doi:10.1016/S0167-2681(01)00188-3
Copyright © 2001 Elsevier Science B.V. All rights reserved.
Equilibrium selection in a nonlinear duopoly game with adaptive expectations
Gian Italo Bischia and Michael Kopel
,
, b
a Istituto di Scienze Economiche,University of Urbino, Urbino, Italy
b Department of Managerial Economics and Industrial Organization, University of Technology, Vienna, Austria
Received 14 December 1999;
revised 22 August 2000;
accepted 29 August 2000
Available online 17 August 2001.
References and further reading may be available for this article. To view references and further reading you must
purchase this article.
Abstract
We analyze a nonlinear discrete time Cournot duopoly game, where players have adaptive expectations. The evolution of expected outputs over time is generated by the iteration of a noninvertible two-dimensional map. The long-run behavior is characterized by multistability, that is, the presence of coexisting stable consistent beliefs, which correspond to Nash equilibria in the quantity space. Hence, a problem of equilibrium selection arises and the long run outcome strongly depends on the choice of the players’ initial beliefs. We analyze the basins of attraction and their qualitative changes as the model parameters vary. We illustrate that the basins might be nonconnected sets and reveal the mechanism which is responsible for this often-neglected kind of complexity. The analysis of the global bifurcations which cause qualitative changes in the topological structure of the basins is carried out by the method of critical curves.
Author Keywords: Equilibrium selection problem; Duopoly games; Adaptive expectations; Basins of attraction; Critical curves
JEL classification codes: C72; D43; D83
Fig. 1. For μ1=μ2=3.4 the duopoly game with quadratic reaction functions and naive expectations has two stable steady-states (Nash equilibria) E1 and E2 and a stable cycle C2 of period two. These attractors are represented in the strategy space . The light and dark grey regions represent the basins of E1 and E2, respectively, the white region represents the basin of C2.
Fig. 2. Space of the parameters Ω={(μ,α)
μ>0,0≤α≤1} for the map T under the assumption of homogeneous players. Ωs(O) represent the set of parameters such that the fixed point O is asymptotically stable, Ωs(ES) represent the set of parameters such that the fixed point ES is asymptotically stable, Ωs(Ei) represents the common stability region of E1 and E2, Ωs(Ei,C2) represents the subset of Ωs(Ei) where the stable cycle C2 coexists with the two stable Nash equilibria E1 and E2. The portion of the curve of equation α=1/(μ+1) included inside Ωs(Ei) represents the set of parameters at which the transition between simply connected and nonconnected basins of the stable Nash equilibria E1 and E2 occurs.
Fig. 3. Representation of the basins of the equilibria E1 and E2 in the case of homogeneous behavior. The meaning of the colors is the same as in Fig. 1. (a) With parameters μ1=μ2=μ=3.4 and α1=α2=α=0.2<1/(μ+1), the fixed point O=(0,0) belongs to the region Z2 between the two branches LC(a) and LC(b) of the critical set LC, and the basins of E1 and E2 are simply connected sets. (b) With parameters μ=3.5 and α=0.5>1/(μ+1), the fixed point O=(0,0) belongs to the region Z4 bounded by LC(b), and the basins of E1 and E2 are nonconnected sets.
Fig. 4. Basins of E1 and E2 when players are heterogeneous with respect to their adaptive expectation rules. The colors have the same meaning as in the previous figures. (a) With μ1=μ2=3.6 and α1=0.55, α2=0.7 the stable set of the saddle point ES, which constitute the boundary between the two basins and , is entirely included in the regions Z2 and Z0. The basins are connected sets. (b) For μ1=μ2=3.6 and α1=0.59, α2=0.7 a portion of the stable set of the saddle point ES belongs to the region Z4, hence the preimages of the portion H0 of inside Z4 constitute nonconnected portion of nested inside .
Fig. 5. With μ1=μ2=μ=3.8 and α1=0.85, α2=0.8, the equilibria E1 and E2 are unstable. Two chaotic attractors A1 and A2 coexist, with basins and , represented by light and intermediate grey, respectively.
Fig. 6. (a) Critical curves of rank-0, obtained as the locus of points such that det(DT(x,y))=0. (b) Critical curves of rank-1, obtained as LC=T(LC−1). These curves separate the plane into three regions, denoted by Z4, Z2 and Z0 whose points have four, two or no rank-1 preimages, respectively. (c) Riemann foliation of the plane. With each point of the region Z4 four distinct inverses are associated, each defined on a different sheet of the foliation, whereas points of Z2 are associated with two sheets. The projection on the phase plane of the folds connecting different sheets are the critical curves LC.