Can deregulation of CNG reverse the outcome of regulation?
Evidence from Thailand’s transport sector
Thanicha Ruangmas∗ and Corbett Grainger†
April 14, 2017
Preliminary and incomplete - please do not cite
Abstract
Fossil fuel subsidies are being criticized for causing excessive fuel consumption which contributes
to more air pollution and greenhouse gases, as well as being inequitable as it benefits higher
income groups. Not all fossil fuel subsidy polices are the same, some may have a net benefit.
This research uses an atmospheric science model to control for meteorological biases, extensive
individual fueling station data to control for sources of transport fuel, and data on CO, NO2,
O3, SO2, and PM10 concentrations. We find that that increased compressed natural gas (CNG)
availability in Thailand improves air quality. However, the effect of CNG price on air quality is
unclear. If prices do not affect consumer decisions to use CNG, then it is beneficial that CNG
subsidies are removed. More work is being done to confirm this statement by analyzing the
effect of car price and fuel price on fuel adoption.
Keywords: Compressed natural gas; Pollution Dispersion; Fossil Fuel Subsidy; Transportation
1
Introduction
Compressed natural gas, or CNG, is adopted in road vehicles in cities like New Delhi, Dhaka,
Mexico City, or in this case, Thailand. Unlike other types of fossil fuels, adoption of CNG could be
beneficial. CNG fueled vehicles emit greenhouse gases with less global warming potential and less
air pollutants than gasoline-fueled vehicles (Dholakia et al. 2013). Adoption of CNG could reduce
air pollutants which could improve the human health (Kumar and Foster, 2007).
To decrease dependence on global fuels, the Thai government introduced CNG as an alternative
fuel choice for cars in 2004. The price of CNG in Thailand have been regulated in two separate
parts. First, the government subsidized PTT Plc., Thailand’s sole distributor of CNG, to sell CNG
∗
Corresponding Author. Graduate student. Department of Agricultural and Applied Economics, University
of Wisconsin - Madison. 427 Lorch Street, Taylor Hall Madison WI, USA 53706. [email protected]. Phone:
0016083389445
†
Assistant Professor. Department of Agricultural and Applied Economics, University of Wisconsin - Madison. 427
Lorch Street, Taylor Hall Madison WI, USA 53706. [email protected]
1
to retail stations at a lower price. Second, the Thai government has been fixing the retail price of
CNG under the market price and repays retail fueling station owners for their losses.
CNG subsidy removal, as part of the global movement towards fossil fuel subsidy removal, has
pressured the Thai government to remove the subsidies. Such price regulations are being criticized
for causing excessive fuel consumption which contributes to more air pollution and greenhouse
gases, as well as being inequitable as it benefits higher income groups (Coxhead and Grainger 2014;
Yusuf and Resosudarmo 2010). The Thai government has stopped subsidizing CNG distributors in
2012.
Similar to the experiences of fellow Southeast Asian countries, social and political pressure
make deregulation not as easy. CNG subsides today are less than it was before, however, plans to
completely deregulate the market have kept postponing.
This research tries to see if CNG adoption in the transportation sector has contributed to
cleaner air and will its price deregulation reverse this effect. To answer this question, we have to
understand the mechanisms underlying the set up of CNG retail stations and the adoption of CNG
fuel in private road vehicles. To our knowledge, the only paper that have looked at the mechanisms
of fuel adoption and its impact on air pollution is Auffhammer and Kellogg (2011). Although
policies to make cars greener have been extensively studied, Anderson and Sallee (2017) pointed
out that there is inadequate research in middle-income countries.
Research on CNG adoption and air pollution is not new. However, existing research have used
poor data as CNG are usually adopted in developing countries with few air pollution monitors and
little available information. This research uses monthly data from more than 40 air pollution monitors across Thailand, individual retail fueling stations including 516 CNG fueling retail stations,
natural gas pipelines, and car sales by fuel type. Our data set spans from 2003 to 2014, from even
before CNG was introduced until its subsidies have been slowly removed.
