The impact of agriculture, energy consumption and economic growth on ecological footprint: testing the agriculture-induced EKC for Türkiye

Abstract

The main purpose of the current study is to empirically analyze the impact of agriculture, energy consumption, and economic growth on environmental degradation in Türkiye for the period of 1968–2020. Against the backdrop, we tested agriculture-induced EKC hypothesis employing Ecological Footprint (EFP) instead of emissions because EFP better represents the sustainability. Our results from ARDL model indicate the validation of the EKC hypothesis for Türkiye. Hence environmental deterioration worsens with economic growth and improves with higher level of GDP. Moreover, increase in both the agricultural value added per capita and energy use per capita increases the EFP per capita. In other words, agricultural production, and energy usage are important drivers for environmental quality. The causality results show the presence of growth hypothesis between agriculture and GDP and energy-EFP for Türkiye. Therefore, government should envisage the policies towards sustainable economic growth as discussed under the conclusion part.

1 Introduction

In recent years, concerns about increased energy price volatility (energy security) and environmental pollution and/or degradation have become major challenges all around the world. Following the agricultural and industrial revolutions, countries have accelerated economic activities from different sectors (i.e. manufacturing, energy, transport, agriculture, etc.) to achieve economic growth (hereafter EGR) and development. These activities, however, require much more energy and put pressure on environment (Udemba, 2020). There are international efforts to combat increasing GHG emissions like Kyoto Protocol in 1987 and Paris Agreement in 2015. However, it’s predicted that world economy will double by 2050 and stronger GDP and population growth will boost the energy demand around 50% (EIA, 2021). Moreover, apart from the Covid-19 induced decrease in 2020, global emissions have increased in every year since 2015, the year of Paris Conference of Parties (BP, 2022). Energy efficiency and renewable energy (hereafter RE) sources are considered to be an alternative to mitigate environmental degradation and ensure sustainable economic development. In this context, many countries have implemented incentive policies (such as feed-in tariffs, subsidies, loans, tenders, taxes etc.) to accelerate the deployment of RE in their energy use. Like energy, demand for agricultural products have been increasing to meet necessities of the growing population globally. World population will excess 9 billion (34 percent higher than present) and 70 percent of them will live in urban (around 49 percent higher than today) by 2050 (IMF, 2020). According to projections of Food and Agriculture Organization (FAO, 2023) food production must increase around 70 percent to meet necessities of word population by 2050. However, agriculture is also assumed as one of the major contributors to environmental pollution and/or degradation in the world (Balogh, 2019). The increase in agricultural production leads greater use of energy (especially fossil fuels), fertilizer, pesticides and resulting CO₂ emissions and/or pollutants create environmental problems. Moreover, incremental agricultural farming requires expanding agricultural land, which creates deforestration (Khan et al., 2023a; Muoneke et al., 2022). Indeed, agricultural activities (use of fertilizers and pesticides, land reclaiming, bush burning etc.) could generate significant environmental problems such as water and soil pollution and/or degradation, overuse of water sources, habitat loss, GHG emissions, acid precipitation, ozone depletion, etc. (Cetin et al., 2022; Udemba, 2020). For example, FAO (2020) reports that nearly one fourth of GHG emissions (17%) is emitted from agriculture and agriculture sector is expected to be second highest emitter of GHGs in near future. Furthermore, agriculture activities (such as crop cultivation, livestock production, irrigation, transportation of harvest, etc.) use more than 70% of freshwater and it consumes about 5% of total energy in the world (Muoneke et al., 2022; Salari et al., 2021). The growing world demand for food expands the agricultural frontier, harms the soil’s ability to regenerate and negatively affects the future generation’s sustainability. As known, United Nations (UN) establishes 17 objectives that quide country’s policies towards sustainable development. The achievement of SDG1, SDG 2, SDG 5, SDG 6, SDG 7, SDG 11, SDG 12, SDG 14 and SDG 15 are totally or partially related to interaction between agricultural production and environmental quality (Ali et al., 2023; Alvarado et al., 2021). Among the others, espeacilly “zero hunger-SDG 2” target is directly linked to agricultural production (Behera et al., 2023). As discussed in some studies (see Schoeffel, 2019; Di Vaio et al., 2023), technological advances such as blockchain are a secure method for efficient agricultural production and food security through establishing digital identity and ensuring efficient communication and property rights in agricultural practices. As highlighted by Di Vaio et al. (2023), blockchain technology has the potential not only to promote the sustainable agriculture but also to expand females’ role in the production system and support empowerment of women, improve their health and education. Therefore, the agricultural sector, which is using blockchain, information and communication technologies more intensively every day, becomes a key sector in establishing the relationship between technology and the social dimension of sustainability as targeted by SDG 5. Hence the interaction between agriculture sector, economy, energy use, and environment have attracted the researchers and policy makers attention to achieve the sustainable economic development (Cetin et al., 2022).

