2.2.7 'MENTORING' Y 'COACHING': ¿IGUALES O DIFERENTES?
5. Fomentar la inclusión de tecnologías de la información y la comunicación (TIC) en la práctica cotidiana de los maestros para usos académicos y administrativos.
2.2.14. PERFIL DE LA FORMACION DOCENTE EN EL PERU.
The unob served idiosyncratic firm features captured by αican be wiped out with panel data by running the first difference equation:
' '
, , 1 ( , 1 , 2) ( , , 1) , , 1
i t i t i t i t i t i t i t i t
PRE −PRE − =γ PRE − −PRE − +β X −X − +ε −ε − (4.7) In equation (4.7) , the difference of the lagged termPREi t, 1− −PREi t,−2 is correlated
with the difference of the disturba nce εi t, −εi t, 1− , i.e., E dPRE( i t, 1−dεit)≠0 . The
conventional ordinary least squares (OLS) estimation is no longer appropriate as
biased and inconsistent estimator will result.
Anderson and Hsiao (1982) suggested using the level PREi,t-2,or the lagged difference
dPREi,t-2, as instruments for the differenced lagged endogenous regressor dPREi,t-1.
Assuming there is no serial correlation, these instruments can be expected to be
uncorrelated with the differenced error term:
, 2
( i t it) 0
E PRE − dε = and E dPRE( i t,−2dεit)=0 (4.8)
method-of-moments (GMM) procedure is proposed by Arellano and Bond (1991), and
then further generalized and extended by Ahn and Schmidt (1995), Arellano and Bover (1995) and Blundell and Bond (1998) just to mention a few. Arellano and Bond
(1991) argue that additional instruments can be obtained in a dynamic panel model if
one utilizes the orthogonality cond itions that exist between lagged values of
dependent variable and the disturbances. Following this, values of PRE lagged one
period or more qualify as instruments in the first-differenced system, implying the following moment conditions:
, ,
( i t s i t) 0
E PRE − dε = , t=3,…,T; s>=2 (4.9)
GMM estimation based on (4.9) alone could be highly inefficient. In most cases, it is
necessary to make use of the explanatory variables as additional instruments. Under
the assumption that X’i,t are exogenous variables, E X( 'i t, εi s, )=0for all t, s=1,2,…,T, the explanator y variables can also be considered as valid instruments:
, ,
( 'i t s i t) 0
E X − dε = , t=3,…,T; s>= 1 (4.10)
However when the va lue of γ increases towards unity, and as the relative variance of the fixed effects αi increases, the instruments used in the standa rd first-differenced GMM estimator become less informative. Blundell and Bond (1998) derived a
consistent estimator for this problem by allowing use of an extended system GMM
estimator. In addition to lagged levels of dependent variable as instruments for
equations in first differences, they use lagged differences of dependent variable as
, 1 ,
( i t i t) 0
E dPRE −ε = , for t=3,…,T (4.11)
When X’ are exoge nous, the follow ing level moment conditions can also be used as additional instruments:
, ,
( 'i t s i t) 0
E dX − ε = , t=3,…,T; s>=1 (4.12)
This technique is especially designed for situations with “small T, large N” panels,
meaning few time periods and many individuals.
The consistency of the GMM estimator s depe nds on the validity of two assumpt ions:
firstly, the error terms are assumed not exhibit serial correlation, and secondly, the
instruments are not correlated with error terms. To address these issues, two
specification tests suggested by Arellano and Bond (1991), Arellano and Bover (1995)
and Blunde ll and Bond (1998) are used here. The overall validity of the moment
conditions is checked by the Sargan test8
8
For more details, p lease see Baltagi, Econom etric Ana lysis of Pan el Data, 3rd Edition, p.141.
