Climate change and agriculture nexus in Bangladesh: Evidence from ARDL and ECM techniques

The erratic weather puts farming households of Bangladesh at high production risk with significant consequences on food production, income, and livelihood. This study attempts to find the effect of various climate change indicators on agriculture in Bangladesh over the period 1980–2014. The study used the ARDL bounds testing approach to assess the long-run associations and the Granger causality test to determine the causal relationships between the regressors and dependent variables. The outcomes revealed that the first lag of agricultural value-added, second lag of carbon emissions, and average rainfall have a positive impact while the first lag of carbon has negative and significant impacts on agricultural production in the long run; in the short run-past realizations of carbon emission have a negative and significant impact on agricultural value-added. Additionally, the results show a unidirectional causality from carbon emission to agricultural output, agricultural output to average rainfall, and agricultural output to energy consumption. The study fills the gap in the climate change literature by applying the ARDL method to establish the nexus between climate change and agricultural output in Bangladesh.


Introduction
Globalization has led to rigorous, effective sustainable development and growth strategies to meet human needs directly or indirectly related to agriculture.Agriculture's physical and economic ties impede economic development and growth in emerging economies like Bangladesh.Agriculture prevails to measure climatic stress based on the impact of climate variability [1,2].Unpredictable climate occurrences, such as rising temperatures and decreasing precipitation, are also a factor in the sector [3].From 1.8 to 5.8˚C, and from 0.09 to 0.88 mm, the world saw the harshest effects of climate change in the last century [4].Furthermore, only South Asia faces a 0.016˚C to 1˚C temperature rise [5].However, simply 0.5˚C can reduce climate-related production, notably agricultural production, by 5.14 percent, and a 3˚C increase puts 600 million people at danger [6,7].There is an alarming consequence of environmental decadence, which is based on three major pillars of sustainable development.
Bangladesh's economy has achieved 6% GDP growth rate and lower-middle-income country status.Agriculture and livestock and fisheries contribute 10.11% and 3% to GDP respectively and employs 39.71 percent of active workers [8].Also, agro-based industries contribute 8% of manufacturing and 1% in GDP [9].However, Bangladesh has 1% of contribution to climate change, it has experienced damage in agricultural output, reduced livelihood possibilities, and exacerbated poverty.Annual GDP decline of 0.5-1%, projected loss will be $12 billion in coming 40 years [10].Global warming reduces rice yield by 28% and wheat production by 66% if temperatures climb by 4˚C [11,12].According to [13], water stress caused 8% and 32% yield losses in rice and wheat crops, respectively.Population growth increased agricultural productivity, increased energy use, economic growth, increased climate occurrences, and the fourth industrial revolution have all impacted agriculture.Agricultural output also provides some industrial raw materials, and industries contribute to GDP.Ensuring food security and raw materials in non-agricultural enterprises is essential.To accommodate rising demand, each stage of production uses more fossil fuels including electricity, gas, and oil.With rising energy demand, the sector contributes 1.23 percent to Bangladesh's GDP in power and 1.67 percent in gas, coal, oil, and other minerals.
Recently, agriculture has increased its usage of energy, in form of chemical fertilisers, irrigation, pesticides, electricity, oil, and gas from fossil fuels as well as human [14].However, increased energy demand in agricultural and non-agricultural output may result in increased carbon emissions [15].Agriculture emitted carbon by 23% itself and also responsible for climate change [16].CO2 emissions are blamed for persistent climate changes such increased temperatures, less rainfall, and groundwater degradation [17,18].However, CO2 is the most important ingredient for photosynthesis.[19] claim that more CO2 in the air helps retain water and acts as a fertilizer.This dual impact of CO2 has drawn attention to the researcher to study emissions, rainfall, temperature, and agricultural production.In recent years, researchers have examined the relationship between CO2 emissions and agricultural output, economic growth, gross domestic product (GDP), population, renewable and non-renewable energy consumption [20][21][22][23].The study aims to focus on the relationship between the agriculture sector and climate change in terms of GDP and energy consumption on carbon dioxide (CO2) emission in Bangladesh.The study also tries to explore both short-and long-run relationships into the series by autoregressive distributed lag (ARDL) at historical time frame 1980 to 2014.

