The Distributional Consequences of Tariffs
A surprising null
Tariffs are generally agreed upon by the economics profession to be bad. Not only do they distort the flow of goods across borders, they also impair the transmission of ideas, make war less costly, and reduce the total factor productivity of firms. While there is room for large countries to extract excess profits by imposing a tax, the actual policies imposed diverge wildly from what is optimal, and serve mainly to protect some domestic industries over others. The individually optimal tariff is also worse for the world as a whole compared to no taxes on commerce. Finally, it seems deeply wrong for the United States to eke out a few billion in consumer surplus at the cost of much larger losses in poor and developing countries. If you believe that marginal utility is decreasing as a function of wealth, then you would hardly want to transfer from people at the bottom (where money increases utility dramatically) those at the top, who gain less and less for every dollar of income.
This concern about distribution also applies to the effects of tariffs within the United States. If tariffs redistribute from poor to rich, they could be bad even if it were beneficial to the GDP of the country as a whole. The literature is conflicted on the distributional effects of tariffs, however. The traditional view is that tariffs are regressive. The poor consume a greater part of their income in the form of goods, which are also more likely to be traded goods. The rich spend considerably more of their income on services, which are by definition not taxed. (You can argue that immigration restrictions are a de facto tax on services, but we care about the marginal changes here, not the system as a whole.)
There are two sides to this. First, we need to figure out how consumption differs by income. Here it’s pretty clear that lower-income people consume more goods which are traded, and the goods which they consume have a lower elasticity of substitution (Fajgelbaum and Khandelwal, 2016). Put concretely, this is because the poor consume fewer luxuries which can’t be exchanged for other goods. In practice, the distributional effects in the United States are particularly large, because even within relatively narrow categories of goods lower quality goods face higher tax rates. (Acosta and Cox (2025) attribute this to a peculiarity of trade negotiations. The within category rates were shaped by negotiations in the 1930s, after which we no longer negotiated individual items but instead shifted all items within a category by a fixed percentage).
However, we need to take into account the effect of who gets more jobs. If it reallocates production to low-skill industries primarily employing the poor, then it may redistribute from top to bottom. Borusyak and Jaravel (2023) compare the two, and show that almost all of the redistribution is occurring within income deciles. Tariffs are costly to the consumer, and are indeed disproportionately costly to lower income consumers, but it does so by reallocating jobs to primarily lower income workers. Taking into account both effects, the distributional consequences of tariffs and trade are approximately nil.
They are able to do this exercise for the United States only, as it requires extremely detailed information on input-output tables. I think I should spend a bit of time discussing what these are, as recent trade work has begun returning to using them. Let’s suppose you’re interested in the effect of trade on the economy as a whole. If you simply look at the changes in more or less affected sectors, you will only be able to say something about the difference between sectors, without any knowledge of whether things got better or worse on the whole. (I discuss this at length in my article on “The China Shock”). An input-output table is a giant matrix, or system of equations, which contains all of the inputs and their ratios, and all of the outputs of an industry.
I am not entirely sure that I believe them! I-O tables can only include the inputs, but we cannot infer how much production will vary. The trouble is that price and quantity are jointly determined by supply and demand. If an input gets cheaper, we can’t say anything about how quantity will change without knowledge of demand. We can estimate this if we have plausibly exogenous shocks to use as an instrumental variable, and there is a whole line of literature (see BLP 1995, Nevo 2001) doing this. However, this is incredibly data intensive, and obviously cannot be done for every combination of industry.
So, this is the best we’ve got. I would simply urge you to have uncertainty around anything to do with I-O tables. If you are interested in using them, the BEA maintains them here.
I do not think that Borusyak and Jaravel fully obviates distributional concerns with regard to the present administration’s tariffs. There will still be redistribution away from non-workers to workers even within decile. We should expect this to penalize having children, who are notably non-working individuals. It will also follow from curved utility that an equal and negative monetary shock to different deciles will have different effects on utility. Finally, changes in employment are a long run change, while changes in consumption are immediate. I do not expect them to last indefinitely, and neither does the stock market. People won’t make the investments needed to change where production occurs unless they believe they’re here to stay.