Assembling real networks from synthetic and unstructured subsets: the corporate reporting case.

The analysis of interfirm business transaction networks provides invaluable insight into the trading dynamics and economic structure of countries. However, there is a general scarcity of data available recording real, accurate and extensive information for these types of networks. As a result, and in common with other types of network studies - such as protein interactions for instance - research tends to rely on partial and incomplete datasets, i.e. subsets, with less certain conclusions. Here, we make use of unstructured financial and corporate reporting data in Japan as the base source to construct a financial reporting network, which is then compared and contrasted to the wider real business transaction network. The comparative analysis between these two rich datasets - the proxy, partially derived network and the real, complete network at macro as well as local structural levels - provides an enhanced understanding of the non trivial relationships between partial sampled subsets and fully formed networks. Furthermore, we present an elemental agent based pruning algorithm that reconciles and preserves key structural differences between these two networks, which may serve as an embryonic generic framework of potentially wider use to network research, enabling enhanced extrapolation of conclusions from partial data or subsets.