Abstract
Two players, A and B, bargain to divide a perfectly divisible pie. In a bargaining model with constant discount factors, \(\delta _A\) and \(\delta _B\), we extend Rubinstein (Econometrica 50:97–110, 1982)’s alternating offers procedure to more general deterministic procedures, so that any player in any period can be the proposer. We show that each bargaining game with a deterministic procedure has a unique subgame perfect equilibrium (SPE) payoff outcome, which is efficient. Conversely, each efficient division of the pie can be supported as an SPE outcome by some procedure if \(\delta _A+\delta _B\ge 1\), while almost no division can ever be supported in SPE if \(\delta _A+\delta _B < 1\).
Similar content being viewed by others
Notes
For any t, the subgame that starts from period \(t+2\) has exactly the same structure as that which starts from period t.
In this paper, a payoff outcome \((u_A,u_B)\) where \(u_i\ge 0\) is said to be efficient if \(u_A+u_B=1\).
See, for example, Fudenberg and Tirole (1991, section 4.2).
References
Anesi, V., & Seidmann, D. J. (2014). Bargaining over an endogenous agenda. Theoretical Economics, 9, 445–482.
Binmore, K., Rubinstein, A., & Wolinsky, A. (1986). The Nash bargaining solution in economic modelling. The RAND Journal of Economics, 17, 176–188.
Binmore, K. G. (1987). Perfect equilibria in bargaining models. In P. Dasgupta & K. G. Binmore (Eds.), The economics of bargaining. Oxford: Basil Blackwell.
Chatterjee, K., Dutta, B., Ray, D., & Sengupta, K. (1993). A noncooperative theory of coalitional bargaining. Review of Economic Studies, 60, 463–477.
Fershtman, C. (1990). The importance of the agenda in bargaining. Games and Economic Behavior, 2, 224–238.
Fudenberg, D., & Levine, D. (1983). Subgame-perfect equilibria of finite- and infinite-horizon games. Journal of Economic Theory, 31(2), 251–268.
Fudenberg, D., & Tirole, J. (1991). Game theory. Cambridge: The MIT Press.
Hyndman, K., & Ray, D. (2007). Coalition formation with binding agreements. The Review of Economic Studies, 74, 1125–1147.
Ju, Y., & Wettstein, D. (2009). Implementing cooperative solution concepts: A generalized bidding approach. Economic Theory, 39, 307–330.
Muthoo, A. (1990). Bargaining without commitment. Games and Economic Behavior, 2, 291–297.
Muthoo, A. (1999). Bargaining Theory with Applications. Cambridge: Cambridge University Press.
Nash, J. (1953). Two-person cooperative games. Econometrica, 21, 128–140.
Okada, A. (2000). The efficiency principle in non-cooperative coalitional bargaining. The Japanese Economic Review, 51, 33–50.
Osborne, M., & Rubinstein, A. (1990). Bargaining and markets. Dublin: Academic Press.
Osborne, M., & Rubinstein, A. (1994). A course in game theory. Cambridge: The MIT Press.
Rubinstein, A. (1982). Perfect equilibrium in a bargaining model. Econometrica, 50, 97–110.
Shaked, A., & Sutton, J. (1984). Involuntary unemployment as a perfect equilibrium in a bargaining model. Econometrica, 52, 1351–1364.
Ståhl, I. (1972). Bargaining theory. Stockholm: Stockholm School of Economics.
Acknowledgments
I am indebted to Xinqiao Ping for his guidance and encouragement. I also thank Adam Brandenburger, Chih Chang, Sen Wei, Lixin Ye, the audience of several conferences, and two anonymous reviewers for very helpful comments and suggestions. This research is supported by the National Natural Science Foundation of China (Grant No. 71471119).
Author information
Authors and Affiliations
Corresponding author
Appendix
Appendix
Proof
(Proof of Theorem 1)
By symmetry, we only consider \(\omega \in \varOmega _A\). The theorem can be easily proved if \(\omega =A\), so we only consider procedures \(\omega \), such that \(r(w)\ge 1\).
