Income and Emission: A Panel Data based Cointegration Analysis

Similar documents
Vectorautoregressive Model and Cointegration Analysis. Time Series Analysis Dr. Sevtap Kestel 1

Unit Root Time Series. Univariate random walk

Time Series Test of Nonlinear Convergence and Transitional Dynamics. Terence Tai-Leung Chong

A Specification Test for Linear Dynamic Stochastic General Equilibrium Models

Methodology. -ratios are biased and that the appropriate critical values have to be increased by an amount. that depends on the sample size.

R t. C t P t. + u t. C t = αp t + βr t + v t. + β + w t

Lecture 5. Time series: ECM. Bernardina Algieri Department Economics, Statistics and Finance

Ready for euro? Empirical study of the actual monetary policy independence in Poland VECM modelling

Licenciatura de ADE y Licenciatura conjunta Derecho y ADE. Hoja de ejercicios 2 PARTE A

How to Deal with Structural Breaks in Practical Cointegration Analysis

Dynamic Econometric Models: Y t = + 0 X t + 1 X t X t k X t-k + e t. A. Autoregressive Model:

Introduction D P. r = constant discount rate, g = Gordon Model (1962): constant dividend growth rate.

Econ107 Applied Econometrics Topic 7: Multicollinearity (Studenmund, Chapter 8)

DEPARTMENT OF STATISTICS

Diebold, Chapter 7. Francis X. Diebold, Elements of Forecasting, 4th Edition (Mason, Ohio: Cengage Learning, 2006). Chapter 7. Characterizing Cycles

The Brock-Mirman Stochastic Growth Model

Department of Economics East Carolina University Greenville, NC Phone: Fax:

Chapter 16. Regression with Time Series Data

Robust estimation based on the first- and third-moment restrictions of the power transformation model

Granger Causality Among Pre-Crisis East Asian Exchange Rates. (Running Title: Granger Causality Among Pre-Crisis East Asian Exchange Rates)

ACE 564 Spring Lecture 7. Extensions of The Multiple Regression Model: Dummy Independent Variables. by Professor Scott H.

Bias in Conditional and Unconditional Fixed Effects Logit Estimation: a Correction * Tom Coupé

ECON 482 / WH Hong Time Series Data Analysis 1. The Nature of Time Series Data. Example of time series data (inflation and unemployment rates)

A New Unit Root Test against Asymmetric ESTAR Nonlinearity with Smooth Breaks

ACE 562 Fall Lecture 4: Simple Linear Regression Model: Specification and Estimation. by Professor Scott H. Irwin

GDP PER CAPITA IN EUROPE: TIME TRENDS AND PERSISTENCE

Solutions to Odd Number Exercises in Chapter 6

Problem Set 5. Graduate Macro II, Spring 2017 The University of Notre Dame Professor Sims

Macroeconomic Theory Ph.D. Qualifying Examination Fall 2005 ANSWER EACH PART IN A SEPARATE BLUE BOOK. PART ONE: ANSWER IN BOOK 1 WEIGHT 1/3

Exercise: Building an Error Correction Model of Private Consumption. Part II Testing for Cointegration 1

Cointegration and Implications for Forecasting

14 Autoregressive Moving Average Models

Why is Chinese Provincial Output Diverging? Joakim Westerlund, University of Gothenburg David Edgerton, Lund University Sonja Opper, Lund University

A multivariate labour market model in the Czech Republic 1. Jana Hanclová Faculty of Economics, VŠB-Technical University Ostrava

Comparing Means: t-tests for One Sample & Two Related Samples

Testing for a Single Factor Model in the Multivariate State Space Framework

Physics 235 Chapter 2. Chapter 2 Newtonian Mechanics Single Particle

Do Steel Consumption and Production Cause Economic Growth?: A Case Study of Six Southeast Asian Countries

Forecasting optimally

Kriging Models Predicting Atrazine Concentrations in Surface Water Draining Agricultural Watersheds

International Parity Relations between Poland and Germany: A Cointegrated VAR Approach

ACE 562 Fall Lecture 8: The Simple Linear Regression Model: R 2, Reporting the Results and Prediction. by Professor Scott H.

1. Diagnostic (Misspeci cation) Tests: Testing the Assumptions

ESTIMATION OF DYNAMIC PANEL DATA MODELS WHEN REGRESSION COEFFICIENTS AND INDIVIDUAL EFFECTS ARE TIME-VARYING

On Measuring Pro-Poor Growth. 1. On Various Ways of Measuring Pro-Poor Growth: A Short Review of the Literature

( ) (, ) F K L = F, Y K N N N N. 8. Economic growth 8.1. Production function: Capital as production factor

A Dynamic Model of Economic Fluctuations

Econ Autocorrelation. Sanjaya DeSilva

OBJECTIVES OF TIME SERIES ANALYSIS

LONG MEMORY AT THE LONG-RUN AND THE SEASONAL MONTHLY FREQUENCIES IN THE US MONEY STOCK. Guglielmo Maria Caporale. Brunel University, London

The general Solow model

Remittances and Economic Growth: Empirical Evidence from Bangladesh

A unit root test based on smooth transitions and nonlinear adjustment

Properties of Autocorrelated Processes Economics 30331

3.1.3 INTRODUCTION TO DYNAMIC OPTIMIZATION: DISCRETE TIME PROBLEMS. A. The Hamiltonian and First-Order Conditions in a Finite Time Horizon

Chickens vs. Eggs: Replicating Thurman and Fisher (1988) by Arianto A. Patunru Department of Economics, University of Indonesia 2004

Time series Decomposition method

Regression with Time Series Data

STRUCTURAL CHANGE IN TIME SERIES OF THE EXCHANGE RATES BETWEEN YEN-DOLLAR AND YEN-EURO IN

The Validity of the Tourism-Led Growth Hypothesis for Thailand

Outline. lse-logo. Outline. Outline. 1 Wald Test. 2 The Likelihood Ratio Test. 3 Lagrange Multiplier Tests

Vehicle Arrival Models : Headway

Explaining Total Factor Productivity. Ulrich Kohli University of Geneva December 2015

Choice of Spectral Density Estimator in Ng-Perron Test: A Comparative Analysis

E β t log (C t ) + M t M t 1. = Y t + B t 1 P t. B t 0 (3) v t = P tc t M t Question 1. Find the FOC s for an optimum in the agent s problem.

(a) Set up the least squares estimation procedure for this problem, which will consist in minimizing the sum of squared residuals. 2 t.