Environmental Kuznets Curve (hereafter EKC) hypothesis constitutes the theoretical basis of economy-environment nexus since the pioneering studies of Grossman and Krueger (1991a, 1991b) and presents potential solution to environmental degradation. If EKC is valid, then environmental problems will disappear at higher living standards. Accordingly, EKC hypothesis argues that income increase has a detrimental quality at the initial stages of economic development since production structure mainly depends on pollutant technologies. In other words, EKC hypothesis suggests that environmental degradation first worsens with the income increase then improves with the economic progress in an inverted U-shaped link. This EKC link between the EGR and environmental quality emerges due to scale, technique, and composition effect¹(Cağlar et al., 2021). Many studies tested the validation of EKC and investigated the link between EGR and environmental pollution with some control variables such as energy consumption (hereafter ENCON), renewable and fossil energy use, foreign direct investment (hereafter FDI), urbanization, globalization, trade, financial development, income distribution, energy prices, information, and communication technologies (hereafter ICT), economic complexity etc. (See Caglar et al., 2021; Saud et al., 2020; Shahbaz & Sinha, 2019). In these studies, however, environmental degradation mostly has been represented by CO₂ emissions. Therefore, since GHG emissions constitute the only the air part of environmental deterioration, this approach is heavily criticized in energy and/or environmental economics. In recent years, EKC studies have employed the Ecological Footprint (hereafter EFP) which is more comprehensive indicator of environmental degradation than emissions (Danish et al., 2019; Sharif et al., 2020; Sabir & Gorus, 2019; Destek & Sinha, 2020, among the others). EFP, which is proposed by Wackernagel (1994) and Wackernagel and Rees (1996), is a cumulative environmental indicator that combines the impacts of human activities on air, water, and soil. These impacts are being calculated with six components: carbon, cropland, grazing land, built-up land, fishing grounds, and forest footprints (Pata, 2021). Hence, EFP is accepted to be more powerfull indicator of sustainability. However, only small number of empirical studies have examined the role of agriculture sector on EFP in EKC literature.

The agriculture sector is of key importance to Türkiye both in economic and social aspects. For example, agriculture sector constitutes the 6% of Türkiye’s Gross Domestic Product (GDP), while employing 18% of total labor force (TurkStat, 2022). Using Input–Output (I–O) analysis, Karkacier and Bölük (2017) demonsrated that agriculture and energy sector have strong inter-industrial linkages and agriculture is the most income generating sector in Turkish economy. Energy consumption in the agriculture sector in Türkiye has increased significantly over the 40 years in parallel with the increases in agricultural production and export. For example, while oil consumption in Türkiye’s agricultural sector was 1906 thousand toe in 1990 and it reached 3193 thousand toe in 2022. Similarly, electricity consumption in agriculture sector increased from 49 thousand toe to 1184 thousand toe in the same period. The 24-fold increase in electricity consumption, almost twofold increase in oil consumption in 1990–22 period shows that the Turkish agriculture sector is highly dependent on energy. Moreover, the use of persticides and fertilizer increased by 67% and 19%, respectively in last 40 years (TurkStat, 2023). Indeed, while contribution of the agriculture sector to GHG emissions was around 8.8% in 2010, this share nearly doubled in 2020 (%14) (TurkStat, 2021, 2022). Although agriculture sector directly and indirectly may have detrimental impact on environmental quality of country, there are only few studies that examine the role of agriculture sector on CO₂ emissions in EKC framework in Türkiye (See Dogan, 2016; Ongel, 2020; Bas et al., 2021). These studies, however, focus only on the CO₂ emissions and ignore the impacts of agricultural activities on EFP. Furthermore, these studies reveal controversial results about both the validity of the EKC hypothesis and the agriculture-environment relationship.

As summarized under the literature survey in detail, previous studies provide limited information about whether EGR and agriculture sector have effect on environmental quality since they just focus on air pollution. Unlike the previous studies, this is the first study that aims to test the validity of agriculture-induced EKC hypothesis in Türkiye during the 1968–2020 by using ARDL methodology and EFP. The contributions of current study are as follows. First, we employ EFP which is used as a more powerful and comprehensive indicator of environmental quality and sustainability in environmental economics to evaluate the agriculture (hereafter AGR) on environmental quality. Second, we focus on Türkiye since there is no study that empirically analyzes the role of AGR on EFP in Türkiye. Third, we employ multivariate model using energy as an explanatory variable since bivariate model may produce biased results. Since energy has a strategic role in existence of the inverted U-shaped relationship, excluding it in the EKC modelling, leads to inconsistent findings. Therefore, our findings will provide important results to formulate efficient policies in agriculture and energy sectors towards sustainable economic development in Türkiye. Furthermore, the studies of EFP in agri-food production have a particular importance of sustainable agriculture and future food security. Lastly, we discuss causality relationships between EFP per capita (hereafter $EFPPC$), agiricultural value added per capita (hereafter $AGRVPC$), energy consumption per capita ($ENGYPC$) and EGR (represented by GDP per capita) since the causal directions are key for policy formulation towards sustainability.

The structure of the current paper is organized as following. The second section discusses the agriculture-induced EKC in the literature and summarizes the results of previous empirical studies that analyzes the impact of AGR on CO₂ emissions and EFP. The third section presents the data and econometric modelling procedures. Fourth section provides obtained results and discussions. Final section concludes and develops some policy recommendations.

2 Literature review

The two main research strands emerged in the EKC literature that investigate the link between EGR, agriculture, and environmental quality. The first strand focuses on the interaction between GHG emissions (represented by CO₂ emissions), agriculture and EGR. Although GHGs are made up of CO₂, methane, nitrous oxide, sulfurhexa fluorite, perfluora carbons, CO₂ is often used in EKC modelling studies because it accounts for the largest share (76.7%) (Gokmenoglu & Taspinar, 2018). Second strand of the EKC studies discusses the link between agriculture, EGR and EFP because EFP is more comprehensive indicator than the air pollution.