Literature review
Agriculture productivity and efficiency are being challenged as well as food security and hunger prevention and malnutrition are being exacerbated by climate change due to the global production framework, consumption patterns, agriculture productivity and per capita CO2 emissions [24][25][26][27].However, CO2 may diminish the basic physical responses to water stress [28].Adaptation process work as a shield against the negative effects of climate change on different scales [29].However, adaption plans are incomplete without addressing the mitigation process [30,31].The agricultural sector is expected to work as part of mitigation by absorbing greenhouse gases, carbon, and other hazardous substances [32].However, some studies [22,33] identified ambiguous link between agriculture, GDP, and climate change and claimed agriculture including cattle, rice farming, fermentation, chemical fertilizer and also found that energy consumption are responsible for 21% of carbon emissions.
[34] used Fully Modified Ordinary Least Square (FMOLS) and Dynamic Ordinary Least Square (DOLS) to analyse economic growth, crop production, and livestock production contribute to CO2 emissions 17 percent, 28 percent, and 28 percent, respectively.The study also highlighted 1% of energy consumption improve environment.[35] explores the short and long-run impact of environmental pollution by Portuguese agriculture and energy consumption through carbon dioxide emissions, applying autoregressive distributed lag (ARDL), and Granger causality and Newey-West Standards Errors Regressions as well as Auto-Regressive Integrated Moving Average (ARIMA).An interrelation of South African carbon emission, agricultural output, and industrial output was established by [36].The investigation employed the ARDL technique to trace the significant influence in agrarian production by carbon emission and industrial production.An empirical study was performed on the Turkish economy about the amplification of carbon dioxide emission through economic growth by [23].Moreover, the study [23] explored that in the long-run GDP per capita, electric consumption, fiscal development, and trade openness will raise carbon dioxide emission at 0.14, 0.52, 0.09, and 0.20%, correspondingly using ARDL and Johansen cointegration.The study also revealed the validity of the Environmental Kuznets Curve (EKC).[24] employed ARDL and ECM as econometric tools to examine the impact of climate change on India's agricultural productivity and economic growth.He found a negative effect of carbon emission on economic development, whereas a positive association between farm output and economic growth.Furthermore, another assessment made by [37,38] used the Johansen cointegration approach (JCA) and vector error correction model (ECM) to examine the relationship of GDP on participation, temperature, and arable land over the period 1983-2013 in Malaysia.The result indicated a long-run association among the study variables and one-way causality running from temperature and arable land to GDP.Time-series data from 1960 to 2013 was employed to analyze long-run linkage across carbon dioxide emission and crop production and livestock production index for Ghana [Sarkodie et al., 2017].The study used both fit regression and ARDL models, which indicated bidirectional causality into crop production index and carbon dioxide emission by 0.52% and unidirectional cause of livestock production index and carbon dioxide emission by 0.81%.
In the context of Bangladesh, various studies have been done with carbon emission, energy consumption, GDP and climatic variability.[39] focus on the relationship between carbon dioxide emission, economic growth, and energy consumption with their development policies through different time series trends, using ARDL and Johansen cointegration model.Some studies [39,40] has got long run unidirectional causality between energy consumption and GDP.found a strong relation between mitigation, agriculture and carbon emission in Bangladesh.[41] shows that agricultural sector has experienced significant impact of exergy loss on energy sustainability.[42] employ Johansen co-integration to investigate long run interaction between energy consumption and industrialization with capital formation, infrastructure and manufacturing.[43] used Tapio decoupling method, and the logarithmic mean Divisia index to reveal strong and weakly decoupled to agricultural sector from sectoral development.The study found that population factor, the agricultural energy intensity factor, and the agricultural economic activity factor are liable for higher carbon emission.
All studies have concluded either long run or short run relationship between particular crop production and limited climatic variability, economic growth and carbon dioxide emission, deforestation, energy consumption, industrialization, population factor, energy intensity by using ARDL model, johansen cointegration, ECM, and ARIMA.As per our knowledge, no study particularly has contributed on how GDP relates with climatic variability, energy consumption and carbon emission at bidirectional short and long-run periods.

