No More Learning

yUniversity of California at Berkeley and NBER; mschwartz@berkeley. edu. Part of this paper was written
while this author was a National Fellow at the Hoover Institution at Stanford. He would like to express his
gratitude to the Hoover Institution for enabling him to spend a productive year at Stanford.
z Institute for Advanced Study, Princeton, NJ, and New Economic School/CEFIR, Moscow;
ksonin@ias. edu.
? 1
1 Introduction
When the ria? e was invented, bows and arrows became obsolete. After the invention of nuclear weapons many observers expected that the conventional weapons would go the way of bows and arrows. For instance, one of the leading columnists of his time, Drew Pearson, declared in the Fall of 1945 that: i`the Navy and huge armies are now obsoletei^. 1 Later it became apparent that nuclear weapons do not give expected political leverage. In 1957, rea? ecting on the time when the US had a nuclear monopoly, Henry Kissinger commented with a measure of surprise that i^We never succeeded in translating our military [nuclear] superiority into a political advantage. i^Today, the verdict of history is clear: conventional arsenal remains important in the nuclear era.
Unfortunately, history is an infallible judge that never writes opinions to explain its ver- dicts. This paper o? ers a framework that deepens our understanding of the role of threats in the environments ranging from international negotiations to negotiations among or within a O? rm. We show that an i`indivisiblei^threat such as such as a nuclear capabilities gives a party far less bargaining power than a threat that is divisible, such as a large arsenal of con- ventional weapons. We also show that a self-enforcing peace agreement between negotiating parties should rely on a sequence of transfers or pre-negotiated concessions (as opposed to one large concession that buys peace). Fearon (1995) argues that indivisibility of the object of bargaining might be a rationalist explanation for cona? icts,2 while Powell (2004) argues that the commitment problem lies in the heart of the i? indivisibility argumenti? : if commitment is possible, a lottery over the indivisible object would be Pareto-superior to any cona? ict. We focus on how the ability to make observable probabilistic or divisible threats a? ects partiesi? bargaining power.
The study of blackmail received little attention from economists. At the O? rst glance, this may seem justiO? ed because relevance of blackmail may appear to be limited to small time crooks and nations with nuclear capabilities. However, in a broader sense many economic relationships involve an element of a blackmail. For instance, a manager induces a worker
1 For a full text of the quote and for analysis of evolution of conventional wisdom about the role of nuclear
weapons in national defence see Besse and Lasswell (1950).
2Early studies of war as a bargaining process are Schelling (1966), Wagner (1982), and Pillar (1983).
Recent models of negotiations in/and wars include Fearon (1995), Filson and Werner (2002), Smith and Stam (2003), Powell (2002, 2004), Slantchev (2003), Wittman (2003), Baliga and Sjostrom (2004).
? 2
to put forth more e? ort by a threat of O? ring, a tenant wants the landlord to do timely maintenance threatening to terminate the lease etc. Legal scholars devoted considerable attention to blackmail. 3 The celebrated Coase theorem (Coase, 1962) was in part an outcome of a decades-long intellectual inquiry of the author in legal questions related to blackmail (Coase, 1988). 4 A deO? ning feature of a successful blackmail is that the threat must be credible (e. g. , Shavell, 1992). Shavell and Spier (2002) analyze blackmail in the absence of binding commitment. In particular, they O? nd that if carrying out a threat is costly, then extracting a payment from a victim is impossible even in an inO? nite horizon setting. The main innovation of our paper relative to Shavell and Spier (2002) is that we introduce brinkmanship into the model. We deO? ne brinkmanship as an ability of the aggressor to undertake an observable action that will lead with positive probability to a war or some other mutually undesirable outcome; i`shooting over the headi^is a metaphor for the kind of brinkmanship we have in mind. We show that being able to engage in this kind of visible brinkmanship may allow the potential aggressor to extract the whole surplus from the victim.
In this paper, we employ the international-relations terminology, though the applications of this idea stretch far beyond the nuclear power politics. For example, in a relationship between a manager and an employee we observe the same pattern: if the only threat available to the manager is to O? re an employee, then manageri? s ability to extract e? ort from the employee may be less than in the case where the manager can make the punishment inO? nitely divisible or probabilistic. Intuitively, if the only options available to a manager are to O? re or to retain a worker, a worker only has to put forth enough e? ort to make the manager indi? erent between O? ring and retaining him. All the bargaining power resides with the worker. If the manager can impose arbitrarily small penalties on the worker he can induce a higher e? ort level (though it might be that in equilibrium penalties are never imposed).
