4. Resultados
4.4. Línea jurisprudencial del desarrollo del homicidio por piedad y
It is arguable that the results from the previous section which are found to be significant can be due to the impact of persistence or high autocorrelation in military and civilian dummy variables. Such a persistence in dummy variables typically occurs when stock returns are regressed on the dummy variable that is highly autocorrelated through time and this leads to a spurious relation between stock returns and a dummy variable when none actually exists (Ferson, Sarkissian, and Simin, 2003 and Powell et al., 2007). This sub-section therefore takes into account such a problem and analyses whether the results are being influenced by the spurious regression problem.
Firstly, this study follows the suggestions of Powell et al. (2009) by reporting the summary statistics for the dependent and dummy regression variables of each sample market in Table 4.17 and 4.18, respectively. Then, the t-statistics that are reported in Table 4.12 together with the level of autocorrelation (AC) and transition probability of dummy variables (q) in Table 4.17 and 4.18 are checked with the table for the cut-off t-statistics that is provided in Powell et al. (2009) whether there is any prospect for a spurious regression problem. It is found that all of the sample markets pass the critical cut-off values for spurious regression bias.
141 Table 4.17
Summary statistics for dependent regression variables
The table presents the monthly summary statistics for the real stock returns variables of each sample markets. The sample periods vary for each sample market. This is according to the year that each stock market starts its trading. However, the end of sampling period is set at December 2007 for all markets. The number of observations (N) in month, mean, standard deviation (SD), and autocorrelation (AC) are reported, respectively. The mean and standard deviation of returns are in percentage term.
Country N Mean (%) SD (%) AC Argentina 492 1.18 19.42 0.073 Bangladesh 339 3.32 48.94 -0.097 Chile 1128 0.67 9.13 0.296 Colombia 971 0.03 6.47 0.214 South Korea 552 1.16 13.62 0.196 Pakistan 587 0.22 5.44 0.176 Peru 972 0.45 11.77 0.060 Thailand 392 0.50 8.51 0.100 Uruguay 851 -0.24 16.24 -0.064 Venezuela 936 0.14 8.08 0.058
142 Table 4.18
Summary statistics for dummy regression variables
The table presents the summary statistics for the Military dummy regression variables of each sample markets. The sample periods vary for each sample market. This is according to the year that each stock market starts its trading. However, the end of sampling period is set at December 2007 for all markets. The number of observations (N) in month, mean, standard deviation (SD), autocorrelation (AC), and transition probability of military dummy variable (q), are reported, respectively. Transition probability is calculated by q = (1+ AR(1))/2 . The mean and standard deviation of dummy variable are in percentage term.
Country N Mean (%) SD (%) AC Q Argentina 492 34.55 47.60 0.983 0.9915 Bangladesh 339 30.24 46.00 0.959 0.9795 Chile 1128 25.74 43.74 0.978 0.9890 Colombia 971 6.48 24.63 0.983 0.9915 South Korea 552 55.37 49.76 0.987 0.9935 Pakistan 587 65.78 47.49 0.982 0.9910 Peru 972 28.45 45.14 0.972 0.9860 Thailand 392 16.03 36.74 0.914 0.9570 Uruguay 851 18.38 38.75 0.992 0.9960 Venezuela 936 30.02 45.86 0.982 0.9910
Secondly, to further confirm the robustness of the results, the issue of persistence in dummy variable is also addressed by using the approach advocated by Powell et al. (2007). In this approach, it is suggested that the regression analysis should be repeated by using one return and one dummy variable per political term. Anderson et al. (2008) suggest that such an
143 approach helps to mitigate the spurious regression problem and reduces the autocorrelation in the dummy variable. Therefore, the regression is being re-estimated with one return and one dummy variable per military/civilian regime. Notably, this robustness test is carried out exclusively for Pakistan and Thailand since they are the two markets with significant differences in stock returns between the two political regimes. The results are reported in Table 4.19, as follows:
Table 4.19
Robustness check: Persistence in dummy variable
The table presents the results from estimating regression model [4.2] on the stock market of Pakistan and Thailand by using one return and one dummy variable per each military/civilian regime. Mt and Ct denote
political dummy variables where M=1 or C=1 if a military or civilian government is in office at time t,
respectively,M=0 or C=0, otherwise. There are nine political shifts between military and civilian regime for Pakistan and 19 for Thailand. The sample period is from 1960 to 2007 and 1975 to 2007 for Pakistan and Thailand, respectively. The results are annualised and they are presented in percentage term. Standard errors are corrected using the Newey and West (1987) procedure for both autocorrelation and heteroskedasticity of an unknown form. T statistics appear in parentheses and are a test of the null hypothesis that the coefficient is equal to zero. F-statistics are also reported in parentheses in the last column for the test for equal coefficients. *, **, *** indicates significance at the 10%, 5%, and 1% level, respectively.
Country α 1 (Mt) α 2 (Ct)
Difference
(α 1 - α 2) = 0 R2
Pakistan 8.906 (2.03)* -7.459 (-1.57) 17.561 (4.14)* 0.355 Thailand 32.022 (2.77)** -8.084 (-0.85) 43.242 (5.58)** 0.195
With this approach, it is found that the autocorrelation of the dummy variables reduces considerably from 0.982 to 0.122 for Pakistan and from 0.914 to 0.117 for Thailand. The results from Table 4.19 also show that the annualised percentage stock returns are significantly greater under military governments for both Pakistan and Thailand. Moreover, these returns are significantly different from each other under the two political regimes. The power of the statistical tests is also shown to be reasonably strong under this approach. Therefore, the findings of higher stock returns under military regimes for Pakistan and Thailand are robust to the use of one return and one dummy variable per military/civilian regime.
144 Despite this, there remains the chance that the differences in the stock returns between the two political regimes are due to a too small sample problem. This issue is therefore addressed next.