2.1 The nexus between agriculture and GHG emissions

Although many EKC studies have been carried out with many explanatory variables following the pioneering studies of Grossman and Krueger (1991a, 1991b), the relationship between AGR and environment has begun to be investigated in 2000s. Agboola and Bekun (2019) tested the existence of agriculture-induced EKC and examined the role of AGR, foreign direct investment (FDI), trade openness and energy use on CO₂ emissions by using data covering 1981–2014 period and cointegration methodology in Nigeria. Their results confirmed the existence of EKC hypothesis for Nigeria. Moreover, authors found that while agriculture and energy use increase the pollution level, FDI inflow mitigates the CO₂ emissions. Similarly, Cetin et al. (2022) tested the agriculture-induced EKC hypothesis in 47 developing countries for 1976–2017 period by using dynamic panel data (DOLS and FMOLS estimators). Their results supported the validation of the EKC hypothesis between income and CO₂ emissions and detrimental impact of agriculture on environmental quality. Doğan (2019) investigated the role of agriculture sector and ENCON on CO₂ emissions and confirmed the inverted U-shaped link and harmful impact of agricultural production on environment for China. Ongel et al. (2020) analyzed the effects of economic sectors such as agriculture, energy and waste and manufacture on emissions in Türkiye in the framework of EKC hypothesis by using ARDL method. Interestingly, authors found no evidence supporting the EKC hypothesis and meaningful relationship between agriculture and manufacture sector on CO₂ emissions. By developing and using Eco-efficiency index for agriculture and data covering 1999–2002 period, Vlontzos et al. (2017) tested the inverted-U shaped relationship for EU countries. Authors found N-shaped relationship between variables and agriculture is far from sustainability. By using annual data covering the 1971–2014 and ARDL model, Gokmenoglu et al. (2019) confirmed the validation of the EKC hypothesis and negative impact of agriculture, EGR, and energy use on GHG emissions in China. Selcuk et al. (2021) examined the role of agricultural activities on CO₂ emissions and tested the EKC hypothesis with some control variables such as FDI, trade openness and ENCON for Next Eleven countries. By using panel cointegration method and data covering 1991–2019 period, authors found mixed results. For example, while agriculture sector mitigates the CO₂ emissions in Bangladesh and Türkiye, it aggravates the air pollution in Mexico. Moreover, EKC is validated only for Bangladesh, Mexico, Nigeria, and Türkiye. Similarly, by using dynamic panel data covering the 1990–2014 period, Parajuli et al. (2019) analyzed the role of agriculture and forest area on CO₂ emissions for 86 countries. Their results showed that 1% increase in forest area decreases emissions around 0.11% while agriculture sector stimulates the CO₂ emissions. Kara et al. (2021) investigated the role agricultural land utilization on different kind of agricultural products for Türkiye in 1988–2019 period. Authors found that while land using farmland for olive plantation and fallow increases the emissions, use of farmland for fruits and beverages and vineyard mitigates the CO₂ emissions in the country. Ehigiamusoe et al. (2023) analyzed the role of three main sectors on environmental quality in Malaysia. Using by ARDL methodology authors found that while banking sector mitigates the CO₂ emissions, agriculture, industry, and other financial sectors deteriorate the emissions. Similarly, employing the panel quantile regression, Khan et al. (2023b), examined the role of agriculture sector including the feed cropping with some control variables such as industrial output, RE, GNP and international trade in nine developing and thirteen developed countries. Authors found that agricultural activities increase the CO₂ emissions around 19.12% and the EKC is validated just for the high quantile countries.

Conversely, some studies found positive impact of agriculture on air quality. For example, Gurbuz et al. (2021), using by ARDL method and data covering 1992–2014, investigated the role of agriculture, ENCON and EGR on CO₂ emissions in Azerbaijan. Their results confirmed the EKC hypothesis and mitigating impact of agriculture on air pollution in this country. Ozcelik et al. (2012) empirically analyzed the role of agricultural activities on CO₂ emissions for Türkiye in 1970–2000 period. Authors revealed that 1% increase in the number of tractors per 100 square kilometers increases emissions around 0.61%, while 1% increase in agricultural production value decreases CO₂ emissions around 1.7% in Türkiye. Similarly, Ridzan et al. (2020) analyzed the impact of agriculture, RE, urbanization, and EGR on CO₂ emissions in Malaysia using by ARDL method and 1978–2016 data and found mitigating effect of agriculture sector on CO₂ emissions. Liu et al. (2017) examined the role of agriculture, RE and EGR on CO₂ emissions in four ASEAN countries over the 1970–2013 period. Their results did not validate the EKC hypothesis but pointed out the positive impact of agriculture on emissions. Doğan (2016) examined the impact of agriculture sector and EGR on CO₂ emissions by using data covering 1968–2010 period and ARDL. Author confirmed the mitigating effect of agriculture on pollution and validation of EKC hypothesis for Türkiye. Bas et al. (2021) examined the link between agriculture value-added, merchandise value-added, export value-added on emissions pollution for Türkiye in 1991–2019 period. Authors’ findings supported the EKC hypothesis and the moderating effect of agriculture sector on CO₂ emissions.

