The Impact of Renewable Energy Consumption on Economic Growth in Bangladesh: Evidence from ARDL and VECM Analyses

1.png

Figure 1. Types of renewable energies and their share in the electricity generation (made by authors using data from [8])

Table 1. Renewable energy’s share in electricity generation

Types of Reneable Energy	Off-Grid (MW)	On-Grid (MW)	Total (MW)
Solar	319.9	81.36	401.26
Hydro	0	230	230
Wind	2	0.9	2.9
Biogas	0.63	0	0.63
Biomass	0.40	0	0.40
Total Energy	322.93	312.26	635.19

Note: Off-grid and On-grid electricity share of different renewable energies [8].

2.1 Solar energy

Solar power is the world's leading clean and sustainable energy source. Considering its geographical location, Bangladesh has a significant capacity for harnessing solar energy; Halder et al. [16] reported an average daily solar radiation ranging from 4 to 4.5 kWh/m² per day. An estimated 401.26 megawatts (MW) of electricity are currently produced from solar energy sources, contributing 3.2% of the country's overall renewable energy production (Table 1). Off-grid solar home systems significantly contribute to the total solar electricity production. Most solar energy comes from off-grid sources. According to the report of the global energy research organization Renewable Energy Policy Network for the 21st Century [17], Bangladesh has the most extensive solar home system (SHS) globally, with more than four million SHS units across the country. Thus, the report affirms and strengthens Bangladesh's positive outlook on solar energy.

2.2 Hydro energy

At present, hydropower continues to stand out as a primary renewable energy source, holding its position of dominance in the energy sector. In 2018, the total global electricity production from hydropower was 4,210 terawatt hours (TWh), or 15.77% of 26,700 TWh of global electricity [18]. According to Liu et al. [19], the only hydroelectric power plant in Bangladesh, the Karnafuli hydroelectric power plant, was established in 1962. This power plant generates 230 megawatts (MW), or 36.2% of the renewable electricity production (Table 1). Bangladesh has more potential as a hydroelectric energy source.

2.3 Wind energy

Currently, wind power is a rapidly expanding renewable energy source. Bangladesh has the world's longest coastal belt of approximately 724 km along the Bay of Bengal and is another potential renewable energy source. Currently, the total electricity generation capacity is 2.9 megawatts (MW) and accounts for only 0.46% of Bangladesh's overall renewable energy production (Table 1). Wind energy technology is still emerging in Bangladesh and requires more investment and technological development.

2.4 Biomass energy

According to Gumartini [20], biomass energy serves as a direct or indirect energy source for approximately 80% of Bangladesh’s population. Bangladesh’s weather is favorable for generating electricity from biomass. Biomass has a significant potential for electricity generation. Many types of agricultural waste and wood are sources of biomass. Hence, biomass energy has emerged as a favorable renewable energy option, particularly for countries reliant on agriculture, such as Bangladesh [16].

2.5 Biogas energy

Biogas is the most popular renewable energy source in Bangladesh. Animal and municipal wastes are used as raw materials in biogas plants. More than 45,000 biogas plants are currently in operation in Bangladesh. Usually, cow dung is used to produce gas for cooking and to generate power in households [21]. Thus, this source also has enormous prospects in the context of Bangladesh.

4.1 Data

This study employs data from 1980 to 2018 to examine the relationship between economic growth and energy consumption, focusing specifically on renewable energy and domestic natural gas in Bangladesh. The data sources were World Development Indicators (WDI) [43], The U.S. Energy Information Administration (EIA) [44], and British Petroleum (BP) [45] collect data on Gross Domestic Product (GDP) (constant 2010 US$), renewable energy consumption (quadrillion Btu), and natural gas consumption (mtoe).

4.2 Model specification

This study utilizes a multivariate regression model to examine the correlation between renewable energy consumption and economic growth in Bangladesh, allowing for an analysis of both the short- and long-term dynamics between these variables. The model is specified in the following manner:

$R E C_t=\beta_1+\beta_2 G D P_t+\beta_3 N G C_t+\varepsilon_t$               (1)

The log form of the Eq. (1) as:

$\operatorname{LnREC}_t=\beta_1+\beta_2 \operatorname{LnGDP}_t+\beta_3 L n N G C_t+\varepsilon_t$             (2)

where, variables are in natural logarithm form. $L n R E C_t$ represents renewable energy consumption, $L n G D P_t$ represents real GDP, $\operatorname{LnNGC}_t$ represents natural gas consumption and $\varepsilon_t$ represents the error term. In the first stage of the analysis, the long-run relationships among the variables are investigated through the ARDL method developed by Pesaran et al. [12]. The second step of the analysis involved investigating the causal relationships among the variables using the Vector Error Correction (VEC) model-based Granger causality test.

