NetRes seminar series

Abstract: The increase in anthropogenic footprint on both urban and non-urban settings has significant effects on wild animals’ natural behaviour. Social network analysis is an excellent tool that can help us study the effect of human pressures on wildlife social behaviour, dynamical processes and disease transmission. The reliability of estimates obtained through social network analysis, however, are susceptible to the proportion of individual sampled compared to the whole population size. This becomes problematic when studying animal social networks since is mostly unfeasible to track full populations. The aim of this study is to understand the reliability of social network metrics when using a subsample of a wildlife population, and the ability of population subsets to capture ecological responses to human disturbances similarly to the entire population. We carried out social network analysis in two wildlife populations living in very different ecological settings: fallow deer in an urban park (Phoenix Park, Dublin, Ireland) and giraffe in the Namib desert (northwest Namibia). Both populations have most of the individuals individually recognizable (i.e., ear tags and pelage patterns, respectively). We found that metrics are not reliable at depicting ecological differences of wildlife population structures when a small fraction of the entire population is monitored. These findings show how network metrics might not be a good method for estimating social network metrics from population subsets and depict how human perturbations affect animal social behaviour and disease spread. We believe new and more reliable methodologies are needed to improve management and conservation strategies accordingly.

Abstract: The study of financial bubbles is a highly controversial topics in economics and finance. Despite a large number of economic analyses, anecdotal evidence and increased theoretical attention, the quantitative monitoring and modeling of financial bubbles still miss standards broadly accepted by scholars. Our study focuses on the famous dotcom bubble that inflated financial markets during the period 1995-2000. Specifically, we investigate Nokia share ownership during the onset of the bubble and during its aftermath up to 2010 by investigating a unique database that tracks the daily financial ownership of all Finnish legal entities. We document a persistent flow of investment from foreign investors in the Nokia company during the inflation period of the bubble. This is a typical anecdotical scenario observed in the setting of financial bubbles. A second fundamental observation concerns the number of Finnish investors having an open investment position in Nokia at a given day. This number increased more than exponentially during the 1998-2000, reflecting a dramatic raise of attention at a country-wise level during bubble inflation. We exploit the unique combination of studying a multinational company that was among worldwide protagonists during dotcom bubble and a complete coverage of daily financial ownerships for all Finnish investors. The trading decision profile and the distribution of investment gains and losses was strongly inhomogeneous across different categories of investors. Financial professionals and experienced individual investors were better equipped to obtain gains during bubble inflation and limit losses when the bubble bursts. On the contrary, investors with limited financial expertise gained during bubble inflation but incurred in significant losses -- or struggled to limit them -- after the bubble burst.

Abstract:  Financial markets have undergone a deep reorganization during the last 20 years. A mixture of technological innovation and regulatory constraints has promoted the diffusion of market fragmentation and high-frequency trading. The new stock market has changed the traditional ecology of market participants and market professionals, and financial markets have evolved into complex sociotechnical institutions characterized by a great heterogeneity in the time scales of market members’ interactions that cover more than eight orders of magnitude. We analyze three different datasets for two highly studied market venues recorded in 2004 to 2006, 2010 to 2011, and 2018. Using methods of complex network theory, we show that transactions between specific couples of market members are systematically and persistently over-expressed or under-expressed. Contemporary stock markets are therefore networked markets where liquidity provision of market members has statistically detectable preferences or avoidances with respect to some market members over time with a degree of persistence that can cover several months. We show a sizable increase in both the number and persistence of networked relationships between market members in most recent years and how technological and regulatory innovations affect the networked nature of the markets. Our study also shows that the portfolio of strategic trading decisions of high-frequency traders has evolved over the years, adding to the liquidity provision other market activities that consume market liquidity.

Abstract: Due to confidentiality concerns, the interactions between the components of financial systems are frequently unknown. In the case of interbank networks, bilateral exposures are not accessible, while aggregated exposures can be derived from the balance sheets of the corresponding institutions. Our final interest is to analyze the stability and assess the related systemic risk of this network, but first, we need to reconstruct its topology from the accessible partial information. Given only the marginals and the link density, the state-of-the-art in the interbank network case is given by the fitness-induced configuration model.

Since banks in Europe preferentially lend to counterparties of the same country, we extend the state-of-the-art by imposing the BIS volumes of inter- and intra-country exposures as additional constraints to reproduce such country-specific blocks in the network. Then, we investigate the spectral properties of real-world networks and work to generate spectra that are similar to empirical ones. Indeed, the state-of-the-art overlooks the empirical feature of reciprocity (the abundance of pairs of links pointing in opposite directions in a given directed network), and this implies limitations in the reconstructed network properties, especially in the spectrum of the reconstructed adjacency matrix. Since the eigenvalues of adjacency matrices of financial networks crucially affect the properties of dynamical processes being modeled on the network (including the stability of the network under the propagation of financial distress and the resulting level of systemic risk), the idea is to introduce a new reconstruction method that can incorporate the property of reciprocity, by using only public information (such as aggregate assets and liabilities) as input. We also concentrate on the concept of reconstructability, which occurs when the constraints, reproduced on average by the state-of-the-art, are also close to their expected value in each realization in the ensemble. In particular, we investigate how the single realizations fluctuate around the constraints in the sparse regime, first in a theoretical framework and then with empirical data.

