From an email:
The eye-catching result here is they have consumption taxes being *sharply* regressive, e.g. 12% for the lowest income group. I’m not aware of any US state that has state + average local sales rates tax that high. And lots of goods are exempt from sales tax. So how do they get this? Well, suppose someone earns $1k in labor earnings and gets $9k in transfers, and consumes it all paying a 5% sales tax = $500 in tax. What sales tax rate have they paid (as a % of their income)? The method Treasury uses says 500/(1k+9k) = 5% (this is also what Auten-Splinter do). Saez-Zucman exclude transfers from the denominator, and thus say 500/1k = 50%. This is a matter of definition, so it’s hard to call it right or wrong, but it does seem misleading and yield some rather nonsensical implications. For example, it means that if welfare to the poor is increased, this will be measured as an increased tax rate.
Indeed, Saez-Zucman themselves seem to realise that this definition yields extreme numbers at the very bottom, where consumption tax rates can easily exceed >100%. In their appendix – https://eml.berkeley.edu/~saez/SZ2019Appendix.pdf – they note “People with very low pre-tax income (below half the federal minimum wage) earn transfer income (temporary assistance, SNAP, supplemental security income, veteran benefits, etc.), which is not part of pre-tax income. They pay sales taxes on that transfer income when it is consumed. As a result, they have high (sometimes very high) tax rates as a fraction of their pre-tax income. We avoid that problem by restricting the population to adults with more than half the minimum in pre-tax income.“
This is quite remarkable. If the sensible way of defining tax rates involves excluding transfers from the denominator (as they claim), the fact that it leads to very high rates by construction at the bottom should be because this is a sensible summary of reality. Yet, in their own words, it’s a problem. Rather than switching method, they drop the people at the very bottom which conveniently covers up the problem (but leaves a less severe version of the problem in their remaining lower income sample). Of course, they could have just used the standard definition which includes transfers in the denominator, but doing this destroys the entire headline result.
It also seems noteworthy that in choosing to excluding transfers, they nonetheless retain payroll taxes. It seems pretty egregious to call payroll taxes regressive when social security is implicitly an insurance scheme with a very large degree of aggregate progressivity, but this is a minor point by comparison.