First, define a strategy pair \(S^*=(S^*_A,S^*_B)\) as follows. When the game comes to period s, for any history, the proposer \(\omega _s=i\) proposes \(\theta (\omega (s))\) as the share for herself and \(1-\theta (\omega (s))\) as the share for her opponent \(j\ne i\) according to \(S_i^*\); the other player j accepts any offer in which the proposer’s share is no more than \(\theta (\omega (s))\) and rejects all other offers according to \(S_j^*\). It is obvious that if players follow \(S^*\), then the game will end by period \(t=1\) with agreement \(\big (\theta (\omega ),1-\theta (\omega )\big )\). We shall prove that \(S^*\) is an SPE of \(G(\omega ,\delta _{A},\delta _{B})\), and \(\big (\theta (\omega ),1-\theta (\omega )\big )\) is the only payoff pair that could ever be supported in SPE.
To show that \(S^*\) is an SPE, it is sufficient to verify that \(S^*\) satisfies the one-shot deviation principle, which asserts that a strategy profile S is an SPE of an extensive game with constant discount factors, if and only if no player can ever become strictly better off by deviating from S for just one period and then reverting to S.Footnote 7 Suppose the game has come to period \(t=s\). If the proposer \(\omega _s\), without loss of generality assumed to be A, deviates from \(S_A^*\) and suggests an offer in which her own share is larger than \(\theta (\omega (s))\), then player B will reject this offer according to \(S^*_B\), and let the game enter period \(s+1\). If \(\omega _{s+1}=A\), then A will get \(v(s+1)=\theta (\omega (s+1))\) in period \(s+1\) according to \(S^*\); if \(\omega _{s+1}=B\), then A will get \(v(s+1)=1-\theta (\omega (s+1))\) in period \(s+1\). In either case, if \(\delta _B>0\), we have
where \(\sum _{k=1}^{r(\omega (s))}(-1)^k p(\omega _s,k)<0\) due to Lemma 2; if \(\delta _B=0\), then \(\theta (\omega (s))=1\), thus we also have \(\theta (\omega (s))>\delta _Av(s+1)\) due to Lemma 4. This implies that A has no incentive to increase her own share in the offer in period s, since, otherwise, her payoff will decline from \(u_A(\theta (\omega (s)),s)\) to \(u_A(v(s+1),s+1)\). On the other hand, it is obvious that A will also not decrease her own share in the offer in period s. Similarly, we can show that B will neither increase nor decrease her reservation share \(1-\theta (\omega (s))\) for accepting an offer proposed by A in period s. Thus, we have proved that \(S^*\) satisfies the one-shot deviation principle, and hence, it is an SPE.
Now, we turn to the uniqueness part of the theorem, using a technique introduced by Shaked and Sutton (1984). Given \(i=0,1,\ldots ,r(\omega )\), let \(\Lambda _i\) be the subset of [0, 1], so that for each \(x\in \Lambda _i\), there exists an SPE \(f^x\) of the game \(G(\omega ,\delta _{A},\delta _{B})\) satisfying the following condition: when the game has come to period \(t_i=\sum _{j=0}^i n_j+1\) (where \(n_0=0\)), according to \(f^x\), the proposer would suggest an agreement in which his own share is x and his opponent’s share is \(1-x\). Let \(M_i\) and \(m_i\) denote the supremum and infimum of \(\Lambda _i\), respectively.
In period \(n_1+1\), player B can get no more than \(M_1\) in any SPE. Hence, A will offer B no more than \(\delta _BM_1\) in period \(n_1\), no more than \(\delta _B^2 M_1\) in period \(n_1-1\), ..., no more than \(\delta _B^{n_1} M_1\) in period 1. Thus, \(m_0\ge 1-\delta _B^{n_1}M_1\). Similarly, we have
where \(\delta _j= \delta _A\) if j is even, and \(\delta _j= \delta _B\) if j is odd.
If player B rejects A’s offers in period \(t=1,\ldots ,n_1\), he will get no less than \(m_1\) in period \(n_1+1\) in any SPE. Thus, due to Lemma 6, A will offer B no less than \(\delta _B^{n_1} m_1\) in period \(t=1\). Hence, \(M_0\le 1-\delta _B^{n_1}m_1\). Similarly, we have
where \(\delta _j= \delta _A\) if j is even, and \(\delta _j= \delta _B\) if j is odd.