. The null hypothesis of the instrumental
variables are uncorrelated to some set of residuals is rejected if the Sargan test statistic
registers a large value compared with a chi-squared distribution with the degree of
freedom equals to the difference between the number of moment conditions and
number of parameters. To check the serial correlation property of the level residuals, the Arellano-Bond m1 and m2 statistics are calculated. If the level residuals were
indeed serially unc orrelated, then, by construction, the first-differenced residuals in
(4.7) would follow a MA(1) process which implies that autocorrelations of the first
order are non-zero but the second or higher order ones are zero. Based on the
1) in large samples, test the null hypotheses of zero first order and second order
autocorrelation, respectively. A signi ficant m1 and insignificant m2 would suggest valid moment conditions.
The dynamic panel regression results, utilizing both the GMM-difference estimator
and GMM-system estimator, of the Shangha i- and Shenzhen- listed firms are reported
in Tables 4.5 and 4.6 respectively. Models A and B are based on the GMM-difference
estimators and models C and D are based on the GMM-system estimators. In mode l A, the instrument set is equation (4.9), that is values of PRE lagged one period or more
are used as instruments in the first-differenced system; in mod el B, instrumental set
(4.10) is further added to the instrumental set (4.9); in model C, (4.11) is augmented
with (4.9)-(4.10), that is lagged differences of PRE are used as add itional instruments
for equations in levels in add ition to the instruments for equations in first differences; in model D, extra instrumental variables, (4.12), lagged differences of explanatory
variables are added to the instrumental set of mode l C. Since I only have a short T,
i.e., T=11, which is not sufficient to conduct the analysis in sub-periods to take into
account of the po licy changes, I will run the dynamic panel regressions over the ent ire
sample period to get a evidence of how the firm fundamentals attribute to the A- shares premium on the SHSE and SZSE respectively. The tests results are reported in
Tables 4.4 and 4.5.
Table 4.4: Dynamic Panel Regressions of Shanghai-Listed Firms
GMM-DIFF GMM-SYS
l.PREM 0.2086 0.2811 0.1755 0.2215 0.4067 0.3256 0.3411 0.2491 (0.287) (0.148) (0.212) (0.111) (0.000)*** (0.005)*** (0.000)*** (0.009)*** DTE 0.0158 0.0388 0.0089 0.0191 0.0173 0.0102 0.0040 0.0028 (0.646) (0.257) (0.533) (0.180) (0.169) (0.413) (0.304) (0.452) EPS 11.6732 9.2807 10.8950 4.5374 8.4752 6.5915 3.8519 2.4612 (0.375) (0.498) (0.091)* (0.490) (0.195) (0.302) (0.391) (0.568) ROE -0.2528 -0.0702 -0.2181 -0.0544 -0.1364 -0.1300 -0.0897 -0.0852 (0.240) (0.761) (0.046)** (0.625) (0.175) (0.183) (0.204) (0.208) FEG 0.6033 0.6358 0.2130 0.2458 0.2512 0.2272 0.1666 0.1493 (0.013)** (0.011)** (0.062)* (0.030)** (0.036)** (0.051)* (0.068)* (0.088)* FSG 1.8894 3.3845 0.6987 1.6272 1.2797 1.0902 1.4346 1.1022 (0.294) (0.053)* (0.323) (0.018)** (0.065)* (0.108) (0.006)*** (0.029)** DIV -0.0396 -0.0395 -0.0303 -0.0247 -0.0208 -0.0236 -0.0207 -0.0243 (0.185) (0.198) (0.012)** (0.042)** (0.103) (0.056)* (0.029)** (0.008)*** FR 0.3834 -17.9195 3.2923 -13.6204 -2.1802 -5.7103 -3.4687 -5.0374 (0.981) (0.196) (0.714) (0.095)* (0.723) (0.311) (0.291) (0.088)* LMCAP1 -2.3771 -2.3402 0.0165 0.2609 (0.140) (0.005)*** (0.943) (0.073)* LMCAP2 1.5518 2.0521 0.2415 0.3776 (0.406) (0.024)** (0.195) (0.001)*** LIQ 0.1948 0.2918 0.0476 0.1667 0.1839 0.1630 0.2622 0.2116 (0.262) (0.090)* (0.553) (0.034)** (0.006)*** (0.014)** (0.000)*** (0.000)*** Sargan testa 15.7227 16.2872 81.9780 85.0576 81.6467 85.2312 145.7353 152.8732 (0.7850) (0.7533) (0.1974) (0.1393) (0.4590) (0.3523) (0.4439) (0.2906) m1b -2.3084 -2.1357 -3.4044 -3.6209 (0.0210)** (0.0327)** (0.0007)*** (0.0003)*** m2c 0.0276 0.4124 0.3907 0.9827 (0.9780) (0.6801) (0.6960) (0.3258) Wald testd 40.82 37.36 38.58 36.32 113.25 122.5 254.85 281.13 (0.0000)*** (0.0000)*** (0.0000)*** (0.0001)*** (0.0000)*** (0.0000)*** (0.0000)*** (0.0000)***
Notes: a The null hypothesis is that the instruments used are not correlated with the residuals.
