The Kenya Case of Multivariate Causality of Carbon Dioxide Emissions

In this study, an attempt was made to investigate the Kenya case of multivariate causality of carbon dioxide emissions by employing a time series data spanning from 1961-2011 using the ARDL method of cointegration analysis. The long-run elasticities show that, a 1% increase in financial development increases carbon dioxide emissions by 0.28%, a 1% increase in GDP per capita increases carbon dioxide emissions by 1.32% and a 1% increase in urbanization decreases carbon dioxide emissions by 1.14%. There was a unidirectional causality running from financial development, food production index, GDP per capita, industrialization and urbanization to carbon dioxide emissions. The innovation accounting shows that 20% of future shocks in carbon dioxide emissions are due to fluctuations in financial development, 9% of future shocks in financial development are due to fluctuations urbanization and 22% of future shocks in food production index are due to fluctuations in carbon dioxide emissions.


INTRODUCTION
Climate change has taken the centre stage in the developmental agenda in developed and developing countries [1].Global effort has been made through the establishment of the sustainable development goals to promote renewable and clean energy technologies, sustainable agriculture and food security, and mitigate climate change and its impact.There is a close relationship between carbon dioxide emissions, industrialization, urbanization, financial development, economic growth and human well-being [2,3,4,5].This is because industrialization, urbanization, financial development, economic growth and carbon dioxide emissions have significant bearing on human development indicators such as incomes, wage employment, skill formation, improved livelihoods (air quality, clean environment, nutrition and health care), gender parity, and entrepreneurship.Significantly, industrial and technological advancement have played a role in food production, processing, storage, nutrition, food security and agricultural tools and techniques [6,7].
As a result of the high levels of multidimensional poverty, the Government of Kenya is working closely to improving the welfare of Kenyans through industrialization.There has been significant improvement in Kenya's industry policies since mid-1980s leading to its vast contribution to the country's GDP, source of employment opportunities and increasing the industrial output through manufacturing activities [6].Notwithstanding, there are challenges associated with industrialization such as; rapid urbanization and poor environmental and health quality as a result of industrial carbon dioxide emissions [9].
According to UNDP [8], changes in climatic patters as a result of global carbon dioxide emissions are now creating harmful impacts on the Kenyan environment, society and economy.
As a result of uncertainties about weather patterns, economic sectors like tourism and agriculture are accruing a significant economic loss.According to IMF [10], "Kenya remains vulnerable to financial shocks that could have a significant adverse impact on the economy".
It is projected that Kenya will require about US$ 1-2 billion yearly by 2030 to address the current and future climate change effects [11].
Against the backdrop, it is worthwhile to examine the multivariate causality of carbon dioxide emissions in Kenya using a time series data spanning from 1961-2011.To the best of our knowledge, the scope of the study is the first time in Kenya which will contribute to existing literature from the Kenya case and further increase the global debate on climate change from the Kenya perspective.Since carbon dioxide emissions, energy consumption/production and GDP have been proven to be collinear in many studies from different countries [12,13].The current study eliminates energy consumption/production and rather examines the equilibrium relationship between carbon dioxide emissions, food production index, financial development, economic growth, industrialization and urbanization using the ARDL method of cointegration analysis.The study further estimates the Granger-causality and the variance decomposition based on VAR.
The remainder of the study consist of "Literature review", "Methodology", "Results and Discussion" and "Conclusion and Policy recommendations".

LITERATURE REVIEW
Within the last decades, the relationship between environmental pollution, energy consumption and macroeconomic variables (financial development and economic growth) have received considerable attention in scientific literature.
HuangHwang and Yang [54] found no relationship between energy consumption and GDP.
Soytas and Sari [53] found that carbon dioxide emissions Granger cause energy consumption while Zhang and Cheng [61] found no evidence of causality from carbon dioxide emissions or energy consumption to economic growth.GulZouHassan et al. [56] found evidence of a unidirectional causality running from energy-consumption to carbon dioxide emissions.
Jammazi and Aloui [57] found a bidirectional causality between energy consumption and economic growth, and a unidirectional causality between energy consumption and carbon dioxide emissions.
Nevertheless, the scope of the study is sporadic and limited in Kenya.To the best of our knowledge, only Al-MulaliSolarin and Ozturk [30] have examined the validity of the Environmental Kuznets Curve hypothesis in Kenya with a time series data spanning from 1980-2012 using the ARDL method of cointegration analysis.Their study does not support the validity of the EKC in Kenya.Unlike their study, the current study examines the relationship between carbon dioxide emissions, food production index, GDP per capita, financial development, industrialization and urbanization.In addition, the direction of causality and innovation accounting using Cholesky's technique is employed in the Kenya case which were absent in previous study [30].The study contributes to existing literature by expanding the period of the time series data from 1961-2011 compared to previous 1980-2012, in order to provide formidable statistical evidence.Moreover, the study increases the global debate on climate change and its impact from the Kenya context and serve as a policy document for future national planning and strategies on climate change mitigation.

