Relationship between board's characteristics and asymmetric information of listed firms on Ho Chi Minh stock exchange

nient formula for estimating asymptotic average adverse selection for each stock in the KO model. Accordingly, ASCi,KO is estimated by the following formula: , , 1 , 2 1 2 ( , ) 1 1 TM it TM it i KO T qit t Cov RD RD ASC S T       (8) Table 3.1 below summarizes the used measurement model, regression equation, and ASC estimation formulas for the sample and individual stock. Table 3.1. The using asymmetric information measurement models Measurement models Regression equation ASC for sample ASC for stock i 1. Glosten and Harris (1988) model GH model ΔPt = c0ΔQt + c1Δ(QtVt) + z0Qt + z1QtVt + εt 0 1 0 1 0 1 2( ) 2( ) 2( ) z z V c c V z z V     0 1 0 1 0 1 2( ) 2( ) 2( ) i i i i i i i i i z z V c c V z z V     2. George, Kaul and Nimalendran (1991) trade-indicator model GKN trade-indicator model 2RDt = a0 + a1 (Sq)[Qt – Qt–1] + εt 1 – a1 1 2 1 ( )( ) 1 ( ) T it it t T it t x x y y x x         (a) 3. George, Kaul and Nimalendran (1991) serial covariance model GKN serial covariance model 𝑆𝑖 ∗ = b0 + b1Sqi + εi 1 – b1 , , 1 1 2 ( , ) 1 1 TM it TM it T qit t Cov RD RD S T      (b) 4. Kim and Ogden (1996) model KO model 𝑆𝑖 ∗∗ = β0 + β1√𝑆�̅�𝑖 2 + εi 1 – β1 , , 1 2 1 2 ( , ) 1 1 TM it TM it T qit t Cov RD RD S T      (c) Note: (a) xit = (Sqi)[Qit – Qit–1], yit = 2RDTM,it ; (b) Jones et al. (1994), Kim and Ogden (1996) proposed; (c) Kim and Ogden (1996) proposed. Source: Glosten and Harris (1988); George, Kaul and Nimalendran (1991); Jones et al. (1994); Kim and Ogden (1996) 16 3.2 Adoption of asymmetric information measurement model Firstly, the measurement model that satisfies conditions that there are few excluded observations, and the deviation between asymmetric information for individual stock and the sample is low is the basis for suggesting the appropriate model. Secondly, the research examines the similarity between models by estimating the correlation coefficients between the levels of asymmetric information applied the different models for each stock according to Van Ness et al. (2001), De Winne and Majois (2003), Lamoureux and Wang (2015). Then, the study estimates the correlation between variables, including the levels of asymmetric information using different models and the determinants, such as growth opportunity, liquidity trading, and debt ratio for the purpose of testing which model will have an estimate of asymmetric information in accordance with economic theory and related empirical studies. The study expects that asymmetric information will be negatively correlated with liquidity trading (Acker et al., 2002; Draper and Paudyal, 2008) and debt ratio (Ross, 1977; Jensen, 1986; Degryse and Jong, 2006), and positively correlated with growth opportunity (Krishnaswami et al., 1999; Hegde and McDermott, 2004; Fosu et al., 2016). Finally, the study tests the change in the level of asymmetric information before and after the period of adjusting price limit range. In Vietnam, the price limit range was widen from 5% to 7% on January 15, 2013 according to Regulation 01/2013/QD-SGDHCM of HOSE; therefore, according to Anshuman and Subrahmanyam (1999), Berkman and Lee (2002), the study expects that asymmetric information will increase after the price limit range is expanded. 3.3 Research framework The quantitative studies of Cai et al. (2006), Armstrong et al. (2014), Goh et al. (2016), Abad et al. (2017) illustrate that, the board's characteristics are likely to impact on asymmetric information. In addition, the impact of non-executive directors and education level of board members on asymmetric information could depend on type of firms with state-owned firms, especially in the context of a developing market (Barberis et al., 1996; Buck et al., 2008; Wang, 2012; Wang et al., 2016). Based on the theoretical analysis framework and related empirical studies, the empirical research framework on the relationship between the board's characteristics and asymmetric information is built as Figure 3.1 below. 