b
The null hypothesis is that the errors in the first difference regression exhibit no first-order serial correlation.
c
The null hypothesis is that the errors in the first difference regression exhibit no second-order serial correlation.
d
Wald test for all explanatory variables are jointly significant.
Values in the parentheses are p-values. ‘*’ indicates significant at the 10% level; ‘**’ ind icates significant at the 5% level; and ‘***’ ind icates significant at the 1% level.
In all mode ls, validity of the instrument variables is confirmed by bot h the Sargan test
and the Arellano-Bond serial correlation test. Highly statistically significant Wald
in explaining the premium. Significant coefficients of LPREM in mode ls C and D
suggest first-order autocorrelation exists in the premium. Though not statistically significant at any conventional level, DTE shows positive sign, as the H. I expected,
in all mode ls. As measurements of firm profitability, EPS and ROE are found not to
be contributing to the premium at any conventional significance level with only one
exception in model B and when LMCAP1 is included, and so I would draw the
conclusion that firm profitability is not an impor tant factor causing the A-shares price premium, and local Chinese and foreign investors do not have muc h difference in
valuing firm profitability. At the 10% significance level, all mode ls confirm FEG
drives the premium, however the sign of which is not what is expected. I find FEG is
positively, rather than negatively, associated with PREM, suggesting it is actually A-
shares investors who care more about firms’ long term performance. This surprising result is also found with FSG, firm’s prospect measured by the forecast sales growth,
as supported by most of the models. DIV is found to be significantly negatively
related to the premium as the H. IV expected. FR is found to be insignificant at the 5%
level in all mode ls. Because of the existence of the unique untradeable shares, firm
size is measured by two approaches, one includes the untradeable shares into calculation of the market capitalization, and one does not. Most of the models show a
positive relationship between the premium and firm size, although some of which are
not statistically significant. This finding is contradictory with both previous evidence
and the theoretical prediction discussed in the H. VI. In interpreting this, I would
rather believe that the result suggests both local and foreign investors prefer to invest in larger firms, though the local Chinese pay relatively more attention to the firm size
factor. Relative liquidity enters the premium pos itively as expected, and confirms
previous evidence that the A-shares investors prefer liquid shares.