Abstract: In the recent years, the energy consumption, the transfer of resources through the international trade, the transition towards renewable energies, and the environmental sustainability appear as key factors to evaluate the resilience of energy systems. In such a complex scenario, we provide a methodological approach for analysing the network reliability and resilience by the use of a directed and weighted temporal multilayer network based on the environmentally extended input-output analysis. Nodes are connected through weighted arcs representing the embodied energy flow exchanged between sectors in possibly different economies (countries). The analysis is performed by considering different types of embodied energy based respectively on renewable and non-renewable sources. This approach allows us to unveil the role during time of sectors and economies in the system identifying key elements by the use of an extension of hub and authority centrality measures, called Multi-Dimensional HITS (MD-HITS). Finally, we evaluate central arcs in the network at each time period by proposing a novel topological indicator based on the maximum flow problem. In this way, we provide a full view of economies, sectors and connections that play a relevant role over time in the network and whose removal could heavily affect the stability of the system. The proposed approach has been tested through a numerical analysis based on the embodied energy flows among countries and sectors in the period from 1990 to 2016. Results prove that the methods are effective in catching the different patterns between renewable and non-renewable energy as well as the strategic role of sectors and countries.

Abstract: In this talk, the author will give a broad overview about mathematical models of spreading processes on network/graph structures, define certain limitations for the inverse problems and discuss process predictability. A novel framework is presented, to model continuous, discrete, and hybrid forms of (non-)Markovian susceptible-infected-recovered (SIR) stochastic processes on networks. The second part of the talk will focus on the inverse problem of inferring the true number of infected nodes of SEIR epidemic spreading under the noisy measurement procedure. Finally, predictability of short-term COVID-19 fatality forecasts of ensemble models from the Centers for Disease Control and Prevention (CDC) and the European CDC (ECDC) is analysed. 

Abstract: The structure of many complex systems can be described with a complex network (graph), that has non-trivial topological features, not purely random nor completely regular. In order to use network/graph as structural prior in different machine learning tasks, that operate in vector spaces, different approaches are taken, such as node embeddings to specific geometry, direct learning on sets or graph neural networks. Additionally, when problem of interest is formulated as dynamics, additional more complicated frameworks are needed such as Neural ODE/SDE models. In this talk, AI Pontryagin, a versatile control framework based on neural ordinary differential equations that automatically learns control signals that steer high-dimensional dynamical systems towards a desired target state within a specified time interval is presented.  Finally, certain open problems of learning complex dynamics are presented and discussed. 

Abstract: In this paper, we characterize the geography of Colombian exporting clusters and analyze how the COVID-19 crisis has affected Colombian exporters. We contribute to the industrial clusters literature by defining exporting clusters with bipartite network analysis and community detection tools. The methodology allows us to empirically detect product clusters, which are compared with an alternative definition of industrial clusters, and to consider the centrality of firms within clusters. Then, we analyze the firms' trade margins during the COVID-19 crisis to evaluate whether belonging to an exporting cluster can be a source of resilience for firms. We find that clusters do not automatically lead to higher resilience and that there are differences in how firms react to a crisis within clusters. Identifying the relevant firms' characteristics can guide policymakers to activate the mechanisms that generate resilience.

Abstract: Production networks are integral to economic dynamics, yet dis-aggregated network data on inter-firm trade is rarely collected and often proprietary. Here we situate company-level production networks within a wider space of networks that are different in nature, but similar in local connectivity structure. Through this lens, we study a regional and a national network of inferred trade relationships reconstructed from Dutch national economic statistics and re-interpret prior empirical findings. We find that company-level production networks have so-called functional local connectivity structure, as previously identified in protein-protein interaction (PPI) networks. Functional networks are distinctive in their over-representation of closed squares, which we quantify using an existing measure called spectral bipartivity. Shared local connectivity structure lets us ferry insights between domains. PPI networks are shaped by complementarity, rather than homophily, and we use multi-layer directed configuration models to show that this principle explains the emergence of functional structure in production networks. Companies are especially similar to their close competitors, not to their trading partners. Our findings have practical implications for the analysis of production networks and give us precise terms for the local structural features that may be key to understanding their routine function, failure, and growth.