In the past year, more than 1 million migrants have entered Europe, many fleeing violence in Syria. The continent has been swamped with asylum claims: In 2015, the European Union received 1,321,560 claims, of which more than 476,000 were for Germany alone. This has been a humanitarian and logistical nightmare. Under the European Union’s current asylum system, a small number of countries — Greece, Hungary and Italy — do most of the work to register and process asylum seekers. Refugees have little say over where they end up, and many states and local communities feel they are being asked to take in more than they are able.
Two Oxford academics, Will Jones and Alexander Teytelboym, believe they have found a way to inject more choice into the process: using a system known as “matching markets,” whereby an algorithm would attempt to match a refugee’s preference with that of a particular country or city. (Similar algorithms are used in the United States to assign medical residents to hospitals.) After being accepted into Europe, refugees would be asked to create a list of places they would like to live; at the same time, states and local communities would be asked which refugees they most want to host, which might depend on particular skill gaps they need to fill or particular capacities, such as at schools or hospitals, that they are able to handle. On a recent afternoon, Jones and Teytelboym discussed how matching markets operate, what such a market for refugees would look like, and what makes a given algorithm fail or succeed.
This interview has been edited and condensed.
You’ve advocated for the use of matching markets to control the flow of refugees into Europe. What, exactly, is a matching market?
AT: Matching markets are markets in which all the parties have to agree to the transaction. If you go to the supermarket, and you buy an apple, the apple doesn’t need to consent to that transaction. But in a matching market, you typically have two parties, and both of the parties have to consent to the particular match. In the refugee crisis, the two parties are refugees and local authorities. A refugee would have to agree to go to a particular local authority, and that local authority should want to take the refugee or the refugee family.
Matching markets tend to be applied to things like where kids go to school, in Britain and in the United States, and in how residents are matched to their hospitals: They express preferences for hospitals, and hospitals express preferences for them. The reason to create a matching market is to aggregate all the information in one system and ensure that the outcome is one in which all of the parties are, to some extent, satisfied. There are no easy gains, and the system is stable enough that no one wants to go out and re-match.
What about the refugee crisis makes it a good candidate for matching markets?
WJ: Currently, some states and local authorities are exceptionally willing and able to host refugees, and refugees are in need of protection. Right now, we often focus on the headline number: how many refugees need to be placed somewhere, and then we randomly allocate refugees to a particular place. But refugees are as diverse as any other set of human beings. They have diverse preferences about how they wish to live, what they want to do, what skills they can bring, how they want to pursue their conception of the good. And not all communities can host all refugees equally effectively. Refugee-hosting communities have information about who they could be in the best position to host, but right now, that information doesn’t get shared.
You’ve written that matching markets would be more efficient and humane than the current system
WJ: The short reason that it’s more efficient is that it’s more efficient to actually ask people what they want, rather than to work out, through a long, bureaucratic process, what’s best for them — to let their preferences, rather than our guesses about their preferences, be decisive. The reason why it’s more humane is that this is a very effective system for centralizing, collecting and collating information about what protection capacities a particular region has. Not all refugees have the same protection needs, and not all areas are equally able to provide protection for refugees. A local community that can protect a particular kind of refugee, because it has a particular health-care capacity or educational capacity, will be able to protect those particular refugees.
AT: One definition of efficiency would be that the preferences of neither side can be improved without hurting someone else. But for that, you need to know refugees’ preferences and the capacities of local authorities. Once you have those two sides, the algorithms and the systems that matching theorists build can try to satisfy some version of a best outcome.
So you need some kind of a central body to collect all the information?
WJ: You need some kind of a central body to administer it, but one of the things that makes this better than conventional refugee resettlement efforts is that you don’t need a centralized bureaucracy to go out and collect all the information. What happens now is that a government employee will go to a refugee settlement and conduct an interview, and on the basis of that interview, they will try to work out what a refugee wants. Then they will have to collect and process information from local authorities and match all the preferences up manually in a bespoke manner. The matching markets system requires a centralized algorithm, but it doesn’t require centralized processing. It allows refugees and local authorities to speak to each other directly.
AT: What’s important is that the information is standardized. So we would collect bits of information that would be the same for each family, in terms of their needs and preferences, and we would collect standardized information from local authorities.
When would this information be collected, and in what form?
WJ: Let’s imagine a scenario in which you are a refugee at a camp in Turkey. You’ve been identified as a candidate for resettlement. There is often an extremely long waiting period, let’s say, optimistically, six weeks, before we are able to put you on a plane to somewhere in Europe. That is a very long period in which we could say to you: We are going to resettle you, that plane is booked, and it is taking off in six weeks. In the meantime, here is a sheet of paper, please indicate your preferences as to where you wish to go.
It seems fairly intuitive why refugees would want to have more choice in this process. What incentive do host states or communities have to participate?
WJ: The short answer is that it gives the hosting state more control. Britain, for example, wants to resettle refugees, but it doesn’t want to resettle them in places that will end up being a huge drain on the resources of the state because they don’t integrate properly or because they don’t get jobs. The state and the refugees have the same interests here. The state doesn’t want to impose an undue burden on areas that have to deal with refugees who are mismatched to their capacities. Right now, there is a lot of humanitarian goodwill in various local government associations in Britain. What the government doesn’t want to do is assign refugees to them that they can’t deal with. The British government will end up spending less money, in the immediate term, putting together refugees and communities willing to host them, and in the long term, it will end up hosting refugees that are more likely to integrate successfully.
AT: We have a very good empirical case study of random refugee allocation in Sweden, where they found that putting refugees in more deprived areas caused them to be worse off in the long run. The refugees were distributed randomly with the best intent — to disperse the refugees, to make sure they integrated into Swedish society as much as possible. But because they were allocated randomly, it created great inequality in outcomes, and refugees that ended up in deprived areas were, in the end, much worse off.
