Table 13.
Annualized portfolios returns based on daily bullish sentiment (unique cashtags)
| Panel A: Twitter | ||||||
|---|---|---|---|---|---|---|
| Raw return | Risk-adjusted return | |||||
| Short | Long | Long-Short | Short | Long | Long-Short | |
| Harvard-IV | 12.90 | 14.35 | 1.45 | −1.17 | 0.47 | 1.49 |
| (2.61) | (3.10) | (0.82) | (−0.91) | (0.35) | (0.86) | |
| LM | 10.87 | 15.13 | 4.26 | −2.99 | 1.08 | 3.92 |
| (2.13) | (3.21) | (2.27) | (−2.06) | (0.82) | (2.24) | |
| L1 | 11.77 | 15.73 | 3.97 | −2.31 | 1.72 | 3.88 |
| (2.42) | (3.55) | (2.07) | (−1.55) | (1.19) | (2.10) | |
| L2 | 11.92 | 17.81 | 5.89 | −2.15 | 3.92 | 5.92 |
| (2.37) | (3.72) | (2.81) | (−1.56) | (2.74) | (3.04) | |
| VADER | 13.19 | 15.91 | 2.72 | −0.59 | 1.99 | 2.42 |
| (2.66) | (3.36) | (1.49) | (−0.39) | (1.51) | (1.33) | |
| Naive Bayes | 13.01 | 16.51 | 3.50 | −1.11 | 2.55 | 3.51 |
| (2.67) | (3.68) | (1.86) | (−0.79) | (1.64) | (1.89) | |
| Max. entropy | 13.31 | 16.49 | 3.18 | −0.81 | 2.51 | 3.17 |
| (2.67) | (3.69) | (1.64) | (−0.59) | (1.57) | (1.69) | |
| Deep-MLSA | 10.36 | 15.40 | 5.04 | −3.64 | 1.42 | 4.91 |
| (2.14) | (3.17) | (2.74) | (−2.72) | (1.14) | (2.81) | |
| DeepMoji | 12.04 | 17.02 | 4.98 | −1.85 | 2.70 | 4.40 |
| (2.39) | (3.60) | (2.39) | (−1.21) | (1.70) | (2.18) | |
| Panel B: StockTwits | ||||||
|---|---|---|---|---|---|---|
| Raw return | Risk-adjusted return | |||||
| Short | Long | Long-Short | Short | Long | Long-Short | |
| Harvard-IV | 12.30 | 16.04 | 3.74 | −1.62 | 2.04 | 3.51 |
| (2.56) | (3.43) | (2.42) | (−1.34) | (1.61) | (2.23) | |
| LM | 11.50 | 13.97 | 2.47 | −2.51 | −0.13 | 2.22 |
| (2.25) | (2.96) | (1.35) | (−1.92) | (−0.10) | (1.27) | |
| L1 | 10.47 | 14.79 | 4.32 | −3.77 | 1.11 | 4.74 |
| (2.23) | (3.16) | (2.40) | (−3.12) | (0.78) | (2.89) | |
| L2 | 11.49 | 18.25 | 6.77 | −2.52 | 4.52 | 6.89 |
| (2.37) | (3.77) | (3.67) | (−1.94) | (2.96) | (3.72) | |
| VADER | 12.60 | 16.73 | 4.12 | −1.23 | 2.63 | 3.71 |
| (2.55) | (3.64) | (2.45) | (−1.01) | (2.10) | (2.22) | |
| Naive Bayes | 12.80 | 15.51 | 2.71 | −1.41 | 1.86 | 3.12 |
| (2.73) | (3.35) | (1.60) | (−1.20) | (1.22) | (1.83) | |
| Max. entropy | 12.59 | 15.35 | 2.76 | −1.57 | 1.75 | 3.17 |
| (2.65) | (3.36) | (1.64) | (−1.31) | (1.18) | (1.82) | |
| Deep-MLSA | 9.43 | 15.15 | 5.71 | −4.90 | 1.22 | 5.97 |
| (1.81) | (3.30) | (2.62) | (−3.19) | (0.82) | (3.05) | |
| DeepMoji | 10.96 | 16.51 | 5.55 | −3.23 | 2.58 | 5.66 |
| (2.32) | (3.53) | (3.06) | (−2.39) | (1.72) | (3.01) | |
Note: The table depicts the annualized raw and risk-adjusted returns (in %) of three portfolios based on daily bullish sentiment estimated from short messages published on Twitter (Panel A) and StockTwits (Panel B) mentioning a unique cashtag. The first portfolio (Short) contains the stocks for which the estimated online investor sentiment on the previous (trading) day is smaller than the 10% cross-sectional quantile. Conversely, the second portfolio (Long) contains the stocks for which the estimated online investor sentiment on the previous trading day is larger than the 90% cross-sectional quantile. A raw long-short portfolio return (Long-Short) is obtained as the difference between these two raw portfolio returns. The stocks are held in the portfolio for 1 trading day. The risk-adjusted returns are defined as the intercept of the regression of portfolio returns on the three risk factors introduced by Fama and French (1993) and the momentum factor of Carhart (1997). The t-statistics, reported in parenthesis, are constructed using Newey-West (1987) standard errors. Returns which are statistically significant at the 5% level are highlighted by boldfaced numbers