| When neoclassic economists analyze determinable factors of many macroeconomic variables like gross product, price level and unemployment, there are always two hypotheses: the one is that agents in the economic system are perfect rational and they have rational expectation to pursuit maximum utility; another one is that economy converges to a equilibrium. We deduce our macroeconomic system based on those hypotheses. Under restraint of computational ability, traditional economic models have been fully simplified, but all these models are different from real economy.Inspired by theory of evolution of Darwinian, Holland developed GA in 1975. Basic GA include: (1) representative of agent (or its potential behavior) by strings; (2) fitness function to evaluating adaptation of strings; (3) GA operators to generating new strings. All strings in environment constitute a population. GA applies operators to generate new population from the old and the new one can be more adaptation to environment. Basic GA operators are reproduction, crossover and mutation, the extend GA version adds election.Economic meanings of GA include: (1) basic components of GA (strings) can be explained as agents or behaviors of agent, population composed by all strings constitute a miniature society; (2) fitness function can be interpreted naturally as selection mechanism, different performance receive different payment; (3) reproduction represent imitation behaviors of agents, strings with high fitness value indicate that agent have higher fitness value in present condition and believes of those agents will be copied by many other agents; strings with low fitness value means that agent is not adapt to environment and has few copies (or no copy) in new generation; (4) crossover can be explained as reconstruction of behaviors or exchange of information, agents adopt part behavior pattern of other agents; (5) random innovation or random mistake can explain mutation; (6) election examines new created behavior that will implement in new generation only when it behaves better than old one; (7) meta mutation represents a trend of innovation of agents, the trend of agents' innovation have negative correlation with behavior inertia; (8) and process of GA running is similar to coevaluation of economy.On the basis of research of foreign scholar, this paper constructs a CobWeb model with GA to study equilibrium learning. Rational expectation considers the CobWeb model has unique equilibrium that determined by parameters of model. Natural expectation concludes that when the CobWeb is steady, model converges to rational expectation equilibrium and when the CobWeb is unsteady, model doesn't have any equilibrium; least square expectation has similar conclusion. Lucas suggests that human subject experiment is a important standard to examine varies learning algorithm. 1989, Wellford conducted human subject of CobWeb, and results are (1) whether CobWeb is steady or unsteady, series of price converges to rational expectation equilibrium and (2) in the process of convergence, price fluctuates with the equilibrium and (3) fluctuation of price with the equilibrium under steady CobWeb is small than unsteady one. The results of this paper are CobWeb model with GA captures the features of human subject experiment, and GA can describe the formation and transmission of equilibrium exactly.With a overlapping generations economy with single money based on GA, which has two monetary policies, this paper studies the selection of equilibrium. Under fixed money supply, GA model has different results from rational expectation model, but the results coincide with human subject experiment. Similarly, under fixed deficit policy, the results of GA model are in accordance with human subject experiment. Results of this paper point that among the learning of rational expectation, lease square and GA expectation, GA model captures the features of human subject better, and illustrates fluctuation and convergence process of equilibrium selection exactly.From the l...
|
PDF Full Text Request