The sideways view

I’d prefer it if states took from the rich to give to the poor. But I wish they did it in a reasonably efficient way, such that poor people get about $1 of utility for every $1 of utility a rich person loses. I think we often fall short of that goal.

(An expansion of some points from my prior policy post.)

In this post I’ll make four claims:

1. Redistribution should mostly take the form of explicit transfers given to poor people, financed by tax revenue. Other policies should treat a $1 benefit to a rich person the same as a $1 benefit to a poor person.
2. The distinction between benefit phase-out and income tax is not economically meaningful. For conceptual clarity we should think about redistribution as a payment that doesn’t depend on income, together with the effective income tax schedule.
3. Transfers should mostly be cash payments that depend on age and health/disability, rather than depending on any choices made by recipients. The most plausible exceptions are in-kind healthcare for the sick, and in-kind transfers to parents to care for their children.
4. As an important special case, we should avoid work requirements and large work incentives.

1. Redistribute with transfers

First claim: for the purpose of most policy decisions, I think we should tally up the costs and benefits to different people, measured in dollars, and adopt the policy if the sum of the benefits exceeds the sum of the costs. We should then use taxes and transfer payments to move money from the rich to the poor.

Consider some policy P (like an increase in the minimum wage). We can evaluate P by thinking about how much it helps or hurts each person. If the sum of these effects is positive, then the following set of changes is a Pareto improvement:

• Implement P
• Tax the “winners” from P
• Send money to the “losers” from P
• Take the extra revenue, and distribute it evenly

That is to say: regardless of what you value, your best strategy always involves implementing policy P.

(ETA: this isn’t true, either for these “positive-sum” trades nor for Pareto improvements, if the best strategy involves making other changes such that P no longer passes this cost-benefit test—which could just be e.g. moving money from rich people in a way that decreases their willingness to pay. H/T Toby Ord for raising this caveat.)

Objection 1: taxing the “winners” is complicated

It’s hard to determine who gained or lost value from the policy P. Fortunately we don’t really have to do it.

At the end of the day, the purpose of taxes and payments is to make the world better, not to ensure that every particular change was a Pareto improvement. For that purpose, all we care about is the marginal value of a dollar in each person’s pocket (according to whatever collective preferences our government is attempting to satisfy). If we think it’s preferable for Alice to have $1 than for Bob to have $1, we can raise taxes on Bob and give the revenue to Alice—no detailed reasoning about winners and losers of past policy changes is is required.

Of course this is exactly the same problem that our tax code already needed to solve. So we should expect passing efficient policies to be good, without inherently making the tax problem any more complicated than it already was.

Objection 2: taxes distort behavior

Raising taxes on the rich to give money to the poor isn’t “free”—it reduces the incentives for people to become rich, which in turn distorts behavior and reduces efficiency.

If this effect were very strong, you might think that it could be idea to adopt inefficient policies that benefit the poor, rather than adopting efficient policies + taxing.

But on closer inspection this effect seems to be a mirage. Any policy which has the net effect of moving value from rich people to poor people will reduce the incentives to become richer—in fact by exactly the same amount as a tax.

(I think that efficient policies + taxes would probably win anyway, at least for current income tax rates in the US, because the disincentive effects from higher taxes probably don’t actually cause much efficiency loss compared to efficiency losses from other redistributive policies. But that’s a more complicated empirical question.)

For example, suppose I pass a minimum wage, and hope that it will have the positive effect of moving money from rich capitalists to poorer laborers. This reduces the incentive to be an investor, and (if investors are typically rich) therefore the incentive to be rich. The total size of the disincentive is the same as the amount of money that it takes from rich people. If the minimum wage isn’t economically efficient, then this is larger than the disincentive from simply raising income taxes and handing the money to poor people. (In the case of the minimum wage, my view is that marginal increases are maybe ~2x as socially costly as taxes that achieve the same amount of redistribution.)

One subtlety: a complete evaluation of economic efficiency might already consider these disincentive effects. So we need to be careful: when evaluating the economic efficiency of a policy, we shouldn’t consider the negative incentive effects from making life worse for the rich, because those will typically be offset by tax rate changes.