Box-Jenkins Modelling of Nigerian Stock Prices Data

DEPARTMENT OF ECONOMICS AND FINANCE COLLEGE OF BUSINESS AND ECONOMICS UNIVERSITY OF CANTERBURY CHRISTCHURCH, NEW ZEALAND

SUPPLEMENTARY INFORMATION

Wednesday, November 7 Handout: Heteroskedasticity

23.2. Representing Periodic Functions by Fourier Series. Introduction. Prerequisites. Learning Outcomes

Hypothesis Testing in the Classical Normal Linear Regression Model. 1. Components of Hypothesis Tests

Mean Reversion of Balance of Payments GEvidence from Sequential Trend Break Unit Root Tests. Abstract

A note on spurious regressions between stationary series

Lecture Notes 2. The Hilbert Space Approach to Time Series

UNIVERSITY OF OSLO DEPARTMENT OF ECONOMICS

The Brock-Mirman Stochastic Growth Model

Two Coupled Oscillators / Normal Modes

Final Exam Advanced Macroeconomics I

Robust critical values for unit root tests for series with conditional heteroscedasticity errors: An application of the simple NoVaS transformation

Has the Business Cycle Changed? Evidence and Explanations. Appendix

20. Applications of the Genetic-Drift Model

10. State Space Methods

GMM - Generalized Method of Moments

= ( ) ) or a system of differential equations with continuous parametrization (T = R

Essential Microeconomics : OPTIMAL CONTROL 1. Consider the following class of optimization problems

Stationary Time Series

Exponential Weighted Moving Average (EWMA) Chart Under The Assumption of Moderateness And Its 3 Control Limits

Testing for linear cointegration against nonlinear cointegration: Theory and application to Purchasing power parity

This document was generated at 7:34 PM, 07/27/09 Copyright 2009 Richard T. Woodward

DEPARTMENT OF ECONOMICS

Temporal Causality between Human Capital and Real Income in Cointegrated VAR Processes: Empirical Evidence from China,

The Properties of Procedures Dealing with Uncertainty about Intercept and Deterministic Trend in Unit Root Testing

Appendix 14.1 The optimal control problem and its solution using

Final Spring 2007

Nonstationary Time Series Data and Cointegration

Financial Econometrics Jeffrey R. Russell Midterm Winter 2009 SOLUTIONS

4.1 Other Interpretations of Ridge Regression

ACE 562 Fall Lecture 5: The Simple Linear Regression Model: Sampling Properties of the Least Squares Estimators. by Professor Scott H.

d 1 = c 1 b 2 - b 1 c 2 d 2 = c 1 b 3 - b 1 c 3

Transcription:

Income and Emission: A Panel Daa based Coinegraion Analysis SOUMYANANDA DINDA S. R. Faepuria College, Beldanga, Murshidabad, W.B., India. Economic Research Uni, Indian Saisical Insiue, Kolkaa-08. India. DIPANKOR COONDOO Economic Research Uni, Indian Saisical Insiue, Kolkaa-08. India. November, 2002; March, 2003. Absrac This paper presens he resuls of an invesigaion of he causaliy issue of incomeemission relaionship based on ime series economeric echniques of uni roo es, coinegraion and relaed error correcion model for a panel daa se. Here, he naure of causaliy beween per capia CO 2 emission (PCCO2) and per capia GDP (PCGDP) has been examined using a cross counry panel daa se covering 88 counries for he period 960-90. Using he panel uni roo es procedure of Im e al. (997) (IPS), we have found ha he hypohesis of uni roo (i.e., non-saionariy) of he ime series of PCGDP and PCCO2 can no be rejeced for individual counry groups. As boh he variables are found o follow I() process, we nex have performed he panel daa co-inegraion es and finally, we have esimaed he ECM (for hese counry groups for which significan income-emission coinegraion was obained) o explore he naure of dynamics implici in he given panel daa se. Our findings sugges ha here is more or less a bi-direcional causal relaionship beween income (PCGDP) and CO 2 emission (PCCO2) for Africa, Cenral America, America as a whole, Easern Europe, Wesern Europe, Europe as a whole and he World as a whole. Tha means, he movemen of he one variable direcly affecs he oher variable hrough a feedback sysem. Thus, he policy makers should be cauious o make proper decision abou he conrol of emission level. JEL Classificaion: C33, O40, and Q25. Keywords: Panel daa, Uni Roo, IPS, CO 2 emission, GDP, co-inegraion, causaliy, ECM. ----------------------------------- Corresponding Auhor: Soumyananda Dinda, c/o, Dipankor Coondoo, Economic Research Uni, Indian Saisical Insiue, 203, B. T. Road, Kolkaa-08, India. E-mail: s_dinda@homail.com and sdinda2000@yahoo.co.in, dcoondoo@isical.ac.in (D. Coondoo), Fax No: 9-033-577-8893.

. Inroducion Coondoo and Dinda (2002) examined he naure of causaliy beween CO 2 emission and income using a cross-counry panel daa se covering 88 counries and he ime period 960-90. Briefly, in ha sudy he presumpion of he Environmenal Kuznes Curve (EKC) hypohesis viz., ha an income o polluion (CO 2 emission, more specifically) causal relaionship holds universally was examined. However, he resuls based on he Granger causaliy es (GCT) did no lend much empirical suppor o ha presumpion. Insead, for individual counry groups well-defined and disincive paerns of causaliy were observed. For example, for he developed counry groups of Norh America and Wesern Europe (and for ha maer, Eas Europe also), causaliy was found o run from emission o income. For Japan, he developing counry groups of Cenral and Souh America and Oceania, on he oher hand, causaliy in he opposie direcion was observed. Finally, for counry groups of boh Asia and Africa causaliy urned ou o be bi-direcional. Inerpreaion of hese observed causaliy paerns was given in erms of iner-emporal changes in he raes of growh of income and emission. This inerpreaion made i clear how shocks in he rae of growh of income or emission migh affec each oher depending on he prevailing naure of causaliy. The GCT has been used in many empirical sudies on EKC and relaed issues 2. This echnique alone, however, can deec presence and direcion of causaliy for a pair of variables only in a limied sense (viz., in respec of heir shor run emporal movemens). The noion of causaliy beween income growh and polluion ha underlies he EKC A closer examinaion of he counry-wise daa for Asia and Africa revealed ha while some counries had causaliy in one direcion, ohers had causaliy in he opposie direcion. Possibly his heerogeneiy in he paern of causaliy led o he observed bi-direcional causaliy a he level of counry-groups for hese wo coninens. 2 See, e.g., Yu and Choi (985), Cheng (996), Cheng and Lai (997) and Yang (2000). 2

hypohesis, on he oher hand, is essenially a longer run concep 3. Thus, furher probe ino he issue of causaliy using comprehensive economeric ools for exploring presence of any long run equilibrium relaionship among income and polluion, viz., he coinegraion analysis, may help verify conclusions abou causaliy ha we have reached so far 4. In his paper, we repor he resuls of an analysis of he relaionship beween per capia GDP (PCGDP) and per capia CO 2 emission (PCCO2) obained by using non-saionary panel daa echniques o a cross-counry panel daa se on hese variables. For convenience of exposiion, henceforh we shall call hese variables income and emission, respecively. To be precise, here we firs used he panel daa uni roo es procedure of Im, Pesaran and Shin (997) (henceforh referred o as IPS) o examine wheher he observed counry-specific ime series daa on income and emission possessed sochasic rend or no. Nex, on finding evidences of presence of such rend in he daa se, we performed he Engle-Granger bivariae coinegraion analysis 5 o examine wheher he pair of variables was coinegraed (i.e., wheher hey obeyed any long run equilibrium relaionship beween hemselves). Finally, we esimaed he Error Correcion Model (ECM) for hose counry groups for which income-emission coinegraion was obained o explore he naure of dynamics implici in he panel daa se for hose counry groups. 3 See, Coondoo and Dinda (2002) for a discussion on his issue. 4 There are ineresing applicaions of ime series economeric ools like vecor auoregression model (VAR) and coinegraion analysis on environmen-relaed daa. See, e.g., Sern (993, 2000) for sudies on causal relaionship beween GDP and energy use for he USA for he period 947-990 based on GCT in a VAR se up, single equaion saic co-inegraion analysis and mulivariae dynamic co-inegraion analysis. See also Cheng (999) for an applicaion of Johansen co-inegraion es o he daa on energy consumpion, economic growh, capial and labour for he Indian economy. 5 Johansen s mehod of coinegraion analysis, which is more comprehensive, could no be used, because we could no access sofware for applicaion of Johansen s mehod o panel daa se. 3