For an economist, the phenomenon of war is in striking contradiction with the so-called i? Coase theoremi? that says that rational agents will always negotiate a Pareto e? cient arrange- ment. 5 At the O? rst glance, it may seem that the weaker party can avoid a cona? ict by trans-
3The fasciation of legal scholars with blackmail dates to at least the O? rst half of the 20th century, as
Campbell (1939) puts it: i^The law of blackmail has something in common with the blackmailer: it allows
its student no peace of mind. i^
4 Other works on economics of blackmail include Ginsburg and Shechtman (1993) and Posner (1993). 5Theories explaining failure of Pareto-e? ciency abound (see, e. g. , very di? erent perspectives in Coase,
1962, Myerson and Satterthwaite, 1983, Acemoglu, 2003). However, neither transaction costs, nor informa-
? 3
ferring resources to the stronger side. 6;7 However, in the absence of external enforcement, no peaceful agreement could ensure that a strong party would not ask for more concessions in the future. Furthermore, the more transfers the strong party receives, the stronger it becomes. One problem is that the commitment power of parties involved in any destructive cona? ict is very limited. 8 Indeed, a binding agreement that calls for a lump sum wealth transfer from one nation to the other in exchange for a promise of peace, can make both parties better o? . However, this type of agreement is not self-enforcing, except for an unlikely scenario, where the pay-o? to the potential aggressor from starting a war decreases substantially as a result of receiving a transfer. 9 Consequently, due to the lack of commitment power, a lump sum transfer in exchange for peace is unlikely to be viable. 10 Rajan and Zingales (2000) formalize this point and show that the lack of commitment power leads to ine? cient outcomes rang- ing from wars to wasteful political cona? icts in organizations. Acemoglu (2003) emphasizes that i^parties holding political power cannot make commitments to bind their future actions because there is no outside agency with the coercive capacity to enforce such arrangements. i^
This paper builds on Rajan and Zingales (2000). We show that their results need not hold
in dynamic setting where a continuos stream of transfers is feasible. Proposition 2 establishes
that resource transfers can form a basis for lasting peace. Unavailability of external contract
enforcement can be mitigated by spreading transfers over time. For example, a potential
aggressor, who receives a transfer every period, would O? nd it proO? table to refrain from
tional asymmetry can fully explain the phenomenon of war (e. g. , Fearon, 1995).
6Lengi? s (1993) typology of ina? uence strategies refers to such a strategy as appeasing. In Huth (1988)
typology, any policy of concessions is conciliatory. Hensel and Diehl (1994) analyze various non-militarized
responses on an immediate military threat. Wittman (2003) considers appeasement in a static setting and
argues that it should be possible to redraw the map so that peace becomes a self enforcing outcome. 7There is a strand of literature that argues that most wars are wanted from the ex-ante perspectives
(e. g, Blainley, 1988). Still, this case is much harder to make from the Pareto-e? ciency perspective: in rare
circumstances, a war is wanted on both sides.
8Barzel (2002) considers the absence of commitment a major cause for violent cona? icts: i^The problem
of . . . the inability to commit is also the root of theft and some warsi^. Classic work that focused on the problem of commitment is North and Weingast (1989) and Shepsle (1991). Most recently, the problem has become the central element in dynamic models of development suggested by Acemoglu and Robinson (e. g. , Acemoglu, 2003). Powell (2004) discusses commitment problems in various contexts close to ours.
9This is not entirely impossible. According to a legend when Attilla the Hun came to Rome, the citizens
of Rome gave him all the gold of Rome in exchange for peace, thus making conquest of the city less attractive
for him.
10Shavell and Spier (2002) invoke Vikings history to illustrate the importance of succession of payments.
