## Breeding hybrid strategies: optimal behavior for oligopolists (1992)

Venue: | Journal of Evolutionary Economics |

Citations: | 32 - 9 self |

### BibTeX

@ARTICLE{Marks92breedinghybrid,

author = {Robert E. Marks},

title = {Breeding hybrid strategies: optimal behavior for oligopolists},

journal = {Journal of Evolutionary Economics},

year = {1992},

volume = {2},

pages = {17--38}

}

### Years of Citing Articles

### OpenURL

### Abstract

OLIGOPOLISTIC pricing decisions—in which the choice variable is not dichotomous as in the simple Prisoner’s Dilemma but continuous—have been modeled as a Generalized Prisoner’s Dilemma (GPD) by Fader and Hauser, who sought, in the two MIT Computer Strategy Tournaments, to obtain an effective generalization of Rapoport’s Tit for Tat for the threeperson repeated game. Holland’s genetic algorithm and Axelrod’s representation of contingent strategies provide a means of generating new strategies in the computer, through machine learning, without outside submissions. The paper discusses how findings from two-person tournaments can be extended to the GPD, in particular how the author’s winning strategy in the Second MIT Competitive Strategy Tournament could be bettered. The paper provides insight into how oligopolistic pricing competitors can successfully compete, and underlines the importance of “niche ” strategies, successful

### Citations

7342 |
J.H.: Genetic Algorithms and
- Goldberg, Holland
- 1988
(Show Context)
Citation Context ...t mutation is capable only of providing a random walk through the space of possible structures. The basic concepts of GAs were developed by Holland (1975) and his students (De Jong 1980; Bethke 1981; =-=Goldberg 1988-=-; and others). Theoretical considerations concerning the allocation of trials to structures (Holland 1975; De Jong 1980) show that genetic techniques provide a nearoptimal heuristic for information ga... |

2827 |
Adaptation in natural and artificial systems
- Holland
- 1975
(Show Context)
Citation Context ...string can be looked at as a collection of substring components or schemata which together account for the good or bad performance of the individual structure. Then Holland’s Schema Sampling Theorem (=-=Holland 1975-=-, Davis 1991) demonstrates that schemata represented in the population will be sampled in future generations in relation to their observed average fitness, if we can assume that the average fitness of... |

1788 | The Complexity of Cooperation
- Axelrod
- 1997
(Show Context)
Citation Context ...uter algorithms. He called for entries of strategies (for the repeated PD) coded as computer algorithms, and ran successive tournaments that attempted to reveal the “best” (highest scoring) strategy (=-=Axelrod 1984-=-, Axelrod and Dion 1988). In essence the tournaments were an attempt to search the strategy space by asking researchers in diverse disciplines to devise and submit strategies. As is now widely known, ... |

778 | Evolution and the Theory of Games - Smith, J - 1982 |

399 |
Reexamination of the Perfectness Concept for Equilibrium Points in Extensive Games
- Selten
- 1975
(Show Context)
Citation Context ...wo-person, normal-form games of complete information. He proves that, under certain properties of the evolutionary process, equilibrium strategies will be supported that are “trembling-hand perfect” (=-=Selten 1975-=-, 1983; Binmore and Dasgupta 1986), a subset of Cournot–Nash equilibrium. Our results support the contention that “repetition breeds coöperation”, at least for two-person games with unique Nash–Courno... |

384 |
Optimization of control parameters for genetic algorithms
- Grefenstette
- 1986
(Show Context)
Citation Context ...lland 1975; De Jong 1980) show that genetic techniques provide a nearoptimal heuristic for information gathering in complex search spaces. A number of experimental studies (De Jong 1980; Bethke 1981; =-=Grefenstette 1986-=-) have shown that GAs exhibit impressive efficiency in practice. While classical gradient search techniques are more efficient for problems which satisfy tight constraints (e.g., continuity, low-dimen... |