The paper is organized as follows: secion 2 explains he moivaion for using coinegraion analysis on he income-emission daa in he presen exercise; secion 3 describes he daa, presens and discusses he empirical resuls, secion 4 inerpres he resuls and secion 5 draws some concluding observaions. Finally, he mehodology of uni roo es, coinegraion analysis and ECM esimaion based on panel daa ha we have acually used in he presen exercise is briefly explained in he Appendix. 2. Moivaion To help jusify he use of coinegraion analysis on he se of cross-counry panel daa on income and emission for examining he naure of causaliy ha may exis beween his pair of variables, le us consider he following simple heoreical consruc. Consider a one-good economy for which environmen E, undersood as a sock variable, affecs boh uiliy and producion level of he represenaive agen. Le C(), E() and K() denoe consumpion, environmen and capial sock a ime. Leθ() (0<θ()<) porion of capial sock be used for commodiy producion a ime and he remaining (-θ()) porion be used for upgrading he environmen. Finally, le γ (>0) be he rae of polluion (i.e., emission or degradaion of environmen per uni of oupu produced). The infinie ime horizon iner-emporal consumpion choice problem for his economy may be specified as 0 ρ Maximize W e U ( C( ), E( )) d () subjec o he accumulaion consrains K & ( ) f ( θ ( ) K( ), E( )) C( ) (2) 4

and E & ( ) g(( θ ( )) K( ), E( )) γf ( θ ( ) K( ), E( )) (3) where ρ>0 is he rae of ime preference and f(.) and g(.) are he producion funcion and he environmen upgrading funcion of he economy. Clearly, he firs consrain relaes o physical capial accumulaion while he second relaes o ne environmenal change due o producion and environmenal upgrading. Treaing C() and θ () as conrol variables and K() and E() as sae variables, he opimaliy condiion for he above problem urns ou o be C & E& α + β φ C E (4) where CU U CC α, C EU U CE K K β and φ ( + ρ) C f g g + γf K K, U, being he second CC U CE order parial derivaives of U(.). Noe ha he above condiion suggess ha opimal ime pah of C and E should generally be inerdependen. This, hus, means a wo-way causal relaionship beween income and emission, in general. If, however, α(β ) urns ou o be idenically zero, he opimal ime pah of C (E) will be auonomous and he naure of he opimal ime pah of E (C ) will depend upon wha he opimal pah of he oher variable is. Le us nex search for a long run equilibrium relaionship beween income (C) and emission (E), underlying he above opimizaion problem. To do so, consider he seady sae soluion where &E μ& 0 i.e., he siuaion where he environmenal sock reaches a sable level. Now, E & 0 implies g(( θ ) K, E) γf ( θk, E) (5) 5

i.e., he rae of environmenal degrading due o producion mus equal he rae of environmenal upgrading. Clearly, eq.(5) defines a relaionship beween K and E say, h ( K, E) 0, (6) for given θ. Nex, le a he seady sae K & σ, a consan. This implies f ( θ K, E) C σ h ( K, E, C) 0, (7) 2 for given θ. Combining eq.s (6) and (7), we obain wha may be called a long run equilibrium relaionship beween C and E, say, h ( C, E) 0, or equivalenly, E h(c), (8) 3 which may be recognized as he long run relaionship beween income (C) and environmen (E). I should now be sraighforward o use he above heoreical consruc o raionalize coinegraion analysis of a bivariae ime series/panel daa se on income and emission, as we have done in he presen paper. Le { C E } denoe ime series of observed, consumpion and environmen variable, where C + ε and E + ε - C C E E C, E being corresponding (unobserved) opimal values and ε C, ε E being random disurbances. In case he observed daa se is consisen wih opimizaion, and should differ from he corresponding opimal values only by saionary random disurbances (i.e., C E εc and ε should be saionary random variables). Also, and, being consisen wih E C E opimizaion, should be coinegraed as hey mus obey eq. (8), bu for saionary deviaions. Granger causaliy beween C and E, which is essenially a shor run noion, is ofen examined wih he help of he ECM as a par of he coinegraion analysis. When ime 6

C E series and are non-saionary and are inegraed of order one (i.e., he corresponding ime series of firs differences are saionary) and he variables are coinegraed, hey admi he Granger represenaion 6 and he ECM can be expressed as ΔC m m βciδc i + γ CiΔE i ηc E h( C )) + i i ( ν C (9) or, equivalenly as ΔE m m β EiΔC i + γ EiΔE i ηe E h( C )) + i i ( ν E (0) where ν C and ν E are pure whie noise random disurbances and β Ci, β Ei, γ Ci, γ Ei, ηc and ηe are he parameers of he ECM. Noe ha ( E h( C )), which is called he error correcion erm, is a measure of he exen by which he observed values in ime - deviae from he long run equilibrium relaionship. Since he variables are coinegraed, any such deviaion a ime - should induce changes in he values of he variables in he nex ime poin in an aemp o force he variables back o he long run equilibrium relaionship. The coefficiens η C and η E of he error correcion erm in he wo equaions (which measure he rae of his adjusmen process) are herefore called he adjusmen parameers and are expeced o be posiive. The parameers γ Ci s in eq. (9) and β Ei s in eq. (0) deermine he naure of causaliy beween C and E. More specifically, if γ 0 Ci for a leas one i ( i, m) and β 0 for all i ( i, m), hen E is said o Granger cause Ei C. On he oher hand, if γ 0 for all i ( i, m) and β 0 for a leas one i ( i, m), Ci hen C is said o Granger cause E. In case γ 0 and β 0 for a leas one i ( i, m), Ci Ei Ei 6 See Hamilon (994) for deails. 7

he causaliy beween C and E is defined o be bi-direcional. Finally, when γ 0 and β 0 for all i ( i, m), Granger causaliy beween C and E is said o be absen 7. The Ei absence of Granger Causaliy for coinegraed variables requires he addiional condiion ha he speed of adjusmen coefficien be equal o zero. In his se up, saisical significance of he esimaed adjusmen parameers η C and η E should help qualify furher he naure of causaliy relaionship beween C and E. Thus, for example, if H : β 0 for all i ( i, m), η E 0 is no rejeced and a he same ime H : γ 0 for 0 Ei all i ( i, m), η C 0 is rejeced, one should inerpre such a resul as corresponding o a siuaion in which he ime pah of C is auonomously deermined and ha of E being caused by C. Oher possible resuls may be inerpreed in a similar manner ( see Glasure and Lee (997) and also Asafu-Adjaye (2000) for deails). 0 Ci Ci 3. Daa Descripion and Resuls As menioned a he ouse, for he presen exercise we have used cross-counry panel daa on PCGDP (measured in erms of PPP in 985 US dollar) compiled by Summers and Heson (viz., he RGDPCH series of Penn World Table (Mark 5.6)). Corresponding panel daa se on PCCO2 (measured in meric ons) was obained from he web sie of Carbon Dioxide Analysis Informaion Cener (CDAIC), Oak Ridge Naional Laboraory of he U. S. A. Combining hese wo daa ses, we compiled a bivariae panel daa se of annual observaions on income and emission covering 88 counries and he ime period from 960 o 990 (for a deailed daa descripion, see Coondoo and Dinda (2002)). For 7 For he specific null hypoheses ha are esed o deec he naure of causaliy in he ECM se up, see Secion A.3 of he Appendix. 8