A second subtlety: income taxes really incentivize minimizing reported income, which might be a different incentive than reducing actual income. If that gap resulted in large efficiency losses, then we could value policies that effectively implemented a harder-to-evade tax. And of course, “has high income” is just a proxy for “has low marginal value of dollars” and any tax that more effectively targeted what we really care about could be better than an income tax. However, I think this is quite rarely the case for the policies people actually consider (and usually there isn’t even a story on offer about how it could be the case).

Objection 3: sometimes you have to settle for second best

In practice, I often hear the following:

In an ideal world we’d raise taxes rather than using less efficient redistributive policies like a minimum wage. But in practice, a minimum wage is more popular than raising taxes, and we need to take what we can get.

As a consequentialist, I’m open to this kind of argument. If a minimum wage is all you can get, and a minimum wage would be good on balance, then you should take a minimum wage.

But even in that world, I think it’s pretty important for “people who should know better” to keep track of the distinction, and to remember that the second best is still only second best.

It’s fine if we want to advocate for an inefficient policy because we think that’s the only way we can score a political win. But somewhere we should be having the discussion about what policies are actually good, since that’s an important part of ending up with good policy. This objection has no place in that discussion.

More rarely, people go further and claim that honest discussion about the inefficiency of a minimum wage is already a bad idea (because it produces political ammunition for our political opponents). For me, this is going way too far towards burning the intellectual commons in order to score a political victory.

2. Unconditional transfers + income taxes

Second claim: there is no economically meaningful distinction between phasing out benefits and charging an income tax. When considering a policy basket, we should think about it as a payment that doesn’t depend on income, plus the effective income tax schedule.

Consequences of this view:

• Objections to a UBI like “Bill Gates would get 13K, which is crazy” don’t really make sense—raising taxes is costly, but benefit phase-out is exactly equally costly. In general the economic distinction between UBI and means-tested programs is not nearly as large as it appears.
• One of the most important arguments about a UBI vs. status quo is whether we should have high effective tax rates on low-earning families, and whether we should have higher tax rates on recipients of disability or childcare benefits. This question is very rarely explicitly discussed.
• If we really endorse the status quo, with reasonably high effective tax rates on low-income families, we should probably change the tax code in other ways as well (probably decreasing the gap between top brackets and lower brackets in general).
• When evaluated through this lens, the current effective tax rate is a little bit crazy, since it is the sum of a bunch of contributing factors that weren’t carefully designed.
• We could probably have a simpler set of redistributive policies, all mediated through changes to the tax code and a very small number of transfer programs.

The argument for this view isn’t very complicated:

• Holding fixed a particular taxpayer’s other features, let T(x) be the amount of money they pay to the government if they earn x of income (it can be negative, if they would receive money from the government).
• A taxpayer doesn’t care much about whether they receive money as a tax reduction or as a transfer payment. They mostly care about the total net payment between them and the government, i.e. T.
• Any given T can be implemented as a lump sum payment of T(0), followed by an income tax with marginal rate dT/dx.
• Thinking about a single policy lever is generally clearer than thinking about a giant basket of levers, so if you have lossless way to transform a giant basket into a single consideration, you should do that.

3. Use cash transfers depending only on age and health

Third claim: transfers should be cash payments, whose size depends only on age and health, with the exceptions of some medical care (which helps target aid to less healthy people) and some in-kind benefits to children (whose parents might not spend the money in the children’s interests).

(The dependence of transfer on income is discussed in the last section. I’m going to separate out “work requirements” into section 3b because that seems like a huge can of worms.)

When I hand people money, I change their incentives. The resulting changes in behavior tend to reduce efficiency.

For example, if I pay people the difference between their rent and 30% of their income, I incentivize them to pay more in rent. They are still better off than they would have been, but they’d be happier if I’d handed them the same amount of money and let them choose how to spend it (they’d end up spending less on rent).

These incentive effects don’t really exist for payments that depend on age (since my age doesn’t depend on my behavior), and are smaller for payments that depend on health/disability (since much of the variation in health is exogenous). But for other properties, there are significant losses.