Data
The study investigates the Kenya case of multivariate causality of carbon dioxide emissions by The World Bank [71] defines Money and quasi money as "the sum of currency outside banks, demand deposits other than those of the central government, and the time, savings, and foreign currency deposits of resident sectors other than the central government", it is therefore used as a proxy for financial development (FD).Moreover, the World Bank [71] defines Industry value added as "the value added to mining, manufacturing, construction, electricity, water and gas", it is therefore used as a proxy for industrialization (INV) [71].

Descriptive Analysis
The study presents the descriptive statistical analysis of the time series variables from 51 observations as showed in Table 1.Information from Table 1 shows that all the variables exhibit a long-right-tail (positive skewness) with INV having the higher skewness.While FD, GDPPC, and INV exhibit leptokurtic distribution, CO2, FPI and URB exhibit a platkurtic distribution.
The correlation analysis shows that all the independent variables have a positive monotonic relationship with CO2.Jarque-Bera test statistic suggests that FD, GDPPC and INV are not normally distributed based on 5% significance level.Therefore, a logarithmic transformation is applied to the variables in order to provide a more stable data variance for the subsequent analysis.At this juncture, let LCO2, LFD, LFPI, LGDPPC, LINV and LURB represent the logarithmic transformation of CO2, FD, FPI, GDPPC, INV and URB.
Figure 1 shows the trend of the study variables.Figure 1 shows that carbon dioxide emissions, financial development, food production index, GDP per capita, industrialization, and urbanization increase periodically which suggest the existent of a strong relationship among them.However, the trend of carbon dioxide emissions decreased over the period 2000-2003 due to a decline of oil imports as a result of Kenya's energy efficiency and conservation policy [2].

Model Estimation
The relationship between carbon dioxide emissions, food production index, financial development, GDP per capita, industrialization and urbanization in Kenya is expressed as a linear function showed in equation ( 1): .
The empirical specifications for the selected ARDL (1, 1, 1, 1, 0, 2) model is quantified as: The study employs the ARDL method of cointegration to estimate the long-run and short-run equilibrium relationship between LCO2, LFD, LFPI, LGDPPC, LINV and LURB.Following the work of Asumadu-Sarkodie and Owusu [7],Asumadu-Sarkodie and Owusu [12] and Al-MulaliSolarin and Ozturk [30], the ARDL co-integrating equation is expressed as: According to PesaranShin and Smith [72], the calculated F-statistic is compared with the critical values of the lower and upper bounds respectively.If the calculated F-statistic goes above the upper bound, the null hypothesis of no cointegration between is rejected.However, if the F-statistic is smaller than the critical value of the lower bound, the null hypothesis of no cointegration cannot be rejected.In addition, if the F-statistic lies between the critical values of the lower and the upper bounds, the null hypothesis of no cointegration become inconclusive, which requires either the estimation of Johansen's test of cointegration [73] or through testing the constancy of the cointegration space using CUSUM and CUSUM of squares of residuals [74].
Unlike Johannsen's method of cointegration approach which employs a set of cointegration equations to analyse the long-run equilibrium relationship between variables, the ARDL method of cointegration by PesaranShin and Smith [72] The joint short-run effect is estimated using the Wald test of linear restrictions to the coefficients of LFD, LFPI, LGDPPC, LINV and LURB in equation ( 2).From equation (2), we derive that = = 0, = = 0, = = 0, = 0 and = = = 0.

RESULTS AND DISCUSSION
This section presents the results and a discussion of the empirical analysis.

ARDL Co-integration
Having determined the integration of variables at I(1), the study selects an optimal model using the Schwarz Criteria.As stated in equation ( 2), the selected model using the Schwarz Criteria is ARDL (1, 1, 1, 1, 0, 2) as shown in Figure 2. Using the optimal model, the ARDL bounds testing is estimated as showed in Table 3. Table 3 shows that the F-statistic lies above the critical values of the upper bound at 10, 5 and 2.5% significance level, therefore the null hypothesis of no long-run relationship is rejected at 5% significance level.Table 3 further presents the error correction, long-run elasticities and short-run equilibrium relationship.Table 3 shows that the speed of adjustment [ECT (-1) = -0.37] is negative and significant at 5% level, meaning that a long-run equilibrium relationship exist running from LFD, LFPI, LGDPPC, LINV and LURB to LCO2.The joint test of linear restrictions of the coefficient in the shortrun estimates shows that LFD, LFPI, LGDPPC, LINV and LURB affect LCO2 in a short-run.
The evidence from the long-run elasticities in Table 3 has policy implications for Kenya.