? 4
starting a war as long as the present value of the stream of transfers exceeds the payo? from a war. Similarly, the other side would be willing to make transfer payments if the following two conditions are satisO? ed: (i) Present value of transfers is less than the expected loss from a war. (ii) The i? weaki? side believes that reducing the size of transfers would prompt the war. 11 We identify the conditions under which an appeasement strategy can succeed. We show that if parties have constant marginal utility of consumption then there exists a continuous stream of transfers that supports a sub-game perfect peaceful equilibrium. Constant marginal utility of consumption is mathematically equivalent to risk-neutrality. This assumption insures that any stream of transfers corresponds to a Pareto e? cient outcome.
If agents are risk-averse, then a self enforcing peace agreement may not be viable because concession will shift the balance of powers and increase future expected losses of the weaker party. 12 Risk aversion assumption implies diminishing marginal utility. Consequently, the weaker party may be very averse to being completely wiped out and the stronger party may gain very little from swallowing the weaker side completely. In this case, there may not exist a viable appeasement strategy, since appeasement may lead to a shift in the balance of power that makes the destruction of the weaker party inevitable. This is consistent with Fearon (1995) who argues that preventive wars should be understood as a result of commitment problems. 13 Slantchev (2001) analyzes the functioning of the Concert of Europe during the O? rst half of the 19th century. He concludes that "the pattern of cooperative behavior is seen to result from the commitment to uphold the settlement, which hinged on the credibility of enforcement threats. . . ".
This logic helps us understand many historic events. Every year since 1933, then- European powers, including England, France, and Poland faced a dilemma: either to O? ght Hitler now, or appease him and wait (Taylor, 1961). First concession was postponement of the World War I debt payments, next came tolerance of mobilization in Germany and launching of German air forces, and then re-occupation of Ruhrgebiet. The Munich treaty
11In fact, arrangements of this nature were practiced throughout history. A number of medieval cities bought peace by making annual contributions to their stronger neighbors. Above arguments suggest that a self-enforcing peace agreement may be viable, provided wealth transfers have a su? ciently small ina? uence on the balance of powers.
12Addressing the moral side of the problem, Walzer (1992) draws a parallel between international negotia-
tions and negotiations with kidnappers and extortionists. Our argument might explain the underlying logic
of legal obstacles to the former negotiations.
13Powell (2004) provides a survey of the literature related to the commitment problem.
? 5
of 1938 year, where the (future) allies agreed with occupation of Sudetenland, a part of Czechoslovakia, is probably the most prominent concession (e. g. , Walzer, 1992; for a game- theoretic analysis, see Hirshleifer, 2001). On October 5, 1938 Winston Churchill, the leader of the opposition in the British parliament, said that i`it seems to me that all those countries of Middle Europe, all those Danubian countries, will, one after another, be drawn into this vast system of power politics - not only power military politics but power economic politics n? radiating from Berlin, and I believe this can be achieved quite smoothly and swiftly and will not necessarily entail the O? ring of a single shot. . . . i^In 1940, it became too clear that Germany is already too strong, and England actually entered the World War II out of the fear that Germany would be even stronger in the future. 14 On May 13, Churchill, a prime-minister for O? ve days, said: i`What is our aim? I can answer in one word: Victory - victory - at all costs, victory, . . . however long and hard the road may be; for without victory, there is no survival. i^
The same reasoning may explain why rulers often miss opportunities to make concessions to the opposition in the days preceding revolutions. Why would not the Russian czar agree to share power with the elected parliament (where the majority was very far from any leftist movement) in January 1917? Why Louis XVI gambled on counter-revolutionary plots during the O? rst years of French revolution, instead of trying to work out a real political compromise? The answer is that the reluctance of these rulers can be explained by the fear that any concession would have empowered the opposition to ask for even more concessions. Then, they would have a choice between making an escalating sequence of greater and greater concessions or an open struggle under less favorable conditions. Nikolai II of Russia and Louis XVI were the ultimate losers, but there are examples of rulers who ultimately prevailed. 15
In Section 4 we show that the bargaining power of the potential aggressor increases dramatically if she is able to make her threat divisible. For instance she may achieve so by taking observable action that lead to a war with positive probability. What might come as a surprising result is that ability of the aggressor to make probabilistic threats may dramatically increase her bargaining power. Here probabilistic threat is an observable action
14Formally, Britain announced that she is in war with Germany on September 3, 1939, but the active
military stage begun no earlier than Summer 1940.
