Impact Of Buyer Search Costs On Sellers Strategies: Simulation Of An Internet Agent-Based Market

During the last years, the remarkable growth of the Internet, caused an increasingly demand for more advanced tools, capable to assist netsurfers in their search for useful information. In the field of electronic commerce, the development of buying agents promise a great future to consumers which hope for better products at lower prices. At the present time, most of these new electronic intermediaries whose role is to interpret and synthesize information on behalf of the users, offer their services for free. In the future, as the agents' performances will be improved, some of the agent companies will ask a price for the services provided by their agents. And even if some other companies will offer free use of agents, the user will most probably have to pay the communications costs of her agent. In any case, a buying agent is expected to cause to her user some search costs for each negotiation process with a seller. Even if this cost will be low compared to the price of the product to be commercialized, it is clear that, in the long run, if the buying agent is to maximize her user's utility, she should consider the incurred search cost in her decisions.At our knowledge, the research available on agent markets considered search costs negligible in the analysis of market evolution. In the present paper, we generalized a market model for electronic agents proposed by Greenwald & Kephart [1] by explicitly introducing search costs for buyers. Thus, we modeled a commodity market where buyers can learn the price of a seller by paying a fixed fee. A lot of research in economics [2][3] was concerned with game theoretic equilibrium of markets where buyers' search of information is costly. We completed this theoretical approach by studying empirically the dynamics of such markets. In this article, we simulated a market and analyzed the possible strategic behavior of sellers when they are confronted with searching buyers.In our market, we considered that buyers use either a fixed-sample-size rule (more precisely, the buyer commits a priori to visiting a number of sellers, then buys for the lowest price if lower than her reservation value) or a reservation price rule (the buyer keeps searching if the expected gain of an additional search is higher than the search cost). As classical models are based on a probability distribution of the prices on the market which reflects buyer beliefs, our agent needed to be endowed with some sort of beliefs. In our simulations the considered beliefs were of two kinds: either the agent used the theoretical probability distribution of the prices at the equilibrium obtained by solving the game or the agent was endowed with the ability to learn provided that some information about the past evolution of the market is revealed to the agent. Moreover, we considered that seller agents, confronted with this buyer population, will dynamically set the prices using one of the following four strategies : game-theoretic, myopically optimal, derivative-following and reinforcement learning. In this setting our work was mainly concerned by two questions : What type of search rule is more efficient from the buyer's point of view? and What type of price- setting algorithm performs better on a market where buyers use different search rules? The experimentation confirmed some expected theoretical results, namely that higher search costs allow for higher sellers' profits and that when search of information is costly, buyers will pay a higher price for the product. However, while economic theory considers reservation-price rules as optimal, our simulation showed that, in some market structures, buyers using fixed-sample rules performs better than those using reservation price rules. The main reason for this result seams to be that the optimality property is highly dependent of the accuracy of buyer's beliefs relative to the market structure. Consequently, the learning ability of the agent is extremely important. From the sellers point of view, the strategy which allows them to win more money is derivative-following strategy. This strategy sustains a tacit collusion between sellers independently of the buyer search rule.However the derivative-following strategy is stable only in homogenous settings (all sellers use the same strategy). Otherwise, some seller has an incentive to deviate to myopically-optimal strategy and the market evolves to the theoretical equilibrium. The most stable market is obtain when buyers use the game theoretical distribution and sellers use myopically-optimal strategies. This stability is due to the compatible belief system of agents.