The Causal-Effect between Carbon Dioxide Emissions and Forestry Production and Trade : A Case Study in Ghana

In this study, the causal-effect between carbon dioxide emissions and forestry production and trade was investigated in Ghana by employing a data spanning from 1961 to 2014 by using the VECM and ARDL model. Evidence of the long-run equilibrium relationship in the VECM shows that, a 1% increase in veneer sheet production reduces carbon dioxide emissions by 1.47% in the long-run. There was evidence of a bidirectional causality between carbon dioxide emissions and veneer sheet production, carbon dioxide emissions and wood charcoal production, and a unidirectional causality running from carbon dioxide emissions to wood fuel production and plywood production to carbon dioxide emissions. Evidence from the long-run equilibrium relationship in the ARDL model shows that; a 1% increase in plywood production will increase carbon dioxide emissions by 0.17% in the long-run, a 1% increase in sawnwood production will increase carbon dioxide emissions by 0.17% in the long-run, a 1% increase in wood charcoal production will increase carbon dioxide emissions by 0.36% in the long-run and a 1% increase in wood fuel production will increase carbon dioxide emissions by 0.37% in the long-run.


Introduction
Greenhouse gas emissions have become a global concern which has attracted attention from researchers and policy makers within the last decades [1][2][3][4][5][6].According to UN-REDD [7], "deforestation and forest degradation, through agricultural growth, conversion to pastureland, infrastructure development, destructive logging, fires, and among others, account for nearly 20% of world greenhouse gas emissions, more than the whole international transportation sector and second solely to the energy sector.It is now clear that in order to constrain the impacts of climate amendment among limits that society can reasonably be able to tolerate, the worldwide average temperatures should be stabilized at intervals 2°C.This will be practically not possible to attain while not reducing emissions from the forest sector, in addition to different mitigation actions".
Illegal chainsaw operators have played a role in Ghana's deforestation accounting for a deforestation rate of 65,000 hectares per annum, in addition, the illegal chain saw milling accounts for 84% of Ghana's annual lumber supply of 497,000 cm 3 and a market value of more than US$ 200 million [8].Due to illegal chainsaw, Ghana's estimated forest reserve that stood at 8.2 million hectares in the 20 th century has eventually dropped to 1.6 million hectares [8,9].Nevertheless, the environmental pollution and the impact of the deforestation in Ghana has not been investigated to the best of our knowledge.
Accordingly, the study aims at analysing the causal-effect between carbon dioxide emissions and forestry production and trade: a case study in Ghana.Previous studies examined the relationship between carbon dioxide emissions, energy consumption, population and GDP in Ghana [10,11] while another study focused on the causal nexus between carbon dioxide emissions and agriculture in Ghana [12].In both cases, there were evidence of a long-run equilibrium relationship between carbon dioxide emissions and environmental pollution in Ghana.Nevertheless, the current study makes an attempt to examine the role of forestry production and trade in environmental pollution by employing a time series data spanning from 1961-2014 by employing both Vector error correction and ARDL models.As a contribution to literature, the study employs the Kendall's tau-b correlation and bootstrapping test for nonparametric and parametric estimates to examine the strength of association in the descriptive statistical analysis.In addition, the study examines the random innovations of variables in the VAR by employing the Cholesky impulse-response test.The study will increase the global debate on the role of forestry production and trade in environmental pollution from the Ghana case.Significantly, the study will serve as an information tool for future national policies, strategies and planning in Ghana while playing a role in climate change mitigation and sustainable development.
The remainder of the study consists of "Methodology", "Results and Discussion", "Conclusion and Policy Recommendations".

Methodology
The study examines the causal-effect between carbon dioxide emissions and forestry production and trade: a case study in Ghana by using both Vector Error Correction Model (VECM) and Autoregressive Distributed Lag (ARDL) model.A time series data spanning from 1961 to 2014 were employed from the FAO database [13].Eight variables were used in the study which comprise; CO2 -Carbon dioxide emissions (kt), WC-Wood Charcoal Production (Tonnes), SV-Sawlogs & Veneer Logs Production (Tonnes), WF-Wood Fuel Production (Tonnes), S-Sawnwood Production (Tonnes), VS-Veneer Sheets Production (Tonnes), P-Plywood Production (Tonnes) and WBP-Wood-Based Panels Production (Tonnes).

Descriptive Statistical Analysis
The study examines the characteristics of the datasets before proceeding to the model estimation.Table 1 presents the descriptive statistical analysis of the study variables.Evidence from Table 1 shows that S and SV are negatively skewed while CO2, P, WBP, WC and WF have a long-right tail (positive skewness).While CO2, SV, WBP, WC and WF exhibit a platkurtic distribution, S and P exhibit a leptokurtic distribution.The Jarque-Bera statistic test shows that CO2, P, VS, WBP and WF are not normally distributed.As a result, logarithmic transformation is applied to have a more stable data variance.At this moment, let LCO2, LP, LS, LSV, LVS, LWBP, LWC and LWF represent the logarithmic transformation of CO2, P, S, SV, VS, WBP, WC and WF.Table 2 presents the Kendall's tau-b correlation test with bootstrap results based on 1000 bootstrap samples.Evidence from Table 2 shows that with the exception of LSV, the linear relationship between the dependent variable (LCO2) and the independent variables (LP, LS, LVS, LWBP, LWC and LWF) are significant at 1% with less than 1% reported standard error and 0.3% bias reported by the bootstrap.It is evident from Table 2 that the correlation coefficient for the relationship between LCO2 and all the variables is less than 0.90 which indicates a less possibility of multicollinearity among the study variables.Fig. 1 shows the trend of the variables.

