Sample data analysis .1 Results of Propensity Score Matching

285 errors to do statistical inference. 1551 observations are left to after calculating the 95 confidence intervals. Then, we employ the nearest neighbor matching procedure to get the treated and control groups. It can be seen from Table 4 this step shrank the sample to 102 treated and 89 untreated firms. Ta le : Des riptio of Esti ated prope sit s ore Obs Mean Std. Dev. Skewness Kurtosis 1551 .0662039 .0601257 8.11689 116.3705 Ta le : Des riptio of A erage Treat e t Effe t No. treat. No. contr. ATT Std. Err. t-statistics 103 89 - 226.845 204.836 - 1.107

3.3.2 Results of Matched Groups Ta le : Des ripti e su

ar statisti s Varia les O s Mea Std. De . Mi i u Ma i u Ske ess Kurtosis IR 166 0.5613 0.6323 - 0.1576 4.2816 2.2051 11.0598 AGE 166 6.5128 4.6883 0.0329 20.0000 0.6700 2.5896 IPRICE 166 21.1579 14.8825 2.6000 69.0000 1.3740 4.5662 DAY 166 13.0970 6.9064 7.0000 77.0000 5.6168 48.1577 LASSET 166 5.9848 4.1758 0.6990 11.6295 - 0.3262 1.1618 LSALES 166 8.9701 0.5314 7.7071 11.4279 0.5968 5.1627 LNUM 166 3.5573 0.3491 3.0449 4.8946 1.8274 7.1663 LPROCEEDS 166 4.7717 0.3980 3.8573 6.1957 0.6668 3.8155 PERCENT 166 0.4261 0.2787 0.0000 0.7500 - 0.5644 1.7987 VCDUM 166 0.5879 0.4937 0.0000 1.0000 - 0.3571 1.1275 Table 5 presents the descriptive statistics for all the variables in this research, including mean, minimum, maximum, standard deviation and skewness and kurtosis. In total, there are 166 observations investigated in this analysis. On average, the sample firms operated before the IPO for 6.5 years and the minimum of AGE is around 12 days, which indicates that the period between establishment and the IPO is relatively short and may have the risk that they go public too prematurely. It is obvious that the kurtosis of Day is very large with 48.1577. The average days from the offering to listing is 13, ranging from 7 to 77. It is inconsistent to previous study that, it normally takes more than 2 months in China for the new offered to become listed Su and Fleisher, 1999. Table 6 shows a correlation coefficient table among ten variables. It is reported that the maximum correlation between two independent variables should be 0.80; if it exceeds 0.80, two variables will be suspected of the existence of multicolinearity Bryman and Cramer, 1997. Multicolinearity means the independent variables are significant correlated with one another. Therefore, these variables with high degree of correlation have to be deleted from the sample. Fortunately, the correlation table presents the correlation coefficient between these variables is all below 0.08, with the maximum of correlation 0.667 between LPROCEEDS and LNUM, which demonstrates that the correlation among these is not very strong. The correlation between IR and LSALES is positive but not significant, which is 0.1594. This confirms H2 that AGE is positively related to IR of VC-backed IPOs. Except for those significant correlation mentioned above, most of the coefficients shown in the matrix are non-significant 286 and weak. The multicolinearity effect is quite small among the independent variables, and the degree of multicolinearity is not possible to have any impacts on the regression results. 287 Ta le : Whole sa ple orrelatio oeffi ie t et ee the aria les Variables IR AGE PERCEN T DAY IPRICE LPROCE EDS LNUM LSALES LASSET VCDUM IR 1.0000 AGE - 0.1820 1.0000 PERCENT -0.4986 0.2923 1.0000 DAY 0.2654 -0.2426 - 0.4468 1.0000 IPRICE - 0.3614 0.1101 0.4812 - 0.2965 1.0000 LPROCEE DS - 0.4157 0.2615 0.3482 - 0.3319 0.4771 1.0000 LNUM - 0.0902 0.1273 - 0.1085 - 0.0102 - 0.2619 0.6675 1.0000 LSALES 0.1594 -0.1694 - 0.2609 0.1501 - 0.0808 0.3566 0.5094 1.0000 LASSET - 0.0986 0.2992 0.1416 - 0.2026 0.0352 0.1388 0.0939 - 0.2011 1.0000 VCDUM - 0.1141 0.2948 0.1768 - 0.2153 0.0556 0.1055 0.0327 - 0.2633 0.6044 1.0000 288

