Macedonian Border Closings in the Kosovo Refugee Crisis: A Game-Theoretic Perspective

Abstract

The 1999 Kosovo crisis generated more than 800,000 refugees, the majority of whom fled into Albania or Macedonia. While Albanians welcomed their ethnic kin, Macedonia closed its border three times. We provide a game-theoretic perspective on strategic interactions between Macedonia and NATO. We use narrative accounts of the crisis to eliminate many configurations of payoffs. Among those remaining, the theory of moves isolates one game that can satisfy the necessary conditions for credible threats: the prisoners' dilemma. Credible threats of border closings in a prisoners' dilemma alter donor incentives and lead to international sharing of asylum burdens in repeated play.

Keywords:

• refugees
• asylum
• Kosovo
• Macedonia
• theory of moves
• prisoners' dilemma

The authors wish to thank Steven Brams, Glenn Palmer, and two anonymous referees for valuable comments and suggestions. We retain full responsibility for any remaining errors or other shortcomings of the paper.

Notes

¹ A shorter version of this paper (Williams & Zeager, 2004) focuses on the structure of the model and its applicability to other crises, with less attention to the details of the specific case considered here.

² Brams and Togman (1998) also use threat power analysis in TOM to understand the role of threats in reaching a cooperative solution to the Northern Ireland conflict.

³ Since the publication of Weiner (1996), some authors have suggested adding a category for environmental and economic refugees (Nafziger, Stewart, & Varynen, 2000, passim). However, following the UNHCR (2000, 23), these people can be considered economic migrants rather than refugees. While their plight may be dire, they lack the “well-founded fear of political persecution should they return home” required by international refugee law.

⁴ Two sources from the UNHCR (UNHCR, 2000, 234; del Mundo & Wilkerson, 1999, 19) attribute this leverage to NATO's need for consent to operate in Macedonian territory. Yet, NATO operated within the territory of other states in the region (to set up refugee camps) that did not exercise the same leverage and were not offered humanitarian evacuation programs. The refugees themselves seemed to recognize the difference in treatment when they had little interest in a proposal from the UNHCR to transfer them from Macedonia to Albania, which would have eliminated any opportunity for evacuation (USCR, 2000, 255).

⁵ The exact proportion of Albanians is not known, but probably ranges from 80% (Daalder & O'Hanlon, 2000) to 85% (IICK, 2000).

⁶ The Serbs also claim that several hundred thousand Serbs had been displaced from Kosovo between 1968 and 1988, while Albanians were moving in, though the actual figures are said to be exaggerated (Van Selm, 2000, 4).

⁷ Failure to resolve problems related to the Nagorno-Karabakh conflict between Armenia and Azerbaijan, including the enormous Azeri refugee problem, had raised questions about the OSCE's ability to intervene successfully in international disputes.

⁸ They were engaged at the ground level, but were overwhelmed by the flood of refugees and had to rely on NATO to work out the relationship with Macedonia (Minnear, 2000). This was true even of the UNHCR, which had rejected a NATO proposal for consultations on refugee management in 1998.

⁹ The UNHCR's role in this crisis was controversial, because the organization traditionally cooperates only with neutral, humanitarian military operations. Since NATO was a belligerent, the UNHCR activities could be seen as part of the anti-Serbian campaign. However, given the numbers of refugees involved and the speed with which NATO forces could respond, local UNHCR representatives had little choice but to go along (Surkhe et al., 2000, 109).

¹⁰ Several sources comment on the scope of the relief effort, as military units from several countries, NATO and non-NATO alike, built facilities to house the refugees (Minnear, 2000). While quality and performance varied, results were often impressive. There was considerable resentment within the UNHCR and NGO community (Minnear, 2000).

¹¹ The UNHCR's efforts were felt more among the organizations providing indirect funding, such as the International Monetary Fund (Minnear, 2000), but this is peripheral to our story of strategic behavior.

¹² The theory of moves can accommodate more than two players (e.g., Zeager, 1998), but it then sacrifices the capacity to analyze strategic threats, which is a key feature of the Kosovo crisis.

¹³ These longer term considerations help to explain why NATO's influence with Macedonia was greater than the assistance providers' influence. During the conflict there was no expectation that assistance providers would have a relationship with Macedonia after the conflict was resolved.

¹⁴ First asylum places a moral, and to some extent a legal, obligation on countries not to turn away a refugee if one appears first on their territory. Otherwise, countries could simply refuse entry or pass the refugee on to another country; the refugee could be handed from one country to another indefinitely with no one accepting responsibility.

¹⁵ Two games are equivalent if one game can be transformed into another by switching the row player's strategies, the column player's strategies, the players themselves, or any combination of these (Brams, 1994, 215).