17 Source: Cai et al. (2006), Armstrong et al. (2014), Goh et al. (2016), Abad et al. (2017); Barberis et al. (1996); Buck et al. (2008); Wang (2012); Wang et al. (2016) Figure 3.1. Empirical research framework Figure 3.1 denotes a impact of the board's characteristics, including number of board members, female directors, educational qualification of directors, duality, personal ownership of directors, and outside directors on asymmetric information. Besides, the impact of outside directors and the educational qualification of board members on asymmetric information is likely to depend on the type of firms, namely state-owned firms. Control variables, including factors related to characteristics of market, such as liquidity trading, stock price volatility, growth opportunity, period of adjusting price limit range, and factors related to characteristics of firm, such as debt ratio, firm size, and effect of industry sectors are also considered. 3.4 Research hypothesis Based on quantitative research, related theories, empirical research framework and the context of Vietnam stock market, the hypotheses on the relationship between the board's characteristics and asymmetric information are constructed as follows: H1: There is a positive relationship between the number of board members and asymmetric information. H2a: There is a negative relationship between non-executive directors and asymmetric information. - Number of board members - Female directors - Duality - Shareholder of board members Asymmetric Information - Inderpendent and non- executive board members - Educational qualification of board members - Type of firm (state- owned and non-state firms) - Market characteristics - Firm characteristics : Dependent variable : Explanatory variables : Control variables 18 H2b: There is a difference in the impact of the independent members on asymmetric information between the state-owned and non-state-owned firms. H3: There is a negative relationship between female directors and asymmetric information. H4a: There is a negative relationship between the educational qualification of board members and asymmetric information. H4b: There is a difference in the impact of the educational qualification of board members on asymmetric information between the state-owned and non-state-owned firms. H5: There is a positive relationship between duality and asymmetric information. H6a: There is a negative relationship between the shareholding ratio of board members and asymmetric information. H6b: There is non-linear relationship between the shareholding ratio of board members and asymmetric information. 3.5 Research method This study uses quantitative method to estimate the relationship between the board's characteristics and asymmetric information of listed firms on Vietnam stock market. The following is a description of how to conduct the study, including data collection, measure of research variables, and method of data analysis. 3.5.1 Data collection We collect data of listed firms on the Ho Chi Minh Stock Exchange (HOSE) from 2009 to 2015. The data of transaction price and order statistics of listed firms on HOSE is collected at the time of Quarter 1, from January 1 to March 31 to measure asymmetric information and market-related factors that include liquidity trading, stock price volatility, and growth opportunity. Data related to board's characteristics, debt ratio, and size of firm is collected at the time firms disclose at the end of the year. Finally, listed firms whose financial year does not coincide at the end of the year will be excluded in the sample. The research sample does not include listed firms with a fiscal year that do not coincide at the end of the year; firms subject to special warning, control, compulsory cancellation of listing or voluntary delisting; financial institutions such as banks, securities companies, insurance companies, and investment funds; and firms that do not disclose information 19 related to research variables. Industry classification benchmark is based on the North American Industry Classification System (U.S. Census Bureau, 2017). 