Partial information, market efficiency, and anomalous continuous phase transition

It is a common belief in economics and social science that if there is more information available for agents to gather in a human system, the system can become more efficient. The belief can be easily understood according to the well-known efficient market hypothesis. In this work, we attempt to challenge this belief by investigating a complex adaptive system, which is modeled by a market-directed resource-allocation game with a directed random network. We conduct a series of controlled human experiments in the laboratory to show the reliability of the model design. As a result, we find that even under a small information concentration, the system can still almost reach the optimal (balanced)state. Furthermore, the ensemble average of the system's fluctuation level goes through a continuous phase transition. This behavior means that in the second phase if too much information is shared among agents, the system's stability will be harmed instead, which differs from the belief mentioned above. Also, at the transition point, the ensemble fluctuations of the fluctuation level remain at a low value. This phenomenon is in contrast to the textbook knowledge about continuous phase transitions in traditional physical systems, namely, fluctuations will rise abnormally around a transition point since the correlation length becomes infinite. Thus, this work is of potential value to a variety of fields, such as physics, economics, complexity science, and artificial intelligence.