Working Paper: NBER ID: w23293
Authors: Vahid Gholampour; Eric Van Wincoop
Abstract: We use 633 days of tweets about the Euro/dollar exchange rate to determine their information content and the profitability of trading based on Twitter Sentiment. We develop a detailed lexicon used by FX traders to translate verbal tweets into positive, negative and neutral opinions. The methodologically novel aspect of our approach is the use of a model with heterogeneous private information to interpret the data from FX tweets. After estimating model parameters, we compute the Sharpe ratio from a trading strategy based on Twitter Sentiment. The Sharpe ratio outperforms that based on the well-known carry trade and is precisely estimated.
Keywords: Eurodollar; Twitter Sentiment; Exchange Rates; Trading Strategies
JEL Codes: F31; F41; G12; G14
Edges that are evidenced by causal inference methods are in orange, and the rest are in light blue.
| Cause | Effect |
|---|---|
| twitter sentiment (Z13) | predictability of exchange rate changes (F31) |
| tweets from accounts with a large number of followers (C55) | high-quality signals (L15) |
| high-quality signals (L15) | predictability of exchange rate changes (F31) |
| informed group (C92) | strong directional predictability (G17) |
| uninformed group (Y40) | lack of predictability (D80) |
| trading strategy based on twitter sentiment (G13) | Sharpe ratio (G11) |