American Political Science Review Vol.83 great powers of any bipolar system.As a uncertainty over time,they are more like- consequence,each is likely to appeal to a ly to occur in an n-power world than in a very different set of potential allies,and two-power world. the allies themselves will be highly con- Finally,exit risks are likely to be lower strained both with respect to their initial in a two-power system because realign- choice of an alliance partner and with ment is impossible for either great power. respect to their ability to shift alliances By default,alliance stability is also the ex- thereafter. clusive responsibility of each:neither of The addition of one or more states to the two great powers in such a system can the system automatically increases the expect any other state to prevent the range of choice available to allies.Because defection of an ally from within its bloc this change necessarily reduces the dis- (Snyder 1984).In a multipolar system,the tance between states relative to a bipolar interest in preserving alliance stability and system,it also increases the range of alli- the incentive to do so can be distributed ance choices and the opportunities for across more than one great power.As a alliance shifts relative to a two-power consequence,each may seek to transfer world.As a result,coalitions in an the burden of maintaining the alliance to n-power system,unlike alliances in a bi- the other. polar world,tend to be"unstable and vul- Thus,a strong analytic argument,as nerable to policy disagreement"(Snyder well as empirical data,supports the hy- 1984,415). pothesis that the risk of exit varies across A second reason that different exit risks international systems.This variation,in exist is that the distribution of informa- turn,induces variation in the discount tion about potential allies is likely to vary rates of alliance members.Allies in a across systems.The industrial organiza- multipolar system will tend to discount tion literature distinguishes between the future benefits accruing from open search goods and experience goods.Ap- markets among them more heavily than plied to international alliances,this dis- will their bipolar counterparts.The rela- tinction illustrates the impact of varia- tively lower value they place on the future tions in information on coalitional stabil- relative to the present reflects their recog- ity.In economics,consumers have com- nition of the "likelihood that the future plete information about the characterist- will not come"(Levi 1988,13).The dis- ics of search goods before purchase (a count factors of allies in a bipolar system, dress,for example);experience goods are in contrast,are not subject to the same products whose characteristics become downward bias:the greater stability of bi- fully known to consumers only after pur- polar coalitions allows the value of future chase(a restaurant's quality,for example) to approximate present benefits more (Tirole1988,106). closely. Because the amount of reliable informa- An analysis of relative exit risks sug- tion about potential great power allies is gests that the security externalities of any likely to be inversely related to their num- free trade agreement are more likely to re- ber,alliances are more likely in a bipolar main internalized within the alliances of system to resemble search goods and in a bipolar systems than they are within alli- multipolar system to resemble experience ances of multipolar ones.As a result,both goods.Thus,the passage of time leads to the economic and political benefits of in- greater learning about alliance partners in traalliance free trade are likely to be dis- a multipolar world than it does in a bi- counted less heavily in a two-power than polar world.To the extent that alliance in an n-power world.Thus,the dimension shifts are a function of the reduction of of exit advantages a bipolar system with 1250American Political Science Review Vol. 83 great powers of any bipolar system. As a consequence, each is likely to appeal to a very different set of potential allies, and the allies themselves will be highly con￾strained both with respect to their initial choice of an alliance partner and with respect to their ability to shift alliances thereafter. The addition of one or more states to the system automatically increases the range of choice available to allies. Because this change necessarily reduces the dis￾tance between states relative to a bipolar system, it also increases the range of alli￾ance choices and the opportunities for alliance shifts relative to a two-power world. As a result, coalitions in an n-power system, unlike alliances in a bi￾polar world, tend to be "unstable and vul￾nerable to policy disagreement" (Snyder 1984, 415). A second reason that different exit risks exist is that the distribution of informa￾tion about potential allies is likely to vary across systems. The industrial organiza￾tion literature distinguishes between search goods and experience goods. Ap￾plied to international alliances, this dis￾tinction illustrates the impact of varia￾tions in information on coalitional stabil￾ity. In economics, consumers have com￾plete information about the characterist￾ics of search goods before purchase (a dress, for example); experience goods are products whose characteristics become fully known to consumers only after pur￾chase (a restaurant's quality, for example) (Tirole 1988, 106). Because the amount of reliable informa￾tion about potential great power allies is likely to be inversely related to their num￾ber, alliances are more likely in a bipolar system to resemble search goods and in a multipolar system to resemble experience goods. Thus, the passage of time leads to greater learning about alliance partners in a multipolar world than it does in a bi￾polar world. To the extent that alliance shifts are a function of the reduction of uncertainty over time, they are more like￾ly to occur in an n-power world than in a two-power world. Finally, exit risks are likely to be lower in a two-power system because realign￾ment is impossible for either great power. By default, alliance stability is also the ex￾clusive responsibility of each: neither of the two great powers in such a system can expect any other state to prevent the defection of an ally from within its bloc (Snyder 1984). In a multipolar system, the interest in preserving alliance stability and the incentive to do so can be distributed across more than one great power. As a consequence, each may seek to transfer the burden of maintaining the alliance to the other. Thus, a strong analytic argument, as well as empirical data, supports the hy￾pothesis that the risk of exit varies across international systems. This variation, in turn, induces variation in the discount rates of alliance members. Allies in a multipolar system will tend to discount the future benefits accruing from open markets among them more heavily than will their bipolar counterparts. The rela￾tively lower value they place on the future relative to the present reflects their recog￾nition of the "likelihood that the future will not come" (Levi 1988, 13). The dis￾count factors of allies in a bipolar system, in contrast, are not subject to the same downward bias: the greater stability of bi￾polar coalitions allows the value of future to approximate present benefits more closely. An analysis of relative exit risks sug￾gests that the security externalities of any free trade agreement are more likely to re￾main internalized within the alliances of bipolar systems than they are within alli￾ances of multipolar ones. As a result, both the economic and political benefits of in￾traalliance free trade are likely to be dis￾counted less heavily in a two-power than in an n-power world. Thus, the dimension of exit advantages a bipolar system with