2.2 The nexus between agriculture and ecological footprint

Using by dynamic panel data, Usman and Makhdum (2021) analyzed the impact of agriculture, RE and non-renewable energy, forest area, financial development, and EGR on EFP in BRICS-T countries (Brazil, India, China, South Africa, and Türkiye) in 1990–2018 period. Authors found that 1% increase in agricultural value added increases the EFP around 0.22%. Moreover, their findings indicated the feedback effect between agriculture, financial development and EFP in these countries. Alvarado et al. (2021) analyzed the link between EFP, R&D, agriculture, and international trade for 77 countries from different regions in 1996–2016 period by using FMOLS method. Their empirical findings indicated that agricultural value added, and trade openness are responsible for high levels of environmental degradation represented by EFP in North America, Europe, and Central Asia. They also found a causal relationship between agriculture and trade openness in these countries. Using by linear and non-linear ARDL Udemba (2020) analyzed the role of FDI, agriculture, energy use, population, and income on environment for India in 1975–2016 period. Author found one-way causal relationship from agriculture, FDI, energy use and population to EFP. Again Olanipekun et al. (2019) confirmed the detrimental impact of agricultural activities on EFP in eleven Central and West African countries. Muoneke et al. (2022) analyzed the impact of agriculture and globalization on environmental quality represented by two different indicators (EFP and CO₂ emissions) and some control variables such as fossil energy use, EGR, urban population, and financial development for Philipinnes. In this study it has been found evidence in favor of agriculture-induced EKC, arguing that EFP first increases and then improves as the agricultural sector develops. Similarly, Pata (2021) examined the link between RE generation, globalization, agricultural activities on environmental pollution represented by both EFP and CO₂ emissions for BRIC countries. The empirical findings highlighted that agricultural production is important driver for environmental problems in these countries. Soltani et al. (2021) calculated and compared the resource food prints of different kind of agricultural products by using “the system for integrated assessment of water, land, food and environment-SEA model” and data of 31 Iranian provinces in 2013–2017 period. Author found that among the others production of oil grains has highest footprints of energy. Moreover, while the livestock and poultry production have the highest EFP, wheat production has the maximum source such as water, energy etc. Salari et al. (2021) analyzed the role of RE consumption, globalization, agriculture, and trade openness on EFP in 21 emerging countries by using panel fixed effect quantile regression model. The researchers found that except for globalization all other explanatory variables deteriorate the environmental quality in these 21 emerging countries. Conversely, some studies found positive impact of agriculture on EFP. For example, Aziz et al. (2020) tested the EKC by using quantile ARDL and analyzed the role of agriculture, renewable and non-renewable energy, forestry, and EGR on EFP for Pakistan in 1990–2018 period. Their results validated the EKC hypothesis and indicated the mitigating effect of RE, forestry and agriculture on EFP. Using by ARDL, ul Haq (2021) investigated the role of agricultural export and financial openness on EFP in Pakistan and concluded that agricultural export improves the environmental quality.

As discussed in detail, the empirical literature mainly focuses on CO₂ emissions and reveals mixed results. Moreover, empirical findings related to detrimental impacts of AGR on EFP is lack of theoretical basis since they generally assume linear relationship between them. Therefore, since very few studies have been conducted on impact of AGR on EFP, our study will fill this gap and support the better understanding of the linkage between GDP, ENGY, AGR and EFP in Türkiye.

3 Methodology

3.1 Data, variables and model

Based on literature (See Aziz et al. 2020; Muoneke et al., 2022) it is possible to analyze whether there is a long-run relationship between EFP, EGR, ENCON and agricultural value added in a quadratic functional form. For this purpose, the following equation has been set to test the validation of EKC hypothesis:

$$EFPPC=f\left(GDPPC,{GDPPC}^{2},ENGYPC,AGRVPC\right)$$

(1)

In Eq. (1), the dependent variable $EFPPC$ is Ecological Footprint per capita (measured in gha terms), $GDPPC$ is Gross Domestic Product per capita (in constant 2000 US Dollars) as the proxy for EGR, ${GDPPC}^{2}$ is the square of $GDPPC$, $ENGYPC$ is the energy consumption per capita (in mtoe) and $AGRVPC$ is agricultural value added per capita (in constant 2015 US Dollars). The annual series for variables were obtanined from World Development Indicators for Türkiye over the period of 1968–2020. The variables were transformed into their natural logarithm by applying natural logarithm to Eq. (1) to obtain the growth rate of variables of interest and coeffcients of elasticities. Therefore, the model in Eq. (1) was rewritten and converted into the linear-logarithmic form (Eq. 2) as following;

$$ \ln EFPPC_{t} = \propto_{0} + \beta_{1} \ln GDPPC_{t} + \beta_{2} \ln GDPPC_{t}^{2} + \beta_{3} \ln ENGYPC + \beta_{4} \ln AGRVCPC_{t} + u_{t} $$

(2)

Under the EKC hypothesis, coefficients of ${\beta }_{1}$ and ${\beta }_{2}$ are expected to have a positive and negative sign, respectively. This means that, environmental deterioration worsens at the gerater level of income, but later (after a turning point), this degradation decreases along with the income levels. The expected signs of ${\beta }_{3}$ and ${\beta }_{4}$ are positive because greater use of energy and other natural resources such as land, water etc. creates pressure on environmental quality.