he purpose of he exercise, we grouped he counries ino 2 counry groups. Table shows he composiion of hese counry groups. The empirical exercise has been done separaely for each of hese counry groups based on he bivariae panel daa ses for he individual counry groups 8. Table 2 presens he counry-group-specific resuls of uni roo es for logarihm of PCGDP and logarihm of PCCO2 (i.e., income and emission respecively, in our erminology) based on he IPS mehod. In each case he es was done wice viz., once assuming presence of a deerminisic ime rend in he daa generaing process and again wihou making such an assumpion. The resuls show ha a 5 per cen level of significance he null hypohesis of uni roo canno be rejeced in any of he cases, excep for income for Easern Europe when presence of a deerminisic ime rend in he daa generaing process is no assumed 9. One may hus conclude ha he counry groupspecific ime series of boh he variables under consideraion are by and large nonsaionary. A repeiion of he same es on he firs-differenced daa se showed rejecion of he null hypohesis of uni roo in all he cases. I hus indicaes ha he counryspecific ime series of boh income and emission were inegraed of order (i.e., hey were I(), symbolically). In he nex sep, we examined wheher or no for individual counry groups he null hypohesis ha income and emission were no coinegraed migh be rejeced. As 8 I may be poined ou here ha he saes/regions covered by he erswhile U S S R have been lef ou from his exercise as pas daa for hese saes/regions are no available. I should be noed ha counries falling ino he same group are more or less in a similar sae of economic developmen. 9 In his case he es urned ou o be marginally significan a he 5 per cen level in he wihou ime rend case and was non-significan in he wih ime rend case. Such a resul may be possible only if an increasing (decreasing) deerminisic ime rend ges canceled wih a decreasing (increasing) sochasic ime rend. 9

explained in he Appendix, he bivariae Engle-Granger mehodology of coinegraion 0 and he IPS uni roo es procedure was used for his examinaion. The resuls of his es are presened in Table 3. Following he Engle-Granger convenion, for each counry group we esed coinegraion wice, viz., once reaing income as he dependen variable and emission as he independen variable and again inerchanging he dependenindependen saus of hese wo variables. The enries under he column heading income (emission) are he compued IPS -saisic values for he coinegraion uni roo es when income (emission) was aken as he dependen variable. Here also in each case he coinegraion es was done wice viz., once assuming presence of a deerminisic ime rend in he residuals of he coinegraing regression equaion and again wihou making such an assumpion. In Table 3 counry group-specific values of hese four es saisics are presened. Table 3 may be summarized as follows: The resuls of coinegraion appear o be sensiive o wheher or no presence of deerminisic ime rend in he e i s (i.e., he regression residuals defined in relaion (A3) of he Appendix) is assumed. When s were assumed no o conain any deerminisic ime rend, in mos of he cases he resul of coinegraion was observed o depend upon wheher income or emission was aken as he dependen variable. Excepions were Cenral America, America as a whole and Easern Europe. In all hese cases he hypohesis of coinegraion was no rejeced irrespecive of wheher income or emission was used as he dependen variable. On he e i oher hand, when presence of deerminisic ime rend in e i s was assumed, he 0 In Engle and Granger's (987) original definiion, coinregaion refers o a linear relaionship beween non-saionary variables. Holz Eakin and selden (995) show he evidence suggesing a linear relaionship beween per capia income and CO 2 emission. We also observe he monoonic relaionship beween income and emission. Tha is, he uni roo es of he residuals of he esimaed long run relaionship beween y i and x i. 0

coinegraion resuls obained by reaing income as he dependen variable mosly agreed wih he corresponding resuls obained by reaing emission as he dependen variable 2. Thus in his case irrespecive of wheher emission or income was aken as he dependen variable, he null hypohesis of coinegraion was no rejeced (equivalenly, he null hypohesis of uni roo of e i s was rejeced) for Africa, Wesern Europe, Europe and he World. In oher words, for hese counry groups ime series of income and emission seemed o obey a long run equilibrium relaionship. For Norh America, Souh America, Asia, Asia excluding Japan and Oceania, on he oher hand, he null hypohesis of coinegraion was rejeced (i.e., he null hypohesis of uni roo of e i s was no rejeced). For he remaining counry groups (viz., Cenral America, America and Easern Europe) he null hypohesis of coinegraion was no rejeced when emission had been aken as he dependen variable, bu i was rejeced when income had been aken as he dependen variable. Nex, using he counry group-specific panel daa, we esimaed he alernaive versions of he ECM - viz., equaion (A5) and (A6) of he Appendix, which we referred o as models I and II, respecively. This esimaion was done only for hose counry groups for which he null hypohesis of coinegraion was no rejeced (viz., Africa, Cenral America, America as a whole, Easern Europe, Wesern Europe, Europe as a whole and he World). In each case he ECM was esimaed using hree differen economeric specificaions of he panel daa regression equaion viz., ordinary leas squares (OLS), 2 I is well known ha in case of he Engle-Granger mehodology he resul of he coinegraion es may be sensiive o he choice of he dependen variable of he coinegraion regression in case of no large enough samples. The power of he uni roo es, on he oher hand, may also depend on wheher or no a deerminisic rend is presen in he daa generaing process and has been incorporaed in he regression model used o es uni roo. Someimes i is suggesed ha when he regression model esimaed for esing uni roo conains a deerminisic rend componen and he es rejecs he null hypohesis of presence of a uni roo, ha is a sufficien indicaion of absence of an uni roo (see, Enders (995) pp. 254-258).

fixed effecs (FE) model and random effecs (RE) model 3. In our exercise he FE model urned ou o be he appropriae choice for almos all he counry groups. The counry group-specific esimaes of he regression coefficiens of he wo versions of he fixed effecs ECM (viz., models I and II) are presened in Table 4. I may be noed ha he esimaed adjusmen parameers (i.e., he coefficien of he EC erm) in Table 4 are all saisically significan wih he expeced negaive sign (in all cases excep for Wesern Europe when emission is aken as he dependen variable). Since in all hese cases income and emission are coinegraed, such a resul is only o be expeced. This is because of he following reason: as he pair of variables is coinegraed, over a long period of ime hey end o move in unison. This means ha if moves over ime always rying o be on he long run equilibrium relaionship. As is well known, he ECM ries o explain he observed shor run variaions of he dependable variable in erms of variaions of he lagged value of he dependen variable and he oher explanaory variable of he model. Following he explanaion given in Secion 2 and he Appendix, he naure of Granger causaliy beween he variables under sudy underlying he given daa se may be examined by esing null hypoheses specifying relevan parameric resricions on he esimaed ECM (See Table 6a). 4. Inerpreaion of Resuls In Table 4 he counry group/coninen-specific FE esimaes of he pair of ECM equaions (i.e., equaions (A5) and (A6) of Appendix) based on panel daa have been 3 OLS is known o be generally inefficien for panel daa regression esimaion. Choice beween FE and RE depends upon wheher or no he null hypohesis H : α i α for 0 i,2,..., N, is rejeced, where α i denoes he inercep for he ih uni. FE is chosen if H 0 is no rejeced. 2

reproduced. We shall now aemp o explain he resuls of Table 4 from he poin of view of causaliy 4 due o shor run flucuaions along wih long run equilibrium relaionship. As is well known, he Error Correcion Model (ECM) depics he shor-run dynamics of he variables of a sysem when heir variables deviae from equilibrium relaion(s) governing heir long run movemens. The dependen variables of equaion (A5) and (A6) of Appendix are and measuring growh rae 5 of income and emission, respecively. So, in general, we may wrie T T2 r i i j j y i j equaion (A5) as α r + β r + η EC + u and equaion (A6) T2 T22 r 2i i 2 j j x 2 i j as α r + β r + η EC + u, where EC is error correcion erm, and u 2 are whie noise error erms wih zero expecaions. As we have already, seen, he esimaed coefficien of he EC erm in Table 4 are all saisically significan wih an expeced negaive sign (in all cases excep for Wesern Europe, in which significan (viz., a 0%) level is low, when emission is aken as he dependen variable). Now, for a specific counry group hese equaions ake specific form depending on he saisical significance of he individual parameers of he above pair of equaions. We discuss hese cases below and also examine heir implicaions for shor run movemen from he poin of view of causaliy. Consider firs he case of Africa for which no all he esimaed parameers are r r u significan. Thus, we have r r + α 2r 2 η y EC + u α, α, α 2, 0 and η y > 4 I should be noed ha in our earlier sudy, (See Coondoo and Dinda 2002) in which, we find he causal relaionship beween income and emission using Granger Causaliy Technique which remain same in his sudy in shor run bu differ in long run. 5 and. Δ Y Δlog( PCGDP) r ΔX Δlog( PCCO2) r 3