Granger-Causality
Due to the inability of the ARDL model to estimate the direction of causality, the study employs the Granger-causality based on VAR to examine the direction of causality among the variables.

Innovation Accounting
The ARDL method is able to examine the long-run and short-run equilibrium relationship while the Granger-causality test examines the direction of causality.Nevertheless, the impulseresponse function that traces the effect of a shock from one endogenous variable on the other variables is uncertain in both ARDL and Granger-causality.Against the backdrop, the study employs the innovation accounting based on Cholesky's technique in order to analyze the variance decomposition of each random innovation affecting the variables in the VAR.
Table 5 shows that 20% of future shocks in LCO2 are due to fluctuations in LFD, 12% of future shocks in LCO2 are due to fluctuations in LURB, 12% of future shocks in LCO2 are due to fluctuations in LGDPPC, 12% of future shocks in LCO2 are due to fluctuations in LFPI and 10% of future shocks in LCO2 are due to fluctuations in LINV.
Table 5 shows that 9% of future shocks in LFD are due to fluctuations in LURB, 7% of future shocks in LFD are due to fluctuations in LFPI, 5% of future shocks in LFD are due to fluctuations in LCO2, 4% of future shocks in LFD are due to fluctuations in LINV and 3% of future shocks in LFD are due to fluctuations in LGDPPC.
Table 5 shows that 22% of future shocks in LFPI are due to fluctuations in LCO2, 17% of future shocks in LFPI are due to fluctuations in LINV, 4% of future shocks in LFPI are due to fluctuations in LFD, 3% of future shocks in LFPI are due to fluctuations in LGDPPC and 3% of future shocks in LFPI are due to fluctuations in LURB.
Moreover, evidence from Table 5 shows that 16% of future shocks in LGDPPC are due to fluctuations in LFD, 15% of future shocks in LGDPPC are due to fluctuations in LFPI, 6% of future shocks in LGDPPC are due to fluctuations in LCO2, 6% of future shocks in LGDPPC are due to fluctuations in LURB and 3% of future shocks in LGDPPC are due to fluctuations in LINV.
Table 5 shows that 43% of future shocks in LINV are due to fluctuations in LGDPPC, 22% of future shocks in LINV are due to fluctuations in LFPI, 13% of future shocks in LINV are due to fluctuations in LFD, 8% of future shocks in LINV are due to fluctuations in LCO2, and 7% of future shocks in LINV are due to fluctuations in LURB.
Table 5 shows that 22% of future shocks in LURB are due to fluctuations in LFPI, 17% of future shocks in LURB are due to fluctuations in LFD, 11% of future shocks in LURB are due to fluctuations in LCO2, 4% of future shocks in LURB are due to fluctuations in LINV and 3% of future shocks in LURB are due to fluctuations in LGDPPC.