In Europe, what would be the central body to administer the program?
WJ: If you were doing it across the European Union, it would be the European Asylum Office. If you were doing it within a particular country, it would be something like the British Home Office.
So a national government could set up a matching market on its own?
WJ: You could absolutely do it within one state. Matching markets have huge advantages in working out whether a certain refugee should go to Sweden or to Portugal, but also whether they should be in London or in Middlesbrough.
AT: For countries that have very secure borders that they can easily control — the United States, Canada, Britain, Australia — running a matching system like this is a complete no-brainer.
Why aren’t more countries setting up matching markets then?
AT: Even in the school system, matching markets have been surprisingly slow to take up. There are good algorithms, and there are bad algorithms, so matching theorists have pushed a lot within the school-choice context to use particular algorithms that are better than others. But that’s a very simple context. The refugee context is actually a much more difficult algorithmic problem to solve. You’re not just filling up one school place with one child. You have a family, the family has diverse needs, and the needs occupy capacities across a variety of services, and a local authority may have very different capacities for these services. So the problem itself is rather more challenging. The theory would be very new for this. But the principles we have for running matching systems would still apply.
Is there any concern that refugees would all express the same preferences, that there wouldn’t be enough diversity of choice?
AT: This is the question we get every time. Let’s say every refugee wants to go to Germany. Germany can’t take all of them, obviously, and so inevitably some will be turned down. What will happen then is that we’d try to satisfy the refugee’s second choice. Now this is exactly why you may want to ask refugees about where they want to go. Even if you guess that every refugee wants to go to Germany, and you might be right, you would really struggle to name their second choice. At the moment, 90 percent of the refugees coming into Europe want to go to Germany, and about 10 percent want to go to Sweden — this is their stated preference in Greece, when they come in. But once we go down to second, third and fourth preferences, which could, in principle, be the only ones that are satisfied, we really have no idea, which is why we should just ask them.
Even within a particular country, or across countries, it is often quite straightforward to work out someone’s first choice. But you can’t always satisfy that. For example, in Britain, a lot of refugees may want to go to London, Manchester, Birmingham — bigger cities where there are already bigger communities. And these communities will naturally have greater capacities to host them. But some refugees will inevitably not be able to end up there and will have to go elsewhere, so we’d want to try to satisfy, as much as possible, their second, third, fourth choices. There’s no way to figure out where a refugee family would want to go — it’s a decision that the family would discuss for several days. That’s why you ask for a preference list, not just of their first choice, but of their top dozen or so localities out of the ones where their needs could definitely be met. If you have a medical need, for example, you would only have to list choices that are near hospitals.
Have policymakers been interested in matching markets?
AT: Yes. We didn’t expect it to take off in the way that it has. Perhaps because of the deal with Turkey, people are now asking whether this could be run on a European level. There will be some refugees who are settled outside of Turkey, and in principle, we could ask them where in Europe they want to go.
In Britain, the idea is getting a lot of traction. The British government is very interested, and Britain offers a very nice context to do refugee matches in. It’s probably the nicest context right now in the world to apply the system. The numbers are not dramatic. Britain has control of the border, and they’re taking 20,000 refugees over the course of the Parliament, and so this is an easily manageable number that they will be able to deal with. We’ve talked to lawyers in Canada and the United States about it, too. There’s interest everywhere simply because it’s a system that ensures that no matter how many refugees you take, they end up in places where they will be well hosted and well protected.
Matching markets haven’t always been successful. When do they fail?
AT: One very famous example is in the case of school choice. The way that the school-choice system used to be run in Boston, prior to 2003, seemed very sensible. There was a centralized system, the preferences of the parents would be collected, and then there would be a matching algorithm that would allocate kids to schools. But the way the algorithm worked, it allowed parents to game the system, and over time, parents realized this. The parents who learned they could game it, and who learned how, tended to be richer. Parents would get together in groups, and they would strategize about how they should alter their actual preferences in order to get their kids into good schools. This was a centralized matching algorithm — it was something that should have worked well, but it was a poor algorithm, it had poor properties. The system is now known as the “Boston mechanism.” When economists got wind of what was happening and went to work with the Boston school board, they changed the algorithm. The algorithm now does not allow you to game it. You have no incentive to state any preferences except for your true preferences. Then, the algorithm was applied in New York, and it is still being used there and elsewhere. So small design changes can make a huge difference to the outcome. Our jobs as matching theorists is to design the system in a way that ensures that it works well, that it’s a good one.
WJ: It’s worth pointing out that the system can be run indicatively. You could run the system, see what it would suggest, and if you don’t like the results it has yielded, you could ignore it.
AT: In the case of the residency match, when a doctor gets matched to a hospital, it’s a contract. You’re tied to that outcome. Same with schools. It’s very hard for a school board to renege on a particular outcome. In the case of refugees, however, there may be other considerations — sudden changes in the health-care capacities of a given area, for example. Let’s say that a given area was in a position to offer health care to children with PTSD, which is a very common need, but now they no longer are in a position to offer that. You can have sudden changes that you may not have built into the system.
Now remember, this has a trade-off. If you tamper with a system, the rules of which you have made clear to the refugees, then refugees will not tell you the truth, and you’re not going to run a very good matching market. It’s better not to tamper. But of course, if you have to improve the algorithm over time, you can do so. No one ever gets a matching market right the first time around. But after several iterations, you can have a system that works well.
Nikita Lalwani is a staff editor at Foreign Affairs.
Sam Winter-Levy is an assistant editor at Foreign Affairs.