4.3 Cointegration analysis (Autoregressive Distributive Lag)

Pesaran and Shin [46] introduced the Autoregressive Distributive Lag (ARDL) bounds testing method, which was later refined and expanded upon by Pesaran et al. [12]. This method is widely utilized to examine the relationships between variables and has proven to be a valuable analytical approach. Compared with other commonly used cointegration tests, this method offers several advantages. This method can be used independently of the order of integration of the series, regardless of whether they are integrated with order zero (I(0)) or order one (I(1)). This suggests that there is no need for unit-root pre-testing in the ARDL approach. Second, this method can use a small sample, whereas larger samples require other conventional cointegration tests. Finally, this approach allows different variables to have different lags than those in other conventional cointegration tests [23].

4.4 Unit root test

The application of the ARDL method is contingent on the integrated order of the series, which should be either I(0), I(1), or a combination of both I(0) and I(1). However, unit-root pretesting does not require the ARDL method for cointegration. It is crucial to ascertain that none of the variables exhibits an integrated order of two (I(2)) before proceeding with the analysis. Augmented Dickey-Fuller (ADF) [47] and Phillips-Perron (PP) [48] tests were conducted to assess the stationarity and integration order of all series, and these tests help evaluate the characteristics of stationarity and integration exhibited by the series being examined. Moreover, the robustness of the results was verified using both tests.

4.5 Cointegration test

The following equations express the ARDL model using a cointegration relationship:

$\begin{aligned} & \Delta L n R E C_t=\alpha_0+\sum_{i=1}^p \alpha_{1 i} \Delta L n R E C_{t-i} \\ & +\sum_{i=0}^q \alpha_{2 i} \Delta L n G D P_{t-i} \\ & +\sum_{i=0}^r \alpha_{3 i} \Delta L n N G C_{t-i} \\ & +\alpha_4 \operatorname{LnREC_{t-1}}+\alpha_5 \operatorname{LnGDP_{t-1}} \\ & +\alpha_6 L n N G C_{t-1}+u_t \\ & \end{aligned}$                     (3)

$\begin{aligned} & \Delta L n G D P_t=\beta_0+\sum_{i=1}^p \beta_{1 i} \Delta L n G D P_{t-i} \\ & +\sum_{i=0}^q \beta_{2 i} \Delta L n N G C_{t-i} \\ & +\sum_{i=0}^r \beta_{3 i} \Delta \operatorname{LnREC_{t-i}} \\ & +\beta_4 \operatorname{LnREC}_{t-1}+\beta_5 \operatorname{LnGDP_{t-1}} \\ & +\beta_6 L n N G C_{t-1}+u_t \\ & \end{aligned}$             (4)

$\begin{aligned} & \Delta L n N G C_t=\gamma_0+\sum_{i=1}^p \gamma_{1 i} \Delta L n N G C_{t-i} \\ & +\sum_{i=0}^q \gamma_{2 i} \Delta \operatorname{LnREC_{t-i}} \\ & +\sum_{i=0}^r \gamma_{3 i} \Delta L n G D P_{t-i} \\ & +\gamma_4 L n R E C_{t-1}+\gamma_5 \operatorname{LnGDP_{t-1}} \\ & +\gamma_6 L n N G C_{t-1}+u_t \\ & \end{aligned}$                (5)

where, each variable is expressed in its natural logarithmic form. LnREC, LnGDP, and LnNGC are renewable energy consumption, real GDP, and natural gas consumption respectively. $u_t$ is the error term, and $\Delta$ is the first difference operator. We determined the proper lag for the study using the Akaike Information Criterion (AIC).

The cointegration analysis employs the bounds test, which relies on the combined F-statistic. The null hypothesis of no cointegration in Eqs. (3)-(5) can be expressed as H₀: α_k = 0, β_k = 0, γ_k = 0 for k = 4, 5 and 6, respectively. Cointegration was determined using two sets of critical values. The null hypothesis was rejected when the test statistic exceeded the corresponding critical value and cannot be rejected if the test statistic is lower than the critical value. If the test statistic falls within the critical range, the co-integration test does not yield definitive evidence and is considered inconclusive. We examined the error correction term within the ARDL model to confirm the existence of cointegration among the variables. Two sets of critical values obtained from different sources were used for assessment. Pesaran et al. [12] specify critical values for sample sizes of 500 and 1000 observations, while Narayan [49] offers critical values that cover a range of sample sizes from 30 to 80 observations. Critical values were to use the sign and statistical significance of the error-correction. This study was based on a dataset comprising annual data, with a limited sample size. Hence, this study employs the critical values from Narayan [49], assuming that one variable in the ARDL model is I(0), whereas the remaining variables are assumed to be I(1). When cointegration exists among the variables, the long-term model and short-term dynamics can be described as follows:

Long-run models

$\begin{aligned} \operatorname{LnREC}_t=\alpha_0+ & \Sigma_{i=1}^p \alpha_{1 i} \operatorname{LnREC}_{t-i} \\ & +\sum_{i=0}^q \alpha_{2 i} \operatorname{LnGDP}_{t-i} \\ & +\sum_{i=0}^r \alpha_{3 i} \operatorname{LnNGC}_{t-i}+v_t\end{aligned}$                          (6)

$\begin{aligned} \operatorname{LnGDP}_t=\beta_0+ & \sum_{i=1}^p \beta_{1 i} \operatorname{Ln} G D P_{t-i} \\ & +\sum_{i=0}^q \beta_{2 i} \operatorname{Ln} N G C_{t-i} \\ & +\sum_{i=0}^r \beta_{3 i} \operatorname{LnREC}_{t-i}+v_t\end{aligned}$                    (7)

$\begin{aligned} \operatorname{LnNGC}_t=\gamma_0+ & \Sigma_{i=1}^p \gamma_{1 i} \operatorname{LnNGC}_{t-i} \\ & +\sum_{i=0}^q \gamma_{2 i} \operatorname{LnREC}_{t-i} \\ & +\sum_{i=0}^r \gamma_{3 i} \operatorname{LnGDP}_{t-i}+v_t\end{aligned}$                (8)

Short-run models

$\begin{aligned} \Delta \operatorname{LnREC}_t=\alpha_0+ & \sum_{i=1}^p \alpha_{1 i} \Delta \operatorname{LnREC}_{t-i} \\ & +\sum_{i=0}^q \alpha_{2 i} \Delta \operatorname{LnGDP}_{t-i} \\ & +\sum_{i=0}^r \alpha_{3 i} \Delta \operatorname{LnNGC}_{t-i}+\delta_1 \epsilon_{t-1} \\ & +e_t\end{aligned}$                         (9)

$\begin{aligned} \Delta \operatorname{LnGDP}_t=\beta_0+ & \sum_{i=1}^p \beta_{1 i} \Delta \operatorname{LnGDP}_{t-i} \\ & +\sum_{i=0}^q \beta_{2 i} \Delta \operatorname{LnNGC}_{t-i} \\ & +\sum_{i=0}^r \beta_{3 i} \Delta \operatorname{LnREC}_{t-i}+\delta_2 \epsilon_{t-1} \\ & +e_t\end{aligned}$                 (10)

$\begin{aligned} \triangle L n N G C_t=\gamma_0+ & \sum_{i=1}^p \gamma_{1 i} \Delta \operatorname{LnNGC}_{t-i} \\ & +\sum_{i=0}^q \gamma_{2 i} \Delta \operatorname{LnREC}_{t-i} \\ & +\sum_{i=0}^r \gamma_{3 i} \Delta \operatorname{LnGDP}_{t-i}+\delta_3 \epsilon_{t-1} \\ & +e_t\end{aligned}$                    (11)

where, $\delta$ is the coefficient of error-correction term $\left(\epsilon_{t-1}\right)$ acquired from the long-run equilibrium model. It provides the indication of the rate at which variables move towards equilibrium, and the coefficient with a negative sign is expected to have a statistically significant value, as highlighted by Kremers et al. [50].

4.6 Causality analysis

ARDL cointegration was employed to ascertain long-run relationship between the variables. Engle and Granger [51] proposed that the cointegration between variables indicates a possible causal relationship in at least one direction. Granger causality refers to the concept that if the previous values of a time series, $X_t$ precede the value of another time series, $Y_t$, then $X_t$ is considered to have a Granger-causal relationship with $Y_t$. Nonetheless, the ARDL model does not provide information regarding the causal direction between variables. Furthermore, conducting a causality analysis using a vector model with an error term confirms the existence of a causal relationship, even if the estimated coefficients of the lagged variables of interest are not statistically significant. The VEC model is employed to investigate the presence of Granger causality in both the short and long terms between pairs of variables. The Eqs. (12)-(14) of the VEC model were reformulated to incorporate the relationships described in Eqs. (9)-(11).