4. Empirical Results and discussion

4.1 IPO Performance 4.1.1 Initial Returns of the IPOs Ta le : Mea i itial retur of VC- a ked IPOs a d No -VC- a ked IPOs IPO cases All IPOs VC-backed IPOs NON-VC-backed IPOs Number 166 97 71 Mean IR 0.5778 0.5011 0.6857 t-statistics: difference from zero of mean 11.189 8.1887 7.7425 P-value Note: differe e et ee ea IR of o -VC- a ked a d VC- a ked IPOs: t- statisti s . Table 7 reflects that on average all 166 IPOs are underpriced by a significant amount of 57.78. VC-backed IPOs are less underpriced than Non-VC-backed IPOs and the difference is statistically significant at the 10 confidence interval, with t-statistics of 1.7737. This provides evidence to against the certification model as it fails to add value by underpricing more. Thus, we could reject H1 that on average the IPOs backed by VCs and non-VC-backed IPOs have the same level of underpricing. Our conclusion that VCs tend to underprice less at the IPO, which is in line with VCs certify to alleviate informational asymmetry about the IPO issue at the issue time Hamao et al., 2000. This finding supports the certificationmonitoring model is evidenced.

4.1.2 Determinants of Underpricing

The empirical results from multiple regression analysis were presented in Table 8. The coefficient shown in the table indicates positive or negative relationship between the underpricing and these independent variables. Following Wong and Wong’ model 2008, the model estimation in illustrated in the column 1. LSALES and LASSET have a significant positive relationship to initial return as assumed in H2 but LPROCEEDS and VCDUM is negatively related and statistically significant, which reject the assumption in H2. The results indicate that IPOs funded by VCs significantly reflect a lower underpricing level than non-VC-funded IPOs. When the IPOs are VCs funded, decrease in the underpricing will be close to 45 in column 2 t-statistics = -0.3993 and 200 in column 1 t- statistics = -1.9582. The existence of VCs contributes to the underpricing by a significant amount. The conclusion is consistent with an amount of studies such as Bottazzi et al. 2008 and Suchard 2009 to the certification monitoring model that VCs is value-added via underpricing less. As suggested by Chen and others studies 2004, in the column 2 we add ISSP, DAY and LNUM to this model. After making the adjustments to this model, the R-square increases by a great amount from 30.07 percent to 35.72 percent. DAY and LNUM are separately negative and positive related to the underpricing in the column 2. R-squared is relatively high with 0.3729, suggesting that the rest variables are better predictors of IR. The conclusion is conflict to Chen and other researchers’ findings 2004 that the underpricing of IPOs is positively associated with the lasting days from the offering to the listing and negatively related to the amount of shares issued. From table 8 it can be told that PERCENT has a significantly negative impact to the IPO performance. It confirms much previous literature that documented non-tradable shares lead to the IPO underpricing in Chinese market e.g., Chen et al., 2004, Tan et al., 2013. In China the great underpricing is partly caused by the large demand from general investors who do not have 289 other investment channels and the limited supply of publicly available shares. The reason is that shares owned by the government or legal institution were not allowed to trade publicly after the IPO, which is represented by PERCENT here. Ta le : Li ear regressio a al sis et ee the IPO u derpri i g a d fir spe ifi aria les a d o trol aria les Dependent Variable IR Column1 Column2 Independent constant 1.6828 1.8201 variables 1.8106 -1.9225 IPO characteristics: LPROCEEDS - 0.9000 - 0.9644 -7.3816 -2.5626 VCDUM - 2.0085 - 0.4480 -1.9582 -0.3993 LNUM 0.3133 0.8064 DAY - 0.0048 -0.6860 PERCENT - 0.6730 -3.3594 ISSP 0.0051 0.7235 Firm-level features: LSALES 0.3195 0.2602 2.9207 2.3906 LASSET 0.2414 0.0568 2.0131 0.4319 AGE - 0.0004 0.0058 -0.0004 0.5882 R-squared 0.3007 0.3572 Adjusted R-squared 0.2787 0.3199 F-statistics 13.6735 9.5719 P-value 0.0000 0.0000 Notes: Significant at: 10, 5, and 1 percent levels, respectively. T - statistics appear in parentheses.