This research also builds upon existing research on the air pollution effects of CNG adoption by
using a Hybrid Single Particle Lagrangian Integrated Trajectory Model (HYSPLIT; see (Draxler
and Hess 1997; Draxler and Hess 1998; Draxler 1999)) to identify sources of air pollution measured
at each monitor. This corrects for meteorological effects that could deviate estimated coefficients.
We find that an increase in CNG availability improves air quality. Our work in progress, which
will also control for selection bias from the fact that CNG stations are built in polluted areas, are
expected to increase the effect of CNG availability on clean air.
2
Figure 1: Total cars in Thailand by each fueltype
2
Background
Natural gas in Thailand comes from the Gulf of Thailand, Myanmar, and other countries in the
form of liquefied natural gas (LNG). Natural gas is mostly used in electricity production and in
other manufacturing industries.
High world oil prices in the early 2000s lead households to start adopt liquefied petroleum gas
(LPG) in their cars. As LPG is highly reactive and denser than air, it is very dangerous. The
Thai government then decided to introduce CNG in road vehicles in 2004. The first CNG fueling
station was built in 2006. PTT Plc., Thailand’s sole distributor of CNG, was distributing CNG to
retail fueling stations under the market price. To alleviate PTT of such burden, the government
subsidized PTT until 2012.
Traditional fuel options for road vehicles in Thailand include gasoline and diesel fuels. Common
alternative fuel options include LPG and CNG, which is generally adopted by retrofitting a gasolinefueled car. Figure 1 shows the number of cars by fuel type. As gasoline fueled cars can be converted
to use either CNG or LPG fuel, the number of CNG and LPG cars in the figure is underestimated.
2.1
CNG price policies
From 2003 to 2006, retail price of CNG was fixed at 50 percent of diesel prices. In 2007 its price
was fixed at 8.5 Baht per kilogram with plans to slowly lift the price ceiling and float its price in
3
2011.
In November 2007, while retail prices are still fixed 8.5 Baht per kilogram, the Thai government
restructured the CNG market. For every unit of CNG sold at fueling stations, revenue received by
fueling station retailers are calculated from the following equation
PCN G = W HP ool2 × (1.0175) + T dZone1+3 + T c − OilF und + StationCost + M arketingCost
where W HP ool2 is well-head gas price (in Baht per million British Thermal Unit, or BTU)
obtained from the Gulf of Thailand, Myanmar, and other countries. T d is fixed cost for pipeline
use. T c is unit transport cost in Baht per BTU of gas transported.
OilF und is a pool of government revenue collected from taxing diesel and gasoline fuel sales
and is used to subsidize ethanol and CNG. CNG was first subsidized at 2 Baht per kilogram, but
was reduced to 1 Baht per kilogram in 2011. CNG is no longer subsidized after February 2012.
StationCost is around 1.00 to 1.12 Baht per kilogram of gas depending on how CNG is transported to the retail station. M arketingCost is around 1.73 to 2.33 Baht per kilogram depending
on the size and type of CNG station.
In conclusion, CNG station owners get PCN G , while a consumer pays the regulated price of
CNG PRegulated . It is suspected that PRegulated ≥ PC CN G and that PTT, the wholesaler of CNG
bears this cost.
Figure 2 shows the difference between the Henry Hub spot price of CNG (converted to Baht per
kilogram with each month’s exchange rate), the estimated subsidized price to retail station owners
(Henry Hub spot price minus unit subsidy), and the regulated retail price of CNG. Henry Hub spot
price is used because the Thai government also regulates the imported price of CNG, and that is
based on a one month lag of the Henry Hub spot price. Because CNG subsidy to retail station
owners ends in 2012, the subsidized price coincides with the world price afterwards.
The Thai government first agreed to lift CNG subsidies in September 2011 and completely
deregulate retail CNG prices in April 2012. Similar to the experiences of fellow Southeast Asian
countries, social and political pressure make implementation not as easy. Although CNG subsides
today are less than it was before, and CNG retail prices are closer to the market price, plans to
completely deregulate the market have kept postponing. CNG price today is regulated at 13.50
Baht per kilogram and the government have stopped announcing new plans to float CNG prices.
