Mirlees is the basis for a lot of how economists think about taxation (he won a Nobel Prize for this body of work in 1996). Famously, while he was a supporter of the Labour Party in the UK, and generally what we would now call progressive, he believed his own results on taxes. Namely that most forms of corporate or capital taxation were a bad idea, and that rates are best kept at zero.
Of course, that doesn’t work politically in most places.
Kocherlakota published an article in 2003 in Econometrica expanding on Mirlees work. It’s one of the basic theoretical papers in taxation.
He set up a model with the following features:
- Agents have skills that determine their income.
- Agents can save, and can finance consumption out of wealth if needed.
- Skills evolve through time with persistence, sometimes getting worse; but you’re self-insured to the extent that you saved something.
- The government is interested in providing a social safety net to the income-poor by taxing the wealth-rich, and redistributing the proceeds as income to the needy.
In this framework, what would the optimal wealth tax look like? The answer is anything but what politicians commonly propose.
- The expected tax rate that an agent expects now on the wealth they may have in the future … should be zero.
- The gross amount of total tax revenue collected on wealth … should always be zero.
- But, the actual tax rate that an agent should face on their wealth once they accumulate it … should be decreasing in their consumption.
- The dispersion of tax rates on wealth between the wealth-rich and wealth-poor should be bigger if the government is bigger.
If you told that to anyone in D.C. (or any other capital) they’d think you were nuts. It means the wealthy should be subsidized, and the not-so-wealthy should be taxed. Further, that should be accentuated when the government is bigger (as they pretty much all have done over the last century).
Economists anger people by pointing out unintended consequences. And there’s a doozy in the framework established by that first set of bullet points (which honestly seemed pretty plausible when you first read it, right?).
It turns out that the dominating effect comes from people who are wealthy now, and who suffer a negative skill shock in the future, so their income drops. Their optimal choice is to … stop working completely … collect the government’s welfare support … and keep their nest egg relatively untouched.
Does this sound like any seniors you know? Wealth rich, income poor, choosing not to work because they have relatively more aches and pains than other seniors, and receiving checks from the government?
Before you complain, let’s look at how an optimal wealth tax might differ from the one’s we have or might have. Current wealth taxes (and those proposed) simply take a percentage of wealth. What Kocherlakota is saying is that you can and should take a fraction of wealth, provided that you adjust that for current income; you take more from someone with low income because they might be a slacker, and you take less from those with high income because you know they’re probably supporting the government more with the high income taxes they pay.
Do note that we already do something like this in certain situations. For example, many welfare programs have asset tests. No one likes this, but in practice this is what is happening when an old person with a home needs to go into a nursing home, and the government requires that the home be sold and the proceeds used up before they’ll start covering the cost of the nursing home. In short, if you self-insured, the wealth tax rate should function to make sure that you use your self-insurance when it’s needed.
The paper is not required. It appeared in Econometrica, which is one of the toughest journals. And the author, Narayana Kocherlakota, has been a top macro theorist at theory-heavy schools for most of his career. By all means check it out, but don’t make yourself feel bad just because you’d need years of math and theory classes to get it all (when all boiled down, the model is 58 equations and 58 unknowns).