臺大管理論叢 NTU Management Review VOL.27 NO.4

19 /274

19
臺大管理論叢 第 27 卷第 4 期
Table 4 Results of the Role of QFIIs in Earnings Informativeness of Income
Smoothing-Zero Cost Portfolio Return Examinations
Xt3_1 (low Xt3)
(n = 1,922)
Xt3_2 (middle Xt3)
(n = 1,922)
X t3_3 (high Xt3)
(n = 1,922)
Panel A: Return Difference for High Income Smoothing/High QFII Group (9) and High Income
Smoothing/Low QFII Group (7)
H_IS_H_QFII return (9) 0.0325 0.0258 0.0608
H_IS_L_QFII return (7) 0.0796 0.0495 0.0777
Zero-cost-portfolio return (9-7) -0.0471*** -0.0247* -0.0169
Panel B: Return Difference for High Income Smoothing/High QFII Group (9) and Low Income
Smoothing/Low QFII Group (1)
H_IS_H_QFII return (9) 0.0325 0.0258 0.0608
L_IS_L_QFII return (1) 0.0619 0.0286 0.0859
Zero-cost-portfolio return (9-1) -0.0294* -0.0028 -0.0251**
Panel C: Return Difference for High Income Smoothing/High QFII Group (9) and Low Income
Smoothing/High QFII Group (3)
H_IS_H_QFII return (9) 0.0325 0.0258 0.0608
L_IS_H_QFII return (3) 0.0616 0.0259 0.0645
Zero-cost-portfolio return (9-3) -0.0291* -0.0001 -0.0037
Legends:
1. The portfolio is carried out by a two-way sequential sorting. Firstly, the firms are sorted into three
groups based on earnings (X t3), which is denoted as low Xt3 , middle Xt3 , and high Xt3 groups,
respectively. Secondly, within each tercile, first sort income smoothing, then sequentially sort on
QFIIs’ participation level. This procedure results nine subsamples (3*3) from the low income
smoothing with low QFII ownership to the high income smoothing with high QFII ownership in each
earnings tercile.
2. “***”, “**”, and “*” denote the significance on 1%, 5%, and 10% levels respectively, based on one-
tailed tests.
4.3 Regression Results
The estimation process of this study begins with least-squares regression of the pooled
data followed by an assessment of the validity of the pooled model’s assumption of a single,
overall intercept term. The Lagrange Multiplier Statistic (LM test) rejects the pooled model
(which implies heterogeneous intercepts), thus panel data models, as conjectured, offer a
more powerful approach. Subsequently, the estimation proceeds to panel data analysis and a
choice between the fixed effect and random effect. The Hausman specification test
(Hausman, 1978) reveals the potential for omitted variables bias and the importance of firm-
specific effects in this setting (x 2 = 631.47 in IS model, x 2 = 649.60 in IS_QFII model, and x2
= 660.57 in IS_HH_HL model), thus we anticipate the need to use the fixed-effect approach