3.5.2 Definition and measure of variables Table 3.2 below summarizes the definition and measure of research variables. Table 3.2. Definition and measure of variables Variables Definition Measurement ASC Adverse selection component proxies asymmetric information Using the suitable asymmetric information measurement model under the context of Vietnam stock market BoardSize Board members Number of board members Outd Independent non-executive directors The proportion of independent non-executive board members Gender Female directors The proportion of women on the board Edu Educational qualification of board members The proportion of board members with postgraduate education Dual Duality Dual = 1, if Chairman also holds the positon of CEO; otherwise Dual = 0 Own Directorial ownership The personal shareholding ratio of board members Gov Type of firm Gov = 1, state-owned firm Gov = 0, non-state-owned firm Depth Liquidity trading The average of shares available at both the best bid and ask prices divided by number of outstanding shares Volatility Volatile stock price Standard deviation of the bid and ask midpoint Opp Level of growth opportunity Opp = 1 if TobinQ > 1, high growth opportunity Opp = 0 if TobinQ < 1, low growth opportunity where: TobinQ = [market value of stock + total debt] / total assets Debt Total debt ratio Ratio of total of debt to total assets Bank Bank loan ratio Ratio of bank loan to total assets Bank_St Short-term bank loan ratio Ratio of short-term bank loan to total assets Bank_Lt Long-term bank loan ratio Ratio of long-term bank loan to total assets DumYear The period of adjusting price limit range from 5% to 7% DumYear = 1; period 2013-2016 (price limit range is 7%) DumYear = 0; period 2010-2012 (price limit range is 5%) FirmSize Size of firm Natural logarithm of total assets Industry Effect of industry sectors Dummy variables indicate the effect of industry sectors 3.5.3 Regression analysis 3.5.3.1 Regression equation 20 Firstly, to estimate the relationship between the characteristics of board and asymmetric information, we use the regression equation as follows: 0 1 2 3 4 5 6 , 1 it it it it it J it it j j it it j ASC BoardSize Outd Gender Edu Dual Own ControlVar                          (3.9) Next, to test whether the impacts of independent directors and the education level of board members on asymmetric information depend on the type of firms, state-owned and non-state-owned firms, applying the method of DeMaris (2004), we, in turn, add two interaction variables Gov*Outd and Gov*Edu into equation (3.9) to get the regression equation (3.10) and (3.11) below as follows: 0 1 2 3 4 5 6 7 , 1 * it it it it it J it it it it j j it it j ASC BoardSize Outd Gender Edu Dual Own Gov Outd ControlVar                             (3.10) 0 1 2 3 4 5 6 8 , 1 * it it it it it J it it it it j j it it j ASC BoardSize Outd Gender Edu Dual Own Gov Edu ControlVar                             (3.11) The estimated results of the regression coefficient β7 of the Gov*Outd variable in equation (3.10) and β8 of Gov*Edu variable in equation (3.11) could be the basis to reject or accept the research hypothesis. Moreover, to clarify the difference in the impacts of Outd and Edu on asymmetric information under type of firms, we divide the