Variables and expectations
The study combines six variables from the [44], World Development Indicators (WDI), and the Bangladesh Metrological Department [45] to examine the relationship between climate change factors and agricultural productivity in Bangladesh from 1980 to 2014.Data on agrarian value-added (constant 2010 US$), carbon emission per capita, energy consumption per capita, and rural population (percentage of the total population) are obtained from WDI, while average temperature and average rainfall are from Bangladesh Metrological Department.

Econometric methodology
The current study attempts to investigate the relationship between agricultural value-added and carbon emission, average temperature, average rainfall, and energy consumption by utilizing the autoregressive distributed lag (ARDL) cointegration analysis introduced by [46,47] as this technique has a variety of flexibilities over traditional cointegrating methods proposed by [48,49].Traditional cointegration methods require the same order of integration of the variables; hence, the need to apply the ARDL methods, which allows the use of I(0) or I(1) or fractionally integrated variables [45].In addition, the ARDL technique is applicable for small samples and may be applied to obtain consistent estimates [50].Also, the ARDL model handles endogeneity in variables and gives more flexibility in selecting lags for both dependent and independent variables.These admirable properties endear the popularity of the ARDL approach among researchers.The study attempt to explain short run dynamics and long run relationships among the series with order of I(0) and I(1) together.In this case, ARDL method is appropriate for performing this study.To explore the relationship of climatic indicators such as carbon emissions, average temperature, and rainfall on agricultural output in Bangladesh as the empirical model is specified in implicit form as: Where, AVA represents agriculture value-added, CO 2 is CO 2 emissions, AT represents average temperature, AR signifies the average rainfall, EC is energy consumption.Eq [1] can also be written explicitly as: To limit the multicollinearity problem and control for outliers, all variables are transformed into natural logarithm and the log-linear equation is specified as: The ARDL bounds testing technique is applied to Eq [3] to inspect whether the study variables have cointegration as well as long-and short-term dynamics.Following [51][52][53][54] the ARDL (p, q,. ..,q) model is indicated as: Where α i and γ i (i = 1 to 5) are short-and long-run coefficients, α 0 is the constant, p and q are optimal lag orders, Δ represents the first difference operator and ε t is the error term that is independently and identically distributed.To test the long-run association between variables, the null hypothesis is stated as . Now the decision of rejection or not rejecting the null hypothesis depends on the value of F-statistic if it is above or below the upper or lower critical value.The null hypothesis of no cointegration is rejected when the F-statistic is higher than the critical value of the upper bounds of the I(0) and I(1) series and not rejected if otherwise [52].But the decision is inconclusive if the statistic calls between the upper and lower bounds.If the longrun association amongst the variable is established, then the short-run parameters are estimated by the error correction mechanism (ECM) depicted as: The error correction method explains the speed of adjustment required to restore long-term equilibrium after a shock in the short-run.Eq [5] shows that agricultural value-added depends on its lag, the contemporaneous and lags of the regressors, and the lag of the equilibrium error term, θ needed is expected to be negative (lies between 0 and -1) because it indicates how much equilibrium is restored on its absolute.However, if it is positive, then the model is out of equilibrium, that is, explosive without reversion to long-run equilibrium.The optimal lag orders for each variable are derived from the Akaike information criteria (AIC) and shown in Table 3.