4
Figure 2: CNG price regulation timeline
3
Literature Review
Natural gas adoption in road vehicles is not a new policy. Outside Thailand, cities like New Delhi,
Dhaka, and Mexico City have encouraged all taxis and buses to run on natural gas (Narain and
Krupnick, 2007; Schifter, Diaz, Lopez-Salinas, and Avalos, 2000; Wadud and Khan, 2013). Papers
that have evaluated this policy can be divided into two categories based on their approaches;
bottom-up1 and top-down estimation. Bottom-up papers are used to predict the air pollution
consequences of such policies whereas top-down approach use existing data to analyze what had
happened. As this paper uses econometric models to analyze what had happened, only top-down
paper will be reviewed.
Some papers have compared ambient concentrations before and after a natural gas adoption
policy (Ravindra et al. 2006; Suthawaree et al. 2011). Specific contents of natural gas such as
polycyclic aromatic hydrocarbons (PAHs) (Ravindra et al. 2006), or specific characteristics of onroad emissions such as CO-NOx or SO2-NOx ratios were used as indicators to ensure that natural
gas adoption has attribute to changes in ambient concentrations (Suthawaree et al. 2011). Other
top-down literature have used econometric models to evaluate air pollution impacts of the policy
(Narain and Krupnick 2007; Kumar and Foster 2007). In general, existing literature have found
1
Bottom up approaches usually starts with estimating emission factors (quantities of emissions per unit of fuel
use) for different types of vehicles with different types of fuels. The number of vehicles as well as distance traveled
is then estimated based on pre-defined energy scenarios. Atmospheric models are used to evaluate the changes in
ambient concentrations.
5
that natural gas adoption can reduce particulate matter, SO2, and NOx. The effect on CO is
unclear.
Except for Kumar and Foster (2007), all top-down papers have relied on data from only one
air pollution monitor. As pollution monitors are placed in highly polluted areas, this selection bias
could underestimate the effect that CNG adoption has on air pollution concentrations.
4
Data
4.1
Air pollution monitor
The Pollution Control Department releases month average concentrations from the air pollution
monitors. In January 2003, there are 41 CO and NO2 monitors, 40 SO2 monitors, 38 PM10
monitors, and 35 O3 monitors across Thailand. By 2014 the total number of air pollution monitors
have increased to 63, with each monitor measuring multiple pollutants.
Table 1 shows average pollution concentrations for existing monitors since January 2003 (old
monitors), monitors that were built after January 2003 (new monitors), and all monitors. Except
for O3, older monitors have higher pollution concentrations than the newer monitors as they are
placed in areas with the highest concern for air pollution. Data from air pollution monitors that
are built after January 2003 are dropped. We also dropped 8 hr. CO and 8 hr. O3 measurements
from analysis as there are too little observations.
4.2
Fueling station data
Three fuel specific data sets are available; one for CNG stations, one for LPG stations, one for
gasoline and diesel. Each of these data set do not cover the entire period of interest, from 2003 to
2014.
A fourth data set contains information about all retail stations is used to supplement each data
set. This data set, however, does not identify which fuels are sold at each station. We will call this
data set the “general retail station data set”. This section discusses how each data set is obtained,
cleaned, and combined.