15Louis XIV of France strategy during the Fronda years might be such an example. John the Landless
policy is a textbook example of an appeasement strategy towards the political opposition. 6
? that leads to war with small exogenously given probability. (If the weak party believes that the strong party does not start a war, it does not transfer any resources to appease the potential aggressor. ) If the observable action that might leads to a war is taken, but the war does not occur, the other party sees the aggressori?
s commitment to action. In this case, there may exist a subgame perfect equilibrium where the aggressor extracts the whole surplus from the weak party. We can view brinkmanship as an instance of a probabilistic threat. Divisible threats may appear in blackmail practices, where the blackmailer have a possibility to divide the harm he could impose on the victim inO? nitely. 16
Economic science is relatively successful in modeling and explaining bargaining processes such as litigation or negotiation between O? rms and trade unions with a strike as a threat point. 17 In spite of apparent similarity, there are also profound di? erences between litigation and armed cona? icts. In order to accurately model negotiations leading to a war, the essential distinctions between wars and related phenomena must be identiO? ed.
First, international agreements must be self-enforcing, while settlements of legal disputes may be enforced externally. Of course, there are many other economic relationships where actions can not be veriO? able by courts and hence parties are constraint to follow self enforcing strategies. For instance, a married couple or an employer and employe may have a contract that speciO? es what happen in case of a brake up of the relationship, however, a contract that speciO? es behavior of both parties within the relationship is often not feasible. Consequently, the results of economic research on negotiations and litigation are often not directly relevant for understanding of the origins of war and causes for ine? cient outcomes in many important economic relationships. Second, while the possibility of a war serves as a threat point during negotiation process, a war itself is largely outside of negotiation process. Unlike a military cona? ict or divorce, a labor strike may be viewed as a continuous process of resource destruction, where a settlement is possible every step of the way. Indeed, settlements in legal or labor disputes are routinely negotiated during resource destructive stage of a cona? ict. A lengthy trial or a strike is not only threat point in negotiations but, in fact, an integral part of the bargaining process. In contrast, O? ring of an employe leads to an end of an employment relationship and the beginning of a war often marks an end to the negotiation
16If the blackmailer can impose harm only once (or do this a O? nite number of times), there is no sub-game
perfect equilibrium, where the victimi? s transfers are non-zero.
17Most notably, Rubinstein (1984) and Samuelson (1984) where among the O? rst to study bargaining under
asymmetric information. Arrow et al (1995) provides a general economic background for cona? ict resolution. 7
? process and makes a quick settlement unlikely. Therefore, a war can be treated as a threat point rather than a stage of the bargaining process: once a war starts, it cannot be stopped in the middle. 18
There are two related strands in the theoretical literature: the O? rst one is concerned with determinants of wars between countries (see Powell, 1998 and 2002 and references therein). The second is related to the famous Coase theorem (Coase, 1962). Acemoglu (2003) illustrates the failure of bargaining between social groups toward an e? cient policy (see also Becker, 1983, and Dixit et al, 2000). Fearon (1995) surveys massive literature on origins of cona? icts. Wilkenfeld (1991) shows that the two parties relative military capacity ina? uence the likelihood of weaker party concessions. Garnham (1976) and Weede (1976) demonstrated empirically that two states are less likely to have a war when one of them have a substantial military advantage. Blainley (1988) and Holti (1991) investigate the origins of wars. In particular, they suggest that a war may be caused by expectations of a shift in the balance of power. Undoubtedly, changes in the balance of power are important component in emergence of international cona? icts. We show that even if the balance of powers does not have an exogenously given time trend, a self-enforcing peace agreement may not be feasible.
The rest of the paper is organized as follows. Section 2 o? ers a description of a bargaining game that captures some essential features of wars. In Section 3, we analyze conditions that allows peace to prevail. The main result is in Section 4 that introduce an additional feature to the model that captures the notion of brinkmanship. Namely we consider a possibility that the potential aggressor can take an observable actions that will lead to war with small positive probability. Section 5 concludes.
