Ordinary Least Squares (OLS) Estimation
We first test to what extent UEO affects IPO underpricing by using an ordinary least squares (OLS) regression. To control for the market effects, 30-day and 60-day windows are included respectively. The specification is as follows:
$$ {INIRE}_i={\beta}_0+{\beta}_{UEO}\times {UEO}_i+{\beta}_i\times {X}_i+{\varepsilon}_i $$
(4)
where X
i
is a vector of other explanatory variables and β
i
are the coefficients to be estimated. Firstly, we check for homoskedastic error terms by using the Breusch-Pagan test. As we reject the null hypothesis that the variance of the error term is constant across observations, the standard errors presented below the coefficients are corrected for heteroskedasticity.
The main results are presented in Table 2. There are two sets of OLS regressions – Group A controls for 30-day market variables and Group B controls for 60-day market variables. As expected, the coefficients for urban economic openness are negative throughout all specifications and they are significant at 5% level. A 10% increase in the UEO results in a 2.56% to 3.08% decrease of IPO initial returns depending upon the specifications. This supports our research hypothesis that real estate companies with investments in more trade-open areas have less incentive to underprice their IPOs. Holding other factors constant, a real estate company would experience 2.56% to 3.08% less IPO underpricing if it runs its real estate business in a city where the level of trades with foreign countries is 10% higher. The greater regional trade openness increases the future profitability and reduces the valuation uncertainty, a key determinant of the level of IPO underpricing. According to the Balassa-Samuelson effect, trade openness will eventually increase the prices of non-tradable products including real estate assets. Besides, trade openness positively affects the productivity which is an important determinant of a company’s profitability. In addition, the demand for real estate could rise as a direct consequence of the increase in foreign and domestic investments into real estate given an improvement in capital flows. Therefore, investors in companies operating in areas with higher UEO tend to be more confident about the local real estate market and the company’s future profitability, which leads to less uncertainty about the company’ valuation. As a result, issuers will have less incentive to underprice the IPO shares.Footnote 6
Table 2 Baseline results: IPO underpricing and UEO
Consistent with the majority of IPO studies, we find that the time lag between issuing and listing positively affects the initial returns of IPOs – i.e. the longer it takes to reach the listing after the offering, the more uncertainty and the higher IPO underpricing we expect. As the time lag is log-transformed, holding other explanatory variables constant, the expected return difference of an IPO between two periods in time (t
1 , t
2) is represented as follows:
$$ {INIRE}_{i,1}-{INIRE}_{i,2}={\beta}_{LNLAG}\times \left[ \ln \left({t}_1\right)- \ln \left({t}_2\right)\right]={\beta}_{LNLAG}\times \ln \kern.3em \left(\frac{t_1}{t_2}\right) $$
(5)
From Eq. (5) it becomes clear that the relative change of the time lag affects the initial returns regardless of the baseline of time. If β
LNLAGA
is equal to 44.573 as shown in specification A1, then a 10% increase in the time gap between the issuance and the listing (around 4 days considering the average time of 37 days – see Table 1) will result in a 4.25% increase in the underpricing.
Another significant determinant is the firm size, which serves as a proxy for the degree of asymmetric information associated with the IPO. Unlike some of the previous Chinese IPO studies, the results in this study are in line with the information asymmetry theory: the larger the firm, the smaller the uncertainty is and therefore the lower the underpricing is. The coefficients on LNPRO are significant throughout all regressions at 95% or 90% confidence level. In specification A1, a 10% increase in the proceeds leads to a 4.37% decrease in IPO underpricing. Interestingly, we find that the return on real estate assets (ROREA) positively affects the IPO underpricing at 10% level (see specification A1 and A2), with a 2.79% to 2.99% increase in the IPO initial returns when ROREA increases by 10%. In fact, when the return on real estate assets of an IPO company is relatively high, investors read this information as a signal of “good” firm quality and hence they are more willing to participate in the IPO with a higher after-market bidding price.