The descriptive statistics of the variables over the period 1968–2020 are summarized in Table 1. The statistics show that all variables exhibit positive mean values and exceed their standard deviations as well. The kurtosis value of all variables is less than three. Only the $EFPPC$ and $ENGYPC$ variables have a negative skewness value. On the other hand, the coefficient of variation $(C.V.)$ of these two variables is the highest. According to the Jarque–Bera test statistics, all variables show a normal distribution at the 1% level of significance. The results of the correlation matrix show that the explanatory variables are generally and positively correlated except for $AGRVPC$ (see Table 7 under the Appendix part).

Table 1 Descriptive statistics of variables

As seen in Fig. 1, the $EFPPC$ variable has an increasing trend. However, the variability is greater due to more frequent fluctuations. The $EFPPC$ variable decreased suddenly in 2000 and increased to a higher level than the previous level in 2007 and continued to increase. $GDPPC$ shows a gradual upward trend. It is observed that there was a level shift in the series in 1980, 2001 and 2009. There is less fluctuation and a flatter increase in the series of GDP. The $ENGYPC$ variable shows a gradual upward trend. While the $ENGYPC$ series show little fluctiation, it is observed that there is a level shift in the series in 1979 and 2001.

Fig. 1

Illustration of the time series plots of the variables used in the current study

3.2 Cointegration methodology

Before analyzing the cointegration relationship, firstly we employed Augmented Dickey–Fuller test (ADF) by Dickey and Fuller (1981) and Kwiatkowski–Phillips–Schmidt–Shin (KPSS) test by Kwiatkowski et al. (1992) to have an idea about presence of stochastic stationary in our data. If the variables in our model have different degrees of integration, the traditionally used cointegration approaches such as Engle and Granger (1987) and Johansen (1988), which require the regressors to be integrated in the same order, cannot be applied to the time series of our model in this article. Therefore, the bound test procedure developed by Peseran et al. (2001) is employed by the current study. This test procedure has two advantages: (i) since unrestricted ECM (error correction model) is used under the ARDL methodogy, this procedure ensures better statistical properties than the Engle-Granger test, (ii) the ARDL has more reliable results than the Johansen ve Engle-Granger tests, especially for small samples (Narayan & Narayan, 2004).

Therefore, Eq. (2), the constant model in Eq. (3) and the constant and trend model in Eq. (4) can be rewritten as an ARDL formula as follows:

$$ \begin{aligned} \Delta \ln EFPPC_{t} = & \propto_{0} + \mathop \sum \limits_{i = 1}^{n} \beta_{1i} \Delta \ln GDPPC_{t - i} + \mathop \sum \limits_{i = 1}^{n} \beta_{2i} \Delta \ln GDPPC_{t - i}^{2} + \mathop \sum \limits_{i = 1}^{n} \beta_{3i} \Delta \ln ENGYPC_{t - i} \\ \quad & + \mathop \sum \limits_{i = 1}^{n} \beta_{4i} \Delta \ln AGRVPC_{t - i} + \beta_{5i} \ln EFPPC_{t - 1} + \beta_{6i} \ln GDPPC_{t - 1 } \\ \quad & + \beta_{7i} \ln GDPPC_{t - 1}^{2} + \beta_{8i} \ln ENGYPC_{t - 1} + \beta_{9i} \ln AGRVPC_{t - 1} + u_{t} \\ \end{aligned} $$

(3)

$$ \begin{aligned} \Delta \ln EFPPC_{t} = & \propto_{0} + \propto_{1} t + \mathop \sum \limits_{i = 1}^{n} \beta_{1i} \Delta \ln GDPPC_{t - i} + \mathop \sum \limits_{i = 1}^{n} \beta_{2i} \Delta \ln GDPPC_{t - i}^{2} + \mathop \sum \limits_{i = 1}^{n} \beta_{3i} \Delta \ln ENGYPC_{t - i} + \mathop \sum \limits_{i = 1}^{n} \beta_{4i} \Delta \ln AGRVPC_{t - i} \\ \quad & + \beta_{5i} \ln EFPPC_{t - 1} + \beta_{6i} \ln GDPPC_{t - 1 } + \beta_{7i} \ln GDPPC_{t - 1}^{2} + \beta_{8i} \ln ENGYPC_{t - 1} + \beta_{9i} \ln AGRVPC_{t - 1} + u_{t} \\ \end{aligned} $$

(4)

where Δ shows the difference operator, ${\beta }_{1}-{\beta }_{4}$ are the coefficients for short-term and ${\beta }_{5}-{\beta }_{9}$ are the coefficients for long-term.

The cointegration test is performed by testing the joint significance level of Eq. (3) the ARDL model variables using the Wald statistic. Testing the presence of cointegration relationship between variables by Pesaran et al. (2001) ARDL bounds testing, requires the estimation of conditional ECM given in Eq. (3) and Eq. (4). In testing the cointegration relationship using bounds testing approach, the null hypothesis is that coefficients of the lagged variables are jointly equal to zero. In testing the cointegration relationship using the bounds test approach, the null hypothesis of no cointegration relationship is tested by using the “the no-cointegration null hypothesis” equation as follows;

$${H}_{0}:{\beta }_{5}={\beta }_{6}={\beta }_{7}={\beta }_{8}={\beta }_{9}=0$$

$${H}_{1}:{\beta }_{5}\ne {\beta }_{6}\ne {\beta }_{7}\ne {\beta }_{8}\ne {\beta }_{9}\ne 0$$

If the estimated F-statistic value exceeds the upper bound critical value I(1) then, we accept the H₁ hypothesis which indicates the cointegration relationship between variables. On the contrary if the F-statistic is lower than the lower bound critical value I(0), we accept there is no cointegration relationship between the variables. If the calculated F-statistic falls between the I(0) and I(1), it means that the test is inconclusive about cointegration.

