Reconstruction of production networks and other studies in complexity economics

Thesis

Recent research portrays the economy as a complex, adaptive system composed of heterogeneous agents interacting with one another and their environment. This thesis contributes to this research effort, focusing on production networks and financial markets.

International supply chains are a remarkable example of complex networks, and their investigation is crucial to understanding our economies. The first part of this thesis is devoted to the reconstruction and modelling of these networks. In Chapter 1, we briefly survey the literature on network reconstruction and how it has been applied to the case of production networks. In Chapter 2, we use Machine Learning to reconstruct the production network -i.e., to infer which firms are linked by commercial relationships. Using some sensible economic and financial properties as inputs, we show that our algorithm outperforms some well-known benchmarks, investigate which features are important for accurate predictions, and study to what extent our approach can be used in real-world tasks. In Chapter 3, we focus on firms' sales time series and show that the production network has a visible impact on their correlation structure. We build on this finding and develop a method to reconstruct production networks from firms' dynamics. Chapter 4 outlines an agent-based model designed to study the impact of exogenous shocks on the real production network. The primary agents of our model are firms connected by trade and credit relationships. We explain the model, provide some analytical results and simulations, and describe a simple approach to generate weighted production networks that match some aggregate (and observable) properties of global trade.

In the second part of the thesis, we pivot our focus away from production networks and onto financial markets. Chapter 5, provides empirical evidence for the Market Ecology hypothesis. This hypothesis models financial markets as ecologies where trading strategies compete to generate returns. Using a large dataset of U.S. stock prices, funds' portfolios, and funds' trading strategies, we find a correlation between wealth allocation across such strategies and market volatility, as recently proposed in simple stylized models. In Chapter 6, we analyze the cryptocurrency market through the framework of its investors' network. Our findings reveal that the structure of this network closely reflects the correlation patterns within the cryptocurrency market.

Authors: Mungo, L

• Alternative title: Reconstruction of production networks
• DOI: 10.5287/ora-m8x0rdpqx
• Type of award: DPhil
• Level of award: Doctoral
• Awarding institution: University of Oxford
• Language: English
• Keywords: network science; complexity economics
• Subjects: network science; complexity economics