Uses and Abuses of Pareto Optimization in Economics

Simon Wren-Lewis posts on Pareto optimization here.  The post is worth a read, as are the comments that follow.  Oddly, no one showed up in the comment thread to defend Pareto optimization in economics, though a few offered the standard criticisms of aggregate social utility functions.  However, there seems to be a few different interpretations of what it means to make Pareto efficiency the goal in economics, and what implications follow from it.  There is a widely held misconception that this road leads to Libertarianism.  On the contrary, accepting this goal as a premise leads to virtually no interesting conclusions about what we should favor in economics whatsoever.

Let's start with some definitions:

Given an initial allocation of goods among a set of individuals, a change to a different allocation that makes at least one individual better off without making any other individual worse off is called a Pareto improvement. An allocation is defined as "Pareto efficient" or "Pareto optimal" when no further Pareto improvements can be made.

This definition suggests what I will call the Weak Pareto Criterion: As a first pass when considering economic policies available to us, we should discard any that lead to outcomes that are not Pareto optimal, because there are superior options among the ones we don't throw out.  Pareto optimality allows us to narrow a large set of options down to a smaller set of options, but we still may reasonably consider making non-Pareto-improving changes that move us from one Pareto efficient outcome to another, as it is not the case that every Pareto efficient outcome is just as good as any other.  This is a very weak criterion indeed, but can easily be abused in ways I will come back to later.

The Strong Pareto Criterion

However, Simon Wren-Lewis and many of the commenters on his post are arguing against a far stronger criterion, which I will call the Strong Pareto Criterion: only consider options that lead to Pareto improvements.  This stronger criterion might just be a straw man, but it's worth knocking down just in case.  This criterion doesn't really have much to do with optimization concerns; it is better understood as a particular definition of fairness or justice, which starts with the Hippocratic "first do no harm": it is unfair to harm anyone against their will, regardless of the benefit that may be realized for others.  Simon Wren-Lewis sees in this a questionable argument for Libertarianism, but if taken seriously this principle doesn't work at all as an argument for Libertarianism: it is an argument for extreme status quo bias.

It is fun to imagine a parallel universe where economists consistently applied the strong criterion. For example, free trade policies have been enacted that have sacrificed the wellbeing of many for the greater good. Where were the economists arguing that NAFTA should not be done because no change should ever be made unless it is a Pareto improvement? Who argues that deregulation and tax cuts should never be done unless you know there is not a single person who will be made worse off?  No one makes these arguments; they decide the benefits outweigh the harms or vice versa even if the benefits and harms aren't to the same people.

Like I said, this one might be a straw man, but even if you do find someone who believes they believe in this criterion, if you look at their actual economic preferences you are likely to find they only follow their principles opportunistically.

The Libertarian Variant

To warp the strong criterion into a Libertarian criterion, you have to ignore the status quo and instead imagine starting over from scratch. You ignore the harms that will be inflicted on people by the changes you would like, and instead ask if you can find anyone who would be harmed in the parallel universe where we were considering moving from a libertarian economy to this one.  This is the Libertarian Variant of the Strong Pareto Criterion:  First do nothing that isn't a Pareto improvement over Libertarianism.  Thus it is fine to harm many if some benefit if you are making the economy more libertarian, whereas it is not fine to support the status quo if some are any worse off than they would be if the economy were more libertarian. Needless to say this kind of thinking is not value neutral at all; it holds Libertarianism up as the ultimate ideal that everything is to be compared to, and alternatives are only to be accepted if they are unarguably better in every way for everyone.  We can safely discard the Libertarian variant as question begging; anyone who's not already convinced of the superiority of Libertarianism has little reason to consider an argument which assumes that the fairness of any outcome should be measured by comparing that outcome to Libertarianism.

The Weak Pareto Criterion

The Weak Pareto Criterion is widely accepted but roughly useless.  It forms the basis of the first and second welfare theorems, which are frequently used to argue for government without economic regulation, but optionally with taxes, spending, and transfers.  However these arguments are logically questionable for reasons I will get into.  These welfare theorems are problematic partly because they make far weaker claims than they seem, but also because their existence and the elegance of their proofs make many economists long to believe that they have shown something worth showing -- that Pareto efficiency is a far more desirable and useful goal than it actually is.

