Analyzing the Elasticities of Coal and Hydropower Technology Towards Malaysian Electricity Generation

Model Construction and Data Sources

Theory is crucial for research as it is the foundation for model development and testing. Thus, this study applies the augmented Cobb–Douglas (AC-D) production function by including capital, labor, GDP and energy (coal and hydropower technology) to examine the relationship among these variables (Dogan & Ozturk, 2017; Shahbaz et al., 2015) by investigating the electricity generation as the dependent variable (Yoo & Kim, 2006). The specific form of AC-D production function is constructed as in Eq. (1).

$EG=A{K}^{{\propto }_1}{L}^{{\propto }_2}{Y}^{{\propto }_3}{C}^{{\propto }_4}{H}^{{\propto }_5}{e}^u$ (1)

where EG is electricity generation; α₁, α₂, α₃, α₄ and α₅ are coefficients of capital (K), labour (L), economic growth (Y), coal (C) and hydropower (H), respectively. Then, all the series are converted into natural logarithms to linearize the form of Eq. (1). This transformation will capture the interpretation of elasticities (Bekhet & Harun, 2017; 2018; Shahbaz & Lean, 2012) to the response of EG with respect to the aforesaid variables. The natural log-linear functional form of the AC-D production function is modeled as in Eq. (2).

${EG}_t={\propto }_0+{\propto }_1{lK}_t+{\propto }_2{lL}_t+{\propto }_3{lY}_t+{\propto }_4{lC}_t+{\propto }_5{lH}_t+e_t$ (2)

Subsequently, the Eq. (2) is stretched to Autoregressive Distributed Lag (ARDL) model to assess the short-run and long-run elasticities. So, the ARDL model is derived as in Eq. (3).

$∆l{EG}_t={\beta }_0+\underset{s=1}{\overset{p}{\sum }}{\beta }_1{lEG}_{t-s}+\underset{s=0}{\overset{p}{\sum }}{\beta }_2{lK}_{t-s}+\underset{s=0}{\overset{p}{\sum }}{\beta }_3{lL}_{t-s}+\underset{s=0}{\overset{p}{\sum }}{\beta }_4{lY}_{t-s}+\underset{s=0}{\overset{p}{\sum }}{\beta }_5{lC}_{t-s}+\underset{s=0}{\overset{p}{\sum }}{\beta }_1{H}_{t-s}+{\propto }_1{lK}_t+{\propto }_2{lL}_t+{\propto }_3{lY}_t+{\propto }_4{lC}_t+{\propto }_5{lH}_t+e_t$ (3)

where lEG , lK , lL , lY , lC , and lH are the natural logarithm of the variables. ${\beta }_0$ is a constant value, ${\beta }_{i}s$ and ${\propto }_{i}s$ denotes the short-run and long-run elasticities of variables, respectively. e specifies the stochastic error terms presumed to be normally distributed, alongside white noise (Bento & Moutinho, 2016), and t signifies the yearly time-series data. The data set covers the 1978–2018 period consist of selection for 6 variables. The data of electricity generation (EG), coal (C) and hydro (H) are assembled from Energy Commission, Malaysia and assessed by kilo tonne of equivalent (ktoe). The proxy of capital (K) is Gross Fixed Capital Formation (GFCF) and the proxy of Y is GDP, which were obtained from the Malaysian Department of Statistics. Labor is proxied by labor force which was extracted from World Development Indicator (WDI).

Research Methodology

Data quality tests were used to identify the basic features of the data that align with the theory. The augmented Dickey–Fuller (ADF, 1979) and Phillips–Perron (P-P, 1988) assessments were used to validate the stationarity of data series results. When these tests are accomplished, the F-Bound test was employed to determine the long-run dynamic relationship among the variables. This test is advantageous in comparison with other traditional tests (Bekhet & Harun, 2017; Marques et al., 2016). This enables the utilization of a mixture of stationary of I (0) and I (1) variables. This test is also superior in small samples, 30 ≤ t ≤ 80. This test also enables the study to obtain different optimal lags of the variables. Finally, this study eradicates endogeneity issues linked with the Engle–Granger method based on the assumption that all variables are endogeneous variables.