4.2.1
CNG stations
Before January 2013, CNG was considered a hazardous material. All CNG fueling stations must
obtain a permit from the Department of the Energy Business (DOEB). The first data set shows
6
Table 1: Air pollution summary statistcs
1 hr. CO (ppm)
Observations
8 hr. CO (ppm)
Observations
24 hr. PM10 (ug/m3)
Observations
1 hr. SO2 (ppb)
Observations
1 hr. NO2 (ppb)
Observations
1 hr. O3 (ppm)
Observations
8 hr. O3 (ppm)
Observations
(1)
Old monitors
0.69
(0.01)
5800
0.76
(0.03)
5788
49.91
(0.39)
5680
3.66
(0.04)
5448
16.72
(0.14)
5594
18.54
(0.12)
5242
21.12
(0.28)
1272
Standard errors in parentheses
7
(2)
New monitors
0.52
(0.01)
1092
0.52
(0.01)
1093
41.23
(0.71)
1287
2.07
(0.05)
881
11.70
(0.27)
886
21.72
(0.30)
981
22.44
(0.51)
403
(3)
All monitors
0.67
(0.00)
6892
0.72
(0.03)
6881
48.31
(0.35)
6967
3.44
(0.04)
6329
16.04
(0.13)
6480
19.04
(0.11)
6223
21.44
(0.25)
1675
476 CNG permits that were given out to fueling stations from May 2009 to August 2012. The data
set includes permit issue date2 , permit expiration date, fueling station address, and fueling station
type. The data set does not indicate all CNG stations that have closed. This overestimates the
number of CNG stations in our data set.
CNG stations that are built after 2013 are recorded in the general retail station data set. We
use the August 2015 version of this data set, and identified 40 new CNG stations that have opened
after August 2012 until August 2015.
4.2.2
LPG stations
LPG fueling station data from the DOEB provides a list of existing LPG fueling stations and its
addresses, but not the date that it starts its business. We merge LPG business names with names
from the general retail station data set to find dates in which each LPG fueling station starts its
business. Of the 1,852 LPG fueling stations that existed in September 2014, we were able find 1,320
LPG stations (72 percent of all LPG stations) with dates in which each station starts is business.
We only included the 1,320 LPG stations. These are considered LPG stations that have opened
and remained in business until September 2014.
4.2.3
Regular gas stations
By using the August 2015 list of gas station from the DOEB, we assume that all other stations
that do not sell CNG or LPG sells gasoline and diesel fuel. Because we ignored the fact that some
gas stations have closed, the number of regular gas stations in our estimation is overestimated.
4.2.4
Natural gas factories
The DOEB also provides the names, and addresses, and dates that each factory has converted from
other types of fuel to natural gas. However, we do not know when each factory is set up and what
type of fuels do they use prior converting to natural gas.
4.3
Weather covariates
Weather data from 52 weather stations from year 2003 to 2014 in Thailand are obtained from
the National Climatic Data Center (NCDO). Weather covariates includes monthly mean temperature, monthly mean minimum, monthly mean maximum temperature, monthly extreme minimum
2
The permit issue date represents the date in which each CNG fueling station starts its business
8
temperature, monthly extreme maximum temperature, monthly total precipitation, and monthly
extreme maximum daily precipitation. In addition, we interact monthly mean temperature with
monthly total precipitation.
4.4
Fuel prices
Average monthly fuel price data are available from the Energy Policy Planning Office (EPPO).
Natural gas price is fixed across the country as described in the policy context section. Prices for
other types of fuels are based on a daily fuel price, retail station brand, and district-level markups.
There is very low spatial variability in fuel prices.
Different mixes of gasoline with different blends of ethanol are available over time, at varying
prices. We use the price of unleaded gasoline with an octane level of 95 (ULG95) to represent the
price of gasoline as it is the only type of gasoline that is available from 2003 to 2014. Although
ULG95 is the most expensive and least used type of gasoline, we find that its price strong correlates
with other types of gasoline and ethanol blends.
4.5
Provincial covariates
Annual province-level data comes from the National Statistics Office. Province specific variables
include annual population, annual average income, total gross provincial product (GPP) in the
manufacturing, utility, and automotive sector.
4.6
Pipelines
Thailand’s first national gas blueprint allows for the establishment of four major natural gas separation plants in the 1980s. The first three gas separation plants were built to accommodate domestic
gas from the Gulf of Thailand, or import LNG from other countries that are transported by ships.3
As part of the revised national natural gas pipeline blueprint announced in May 17, 2003, the
Thai government gave rights to PTT Plc., to invest in natural gas pipeline construction with the
overall objective of decreasing dependence of fossil fuel imports, increasing reliance on cheaper fuels,
and reducing air pollution. As a result, 40 new natural gas pipelines were being built during 2010
to 2014. The construction of each pipeline was documented with details on each pipeline’s objective
3
These three plants are Khanom Gas Separation Plant in Nakhon Si Thammarat, Trans Thai-Malaysian (TTM)
Gas Separation Plant in Songkla, and Rayong Gas Separation Plant in Rayong. The forth plant, TBW Gas Separation
Plant in Kanchanaburi, is to accommodate imported natural gas from Myanmar.
