Table 8.

Fama-MacBeth (1973) regression coefficients for daily bullish sentiment (unique cashtags)

Panel A: Twitter
	$t+1$	$t+2$	$t+3$	$t+4$
Harvard-IV	0.0010	0.0004	0.0003	0.0002
	(1.49)	(0.56)	(0.51)	(0.32)
LM	0.0028	0.0014	0.0012	0.0011
	(4.22)	(2.06)	(1.68)	(1.53)
L1	0.0014	0.0005	−0.0003	0.0003
	(1.83)	(0.63)	(−0.41)	(0.47)
L2	0.0027	0.0025	0.0019	0.0017
	(3.62)	(3.18)	(2.55)	(2.26)
VADER	0.0011	0.0007	0.0011	0.0007
	(1.50)	(1.00)	(1.53)	(0.99)
Naive Bayes	0.0031	0.0021	0.0006	0.0015
	(4.29)	(2.76)	(0.71)	(1.77)
Max. entropy	0.0025	0.0015	0.0006	0.0006
	(3.31)	(2.01)	(0.78)	(0.76)
Deep-MLSA	−0.0004	−0.0006	−0.0003	−0.0008
	(−0.68)	(−0.87)	(−0.41)	(−1.30)
DeepMoji	0.0013	0.0001	−0.0007	−0.0002
	(1.75)	(0.17)	(−0.73)	(−0.26)

Panel B: StockTwits
	$t+1$	$t+2$	$t+3$	$t+4$
Harvard-IV	0.0028	0.0019	0.0019	0.0011
	(4.44)	(2.69)	(2.42)	(1.44)
LM	0.0034	0.0025	0.0011	0.0012
	(5.42)	(4.18)	(1.50)	(1.60)
L1	0.0022	0.0015	0.0008	0.0006
	(2.79)	(1.99)	(0.96)	(0.71)
L2	0.0024	0.0016	0.0002	0.0001
	(3.37)	(2.26)	(0.35)	(0.09)
VADER	0.0025	0.0019	0.0017	0.0010
	(3.54)	(2.61)	(2.14)	(1.24)
Naive Bayes	0.0003	0.0001	−0.0001	−0.0007
	(0.34)	(0.08)	(−0.19)	(−0.98)
Max. entropy	−0.0002	−0.0002	−0.0012	−0.0007
	(−0.21)	(−0.21)	(−1.50)	(−0.96)
Deep-MLSA	0.0000	0.0003	−0.0011	−0.0015
	(−0.02)	(0.41)	(−1.76)	(−2.37)
DeepMoji	0.0019	0.0003	0.0007	0.0006
	(2.42)	(0.40)	(0.93)	(0.76)

Note: The table reports average cross-sectional regression coefficients (see Fama and MacBeth 1973) for daily bullish investor sentiment estimated from short messages published on Twitter (Panel A) and StockTwits (Panel B) mentioning a unique cashtag. The rows refer to the respective investor sentiment measure. The columns represent the dependent variable, being the h-day ahead retail investors’ order imbalance, for $h=1,\cdots ,4$. Newey-West (1987) standard errors are used to construct t-statistics, which are reported in parentheses below the respective coefficient estimate. All covariates are standardized such that the reported parameters can be interpreted as the effect of a one standard deviation change in that variable