Chapter 4: Green bond market drivers and implications for sustainability
4.4. Results
4.4.1. Explanatory factor analysis (EFA) results
121
Governance Indicators (WGI) (2019b). The sources and descriptions of each variable included in this study are outlined in the Table 4.3.1.
122 Table 4.4.1: Cronbach’s alpha
Item Obs. Sign item-test correlation
item-rest correlation
average inter-item covariance
alpha
TF 2411 + 0.4082 0.3782 87.3049 0.6987
RL 2411 + 0.3902 0.3868 90.29595 0.7063
RQ 2411 + 0.4465 0.4435 90.27126 0.7063
GDP 2411 + 0.8339 0.8307 88.24597 0.7007
ST 2114 + 0.7866 0.779 86.26103 0.6976
TO 2411 - 0.7616 0.671 61.75313 0.6337
NDC 2411 + 0.6023 0.5898 87.49668 0.6988
NDCR 2411 + 0.5288 0.4603 79.66495 0.681
OECD 2411 + 0.4055 0.404 90.52142 0.7069
REER 2411 + 0.5933 0.4739 71.11419 0.666
IRS 2411 - 0.2325 0.211 89.3499 0.7041
INF 2411 - 0.1347 0.1153 90.08954 0.706
MCDC 2411 + 0.9186 0.8042 33.70205 0.6005
DCPSB 2411 - 0.8508 0.6876 41.5877 0.6238
Test scale
77.70172 0.7033
Factor loadings for the real effective exchange rate (REER), interest rate spread (IRS), market capitalization of domestic companies (MCDC), and domestic credit provided to the private sector by banks (DCPSB) were below the 0.6 cut-off line. The uniqueness of the factor loading for IRS was also above the 0.6 threshold of acceptance. Among the variables with acceptable factor loadings, trade freedom (TF), rule of law (RL), regulatory quality (RQ), and Organization for Cooperation and Economic Development membership (OECD) were manifest variables of Factor 1; gross domestic product (GDP), stocks traded (ST), trade openness (TO), nationally determined contributions to the Paris Agreement (NDC), and the rank of those nationally determined contributions (NDCR) were manifest variables of Factor 2; and inflation (INF) was the sole manifest variable of Factor 3.
The latent factor correlation matrices revealed the highest correlation between any two factors to be the 0.4105 between Factor 1 and Factor 2. The Bartlett test of sphericity and
Kaiser-123
Meyer-Olkin Measure of Sampling Adequacy also indicated good model fit, the former showing a statistically significant p-value (0.000) and the latter displaying a value above the acceptable minimum of 0.70.
Table 4.4.2: EFA results, stage 1
Variable Factor 1 Factor 2 Factor 3 Uniqueness
TF 0.9744 -0.1076 0.079 0.0986
RL 0.9703 -0.057 0.097 0.0656
RQ 0.9674 -0.007 0.0372 0.0582
GDP 0.0427 0.8658 0.2491 0.1175
ST 0.0552 0.8362 0.3282 0.1015
TO 0.0592 -0.8413 0.0843 0.3329
NDC 0.1668 0.7307 -0.2186 0.3272
NDCR -0.3739 0.886 -0.0218 0.3477
OECD 0.9093 -0.0073 0.0963 0.1446
REER 0.1512 0.4794 0.496 0.3813
IRS -0.4987 0.0231 -0.247 0.6655
INF -0.2694 0.0647 -0.6966 0.4069
MCDC 0.4325 0.566 0.0131 0.2886
DCPSB -0.5053 -0.461 0.511 0.1915
Table 4.4.3: Correlation matrix of Kaiser promax-rotated common factors, stage 1 Factors Factor 1 Factor 2 Factor 3
Factor 1 1.0000
Factor 2 0.4105 1.0000
Factor 3 0.1409 0.08291 1.0000
Accordingly, it is appropriate to perform a second round of analysis after eliminating the aforementioned variables with factor loadings below the 0.6 cut-off mark. Tables 4.4.5, 4.4.6.,
124
and 4.4.7 display the factor loadings, latent variable correlations, and fitness indices of the second stage of analysis, respectively.
Table 4.4.4: Determinant of the correlation matrix, Bartlett test of Sphericity, and Kaiser-Meyer-Olkin Measure of Sampling Adequacy, stage 1
Determinant of the correlation matrix Det = 0.000
Bartlett test of sphericity
Chi-square = 36674.963 Degrees of freedom = 91 p-value = 0.000 H0: variables are not intercorrelated
Kaiser-Meyer-Olkin Measure of Sampling Adequacy
KMO = 0.735
Table 4.4.5: EFA results, stage 2
Variable Factor 1 Factor 2 Uniqueness
TF 0.9608 -0.0545 0.1077
RL 0.9724 -0.0066 0.0584
RQ 0.9559 0.0299 0.0668
GDP 0.1601 0.8622 0.1418
ST 0.193 0.8328 0.1654
TO 0.0329 -0.8263 0.3337
NDC 0.149 0.7246 0.383
NDCR -0.3495 0.8773 0.3062
OECD 0.9159 0.0534 0.1267
INF -0.4719 0.0185 0.7826
125
Table 4.4.6: Correlation matrix of Kaiser promax-rotated common factors, stage 2
Table 4.4.7: Determinant of the correlation matrix, Bartlett test of Sphericity, and Kaiser-Meyer-Olkin Measure of Sampling Adequacy, stage 2
Determinant of the correlation matrix Det = 0.000
Bartlett test of sphericity
Chi-square = 27461.809 Degrees of freedom = 45 p-value = 0.000 H0: variables are not intercorrelated
Kaiser-Meyer-Olkin Measure of Sampling Adequacy
KMO = 0.772
Of the ten variables assessed, only INF had unacceptable factor loadings and uniqueness.
