Analysis of multi-input multi-output transactions in the Bitcoin network

• Main Authors: Phetsouvanh, Silivanxay, Datta, Anwitaman, Oggier, Frédérique
• Other Authors: School of Physical and Mathematical Sciences
• Format: Article
• Language: English
• Published: 2022
• Subjects: Science::Mathematics; Bitcoin Network; Graph Analysis; Mixing
• Online Access: https://hdl.handle.net/10356/154816
• Institution: Nanyang Technological University

Distinct transactions among different and unrelated users are combined together to create a single Bitcoin transaction (mixing transaction) to obfuscate the relationships among the actual *Anwitaman Datta Email: participants (more specifically, the wallet addresses used for the transactions). We consider multi- [e-mail] input multi-output transactions with at least two inputs and three outputs as proxy, to analyze Present Address four characteristic periods of ∼50 days each, representing periods before the introduction of School of Computer Science and Engineering mixing, in its early days, during its growth, and after the volume of such multi-input multi-output Blk N4 (office: 02A-18), North Spine Nanyang Avenue, Nanyang Technological University Singapore 639798 transactions became more or less stabile. Structural properties and characteristics of the transac- tion and wallet address networks are computed and compared, through standard tools, but also via the introduction of two novel techniques that provide indicators of mixing-like behaviors: (i) an entropy characterization to detect abnormally uniform inputs and/or outputs, and (ii) a con- nected component analysis of subgraphs formed by only multi-input multi-output transactions (showing cascades of such transactions). The contributions of this exploratory Bitcoin network analysis paper can thus be seen as two-fold. At a macroscopic level, the growth and stabilization periods are shown to stand out with respect to most considered metrics, while at a microscopic level, chains of multi-input multi-output transactions, and transactions with outlier behavior in terms of input/output entropies are identified for further investigation.