^{1}

^{*}

^{1}

We study the relationship between energy consumption and economic growth in the case of USA by using an asymmetric ARDL bounds test approach to achieve the actual model. The quarterly data set covers the period of 1973:1- 2013:4. The findings indicate that the effect of energy consumption is asymmetric in the long term but not in the short term. In the long run, the effect of negative component of energy consumption on economic growth is small and statistically insignificant. The coefficient of the positive component of energy consumption is found about 0.9 and statistically significant at 1% level. We conclude that energy saving policies such as technological progress and organizational rearrangements may have the dimmer effect for the impact of a negative component of energy consumption and the booster effect for impact of the positive component of energy consumption. Thus, energy saving policy should be tightly followed by the goal of high economic growth.

According to the neoclassical theory, energy inputs are accepted as auxiliary inputs in production. It is possible to assume that not only, there is a certain effect of energy inputs on a production level, but also the unidirectional causal nexus runs from economic growth to energy consumption [

On the contrary to neoclassical model, the ecological economics intellectual tradition highlights that scarce resource may create a restraining impact on the growth especially in industrialized economies [

The schools of ecological economists claim that there is a certain impact on the economic development created by energy, and energy significantly and directly affected industrial revolution as well. Besides, this school of thought supposes that although substation of capital and energy by technological improvements may support overcoming the scarce resources, the role of this substation can only produce a minor positive impact [

Despite the certain distinction on the impact of energy consumption in economic growth between the approaches of neoclassical and ecological schools, it is possible to state that neoclassical economics and ecological economics admit that there is a certain long-standing relation between economic development and use of energy. Against this consensus, the disagreement between these two approaches is to the short and long term direction of these mutual influences. There are major studies, which question the presence and direction of the causative nexus between energy consumption and economic growth through Granger causality techniques [

The causality methodology analyses two main hypotheses as for the energy-led growth hypothesis and the growth-led energy hypothesis. In this context, Granger-causality produces four different alternatives on the possible consequences of causality. The first of is the growth hypothesis assuming that there is one-way causal nexus running from energy consumption to economic growth. The second is the conservation hypothesis suggesting that there is a one-way causal relation from economic growth to energy consumption. According to the feedback hypothesis, bi-directional causality from energy consumption to economic growth, when the last one, the neutrality hypothesis, assumes that there is no causal nexus between energy consumption and economic growth [

By the growth hypothesis, conducting energy saving policies with the purpose of protecting the environment may create an adverse impact on the present state of economic growth if energy consumption Granger causes economic growth. When the growth, conservation and feedback hypotheses highlight some linear causal relations between energy consumption and economic growth, they neglect the effect of technological progress on the energy-growth nexus. In the highly industrialized economies, technological advances mainly originate, this relationship can change. That is to say technological advancement improves energy efficiency, leading to lower energy prices and more energy consumption using “rebound” or “take-back” effect, as first postulated by [

According to [

As a result, decreased energy consumption due to technological advancement may not necessarily lead to economic contraction, while increased energy consumption concurrent with energy efficient technological progress may cause an extraordinary level of economic growth. These are the mechanisms which this study refers to as dimmer and booster effects of energy efficiency improving technological advances. As a result of these mechanisms, there may be an asymmetric causal relation between energy consumption and economic growth.

The purpose of this study is to contribute to the empirical literature and generate policy suggestion concerning the energy consumption and economic growth nexus. The contributions of our study are threefold. First, following the modeling strategy developed [

The rest of the paper is organized as follows: The next section summarizes the empirical literature, testing the energy consumption-economic growth nexus. The used data and modeling strategy, followed in this study, is described in Section 3. Empirical findings are presented in Section 4. The last section consists of conclusion and policy implications of the paper.

