Table 12.

Differences in risk-adjusted annualized returns of long-short portfolios

Panel A: Twitter
	(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)	(9)
Harvard-IV (1)	–	−0.97	−1.08	−2.09	−0.31	−0.36	−0.38	−1.28	−0.69
LM (2)		–	−0.04	−1.04	0.78	0.59	0.59	−0.32	0.29
L1 (3)			–	−1.33	0.80	0.76	0.72	−0.27	0.39
L2 (4)				–	1.86	1.85	1.80	0.69	1.52
VADER (5)					–	−0.09	−0.09	−1.01	−0.48
Naive Bayes (6)						–	0.00	−0.95	−0.36
Max. entropy (7)							–	−0.97	−0.35
Deep-MLSA (8)								–	0.58
DeepMoji (9)									–

Panel B: StockTwits
	(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)	(9)
Harvard-IV (1)	–	−1.68	−1.60	−3.24	−1.84	−1.61	−1.25	−2.42	−1.86
LM (2)		–	0.08	−1.40	−0.01	−0.01	0.36	−0.89	−0.15
L1 (3)			–	−1.77	−0.10	−0.09	0.34	−0.90	−0.26
L2 (4)				–	1.43	1.69	2.08	0.60	1.52
VADER (5)					–	0.01	0.43	−0.81	−0.17
Naive Bayes (6)						–	0.57	−0.79	−0.20
Max. entropy (7)							–	−1.13	−0.63
Deep-MLSA (8)								–	0.69
DeepMoji (9)									–

Note: The table depicts the t-statistics of the average differences in risk-adjusted returns of long-short portfolios based on daily bullish sentiment estimated from short messages published on Twitter (Panel A) and StockTwits (Panel B). More precisely, we take the difference between returns obtained from methodologies reported in the rows with those reported in columns. Risk-adjusted returns are defined as the intercept plus the residuals of the regression in Equation (5). The t-statistics are constructed using Newey-West (1987) standard errors. Differences which are statistically significant at the 5% level are highlighted by boldfaced numbers