You might think that transfers based on other properties could still be a good idea on balance, for several reasons:

1. We want to give money to those who need it the most. We can’t directly observe that, so we need to use proxies. The extra welfare gains from better-targeted redistribution may offset the efficiency losses from bad incentives.
2. Many behaviors have positive externalities, and so we may want to pay people money to help internalize those externalities.
3. People may make bad choices by default, and so incentivizing or forcing them to spend money in a certain way (or change their behavior) might make them better off.

Overall, I find justifications 1-3 quite weak compared to the efficiency losses. Justification #4 seems plausible and would sympathize with child support credits taking the form of vouchers for housing, food, and education.

Justification 1: Better targeting redistribution

Examples of this justification:

• People who would spend marginal dollars on food are more likely to be really need the money.
• People who lose work through no fault of their own are more likely to be falling on particularly hard times.
• People living in more expensive areas need more money to reach a fixed quality of life.

Any of these objections could fly in principle, but in almost all cases they just don’t seem to hold up (and usually no attempt is made to balance the costs and benefits). In the three cases given above:

• There isn’t that much exogenous variation in how much food people need. So most of the variation in food spending is driven by people’s choices rather than bad luck that they need to be insured against. So targeting payments to people who spend more on food is probably a mistake.
• Unemployment insurance straightforwardly decreases the incentives to find new work. The recently unemployed will have lower income, which certainly increases our estimate for how much they’d value marginal dollars, and beyond that I think those who have recently been working are more likely to really need money, but the quantitative size of the effect seems small relative to the obvious and large incentive effect.
• In the short-term living in an expensive area is more like bad luck, but over the long term it seems to be pretty elastic to the costs and benefits (and most of the initial luck in where you were born has plausibly washed out / was itself the product of similar incentives). Incentivizing more people to congregate in expensive areas is a particularly troubling and risky distortion. So absent some real argument about this I’d be super careful about messing up these incentives.

There is one family of justifications like this that I do like: unhealthy and disabled folk can easily have much higher values for money. And some kinds of health/disability status are not too sensitive to incentive effects. So I’m pretty excited about redistributive programs that cover a substantial fraction of the plausibly “bad luck” health conditions. (With higher fractions, approaching 1, for those that are least sensitive to incentives.) Note that health care spending seems extremely sensitive to incentives though, and I think that it’s easy to end up spending too much on healthcare if you use “healthcare spending” as a proxy for “health.”

I think my ideal policy would be something like: pay sick people based on the average spending of people with similar features. Prefer to pay people in terms of vouchers for relevant healthcare (since those are less valuable for people who aren’t sick, reducing the incentives to misrepresent health status) but don’t allow vouchers to ever cover more than ~50% or so of the total cost for procedures that people might not be willing to pay for out of pocket (since that makes spending decisions highly inefficient).

Justification 2: positive externalities

Some activities have externalities. Just as it’s great to raise money by taxing activities with negative externalities, it can be good to distribute money while creating incentives to do good things.

The positive externalities case is rarer, since positive externalities will be internalized if there is any way to do so, in contrast with negative externalities which will only be internalized if there is no way to avoid it.

More seriously, whatever redistributive effect you achieve via incentive payments is going to have some implicit income tax component, subject to the same discussion as in section 1 above. The real upshot is the same as before: when evaluating a subsidy policy for efficiency, we can neglect the inefficiencies from the implicit income tax. This is usually small compared to other efficiency considerations though.

This suggests that we can consider the distributional effects of subsidies separately from the efficiency effects (and that transfers for distributional purposes can mostly remain unconditional cash transfers).

Setting aside this theoretical argument, I think the arguments given for positive externalities are often quite weak relative to the relevant consequentialist bar. As an extreme example, I think there is a plausible case that higher education has positive externalities, but the magnitude of these effects seems very unlikely to be large enough to justify free college education as an efficient policy (and I think the case for small or negative externalities is also reasonably plausible).

Justification 3: people don’t know what’s good for them

Voters feel more comfortable giving poor people food and housing than giving them cash—not because they think that this corrects externalities or provides better incentives, but because they think that people would spend cash in ways that are bad for themselves.