If \(\omega \) is finite, by repeatedly using (3) and (4), we have
where \(\overline{m}_{r(\omega )}= m_{r(\omega )}\), \(\underline{m}_{r(\omega )}= M_{r(\omega )}\) if \(r(\omega )\) is even, and \(\overline{m}_{r(\omega )}= M_{r(\omega )}\), \(\underline{m}_{r(\omega )}= m_{r(\omega )}\) if \(r(\omega )\) is odd. However, in the last period of the game \(t_{r(\omega )}=T(\omega )\), in any SPE, the proposer always suggests the agreement in which her own share is one. Hence \(m_{r(\omega )}=M_{r(\omega )}=1\). Since \(m_0\le M_0\), we have \(m_0= M_0=\sum _{k=0}^{r(\omega )}(-1)^k p(\omega ,k)=\theta (\omega )\).
If \(\omega \) is infinite, we can similarly get
for each \(j\ge 1\), where \(\overline{m}_j= m_j\), \(\underline{m}_j= M_j\) if j is even, and \(\overline{m}_{j}= M_{j}\), \(\underline{m}_{j}= m_{j}\) if j is odd. Note that \(\lim _{j\rightarrow \infty }(-1)^jp(\omega ,j)\overline{m}_j=\lim _{j\rightarrow \infty }(-1)^jp(\omega ,j)\underline{m}_j=0\). Due to Lemma 1, let \(j\rightarrow \infty \), we have \(m_0= M_0=\sum _{k=0}^{\infty }(-1)^k p(\omega ,k)=\theta (\omega )\). Hence, we have proved the theorem. \(\square \)
Proof
(Proof of Theorem 2)
We shall only prove that for any \(x\in [1-\delta _B,1]\), there exists \(\omega \in {\varOmega }_A\), such that the payoff pair supported in SPE by \(\omega \) is \((x,1-x)\). For any \(x\in [0,1-\delta _B)\subset [0,\delta _A)\), a similar proof applies for \(\omega \in {\varOmega }_B\).
We can construct a procedure \(\omega \in {\varOmega }_A\) inductively as follows. If \(x=1\), then \(\omega =A\) suffices for the proof. Otherwise, we have \(1-\delta _B\le x<1\). Let \(n_1\) be the maximal integer, such that \(1-\delta _B^{n_1}\) is not greater than x. That is
If the equality in (5) holds, that is, \(x=1-\delta _B^{n_1}\), we can implement \((x,1-x)\) as an SPE outcome by \(\omega =A^{n_1}B\). Otherwise, we have \(1-\delta _B^{n_1}<x<1-\delta _B^{n_1+1}\). In this case, let \(n_2\) be the maximal integer, such that \(1-\delta _B^{n_1}+\delta _A^{n_2}\delta _B^{n_1}\) is not less than x. That is
If the equality in (6) holds, that is, \(x=1-\delta _B^{n_1}+\delta _A^{n_2}\delta _B^{n_1}\), we can implement \((x,1-x)\) in SPE by \(\omega =A^{n_1}B^{n_2}A\). Otherwise, we have \(1-\delta _B^{n_1}+\delta _A^{n_2+1}\delta _B^{n_1}<x<1-\delta _B^{n_1}+\delta _A^{n_2}\delta _B^{n_1}\), and should continue to construct \(\omega \).
