A Study of Users, Real Cash Flows And Temporal Activity In The Bitcoin Ecosystem

Abstract

Bitcoin is the oldest cryptocurrency, and among the most active ones. All its transaction data is stored in a decentralized ledger - the Bitcoin blockchain - freely accessible to anyone willing to analyze it. Analyzing the content of this data is the purpose of this thesis. The manuscript focuses on two main research questions: the identification of Bitcoin users, and the characterisation of the activity of those users. In the first part, we propose a method for improving the construction of aggregates of Bitcoin addresses belonging to the same user, by identifying the change output of a transaction using supervised machine learning. The quality of the result is evaluated using a ground truth based on on-chain and off-chain data. We show that the results outperform previous work, but also that identifying the change output of a single user might be a better strategy than the usual objective of considering the whole blockchain as a single problem. The second part of the work focus on interpreting the users' activity in the Bitcoin blockchain. It particularly focuses on defining the real economic activity present in the Bitcoin blockchain, as opposed to artificial transactions driven by the protocol or by users moving money from address to address for technical reasons. Heuristics are proposed aiming to classify users in three categories: Frequent Receivers (FR), Neighbors of FR, and Others. The work shows that FR (being a proxy for commercial entities) represent a small fraction of entities, but concentrate most of the payments, showing a centralization in the bitcoin ecosystem. A temporal study is also conducted, allowing us to estimate the geographical location of users. We notably use this information to quantify the bias of a dataset commonly used in the literature for entity tagging.