Causality relationship between growth and energy use: a case study of Vietnam

Table 1 An overview and evaluation of existing literature Paul & Bhattacharya 2004 India 1950–1996 Engle–Granger cointegration approach combined with the standard Granger causality t

Causality Relationship between Growth and Energy Use:

A Case Study of Vietnam

NGUYEN THI TAM HIEN

The University of Danang, Campus in Kon Tum - nttamhien@kontum.udn.vn

NGUYEN THI PHUONG THAO

The University of Danang, Campus in Kon Tum - ntpthao@kontum.udn.vn

VU THI THUONG

The University of Danang, Campus in Kon Tum - vtthuong@kontum.udn.vn

Abstract

Recently, Vietnam is becoming an energy-dependent country In spite of the important contribution of

energy to Vietnam economic growth via import and industrial production, the increase in energy consumption

also raises the concerns of resource scarcity, the overwhelming dependence on energy, and sustainable growth

This study investigates the causal relationship between economy growth and energy consumption in the case

of Vietnam from 1986 to 2013 Through testing different types of Granger Causality based on the Vector Error

Correction Model (VECM), the main finding is the unidirectional Granger causal connection from energy

consumption to economic growth, which is different from previous research in Vietnam In addition, the paper

indicates the negative effect of energy consumption on economic growth

Keywords: energy consumption; Granger causality; Vector Error Correction Model (VECM)

1 Introduction

Over past two decades, the dramatical increase in the demand for energy to meet the rapid economic development in the Asian countries along with inefficient energy use has caused energy scarcity Besides, the high volatility of energy prices along with the rising greenhouse gas emissions

in recent years also has become a challenge to the sustainable development of these countries The building of energy conservation policy intended to ensure energy security as well as promote sustainable growth has attracted many scholars’ concerns It is important in design this policy is that policy makers must understand clearly the causal relationship between energy consumption and economic growth In other words, policymakers have to answer the question whether economic growth boosts energy consumption or whether energy consumption causes economic growth However, up to now, there has a lack of consensus among economists due to the mixed findings from previous studies

In term of energy economics literature, there has a massive body of academic research on the linkage between energy consumption and economic growth These studies have conducted in various countries with different periods and using different econometric methodologies (Ozturk, 2010) However, the directions of this causality are still mixed and controversial among empirical pieces of research Azlina (2012) classifies the causal relationship between energy consumption and economic development into four views The first view suggests that economic development is considered as the main driver for energy demand economy grows It means that economy growth to lead to the energy demand of the economy also increases On the contrary to the first view, the second view points out the important roles of energy in the economic development process In addition to capital, labor, and materials, energy is considered as an input to production The third view shows that there has a two-way causal relationship between energy consumption and economic development The fourth view argues that both energy consumption and economic development are neutral with respect to each other

Over more past two decades, Vietnam has witnessed impressive economic growth in the Southeast Asia However, its consumption of energy also increased tremendously accompanied by high economic growth In particular, before the impact of 2008 World Economic Crisis, the economic growth rates in Vietnam on average reached over 7 percent for the period from 1990 to 2007 At the same time, the energy consumption per capita in Vietnam also increased by 9.3 percent per year1 CIEM (2012) also show that in two last decades, the economic growth in Vietnam has relied heavily

on draining a lot of its natural resources and energy intensity levels has been higher than other countries in the region In particular, to generate $ 1,000 of GDP, Vietnam must consume about 600

kg of oil equivalent, 1.5 times higher than in Thailand and more than 2 times compared to the average level of the world (Hong Quang, 2015) Do and Sharma (2011) also give a forecast that the total

1 Toan, P K., Bao, N M., & Dieu, N H (2011)

energy consumption in Vietnam is projected to rise from 55.6 Mtoe in 2007 to 146 Mtoe in 2025 Vietnam is facing the risk of dependence on imported energy This impressive performance has placed an interesting question to economists and policy makers of whether energy consumption is the cause or effect of economic growth in Vietnam

