Quantitative Analysis of Bitcoin Transferred in Bitcoin Exchange

In this paper, a new Bitcoin address clustering algorithm is proposed for Bitcoin exchanges. The proposed algorithm aims to classify the cold wallets, hot wallets and user wallets in the Bitcoin exchanges, which are verified by off-chain information from the Internet. By analyzing the structures of different Bitcoin exchanges, we find that most Bitcoin exchanges exist weakness in managing the reasonable amounts of bitcoins kept in hot wallets. A large amount of Bitcoins stored in hot wallets can meet the users’ withdrawal demands but may increase the risk of attack. However, a small amount of bitcoins in hot wallet may be inconvenient to frequently transfer for cold wallet. The problem then is modeled as a Bitcoin withdrawal prediction problem for hot wallets. We adopt traditional and classific supervised learning methods to solve the problems. Numerical experiments show that the proposed approach provides reasonable prediction results. Furthermore, we simulate two processes to analyze our results. The first process shows that Facebook-prophet outperforms other methods if there is no transaction occurred from cold wallets to hot wallet. The second process shows that the more transactions from cold wallets to hot wallets, the smaller Bitcoins required for hot wallets. Overall, our work is valuable and useful for the Bitcoin exchanges’ business.