Since the form of energy consumption-economic growth relationship may affect a wide range of policies such as energy, trade, innovation incentive, environmental, resource allocation, urbanization, employment and finance policies, extant research tries to find a proper modeling strategy to obtain robust findings of energy consumption and economic growth relationship. However, the findings in empirical studies are highly conflicting. Most of these studies follow various types of linear models. Few studies adapt non-linear modeling strategies to the relationship between energy consumption and economic growth. Since energy consumption and economic growth series are order one,

As discussed by [

As such, certain studies prefer to use not only linear but also non-linear Granger causality models such as [

However, other studies concentrate on non-linear vector error correction models such as [

[

The criticism of linearity of the classical cointegration model suggested by [

Some authors try to deal with both non-stationarity and non-linearity in the co-integrating modeling strategy, since the mid-1990s. This extended section of co-integration research may be described as the primacy of three regime switching models. The first one is the threshold error correction model developed by [

Alternatively, [

Lastly, [

The non-linear autoregressive distributed lag model suggested by [

where

If Equation (1) is estimated with OLS estimator to calculate the cointegration parameter, some important problems such as weak endogeneity and serially correlated errors can arise. So, estimated cointegrating parameter would be poorly determined especially in ﬁnite samples. The best solution to these problems may be the ARDL approach advanced by [

Following [

where

[

Equation (5) can be estimated by standard OLS since the model is linear. Moreover, the null hypothesis of a symmetric long-run relationship

The existence of an asymmetric cointegration relationship in non-linear error correction model can be tested using F_{PSS} test suggested by [_{PSS} test is_{PSS} statistics under each of these cases for a range of values of k, the number of regressors entering the long-run relationship.

The model specification in Equation (5) is necessary for establishing the properties of the dynamic adjustment mechanism. There is no restriction on both short and long run asymmetries in Equation (5). Three alternative model specifications are possible in Equation (5). Firstly, short-run dynamic asymmetries can be analyzed in the response of economic growth to ﬂuctuations in the energy consumption by implicitly imposing the long-run symmetry restrictions

Second, an asymmetric long-run relation can be investigated by imposing short-run symmetry restriction

Lastly, when one assumes both symmetric short-run and long-run adjustment, the most restrictive speciﬁcation is obtained as Equation (8):

Following the boundstesting approach, the models can be estimated irrespective of whether _{t} are

The NARDL approach starts to distinguish negative and positive components of the energy consumption variable.

Linear ARDL and asymmetric ARDL models were estimated to compare the two models. Firstly, we choose a lag specification for ARDL model, following the general-to-speciﬁc approach. The preferred speciﬁcation is determined by starting with max p = 12 and max q = 8. Later, all insigniﬁcant stationary regressors are dropped from the ARDL model. In

According to _{PSS}) cannot reject the null hypothesis of no co-integrating relationship between energy consumption and economic growth in the symmetric ARDL model. Furthermore, the normality test (

nexus between energy consumption and economic growth. So, asymmetric ARDL modeling may be a more correct modeling strategy than linear ARDL

Variable | Coefficient | Standard Error | t Statistic | Probability |
---|---|---|---|---|

GDP_{t}_{−1 } | 0.003801 | 0.001994 | 1.906557 | 0.0585 |

_{ } | −0.018376 | 0.010321 | −1.780347 | 0.0771 |

_{ } | 0.371896 | 0.072472 | 5.131561 | 0.0000 |

^{ } | −0.200989 | 0.072900 | −2.757067 | 0.0066 |

^{ } | 0.207531 | 0.072842 | 2.849071 | 0.0050 |

_{ } | 0.089599 | 0.029938 | 2.992783 | 0.0032 |

^{ } | 0.060288 | 0.029285 | 2.058644 | 0.0413 |

_{ } |

Note: E represents the natural logarithm of energy consumption.

approach. Since we cannot know the true model specifications, the estimated ARDL model includes both short term and long term asymmetries.