376 |
Agreeing to disagree
- Aumann
- 1976
(Show Context)
Citation Context ...adner (1980, 1986) assumed a type of bounded rationality similar to satisficing. Kreps et al. (1982) assumed incomplete information—they relaxed the assumption that rationality is “common knowledge” (=-=Aumann 1976-=-) among the players. Neyman (1985) and Radner (1986) argued that limited complexity of players’ strategies, and Harrington (1987) argued that limited complexity of players’ beliefs, could result in th... |

310 | The folk theorem in repeated games with discounting or with incomplete information,” Econometrica - Fudenberg, Maskin - 1986 |

252 |
The evolution of strategies in the iterated prisoner’s dilemma
- Axelrod
- 1987
(Show Context)
Citation Context ...d—albeit in more general terms—by Aumann (1985, pp.218–219) and by Binmore and Dasgupta (1986, pp.6–7, 12–14). Axelrod then used the GA to “breed” strategies in the two-person - 3 - repeated PD game (=-=Axelrod 1987-=-, Forrest and Mayer-Kress 1991). Axelrod reported that the GA evolved strategy populations whose median member was just as successful as Tit for Tat, whom they closely resembled. (In 95% of the time, ... |

195 | Rational cooperation in the finitely repeated prisoners’ dilemma - Kreps, Milgrom, et al. - 1982 |

174 |
Games and decisions: Introduction and critical survey. NewYork
- Luce, Raiffa
- 1957
(Show Context)
Citation Context ..., as they learn more about the type of behaviour they can expect from each other and build up a set of commonly held norms of behaviour. An early analysis of successful strategies in the repeated PD (=-=Luce and Raiffa 1957-=-, pp.97–102) suggested that continued, mutual coöperation might be a viable strategy, despite the rewards from defection, but for twenty years no stronger analytical results were obtained for the repe... |

154 | 1838): Recherches sur les Principes Mathématiques de la Théorie des Richesses, Paris. English translation (1897): N - Cournot |

151 |
A Learning System Based on Genetic Adaptive Algorithms
- Smith
- 1980
(Show Context)
Citation Context ...gs which are ever more successful at playing the repeated game. As explained by Schaffer and Grefenstette (1988), our process can be classified as an example of the Pitt approach to machine learning (=-=Smith 1981-=-), in which each string is evaluated for evolutionary fitness (its score in the repeated game), and this score is used to control the selection of strings used to generate a new set of strings. The pa... |

122 |
Nonstationary function optimization using genetic algorithms with dominance and diploidy
- Goldberg, Smith
- 1987
(Show Context)
Citation Context ...on of 50 strategies converged to coöperative behaviour (A, A) after 419 generations of bootstrapping, using 22-round games (w = 0.9545, r = 4.76% per round). _______________ 4. It has been suggested (=-=Goldberg and Smith 1987-=-) that the recessive genes of diploid genotypes are a reservoir of stored information that proved fit in earlier environments, and that GAs which included diploidy and dominance will thus perform bett... |

106 | Money as a Medium of Exchange in an Economy with Artificially Intelligent Agents - Marimon, McGrattan, et al. - 1990 |

102 | Bounded complexity justifies cooperation in the finitely repeated prisoners’ dilemma - Neyman - 1985 |

97 |
The Coevolution of Automata in the Repeated Prisoner’s Dilemma
- Miller
(Show Context)
Citation Context ..., nasty, grudging, forgiving, generous, etc.). As soon as a string gets much over 16 bits long, such recognition is difficult. Perhaps using constructs from the theory of finite automata will assist (=-=Miller 1989-=-, Marks 1990); perhaps there is no way to characterize a complex strategy: the only way to understand it is to watch its behaviour in a repeated game. This representation allows us to use the GA to de... |

81 | Théorie mathématique de la richesse sociale. Journal des savants - Bertrand - 1883 |

65 | No pure strategy is evolutionarily stable in the repeated prisoner’s dilemma game - Boyd, Lorderbaum - 1987 |

59 |
Lectures on Game Theory
- Aumann
- 1989
(Show Context)
Citation Context ...or convenience, except when the implicit discount rate (r = 4.76% per round) was too low, as discussed below. There is a discontinuity for infinitely repeated games (or supergames): the Folk Theorem (=-=Aumann 1989-=-) tells us that any individually rational payoff vector can be supported in infinitely repeated games, for sufficiently low discount rates. (For high discount rates the threat of future punishment may... |