r r β3r 3 ηxec + u2 η x > r r β ; β, β3, 0. Thus, and follow auoregressive processes and are auonomous in shor run, alhough a saisically significan long run relaionship exiss beween hem. For Cenral America and America as a whole, we have r r ηyec + u α and r 2 2 x 2 α r β r β r η EC + u ; α, β, β2, 0. Here,, following a firs η x > r order auo-regressive process, is clearly auonomous. On he oher hand, r significanly depends upon boh r and is own pas values. Thus, we have a case of income o emission causaliy in he shor run. Nex, le us consider he cases of Wesern Europe. We have α α, β, β 3, 0 and r r α 2r 2 + βr β 2r 2 η y EC + u, α 2 η y > r η xec + u2 (coefficien of EC erm is significan a 0% level). These resuls sugges ha he rae of growh of emission has reached a sage of saionariy mainaining a long run equilibrium relaionship wih he rae of growh of income, bu in shor run r significanly depends on boh is own pas value and r. This implies ha any shock in r will cause a corresponding shock in r. Hence, we have a very specific kind of emission o income reverse causaliy for Wesern Europe. Finally, we have α α α, β, 0 for Easern r r + α 2r 3 β 2r 2 η y EC + u, 2 2 η y > Europe and α r α r + β r β r η EC u ; α α, β, β, 0 for r 2 2 2 2 y +, 2 2 η y > Europe as a whole; and r η xec + u2 for boh. Thus, here he growh rae of emission is saionary wih a long run equilibrium relaionship. Growh rae of income, being dependen on he growh rae of emission, is also saionary bu any shock in 4

emission growh rae r would cause a flucuaion in he income growh rae. Hence, in his case also here is reverse causaliy from emission o income. However, in hese cases he emission o income causaliy is supplemened by an addiional auoregressive effec of income growh. This means ha a sudden drop in he emission rae will cause no only a corresponding immediae negaive shock in he income growh rae, he effec will linger due o he significan auoregressive elemen ha governs he income growh rae. Now, le us see he long run income-emission relaionship (as given by he esimaed coinegraing vecor, viz., (, -b 0, -b )) and also he speed of adjusmen (η ) for differen counry groups. As is well known, he coinegraing vecors of differen groups give long run relaionship beween income and emission for individual counry groups. The coinegraing vecors 6 for Africa, Cenral America, America as a whole, Easern Europe, Wesern Europe, Europe as a whole and he World as a whole are presened in Table5. The parameers η y and η x in Table 6b are inerpreed as he speed of adjusmen coefficiens which measure he speed a which he values of y and x come back o long run equilibrium levels, once hey deviae from he long run equilibrium relaionship. These parameers are of paricular ineres in ha hey have imporan implicaions for he dynamics of he sysem. As indicaed above, he adjusmen coefficiens (i.e., he coefficien associaed wih he EC erm) show ha if any deviaion from he long run equilibrium occurs in one period, how much error is correced by ha variable in he nex period. The negaive sign of he esimaed speed of adjusmen coefficiens are in accord 6 A pair of co-inegraing vecors has been repored in Table 5 for individual counry group by changing he saus of dependen and independen variables. Sandard normalizaion process slighly differs in hese cases because of he presence of counry effecs or some oher flucuaions, alhough boh he variable are coinegraed for individual counry groups. 5

wih convergence oward long run equilibrium. The larger he value of η, sronger is he response of he variable o he previous period s deviaion from long run equilibrium, if any. Here we have found ha η is large for Africa (26.3%) and Cenral America (8.6%) and is small for Wesern Europe (2.8%). This implies ha in he case of Wesern Europe any deviaion from long run equilibrium of he value of y and x requires much longer ime o resore equilibrium. Since all he η s are saisically significan for all counry groups in boh he models, any change in one variable is expeced o affec he oher variable hrough a feedback sysem. This implies more or less a bi-direcional causal relaionship beween income and emission for all he counry groups. I should be noed ha if we ignore he EC erm, he resuls of Granger causaliy in our earlier sudy (See, Coondoo and Dinda 2002) remain same in his case also. Considering he EC erm, which is saisically significan and inerpreed as a source of causaliy in he long run sense, he ECM resuls differ from ha of our earlier resuls. In ECM, we find boh long run relaions wih shor run flucuaions. So, he resuls of ECM are qualiaively differen from ha of Granger causaliy. For a comprehensive sudy, we should address he issue of cross secional dependence. For example CO 2 mus be easily ransmied from one counry o he oher hrough rade. We assume ha he openness of an economy can provide he evidence of cross secional dependence. Degree of openness of an economy may also influence he naure of incomeemission causaliy. To be specific, a highly open economy, because of is easy access o fuel hrough inernaional rade, may no face he fuel supply consrain and hence coninue o have he income o emission causaliy problem. 6

The openness measure is defined as a raio of (expor+impor) o GDP a curren inernaional prices. The measure of openness is given in he Penn World Table for individual counry for each year. Using his daa we examine he income-emission relaion for all he counry -groups. Our empirical findings sugges ha openness 7 reduce CO 2 emission in Wesern Europe and Europe as a whole, where as i increases emission in Africa, Cenral America (See, Heige e al. 992). So, here is a clear evidence ha developed counries impor he polluion -inensive producs which are expored by developing or under developed counries (See also, Agras and Chapman 999). 5. Conclusion The basic objecive of his sudy was o examine he naure of causaliy beween income and CO 2 emission using a cross-counry panel daa se. This paper presens he resuls of invesigaion of he causaliy issue based on ime series economeric echniques of uni roo es, co-inegraion and relaed error correcion model esimaion. Using counrygroup specific panel daa on income and emission, we have found ha for seven counry groups (viz., Africa, Cenral America, America as a whole, Easern Europe, Wesern Europe, Europe as a whole and World as a whole) income and emission are coinegraed. Thus, for hese counry groups over a long period of ime income and emission end o move in unison. Examinaion of causaliy based on esimaed Engle-Granger error correcion model gives paern of causaliy which are some ime quie differen from hose given by he sandard Granger Causaliy Tes. Here we find ha bi-direcional causaliy beween income and emissions exis for more or less all he counry groups. 7 Heige e al. (992) find ha oxic inensiy decreases wih openness of he economy, bu he growh rae of he oxic inensiy of manufacuring increased in he poores counries. 7