Once we obtain a cointegration relationship between the variables, then we can estimate ARDL process in the long-run as following (Eq. 5);

$$ lnEFPPC_{t} = \beta_{0} + \mathop \sum \limits_{i = 1}^{n} \beta_{1i} lnGDPPC_{t - i} + \mathop \sum \limits_{i = 1}^{n} \beta_{2i} lnGDPPC_{t - i}^{2} + \mathop \sum \limits_{i = 1}^{n} \beta_{3i} lnENGYPC_{t - i} + \mathop \sum \limits_{i = 1}^{n} \beta_{4i} lnAGRVPC_{t - i} + u_{t} $$

(5)

Then, the short-run coefficients are estimated using ECM based on the ARDL methodology as follows;

$$ \begin{aligned} \Delta \ln EFPPC_{t} = & \beta_{0} + \mathop \sum \limits_{i = 1}^{n} \beta_{1i} \Delta \ln GDPPC_{t - i} + \mathop \sum \limits_{i = 1}^{n} \beta_{2i} \Delta \ln GDPPC_{t - i}^{2} \\ \quad & + \mathop \sum \limits_{i = 1}^{n} \beta_{3i} \Delta \ln ENGYPC_{t - i} + \mathop \sum \limits_{i = 1}^{n} \beta_{4i} \Delta \ln AGRVPC_{t - i} + \beta_{5} ECT_{t - 1} + u_{t} \\ \end{aligned} $$

(6)

where ${\beta }_{5}$ is the coefficient of the error correction term and ${ECT}_{t-1}$ is lagged error correction term. The coefficient of ${ECT}_{t-1}$ should be negative and statistically significant for efficient ECM. In addition, this coefficient ${ECT}_{t-1}$ gives information about how fast the series attain the a long-run equilibrium.

The ARDL method tests the existence of short-run and long-run relationships between the variables$EFPPC$, $GDPPC,$ ${GDPPC}^{2},$ $ENGYPC$ and $AGRVPC$ variables, however it gives no idea about the directions of the causalities. Hence, Toda and Yamamoto (1995)'s technique is used as a causality test, since the series are not cointegrated at the same degree of integration. This test has an asymptotic χ2 distribution with k degrees of freedom at the boundary when VAR [k + dmax] is estimated (dmax is the maximum degree of integration for the series in the system). This method is implemented in two stages. The first stage includes determining the maximum degree of integration (d) and lag length (k) of the variables in the system. After these two parameters are determined, a VAR is estimated at p = [k + dmax] lag level. In the second step, the standard Wald test is applied to perform the Granger causality test on the first k lag coefficient parameters (not all lagged coefficients) of the VAR coefficient matrix (Table 2).

Table 2 Results of unit root test

3.3 Emprical results and discussion

3.3.1 Integration process

Augmented Dickey-Fuller test (ADF) and Kwiatkowski–Phillips–Schmidt–Shin (KPSS) tests are performed for variables at logarithmic levels and logarithmic first differences. While the null hypothesis of KPSS is trend stationarity, the null hypothesis for other unit root tests is that the series has a unit root. The results presented in Table 2 show that all variables have the first difference I(0) integrated degree based on the ADF unit root test. According to the KPSS unit root test, it was determined that the $AGRVPC$ variable had an integrated degree of I(0) in both levels and first differences. On the other hand, according to the KPSS unit root test, in which the intercept and trend are included, it was determined that the $EFPPC$ and $ENGYPC$ variables had I(0) integrated degrees both in level and first differences. Both unit root tests consistently show that none of series of the variables is integrated order two (I (2)). Therefore, our results show that one of the prerequisites of the bounds test approach is met.

Table 3 ARDL bound test

3.3.2 Cointegration Process

In the second stage of the ARDL test, the existence of a long-term cointegration relationship is investigated. After the unit root testing, the ARDL method of cointegration to estimate the relationship between the variables was employed (Pesaran et al., 2001). In the estimation of ARDL models, the Schwarz Bayesian information criterion (SBIC) was used to select the optimum lag length. Therefore, the SBIC-based ARDL (1) model was chosen for the ecological footprint equation.

The ARDL bounds test approach was used to test whether there is a long-term relationship between the ecological footprint per capita and the other explanatory variables. According to Eqs. (3) and (4), the hypothesis of ${H}_{0}={\beta }_{5}={\beta }_{6}={\beta }_{7}={\beta }_{8}={\beta }_{9}=0$ was tested. The F test statistic was calculated to test the validity of the H₀ hypothesis, which states that there is no cointegration relationship between the variables in the model.

To determine the appropriate lag length p and whether a deterministic linear trend is required in addition to the per capita ecological footprint variable, the conditional model (Eqs. 3 and 4) was estimated using Ordinary Least Square (OLS) with and without a linear time trend for p = 1,2, 3,..,5. ARDL boundary test results for Eqs. 3 and 4 are presented in Table 3. The SBIC selects the lag order as 1 in ARDL, and the (Breusch–Godfrey) LM test result shows no first-order serial correlation in the model. The calculated F-statistic is significant at the 1% significance level, regardless of whether the variables are fully I(1), fully I(0), or mutually cointegrated. Calculated F-statistic is greater than both the Narayan (2004) and Pesaran et al. (2001) upper bound critical values at the 5% level of significance. According to the results of the bounds test, the H₀ hypothesis, which states that there is no cointegration relationship between the variables, is rejected. According to this result, it is concluded that there is a long-term relationship between $EFPPC$, $GDPPC$, $ENGYPC$ and $AGRVPC$.