Once the existence of a long-run dynamic relationship is confirmed, the long-run and short-run elasticities of EG responses to the changes of its determinants are examined using the associated ARDL model. Finally, to ensure the error term ( e _t ) is robust and accurate, the diagnostic and stability tests are applied. The diagnostic tests include the Jarque–Bera tests for normality, the Breusch–Godfrey Lagrange multiplier (LM) for serial correlation test, autoregressive conditional heteroscedasticity (ARCH) for heteroscedasticity test and Ramsey regression equation specification error test (RESET) to correct the functional form. Then, the cumulative sum (CUSUM) test was performed to verify the presence of parameter stability. The suitability of methodologies application is advantageous to accomplish the postulated objectives and the hypotheses are formulated as follows:

H₁₁: Significant dynamic relationship exists between EG and its determinants in Malaysia.

H₁₂: Estimation of long-run and short-run elasticities between EG and its determinants in Malaysia

Preliminary Analysis Results

Table 01 presents the descriptive statistics result of these five variables in this study. It reports that the result of Jarque-Bera test strongly accepts normality of aforesaid variables.

Unit Root Tests Result.

While preliminary test results are detected, the unit root tests are examined. Thus, Table 03 shows the results of stationary properties for each variable. It indicates that all variables are not stationary at level, I(0). However, after taking the first difference of all variables, the series are found to be stationary at I(1).

Long-run Dynamic Relationship Result

Despite the fact that all variables are integrated of order one, it is a necessary condition to test a long-run dynamic relationship among the variables in the system. Thus, the F-Bound test is applied, however, it is sensitive to the lag selection; then, we found the appropriate AIC maximum lag length is 3. The outcomes indicate the presence of a long-run dynamic relationship among the variables because the calculated F-statistics (5.301) exceeded the upper critical bound, at 1%, 5%, and 10% levels. Correspondingly, this outcome is permitted to attain long- and short-run elasticities.

Long-run and Short-run Elasticities Result

Table 04 shows that all the elasticities estimation have positive signs indicating that these determinants have positive contribution to the electricity generation in Malaysia except for labour and hydropower technology. The finding reveals that both determinants, labour and hydropower technology have negative elasticity and inelastic, towards the electricity generation in Malaysia.

Furthermore, the finding indicate that coal has a positive elasticity but inelastic. It shows that 1% increase in coal effecting the electricity generation increased by 7%. But, the increasing of coal is less sensitive to the increasing of electricity generation in Malaysia. On the other hand, the result shows the negative and inelastic elasticity for the hydropower technology. Yet, both these results show that Malaysia’s government depending more on the supply of coal rather than hydropower to generate electricity. Alternatively, the hydropower technology is applicable to generate electricity in Malaysia. If the number of hydropower technology is in absolute value, the electricity generation increased by 8.1%, which is 1% higher than coal production (see Table 04 ).

Similar to long-run dynamic elasticities, the short-run dynamic elasticities have a mix sign indicating that all the determinants have positive elasticities except for coal and hydropower (see Table 05 ). This result reveals that the elasticity of electricity generation with respect to coal is negative inelastic. It is noticeable difference in the long-run analysis, then, the effect of coal to the electricity generation is relatively smaller in the short-run, for the absolute term. Similarly, the elasticity of hydropower is negative inelastic, but the effect of hydropower in the long-run is better rather than in the short-run, for the absolute term.

Diagnostic Test and Stability Test Results

The diagnostic test is essential tool to validate the quality of the estimation for the ARDL model. The result of Jarque-Bera test shows that the error term is normal distributed. Meanwhile, the Breusch-Godfrey Serial Correlation LM test is found that the error term is free from serial correlation. Also, the error term is free from heteroscedasticity problem using Breusch-Pagan-Godfrey Heteroscedasticity and ARCH Heteroscedasticity tests. Furthermore, the Ramsey’s RESET indicates that the functional form for the model used is well specified.

Once the model is validated as the best model for goodness of fit and does not violate the condition of OLS assumption, it is necessary to ensure the model is stable. In order to evaluate the stability of the model, the CUSUM statistic plot is applied (Bélaïd & Youssef, 2017; Bekhet & Harun, 2017; Shahbaz et al., 2015). This test capable to detect the stability model by graphically for two statistics are plotted within two straight lines. As a result, the plots of this test are fluctuating within the line and significant at 5 percent for the 1978 to 2018 period. Accordingly, this result suggests that all coefficients in the model [see Eq. (3)] are stable and robust over that sample period.