Moreover, the number of latent factors was truncated to Factor 1 and Factor 2, with TF, RL, RQ, RQ, and OECD observed in Factor 1 and GDP, ST, TO, NDC, and NDCR observed in Factor 2.
There was thus a near-even split in the number of manifest variables in each latent factor, which will provide unique implications for the determining and labelling the latent variables for subsequent structural equation modelling.
The latent variable correlation matrices reveal a 0.3230 correlation between the variables, which is neither exceptionally low nor high. Moreover, the Bartlett test of sphericity p-value of 0.000 and KMO value of 0.772 in both indicate acceptable model fit. Upon dropping the INF
Factors Factor 1 Factor 2 Factor 1 1.0000
Factor 2 0.3230 1.0000
126
variable that lacked appropriate factor loadings and uniqueness, a third and final round of EFA was performed to isolate the remaining variables within the latent variables to which they are most suited. Tables 4.4.8, 4.4.9, and 4.4.10 provide the EFA results, correlation between latent factors correlations, and model fitness indices for this analytical stage.
Table 4.4.8: EFA results, stage 3
Variable Factor 1 Factor 2 Uniqueness
TF 0.9631 -0.0549 0.1030
RL 0.9787 -0.0085 0.0474
RQ 0.9641 0.0271 0.0532
GDP 0.1601 0.8629 0.1419
ST 0.1857 0.8364 0.1672
TO 0.0155 -0.8194 0.3364
NDC 0.1618 0.7195 0.3821
NDCR -0.3511 0.8781 0.3016
OECD 0.9259 0.0499 0.1109
As with the second round of EFA, TF, RL, RQ, GDP, and OECD all revealed high factor loadings above 0.9 for Factor 1. Though not quite as high, factor loadings for GDP, ST, TO, NDC, and NDCR generally remained above 0.80 (with the exception of the 0.7295 loading for the NDC variable), similarly expressing high values. Once again, none of these variables showed unacceptably high uniqueness, so they will be retained. Furthermore, correlations between latent Factor 1 and Factor 2 dropped only slightly to 0.3179, and both Bartlett and KMO values for model fitness remained above acceptable lower limits. As such the observed variables will be retained for SEM analysis.
127
Table 4.4.9: Correlation matrix of Kaiser promax-rotated common factors, stage 3
Factors Factor1 Factor2
Factor1 1.0000
Factor2 0.3179 1.0000
Table 4.4.10: Determinant of the correlation matrix, Bartlett test of Sphericity, and Kaiser-Meyer-Olkin Measure of Sampling Adequacy, stage 3
Determinant of the correlation matrix Det = 0.000
Bartlett test of sphericity Chi-square = 26894.679 Degrees of freedom = 36 p-value = 0.000 H0: variables are not intercorrelated
Kaiser-Meyer-Olkin Measure of Sampling Adequacy
KMO = 0.769
At this stage, latent variables are labelled based upon the nature of their underlying observed variables. It is first worth mentioning that OECD, NDC, and NDCR will be retained as exogenous variables instead of manifest variables of their latent factors for few specific,
exceptional cases. In the case of the former, the majority of green bonds were issued in the U.S., France, and other OECD member states (CBI, 2018), a substantial number and volume of green bonds proceeds were allocated in OECD member states (Tolliver, Keeley, & Managi, 2019), and the majority of green bonds are issued in OECD member states in the Eurozone (Environmental Finance, 2019), including OECD membership as an exogenous dummy variable is reasonable. In the case of the latter two, this study created a normalized index of the strength of Paris NDCs for
128
each of the assessed countries. Due to the novelty of the variables themselves, as well as to a priori observations of the many factors that influence the strength of each country’s NDCs (including those beyond the scope of the latent factors assessed in the previous EFA), including the NDC and NDCR variables as exogenous variables would allow for more meaningful
assessments of the correlation between precedent-setting climate policies (i.e. Paris NDCs) and the green bond branch of environmental finance.
With the aforementioned exogenous variables accounted for, the remaining manifest variables of each Factor were used to determine and label the latent variables assessed in the SEM analysis. Comprised of the TF, RL, and RQ variables that relate to the quality of governance, regulatory procedures, and institutional legitimacy of their underlying societies, Factor 1 was labelled “Institutional and Regulatory Environment.” Consisting of the TO, GDP, and ST variables linked to economic development, financial market maturity, international trade capacity, Factor 2 was labelled “Macroeconomic Conditions.” Both of these labels were
supported by similar factor groupings and labels in existing literature (Bae, 2012). Furthermore, these latent and manifest variable grouping are listed alongside the independently observed exogenous variables in Table 4.4.11.
Table 4.4.11: Latent factors and constructs for green bond issuances Factor 1:
Institutional and Regulatory Environment
Factor 2:
Macroeconomic Conditions
Exogenous Variables
TF TO NDC
RL GDP NDCR
RQ ST OECD
129
The nature of the NDC, NDCR, and OECD variables and the aims of this study justify their inclusion as exogenous variables in the SEM analysis. As investment target frameworks, NDCs are separate from yet interrelated with institutional and macroeconomic drivers, and measuring their direct impacts on green bond issuances is the primary focus of this study.
Though OECD membership reflects an institutional affiliation, it requires that countries possess both macroeconomic features (e.g. rules-based open market economies, stable and transparent financial systems) and institutional features (e.g. governmental, administrative, and monetary institutions and regulations) before becoming member states (OECD, 2017). As such, this study treated it as separate from yet interrelated with both factors and included it alongside NDC as a unique exogenous variable.