I’m open to this argument in principle, but my current feeling is that it’s usually kind of weak (and that people’s intuitions about this justification don’t track its actual strength). Some relevant considerations:

• The actual decisions about how people “should be” spending money are usually quite divorced from the detailed facts on the ground about their situation.
• Losses from earmarking are often quite large, just from administrative overheads, accidental restrictions in the way people can spend their money, etc., beyond the inherent distortion of trying to affect how people spend their money.
• If you give someone $X for food (or rent, or whatever), you will either be in the regime where they were already spending $X—in which case your restriction has no impact—or in the regime where they were spending less than $X—in which case you are really distorting behavior. So the question is: amongst cases where someone was spending less than $X on food (or rent, or whatever), in what fraction are people making a mistake, and how large are the costs of mistakes vs. the savings in cases where they were rationally paying less than $X? Ultimately this is an empirical question, and I have only one or two experiences with people who’ve been on the “rationally paying less than $X” side of the divide. But framed in this way it seems quite likely that the costs exceed the benefits.

I do accept this argument in the case of children: that is, I think it makes sense to give money to children’s parents rather than directly to them, and once we are giving money to children’s parents I think it makes sense to try to tie their parents’ hands by restricting what they spend it on.

4. No work requirement

Fourth claim: we shouldn’t require welfare recipients to work. It might make sense to have slightly negative but probably not large negative income taxes (e.g. the 40% EITC phase-in is probably too much).

The argument against work requirements has two parts. First, that work requirements aren’t efficient:

• People should work if and only if the benefit they provide to others by working is larger than the cost they incur by working.
• The benefit you provide by working is similar to your wage.
• So people should work only if they consider working a good deal at market wages.

Second, that work requirements have particular large negative effects:

• Work requirements empower formal employers and disempower low-wage workers. They make low-wage workers dependent on their employers in a way that we should expect to make conditions worse and lead to unnecessary suffering.
• Work requirements mean that transfers aren’t reaching the absolute poorest people / aren’t reaching people at the times in their lives where they are hardest on their luck and most need money.
• In order to avoid the worst failures, work requirements probably need to make some allowances for “trying to work” or establishing that you are unable to work. But if such checks are strong enough to be a meaningful protection, they are also going to introduce their own distortions and administrative overhead. (In fact even demonstrating that you are working is a serious problem in practice.)

Given that I think work requirements have no real efficiency justification and seem to have significant costs, I tentatively think we should not impose them. The same argument goes for negative income taxes.

Objection 1: transfers reduce the incentive to work, and we need to explicitly offset that effect

Very poor people really want dollars, and so they really need to work even if conditions are bad or wages are low. Unconditional transfer payments might give them enough money to survive, at which point they may drop out of the labor force. Coupling such payments with work requirements/incentives could offset the effect.

My main problem with this argument is that it’s not really an argument. There is no reason that “offsetting” this negative effect is a good idea.  There’s not some magical economic state of nature where the incentives are “natural,” with redistribution moving us from the natural state.

From an economist’s perspective, the important thing is whether the value I create by working exceeds the cost to me of working, and I get a Pareto efficient outcome regardless of whether my starting point is complete desperation or a comfortable income to live on.

Objection: people who most need money will be working

In general we want to give money to those who need it most. It’s reasonable to expect the people who need money the most will be most willing to make sacrifices to get it. Working is a relatively “easy” sacrifice, so if someone isn’t making that sacrifice it’s evidence that they don’t need money as much.

I think this is the best argument in favor of work requirements / incentives. I still don’t find it convincing:

• People who can’t work, for reasons outside of their control, probably need money even more than those who are working. Demonstrating inability to work conclusively to a bureaucracy seems pretty challenging.
• On the flipside, working is only a valuable measure of “needs money more” (rather than “has better opportunities”) if there is a lot of hard-to-observe exogenous variation in how much people need money. But adjustments for age, health and dependents seem like they cover a lot of that, and the residual hard-to-observe variation seems smaller than the hard-to-observe variation in people’s opportunities to work and preferences about work.
• Even if this were a valid test, the improved targeting would need to be compared to the negative effects of work requirements, and quantitatively I don’t think the argument looks good.

Ultimately this is a quantitative empirical question. In this section I’ve described my view but not argued for it.