9
(to distribute natural gas to an electric power plant, a manufacturing power plant, or to natural
gas for vehicle stations), pipeline distance, diameter, district, construction date, and provinces that
it will go through.
4.7
Highways
We also have a GIS map of all land transport routes in Thailand in 2011. This includes railroads,
highways, major roads, and different other routes.
5
Empirical methodology
An ideal experiment would be to look at air quality in one area with a CNG subsidy, and another
area where a CNG subsidy is randomly removed. However, such social experiment does not exist.
Because there is no spatial variation in CNG prices, we exploit the fact that there are costs to
the search of fuel (Houde 2008, Manuszak and Moul 2009). We interact fuel price and different
variables that represent fuel availability, and use this as our variable of focus in our reduced form
estimation. Our data set is at the air pollution monitor-month level.
Preliminary estimation in this section follows reduced form estimates in Schlenker and Walker
(2016). Because CNG retail stations are not the only sources of fuel, we also include LPG fueling
stations, regular fueling stations, and CNG factories as covariates. We begin by estimating the
direct effect of all fuel prices and costs to obtain each fuel type on air quality. Like Greene (1997)
we assume that fuel availability and price represent the costs to obtain fuel and take a quadratic
form. Our regression takes the following form:
AQMmt =
X
0
{αf 1 (αf 2 Pf mt +Af mt )2 +αf 3 Pt +αf 4 At }+Xmt
γ +θm +ηmonth,region +ζyear +mt (1)
f
where AQMmt is the air pollution concentration of monitor m at month-year t. Pf mt is the
spot fuel price of CNG and other fuels f . Af mt is fuel availability. Different variables that are used
to represent fuel availability is described in the next section. αf 2 Pf mt + Af mt would then represent
the costs of obtaining natural gas. We square this term in order to obtain the interaction effect
Pf mt × Af mt which would have both spatial and temporal variation. Xmt are provincial covariates.
This include number of households, population density, household income. To partially control for
emissions from other sectors, we also include gross provincial product from year sector-province10
year. We also include a dummy variable if the air pollution has been moved. To control for regional
weather patterns, we include month-region fixed effects. We also control for monitor fixed effects
for any monitor-specific effects such as monitor placement. To control for national monthly demand
shocks (for example, the July 2011 flood or coup on May 2014), we also include year fixed effects.
Results are not yet clustered.
5.1
Proxies for costs of fuel search
We follow the methodology of Du and Li (2017), and use the average distance (in kilometers) to
10 closest fueling station. Furthermore, we also follow the methodology of Li et al. (2016) and
Manuszak and Moul (2009) which used the total number of fueling stations in a specified area. We
have calculated the number of fueling stations surrounding a 10 km., 50 km., and 100 km. radius
surrounding every air pollution monitor. Summaries of these variables are shown in column (1) to
(4) for Figure 2.
Table 2: Emission sources summary statistics
(1)
Average
distance (in
km.)
CNG stations
Observations
CNG factories
Observations
LPG stations
Observations
Regular stations
Observations
249.61
(2.54)
6246
414.71
(4.39)
6246
33.70
(0.78)
6246
3.87
(0.07)
6246
(2)
Number
of stations
within 10
km.
3.50
(0.12)
6246
10.40
(0.37)
3162
16.62
(0.24)
6246
60.54
(0.58)
6246
(3)
Number
of stations
within 50
km.
26.83
(0.89)
6246
122.48
(1.77)
3162
117.90
(1.68)
6246
528.82
(3.95)
6246
(4)
Number
of stations
within 100
km.