research sample into two groups including state-owned and non-state-owned firms. The results of regression estimation on two groups will illustrate the form and strength of the impact of these two variables on asymmetric information for different types of firms. Finally, to test the hypothesis there is a nonlinear relationship between the shareholding of board members ratio and asymmetric information, we estimate the threshold regression model according to Bai and Perron (2003). The threshold regression model has the following form: , , 0 1 1 ( , ) K L J i k i k i l l i j j i i k l j ASC Own Own X ControlVar               (3.12) Where: τ is threshold value of Own; assuming k potential thresholds (τ) of Own, then producing k+1 regimes; 1( , ) ( )k i k i kOwn Own       is an indicator function with a 21 value of 1 if the condition is satisfied, otherwise the value is 0; Xl (l = 1,5 ) are the explanatory variables whose coefficients do not change across regimes, and represent the board characteristics including: Outd, Edu, BoardSize, Gender, and Dual; ControlVarj are control variables. Regression coefficients α, β, δ and threshold values τ of Own from equation (3.12) are estimated by minimizing the function S(α, β, δ, τ) which have the following equation form: 2 , , 1 0 1 1 ( , , , ) ( , ) n K L J i k i k i l l i j j i i k l j S ASC Own Own X ControlVar                          After estimating the threshold values τ, we test the strength of this value by applying the piecewise regression method according to Morck et al. (1988), Hermalin and Weisbach (1991). With the assumption of finding two threshold values of Own (τ1 and τ2), the piecewise linear regression has the following equation form: 0 1 2 3 4 5 3 , , 1 1 _ it it it it it it J s s it j j it it s j ASC BoardSize Outd Gender Edu Dual Own Thr ControlVar                           (3.13) Where: Own_Thrs (s = 1,3 ) are threshold regimes of Own and are defined as follows: 1 1, 1 1 1 2, 1 1 2 2 1 2 2 3, 2 2 if _ if 0 if < _ if < if 0 if _ if it it it it it it it it it it it it it Own Own Own Thr Own Own Own Thr Own Own Own Own Own Thr Own Own                               The results of estimating the regression coefficients λs of Own_Thrs could be the basis to accept or reject the nonlinear relationship between the shareholding of board members ratio and asymmetric information. 3.5.3.2 Regression method For panel data, we use regression techniques, including Pooled-OLS regression model (Pool), fixed effect model (FEM), and random effects model (REM). The adoption of these models will be considered based on Hausman and Breusch-Pagan tests. 22 CHAPTER 4 EMPIRICAL RESULTS AND DISCUSSION 4.1 Measuring asymmetric information 4.1.1 Degree of asymmetric information Adverse selection component, a proxy for the degree of asymmetric information, is measured by econometric models, including Glosten and Harris (1988) (GH model), George, Kaul and Nimalendran (1991) trade-indicator model (GKN trade-indicator model), George, Kaul and Nimalendran (1991) serial covariance model (GKN serial covariance model), and Kim and Ogden (1996) model (KO model). 4.1.1.1 Glosten and Harris (1988) model Table 4.1 below presents regression results for estimating ASCGH, adverse selection component based on GH model. Table 4.1. Regression results using the GH model Coefficient ∆Pt = c0∆Qt + c1∆(QtVt) + z0Qt + z1QtVt + εt 2015 2014 2013 2012 2011 2010 2009 constant 0.033 *** 0.012 0.024 *** 0.025 *** -0.002 -0.036 *** -0.040 *** c0 0.411 *** 0.373 *** 0.283 *** 0.257 *** 0.167 *** 0.250 *** 0.321 *** c1 -0.030 *** -0.025 *** -0.019 *** -0.018 *** -0.009 *** -0.016 *** -0.022 *** z0 0.200 *** 0.165 *** 0.154 *** 0.104 *** 0.138 *** 0.137 *** -0.661 *** z1 0.022 *** 0.014 *** 0.024 *** 0.029 *** 0.021 *** 0.027 *** 0.129 *** ASCGH 72.2% 63.3% 79.0% 77.9% 