To the best of our knowledge, there have been some studies examining the energy-growth nexus, such as Chontanawat et al (2008), Phung (2011), Le (2011), and Nguyen (2012) However, this causal relationship between energy consumption and economic growth has not been reached to a consensus among economists2

Awareness of the importance of understanding the causal relationship between energy consumption and economic growth in policy implications leads us to continue this issue Performing Granger causality test and Vector Error Correction Model (VECM) using the energy consumption per capita and income per capita from the year 1986 to 2013, this paper aim to answer the following questions: (1) Does there exist causality between economic growth and energy consumption in short-term and in long-short-term? (2) If yes, what is the sign and magnitude of such effects?

The remainder of this paper is organized as follows: Section 2 briefly reviews the literature on the causal relationships between energy consumption and economic growth Section 3 outlines the data and the econometric methodology Then, the econometric results are discussed in section 4 Conclusions follow in Section 5

2 Literature review

The causal linkage between energy consumption and growth has been a widely studied topic in the literature for a long time Since the pioneering work of Kraft and Kraft (1978), which concluded

a unidirectional causality from income to energy consumption in the United States for the 1947-1974 period, many economists have joined the debate with either supporting or conflicting views during the next two decades For example, Abosedra and Baghestani (1989) confirms the Kraft and Kraft’s result by applying the standard test of Granger causality Cheng and Lai (1997) with Taiwanese data from 1955 to 1993 and Cheng (1999) with two time series of India also support the causality running from GDP to energy consumption without feedback Conversely, Akarca and Long (1980) detect the issue of temporal sample instability in Kraft and Kraft’s study Therefore, by replacing the time period, the relationship turned out to be no statistically significant The same neutrality property is proved by Yu and Hwang (1984), Yu and Jin (1992) with updated data to 1979 and 1990, respectively Using standard Granger technique, Yu and Choi (1985) also have the same conclusion for the case of

US, UK, and Poland However, the causal relationship runs from GNP to total energy consumption for South Korea and vice versa for the Philippines Masih & Masih (1996) also found different results when checking the growth-energy consumption nexus for 6 Asian countries The Johansen's

2 Tang, C F., Tan, B W., & Ozturk, I (2016)

multivariate cointegration tests and dynamic vector error correction model (VECM) show mutual effects between development and energy use in Korea, Taiwan, and Pakistan In sum, most studies

in this period used bivariate models and Granger causality approach Hence, major reasons for these contradicted findings may be the usage of different tests and lag terms for time series, the diversification in data of various countries and various time scales

In the recent years, the debate on the growth-energy nexus is even more extensive and diversified with various research directions and economic techniques Both country-specific and multi-country provide a broad context of the research issue with the aim of drawing a definite conclusion on the relationship and its direction between economic growth and energy consumption This ambitious purpose has still not been achieved due to a consensus on the subject matter so far Table 2.1 provides

a summary of controversial arguments in the last 15 years

From the review of recent literature, some noticeable points are summarised as follows:

- The previous two decades experience a vigorous debate of a large number of researches worldwide Some studies focus on a specific nation while others assess a group with certain common properties such as developing countries, industrialized countries, G-7 group, oil-exporting countries, same region or same continent countries

- Except for the study of Stern and Enflo (2013) which uses 150-year time series for Sweden, the examined period often ranges from 3 to 5 decades, assuring the reliability of findings However, the most recent year is 2011 that is relatively out of date In the context of increasingly urgent environmental issues, research with more updated data should be carried out to draw conclusions that are more suitable

- In terms of methodology and techniques, most of researches implement 3 main steps:

First, test the stationarity of the series or their order of integration in all variables, using Augmented Dickey-Fuller (ADF) test and Phillips-Perron (PP) test Some authors apply more tests such as Kwiatkowski-Phillips-Schmidt-Shin (KPSS) test (Ang, 2008), Zivot and Andrews test (Altinaya & Karagol, 2004) to verify their findings