Now, suppose by induction that \(n_1,\ldots ,n_r\) have already been defined, where \(r\ge 2\). If \(x=\sum _{k=0}^r(-1)^kp(\omega ,k)\), then \((x,1-x)\) can be supported in SPE by \(\omega =A^{n_1}B^{n_2}\cdots A^{n_r}B\) if r is odd, or by \(\omega =A^{n_1}B^{n_2}\cdots B^{n_r}A\) if r is even. Otherwise, if r is odd, then
while if r is even, then
In either case, we need to define \(n_{r+1}\). If r is odd, let \(n_{r+1}\) be the integer, such that
If r is even, let \(n_{r+1}\) be the integer, such that
We need to show that \(n_{r+1}\) is well defined, that is, there exists exactly one integer \(n_{r+1}\), such that (8) or (9) holds. If r is odd, \(\sum _{k=0}^{r+1}(-1)^kp(\omega ,k)\) is decreasing in \(n_{r+1}\). When \(n_{r+1}=1\), according to (7), we have
while as \(n_{r+1}\rightarrow \infty \), \(\sum _{k=0}^{r+1}(-1)^kp(\omega ,k)\rightarrow \sum _{k=0}^{r}(-1)^kp(\omega ,k)<x\). Thus, there exists a unique integer \(n_{r+1}\) that satisfies (8), that is, \(n_{r+1}\) is well defined. Similarly, we can prove \(n_{r+1}\) is well defined if r is even. We have thus finished defining \(n_{r+1}\).
Given \(x\in [1-\delta _B,1]\), if there exists an integer h, such that for \(n_1,\ldots ,n_h\) defined above, \(x=\sum _{k=0}^{h}(-1)^kp(\omega ,k)\), then from Theorem 1, the payoff pair \((x,1-x)\) can be supported as an SPE outcome by \(\omega =A^{n_1}B^{n_2}\cdots A^{n_h}B\) if h is odd, or by \(\omega =A^{n_1}B^{n_2}\cdots B^{n_h}A\) if h is even. If such a finite sequence of integers cannot be found, we will eventually get an infinite sequence \(n_1,n_2,\ldots \) and a corresponding infinite procedure \(\omega =A^{n_1}B^{n_2}\cdots \). Let
The elements in the sequence \(\{z_1,z_2,\ldots \}\) are alternately larger and smaller than x. According to Lemma 1, this sequence converges, and thus, its limit is x. That is, \(x=\sum _{k=0}^{\infty }(-1)^kp(\omega ,k)\). It follows from Theorem 1 that \((x,1-x)\) can be supported as an SPE outcome by \(\omega \). Thus, we have proved the theorem. Note that the condition \(\delta _A+\delta _B\ge 1\) is used in the middle inequality of (10). \(\square \)
Proof
(Proof of Theorem 3)
The proof of the theorem proceeds in steps.
Step 1: We shall show \((\delta _A,1-\delta _B)\cap \varGamma (\delta _A,\delta _B)=\emptyset \), that is, if \(\delta _A<x<1-\delta _B\), then there does not exist \(\omega \in \varOmega \), such that \((x,1-x)\) can be supported as an SPE outcome by \(\omega \). Suppose, on the contrary, that we can find such a procedure \(\omega \), and assume without loss of generality that \(\omega \in \varOmega _A\). According to Theorem 1 and Lemma 3, \(x=\theta (\omega )=\sum _{k=0}^{r(\omega )}(-1)^kp(\omega ,k) \ge 1-\delta _B^{n_1}\ge 1-\delta _B\), which contradicts \(x<1-\delta _{B}\). Thus, \((\delta _A,1-\delta _B)\cap \varGamma (\delta _A,\delta _B)=\emptyset \). Therefore, \(m[\varGamma (\delta _A,\delta _B)]\le 1-m[(\delta _A,1-\delta _B)]=\delta _A+\delta _B\).
Step 2: We shall first prove that \((1-\delta _B+\delta _B\delta _A,1-\delta _B^2)\cap \varGamma (\delta _A,\delta _B)=\emptyset \). Suppose, on the contrary, that we can find \(x\in (1-\delta _B+\delta _B\delta _A,1-\delta _B^2)\) and \(\omega \in \varOmega \), such that \((x,1-x)\) can be supported as an SPE outcome by \(\omega \). Since \(x>1-\delta _B\), we have \(\omega \in \varOmega _A\), that is, \(\omega _1=1\). Again, due to Theorem 1 and Lemma 3, if \(\omega _2=1\), then \(x\ge 1-\delta _B^{n_1}\ge 1-\delta _B^{2}\); if \(\omega _2=2\), then \(x\le 1-\delta _B+\delta _B\delta _A^{n_2}\le 1-\delta _B+\delta _B\delta _A\). This contradicts \(x\in (1-\delta _B+\delta _B\delta _A,1-\delta _B^2)\). Thus, \((1-\delta _B+\delta _B\delta _A,1-\delta _B^2)\cap \varGamma (\delta _A,\delta _B)=\emptyset \). Similarly, we can prove \((\delta _A^2,\delta _A-\delta _B\delta _A)\cap \varGamma (\delta _A,\delta _B)=\emptyset \). Hence, \(m[\varGamma (\delta _A,\delta _B)]\le \delta _A+\delta _B-m[(1-\delta _B+\delta _B\delta _A,1-\delta _B^2)]-m[(\delta _A^2,\delta _A-\delta _B\delta _A)]=(\delta _A+\delta _B)^2\).