Thus, any change in one variable is expeced o affec he oher variable hrough a feedback sysem. Le us enumerae he limiaions of he presen sudy. A comprehensive analysis of income-emission relaionship would necessarily call for an examinaion of he effecs of such deerminans as he ype of fuel used, he secoral composiion of income/gdp, available echnology and he price of fuel, among oher hings. We hope o underake a follow up sudy looking ino his aspec of he problem. Furher, any meaningful policy discussion for conrol of global emission should require a careful examinaion of he cross-counry disribuional paerns of global income and corresponding aggregae emission and heir changes over ime, keeping in mind he naure of causaliy ha is operaive in individual cases. Such a sudy should be nex on our research agenda. 6. References. Agras, J., and Chapman, D., 999, a dynamic approach o he Environmenal Kuznes Curve hypohesis, Ecological Economics, vol.-28, 267-277. 2. Asafu-Adjaye, J., 2000, The relaionship beween energy consumpion, energy price and economic growh: ime series evidence from asian developing counries., Energy Economics, vol.-22, 65-625. 3. Balagi, B.H. and J. M. Griffin, J.M., 995, A Dynamic Demand Model for Liquor : The case for pooling, Review of Economics and Saisics, vol.- 77, 545 553. 4. Balagi, B.H. and J. M. Griffin, J.M., 997, Pooled esimaors Vs. heir heerogeneous counerpars in he conex of dynamic demand for gasoline, Journal of Economerics, vol.-77, 303-327. 8

5. Balagi, B.H., 999, Economeric Analysis of Panel daa, John Wiley & Sons. 6. Balagi, B.H., Griffin, J. M. and Xiong, W. 2000, To Pool or no o Pool: Homogeneous versus Heerogeneous esimaors applied o cigaree demand, Review of Economics and Saisics, vol.-82(), 7 26. 7. Cheng, B. S., 996, An invesigaion of coinegraion and causaliy beween energy consumpion and economic growh, The Journal of Energy and developmen, vol.- 2(), 73-84. 8. Cheng, B. S. and Lai, T. W., 997, An invesigaion of coinegraion and causaliy beween energy consumpion and economic aciviy in Taiwan, Energy Economics, vol.- 9, 435-444. 9. Cheng, B. S., 999, Causaliy beween energy consumpion and economic growh in India: an applicaion of coinegraion and error correcion modeling., Indian Economic Review, vol.-34, 39-49. 0. Coondoo, D. and Dinda, S., 2002, Causaliy beween income and emission: a counry group-specific economeric analysis, Ecological Economics, vol.-40(3), 35-367.. Emerson, J. and Kao, C., 2000, Tesing for srucural change of a ime rend regression in panel daa, mimeo, Syracuse Universiy. 2. Enders, W.,995, Applied Economeric Time Series, John Wiley & Sons, Inc., U.S. A. 3. Engle, R.F. and Granger, C.W.J., 987, Coinegraion and error correcion represenaion, esimaion, and esing., Economerica, vol.-55(2), 25-276. 4. Funk, M. and Srauss, J., 2000, The long-run relaionship beween produciviy and capial, Economics Leers, vol.-69, 23 27. 9

5. Glasure, Yong U. and Lee, Aie-Rie, 997, Coinegraion, error correcion, and he relaionship beween GDP and Energy: he case of Souh Korea and Singapore., Resource and Energy Economics, vol.-20, 7-25. 6. Hamilon, J.D., 994, Time Series Analysis, Princeon Universiy press, New Jersey, USA. 7. Heihe, H., Lucas, R. E. B., and Wheeler, D., 992, The Toxic Inensiy of Indusrial Producion: Global paerns, Trends and Trade Policy, American Economic Review, vol.- 82, 478-48. 8. Holz Eakin, D. and Selden, T. M., 995, Soking he Fires? CO 2 Emissions and Economic Growh, Journal of Public Economics, vol.-57, 85-0. 9. Hsiao, C., Pesaran M. H. and Tahmiscioglu, A. K., 997, Bayes Esimaion of shorrun coefficiens in dynamic panel daa models, mimeo, Triniy College, Cambridge. 20. Hsiao, C., M., 986, Analysis of Panel daa, Cambridge Universiy Press. 2. Im, K. S., Pesaran M. H. and Shin, Y., 997, Tesing for Uni Roos in Heerogeneous Panels, mimeo, Dep. of Applied Economics, Universiy of Cambridge. 22. Kao, C., 999, Spurious regression and residual-based ess for coinegraion in panel daa, Journal of Economerics, vol.-90, 44. 23. Kao, C. and Chiang, M., 998, On he Esimaion and Inference of a Coinegraed Regression in Panel Daa, mimeo, Syracuse Universiy. 24. Kao, C., Chiang M. and Chen, B., 999, Inernaional R&D spillovers: An applicaion of esimaion and inference in panel co-inegraion, Oxford Bullein of Economics and Saisics, vol.-6(4), 693 7. 20

25. Levin, A. and Lin, C. F., 993, Uni Roo Tess in Panel Daa: New Resuls, Working paper, Universiy of California, San Diego. 26. McCoskey, S. and Kao, C., 999, Tesing he sabiliy of a producion funcion wih urbanizaion as a shif facor, Oxford Bullein of Economics and Saisics, vol.-6(4), 67 69. 27. McCoskey, S. and Kao, C. 998, A residual-based es of he null of coinegraion in panel daa, Economeric Reviews, vol.-7, 57-84. 28. Oak Ridge Naional Laboraory, CDIAC, 998, Esimaes of global, regional and naional CO 2 emissions from fossil fuel burning, cemen manufacuring and gas flaring: 755-996, available a hp://www.cdiac.esd.orgl.gov/epubs/ndp030/global97.ems 29. Quah, D., 994, Exploiing cross-secion variaion for uni roo inference in dynamic daa, Economics Leers, vol.-44, 9 9. 30. Sern, D. I., 993, Energy and economic growh in he USA, a mulivariae approach., Energy Economics, vol.-5, 37-50. 3. Sern, D. I., 2000, A mulivariae coinegraion analysis of he role of energy in he US macroeconomy, Energy Economics, vol.-22, 267-283. 32. Yang, Hao-Yen, 2000, A noe on he causal relaionship beween energy and GDP in Taiwan, Energy Economics, vol.-22, 309-37. 33. Yu. S. H. Eden, and Choi, J., 985, The causal relaionship beween energy and GDP: an inernaional comparison, The Journal of Energy and Developmen, vol.-0(2), 249-270. 2

Appendix Economeric Mehods used As already menioned, in his exercise we have examined wheher income-emission daa for differen counry groups were coinegraed using he Engle-Granger bivariae coinegraion analysis framework and esimaed ECM for counry groups for which coinegraion was observed o be significan, using economeric echniques appropriae for a panel daa se 8. The economeric exercise involved hree seps. In he firs sep, he uni roos es was performed o ascerain wheher or no he ime series of he variables (i.e., naural logarihm of PCGDP and PCCO2, henceforh denoed by y and x, respecively) conained sochasic rend. In he second sep, coinegraion of income and emission was examined. Finally, in he hird sep, he ECM was esimaed for hose counry groups for which coinegraion of income and emission had been found. In he firs sep he IPS panel daa uni roo es procedure was used o es presence of uni roo in he ime series daa ses for individual counry groups. The same procedure was also used in he second sep while performing he Engle-Granger bivariae coinegraion analysis. Finally, he ECM in he hird sep was esimaed by using panel daa regression echnique. In wha follows, we describe briefly he economeric procedures ha we have used in he hree seps of he presen exercise. A. IPS Uni Roo Tes For a balanced panel daa se ( y i, i,2,..., N;,2,..., T ), where i and denoe crosssecional uni and ime, respecively; Im e al. considered he following linear regression se up for developing heir panel uni roo es 8 As is well known, he ECM is a comprehensive linear regression equaion specificaion which provides a descripion of he possible naure of inerdependence of he shor run movemens of a pair of co-inegraed variable keeping in view he fac ha hey bear a long run equilibrium relaionship. 22