Second, examine the presence of a long-run relationship, utilizing the popular approaches of Johansen (1991) or Pedroni (1997, 2004) A few studies are distinguished by some newly developed cointegration tests, for example, test of Pesaran et al (2001) and a modified version of the Granger causality test due to Toda and Yamamoto (1995) in the studies of Wolde-Rafael (2006), Fatai et al (2004)

Third, identify the direction by employing Granger causality tests (VAR model) or vector error correction model (VECM) Some other techniques are also applied depending on data’s characteristics like Hsiao’s version of Granger causality tests (Aqeel & Butt 2001)

- There are two types of models: (1) bivariate model in which two main variables are energy and growth (total or per capita) and electricity consumption seems to be the most popular measure energy use Other energy sources such as oil, gas are also mentioned in several studies (2) A multivariate model which usually adds energy price, labor, capital as exogenous factors

- Although the results are mixed and contradicting, there are three main strands: (1) unidirectional causality which can be energy consumption – growth or growth – energy consumption directions; (2) bi-directional causality; (3) no relationship Since most of researches lead to policy suggestions accordingly to their findings, evidence in either direction is useful and important To be specific, if there is unidirectional causality running from energy consumption to economic growth, conducting and promoting energy preservation programs could harm the economy Similarly, with bi-directional causality, economic growth demands more energy at the same time energy levels also affect economic growth Energy-saving strategy, therefore, may slow down the momentum of development On the other hand, if the growth - energy consumption direction is found, policies aiming at reducing energy use may be implemented with little or no negative influence on economic growth The same suggestions will be made in the case of no causality

Several studies on the growth-energy consumption relationship in Vietnam have only been carried out recently, including Phung (2011), Le (2011) and Nguyen (2012) Three authors used the bivariate model to study the relationship between energy consumption/ electricity consumption and economic growth in periods 1976 – 2010, 1975 – 2010, and 1986 – 2006 respectively Despite the different time scales, they came up with a unidirectional relation from growth to energy consumption These results could not explain the energy-led economy in Vietnam This makes difficult to understand the energy as a determinant of economic growth and giving implications to policymakers in energy policies or sustainable growth policies

Table 1

An overview and evaluation of existing literature

Paul &

Bhattacharya

(2004)

India 1950–1996 Engle–Granger cointegration

approach combined with the standard Granger causality test

Multivariate model Bi-directional causality between

energy consumption and economic growth

Ghosh (2006) 1970-2002 Test stationary: ADF test

Test co-integration: Johanson test VECM

Bivariate model Unidirectional causality from GDP

to petroleum consumption

+ In-sample forecasts till 2011–

2012

Soni, Singh &

Banwet (2013)

1981-2011 Test stationary

Test co-integration Granger causality & vector error correction mechanisms (VECM)

Bivariate model Unidirectional causality: economic

growth to electricity consumption)

+ Implications on policy

No forecast

Ang (2008) Malaysia 1971–1999 Unit root tests: ADF, PP, KPSS tests

Johansen approach for the VARs constructed in levels

Causality tests

Multivariate model:

add CO2 emissions

Unidirectional causality: economic growth to energy consumption

+ KPSS test has higher power

of rejecting the null

Loganathan,

Nanthakumar

&

Subramaniam

(2010)

1971-2008 OLS Engel-Granger (OLS-EG)

Dynamic OLS (DOLS) Autoregressive Distributed Lag (ARDL)

Bounds testing approach & VECM

Bivariate model Bi-directional causality between

total energy consumption & GDP

+ Combination of different methods OLS-EG, DOLS, ARDL & VECM

Azlina (2012) 1960-2009 Test stationary: ADF & PP tests

Test co-integration VECM

Multivariate model:

add share of industry

in GDP capital

Unidirectional causality: income to energy consumption

+ Implications on policy + Consider both demand and supply side of energy Altinaya &

Karagol

(2004)