41.32
(1.26)
6246
236.09
(3.26)
3162
179.21
(2.23)
6246
1438.45
(9.16)
6246
(5)
HYSPLIT
weights
82.45
(2.68)
6246
149.75
(5.10)
6246
348.44
(5.38)
6246
1700.25
(15.36)
6246
Standard errors in parentheses
However, by just tracking the set up of fueling station and fuel prices around a air pollution
monitor could not accurately measure its effect on air quality at a specific area. Factors such as
wind and precipitation could also change air pollution concentration measurements, resulting in a
bias in our coefficients (Sullivan 2016).
11
Figure 3: Sample HYSPLIT simulation
To ensure sources of air pollution at each air pollution monitor, a Hybrid Single Particle Lagrangian Integrated Trajectory Model (HYSPLIT; see (Draxler and Hess 1997; Draxler and Hess
1998; Draxler 1999)) is used. HYSPLIT could identify sources of air pollution on each monitor
every day of the year.
We run HYSPLIT for every day, two times a day of a representative year. We find the aggregate
impact that each 0.01 latitude and 0.01 longitude grid 4 on each air pollution monitor of each month.
We then scale its impact such that all non-zero values for each month-monitor sums up to one.
We call this the “HYSPLIT weights”. The higher the HYSPLIT weights, the higher the impact on
each air pollution monitor. Figure 3 shows the HYSPLIT weights of from simulations of a sample
pollution monitor in Bangkok.
4
This is similar to a square block of 1.2 square kilometers
12
We then count how many fueling stations are in each 0.01 latitude and 0.01 longitude grid of
each month. The effect that each type of fueling station has on an air pollution monitor is the
sum product of the number of fueling stations and the HYSPLIT weights. Column (5) in Table 2
summarizes HYSPLIT weighted number of fueling stations.
5.2
Inverse distance weighting to accompany average distance to fueling stations
This section describes how prices and other covariates are calculated when distance to fueling
station is used to represent fuel availability. Inverse distance from each air pollution monitor to
each province centroid is used to weigh province-level covariates.5
The same steps are done to find inverse distance weighted weather data, using distances from
each air pollution monitor and weather stations instead. Summary statistics of such variables are
found in column (1) of Table 3.
5.3
Province fraction weighting to accompany number of stations within a given
radius
This section describes how prices and other covariates are calculated when the number of stations
within a given radius represents fuel availability. First, we draw a radius of 10 km., 50 km., and
100 km. around an air pollution monitor. Provincial covariates at the monitor level are weighted
averages from provinces that surround that air pollution monitor.6 Summary statistics of such
variables are found in column (2) to (4) of Table 3.
5.4
HYSPLIT weighted price and covariates
We find the monthly aggregate impact each 0.01 latitude and longitude grid point has on each air
pollution monitor. For each monitor and month, we identify the 0.01 latitude and 0.01 longitude
grid point that coincides with a province’s centroid. The HYSPLIT weight of that latitude and
longitude grid point would then represent the impact that each province has on each air pollution
1
gaspricept where Distancemp is the distance from monitor
ˆ mp
Distance
P
1
m to province p. It is scaled such that
= 1. gaspricept is gas price in province p at time t.
p
ˆ mp
Distance
6
For example, the province fraction weighted values of population density in period t is
P areamp
densitypt whereareamp is the area of province p that is within some radius of monitor m. aream
p area
m
is the total area surrounding monitor m. densitypt is the population density of province p at time t.
5
For example, gas price in period t is
P
p
13
Figure 4: Marginal effect of CNG availability on pollution
monitor. For each monitor and month, we scale the HYSPLIT weights of all province centroids
such that it sums up to unity. We then find the sum-product of the scaled HYSPLIT weights and
data from corresponding province and month to find monitor-month HYSPLIT weighted data.7
Summary statistics of HYSPLIT weighted variables are found in column (5) of Table 3.
