Inferring supply networks from mobile phone data to estimate the resilience of a national economy

(a) Probability p(s|c) to find a supply link, given that there exists communication. (b) Cumulative distribution function for the degree. The degree distribution of the HSN is much more similar to the RSN than the HCN.

Abstract

National economies rest on networks of millions of customer-supplier relations. Some companies – in the case of their default – can trigger significant cascades of shock in the supply-chain network and are thus systemically risky. Up to now, systemic risk of individual companies was practically not quantifiable, due to the unavailability of firm-level transaction data. So far, economic shocks are typically studied in the framework of input-output analysis on the industry-level that can’t relate risk to individual firms. Exact firm-level supply networks based on tax or payment data exist only for very few countries. Here we explore to what extent telecommunication data can be used as an inexpensive, easily available, and real-time alternative to reconstruct national supply networks on the firm-level. We find that the conditional probability of correctly identifying a true customer-supplier link – given a communication link exists – is about 90%. This quality level allows us to reliably estimate a systemic risk profile of an entire country that serves as a proxy for the resilience of its economy. In particular, we are able to identify the high systemic risk companies. We find that 65 firms have the potential to trigger large cascades of disruption in production chains that could cause severe damages in the economy. We verify that the topological features of the inter-firm communication network are highly similar to national production networks with exact firm-level interactions.

Publication
Inferring supply networks from mobile phone data to estimate the resilience of a national economy

Figure description:

(a) Probability $p(s|c)$ to find a supply link, sij , given that there exists a communication link, cij, between firms i and j for communication links exceeding a given call duration, dij. Error bars denote the quartiles of a bootstrap simulation described in SI Text 1.

(b) Cumulative distribution function $p(ki > k)$ for the degree k of the reconstructed supply network RSN (blue dots), hungarian supply network HSN (orange x’s) and human communication network HCN (green pluses). The degree distribution of the HSN is much more similar to the RSN than the HCN. Errorbars denote the quartiles of a bootstrap simulation described in SI Text 1.

This publication is currently a preprint.

Georg Heiler
Georg Heiler
PhD candidate & data scientist

My research interests include large geo-spatial time and network data analytics.

comments powered by Disqus

Related