Short- and long-term coefficient estimates for the ARDL (1) model are presented in Table 4. Based on the R² (0.827), corrected R² (0.806) and F-statistics probability values (0.000) of the model, it was understood that the model as a whole was found to be significant. Since the Durbin-Watson test has a value of 2.161, this means that there is no autocorrelation problem in the data. The error correction term (ECM (− 1)) is statistically significant at the 1% significance level with the expected sign, confirming the cointegration relationship between these variables. The coefficient value of the error correction term which indicates the speed of adjustment when the balance experiences a shock, is − 1.228. This value implies that the rate of conversion to equilibrium is 123% and that discrepancies between shocks and trend are corrected in less than one year. If the coefficient of error correction is greater than 1, it shows that the system fluctuates (Narayan & Smyth, 2006). This fluctuation decreases each time, making it possible to return to the balance in the long run.

Table 4 Long-run and short-run coefficient estimates for the selected model ARDL (1)

The long-term results of the ARDL (1) model test show that there is a significant relationship between $EFPPC$ and other variables at the 5% significance level. The $ENGYPC$ variable is statistically significant at the 5% significance level, and 1% increase in$ENGYPC$, causes an increase around 0.401% in$EFPPC$. As argued by Marrero (2010), energy is a very important variable in validation of EKC hypothesis and it’s not surprising that energy use and power plants produce environmental pollution. $ENGYPC$, however, has a lower impact on ecological footprint than agricultural value added per capita variable. The effect of agricultural value added on ecological footprint is positive and statistically significant at 1% significance level. A 1% increase in agricultural value added per capita increases $EFPPC$ by 0.429% on average, other things being constant. Our estimation results show that agricultural prouduction in Türkiye deteriorates the environmental quality. This can be explained by the study of Tabar et al. (2010) which shows that agricultural activities lead to use of fossil energy (diesel or oil) and increased use of nitrogen rich fertilizer to protect yields. Again as highlighted by the study of Holly (2015), methane (CH₄) and nitrous oxide from livestock and crop production are among the main reasons for air pollution. Hence as stated by Shah et al. (2022) and Uddin (2020), agricultural production can lead to resource depletion (water, soil) and deforestration at a higher rate than regeneration rate especially in developing countries like Türkiye. Considering the our findings, our obtained results that agricultre and energy deteriorate the environmental quality are in line with the results of Usman and Makhdum (2021), Alvarado et al. (2021), Udemba (2020), Olanipekun et al. (2019), Pata (2021) and Salari (2021) but contradict with the results of Aziz et al. (2020) and ul Haq (2021). According to ARDL long-term coefficients, a 1% increase in GDPPC causes an increase of 6.411% in EFPPC at 1% significance level, and a 1% increase in ${GDPPC}^{2}$ causes a 0.35% decrease in EFPPC in Türkiye. Interestingly, while EFPPC and GDPPC have a significant positive relationship, ${GDPPC}^{2}$ and EFPPC have a robust negative coefficient revealing that EGR and EFPPC have a nonlinear inverted U-shaped linkage. Therefore, our obtained results confirm the validation of EKC for Türkiye. Our result is compatible with the result of Muoneke et al. (2022) who validated the agriculture-induced EKC for Philipinnes.

From the long-run regression equation, the turning point of $GDPPC$ is calculated to be ${\beta }_{1}/\left|{2\beta }_{2}\right|\cong 8184$ US Dollars. Since this income level is lower than the $GDPPC$ in our data sample, the EKC’s turning point is inside of our observation values. Accordingly, Türkiye has passed the turning point of EKC, with any further increase in income level, environmental degredation will be alleviated.

3.3.3 Diagnostic test

To evaluate the goodness of fit of ARDL, we performed the Breusch–Godfrey LM test for serial correlation, the Breusch–Pagan–Godfrey heteroscedasticity test for varying variance, and the Jarque–Bera normality test for normality of the residual components and the Ramsey RESET test for model specification are listed in Table 5. In addition, CUSUM and CUSUM of squares tests were performed to investigate the stability of the ARDL (1) model. Diagnostic tests show that the error terms of the short-run models are normally distributed, serial correlation and heteroskedasticity problems were not encountered in our model. The Ramsey zero test shows that the functional form is well defined for short-run models. The marginal probability values of Breusch–Godfrey LM test, Breusch–Pagan–Godfrey heteroscedasticity test, Jarque–Bera test, and Ramsey RESET tests are all greater than 0.10.

Table 5 Diagnostic test results of the ARDL model

We performed the CUSUM and CUSUMQ tests to evaluate the stability of the coefficients of the variables and to determine whether there is a structural break in the ARDL (1) model. As seen in Fig. 2, estimated parameters are stable over time since CUSUM and CUSUMQ statistics do not exceed the critical values at the 5% significance level. Therefore, there is no structural break.