Assume by induction that by the first \(k\ge 1\) steps, we have deleted the following \(2^{k}-1\) disjoint intervals from \(\varGamma (\delta _A,\delta _B)\): at step one, we delete \(A^1_1=(\delta _A,1-\delta _B)\), at step two, we delete \(A^2_1=(\delta _A^2,\delta _A-\delta _B\delta _A)\) and \(A^2_2=(1-\delta _B+\delta _B\delta _A,1-\delta _B^2)\), \(\cdots \), at step k, we delete \(A^k_1=(\delta _A^k,\delta _A^{k-1}(1-\delta _B))\), \(\ldots \), \(A^k_{2^{k-1}}=(1-\delta _B^{k-1}(1-\delta _A),1-\delta _B^k)\). The exact structure of each \(A^k_i\) can be derived according to the inductive algorithm described at step \(k+1\) below. There remain \(2^{k}\) intervals, whose total measure is \(1-\sum _{s=1}^k\sum _{i=1}^{2^{k-1}}m[A^s_i]=(\delta _A+\delta _B)^k\). Thus, \(m[\varGamma (\delta _A,\delta _B)]\le (\delta _A+\delta _B)^k\).
Now, at step \(k+1\), we further delete \(2^{k}\) disjoint intervals \(A^{k+1}_i\), \(i=1,\ldots ,2^{k}\), from \(\varGamma (\delta _A,\delta _B)\). More specifically, in each remaining interval \([P_i,Q_i]\) after step k, we delete \(A^{k+1}_i=\big (P_i+\delta _A(Q_i-P_i),Q_i-\delta _B(Q_i-P_i)\big )\). There remain two disjoint intervals in \([P_i,Q_i]\): \([P_i,P_i+\delta _A(Q_i-P_i)]\) and \([Q_i-\delta _B(Q_i-P_i),Q_i]\), whose total measure is \((\delta _A+\delta _B)(Q_i-P_i)\). Therefore, after step \(k+1\), there remain \(2^{k+1}\) intervals, whose total measure is \((\delta _A+\delta _B)^{k+1}\).
To show \(m[\varGamma (\delta _A,\delta _B)]\le (\delta _A+\delta _B)^{k+1}\), we still have to prove that \(A^{k+1}_i\cap \varGamma (\delta _A,\delta _B)=\emptyset \), \(i=1,\ldots ,2^{k}\). Without loss of generality, we only consider \(A^{k+1}_1=\big (\delta _A^{k+1},\delta _A^{k}(1-\delta _B)\big )\subset [P_1,Q_1]=[0,\delta _A^k]\). Suppose, on the contrary, that we can find \(x\in A^{k+1}_1\) and \(\omega \in \varOmega \), such that \((x,1-x)\) can be supported as an SPE outcome by \(\omega \). Since \(0<x<\delta _A<1-\delta _B\), we may write \(\omega =B^{n_1}A^{n_2}\cdots \), then \(x=\delta _A^{n_1}-\delta _A^{n_1}\delta _B^{n_2}+\cdots \). If \(n_1\ge k+1\), then \(x<\delta _A^{n_1}\le \delta _A^{k+1}\), which contradicts \(x> \delta _A^{k+1}\). If \(n_1\le k\), then \(x \ge \delta _A^{n_1}(1-\delta _B^{n_2})\ge \delta _A^{k}(1-\delta _B)\), which contradicts \(x<\delta _A^{k}(1-\delta _B)\). Thus, \(A^{k+1}_1\cap \varGamma (\delta _A,\delta _B)=\emptyset \).
