According to Andersson (2017), there are two distinct categories to refugee assignment: the first is assignment through resettlement schemes, i.e., the authorities have information about the number of refugees that will need to be allocated within a specific time interval as well as the refugees’ characteristics. This is known as a static assignment. The other is for those refugees that directly arrive at a location, also known as asylum seekers, without a resettlement scheme, these are dynamic assignments. The introduction of dynamics results in uncertainty around the number of refugees that need to be allocated within the market. In this paper, we intend to explore two approaches suggested by Andersone et al. (2018) and Delacretaz, Kominers, & Teytelboym (2016). We will further discuss how these algorithms compare to those currently in place.
The global refugee population was estimated to be as high as 25.9 million at the end of 2018, up from 10.5 million in 2012, which represents an ever-growing global refugee market. The top ten countries of origin account for 82 percent of all refugees, or 16.6 million in 2018. Each year, there are over 3.5 million asylum seekers, with most going to the USA, Turkey, and Germany. This, together with the overall increase in demand for allocations, means that much more emphasis must be placed on the improvement of current mechanisms to make refugee allocations more efficient and cost-effective for all parties involved.
Improving Refugee Matching Regimes: An Exploration of Algorithms' Relative Merits
December 13, 2019
Introduction to Market
The global refugee population was estimated to be as high as 25.9 million at the end of 2018, up from 10.5 million in 2012, which represents an ever-growing global refugee market (UNCHR, 2018). The top ten countries of origin account for 82 percent of all refugees, or 16.6 million in 2018 (UNCHR, 2018). Each year, there are over 3.5 million asylum seekers, with most going to the USA, Turkey and Germany (UNCHR, 2018). This, together with the overall increase in demand for allocations, means that much more emphasis must be placed on the improvement of current mechanisms to make refugee allocations more efficient and cost-effective for all parties involved.
According to Andersson (2017), there are two distinct categories to refugee assignment: the first is assignment through resettlement schemes, i.e., the authorities have information about the number of refugees that will need to be allocated within a specific time interval as well as the refugees' characteristics. This is known as a static assignment. The other is for those refugees that directly arrive at a location, also known as asylum seekers, without a resettlement scheme, these are dynamic assignments. The introduction of dynamics results in uncertainty around the number of refugees that need to be allocated within the market.
Discussion of Matching Models
In class, we have covered several standard matching models, i.e. top-trading cycles and the deferred acceptance algorithm, that take only one variable into account when creating preferences or priorities. However, in the case of refugee allocations, the fundamental structure is more complicated, as more than one variable has to be taken into account (Delacretaz et. al. , 2016). For instance, not only does nationality play a crucial role, but also family composition, need for health care and other public services, or languages spoken, could be essential variables when allocating a refugee to a location. Delacretaz et. al. (2016) describes this as a multidimensional problem that, because of its complexity, further places constraints on the local authorities themselves, e.g., availability of education for children within the family. Under this framework, assignments become even more complex as complementarities are considered. Hence, standard matching models are insufficient for these allocations.
In the traditional two-sided matching model, we have seen that there are three key desirable attributes: stability, efficiency (fairness) and strategy-proofness. These need to be altered in the context of refugee assignments. In the traditional school matching system, stability is defined such that no student is not matched to a school in favor of another with lower priority. In the case of schools, this takes the form of justifiable envy, where students and schools end up dissatisfied with their current matching (Jones & Teytelboym, 2017). In the refugee case, the matching is stable when no family of a lower priority is matched with a location over a family of higher priority. The second characteristic is efficiency. In the context of school matching, no student can be accepted by a more preferred school without worsening the matching of another student. In the context of refugee matching, a family should not be matched with a more preferred location if it results in another family being matched with a less preferred location (Jones & Teytelboym, 2017). Lastly, strategy-proofness is also desired. Strategy-proofness (for students only) refers to the incentive for students to truthfully report preferences, such that it is not possible to manipulate results in one's favor. For refugees, no location or family should have an incentive to misrepresent their preferences in order to be matched with their most preferred localities.
However, these characteristics do not seem to be fully taken into consideration when looking at current algorithms that are in place to deal with refugee assignments. In the current system in use in the United States, nine agencies that specialize in matching refugees take turns in picking which refugee cases they would like to handle, based on biographical information provided to them. The refugees themselves do not get to express their preferences, making the matching mechanism more of an agent-object pairing than a pairing between two agents. Given that the US is one of the largest refugee-receiving countries, it is unfortunate that the system does not consider refugees' preferences. Jones and Teytelboym (2017) argue that due to information asymmetries, it is not ideal to ask refugees to list preferences of specific locations. However, they argue that refugees should be asked to express preferences over types of places, and their preferences should be taken into equal consideration when matching. This would increase efficiency for both sides of the matching.