Contrary to previous research in developed countries, we also find no significant relationship between the firm age and the underpricing, further indicating that the classic information asymmetry theory may be weakened by the more “immature” Chinese market. Models A1 and B1 include state ownership (STATEO) and control for 30-day and 60-day pre-IPO market conditions respectively. The coefficients for state ownership are both significant and positive suggesting that state-owned companies experience significantly higher IPO underpricing than private companies. This is consistent with the majority of Chinese IPO studies, which blame the extremely high underpricing on the political connections and government interventions—see Tian (2011). For example, Chan et al. (2004) find that the state ownership, including government and legal entity ownership, is positively related to IPO underpricing. Chang et al. (2008) argue that the Chinese government decides the IPO supply and sets the price-to-income limit for offering shares, with both regulations leading to a high level of underpricing. With regards to post-IPO stock returns, Fan et al. (2007) show that companies with more political connections actually underperform those which are loosely connected. Specifications A2 and B2 include the IPO location (China) instead of state ownership status and coefficients are significantly positive, consistent with Wong et al. (2013), where listing in Mainland China (i.e. market less transparent) leads to a greater underpricing.
To include the market factors proposed by Loughran and Ritter’s (2002) and account for market euphoria, most studies control for market returns during the 30 days preceding the IPO. As we argue that a longer momentum effect could exist for Chinese companies,Footnote 7 we also control for 60-day market returns as a robustness check. Group A and B regressions include market returns and number of IPOs issued during respectively 30 and 60 days before the IPO. Consistent with Loughran and Ritter’s (2002) prospect theory, we find that investors’ sentiment (proxied by market returns) positively affects the IPO initial returns. When we pass from a 30- to 60-day period, the coefficient becomes more significant (99% in model B1 from 90% in model A1) and hence we find support for our assumption about the weak efficiency of Chinese markets. The number of IPOs during the period preceding the IPO listing date shows a negative effect on IPO underpricing for both periods (30 and 60 days), but the coefficient is only significant when 60 days are used, supporting Altı (2005) who argues that the unknown common factor about IPO valuation will be revealed by previous IPOs, resulting in less underpricing.
Furthermore, consistent with Wong et al. (2013), we find a significant impact of listing location on IPO initial returns. Companies listed in Mainland China experience significantly higher IPO underpricing than those listed in Hong Kong or Singapore. Wong et al. (2013) argue that the low underpricing is a form of reward to a company choosing to go public in a more competitive and informationally transparent market and it signals a high firm quality. One may argue that the characteristics of companies listed in Mainland China (a) are systematically different from those listed outside (b). The same argument may be applied to the ownership structure of state-owned (a) versus private (b) companies. Hence, we investigate the need to estimate separate models by using a Chow test. Firstly, we split the sample by IPO location or state ownership, and estimate the two following regressions:
$$ {INIRE}_{i, a}={\beta}_{0, a}+{\beta}_{UEO, a}\times {UEO}_{i, a}+{\beta}_{i, a}\times {X}_{i, a}+{\varepsilon}_{i, a}\kern0.75em \mathrm{if}\frac{\mathrm{CHINA}}{\mathrm{STATEO}}=1 $$
(6)
$$ {INIRE}_{i, b}={\beta}_{0, b}+{\beta}_{UEO, b}\times {UEO}_{i, b}+{\beta}_{i, b}\times {X}_{i, b}+{\varepsilon}_{i, b}\kern0.75em \mathrm{if}\frac{\mathrm{CHINA}}{\mathrm{STATEO}}\ne 1 $$
(7)
Secondly, we compare these results with the ones obtained estimating the pooled model from Eq. (5) which assumes that the coefficients are the same across the two groups. The Chow statistic is the output of an F-Test comparing the difference between the above coefficients:
$$ F=\frac{RSS_{pooled}-\left({RSS}_a+{RSS}_b\right)}{RSS_a+{RSS}_b}\times \frac{n-2 k}{k}\sim F\left( k, n-2 k\right) $$
(8)
where \( RSS=\sum_i^N{\varepsilon}_i^2 \) is the residual sum of the squares, i.e. the variation unexplained by the regression model.Footnote 8 Results reported in Panel A of Table 3 indicate that we can estimate the pooled sample in both cases.