Hence, we have proved \(m[\varGamma (\delta _A,\delta _B)]\le (\delta _A+\delta _B)^k\), \(\forall k\ge 1\). Since \(\delta _A+\delta _B<1\), we have \((\delta _A+\delta _B)^k\rightarrow 0\) as \(k\rightarrow \infty \). Thus, \(m[\varGamma (\delta _A,\delta _B)]=0\).
Proof
(Proof of Corollary 1)
In the proof of Theorem 3, we have shown that the set \([0,1]\backslash \varGamma (\delta _A,\delta _B)\) can be constructed as the union of infinitely many disjoint intervals: \(A^1_1=(\delta _A,1-\delta _B)\), \(A^2_1=(\delta _A^2,\delta _A-\delta _B\delta _A)\), \(A^2_2=(1-\delta _B+\delta _B\delta _A,1-\delta _B^2)\),..., \(A^k_1=(\delta _A^k,\delta _A^{k-1}(1-\delta _B))\), \(\ldots \), \(A^k_{2^{k-1}}=(1-\delta _B^{k-1}(1-\delta _A),1-\delta _B^k)\),...,. Let \(\{C_j\}_{j=1,2,\ldots }\) be a sequence, such that (a) for each \(A^k_i\), \(i=1,\ldots ,2^{k-1}\), \(k=1,2,\ldots \), there is one and only one \(C_j=A^k_i\), and (b) \(m(C_j)\ge m(C_{j+1})\), \(j=1,2,\ldots \),. Since \(\varGamma ^\lambda (\delta _A,\delta _B)=\varGamma ^\lambda (\delta _A,\delta _B)\cap [0,1] = \varGamma ^\lambda (\delta _A,\delta _B) \cap \left( \cup _{j=1}^\infty C_j \cup \varGamma (\delta _A,\delta _B)\right) = \cup _{j=1}^\infty \left( \varGamma ^\lambda (\delta _A,\delta _B)\cap C_j\right) \cup \varGamma (\delta _A,\delta _B)\), due to Theorem 3, we have
Since \(\delta _A\in \varGamma (\delta _A,\delta _B)\), \(1-\delta _B\in \varGamma (\delta _A,\delta _B)\) but \(A^1_1\cap \varGamma (\delta _A,\delta _B)=\emptyset \), we have \(A^1_1\cap \varGamma ^\lambda (\delta _A,\delta _B)= (\delta _A,\delta _A+\lambda ]\cup [1-\delta _B-\lambda ,1-\delta _B)\), and thus, \(m\big [C_1\cap \varGamma ^\lambda (\delta _A,\delta _B)\big ]= m\big [A^1_1\cap \varGamma ^\lambda (\delta _A,\delta _B)\big ]=\min \{2\lambda ,1-\delta _A-\delta _B\} = \min \{2\lambda ,m(C_1)\}\). Similarly, for each \(C_j\), \(j=1,2,\ldots \), we have \(m\big [C_j\cap \varGamma ^\lambda (\delta _A,\delta _B)\big ]=\min \{2\lambda ,m(C_j)\}\).
For each \(\varepsilon \), let \(\mu \) be a real number, such that \(\sum _{j=1}^{[{\varepsilon }/{(4\mu )}]}m(C_j) >1-\frac{\varepsilon }{2}\), where \([{\varepsilon }/{(4\mu )}]\) is the integer part of \({\varepsilon }/{(4\mu )}\). Due to the proof of Theorem 3, such \(\mu \) can be found when \(\mu \) is small enough. When \(\lambda <\mu \), according to (11), we have
This is the conclusion we want to prove. \(\square \)
Rights and permissions
About this article
Cite this article
Mao, L. Subgame perfect equilibrium in a bargaining model with deterministic procedures. Theory Decis 82, 485–500 (2017). https://doi.org/10.1007/s11238-016-9577-5
Published:
Issue Date:
DOI: https://doi.org/10.1007/s11238-016-9577-5