Turkey 1950-2000 Test stationary: Zivot & Andrews

test Granger non-causality test

Bivariate model Unidirectional causality: electricity

consumption to income

The standard Granger causality test Lise & Van

Montfor

(2005)

1970-2003 Test stationary: ADF test

Test co-integration: Johanson test VECM

Bivariate model (per capita energy and GDP)

Unidirectional causality from GDP per capita to energy consumption per capita

+ Tests on misspecification: Durbin-Watson, Godfrey, Ramsey’s reset, Lagrange multiplier, Breusch-Pagan Chow & Recursive Chow Jobert &

Karanfil

(2007)

1960-2003 Test stationary: ADF, PP tests

Test co-integration: Johansen approach

Multivariate model:

add industrial value

No relationship between real GDP and total energy

Stern & Enflo

(2013)

Sweden 1850-2000 Tests unit roots

Granger causality tests using VAR model

Test co-integration & VECM

Multivariate models:

add capital, labor

The relationship has changed over time, but mostly energy

causes growth

+ Long period (150 years) + Diversified techniques

Glasure

(2002)

add government expenditure, money supply, oil prices

Bi-directional causality between oil consumption and economic growth

+ Deal with the omitted variables ( control effects of the oil price, money supply, government spending and the oil price shocks

Oh & Lee

(2004)

1970-1999 Unit root test: PP test

Cointegration test: VAR and the corresponding VECM Causality test

Multivariate model:

labour and capital added

Bidirectional causation between energy and GDP

+ The Newey and West (1987) method applied to choose optimal lag lengths

Mozumder &

Marathe

(2007)

Bangladesh 1971-1999 Test stationary: ADF test

Test co-integration: Johanson test VECM

Bivariate model Unidirectional causality: per capita

GDP to per capita electricity use

+ Policy implications

Aqeel & Butt

(2001)

Pakistan 1955-1996 The co-integration technique and

Hsiao’s version of Granger causality tests

Bivariate model Unidirectional causality: economic

growth to energy consumption

+ Hsiao’s Granger Causality test: more robust over arbitrary lag length selection and other systematic methods determining lag length

Yang (2000) Taiwan 1954–1997 Unit root tests

Cointegration panel tests VECM

Bivariate model Bi-directional causality between

total energy consumption and GDP

Lolos &

Papapetrou

(2002)

Greece 1960-1996 Unit root tests

Cointegration panel tests VECM

Multivariate model:

add CPI

Bi-directional causality between energy consumption (total and industry) and GDP

Shiu & Lam

(2004)

China 1971-2000 Unit root tests

Cointegration panel tests VECM

Bivariate model Unidirectional causality: electricity

consumption to real GDP

Fatai, Oxley &

Scrimgeour

(2004)

NZ 1960-1999 Toda and Yamamoto (1995)

approach to check robustness of results

Bivariate model Unidirectional causality: GDP to

energy consumption

(+) comparison with Au, Indo, India, Philippine and Thai

Jumbe

(2004)

Malawi 1970-1999 Standard GC methodology

VECM

Bivariate model Bi-directional causality between

electricity consumption (kWh) and GDP

Unidirectional causality:

nonagricultural-GDP to kWh Morimo &

Hope

(2001)

Sri Lanka 1960-1998 Econometric model developed by

Yang (2000)

Bivariate model Unidirectional causality: electricity

supply to real GDP

Soytas & Sari

(2003)

G-7 countries and 16 emerging markets

1950-1994 Unit root test: DF , ADF ; PP test

Cointegration panel tests VECM

Bivariate model: Bi-directional causality in

Argentina Unidirectional causality from GDP

to energy consumption in Italy, Korea, vice versa in Turkey, France, Germany, Japan

+ Suggest a number of ways to reduce CO2

Narayan &

Smyth (2008)

