In this paper we present the results of experiments and computational analyses of trading in decentralized markets with asymmetric information.
We consider three trading configurations, namely the ring, the small-world, and the Erdos-Renyi random network, which allow us to introduce heterogeneity in the degree, centrality and clustering of nodes, while keeping the number of possible trading relationships fixed. We analyze how the prices of a traded risky asset and the profits of differently informed traders are affected by the distribution of the trading link, and by the location of the traders in the network. This allows us to infer key features in the dynamics of learning and information diffusion through the market. Experimental results show that learning is enhanced by clustering rather than degree, pointing to a learning dynamic driven by interdependent, successive trading events, rather than independent exposures to informed traders. By calibrating a behavioural agent-based model to the experimental data, we are able to estimate the speed at which agents learn, and to locate where information accumulates in the market. Interestingly, simulations indicate that proximity to the insiders leads to more information in regular networks but not so in random networks.