Objection: work has lots of positive externalities

If employee’s salaries typically understate the good they do, then we’d want to incentivize people to work more.

I think this is true, but only to a small extent. Some important observations:

• Most of the value that society produces gets distributed as rent and wages.
• As a result rents and wages reflect the average value of inputs (on average).
• The marginal value of inputs can be larger than the average value, if there are global returns to scale.
• Global economies of scale seem to be real but small—if I had to guess I might say that increasing global labor+capital by 1% increases total output by 1.5% over the long run, owing mostly to innovation / information goods and to a lesser extent thick market externalities.
• I think the lowest-wage workers, for whom work requirements/incentives are most likely to have an impact, create systematically lower positive externalities since they contribute less to information goods.
• Adding additional labor increases the returns to capital. But this is a “pecuniary” externality, mostly compensated by lower wages for other workers. On balance it has no impact except insofar as there are increasing returns to scale overall.

Overall I think this suggests that the “efficient” tax rate might be slightly negative—in the same way that a 10% income tax might destroy value by encouraging work, a 0% subsidy might be leaving some value on the table. Put another way, the 0% tax rate is not a distinguished efficient level of taxes, the actual efficient level maybe more like -30%.

Of course this argument is relevant across the board, and generally means that taxes are worse than they look. But it is most important for people who might have been at a 0% tax rate.

Objection: people underestimate the value of working

We might think that much or most of the value of working comes from building skills, a positive reputation, a network, etc., and that people on the borders of the workforce systematically undervalue these longer-term effects. If so, then work incentives or requirements could make their lives better.

I have the same kinds of objections to this line of argument as to paternalistic redistribution. (I think the argument for paternalism about work requirements is a little bit stronger, because it’s more likely that people are messing up by undervaluing work experience than that they are messing up by not spending enough on housing, but not much stronger).

On the flip side, I think signaling effects will probably lead to too much working (e.g. job applicants look particularly bad when they have gaps in their resume, in a way that’s basically a zero-sum competition amongst job seekers rather than a real reflection of social costs from such gaps; and someone who stays at home and plays games is going to be considered a loser in a way that may be a legitimate judgment but still results in positive externalities).

My best guess overall is that people undervalue working but only by a modest factor, and I think it’s plausible that people work too much rather than too little (ignoring college students). As the economy becomes more “efficient” I expect the size of the non-monetized skill-building to get smaller over time, and so this effect to shrink: Amazon doesn’t want its workers to be partially compensated in the form of skills that those workers won’t appreciate—they’ll try to pass those costs on to workers, or market the job as skill-building, or just optimize the job without regard to useful skill-building.

Moreover, I think the kind of work that someone ends up doing because they are forced to continuously do work (or else lose their transfer payments) is not going to be well-optimized for skill-building. I think that arguments about the beneficial effects of a UBI along these lines are greatly overstated, but I do think it’s a real effect that work requirements can make it harder to fix important problems in your life or develop the skills that would actually pay longer-term dividends.

Conclusion

I think these four arguments are reasonably solid:

1. We should prefer cash payments from rich people to poor people, compared to other policies motivated by distributional benefits.
2. Benefit phase-out is economically equivalent to taxes, and we should design the effective income tax schedule mindfully.
3. Payments should be cash and depend only on age/health. Possible exceptions include health-care (to better target spending at the sick) and services to benefit children (since their parents may underinvest).
4. As a special case, we probably shouldn’t have work requirements or large work incentives.

I think the easiest way to talk about the resulting policy looks basically like a UBI, potentially with subsidized healthcare, with payments depending on age, and with parents potentially receiving restricted transfers to spend on their children.

Overall I like the simplicity of thinking about a UBI, and I could imagine moves in this direction decreasing the complexity of the current US safety net. I also think that some features of the debate are healthy and useful, particularly a more serious discussion about work requirements. But I think talking about a UBI can just as easily obscure the actually important questions for redistributive policy. From my perspective, these key questions are:

• What effective income tax schedule should we adopt, especially at the lower end of the income distribution?
• How should transfers depend on health? Should we provide healthcare?
• How should transfers depend on age?
• How should we support children who are too young to make spending decisions for themselves?