Diagnostic and Stability Checks
Diagnostic and stability checks were performed to examine the independence of the residuals.
The Jarque-Bera test in Figure 3 shows that the null hypothesis of normal distribution in the residuals cannot be rejected at 5% significance level.The ARDL model shows a long-run and a short-run equilibrium relationship running from food production index, financial development, GDP per capita, industrialization and urbanization to carbon dioxide emissions.The long-run elasticities show that, a 1% increase in financial development increases carbon dioxide emissions by 0.28%, a 1% increase in GDP per capita increases carbon dioxide emissions by 1.32% and a 1% increase in urbanization decreases carbon dioxide emissions by 1.14%.The higher effect of per capita GDP and financial development on carbon dioxide emissions in Kenya is due to the higher depletion and degradation of natural resources for industrial purposes in order to achieve the 2030 development strategy of reaching a middle income country status of US$1,000 per capita GDP with an accelerated economic growth of 6% [75].As a global menace, Kenya is experiencing higher levels of urbanization in two industrial hubs namely Nairobi and Mombasa; the former been industrial and services hub for the local and regional markets while the later been a costal industrial hub for the emerging global markets.Nevertheless, Kenya is making the best out of urbanization to improve the labour force leading to an accelerated growth in economy and literacy which serves as a way of mitigating climate change and its impact through awareness creation.
Moreover, there was evidence of a unidirectional causality running from financial development to carbon dioxide emissions, food production index to carbon dioxide emissions, GDP per capita to carbon dioxide emissions, industrialization to carbon dioxide emissions, urbanization The evidence from the long-run and short-run equilibrium relationship, the Granger-causality and the innovation accounting have policy implications for Kenya.All fast-growing developing economies are rapidly urbanizing, therefore Kenya should transform the rate of urbanization into good use through the creation of decent jobs and a strong labour force to increase economic productivity.Achieving higher levels of economic growth and productivity through technological improvements, innovation and creativity, diversification and high-value added to raw materials is a requirement for Kenya to achieve the middle-income status by 2030.
Finally, as extreme levels of carbon dioxide emissions affect food production index, efforts by the Government of Kenya that promotes sustainable agriculture through modern technologies and improved agricultural techniques would boost Kenya's economic productivity, promote food security while mitigating climate change and its impacts.
employing a time series data from the World Bank [71] at a period spanning from 1961-2011 using the ARDL method of cointegration analysis.Six study variables are used in the study which include: CO2 -Carbon dioxide emissions (kt), GDPPC -Gross Domestic Product per capita (current LCU), IND-Industry, value added (current LCU), FPI-Food production index (2004-2006 = 100), FD-Money and quasi money (M2) (current LCU), URB-Urban population.
in year , − 1 and − 2 represents lag 1 and 2 and 's are the elasticities to be estimated.The substituted coefficients from the estimated equation are; 61 (0.00) and = 8.82 (0.00).The estimated coefficients are all significant at 5% level with the exception of and .

,
Preprints (www.preprints.org)| NOT PEER-REVIEWED | Posted: 26 December 2016 doi:10.20944/preprints201612.0127.v1where is the intercept, is the lag order, is the error term and ∆ is the first difference operator.The application of ARDL cointegration among variables can be estimated at either I(0) or I(1) without pre-specification of variables which are either I(0) or I(1).Moreover, ARDL has desirable small sample properties and provide unbiased long-run estimation, even when some endogenous variables behave as regressors.The initial step of ARDL cointegration is the bounds testing procedure which is based on the F-test.The Null hypothesis of no cointegration between LCO2, LFD, LFPI, LGDPPC, LINV and LURB is :

Figure 4 .
Figure 4. (a) CUSUM of Squares and (b) CUSUM of the Residuals

Preprints
(www.preprints.org)| NOT PEER-REVIEWED | Posted: 26 December 2016 doi:10.20944/preprints201612.0127.v1 to carbon dioxide emissions, food production index to GDP per capita, industrialization to GDP per capita and urbanization to GDP per capita.The innovation accounting based on the Cholesky's technique of variance decomposition shows that 20% of future shocks in carbon dioxide emissions are due to fluctuations in financial development.In addition, 9% of future shocks in financial development are due to fluctuations urbanization, 22% of future shocks in food production index are due to fluctuations in carbon dioxide emissions, 16% of future shocks in GDP per capita are due to fluctuations in financial development, 43% of future shocks in industrialization are due to fluctuations in GDP per capita and 22% of future shocks in urbanization are due to fluctuations in food production index.

Table 1 .
Descriptive Statistical Analysis stationarity in the presence of structural breaks, the study estimates the order of integration with Vogelsang's breakpoint unit root test by considering the presence of innovational outliers.Information from Table2shows that, the null hypothesis of non- Figure 1.Trend of Variables3.3 Stationarity TestThe empirical analysis begins with testing for the stationarity properties of the variables.The study employs the Kwiatkowski-Phillips-Schmidt-Shin (KPSS) and Vogelsang's breakpoint tests in order to have a robust result.KPSS test results in Table2shows some different specifications, however the null hypothesis of stationarity is rejected at level in all the variables but cannot be rejected at first difference based on 5% significance level.Since KPSS fails to test the

Table 3 .
ARDL Bounds Test, Error Correction, Long-Run and Short-Run Relationship

Table 5 .
Innovation Accounting based on Cholesky's technique

Table 6 .
ARDL Diagnostic Test Jarque-Bera Test of ResidualsFigure4shows the CUSUM of Squares and CUSUM tests for checking the constancy of the cointegration space in the residuals of the ARDL model.Figure4shows that the CUSUM of Squares and CUSUM plots lie within the 5% significance level.The diagnostic and stability checks shows that the residuals in the ARDL model are independent and have stable parameters to make unbiased statistical inferences.