56 |
Legal Theory
- Friedman
- 1953
(Show Context)
Citation Context ... algorithms. 2. Game Theory and Strategic Behaviour MICRO-ECONOMICS has recently been enriched by studies of strategic behaviour among small numbers of competitors, in oligopolistic markets (see J.W. =-=Friedman 1983-=-, for instance). Following Cournot, these have been in terms of the dynamic adjustment of the competitors’ behaviours, and have been facilitated by the insights from game theory (Schelling 1984; Ulph ... |

48 | Survey of repeated games - Aumann - 1981 |

45 | Adaptive System design: a genetic approach - Jong - 1980 |

43 | Collusive Behavior in Noncooperative Epsilon-equilibria of Oligopolies with Long but Finite Lives - Radner - 1980 |

35 |
Genetic algorithms as function optimizers. Doctoral dissertation
- Bethke
- 1981
(Show Context)
Citation Context ...nes would have been overtaken by genes with better evolutionary strategies,” although there is some dispute about the extent to which such outcomes are optimal (Dupré 1987). The GA can be thought of (=-=Bethke 1981-=-) as an optimization method which overcomes the problem of local fitness optima, to obtain optima which are almost always close to global. Moreover, following biological evolution, it treats many cand... |

29 | Using the genetic algorithm to generate Lisp source code to solve the Prisoner's dilemma - Fujiki, Dickinson - 1987 |

27 |
Optimization strategies gleaned from biological evolution
- Brady
- 1985
(Show Context)
Citation Context ... various forms of random search on more difficult (and more common) problems, such as optimizations involving discontinuous, noisy, high-dimensional, and multimodal objective functions (Holland 1984; =-=Brady 1985-=-). The GAs do not require well-behaved, convex objective functions— indeed, they do not require closed objective functions at all—which provides an opportunity for an exhaustive study of the solution ... |

26 | Evolutionary stability in extensive twoperson games - SELTEN - 1983 |

26 | On repeated games with complete information - Sorin - 1986 |

23 | Market Structure and Behavior - Shubik, Levitan - 1980 |

16 | Evolutionary games in economics, Econometrica 59 - Friedman - 1991 |

14 |
Genetic algorithms and adaptation
- Holland
- 1984
(Show Context)
Citation Context ...techniques and various forms of random search on more difficult (and more common) problems, such as optimizations involving discontinuous, noisy, high-dimensional, and multimodal objective functions (=-=Holland 1984-=-; Brady 1985). The GAs do not require well-behaved, convex objective functions— indeed, they do not require closed objective functions at all—which provides an opportunity for an exhaustive study of t... |

14 | Can bounded rationality resolve the prisoner’s dilemma?, mimeo - RADNER - 1978 |

12 | Evolutionary foundations of solution concepts for finite, twoplayer, normal-form games - Samuelson - 1988 |

11 |
A genetic algorithms tutorial
- Davis
(Show Context)
Citation Context ...looked at as a collection of substring components or schemata which together account for the good or bad performance of the individual structure. Then Holland’s Schema Sampling Theorem (Holland 1975, =-=Davis 1991-=-) demonstrates that schemata represented in the population will be sampled in future generations in relation to their observed average fitness, if we can assume that the average fitness of a schema ma... |

9 |
Implicit coalitions in a generalized prisoner's dilemma
- Fader, Hauser
- 1988
(Show Context)
Citation Context ...of the repeated game, given complete knowledge of one’s own previous moves (prices), one’s own cumulative score (total profits, undiscounted), and the previous moves and scores of both other players (=-=Fader and Hauser 1988-=-). One reason was to explore how Axelrod’s twoperson results generalize to more complex and managerially relevant situations. The model is of sales of differentiated goods. With prices of Pi, Pj, and ... |

8 | Niche strategies: the Prisoner’s Dilemma computer tournaments revisited, http://www.agsm.edu.au/!bobm/papers/niche .pdf - Marks - 1988 |