79.1% 78.5% 89.2% Observations 9454 9239 9295 9690 9488 8680 6669 R 2 adj. 25.1% 22.8% 32.5% 27.4% 33.5% 18.2% 35.2% Durbin-Watson 2.14 2.21 2.07 1.87 1.98 2.03 1.92 F 0.91 0.54 0.95 1.61 *** 1.30 *** 0.92 0.58 Breusch-Pagan 47.70 *** 19.16 *** 42.58 *** 73.33 *** 140.34 *** 134.34 *** 968.18 *** Hausman 21.25 *** 5.81 26.65 *** 1.54 12.29 ** 13.29 *** 8.96 * *** significant at the 1% level; ** significant at the 5% level; * significant at the 10% level Source: Transaction data of listed firms on HOSE Table 4.1 illustrates that the results of F test, Breusch-Pagan, and Hausman recommend FEM method for performing regression equation in 2014, 2012, and 2009. While also with these tests, FEM method will be recommended in the remaining years. ASCGH for the sample in years belongs to (63.3% ; 89.2%) and satisfies the condition 0 < ASCGH < 1. In which ASCGH reached the highest value in 2009 (89.2%), the second-highest in 2011 (79.1%), and lowest value in 2014 (63.3%). 23 4.1.1.2 George, Kaul and Nimalendran (1991) trade-indicator model Table 4.2 below presents regression results for estimating ASCGKN1, adverse selection component based on GKN trade-indicator model. Table 4.2. Regression results using the GKN trade-indicator model Coefficient 2RDTM,it = a0 + a1 (Sqit)[Qit – Qit–1] + εit 2015 2014 2013 2012 2011 2010 2009 a0 0.020 0.024 0.019 0.017 0.021 * 0.024 0.005 a1 0.287 *** 0.283 *** 0.245 *** 0.301 *** 0.433 *** 0.382 *** 0.412 *** ASCGKN1 71.3% 71.7% 75.5% 69.9% 56.7% 61.8% 58.8% Observations 9454 9239 9295 9690 9488 8680 6669 R 2 adj. 51.2% 45.7% 49.2% 51.5% 58.3% 54.5% 60.7% Durbin-Watson 2.99 2.99 2.96 2.89 2.88 2.83 2.86 F 0.01 0.02 0.02 0.03 0.03 0.02 0.01 Breusch-Pagan 186.95 *** 81.70 *** 286.90 *** 305.87 *** 1,039 *** 1,094 *** 5,596 *** Hausman 0.10 1.09 0.80 1.44 1.69 0.86 0.04 *** significant at the 1% level; ** significant at the 5% level; * significant at the 10% level Source: Transaction data of listed firms on HOSE The results of necessary tests in Table 4.2 recommend FEM method for performing regression equation. In addition, ASCGKN1 for the sample in years is about (56.7% ; 75.5%), and satisfies the condition 0 < ASCGKN1< 1. In which ASCGKN1 reached the highest value in 2013 (75.5%), the second-highest in 2014 (71.7%), and lowest value in 2011 (56.7%). 4.1.1.3 George, Kaul and Nimalendran (1991) serial covariance model Table 4.3 below presents regression results for estimating ASCGKN2, adverse selection component based on GKN serial covariance model. Table 4.3. Regression results using the GKN serial covariance model Coefficient Si GKN = b0 + b1Sqi + εi 2015 2014 2013 2012 2011 2010 2009 b0 0.021 -0.272 0.302 ** 0.251 *** 0.061 * 0.172 * 0.344 *** b1 0.346 *** 0.376 *** 0.250 *** 0.304 *** 0.463 *** 0.370 *** 0.344 *** ASCGKN2 65.4% 62.4% 75.0% 69.6% 53.7% 63.0% 65.6% Observations 163 168 169 170 164 155 117 R 2 adj. 72.9% 61.3% 64.2% 85.3% 90.5% 69.5% 77.2% White 41.74 *** 143.50 *** 65.55 *** 51.72 *** 43.17 *** 53.23 *** 17.70 *** *** significant at the 1% level; ** significant at the 5% level; * significant at the 10% level Source: Transaction data of listed firms on HOSE The regression results in Table 4.3 presents that ASCGKN2 for the sample in years is about (53.7% ; 75.0%), and satisfies the condition 0 < ASCGKN2< 1. In which ASCGKN2 reached the 24 highest value in 2013 (75.0%), the second-highest in 2012 (69.6%), and lowest value in 2011 (53.7%). 