6
Results
An increase in fuel availability should induce fuel adoption. We expect an increase of CNG availability to decrease air pollution concentrations. Figure 4 shows the marginal effect of CNG availability
on air pollution concentrations when CNG is priced at 13.5 Baht per kilogram.8 It is evident that
different proxies for fuel availability yield different results. Most notably, we see that the marginal
effect is more negative when CNG availability weighted by HYSPLIT. This shows that HYSPLIT
has corrected for wind bias which is negatively correlated with air pollution concentrations.
The marginal effects using HYSPLIT weights and ordinary least squares implies that increasing
CNG availability (decreasing cost to search for fuel) is correlated with lower levels of air pollution.
An increase in fuel price should hinder fuel adoption. We expect that an increase of CNG
7
P
For example, the household income of monitor m and time t is p HY SP LITmpt incomept where HY SP LITmpt
P
is the monthly aggregate impact province p on monitor m in time t. It is scales such that
HY SP LITmpt = 1.
p
incomept is annual household income of province p at time t.
8
Distance to fueling station is omitted from the Figure 4 as it generates unrealistic results. Full set of results in
tabular form are available from the authors upon request.
14
Table 3: Covariate summary statistics
CNG price(in Baht/Kg.)
Observations
Diesel price(in Baht/Liter)
Observations
LPG price(in Baht/Liter)
Observations
Unleaded gasoline 95 price (in Baht/Liter)
Observations
Annual GPP (in thousand Baht)
Observations
Annual GPP in the automobile industry (in thousand Baht)
Observations
Annual GPP in the logistics industry (in thousand Baht)
Observations
Annual GPP in the manufacturing industry (in thousand Baht)
Observations
Annual GPP in the utilities industry (in thousand Baht)
Observations
Annual household income (in Baht.)
Observations
Population
Observations
Population density (in people/sq. km.)
Observations
Monthly mean temperature (in Celcius)
Observations
Monthly mean minimum temperature (in Celcius)
Observations
Monthly mean maximum temperature (in Celcius)
Observations
Extreme minimum temperature (in Celcius)
Observations
Extreme maximum temperature (in Celcius)
Observations
Total precipitation (in mm.)
Observations
Extreme maximum daily precipitation (in mm.)
Observations
Standard errors in parentheses
15
(1)
Inverse
distance
weighted
7.29
(0.03)
6246
22.22
(0.09)
6246
16.06
(0.05)
6246
31.37
(0.17)
6246
276718.36
(2311.90)
6246
51884.00
(504.22)
6246
33807.15
(342.87)
6246
73115.15
(476.85)
6246
7023.96
(44.38)
6246
164604.48
(797.99)
6246
1132582.70
(4040.34)
6246
583.49
(3.98)
6246
28.71
(0.02)
6246
23.94
(0.03)
6246
33.38
(0.02)
6246
21.35
(0.04)
6246
35.87
(0.02)
6246
134.27
(1.30)
6246
38.82
(0.30)
6246
(2)
Within 10
km. radius
(3)
Within 50
km. radius
(4)
Within 100
km. radius
(5)
HYSPLIT
weighted
7.32
(0.03)
6246
22.20
(0.09)
6246
16.13
(0.05)
6246
31.35
(0.17)
6246
834304.94
(10448.62)
6246
174617.83
(2485.20)
6246
116481.74
(1641.15)
6246
162507.28
(1427.94)
6246
18493.90
(162.73)
6246
274841.22
(2412.53)
6246
2312455.40
(20995.09)
6246
1006.32
(15.76)
6246
28.88
(0.02)
6246
24.22
(0.03)
6246
33.45
(0.02)
6246
21.61
(0.04)
6246
35.94
(0.02)
6246
136.36
(1.42)
6246
39.13
(0.34)
6246
7.35
(0.03)
6246
22.31
(0.09)
6246
16.20
(0.05)
6246
31.53
(0.17)
6246
451229.48
(3865.11)
6246
84765.49
(845.08)
6246
56451.42
(596.11)
6246
123397.34
(933.38)
6246
11756.48
(100.67)
6246
242340.38
(1966.35)
6246
1436133.27
(7419.77)
6246
412.24
(4.68)
6246
28.98
(0.02)
6246
24.35
(0.03)
6246
33.51
(0.02)
6246
21.74
(0.04)
6246
35.91
(0.02)
6246
130.69
(1.43)
6246
38.71
(0.37)
6246
7.33
(0.03)
6246
22.28
