Machine Forecast Disagreement in the Cryptocurrency Market

Gang Chu, Dehua Shen, and Zhaobo Zhu

♦ This paper uses machine learning models to construct a crypto-level measure of simulated investor belief disagreement (MFD) for cryptocurrencies. Simulated investors use their individual models and common data information to form their individual return forecasts. Consistent with Miller (1977), we find a significantly negative relation between MFD and future cryptocurrency returns in the cross section. Past return variables are the top drivers of MFD, suggesting that disagreement is strongly associated with information contained in past returns. Moreover, the negative MFD-return relation is stronger for cryptocurrencies with larger limits-to-arbitrage or more severe overpricing, highlighting the role of mispricing and limits-to-arbitrage.

Read it here.