4.1.1.4 Kim and Ogden (1996) model Table 4.4 below presents regression results for estimating ASCKO, adverse selection component based on KO model. The estimated results show that ASCKO for the sample in years is about (53.9% ; 75.0%), and satisfies the condition 0 < ASCKO < 1. In which ASCKO reached the highest value in 2013 (75.0%), the second-highest in 2012 (70.0%), and lowest value in 2011 (53.9%). Table 4.4. Regression results using the KO model Coefficient Si KO = β0 + β1√�̅�𝒒𝒊 𝟐 + εi 2015 2014 2013 2012 2011 2010 2009 β0 0.021 -0.273 0.299 ** 0.254 *** 0.059 * 0.169 * 0.342 *** β1 0.345 *** 0.375 *** 0.250 *** 0.300 *** 0.461 *** 0.369 *** 0.343 *** ASCKO 65.5% 62.5% 75.0% 70.0% 53.9% 63.1% 65.7% Observations 163 168 169 170 164 155 117 R 2 adj. 73.0% 61.3% 64.2% 85.3% 90.5% 69.8% 77.3% White 41.65 *** 143.55 *** 65.40 *** 51.21 *** 43.38 *** 51.73 *** 17.66 *** *** significant at the 1% level; ** significant at the 5% level; * significant at the 10% level Source: Transaction data of listed firms on HOSE Figure 4.1 below outlines the overall degree of asymmetric information measurement results based on GH model, GKN trade-indicator, GKN serial covariance, and KO. Source: Transaction data of listed firms on HOSE Figure 4.1. Changing in the degree of asymmetric information over the years 50% 55% 60% 65% 70% 75% 80% 85% 90% 95% 2009 2010 2011 2012 2013 2014 2015 GH GKN trade-indicator GKN serial covariance KO 25 4.1.2 Degree of asymmetric information for individual stock Table 4.5 below shows the statistics of measurement results of adverse selection component (ASC), a proxy for asymmetric information, for each stock over the years. Table 4.5. Statistics of ASC measurement results for individual stock Period ASC ASC unsorted ASC sorted (0 < ASC < 1) Mean Min Max n Mean Min Max n 2009-2015 ASCGH 77.6% -107.3% 1100.0% 1106 58.0% 17.4% 99.3% 96 ASCGKN1 63.9% -29.0% 100.0% 1106 64.1% 25.4% 92.3% 1102 ASCGKN2 59.9% -21.6% 91.7% 1106 60.0% 15.4% 91.7% 1105 ASCKO 60.1% -21.3% 91.7% 1106 60.2% 15.5% 91.7% 1105 2015 ASCGH 74.0% 13.6% 257.1% 163 51.6% 24.4% 97.7% 21 ASCGKN1 69.5% 36.2% 90.0% 163 69.5% 36.2% 90.0% 163 ASCGKN2 63.6% 27.2% 88.4% 163 63.6% 27.2% 88.4% 163 ASCKO 63.7% 27.4% 88.4% 163 63.7% 27.4% 88.4% 163 2014 ASCGH 66.6% 12.6% 129.7% 168 50.5% 17.4% 89.7% 13 ASCGKN1 73.3% 31.6% 92.0% 168 73.3% 31.6% 92.0% 168 ASCGKN2 68.7% 17.2% 90.3% 168 68.7% 17.2% 90.3% 168 ASCKO 68.7% 17.4% 90.3% 168 68.7% 17.4% 90.3% 168 2013 ASCGH 79.8% 34.0% 152.5% 169 73.0% 31.8% 98.9% 12 ASCGKN1 69.8% 36.7% 92.3% 169 69.8% 36.7% 92.3% 169 ASCGKN2 64.3% 20.7% 91.7% 169 64.3% 20.7% 91.7% 169 ASCKO 64.4% 21.0% 91.7% 169 64.4% 21.0% 91.7% 169 2012 ASCGH 76.2% 31.8% 145.6% 170 56.8% 25.8% 99.3% 24 ASCGKN1 63.5% 29.0% 85.2% 170 63.5% 29.0% 85.2% 170 ASCGKN2 59.7% 33.5% 83.6% 170 59.7% 33.5% 83.6% 170 ASCKO 60.1% 34.1% 83.7% 170 60.1% 34.1% 83.7% 170 2011 ASCGH 85.2% 20.3% 1100.0% 164 61.2% 24.2% 99.0% 12 ASCGKN1 52.7% 25.4% 100.0% 164 52.4% 25.4% 79.9% 163 ASCGKN2 49.8% 23.5% 90.3% 164 49.8% 23.5% 90.3% 164 ASCKO 50.1% 24.1% 90.4% 164 50.1% 24.1% 90.4% 164 2010 ASCGH 78.6% -107.3% 219.5% 155 51.6% 21.2% 95.1% 7 ASCGKN1 59.5% -29.0% 79.1% 155 60.9% 32.2% 79.1% 152 ASCGKN2 56.4% -21.6% 79.5% 155 57.0% 18.2% 79.5% 154 ASCKO 56.7% -21.3% 79.5% 155 57.2% 18.4% 79.5% 154 2009 ASCGH 85.0% 49.6% 165.1% 117 71.2% 58.7% 96.3% 7 ASCGKN1 56.4% 35.7% 78.2% 117 56.4% 35.7% 78.2% 117 ASCGKN2 55.2% 15.4% 81.1% 117 55.2% 15.4% 81.1% 117 ASCKO 55.4% 15.5% 81.1% 117 55.4% 15.5% 81.1% 117 Source: Transaction data of listed firms on HOSE The results in Table 4.5 show that, sorting ASCGH significantly reduced the number of observations, from 1106 to 96 observations. Yearly, from 2009 to 2015, the number of 26 observations that the ASCGH satisfies is low. The highest value was 24 observations in 2012, and the lowest was 7 in 2010. Meanwhile, classification ASCGKN1, ASCGKN2 và ASCKO hardly reduces the number of observations. Next, Table 4.6 below describes the degree of deviation between the measured ASC for the sample and the measured ASC for each stock applied according to 4 research models. Table 4.6. ASC for the sample and individual stock Year 2015 2014 2013 2012 2011 2010 2009 Panel A. Results o