TAX RATES AND CONSUMER SPENDING
Introduction
In August 2020, The Denver Post published an op-ed piece by economist Steven Pressman, in which he stated that higher taxes reduce consumer (and business} spending. That was not the point of the article, and, indeed, was rather an aside. However, the statement seemed dubious to me because of all the quantitative studies I had already done on taxes and GDP growth: none of them showed any negative correlation between the two variables over multiple decades of U.S. history. If higher taxes don’t affect GDP growth negatively, then I was doubtful that they would do so for consumer spending.
On the Bureau of Economic Analysis website[1] I found data tables for real (adjusted for inflation), per capita consumer spending (specifically, “personal consumption expenditures” in the BEA tables). For tax rate I used the Saez and Zucman tables for total (local, state, & federal) effective tax rates paid by all U.S. taxpayers[2]. When I did a linear regression analysis between these two variables for the 70 year period 1950-2019, I found not a negative correlation (as one might expect if higher taxes caused lower consumer spending) but rather a statistically significant, positive correlation and a positive regression line slope. Figure 1 illustrates the linear regression, and the correlation coefficient had a value of r = 0.47.
I actually sent the figure and the results of the linear regression to Pressman and asked how he knew that higher taxes caused lower consumer spending. Although he was polite enough to answer, he pretty much blew me off, saying essentially that all economists know that, and that my result just showed that when the economy is good people spend more and pay higher tax rates because taxes are progressive. In other words, in his opinion, both of my variables are the reflection of another hidden one: the general state of the economy.
He also specifically referred me to a 1957 paper by Milton Friedman (A Theory of the Consumption Function) which he said proved the point and won Friedman a Nobel Prize. Well, who was I to question a Nobel Prize winning paper or a famous economist like Friedman? But question I did.
FIGURE 1
Friedman
Figuring that you’re never too old to learn something, I actually downloaded an electronic copy of Friedman’s paper and began to read it. Even an amateur like me could quickly see that the paper was trying to establish a relationship between income and consumer spending, not taxes and consumer spending. One advantage of a digital electronic copy was that I could do a word search for “tax”, which also picks up all instances of “taxes” and “taxation”. There are about 15 such occurrences in the 200+ page document. All of them are simply explanations that taxes are included as a component of consumer spending, or are identifications of some income under consideration as being before or after taxes. Nowhere did Friedman prove, or even assert, that higher taxes cause reduced consumer spending. Even if he had, I would not have been overawed, because all of his data, of necessity, was from the first half of the 20th century, and even some from the end of the 19th century. In fact, he never considered the relationship between taxes and consumer spending at all.
Pressman had either forgotten, or never knew, the real content of Friedman’s famous paper. Such ignorance certainly reduced my inclination to defer to him as an expert.
What is Affecting What?
After plowing through A Theory of the Consumption Function I was ready to find a way to test Pressman’s other assertion that the positive correlation of Figure 1 occurred because both variables move up or down together according to whether the general economy is good or bad, rather than because either is affecting the other. First I should note that the tax rate I used in my analysis was total tax rate, not just federal taxes which are progressive. State and local tax rates are not all progressive, and some are regressive (sales tax). This total tax rate therefore does not necessarily increase when the general economy is good.
I tried to think of some quantitative measure of a “good” or “bad” economy. I did not have to ponder too long before realizing that per capita consumer spending itself is such a measure. It is nearly the same thing as GDP/capita, because GDP is just the dollar value of all goods and services bought during the year. Although “personal consumption expenditures” does not include business spending, there is still a 0.999 correlation between consumer spending/capita and GDP/capita. Figure 2 illustrates.
FIGURE 2
Thus, the positive correlation seen in Figure 1 could have been just the effect of per capita consumer spending on tax rate (because of tax progressivity) rather than the reverse as I interpreted it. A booming economy, as measured by consumer spending, could cause a higher effective tax rate because people move into higher tax brackets, justifying Pressman’s criticism of my first calculation.
There is a way to do the linear regression analysis so that any causation can be in only one direction: simply displace one of the variables by a year. So I paired the effective tax rate value in year j with the consumer spending value in year j+1; the former can affect the latter, but the latter cannot affect the former. Indeed, there is reason to suspect that the former does affect the latter, because federal and state taxes for year j are not fully reconciled until the end of the first quarter of year j+1: taxpayers receive their refunds or make their final tax payments for year j in April of year j+1. In this analysis we would expect a significant drop in the correlation value if Pressman was right, because the economic state in year j+1 can’t possibly affect the tax rate in year j. Figure 3 shows the result.
FIGURE 3
The value of the correlation coefficient actually increases slightly: r = 0.50. Because there is no decrease in the correlation value at all, I conclude that Pressman’s interpretation of the previous calculation was wrong; it was not showing the effect of consumer spending on tax rate, and had nothing to do with any tax progressivity.
If we look at just federal income tax[3], which we know is progressive, and consumer spending we still get a positive correlation, although it is not statistically significant: r = 0.22. Figure 4 shows the linear regression. This small positive correlation could be what Pressman was thinking about: a good economy, as measured by consumer spending, causing a higher effective federal income tax rate because of tax progressivity. However he would be wrong here also, because the same procedure used above (displacing the consumer spending numbers one year forward) yields a new correlation value slightly higher: r = 0.23.
FIGURE 4
The Uncontrolled Variables
Pressman also pointed out that I had not controlled for other variables which could be affecting either or both variables in the analysis of Figure 1. I have noticed that economists often retreat to this position when shown unequivocal empirical evidence blatantly contradicting their hypotheses. It would be a good argument if I were asserting that higher taxes cause higher consumer spending. The coefficient of determination associated with Figure 1, COD = r2 = 0.22, tells us how much, at most, of the variation in consumer spending over the 70 years could be ascribed to variation in total tax rate. So if I were saying that higher taxes cause higher consumer spending, then only 0% - 22% of consumer spending increases could be so caused; 78% - 100% must be due to other factors.
But I am not making any such assertion. I am only saying that the empirical evidence contradicts the opposite hypothesis that higher taxes cause lower consumer spending. It is still possible to believe that the uncontrolled variables have conspired over 70 years of recent history to hide the “real” relationship between taxes and consumer spending. I call such a belief the “Conspiracy of the Uncontrolled Variables” theory. This theory says that if all other factors were constant, then the negative relationship between taxes and consumer spending would show up. The problem with this theory is that all other factors are never constant, as is quite clear from the scatter of points in Figures 1 and 3. If I knew what the other influences on consumer spending were, I could easily take account of them in multiple linear regression calculations[4].
One last point: anyone who clings to the Conspiracy of the Uncontrolled Variables theory in the face of Figures 1 and 3, is tacitly admitting that any contribution of higher taxes to lower consumer spending is negligible compared to the effect of other factors on the latter. That is, as a practical matter, higher taxes don’t reduce consumer spending after all.
Gary Waldman
February 2021
[1] www.bea.gov – use the tools menu to go to Interactive data. The required data is in line 13 of Table 7.1 [2] www.taxjusticenow.org - click on “More” at the upper right and choose “Technical Appendix” in the submenu, then pick “Supplementary tables underlying statistics presented in the book”. I used column B of Table A3. [3] Ref. 2 – column C only [4] I have studied one: unemployment rate. It had little to no effect on the relation between total tax rate and consumer spending.
Comments