Transcript Is there an Environmental Kuznets Curve for Energy Use and

Is there an Environmental
Kuznets Curve for Energy Use
and Carbon Emissions?
Amy K. Richmond and Robert K. Kaufmann
US Association for Ecological Economists
Saratoga Springs, NY
May 22, 2003
http://cybele.bu.edu/people/arr.html
Talk Overview
• Omitted variable bias
– Energy Mix
• Model specification
– Quadratic, Semi-Log, Double-Log
• Tests of predictive accuracy
• Questions
– Does the inclusion of energy mix influence turning
points?
– Is there a turning point?
• Conclusion
– Omission of energy variables effects turning point
– Quadratic specification is best
2
Environmental Kuznets Curve
•
Reasons for EKC
include income driven
changes in:
–
–
–
–
composition of
production and
consumption;
preference for
environmental quality;
institutions that are
needed to internalize
externalities;
increasing returns to
scale associated with
pollution abatement
Natural
Resources
Use
and\ or
Emission of
Wastes
Income
3
Turning Points
Natural
Resources
Use
and\ or
Emission of
Wastes
Income
4
Energy Omission
• E/GDP influenced by energy mix
• Different energy types have different CO2
emissions
• Statistical effects of omitted variable bias
5
Methodology: Data
• Panel of International Data
– 36 nations
• 20 developed countries
• 16 developing countries
•
•
•
•
– 1973-1997
Total Economic activity measured by GDP in
1996 US dollars, converted using PPP
indices
Carbon emissions (kg/ million BTU)
Total energy use (BTU’s)
Final energy consumption (BTU’s)
6
Basic Model
Yit     X it   ln(Ztj ) it
• Yij is a measure of energy per capita (TE/Pop) or carbon emissions per capita
(CO2/Pop) by nation i at time t
• X is per capita GDP
• Z is a vector of fuel shares (PCTCOAL, PCTPET, PCTELC)
• µ is the regression error
• α, β, Φ, are regression coefficients
7
Model Specifications
1. Quadratic Specification:
Yit    1 X it  2 X it2  ln(Zit ) it
• EKC if β1 > 0 and β2 < 0
• Turning point = –(β1/ 2β2)
2. Semi Log Specification:
Yit     ln(X it )   ln(Zit ) it
• Diminishing returns
3. Double Log Specification:
ln(Yit )     ln(X it )   ln(Zit ) it
• Constant elasticity
8
Omitted Variable Bias: Fuel Share
PCTCOAL = ln(FINCOAL/TE)
PCTPET = ln((FINOIL+FINGAS)/ TE)
PCTELC = ln((HYRDO+NUCLEAR)/TE)
• Expect coefficient associated with PCTCOAL to
be positive and coefficients associated with
PCTPET and PCTELC to be negative.
• Diminishing Returns
9
Estimation Techniques
• Regression techniques:
– Pooled OLS
Yit    X it   Zit  it
– Fixed Effects or Random Effects Estimator
Yit  i   Xit   Zit  it
– Random Coefficient Model (Swamy, 1970)
Yit  i  i X it  i Zit  it
• Cointegration (Pedroni, 1999)
10
Tests of predictive accuracy
(Diebold and Mariano, 1995)
N
 I d   0.5N

S2a 
t
t 1
0.25N
N N  1
 I  dt rank( dt ) 
t 1
4
S3a 
NN  1(2N  1)
24
N
I (dt )  1
if dt  0
 0 otherwise
11
Results
• Variables generally have correct sign and
statistically significant
• GDP variables have correct sign, quadratic term
not statistically significant
• All variables contain stochastic trend (indicates
modeling variables using time trends is not
sensible)
• Quadratic specification cointegrates
• Energy shares allow diminishing returns
specifications to cointegrate
12
Relation between income and energy consumption
(corrected for changes in energy mix)
Quadratic Model
Semi Log Model
Double Log Model
13
Relation between income and CO2 emissions
(corrected for changes in energy mix)
Quadratic Model
Semi Log Model
Double Log Model
14
Effect of Energy Mix on
Turning Points
• Energy use
– Turning point without fuel shares: $43,767*
– Turning point with fuel shares: $52,296*
• CO2 emissions
– Turning point without fuel shares: $110,600
– Turning point with fuel shares: $29,700
* Calculated even though quadratic term not statistically significant
15
Why do Energy Mix Variables
Effect Turning Points?
Sit    1Xit  2 Xit2  it
Fraction
Coal
Fraction
Petroleum
Fraction
Electricity
GDP
0.00443
(2.89)
-0.031
(8.3)
0.00197
(8.2)
GDP2
-46.53E-04
(1.7)
0.00127
(0.8)
--
Turning
point
$3,390
$12,070
--
Panel ADF
-3.14
-5.45
-6.51
Group ADF
-4.40
-7.13
-9.12
16
Conclusion
• Omission of energy variables effects turning
point
• Quadratic specification generates a more
accurate out of sample forecast
• Modeling relationship between energy use
and income using time trends is not sensible
http://cybele.bu.edu/people/arr.html
17