$\begin{aligned} \triangle \operatorname{LnREC}_t=c_1+ & \sum_{i=1}^n \beta_{11 i} \Delta \operatorname{LnREC}_{t-i} \\ & +\sum_{i=1}^n \beta_{12 i} \Delta \operatorname{LnGDP}_{t-i} \\ & +\sum_{i=1}^n \beta_{13 i} \Delta \operatorname{LnNGC}_{t-i}+\lambda_1 \epsilon_{t-1} \\ & +\omega_{1 \mathrm{t}}\end{aligned}$             (12)

$\begin{aligned} \Delta {LnGDP}_t=c_2+ & \sum_{i=1}^n \beta_{21 i} \Delta \operatorname{LnREC}_{t-i} \\ & +\sum_{i=1}^n \beta_{22 i} \Delta \operatorname{LnGDP}_{t-i} \\ & +\sum_{i=1}^n \beta_{23 i} \Delta \operatorname{LnNGC}_{t-i}+\lambda_2 \epsilon_{t-1} \\ & +\omega_{2 t}\end{aligned}$                 (13)

$\begin{aligned} \Delta {LnNGC}_t=c_3+ & \sum_{i=1}^n \beta_{31 i} \Delta {LnREC}_{t-i} \\ & +\sum_{i=1}^n \beta_{32 i} \Delta {LnGDP}_{t-i} \\ & +\sum_{i=1}^n \beta_{33 i} \Delta {LnNG}_{t-i}+\lambda_3 \epsilon_{t-1} \\ & +\omega_{3 \mathrm{t}}\end{aligned}$                          (14)

Short and long-run causality can be evaluated by testing the variables in each equation. Short-term causality can be examined by considering the variables in their first differences, while long-term causality can be assessed by incorporating the error correction terms derived from the long-run models. If the null hypothesis ( $H_0: \beta_{21 i}=0$ for all i ) is rejected in Eq. (13), indicates the presence of short-run Granger causality from Renewable Energy Consumption (REC) to Gross Domestic Product (GDP). In the short term, the findings provide evidence of a causal relationship between REC and GDP. In contrast, the error correction term ($\epsilon_{t-1}$) signifies the long-term causal relationship in the equation. The coefficient values demonstrate the speed at which the system adjust deviations from the long-run equilibrium, as indicated by the error term $\epsilon_{t-1}$. Thus, long-run causalities can be found by testing $H_0: \lambda_i=0$ for all $i$  in Eqs. (12)-(14). Finally, strong causalities are examined by joint F-tests: $H_0: \beta_{12 i}=\lambda_i=0$ and $H_0: \beta_{13 i}=\lambda_i=0$ for all $i$ in Eq. (12).

The present global energy sector is more diverse than ever before, yet non-renewable energies still dominate, and the role of fossil fuels needs to be addressed. Countries are focusing increasingly on sustainable, efficient, and renewable energy sources. Bangladesh has been increasing its use of renewable energies, which has positively impacted electrification and energy access. This study looked at the shift from conventional energy sources to sustainable energy sources and the short and long run relationship between energy consumption and economic growth in Bangladesh from 1980 to 2018. It found a robust causal association between these two variables, suggesting that an increase in the use of renewable energy stimulates economic growth or contributes to it. This finding raises a concern about policy direction regarding the changing proportion of renewable energy in the overall energy composition. Adopting renewable energy helps decrease reliance on non-renewable sources and improves environmental quality.

Whether renewable energy consumption can drive sustainable economic growth in Bangladesh remains to be seen. This study’s findings suggest a unidirectional long-term relationship and robust causal association from renewable energy consumption to economic growth, without any supporting confirmation of a short-term causal relationship between the two variables. The study affirms the credibility of the conservation hypothesis, suggesting that economic growth is a significant contributor to the increase in the use of renewable energy consumption in Bangladesh.

On the contrary, domestic natural gas consumption also has a direct impact on economic growth in Bangladesh. In the short term, an increase in natural gas consumption positively affects renewable energy consumption, while in the long term, they mutually influence and reinforce each other. This study suggests that adopting energy conservation policies in Bangladesh can lead to long-term reforms in energy consumption and foster economic growth.

The World Bank and United Nations emphasize the importance of energy conservation in sustainable development and Bangladesh's government has adopted an action plan to strengthen energy efficiency strategies. An energy conservation policy can improve mismanagement in the energy sector, save more energy, and reduce the excessive burden of domestic natural gas in Bangladesh. Further investment in sustainable energy sources in this country is necessary to promote the growth of renewable energy consumption. The private sector has contributed significantly to Bangladesh's economic development, but government investment still dominates the energy sector. Energy policy should consider energy security, cost efficiency, environmental friendliness, safety, and utilizing the country's available resources and technology.

Finally, this study acknowledges that data availability and collection circumstances may introduce bias, and including more variables can lead to more diverse conclusions. Although the primary emphasis of this research was examining the use of renewable energy and natural gas, it is crucial to consider other energy sources, such as petroleum and liquefied petroleum gas (LPG), when investigating the relationship between energy consumption and economic growth. Multidimensional research considering various factors would be a valuable reference for establishing policies.