18In some sense, the di? culty of settling a war i^in the middle of the cona? icti^relative to settling a labor dispute is quantitative. However, this distinction is so dramatic that it becomes qualitative. Naturally, as a semantic matter a war can never be stopped in the middle: By deO? nition, whenever an armed cona? ict is settled it is the end and not the middle of a war. But, aside from semantic casuistry, it is the lack of external contract enforcement again, that makes negotiating an end to a war extremely di? cult. Negotiating a self-enforcing peace accord at the time when active military actions are taking place is often impossible due to the nature of warfare technology. Litigation is sometimes described as a form of i^ritualized warfarei^, yet analogy is far from perfect because i^litigation ritualsi^ and rules are enforced externally. In order to foster e? ciency, a settlement during a cona? ict is encouraged by the legal system; the same cannot be said about wars.
? 8
2 The Setup
Consider a world consisting of two countries, A and B. Country A is a potential aggressor, whose expected payo? from a war is greater than that of B. Countryi? s B expected payo? from a war is negative. Consequently, B may be willing to pay tribute (bribes) to A in order to preserve peace. At each instant of time t > 0 player A (the potential aggressor) chooses whether or not to start a war, at 2 fP; W g; while the player B can adjust the rate of transfer to A: 19 The game ends if a war starts. During the pre-war period, the consumption of both parties is determined by the history of transfers. If the war occurs, the post war consumption is determined by the war outcome, while the probability distribution of war outcomes may be in turn ina? uenced by pre-war wealth transfers. 20 Both countries maximize expected discounted utility of consumption.
In a reduced form, a strategy of the weaker country (player B) can be represented by
a non-negative di? erentiable function b : R+ ! R+; t 7! bt: In the event that the game
proceeds until time t without a war, bt represents the rate at which B transfers resources to
A at time t. To deO? ne an aggressori? s strategy, let us summarize the history of the game up to
time T by an element from the set HT of non-negative di? erentiable function with support
on [0;T[:21 Each strategy of player A prescribes her to start or not to start a war for each
history of transfers, i. e. a strategy of player A is a function F : HT ! fW;Pg. The payo?
to playerRi is the discounted utility of the players consumption proO? le cit; the payo? equals
toVi = 1 tU(c )dtwhereU0(:)>0andU00(:)0,i=A;Band isthediscountrate. 22 0iit i i
Instantaneous consumption cit equals to some base rate plus a transfer received. Thus, without loss of generality, we normalize total consumption to zero, so that consumption proO? le prior to war is given by cAt = bt and cBt = bt for t  T; where T denotes the time when agent A selects action W: If a war never occurs, we can say that T = 1. If a war
19The same situation might be modelled assuming full symmetry between countries.
20Gains/losses from a war tend to be spread out over time. For example, the party that prevails in
the cona? ict may receive a stream of reparation payments or a perpetual stream of beneO? ts from concurred
territories. Consequently, we assume that the post war consumption is a constant stream of beneO? ts/losses
determined by the outcome of the war.
21The assumption that the function representing transfers is di? erentiable is essentially equivalent to the
assumption that the agent chooses the level of transfers at time zero, and chooses the rate of change at each
subsequent moment.
22For the sake of simplicity of notations, we assume here that both players are described by the same
utility function U(:).
? 9
occurs, the total consumptions is permanently reduced so that cAt + cBt = L for t > T , where L is the loss of resources due to war. The post-war consumption of players are cAt = x and cBt = x L for all t > T; where x is random variable that represents the outcome of the war. In order to allow for a possibility that transfers shift the balance of power in favor of the recipient we assume that x is distributed according to probability density function g(xjc) where c is the consumption of party A at the time immediately preceding the war, the distribution g(xjc) is atomless, continuous, and dE[xjc] > 0.
dc
We leave a more formal description of the game to the appendix. Throughout the paper, our analysis is conO? ned to subgame-perfect Nash equilibria. A peaceful equilibrium is any equilibrium in which a war occurs with zero probability.
3 Peace and Transfers
In this section, we use the model developed above to obtain insights into the origins of war and international peace treaties. Proposition 1 establishes necessary and su? cient conditions for maintaining peace without one-sided concessions or wealth transfers. Proposition 3 shows that a self-enforcing peace agreement is viable whenever transfers do not change the balance of powers. In such an equilibrium, all gains from peace are captured by the weaker party, i. e. leaving the potential aggressor indi? erent between starting and not starting a war. Even if transfers increase the beneO? t from a war to the potential aggressor, thus increasing his demands for more concessions, a self-enforcing peace agreement between risk neutral parties is feasible (see Proposition 2). However, if parties are risk-averse, a self-enforcing peace agreement may not be feasible.
