Results of research question 3

The third research question is made up of three sub-research questions:

i. “Are the determinants of capital structure significantly different between Malaysia and Indonesia?”

ii. “Are the determinants of capital structure that affect firm financial performance significantly different between Malaysia and Indonesia?”

iii. “Is the relationship between firm leverage and firm financial performance significantly different between Malaysia and Indonesia?”

The questions above were addressed by performing multi-group comparison analysis of the path coefficients of capital structure determinants and firm financial performance for Malaysia 75_.The

questions are typically applied to explore different characteristics such as country or gender. The different characteristics are recognized as heterogeneous data (Hair et al., 2013). The aim of multi-group analysis is to compare parameters (usually path coefficients) across two or more groups of data as to whether there are differences for each of the parameter group estimates (Hypothesis 4). From Table 4.6, I illustrate in Figure 4.3 the simple example of heterogeneity in the context of capital structure, in which the firm leverage (M) depends on two firm specific attributes: firm size (X1) and firm liquidity (X2). If the full data set is used and if I failed to

determine the heterogeneity between the pooled sample (countries), the path coefficient estimates would show substantial bias (Hair et al., 2013). That is, by using the full data set, both path coefficient estimates equal 0.02, leading to a conclusion that firm size and liquidity are equally important across pooled sample when in fact they are not. More specifically, when I split the pooled sample, the effect of liquidity (X2) on firm leverage (M) is much higher

in the Malaysian subsample - Group 1 (

p

(1)x₂

=0.03

; the superscript in parentheses indicates the group) than in the Indonesian subsample - Group 2 (

p

x(2)₂

=0.009

). Firm size (X1) exerts a

75 _{The test approach is also called modeling categorical moderation effects.}

106

greater influence on firm leverage in the Indonesian subsample than the Malaysian subsample. However, the question arises whether the path coefficients between countries are statistically significant. If there is a significant difference, what attributes contribute to the differences between countries? In testing this I assume two cases, a) standard errors of the two samples are equal and b) standard errors are unequal76_.

X1 (Firm size) X2 (Liquidity) Y (Leverage) Path coeff. β1= 0.02 Path coeff. β1= 0.02 X1 (Firm size) X2 (Liquidity) M (Leverage) Path coeff. β1= 0.004 Path coeff. β1= 0.03 X1 (Firm size) X2 (Liquidity) M (Leverage) Path coeff. β1= 0.04 Path coeff. β1= 0.009

Group 1 Malaysia (76.4% of the data)

Group 2 Indonesia (23.6% of the data) Full set of data

Figure 4-3 Heterogeneity in a structural model

76 _{Chin (2000) and Hair et al. (2013) propose that the first step in selecting the appropriate test statistic in PLS-MGA}

is to identify whether the standard errors can be assumed to be equal or unequal in the population. The test for standard error to be significantly different across groups can be assessed by means of Levene’s test (see Mooi and Sarstedt, 2011 that provides a more detailed discussion). In this study, Levene’s test (i.e., a test of homogeneity of variances) verified that the significance of the p-value is lower at p<0.01, which implies that we can reject the null hypothesis (see Appendix B of Table B-8). Thus, we can assume that the standard errors are unequal. However, as a robustness test I also consider the case where the standard errors are equal. Chin (2000), Chin et al. (2010) and Hair et al.(2013) proposed using the Smith-Satterthwaite test (SST) to compute its significance value76_{. The SST is}

computed in two ways, which is for assumed equal standard errors, and assumed unequal standard errors. If the equal standard errors are assumed, the t-value is computed as follows: 1 2

(1) 2 ( 2) 2 ( 1) (1) 2 ( 1) ( 2) 2 1 1 (1) ( 2) (1) ( 2) (1) ( 2) ( 2). ( ) ( 2). ( ) . sample sample n n n n n n n n path path se p se p t ₋θ θ ₋ + − + − − + + =

; and if unequal standard errors are assumed, the t-value is computed as follows: 1 2

(1) ( 2) ( 1) (1) 2 ( 1) ( 2) 2 (1) ( 2) ( . ( ) . ( ) sample sample n n n n path path se p se p t θ θ − − − + =

where: is the path coefficient for group one (Malaysia), is the path coefficient for group two (Indonesia), se(p(1)_{) is the standard error coefficient for group one (Malaysia) and se(}p(2)_{) is the standard error coefficient for}

group two (Indonesia), n(1) _{and n}(2) _{are the number of observations in group 1 and 2.} (1)

θ _θ(2)

107

3.8.3.1 Partial least squares of multi-group analysis (PLS-MGA)

Table 4-9 Multi-group comparison test results (PLS-MGA)

The path correlation coefficient (β) and standard errors (s.e) hypothesized between the exogenous variables and endogenous variables are generated from the PLS path modelling. The Beta (β) coefficient is be measured using resampling from the bootstrapping procedure for 5000 samples for all samples; sample Malaysia with N =5975 and sample Indonesia with N=1844. The SST (t-test) is used to test for the significance of the differences between the path estimators.