Table 3 Chow test for the state ownership vs IPO location and variance inflation factors
Two Stage Least Square (2SLS) Estimation
Since UEO is a company-level variable constructed by using macroeconomic factors (trade openness of Chinese cities), we need to guarantee that it is not correlated with unobserved economic factors which might also affect the individual IPO performance. In fact, if UEO is correlated with the error term ε in Eq. (5), the exogeneity assumption of OLS estimators is violated and we are presented with an omitted variable bias. As a result, the OLS estimation would be inconsistent with
$$ E\left[\beta | X\right]=\beta +{X}^{\prime }{X}^{-1}{X}^{\prime}\varepsilon \ne \beta $$
(9)
A common method to correct for endogeneity is the use of instrumental variables (IVs) which are correlated with the endogenous variable (UEO in our case) but uncorrelated with the error term ε. Previous studies find a significant relationship between the exchange rate volatility and foreign trade volumes. However, Gu and Gao (2007) find that the exchange rate volatility does not significantly affect foreign trade volume in China because, being a developing country, trades may be mainly driven by domestic demand. Hence we expect disposable household income to have a positive relationship with trade volumes and to be unlikely related with IPO returns. Therefore we use the natural logarithm of disposable household income per capita (LNDINC) as an instrument for UEO in our estimation. As a robustness check (see details in the following robustness check section), we also use an alternative instrument (distance between a city and its closest port) and find that results do not change.
The regional disposable household income per capita is collected from the City Annual Statistical Reports. We use a two stage least square estimation (2SLS). In the first stage, we estimate the predictions of the endogenous variable UEO by using the instrumental variable:
$$ UEO={\delta}_0+{\delta}_1{x}_1+\dots +{\delta}_{k-1}{x}_{k-1}+\theta LNDINC+{v}_{UEO} $$
(10)
In the second stage, the fitted values of UEO are used to replace the actual regressor and we estimate the following model:
$$ \kern2.5em INIRE={\alpha}_0+{\alpha}_1+\dots +{\alpha}_{k-1}{x}_{k-1}+\lambda LNDINC+\eta $$
(11)
where η = ε + β
UEO
v
UEO
, a
j
= β
j
+ β
UEO
δ
j
and λ = β
UEO
θ.
Results in Table 4 show that the effect of UEO on IPO initial returns remains negative and significant throughout the estimations, with significance levels reduced to 90%. A 10% increase in UEO leads to a decrease between 10.50% and 20.40% in IPO underpricing. As expected, the efficiency in a 2SLS specification is reduced while standard errors are not significantly different from OLS models. The effect of the time lag remains positive and strongly significant and the 60-day pre-IPO market performance is still preferred to a 30-day market return.
Table 4 Two stage least square estimation of IPO initial returns
Durbin and Wu-Hausman estimates which include the estimated error term from the first stage in the 2SLS estimation as an additional variable are performed to test for the endogeneity of UEO. Under the null hypothesis that all the variables are exogenous, the coefficients on the error term from the first stage should be insignificant in the Durbin and Wu-Hausman tests (otherwise, we should reject the null hypothesis and treat UEO as endogenous). The Durbin and Wu-Hausman tests are performed for each 2SLS regression and statistics are reported at the bottom of Table 4. Apart from the Wu-Hausman statistic in regression D2 (with IPO location and 60-day market variables), all other statistics suggest that we need to treat UEO as endogenous.
As there is no clear definition or test for the weakness of an instrument, we at least report the results of our first-stage regression (Table 5), where the statistical significance of the instrumental variable used to explain the endogenous variable UEO is reported. The positive relationship between disposable household income per capita and UEO is significant throughout the four specifications; the R-squared is between 0.24 and 0.29 while the F statistic ranges from 3.25 to 6.56 The F-test strongly relies on the number of endogenous variables and the number of instruments so that the more additional valid instruments are used, the greater the F statistic of the joint significance of the instruments will be. Overall, our results suggest that the chosen instrument is appropriate for our model.
Table 5 First-stage regression for urban economic openness