In _{LR} is the Wald test which has the null hypothesis of long-run symmetry (_{SR} test is short-run symmetry (_{LR} test. However, WALD_{SR} test cannot reject the null hypothesis of short term symmetry even at 10% significance level. So, one can conclude that there is asymmetric effect of energy consumption on economic growth. Furthermore, when the ARDL model is specified to asymmetric relation, the cointegration test (F_{PSS}) can reject the null hypothesis of no co-integrating relationship between energy consumption and economic growth. The long term coefficient, indicating the effect of positive change in energy consumption on economic growth (

In the next stage of the analyses, impulse response function for the relationship between energy consumption and economic growth is investigated. An impulse response function measures the time profile of the effect of a shock on the behavior of a series. Like that, one can approve an impulse response function as the outcome of a conceptual experiment. The time profile of the effect of a positive unit shock hitting a series at time t can be conceptualized as a simulation. The idea closely resembles Keynesian multiplier analysis. The distinction between the two is that the impulse response function analyzesare carried out on

Variable | Coefficient | Standard Error | t Statistic | Probability |
---|---|---|---|---|

GDP_{t}_{−1 } | −0.08045 | 0.0265 | −3.04 | 0.0280 |

_{ } | 0.07113 | 0.0287 | 2.47 | 0.015 |

_{ } | 0.005 | 0.0156 | 0.32 | 0.750 |

ΔGDP_{t}_{−1} | 0.37334 | 0.0747 | 5.00 | 0.0000 |

ΔGDP_{t}_{−8}^{ } | −0.13440 | 0.0728 | −1.84 | 0.067 |

ΔGDP_{t}_{−9}^{ } | 0.2427 | 0.0751 | 3.23 | 0.002 |

0.0697 | 0.0500 | 1.39 | 0.166 | |

_{ } | −0.0998 | 0.0482 | −2.07 | 0.040 |

_{ } | 0.1048 | 0.0439 | 2.39 | 0.018 |

0.0121 | 0.0589 | 0.29 | 0.770 | |

0.1199 | 0.0553 | 2.19 | 0.030 | |

0.1705 | 0.0588 | 2.90 | 0.004 | |

F_{PSS}: 4.1582 [3.17 - 4.14] | WALD_{SR}: 2.594 [0.110] | |||

WALD_{LR}: 231.8 [0.000] |

Note: E represents the natural logarithm of energy consumption._{PSS} denotes the PSS F-statistic testing the null hypothesis_{PSS} test, attained from [

shocks or innovations of macroeconomic time series, rather than the series themselves, such as investment or government expenditure [

Although we find that the optimal model requires long termasymmetry and short term symmetry specifications, for the aim of comparison, impulse response functions are graphed by estimating an unconstrained and three constrained models. The unconstrained model in Equation (5) allows asymmetries both in the short run and long run (

The

12 quarters. The cumulative negative effect of a decrease in energy consumption on economic growth disappears after about 20 quarters.

A large number of studies in the literature seek to find a proper modeling strategy that captures true dynamic relationship between energy consumption and economic growth. Specification errors in the models lead to findings and policy suggestions incompatible with real policy implementations. For example, the papers, finding evidence in favor of the growth hypothesis, suggest stopping energy saving policies, since a decrease in energy consumption does harm economic growth. However, this policy suggestion conflicts the real life fact of the societies continuous strive towards new ways to reduce energy consumption, be it at the individual, firm or country level. These efforts at improved energy efficiency can be in the form of a new technological achievement such as an electrical automobile, high efficiency motors, replacement of energy losing machines, or organizational rearrangements such as regulating the working hours, developing an efficient energy saving strategy and mandatory regulation for improving energy efficiency. So, one can be doubtful about modeling strategy leading to the suggestion of stopping energy saving policy.

Departing from previous studies in the literature, we assume that the relationship between energy consumption and economic growth is both dynamic and asymmetric. To apply this modeling strategy, we follow the asymmetric ARDL approach suggested by [

The overall assessment of the analysis in this study is that energy saving policy option is independent of the relationship between energy consumption and economic growth. More precisely, energy saving policy may have a positive effect on economic growth rather than negative effect. That is, energy saving policy should be followed in every scenario to reduce the cost of energy and environmental pollution.

Bayramoglu, A.T. and Yildirim, E. (2017) The Relationship between Energy Consumption and Economic Growth in the USA: A Non-Linear ARDL Bounds Test Approach. Energy and Power Engineering, 9, 170-186. https://doi.org/10.4236/epe.2017.93013