Table 4.5: Dynamic Panel Regressions of Shenzhen-Listed Firms
GMM-DIFF GMM-SYS A B C D l.PREM 0.2793 0.2927 0.2167 0.1963 0.3058 0.2571 0.3477 0.3213 (0.086)* (0.082)* (0.053)* (0.082)* (0.001)*** (0.004)*** (0.000)*** (0.000)*** DTE 0.0074 0.0054 0.0034 0.0032 -0.0001 -0.0024 0.0001 -0.0008 (0.401) (0.558) (0.528) (0.547) (0.968) (0.573) (0.972) (0.690) EPS 1.1347 1.2570 1.7633 1.2487 1.7120 1.6437 1.0516 0.9754 (0.628) (0.605) (0.053) (0.184) (0.049)** (0.049)** (0.096)* (0.115) ROE -0.0132 -0.0186 -0.0317 -0.0279 -0.0362 -0.0377 -0.0280 -0.0282 (0.650) (0.535) (0.018)** (0.040)** (0.003)*** (0.001)*** (0.001)*** (0.001)*** FEG -0.1497 -0.1334 0.0222 0.0431 0.0388 0.0359 0.0628 0.0618 (0.373) (0.448) (0.773) (0.579) (0.631) (0.645) (0.307) (0.306) FSG -1.1085 -1.1464 -0.5028 -0.3667 0.2056 0.2472 -0.1004 -0.1330 (0.376) (0.379) (0.422) (0.563) (0.742) (0.682) (0.784) (0.714) DIV 0.0083 0.0042 -0.0010 -0.0005 0.0007 -0.0006 0.0001 -0.0013 (0.571) (0.782) (0.899) (0.949) (0.926) (0.932) (0.989) (0.801) FR 3.9933 -0.2083 -2.5703 -6.1592 -4.6904 -6.0629 -0.5609 -0.7515 (0.554) (0.974) (0.531) (0.105) (0.076)* (0.010)*** (0.521) (0.339) LMCAP1 -1.0120 -0.4590 0.2769 0.1195 (0.112) (0.173) (0.057)* (0.091)* LMCAP2 -0.4194 0.1277 0.3489 0.1426 (0.563) (0.723) (0.003)*** (0.016)** LIQ 0.2882 0.2988 0.0908 0.0922 0.1268 0.1153 0.0742 0.0683 (0.000)*** (0.000)*** (0.011)** (0.010)*** (0.000)*** (0.000)*** (0.000)*** (0.000)*** Sargan testa 20.9027 21.5443 64.1060 65.1666 79.2188 80.3416 128.0907 130.3495 (0.8909) (0.8701) (0.6442) (0.6085) (0.4402) (0.4056) (0.9439) (0.9256) m1b -2.7689 -3.0023 -4.2018 -4.1825 (0.0056)*** (0.0027)*** (0.0000)*** (0.0000)*** m2c 0.8931 0.9898 2.3784 2.3559 (0.3718) (0.3223) (0.0174)** (0.0185)** Wald testd 32.64 28.44 28.79 26.78 133.63 148.24 215.95 227.22 (0.0003)*** (0.0015)*** (0.0013)** (0.0028)*** (0.0000)*** (0.0000)*** (0.0000)*** (0.000)***
Notes: a The null hypothesis is that the instruments used are not correlated with the residuals.
b
The null hypothesis is that the errors in the first difference regression exhibit no first-order serial correlation.
c
d
Wald test for all explanatory variables are jointly significant.
Values in the parentheses are p-values. ‘*’ indicates significant at the 10% level; ‘**’ ind icates significant at the 5% level; and ‘***’ ind icates significant at the 1% level.
In the Shenzhen market, insignificant Sargan test statistics again confirm the overall
validity of instruments used in all models. However, Arellano-Bond’s second-order
serial correlation test fails to reject at the 5% significance level in mod el B. Again,
highly statistically signi ficant Wald statistics confirmed all variables included in the
regressions are jointly significant in explaining the premium. Highly significant and
pos itive coefficients of LPREM in all models indicate strong mean reversion property
of the premium. Same as what is found in the Shanghai market, the financial leverage
is insignificant, at any conventional level, in explaining the premium in the Shenzhen
market. EPS is found significant at the 5% level in Model C but not in other mode ls,
and I would consider it as not statistically impor tant in explaining the premium.
Hence, the result indicates there is not much difference in valuing EPS between investors of the A- and B-shares. However, the other profitability measurement, ROE,
is found to be highly significant and negatively related to the premium in almost all
the models, suggesting compared with the A-shares investors, the B-shares investors,
as the H. II expected, give relatively more weight to ROE when valuing shares. Both the two measurements of firm prospects, FEG and FSG, are found no longer
statistically significant at any conventional level. DIV is found no longer significant in
affecting the premium in any of the models either. As the H. V expected, significant
and negative coefficient of FR in model C, though not in other models, supports the
C and D, the po sitive sign o f firm size is again observed: the LMCAP1 and LMCAP2
are found statistically significant at the 10% and 5% levels respective ly. The highly significant and pos itive liquidity factor again confirms the liquidity preference of the
A-shares investors.