y i ρ y i p i + θ jδyi j + ziγ + ε i j. (A) Here z denoes he deerminisic componen of which may be zero, a common γ i y i consan inercep, a ime-invarian fixed effec μ i or a fixed effec ha varies boh across i and over and ε i s are whie noise equaion disurbance erms. Noe ha in (A) he auoregressive parameer ρ i is allowed o vary across unis 9. The null hypohesis for he IPS uni roo es is H 0 : ρ for all i and he corresponding alernaive hypohesis is H : i ρ i < for a leas one i. As ρ i is allowed o vary across i, he IPS es procedure is based on he average of he uni-specific uni roo es saisics. Specifically, his es uses he average of he uni-specific Augmened Dickey Fuller (ADF) es saisics, which has been called he -bar saisic. This is as given below: N N ρ, i i ρ being he -saisic for esing H 0 : ρ in (). I is shown ha, given N, as T, i i ρ weak ly converges o i it 0 W 0 iz dw W 2 iz iz, where W iz denoes a Brownian moion 20. Assuming it s o be independen and idenically disribued wih finie mean and variance, he IPS es saisic is derived as 2 Quah (994) considered equaion (A) wihou he second and hird erms as he model for his panel uni roo es. Levin and Lin (993) considered a more general model o allow for fixed effecs, individual deerminisic rends and heerogeneous serially correlaed errors. In fac, hey considered equaion (A) wihou he second erm as heir model 2 specificaion. They, however, assumed he unis o be iid (0, σ ε ) and also ρ i ρ for all i. Here H 0 : ρ agains H : < ρ. Levin and Lin s es is hus resricive as i requires ρ i o be he same for all i. 20 Brownian moion is also called Wiener Process (see, Hamilon (994), ch-7, p-478). 23

IPS N ( E( it ; H0 : ρi ). (A2) var( ; H : ρ ) it 0 i So far as he acual es procedure is concerned, IPS provide able of esimaes of E( ; H0 : ρ i ) and corresponding var( ; H0 : ρ i ) for differen values of T it i and p compued by sochasic simulaion for wo versions of he ADF(p) regression viz., it i Δy α + βy p + γ jδy j + j error for he wihou ime rend case and Δy α + δ + βy p + γ jδy j + j error for he wih ime rend case. Given hese and he compued value of for he given panel daa, IPS is calculaed using (A2). The able of corresponding criical values for he given values of N and T and various levels of significance are provided in Im e al (997). A.2 Co-inegraion Tes for Panel daa Given a se of panel daa on (K+) variables y, x j, j, K, he single equaion IPS coinegraion es proceeds as follows: Firs, he linear regression equaion y i K j β x ji ji + error is esimaed separaely for i, 2,, N individual unis and he regression residuals e i y i K j ˆβ x ji ji, i,2,..., N;,2,..., T (A3) are obained, where βˆ ji s denoe he esimaed parameers of he regression equaion for he ih uni. These esimaed linear regression equaions may be aken as esimae of he 24

long run equilibrium relaionship beween y and he x s, in case he variables urn ou o be coinegraed 2. Nex, for each i he following ADF(p) equaion is esimaed: where p i ij ei j z + θ Δ + iγ vip (A4) j e i λe + z i γ is same as defined for equaion () above and ν ip is he equaion disurbance erm assumed o be a whie noise. Here also one may consider wo alernaive specificaions of equaion (A4) - viz., one wihou a ime rend and anoher wih a ime rend. The IPS mehodology of coinegraion 22 es for he se of variables under consideraion hus involves he es of uni roo for he regression residuals { }- i.e., he null hypohesis H0: λ (i.e., no coinegraion) is esed agains he alernaive hypohesis H : λ < (i.e., coinegraion). In our empirical exercise, we have performed he coinegraion es wice, viz., once reaing logarihm of PCGDP (i.e., y) as he dependen variable and logarihm of PCCO2 (i.e., x) as he independen variable and again reversing he saus of hese variables. A.3 Esimaion of ECM from Panel daa Once he pair of variables ( x, y ) has been found o be coinegraed, he nex sep in he Engle Granger mehodology is o model he shor run variaions of he variables. This is done by esimaing he ECM. For a bivariae case as he presen one, he ECM, which is implied by he well known Granger Represenaion Theorem (see Hamilon (994), Ch.9, pp. 58-582), is expressed as eiher of he following linear regression equaions: e i 2 I may be noed ha when he variables are coinegraed, he rue relaionship underlying his linear regression equaion is a long run equilibrium relaionship beween y and he x s. Kao, Chiang and Chen (999) poined ou ha for a se of coinegraed variables he use of OLS o esimae his long run equilibrium relaionship from he given se of panel daa will give biased resuls in a finie sample and recommended he use of Dynamic OLS (DOLS) for minimisaion of such bias. See Kao and Chiang (998) for he definiion of DOLS. 22 Panel daa coinegraion es is also performed by Kao (999), McCoskey and Kao (998). 25

Δy Δx i i T T2 μ + α Δy + β Δx + η ECY + u (A5) yx j i j j j T2 T22 j i j μ + β Δx + α Δy + η ECX + u. (A6) xy 2 j i j j j 2 j i j yx xy i i i 2i Here Δ denoes he difference operaor; T lm, l, m, 2 denoes he number of lagged values of Δ and Δ x ha affec he curren value of hese differenced variables; μ, α, β y i i and η denoe regression parameers; u li, l, 2 are he equaion disurbance erms (ha should be whie noises when he ECM has been adequaely specified); and finally, ECY ˆ φˆ i yi φ0 xi and ECX i xi ˆ ϕ 0 ˆ ϕyi are he error correcion erms (hereafer refereed o as EC erms) measuring deviaion of y i ( x i ) from he corresponding long run equilibrium value, given x i y ). ( i 23 The parameers η yx and η xy in equaions (A5) and (A6) are called he adjusmen parameers. They are expeced o have negaive values 24. In his se up he naure of Granger causaliy is deermined as follows: () if β 0 for all j and η 0, x may be said no o Granger cause y; j yx (2) if α 0 for all j and η 0, y may be said no o Granger cause x; 2 j xy (3) if () holds bu (2) does no, Granger causaliy may be said o be unidirecional from y o x; 23 Noe ha here yi φ 0 + φx i + ε i and xi ϕ 0 + ϕy i + ε 2i are alernaive represenaions of he (populaion) long run equilibrium relaionship beween y and x, where ε s are he saionary error erms. As y and x are coinegraed, by he definiion of coinegraion for some consans, ω 0 + ωyi + ω2x i εi, where εi is a saionary error erm and ω ( ω0, ω, ω2) is he non- normalized coinegraing vecor. Thus, by normalizing ω one may wrie he long run equilibrium relaionship for (y,x) in eiher form as shown above. 24 This is for he following reason. If, for example, ECYi > 0 for some i,, i means ha he realized value of yi exceeded he corresponding long run equilibrium level a -, given x i. Now since y i and x i are coinegraed, once a posiive deviaion from he long run equilibrium level akes place, he acual value mus ry o move in he opposie direcion in subsequen ime poins in an aemp o resore he long run equilibrium and hence he negaive sign of η and η. yx xy 26

