The broad content in the study of complex networks has attracted increasing interest from various scientific communities over the past decade. Its major appealing features lie in the ubiquity of networks in nature and human society, as well as their uniform mathematical description within the unified framework of graph theory. Under such a wider framework, this dissertation focuses on the application of concepts and techniques from network science to problems in international trade and finance, which can broadly be associated with macroeconomics. A main idea is the interpretation of a datadriven approach to networked structures as an extension of the classic statistical analysis, which allows for a self-contained and consistent treatment of higher-order interactions within a complex system. Three major accomplishments will be presented. In the first part, the framework of composite centrality will be introduced, dedicating to the problem of heterogeneities in complex datasets, which is a general phenomenon when investigating real-world networks. A measure standardisation recipe is given, which transforms a unimodal distribution to an approximative standard normal distribution. This can be seen as an ideal starting point for further analyses. The composite centrality framework is not bound to the analysis of complex networks, but offers a general approach to the investigation of multi-variate statistics. In the second part, the analysis of a pair of cross-border portfolio investment networks is studied, covering a time period of the recent global financial crisis 2007-2009. It will be shown that the network characteristics of these large-scale macroeconomic structures can be used to describe aggregated properties of the global financial system. Particularly, an early-warning mechanism is developed for predicting future financial crises, which couples the output of a simple phenomenological model with a macroeconomic reference variable to indirectly measure the interdependence of financial markets, caused through the proliferation of over-the-counter traded financial derivative products. The third part introduces a novel way for the generation of complex networks from data in the form of a generic linear network model. The underlying idea is that related indicators within a dataset should be able to mutually describe each other. An iterative algorithm is designed for constructing a multiple linear regression fit network model from the available data, which is interpreted as a hybrid object, allowing for the application of basic concepts from statistical analysis, such as description and forecast, and from network science, like the analysis of the organisational structure of a given dataset. In summary, the main contributions of this Thesis include: 1. The development of a novel framework for multi-variate statistical data analyses. 2. The design and implementations of several network-based data mining approaches. 3. The investigation of cross-border portfolio investment networks and the identification of early-warning indicators for financial crises. 4. The development of a novel methodology for the generation of relational networks from data. 5. A consistent investigation of heterogeneous multi-layered networks in international trade and finance.