Unit Root Test
To examine the causation among variables using econometric methods, the study first estimates the unit root test, which is a precondition for most co-integration techniques.The study employs the Augmented Dickey-Fuller (ADF) [14], Phillip Perron (PP) [15] and Vogelsang [16] breakpoint unit root tests.which satisfies the pre-condition of Johansen's method of co-integration and the ARDL bounds test approach.

Model Estimation
After fulfilling the pre-condition of Johansen's method of co-integration and the ARDL bounds test approach, the study follows the work of Asumadu-Sarkodie and Owusu [10],Asumadu-Sarkodie and Owusu [11], to estimate the VECM for the study, which is expressed as: Where 2 denotes the dependent variable, , , , , , and are the explanatory variables in year , ∆ represents the difference operator, represents the error correction term resulting from the long-run co-integration relationship, 's, 's and 's are the parameters to be estimated, represents the number of lags and 's are the serially independent error terms.
Following the work of Asumadu-Sarkodie and Owusu [10],Asumadu-Sarkodie and Owusu [12],Ozturk and Acaravci [17],Asumadu-Sarkodie and Owusu [18], the ARDL model for this study is expressed as: Where denotes the intercept, represents the lag order, represents the error term and ∆ represents the first difference operator.In order to estimate the long-run equilibrium relationship in the ARDL Bounds approach, the study employs the F-tests statistic based on the null hypothesis of no co-integration between LCO2, LP, LS, LSV, LVS, LWBP, LWC and

Vector Error Correction Model
Table 4 shows the VAR Lag order selection criteria used in the study.Evidence from Table 4 shows that the optimal lag selected by the selection criteria (LR, FPE, AIC, SC and HQ) for the Johansen test of co-integration is indicated by "*".Using the optimal lag selected, the Johansen co-integration test [20] employs the maxeigenvalue and trace methods for the unrestricted co-integration rank tests presented in Table 5.Evidence from Table 5 shows that the the null hypothesis of no co-integration is rejected at 5% significance level, indicating 3 co-integrating equations.The 3 co-integrating equations are used to estimate the VECM as showed in Table 6.Evidence from Table 6 shows that the error correction term _ce1 L1. = −0.73 is negative and significant at 5% level, showing evidence of a long-run equilibrium relationship running from LP, LS, LSV, LVS, LWBP, LWC and LWF to LCO2.Results in Table 6 indicates that LP _ce1 L1. = 0.52, ρ = 0.11 , LS _ce1 L1. = 0.08, ρ = 0.74 , LSV _ce1 L1. = −0.29,ρ = 0.28 , LWBP _ce1 L1. = 0.01, ρ = 0.97 and LWF _ce1 L1. = 0.00, ρ = 0.96 are not significant at 5% level.However, the relationship between LVS _ce1 L1. = 1.47, ρ = 0.04 and LCO2 is positive and significant at 5% level.The implication is that, a 1% increase in veneer sheet production reduces carbon dioxide emissions by 1.47% in the long-run.Veneer sheet production tend to affect carbon dioxide emissions positively since its one of the most environmentally efficient and economically feasible ways of producing timber.
The relationship between LWC _ce1 L1. = 0.18, ρ = 0.00 and LCO2 is positive and significant at 5% level.The implication is that, a 1% increase in wood charcoal production reduces carbon dioxide emissions by 0.18% in the long-run.

VECM Diagnostic Test
In order to have a robust result, the study examines the independence of the residuals in the VECM.Table 7 presents the diagnostic test for the VECM.Evidence from the Jarque-Bera test in Table 7 shows that the null hypothesis of normal distribution cannot be rejected at 5% significance level.The null hypothesis of no serial correlation at lag order h by the Lagrangemultiplier test cannot be rejected at 5% significance level.The null hypothesis of no ARCH effect by the Heteroskedasticity Test cannot be rejected at 5% significance level.Evidence from Table 7 shows that the residuals in the VECM are normally distributed, have no problems with serial correlation and have a constant variance.
Fig. 2 presents the inverse root of characteristics polynomial to check the stability of the VECM.Evidence from Fig. 2 shows that the eigenvalues of the respective matrix is less than 1 with no unit root outside the unit circle, confirming the VECM satisfying the VAR stability conditions.