Fig. 2

Plots of CUSUM and CUSUMSQ

3.3.4 Causality analysis

Table 6 summarizes the findings of the Toda–Yamamoto causality analysis. In this test, the H₀ hypothesis suggest that there is no causal relationship between the variables, whereas the H₁ hypothesis shows that there is a causal relationship between the variables. Our causality test results show that the null hypothesis cannot be rejected because the probability values between$ENGYPC$, $AGRVPC$ and $EFPPC$ are greater than the critical value of 0.05. However, no causal relationship was found between$ENGYPC$, $AGRVPC$ and $EFPPC$. Bidirectional causality was found between$EFPPC$, $GDPPC$ and${GDPPC}^{2}$, while unidirectional causality was found from $ENGYPC$ to$EFPPC$, $AGRVPC$ to$GDPPC$, $AGRVPC$ to${GDPPC}^{2}$. Our results are in line with the findings of Udemba (2020) who found one-way causal relationhip from energy use to EFP and Usman and Makhdum (2021) who found a feedback effect between AGR and EFP in BRICS countries.

Table 6 Toda–Yamamoto causality between variables

As stated in metedological part, the ARDL estimates gives us the long-run relationships among the variables but do not ensure the directions of causalities. The directions of causalities between the variables are depicted in Fig. 3. It is seeming from the Fig. 3, while $ENGYPC$ and $GDPPC$ are key determinants of environmental quality which is represented by EFP, $AGRVPC$ first stimulates the income and indirectly affects the $EFPPC$.

Fig. 3

Causalities between the variables in the model

4 Conclusions, policy recommendations and future research perspectives

The important gap in the previous agriculture-induced EKC studies in the literature is that, they mostly used CO₂ emissions in validation of EKC hypothesis. However, emissions alone can not represent all environmental degradation. Apart from the previous research studies, using the data spanning from 1968 to 2020 and ARDL approach, our paper empirically examines the role of agricultural production on EFP and tests the agriculture-induced EKC for Türkiye using energy to be control variable for the first time. Moreover, we employ Toda-Yamomoto test to obtain causal directions which are also important for the policy design towards sustainable development for the country.

The main findings of the current study can be summarized as following. Firstly, the results of the ARDL model validate the agriculture-induced EKC (inverted U-shaped relationship) between GDP and EFP for Türkiye. This highlights that GDP growth enable Türkiye to improve the environmental degradation. Accordingly, initially environmental degradation represented by EFP increases as the economy develops and then improves as the economy grows. Since Türkiye’s GDP has already exceeded the turning point, any increase in countrys’ income level will improve environmental quality. Secondly, the energy use seems to deteriorate the environment in the country. Accordingly, a 1% increase in ENGYPC increases the Türkiye’s EFP level by 0.401%. Third, our findings indicate that agricultural production in Türkiye exacerbates the country’s EFP. Accordingly, 1% increase of AGRVA enhances the Türkiye’s EFP level by 0.429%. This result is in line with the data showing that energy and/or electricity consumption have significantly increased in Türkiye’s agriculture sector. When the magnitude of the effect is compared, it is seen that the destructive effect of agriculture on the environment is greater than energy use in Türkiye. This result is not surprising since agriculture not only causes air pollution but also damage to natural resources due to the increasing use of fertilizers and pesticides, as indicated by the data. Fourth, there is a birectional causality between EFP and GDP. In contrast, a unidirectional causality from energy to EFP implies that policies towards to energy conservation are determinant to mitigate the environmental deterioration. It seems that energy saving, and energy efficiency are key issues to mitigate environmental degradation. Moreover, growth hypothesis is observed from agriculture sector to the GDP level.

Based on our findings, some some policy implications arise from obtain results. Since energy use deteriorates the environmental quality, Türkiye should accelerate the transition from fossil energy to RE technologies. Although, there has been important progress in installed RE in Türkiye thanks to implemented incentive policies such as feed-in tariff, tendering, blending ratio etc. under the Renewable Energy Law in 2005 (Law no: 6094), country can not fully exploit its RE potential. Hence, developing RE technologies, implementing additional incentive tools and greater use of RE in energy mix and introducing stricker environmental regulations by government will enable Türkiye to mitigate pollution and ensure to sustainable economic development. Similarly, as agriculture is important determinants of environmental quality, government should design policies to increase cleaner technologies such as solar, wind, biomass, biogas, geothermal in agricultural production and decrease the agricultural land use. Moreover, it’s very important to exploit the advantages of technology in agriculture sector that increases the productivity and reduces the use of fertilizer, energy sources and pesticides to avoid further degradation of environment. As stated by Di Vaio et al. (2020), ICT (information and communication technologies) and/or digital technologies can mitigate emissions and intensity of resources in agricultural production. Hence, circular agriculture model and smart agriculture solutions (Agriculture 4.0) should be adopted to promote sustainable development. In this context, as blockchain technologies encourage the achievement of all SDGs, not only agricultural firms but all firms should exploit the advantages of the technology and use business models to be built on blockchain infrastructure.

The current study is not without some limitations. The main limitation of this study is that the lack of data. The availability of data on energy (fossil and renewable) consumption didn’t let us to extend our sample size beyond 2020. We therefore suggest that studies should examine the energy type issues in detail as data becomes available in the future. Additionally, this study was not able to account for all variables affecting the EFP in the country. Future studies may include some other variables such as ICT, R&D, human capital, RE, agricultural trade, foreign direct investment (FDI) etc.that influence EFP.

Appendix

See Table 7.

Table 7 Correlation between variables