In this paper, we intend to explore two approaches suggested by Andersone et al. (2018) and Delacretaz, Kominers, & Teytelboym (2016). We will further discuss how these algorithms compare to those currently in place.
Delacretaz, Kominers, and Teytelboym (Multidimensional Matching)
In their paper ‘Refugee Resettlement', Delacretaz, Kominers, and Teytelboym (2016) develop their ‘matching with multidimensional constraints' framework. The authors recognize that one of the greatest shortcomings in current matching mechanisms (or lack thereof) is that they are generally blind to the demand and supply for public services that refugees and localities create respectively. If refugees' needs for public services are ill-matched to those that their new communities can provide, it is clear that there will be sub-optimal results for both localities and refugees. As such, they propose several progressively more complex algorithms that are sensitive to service demand and provision in order to improve matching efficiency, incentivize truthful reporting of preferences by refugees, and better enforce localities' priorities, which should have the additional benefit of making communities more open to receiving migrants. The authors develop a multitude of approaches, only some of which we will be able to discuss in this paper.
The first major proposal that they make is the Multidimensional Top Trading Cycles (MTTC) algorithm. It is essentially a modified version of the Top Trading Cycles algorithm that we discussed in class, by which all ‘contracts' (our in-class arrows) are removed depending on the ability of localities to accommodate the service needs of refugees concerned. In theory, these service needs could be based on the evaluation of the refugee matching agency, perhaps with input for the refugees. The mechanism used for this is essentially a more complex version of the idea of ‘capacity' that we addressed in class when discussing the National Residency Matching Program, which is especially clear when the authors include housing as one of the possible constraints. They prove that this algorithm will produce Pareto-efficient results with respect to the preferences of refugees, that it is strategyproof for all parties involved, and further that it can be computed reasonably for the likely numbers of players involved.
While this algorithm would, in principle, seem an ideal solution to the problem that the authors set out to address, the fundamental issue is that it assumes that the social planner's priority here is the guarantee of refugees' preferences over the priorities of localities. This is a problem because, in practice, governments will want mechanisms that place more weight on their own priorities, especially given that it is only by their assent that placements are provided for resettlement. One method of doing so is to provide an exogenous tentative allocation prior to running the MTTC. However, the authors demonstrate that doing so increases the computational load to run the algorithm unreasonably, and so turn to a different solution: the Serial Multidimensional Top Trading Cycle (SMTTC) algorithm.
The SMTTC algorithm runs just like the original MTTC with initial endowments, with the difference that it labels only some localities and households as ‘active' on the basis of service demand and supply compatibility. In so doing, it drastically reduces the computational load and makes a more balanced approach between locality and refugee priorities and preferences possible. Again, we seem to have resolved the problem, but we are now caught between securing a matching that respects both the priorities of localities and the preferences of refugees. As we discussed in class, such a matching may not always exist in these cases. It is for this reason that we have to rely on concepts such as ‘fairness' or a lack of ‘justified envy'. Recognizing this, the authors seek to develop an algorithm that will discover such a matching.
This algorithm is the Top Choice algorithm, which will find a stable outcome if one exists. We do not have space in this paper to discuss the algorithm itself, especially given the critical flaw that determining whether a stable outcome exists is an extraordinarily difficult computational task. As such, the authors introduce the concept of ‘Quasi-Stability'. In short, quasi-stability specifies that a household cannot block a matching if it has the lowest priority among the other households matched to their most preferred locality. This property is not only mathematically convenient, but makes sense in the context of refugee resettlement because it essentially allows for the service provision of desired localities to go unused. As the authors note, given the dynamic nature of resettlement, services that go unused may be allocated to future refugees. Thus, this move to deemphasize the complete use implicit in the algorithms previously discussed does not come at a significant cost.
[...]
- Quote paper
- Anonymous,, 2019, Improving Refugee Matching Regimes, Munich, GRIN Verlag, https://www.grin.com/document/1289866
-
-
-
-
Upload your own papers! Earn money and win an iPhone X. -
-
Upload your own papers! Earn money and win an iPhone X. -
-
Upload your own papers! Earn money and win an iPhone X. -
-
Upload your own papers! Earn money and win an iPhone X. -
-
Upload your own papers! Earn money and win an iPhone X. -
-
Upload your own papers! Earn money and win an iPhone X.