Title: Cryptocurrency-specific lexicon and sentiment projection
Authors: Thomas Renault - Université Paris 1 Panthéon-Sorbonne (France) [presenting]
Abstract: A non-classical asset, cryptocurrencies, draws great attention to a particular subset of investors who possess higher risk preference in order to ride the trend. Due to limited knowledge for its fundamental value, investor opinions (sentiment) in this digital asset class could convey incremental information in terms of price discovery. Using a novel dataset of 1069K messages related to 375 cryptocurrencies posted on the microblogging platform Stocktwits during a 4-year period, we construct a ``crypto-specific lexicon'' in order to precisely capture the semantic orientations and avoid misspecification. The results show that in comparison with the financial lexicon used in the literature, the crypto-specific lexicon achieves 32\% higher accuracy in terms of out-of-sample classification. The opinions quantified through crypto-specific lexicon strikingly drive the movement of cryptocurrency market. A time series return predictability further suggests our tone measure positively predicts Top 200 cryptocurrency index up to 3 days without return reversal.