The O? rst two Propositions of this section replicate and generalize the result of Shavell and Spier (2002) for our setting. Unlike Shavell and Spier (2002), we consider a continuous time model and allow for risk aversion. The results of this sections are important for understanding the main result presented in the next section. Peace without transfers is possible only if no party expects to beneO? t from a war. The following straightforward proposition makes this claim formal.
Proposition 1 A peaceful equilibrium without transfers exists if and only if the agent Ai? s expected instantaneous utility from starting a cona? ict does not exceed her instantaneous utility
? 10
under the status-quo, that is R +1 UA(x)g(xj0)dx  UA(0). If the inequality is strict, any 1
sub-game perfect equilibrium is peaceful.
Now let us consider a more interesting case: a possibility of peaceful equilibrium with positive transfers. A possibility of peace, especially when transfers change the balance of powers is not obvious, because in this case concessions will lead to demands for new and bigger concessions (see Rajan and Zingales, 2000, for an example of a model where ine? cient outcome obtains for exactly this reason). Hirshleifer (2001) argues that this might be the case to start a war as early as possible. The following proposition shows that risk-neutral parties can always negotiate a self-enforcing peace agreement.
Proposition 2 Suppose that players have a constant marginal utility of consumption, U(c) = c: Then there exists an equilibrium transfer stream such that a war does not occur on the (sub-game perfect) equilibrium path.
The idea of the proof is to construct a transfer proO? le that makes party A indi? erent between starting and not starting a war at any point in time. Such strategy gives all rents to B, and so we refer to such proO? le as B-proO? le. By construction Ai? s best response to B-proO? le is to refrain from starting the war. It is also an equilibrium for player B to follow B-transfer proO? le because B will be strictly worse o? if she deviates (because a war will result). Both parties recognize that transfers received by a potential aggressor in the present will allow her to gain more transfers (or wage a more proO? table war) in the future. Indeed, consumption stream resulting from starting a war at period t is the amount xt  E[xjbt]: Player A may be willing to postpone the war at time t even if the transfer that he receives at time t and immediately afterwards is substantially less than the post war consumption level would have been. If transfers shift balance of power, an increase in transfers to A will trigger even greater demands by the potential aggressor tomorrow. In equilibrium future transfers will increase over time to some limit denoted b1 where b1 = limt! 1 bt. For a given probability distribution of war outcomes g(xjc) the asymptotic value of transfer is given by the non-negative root of the following equation E[xjc = b1] = b1. 23
History provides us with a number of examples illustrating that a stream of transfers could
support peace better than lump-sum transfer. For decades since the last days of Chinguiz
23Forinstance,ifE[xjc]=1+ c thensolving1+ b1 =b1weobtainb1=c=1+1p5:Alsonote c+1 b1+1 22
that E[xjc = b1] = b1 only if b1 is O? nite.
? ? ? ? ? ? ? 11
Khan, the Mongol Empire was the largest state in the history of Earth by size (and the largest of the medieval empires by population). Most of the empire parts were controlled by relatively small forces, which collected modest, but regular tributes from occupied lands. Mongols used a threat of war to extract tributes from the territories under their control. In the political domain, a concession by one party often changes the balance of powers that in turn leads to an escalating sequence of concessions. (The history of the civil-rights movement provide a vivid example of this. ) Making a small change in a political institution may shift the balance of power making an institution unstable and leading to a sequence of reforms. Laguno? (2004) investigates the set of stable political institutions in a static setting. Our model suggests that political institutions may also be in a dynamic equilibrium where stakeholders initiating reform (as well as these opposed to it) understand and correctly anticipate that the change will trigger future reforms.
Now let us turn to the case where transfers do not ina? uence the balance of power.
Proposition 3 Suppose that A and B are risk-averse, and that past transfers do not a? ect the probability distribution of war outcomes, i. e. g(xjc) = g(x) for any c: Then there exists a peaceful sub-game perfect equilibrium.
From Proposition 1, we know that there exists a peaceful equilibrium whenever E[UA(X)]  UA(0). Let us consider the case when E[UA(X)] > UA(0); and hence E[X]  0 by the risk- aversion assumption. Consider the following strategy proO? le: the strategy of player B calls for B-proO? le transfer schedule, i. e.