(4) Conversely, if () does no hold bu (2) does, Granger causaliy may be said o be unidirecional from x o y; (5) if boh () and (2) do no hold, Granger causaliy beween x and y may be said o be bidirecional; and finally (6) if boh () and (2) hold, Granger causaliy beween x and y may be said o be absen (see Enders (995), Glasure and Lee (997) and Asafu-Adjaye (2000) for deails). In he presen exercise, equaions (A5) and (A6) (henceforh referred o as model I and model II, respecively) were esimaed separaely for each counry group, using he panel daa se for he counry group. Counry group-specific inference abou he naure of Granger causaliy beween x and y were hen drawn by performing appropriae es of hypohesis for he relevan parameers of model I and II, as laid down above. For example, o es he null hypohesis ha x does no Granger cause y, one should perform an F-es for he null hypohesis H β j 0, j,2,...,, η 0, using model I. 0 : T2 Similarly, o es he null hypohesis ha y does no Granger cause x, an F-es for he null hypohesis H α 0, j,2,..., T, η 0 using model II will be required. Given 0 : 2 j 22 xy he resuls of hese wo basic F-ess, he remaining null hypoheses (3)- (6) laid down above can be esed. yx 27

Table. Coninen-wise lis of counry groups and counries covered Coninen Counry Group Counries Covered Africa Africa Algeria, Cameroon, Cape Verde Island, Cenral African Republic, Comoros, Congo, Egyp, Gabon, Gambia, Ghana, Guinea, Guinea Bissau, Kenya, Madagascar, Mali, Mauriania, Mauriius, Morocco, Mozambique, Nigeria, Senegal, Souh Africa, Togo, Tunisia, Uganda, Zimbabwe. Norh America Canada and USA America Cenral America Cosa Rica, Dominican Republic, El Salvador, Guaemala, Honduras, Jamaica, Mexico, Nicaragua, Panama, Trinidad & Tobago. Souh America Argenina, Bolivia, Brazil, Chile, Colombia, Ecuador, Asia Japan Japan. Paraguay, Peru, Uruguay, Venezuela. Asia (excluding Japan) China, Hong Kong, India, Indonesia, Iran, Israel, Jordan, Korea Republic, Philippines, Singapore, Sri Lank, Syria, Thailand. Eas Europe Ausria, Czechoslovakia, Finland, Greece, Turkey, Yugoslavia. Europe Wesern Europe Belgium, Cyprus, Denmark, France, Wes Germany, Iceland, Ireland, Ialy, Luxembourg, Neherlands, Norway, Porugal, Spain, Sweden, Swizerland, U.K. Oceania Oceania Ausralia, Fiji, New Zealand, Papua Guinea. 28

Table 2. Resuls of Panel Uni Roo Tes : IPS saisic by Counry Group Counry Group Wih Time Trend Wihou Time Trend -bar For Criical Value -bar For Criical Value income emission (5% level) income emission (5% level) Africa Norh America Cenral America Souh America America Japan Asia(excl. Japan) Asia Eas Europe Wes Europe Europe Oceania World -0.289-0.330 2.09.92 2.6 NA -0.734-0.842 3.238-0.70.093-0.250.306-0.376 0.486.025 0.980.498 NA -0.250-0.307.308-0.605 0.67-0.488 0.402-2.45-2.94-2.60-2.60-2.47-2.56-2.54-2.74-2.52-2.47-2.84-2.32 2.469 0.296-0.038.20 0.880 NA 6.068 5.757-0.592 3.283 2.49 0.949 5.526 0.664 -.384-0.302 0.949 0.09 NA 2.35 2.075-2.23 0.022 -.090 0.293 0.75 Noe:. Im e al (997) give Tables of criical values of heir Panel uni roo es saisic for seleced combinaions of N and T values. The criical values shown in he presen Table have been derived from he original Tables by inerpolaion whereever required. 2. NA denoes no available. For Japan, a single counry, he panel uni roo es was no applicable. Hence no resul is shown agains Japan. -.82-2.30 -.99 -.99 -.84 -.94 -.92-2.2 -.89 -.84-2.2 -.68 Table 3. Resuls of Coinegraion Tes : IPS saisic by Counry Group Counry Group wihou ime rend criical value wih ime rend criical value income emission income emission Africa -0.880-2.57 -.82-2.643-4.80-2.45 Norh America -0.608-2.82-2.30 -.665-0.567-2.94 Cenral America -2.05-2.263 -.99 0.905-2.524-2.60 Souh America -0.846 -.09 -.99 -.384-2.23-2.60 America -2.2-2.99 -.84-0.825-3.304-2.47 Japan NA NA NA NA Asia(excl. Jap) 3.428-0.054 -.94 -.862 -.543-2.56 Asia 3.052-0.398 -.92 -.879 -.53-2.54 Eas Europe -2.089-3.523-2.2-2.237-4.649-2.74 Wes Europe 0.572-2.484 -.89-3.088-3.935-2.52 Europe -0.603-3.958 -.84-3.802-5.784-2.47 Oceania -0.363-0.978-2.2-0.922 -.520-2.84 World -0.696-5.203 -.68-4.697-7.744-2.32 Noe:, and denoe he significance level a 0%, 5% and %, respecively. Criical values shown correspond o he 5% level of significance. NA denoes No Applicable. 29

Table 4. Esimaed parameers of he ECM for counry groups for which coinegraion hypohesis was no rejeced Counry group Esimaed coefficien of he explanaory variable ( Δ log) Model income_ income _2 income_3 emission_ emission_2 emission_3 Africa I (3) 0.0 0.0-0.07 0.00-0.00-0.02-0.09 (2.59) (2.56) (-.79) (0.02) (-0.25) (-.92) (-4.77) II (3) 0.05 0.2-0.8-0.20-0.08-0.7-0.26 (0.3) (.3) (-.2) (-4.59) (-.72) (-4.2) (-7.95) Cenral America I(2) 0.92 0.09-0.004 0.06 - -0.0906 (2.993) (0.3) (0.8) (0.8) (-2.62) II(2) 0.782 0.52 - -0.4-0.28 - -0.86 (4.) (0.8) (-5.97) (-4.7) (-3.42) America I(2) 0.229-0.02-0.02 0.0 - -0.059 (5.3) (-0.45) (0.7) (0.68) (-3.6) II(2) 0.666 0.9 - -0.36-0.238 - -0.09 (6.09) (.7) (-8.4) (-5.76) (-3.28) Easern Europe I(3) 0.72-0.056 0.205 0.052-0.38-0.08-0.083 (2.8) (-0.72) (2.7) (.09) (-2.98) (-0.37) (-4.58) II(2) 0.029 0.45-0.04-0.08 - -0.32 (0.22) (.) (0.9) (-0.24) (-4.85) Wesern Europe I (2) 0.24-0.8-0.04-0.04 - -0.03 (4.85) (-3.62) (.97) (-2.2) (-3.33) II (2) 0.6 0.08-0.05-0.07 - -0.03 (.2) (0.59) (0.98) (-.23) (-.74) Europe I (3) 0.23-0.3 0.08 0.04-0.06-0.02-0.04 (5.26) (-2.92) (.8) (2.32) (-3.30) (-0.76) (-4.89) II (2) 0.2 0. - 0.07-0.04 - -0.07 (.0) (.06) (.52) (-0.97) (-4.00) World I (3) 0.2 0.03-0.03 0.02-0.00-0.0-0.04 (5.45) (.55) (-.47) (.9) (-0.2) (-.66) (-5.33) II (3) 0.26 0.27-0.02-0.22-0.09-0.2-0.7 (3.54) (3.76) (-0.32) (-9.40) (-4.06) (-5.44) (-.2) Noe:. Figure in brackes in he model column indicaes he opimum number of lagged variables used as regressors in he ECM as deermined for he given daa se. 2. For each counry group and model he firs row of 3 rd o 9 h column gives he esimaed coefficiens. The corresponding figures in brackes in he nex row of hese columns are he corresponding -raios. EC erm 30