¹⁶ Gilboa (1995), Stone (2001), and Brams (2001) offer contrasting perspectives on TOM as a framework for understanding strategic behavior. Stone's criticisms focus on the idea of moving power, using TOM for inherently simultaneous games (e.g., chicken), and the lack of axiomatic foundations for threat power (and other forms of power) in TOM. Moving power does not arise in this paper, since the games are noncyclic. In all the games considered here, both players have dominant strategies, so the dominant-strategy equilibria are the same, regardless of whether play is simultaneous or sequential. Finally, as Gilboa (1995, 370) argues, treating notions like threat power as primitives makes the analysis more lucid for practitioners.

¹⁷ Repeated play in TOM is different from the usual meaning in the game theory literature, because Brams (1994) does not impose cardinal utilities, use discounting, or assume that a stage game is played, de novo, again and again.

¹⁸ In other approaches to bargaining, the threat state would be called the bargaining solution and the breakdown state would be called the threat (or status quo) point (Mas-Collel, Whinston, & Green, 1995, 839).

¹⁹ Brams (1994, 145) claims that the conditions in (1) and x_ij ≥ 3 are sufficient for a credible threat. However, in private correspondence, he has confirmed that this claim is incorrect and noted that game 36 in TOM provides a counter-example.

²⁰ A point like (2.5, 3.5) has no meaning in Figures 3a–3c, because the players' payoffs only convey rank orderings of outcomes and do not admit cardinal interpretations.

²¹ As a dominant-strategy equilibrium, WF in Figure 3b is also the unique Nash equilibrium of the game. Furthermore, WF is the unique nonmyopic equilibrium for game 11 in the standard (strictly alternating moves) version of TOM (Brams, 1994, 217).

²² As in game 11 above, WF is both the unique Nash equilibrium of game 9 and the unique nonmyopic equilibrium under the standard rules of play in TOM (Brams, 1994, 217).

²³ Keep in mind that some of the Kosovo refugees had ethnic ties within Macedonia, were very numerous, and might be armed. Thus, Macedonia could face a difficult situation if NATO forces left Macedonia and the relief agencies to handle the asylum seekers.

²⁴ The folk theorem for repeated games (Kreps, 1990, 505–515) ensures that cooperation can be sustained in the prisoners' dilemma, provided the future looms sufficiently large in the decisions of the players. The future can be represented by a probability of the game continuing for another round or discount rates for the players. To derive sufficient conditions for cooperation (in terms of the probability of continuing or the discount rates), it is necessary to assign cardinal rather than ordinal payoffs, and it would typically be assumed that the players maximize expected utilities. For an approach to this problem, see Kreps (1990).

²⁵ Computer simulations by Axelrod (1984, 1997) have generated a large literature on the prospects for cooperation in the prisoners' dilemma game under repeated play. For critical reviews of this literature, see Binmore (1998) and Hoffman (2000).

²⁶ Note that the necessary conditions for a credible threat in (9) are satisfied for both players. Also, condition (10) for a deterrent threat is satisfied for Albania, while the condition x₁₂ = 4 for a compellent threat is satisfied for NATO.

²⁷ The Albanian government had been duped into encouraging nationals to invest in a pyramid scheme. When the original sponsors fled with all of the money, widespread rioting and raids on government facilities ensued.

²⁸ In 1994, the Albanian population in Macedonia was officially 21.7% of the overall population of 2.2 million. An Information Ministry official said it might be higher, perhaps 25%. Ethnic Albanians set the percentage much higher (Kamm, 1994).

²⁹ In Greece, opposition led to anti-NATO riots.

³⁰ Traditionally, responsibility for refugee management rested with the UNHCR, whose office handled the earlier crises in Bosnia, but the UNCHR itself was only prepared for about 100,000 refugees (UNHCR, 2000, 234).

³¹ This status was extended several times, with the last period expiring on March 28, 2000 ( Country Report: Macedonia, 2000).

³² The USCR (2000, 255) claims that Macedonia closed its border on March 23, but reopened it the following day. This incident is also noted in Figure 4.

³³ Another source puts the figure of people seeking refuge in Macedonia at 360,000 (Country Report, 2000).

³⁴ Lower-level negotiations continued as well. As the bombing campaign continued into mid-April, NATO and the Macedonian government negotiated camp extensions. The extension was delayed by Macedonia's desire that NATO use a state contractor. As the state contractor was not accustomed to a problem of this scale, NATO preferred more direct efforts. Germany provided $1 million for the camp extensions and assigned its soldiers to the task. The construction could have been handled by a combination of local labor, refugees, and NGOs, but “politically, the German alternative was the easiest” (Surkhe et al., 2000, 134).

³⁵ Proksch (2000) provides an excellent discussion on the Stability Pact.

³⁶ Macedonia had agreed to allow NATO facilities to be established rent-free, but insisted that NATO pay for utilities.

³⁷ For a recent discussion of NATO policy, see http://www.nato.int/docu/update/2002/11-november/e1129a.htm, downloaded 5 December 2002.

∗ Dominant strategy.