The mutual predictability of Bitcoin and web search dynamics

Abstract

Economic theory predicts the price dynamics of an unbacked asset to be inherently unforecastable. The same applies to exchange rates of unbacked currencies. Albeit, empirically investors are found to be driven by online and offline news media. This study analyzes the Bitcoin cryptocurrency price series and web search queries with regard to their mutual predictability and cause‐effect delay structure. Chinese Baidu engine searches and compounded Baidu–Google search statistics predict Bitcoin price dynamics at relatively high frequencies ranging from 2 to 5 months. In the other direction, Granger‐causality runs from the cryptocurrency price to queries statistics across nearly all frequencies. In both directions, the reaction time computed from a phase delay measure for the relevant frequency bands with significant causality ranges from about 1 to 4 months. For either direction, out‐of‐sample forecasts are more accurate than forecasts of a benchmark stochastic process. Bivariate models including the Baidu Search Index slightly outperform competing models that include a Baidu–Google composite index. Predictive power seems less diluted if the September 2017 trade regulations by the Chinese government are controlled for.

Forecasting and trading Bitcoin with machine learning techniques and a hybrid volatility/sentiment leverage

“…Zhang et al (2021) suggest that BTC returns can be used to predict BTC volatility by using aggregation of intraday information. In addition, Süssmuth (2022) provides evidence that Baidu–Google search statistics forecast BTC price dynamics at relatively high frequencies. Milunovich and Lee (2022) show that advanced ML techniques have high accuracy in predicting cryptocurrency exchange activity.…”

“…Yao et al (2019) find that news articles can influence BTC prices at a certain level. Azqueta‐Gavaldón (2020) further find bidirectional causal relationships between narrative sentiment and BTC prices by applying a dynamic system model, whereas Süssmuth (2022) explains that mutual causality exists between web search dynamic and BTC prices before 2018. López‐Cabarcos et al (2021) provide a similar finding that the BTC market is connected with the investor sentiment and this relationship will be more significant in the stable period.…”