Descriptive statistics and correlation analysis
The results of the descriptive statistics are shown in Table 1.Data extracts that the minimum agricultural value-added of 108.61 million US$ was recorded in 1980, while CO 2 0.096 metric tons per capita, average temperature 24.64˚C, average rainfall 158.62mm, and Energy consumption 104.86 Kg oil equivalent per capita were for 1983,2009,1992,1981  The upper panel shows that agricultural value-added, carbon emission, average temperature, average rainfall, and energy consumption are generally distributed with mean zero and variance from the Jarque Bera statistics.Additionally, the estimated outcomes of the pairwise correlation displayed in the lower panel of Table 1 show that except average rainfall, CO 2 emissions, average temperature, and energy consumption have positive and significant association with agriculture at the 1% and 5% levels.

Unit root tests
Examining the stationary properties of the variables to investigate the presence of cointegration is imperative.The most crucial assumption of applying the ARDL method is that the series must be integrated order I(0) and I(1), or all of them are I(1) series.ARDL technique is not applicable if any of the variables are integrated of order 2, that is I(2) series because, at that instance, the mathematical limits of ARDL bounds and F-statistic used for checking cointegration among variables becomes meaningless [55].This study applies [56,57] tests for testing the stationary properties of the variables under study.The mean and variance do not change over time in stationarity [58].Table 2 indicates that the variables have mixed orders of integration, and none of them is integrated of order 2, hence justifies the use of the ARDL model for this analysis.

Lag length criteria
The optimal lag length of the study variables must be determined before applying the ARDL model.Appropriate lags for the study variables should be selected very carefully as incorrect lags may lead to inconsistent that are inappropriate for policy analysis.Akaike information criterion (AIC) and Schwarz Criterion (SC) are two renowned methods for choosing the most appropriate lags.The optimal lag for this analysis is selected by using the AIC criterion and the result of Table 3 indicates that lag 2 suits our sample size hence selected as optimal lag.

Bound cointegration test results
For the ARDL approach, this study used the AIC to select the appropriate lag length [proposed by 46 and 59].The outcomes of ARDL bounds tests shown in Table 4 confirm a long-run association between agricultural value-added (LnAVA) and its regressors.The F-statistic of 8.169, which is statistically significant at 1%, reveals that carbon emissions, average temperature, average rainfall, and energy consumption are the forcing variables that move first when a common stochastic shock hits the system.The above finding implies that agricultural value-added follows changes in these indicators.We test for the robustness of the model with the inclusion of the rural population.The F-statistic of 7.959 is also statistically significant at 1%, and the same conclusion holds: the null hypothesis of no cointegration is rejected.

Long-run and short-run analysis
The results of the long-run analysis are reported in Table 5.At the upper panel, the estimated coefficients of the long-run relationship are significant for the first lag of agricultural valueadded (0.888), the first and second lags of carbon emissions (-0.165, 0.159), and average rainfall (0.079).These variables are strong predictors of agricultural value-added, suggesting that a percentage change in the first lag of agricultural value-added is associated with 0.89 percent increase in agricultural value-added at the 1% significance level, on average, ceteris paribus.In contrast, a percentage change in the first lag of carbon emissions leads to 0.17 decrease in agricultural value-added at the 10%significance level, on average, ceteris paribus.Similarly, a percentage change in average rainfall contributes about 0.08 percent increase in agricultural value-added at the 1% significance level, on average, ceteris paribus.The outcomes of this research are consistent with other studies, evidencing that previous period agricultural valueadded have positive impact on current agricultural value-added as last year's high price motivates the farmers to produce more value-added [60,61].Similarly, studies of [62][63][64] support the positive impact on carbon emission on agricultural value-added, whereas studies of [65,66] indicate adverse effects.Our finding aligns with similar studies that found positive and significant relation between rainfall and agricultural productivity [67,68].These outcomes provide strong evidence on the impact of climate change on agro-productivity.For example, carbon emissions have an asymmetric influence on agriculture, implying both detrimental and good consequences.On the other side, rainfall has been shown to have significant and beneficial effects on agriculture.Productivity will grow with increasing rain.These are added empirical evidence to the agro-climate change literature.The coefficient on the lagged EC term (-0.112) is statistically significant at 1% and suggests a reversion to long-run equilibrium.It also implies that a deviation from the equilibrium level of agricultural value-added during the current period will be corrected by 11.2 percent in the next period.
For the short-run analysis in the lower panel of Table 5, average temperature, average rainfall, and energy consumption appear in the model contemporaneously (they have 0 lags).Hence, they are not reflected in the short-run analysis.Only the coefficient of past realizations of carbon emissions is negative and statistically significant at the 5% level.It implies that a percentage change in past carbon emission realizations is associated with a 0.159 decrease in the current level of agricultural value-added, on average, ceteris paribus.This outcome asserts the significance of carbon emission in affecting agricultural value-added both in the long-and

