A Novel Method for Bitcoin Price Manipulation Identification Based on Graph Representation Learning

Abstract

Bitcoin is a cryptocurrency designed based on the concept of “decentralization,” and its price fluctuation is much larger than those of traditional financial assets, which has raised concerns about intentioned Bitcoin price manipulation. Most of the existing research on Bitcoin price manipulation is limited to the analysis of manipulation behaviors, lacking effective identification methods and performance metrics of identification methods. In this article, we use MtGox real trading data to build a trading network and analyze the diverse price manipulation patterns in the trading network. Based on this, we improve the classical graph representation learning method to effectively identify the price manipulation accounts. Specifically, considering that Bitcoin has special financial investment properties, and its price condenses important information from the financial market, we propose a novel identification algorithm, called Finan2vec, by feeding the time-series information of the price into the transfer strategy of the Node2vec algorithm. This algorithm makes it easier to detect nodes with large trades or significant price-raising behavior. We then use two classification algorithms to complete the identification of abnormal nodes and finally draw on the knowledge of financial market experts to construct a hit rate indicator to measure the effectiveness of the proposed method. The experimental results show that our method can effectively identify price manipulation accounts for Bitcoin and other public blockchain systems.