Let's start with some of the (I suspect) uncontroversial shortcomings of Pareto efficiency as a standard:

• Plenty of Pareto efficient outcomes would be reasonably judged by most people to be terrible outcomes, relative to other possible outcomes.  For example, we can imagine the outcome where I own all property in the world and therefore have claim on its output, and everyone else can earn just enough to eat by working full time for me.  Because no one can be made better off without harming me, this outcome is Pareto efficient.  But you can just as easily imagine bad economies where it is, say, Mitt Romney who owns everything instead of me.
• Pareto optimization is intended for the purpose of taking a large set of possibilities and narrow it to a smaller (but probably still quite large) set of possibilities, the latter set certain to contain the most desirable possibilities but probably undesirable ones as well. Pareto efficiency is not intended for the purpose of justifying any particular outcome in the Pareto set as being a good outcome.  The moment you suggest that a particular economic arrangement that will lead to one particular Pareto optimal outcome is one we should prefer, you have necessarily brought in considerations other than Pareto optimization.
• If outcome A is Pareto efficient and outcome B is not, that itself is no reason to believe A is better than B, unless A itself is a Pareto improvement over B.  Even getting rid of regulations that are causing deadweight loss to make a market clear is likely to not be Pareto improving, as illustrated nicely by Steve Waldman.

The most insidious use of arguments from Pareto efficiency come when options are discarded due to the existence of Pareto better options that aren't actually on the table. You should never throw away option A in favor of option B if option B isn't a particularly plausible option and option A is. But this is quite common in economics, because the Pareto set in the welfare theorems is drawn from the outcomes that are theoretically achievable, whereas no consideration is given to what is practically achievable. As a result, options that are Pareto efficient in practice are thrown out because they are not Pareto efficient in theory. This is problematic.

To give an example, consider the minimum wage. It can be argued in theory that if we got rid of it, if we calculated how much each person benefited or lost from that change the winners would be able to compensate the losers such that everyone would be better off than they are with it. Therefore an outcome with a minimum wage is not part of the theoretical Pareto frontier. However, arguing on this basis for getting rid of the minimum wage without any sort of compensation scheme would be a non sequitur as simply eliminating the minimum wage would certainly not be a Pareto improvement. But a more sophisticated argument might be that we shouldn't consider a minimum wage because there are better ideas we could be considering instead. However the better ideas are unrealistic. No one has ever actually laid out a policy that could be enacted into law that would get rid of the minimum wage in a Pareto improving way; this is not practically possible to do. Furthermore, due to its popularity a minimum wage is not just practically possible to write legislation to enact, it is politically possible to build the support necessary to enact it, something other alternatives may lack. For this reason keeping the minimum wage (or raising it to $10, $15, probably even $35) should be considered part of the practical Pareto frontier even if they are not part of the theoretical Pareto frontier.  Now that doesn't mean that raising it to any of those values would be a good idea. There may be good reasons to support lowering the minimum wage or to oppose raising it by too much, but Pareto efficiency is not among those reasons.

And as for the problems with aggregate social utility and interpersonal utility comparisons?  The problems are as real as they are unavoidable.  If you have ever supported a tax cut or a tax hike, opposed a regulation you believed was protecting one industry's profit at everyone else's expense, supported breaking up a monopoly, supported lower or higher trade barriers, or raising, lowering, or eliminating the minimum wage, then you are guilty of interpersonal utility comparisons.  You have supported harming some to help others; if you had any sound reason at all for your position it necessarily relied on some conception of the greater good.  The values you relied on to make that judgement may be reasonable ones, but make them explicit so they can be part of the conversation; do not insist on a contest to see who is best at sweeping them under the rug.

Update: For more on this, including the problematic Potential Pareto criterion which I was not previously aware of, see Steve Waldman's Welfare Economics series.  In fact, read that series before you read any of the other posts I've linked to, or for that matter before you read this post.