Diagnostics results
Table 6, which shows the estimated results of diagnostic tests for both models (main and robustness), does not provide evidence of higher-order autocorrelation and heteroscedasticity.Moreover, the Jarque-Bera test indicates that the ARDL model is correctly defined, and the residuals are distributed normally.This study used the CUSUMSQ test proposed by [69] to stabilize the long-run parameters for both models.The CUSUMSQ plots are within a 5 percent significance level (See Figs 1 and 2), showing no signs of long-run parameter instability.

Granger causality test
The Granger causality tests show the relationships in pairs, which can be unidirectional, bidirectional, or no causal relation, and the results are shown in Table 7.On causal analysis, the null hypothesis that carbon emissions do not Granger cause agricultural value-added is rejected at the 10% significance level and reveals a unidirectional causality from LnCO2 to LnAVA.Likewise, the null hypothesis that agricultural value-added does not Granger cause average annual rainfall is rejected at 5% level of significance, evidencing unidirectional causality from LnAVA to LnEC.Moreover, the unidirectional causality from LnAVA to LnEC indicates that agricultural value-added Granger causes energy consumption at 1% level of significance.

Conclusion and policy recommendations
From the Sustainable Development Goals 2, 13, and 14, climate change poses a severe threat to global agriculture.Developing economies are more at risk for the repercussions of climate change than developed countries, particularly those who rely on nature and farmlands for their livelihoods [70,71].For instance, Bangladeshi economy in term of GDP, economic growth and employment is susceptible to the threat of climate change, as they rely heavily for their subsistence on weather-dependent agriculture.Consequently, prudent adaptation is pursued to decrease the possible negative externalities in agricultural productivity.This study evaluates the impact of various climate change indicators (such as carbon emissions, average rainfall, average temperature, and energy consumption) on agricultural productivity in Bangladesh over the period 1980-2014.Our findings, amongst others, enrich the climate change literature by providing evidence that carbon emissions have an asymmetric impact while average rainfall has consistent positive effects on agricultural value-added.The results of the causality tests reveal that unidirectional causality is from carbon emission to agricultural output, agricultural output to average rainfall, and agricultural output to energy consumption.Some policy recommendations based on our findings are not far-fetched.Government should implement effective weather forecasting technology and minimize the detrimental effects of climate change on agricultural products such as heat and drought through improved tolerant seedlings.Likewise, improved irrigation systems need to be developed, whereas per unit price of electricity is one of the significant factors related to improved irrigation.Subsidy on electricity consumption may increase the irrigation facilities in the agricultural sector significantly to boost the production of boro (most produced variety) rice.Government may organize training program of farmers to train in use of chemical fertilizer, pesticide and other energies to use energy efficiently.The report also states that stakeholders and policymakers must support adaption and mitigation techniques that would country or region-specific, or crop-specific to reduce the negative impact of climate change on agricultural productivity.Moreover, the government should introduce several new policies regarding improved technologies to achieve sustainable production of key crops.Given data availability, regional-based impact, crop-specific impact, and renewable energy impact on food security in Bangladesh may be taken up in future research.

Table 5 . ARDL and ECM results.
To test the robustness of our results, the log of the rural population is added to the model.The results shown on the right-hand side of Table5are not significantly different from the main results.Hence, this study concludes that climate change indicators are strong predictors of agricultural value-added.