Machine-learning forecasting of successful ICOs

Abstract

The identification of ventures that are more likely to be successful is a complex task for equity investors, such that a mix of assessment criteria is typically employed. Machine learning (ML) techniques may provide valid support to investors in processing the set of available information. This paper applies ML to initial coin offerings (ICOs), which allows firms to raise blockchain finance through public online offerings. After implementing this novel approach to a sample of 383 ICOs launched between August 2014 and December 2019, we found an increase in forecast accuracy, from 54.3% when using standard Logit models to 72.8% when using ML techniques. Therefore, we contribute to the scientific debate and practice by introducing an algorithm-based approach that makes the analysis of whitepaper content available to individual investors. Importantly, we document that the structure of ICO white papers, a so far neglected dimension, is a significant determinant of the success of ICOs.