G7-countries

1972–2002 Panel unit root tests

Panel cointegration tests

Multivariate model:

add capital stock, real

Unidirectional causality: energy to growth (a 1% increase in energy

+ Panel cointegration test proposed by Westerlund (2006) allowing for multiple

Granger causality and long-run structural estimation

gross fixed capital formation per capita

use increases real GDP by 0.12–

0.39%)

structural breaks

Yoo (2006) Asian

countries

1971–2002 Stationary and co-integration tests

Granger-causality method VECM

Bivariate model Bi-directional causality between

electricity consumption per capita and GDP per capita in Malaysia and Singapore

Unidirectional causality: energy to growth in Thailand & Indonesia

+ Use suitable information criteria to select the optimum lag

Mehrara

(2007)

11 exporting oil countries

1971–2002 Panel unit-root tests

Panel cointegration analysis

Bivariate model Unidirectional causality: economic

growth to energy consumption Lee (2005) 18

developing countries

1975-2001 Panel stationary tests with

heterogeneous country effect

Modified ordinary least square techniques (FMOLS)

Multivariate model:

add capital

Unidirectional causality: energy consumption to GDP

Joyeux &

Ripple

(2007)

7 East Indian Ocean countries

1971-2001 Panel unit root tests

Co-integration panel tests of Pedroni (1997, 2004)

Biviarate model No co-integration: income and

electricity consumption (household level)

+ Different research direction Other composite indexes (HDI)

Wolde - Rafael

(2006)

17 African countries

1971–2001 Newly developed cointegration test

proposed by Pesaran et al (2001) Modified version of the Granger causality test of Toda and Yamamoto (1995)

Bivariate model Different relationships and

direction between countries

+ Suitable methods for studies that have small sample sizes

Asafu-Adjaye

(2000)

India, Indonesia, Philippines Thailand

1973–1995 Unit root tests

Cointegration panel tests VECM

Multivariate model:

add CPI

Unidirectional causality: energy to income for India and Indonesia Bi-directional causality: energy to income for Thailand and the Philippines

Source: Authors’ analysis and synthesis

3 Data and econometric methodology

3.1 Data

To carry out the analysis of Granger causal relationship between energy consumption and economic growth, this study uses the energy consumption in kg of oil equivalent per capita (denoted E) and GDP per capita in current $US (denoted Y) These secondary data is collected from World Development Indicators (WDI, 2016) Both data have the time horizons from 1986 to 2013 The logarithm form is applied to both variables to reduce heteroscedasticity

3.2 Econometric methodology

To analyze the causal relationship between the energy consumption and economy growth, we apply the process as follows

3.2.1 Stationary testing

This study investigates the stationary of logY and logE using augmented Dickey and Fuller approach – ADF (Dickey and Fuller, 1979) and Phillips – Perron test – PP (Phillips and Perron, 1988) The purpose of the stationary test is to avoid the spurious regression and then determine the order

of integration of each variable The fitted regression model both logE and logY is expressed as:

∆𝑋𝑡 = 𝑎0+ 𝑎1𝑡 + 𝑎3𝑋𝑡−1+ ∑ 𝑏𝑖∆𝑋𝑡−𝑖

𝑚

𝑖=1

+ 𝜀𝑡

where: ∆ is the first difference of the log of variables

t is the time or trend variables (if available)

𝜀𝑡 is the pure white noise error term

𝑚 is the maximum length of the lagged depentent variable

Under the null hypothesis of stationary testing, the time series contain a unit root or - 𝐻0: 𝑎3 = 0.The alternative hypothesis is 𝐻0: 𝑎3 < 0 For next cointegration test, Granger causality test, and, VECM, logY and logE are expected to be non-stationary at level and stationary at the first difference 3.2.2 Testing for cointegration

The regression among nonstationary variables can lead to the spurious result that is not meaningful to decision making However, in the context that the time series in the study are under cointegration, the spurious regression does not occur In this interesting event, the cointegrated variables show the short- term deviation from their association that must converge to equilibrium in long-term

Granger (1986) and Engle and Granger (1987) introduced the Engle-Granger test for cointegration between two variables, basing on the residual of a linear combination of variables