Renewable energy consumption and agriculture: evidence for cointegration and Granger causality for Tunisian economy

ABSTRACT

This paper uses the vector error correction model (VECM) and Granger causality tests to investigate short and long-run relationships between per capita carbon dioxide (CO₂) emissions, real gross domestic product (GDP), renewable and non-renewable energy consumption, trade openness ratio and agricultural value added (AVA) in Tunisia spanning the period 1980–2011. The Johansen-Juselius test shows that all our considered variables are cointegrated. Short-run Granger causality tests reveal the existence of bidirectional causalities between AVA and CO₂ emissions, and between AVA and trade. There are short-run unidirectional causalities running from non-renewable energy and GDP to AVA and to renewable energy, and running from CO₂ emissions to renewable energy. Interestingly, there are long-run bidirectional causalities between all considered variables. Our long-run parameters estimates show that non-renewable energy, trade and AVA increase CO₂ emissions, whereas renewable energy reduces CO₂ emissions. In addition, the inverted U-shaped environmental Kuznets curve (EKC) hypothesis is not supported. Our policy recommendations are to increase international economic exchanges because this gives new opportunities to the agricultural sector to develop and to benefit from renewable energy technology transfer. Subsidizing renewable energy use in the agricultural sector enables it to become more competitive on the international markets while polluting less and contributing to combat global warming.

KEYWORDS:

• Renewable energy
• agriculture
• trade
• Granger causality
• Tunisia

Disclosure statement

No potential conflict of interest was reported by the authors.

Notes

1. Economic growth and agricultural value added are incorporated into the same specification model because of their low correlation value (0.3).

2. LM-stat denotes the computed statistic value of the Lagrange Multiplier test and p indicates the probability of the null hypothesis. Our results show that the null hypothesis of no serial correlation in the residuals cannot be rejected for lags one, two and three.

3. The null hypothesis assuming that residuals are multivariate normal is verified using Skewness, Kurtosis and Jarque-Bera methods.