Granger-Causality
The VECM reveals the existence of a long-run equilibrium relationship between variables but fails to indicate the direction of the causal relationship.As such, the study employs the Grangercausality test [21] to ascertain the causal relationships between LCO2, LP, LS, LSV, LVS, LWBP, LWC and LWF.
Table 10 presents an estimation of the ARDL regression.Table 10 shows that the error correction term ECT −1 = −0.72 is negative and significant at 5% level, showing evidence of a long-run equilibrium relationship running from LP, LS, LSV, LVS, LWBP, LWC and LWF to LCO2.
Evidence from the short-run estimation in Table 10 shows that; a 1% increase in LP will increase LCO2 by 0.19% in the short-run, a 1% increase in LS will increase LCO2 by 0.26% in the short-run, a 1% increase in LWC will increase LCO2 by 1.14% in the short-run and a 1% increase in LWF will increase LCO2 by 0.63% in the short-run.
Evidence from the long-run estimation in Table 10 shows that; a 1% increase in LP will increase LCO2 by 0.17% in the long-run, a 1% increase in LS will increase LCO2 by 0.17% in the long-run, a 1% increase in LWC will increase LCO2 by 0.36% in the long-run and a 1% increase in LWF will increase LCO2 by 0.37% in the long-run.Evidence from the long-run equilibrium relationship in the ARDL model shows that; a 1% increase in plywood production will increase carbon dioxide emissions by 0.17% in the longrun, a 1% increase in sawnwood production will increase carbon dioxide emissions by 0.17% in the long-run, a 1% increase in wood charcoal production will increase carbon dioxide emissions by 0.36% in the long-run and a 1% increase in wood fuel production will increase carbon dioxide emissions by 0.37% in the long-run.renewable-source fuels and highly efficient kilns can be utilized to make the production more environmentally friendly.Evidence from Fig. 7 shows that a one standard deviation shock to carbon dioxide emissions increases plywood, sawnwood, sawlogs & veneer logs, veneer sheets, wood-based panels, wood charcoal and wood fuel production carbon dioxide emissions to 2-period horizon and decreases over the given period which has a policy implication for Ghana.Evidence from the study shows that if carbon dioxide emission levels in Ghana are not mitigated, Ghana's forest reserve will decline with time which will lead to a lot of environmental hazards such as heat waves, change in weather patterns, poor air quality leading to health hazards and a destroyed ecosystem.

ARDL Diagnostic Test
As a policy recommendation, the Government of Ghana should institute public timber procurement policies that ensures re-planting of trees to replaced extinct ones.There is the need for the creation of new job opportunities that displaces the illegal chainsaw operation in Ghana.
There is the need for an establishment of code of conduct for the lumber industry and enhanced capacity of the Forest Commission of Ghana to monitor the forest regularly.It is essential for the Government of Ghana to incorporate climate change mitigation options into the forest policies, by creating awareness, providing early warnings, climate change adaptations and institutional capacity building.

Fig. 6 .
Fig. 6.Response of LCO2 to Cholesky One S.D. Innovations in other Variables

Fig. 7 .
Fig. 7. Response of other Variables to Cholesky One S.D. Innovations in LCO2

Table 1 .
Descriptive Analysis *. Correlation is significant at the 0.05 level (2-tailed).c.Unless otherwise noted, bootstrap results are based on 1000 bootstrap samples

Table 3 .
Unit Root Test Table 3 presents the results of ADF, PP and Vogelsang's breakpoint unit root tests.Significantly, Vogelsang's breakpoint unit root test provides a robust results since it takes into consideration the innovational outlier which ADF and PP tests fail to test in the presence of structural breaks.Evidence from Table3shows that the null hypothesis of unit root at level cannot be rejected at 5% significance level.Evidence from Table3shows that based on 5% significance level, the null hypothesis is rejected at first difference.The ADF, (1)and Vogelsang's breakpoint unit root tests suggest that the variables are integrated at I(1)

Table 4 .
Lag Selection Criteria

Table 6 .
Vector Error Correction Model

Table 8 .
Table 8 presents the results of the Granger causality tests based on VECM.The null hypothesis that LCO2 does not Granger cause LVS, LCO2 does not Granger cause LWC, LCO2 does not Granger cause LWF, LP does not Granger cause LCO2, LVS does not Granger cause LCO2, and LWC does not Granger cause LEU is rejected at the significance level indicated with '*' in Table 8.Accordingly, there is a bidirectional causality between LCO2 and LVS, LCO2 and LWC, and a unidirectional causality running from LCO2 to LWF and LP to LCO2.Evidence from the joint Granger-causality shows a unidirectional causality from LCO2 to all the variables (LP, LS, LSV, LVS, LWBP, LWC and LWF).Granger causality Wald tests

Table 11
presents a diagnostic tests of the ARDL regression analysis.Evidence from Table11shows that the null hypothesis of no heteroskedasticity by the Breusch-Pagan-Godfrey test cannot be rejected at 5% significance level.The null hypothesis of no serial correlation by the Breusch-Godfrey LM test cannot be rejected at 5% significance level.The null hypothesis of normal distribution by the Jarque-Bera test cannot be rejected at 5% significance level and the null hypothesis of no omitted variables in the model by the Ramsey RESET Test cannot be rejected at 5% significance level.In summary, the residuals in the ARDL model have no problems with heteroskedasticity, have no problems with serial correlation, are normally distributed and have a constant variance.