Panel A [diff] errors assumed Equal standard Unequal standard errors assumed

Asset structure (AS) -> firm leverage (LEV) 0.01 0.7707 0.8327 Growth opportunity (GRW) -> firm leverage (LEV) 0.0036 0.1836 0.2436 Firm size (FS) -> firm leverage (LEV) 0.0408 3.3658*** 3.8257*** Liquidity (LIQ) -> firm leverage (LEV) 0.0189 2.5606*** 2.5937*** Business risk (BR) -> firm leverage (LEV) 0.0023 2.1448** 1.7995* Non-debt tax shield (NDTS) -> firm leverage (LEV) 0.0361 1.9325* 1.7122* Bond market dev. (BMD) -> firm leverage (LEV) 0.0089 0.8802 0.7721 Stock market dev. (SMD) -> firm leverage (LEV) 0.0005 0.7694 0.8490 Economic growth (EG) -> firm leverage (LEV) 0.0717 6.8378*** 5.0107*** Interest rate (INT) -> firm leverage (LEV) 0.035 8.9490*** 7.4474*** Inflation rate (INF) -> firm leverage (LEV) 0.0037 0.2292 0.1670 Panel B

Asset structure (AS) -> firm performance (FFP) 0.0403 2.7155*** 3.0089*** Growth opportunity (GRW) -> firm performance (FFP) 0.0222 0.1786 0.1790 Firm size (FS) -> firm performance (FFP) 0.0004 2.5222** 2.0613** Liquidity (LIQ) -> firm performance (FFP) 0.002 1.1758 0.9037 Business risk (BR) -> firm performance (FFP) 0.0028 0.3917 0.2717 Non-debt tax shield (NDTS) -> firm performance (FFP) 0.0067 0.2419 0.2875 Bond market dev. (BMD) -> firm performance (FFP) 0.0244 1.6208 1.1408 Stock market dev. (SMD) -> firm performance (FFP) 0.0074 0.4795 0.6075 Economic growth (EG) -> firm performance (FFP) 0.0094 0.5015 0.4205 Interest rate (INT) -> firm performance (FFP) 0.044 3.1392*** 2.3985** Inflation rate (INF) -> firm performance (FFP) 0.0009 0.5625 0.4700

Firm leverage (LEV) -> firm performance (FFP) 0.0394 0.2846 0.2368 Notes: ***, **,*Statistically significant at the 1 per cent, 5 per cent and 10 per cent levels, respectively.

Table 4.9 shows the differences in the comparison of the path coefficient estimates (Malaysia vs. Indonesia), and provides the results of the multi-group comparison based on two ways of measuring, i.e., assumed equal standard errors and assumed unequal standard errors (Chin et al., 2010; Hair et al., 2013). I find consistent results for both measurements, which indicates that I cannot reject the null hypothesis that most of the path coefficients are equal across the two countries, Malaysia and Indonesia. In this case, I can conclude that most of the impact of firm- and country-specific attributes on firm leverage and firm financial performance is equal in both countries (Hypotheses 4-E1 and 4-E2). This is consistent with previous studies that make implicit assumptions that the impact of firm-specific attributes on firm leverage is equal (Booth et al., 2001; Deesomsak et al., 2004; Giannetti, 2003)77_{. However, I demonstrate that some}

77 _{They used a single average regression framework for an observation.}

108

attributes are significantly different across countries. I find that the impact of the firm size, liquidity, business risk, non-debt tax shield, economic growth, and interest rate on firm leverage are significantly different between Malaysia and Indonesia. I also find that the relationship of asset structure, firm size and interest rate to firm financial performance are significantly different between the two countries. For the capital structure determinants (which include firm- and country-specific attributes), I reject the null hypothesis that all path coefficient estimates for Malaysia and Indonesia are equal. This result supports Jong et al. (2008) and Psillaki and Daskalakis (2009) who found that the impact of firm specific attributes on firm leverage is not necessarily equal across countries. My results also indicate that there is no significant difference between Malaysia and Indonesia in terms of the effect of firm leverage on firm financial performance (hypothesis 4-E3). This means that the countries tend to have equal impact on firm leverage (particularly using external financing instead of internal financing) to enhance firm financial performance. This is consistent with Harris and Raviv (1991), Jensen (1986) and TOT that predict a positive relationship.

In summary, I acknowledge that some of the impact of firm- and country-specific attributes on firm leverage and firm financial performance differs in terms of sign, magnitude and significance level in Malaysia and Indonesia. From multi-group comparison test results, I reject the hypotheses that some firm- and country-specific coefficients are equal across countries. This indicates that an additional contribution is that the often made implicit assumption of equal impact of such relationships across countries does not hold.

In document Three essays on capital structure determinants (Page 118-121)