5 | Coping With Complexity: The Adaptive Value of Changing Utilities.” American Economic Review - Cohen, Axelrod - 1984 |

5 | Introduction - Rawlins - 1991 |

4 |
Genetic Algorithms, Nonlinear Dynamical Systems, and Models of International Security
- Forrest, Mayer-Kress
- 1991
(Show Context)
Citation Context ...re general terms—by Aumann (1985, pp.218–219) and by Binmore and Dasgupta (1986, pp.6–7, 12–14). Axelrod then used the GA to “breed” strategies in the two-person - 3 - repeated PD game (Axelrod 1987, =-=Forrest and Mayer-Kress 1991-=-). Axelrod reported that the GA evolved strategy populations whose median member was just as successful as Tit for Tat, whom they closely resembled. (In 95% of the time, the evolved rules make the sam... |

3 | The evolution of coöperation revisited - Nachbar - 1988 |

3 | A critical review of genetic algorithms - Schaffer, Grefenstette - 1988 |

2 | An ecological approach to economic games - Nachbar - 1990 |

2 |
Economic puzzles: noisy prisoners, Manhattan locations
- Nalebuff
- 1987
(Show Context)
Citation Context ...ler (1989) and Marks (1989) have both extended Axelrod’s recent work, and examine how the GA can be used in the breeding of strategies to such problems as the twoperson PD with uncertainty (“noise”) (=-=Nalebuff 1987-=-). This paper examines examples of oligopolistic markets, such as the three-person PDs of the price war (Fader and Hauser 1988). The advent of GAs (and machine learning) means that a much more exhaust... |

2 |
What is game theory
- Schelling
- 1967
(Show Context)
Citation Context ...ee J.W. Friedman 1983, for instance). Following Cournot, these have been in terms of the dynamic adjustment of the competitors’ behaviours, and have been facilitated by the insights from game theory (=-=Schelling 1984-=-; Ulph 1987). Strategic behaviour is important because in competition among few agents the individual agent is neither powerless (pure competition) nor powerful (monopoly), and the interaction among c... |

2 |
More Realism in the Prisoner's Dilemma
- To
- 1988
(Show Context)
Citation Context ...om distribution to the uniform coöperative behaviour of (C,C) in 61 generations. 5.2 Extended Prisoner’s Dilemma Games A more realistic PD might allow players to shade their coöperation or defection (=-=To 1988-=-) by choosing actions with payoffs between the two extremes of the simple game. For instance, the row player’s payoff matrix of Figure 2 is a superset of the simple PD payoff matrix, and allows a grea... |

1 |
A User’s Guide to GENESIS. Navy Center for Application Research
- Grefenstette
- 1987
(Show Context)
Citation Context ...either “parent.” If this structure represents a highperformance area of the search space, the evaluation of y 2 will lead to further exploration in this part of the search space. The GENESIS package (=-=Grefenstette 1987-=-), which we use, implements two crossover points per mating. A second operator is mutation: each bit in the structure has a chance of undergoing mutation, based on an interarrival interval between mut... |

1 | Finite rationalizability and coöperation in the finitely repeated Prisoner’s Dilemma - Harrington - 1987 |

1 |
Repeated games and finite automata. AGSM Working Paper 90–045
- Marks
- 1990
(Show Context)
Citation Context ...ging, forgiving, generous, etc.). As soon as a string gets much over 16 bits long, such recognition is difficult. Perhaps using constructs from the theory of finite automata will assist (Miller 1989, =-=Marks 1990-=-); perhaps there is no way to characterize a complex strategy: the only way to understand it is to watch its behaviour in a repeated game. This representation allows us to use the GA to develop what i... |

1 |
personal communication
- Midgley
- 1988
(Show Context)
Citation Context ...th more than two possible actions per player, and (c) games with more than two players. With each of these relaxations our models (slowly) come closer to modeling reality. A further relaxation might (=-=Midgley 1988-=-) be (d) partitioning the players into two or more groups, each with a distinct payoff matrix or function—this raises the interesting question of how the market system will behave with each group faci... |