GDP Growth

I am posting out of order. This post is part of a longer piece discussing Christina Romer’s views on the efficacy of fiscal stimulus.

One of the difficult elements of studying fiscal influences on growth is the lack of counterfactuals and exogenous variables. Large changes to government spending or to taxes almost always occur for a reason and most of the time these reasons are cyclical: the changes are made with the explicitly intent of boosting output. [The big exception to this is spending and tax changes during wars. Krugman, Romer, and other Neo Keynesians wish to exclude war spending and taxation from the current analysis. I disagree but let’s play by their rules.]

In Romer’s words (see her speech here), all studies of fiscal stimulus are plagued by omitted variable bias. Therefore, she does not find it surprising that (almost) all previous empirical studies of relationship between fiscal policy (both spending and tax) have found very small multipliers. In her view, the omitted variables are biasing the results down. Of course, since they are both omitted and unknown they could also be biasing the results up.

To bypass the problem, Romer and Romer (2008) use the narrative record to divide large tax changes into exogenous and endogenous pieces. They use the Congressional record and Presidential speeches to separate tax policy actions taken for reasons related to economic activity and those taken for non-economic reasons. They have the right idea: if we can identify exogenous tax moves, and if there are a sufficient number of them, we can identify the impact of tax changes on economic growth.

Romer and Romer find that a tax cut equivalent to 1 percent of GDP will boost output by 3 percent over ten quarters. While I believe tax cuts can stimulate growth, I do not find their results at all convincing. The tax cuts they identify as exogenous are not: the majority of their tax cuts occur in the depth of recessions. Since economies tend to bounce out of recession very quickly, their model mixes the effects of tax cuts with growth recoveries. In particular, their inclusion of a series of tax cuts in 1981 and 1982 significantly bias their results.

While this paper is an interesting academic exercise, to use these results as the justification for fiscal stimulus is wrong.

The Details

Figure 1 on page 46 in the linked paper shows their time series for exogenous tax changes. The majority of tax changes identified as exogenous are tax cuts, apparently tax hikes are less likely to be exogenous. The largest tax cut occurs in 1947 and is almost 2 percent of GDP. Then, there are clusters of tax cuts beginning in 1964 and 1981. Each of these clusters contains 4 cuts and the cuts are made over about 2 years. The final two significant cuts occur under Bush and are implemented in 2001 and 2003.

Although I applaud their attempt and they do have a good methodology, the dates they have chosen do not seem random in an economic sense. Before we turn to a general look at their dates, let’s take a look at the last two cuts: 2001 and 2003. These I know a bit more about.

Independent of the reasons given in speeches at the time of their passage, these tax cuts were proposed and enacted to counter the 2001 recession. I realize Romer and Romer divide the multitude of tax cuts over this two year period into different categories, some exogenous and some endogenous; and, it is true that the Bush administration was ideologically pro-tax cut; nonetheless, these were all implemented to boost economic output. In the absence of the recession in 2001 and the weak labor market in 2003 these tax cuts would not have occurred.

This highlights the fundamental difficulty Romer faces: the content of speeches and the congressional record, do not always reveal the full story.

Take a look at the picture below. The picture shows GDP growth over different horizons surrounding the key tax cuts in the Romer-Romer data. The first bar in each set shows average GDP growth over the 8 quarters before the tax cut. The second bar shows GDP growth for the four quarters ending at the date of implementation. The third bar is GDP growth for the 8 quarters following the cut and the last two bars show average growth for the 10 years before and after, respectively.
The first thing to notice is that with the exception of 2003 and 1971 growth in the year prior to the tax cut was slower than growth on either side of the cut. That is, whether or not there is a recession, these tax cuts only seem to occur when economic growth is relatively slow. My interpretation of this finding (especially when combined with specific knowledge of the 2001 and 2003 cuts), is that the majority of the tax cuts in the post war period are not completely exogenous in the sense Romer and Romer need them to be.

In particular, take a look at the bars labeled 1981. Reagan, like Bush, was philosophically pro-tax cuts. He would likely have tried to implement tax cuts independent of the economic situation. Whatever the counterfactual, these tax cuts were made in the midst of the (at least until now) deepest downturn of the post-war era.

The four tax cuts which occurred in 1981 likely provide much of Romer and Romer’s identification. The four tax cuts were followed by a monumental acceleration in growth: from almost -2 percent to almost plus 5 percent. Their statistical model attributes all of this acceleration to the tax cuts, not to the fact that the economy was bouncing back from a recession.

So, while the tax cuts may have helped boost growth, the model attributes to much growth to the tax cuts and not enough to the cyclical state of the economy. In other words, Romer and Romer suffer from omitted variable bias just as other studies do. To use these numbers as the basis for fiscal policy is wrong.