(0.09)
6246
16.16
(0.05)
6246
31.47
(0.17)
6246
293308.21
(2163.51)
6246
51363.66
(407.15)
6246
33044.12
(282.21)
6246
92671.27
(693.98)
6246
8440.60
(72.99)
6246
203154.78
(1461.48)
6246
1065502.51
(3435.57)
6246
229.21
(2.02)
6246
29.03
(0.02)
6246
24.38
(0.03)
6246
33.59
(0.02)
6246
21.79
(0.04)
6246
35.97
(0.02)
6246
128.49
(1.40)
6246
38.35
(0.35)
6246
7.30
(0.03)
6246
22.27
(0.09)
6246
16.11
(0.05)
6246
31.44
(0.17)
6246
345822.07
(5095.65)
6246
65838.90
(1148.85)
6246
44307.89
(782.44)
6246
89068.88
(950.73)
6246
8568.04
(96.29)
6246
184147.79
(1551.76)
6246
1290381.18
(9776.02)
6246
658.65
(7.97)
6246
28.75
(0.02)
6246
24.11
(0.03)
6246
33.30
(0.02)
6246
21.53
(0.04)
6246
35.81
(0.02)
6246
145.87
(1.54)
6246
40.76
(0.35)
6246
Figure 5: Marginal effect of CNG price on air pollution
price will increase air pollution concentrations. Figure 5 summarizes the effect of CNG price on
air pollution with price levels and fuel availability at the end of 2014. We see that marginal effects
with HYSPLIT weights are more negative, but overall results are very inconsistent. An increase in
CNG price is correlated with a decrease in O3 and PM10 concentrations which does not match our
hypothesis.
7
Conclusion
This paper have built on existing literature on the effects of CNG adoption and air pollution in
multiple ways. Most notably, we have combined air pollution modeling with econometrics estimation which corrects for meteorological biases. We controlled for other sources of air pollution
using data from individual fueling stations and factories. The effect of CNG adoption is also being
analyzed with five types of pollutants.
We see that an increase in CNG availability improves air quality, but its price effect are unclear.
As selection bias for air pollution monitors and set up of CNG gas station in high polluted area
is not yet controlled for, we postulate that that CNG availability contributes to even more air
pollution reduction.
Because the Thai government explicitly states that CNG have been introduced to improve
air quality, CNG fueling stations may be set up in high polluted areas. This selection bias issue
16
could cause a positive bias in the CNG fueling station coefficient. To control for this, we use an
instrument. We exploit the fact that PTT states that it will distribute CNG to fueling stations
that are close to CNG pipelines, next to major roads and highways, with no other CNG fueling
stations nearby.
We claim that an interaction effect between a natural gas pipeline and major highways satisfies
the exclusion restriction because only CNG stations could only be affected by both factors. Factories
that convert to natural gas are influenced by natural gas pipelines, but not major roads and
highways. Alternatively, LPG and regular gas stations should be influenced by placements of
highways but not natural gas pipelines.
Although we can conclude that an increase in CNG availability has contributed to cleaner air,
the effect of CNG price on air quality is unclear. If prices do not affect consumer decisions to use
CNG, then it is beneficial that CNG subsidies are removed. More work has to be done, especially
on the mechanisms involving fuel price, car price, and fuel adoption.
17
8
References
1. Anderson, Soren T. and Sallee, James M. June 2011. Using Loopholes to Reveal the
Marginal Cost of Regulation: The Case of Fuel-Economy Standards" American Economic
Review, 1375-1409
2. Auffhammer, M. and Kellogg R. 2011. "Clearing the Air? The Effects of Gasoline
Content Regulation and Air Quality". American Economic Review 101:2687-2722.
3. Coxhead, I. and Grainger C. 2014. "The incidence of energy policy reform: fossil fuel
subsidies in Southeast Asia". Working paper
4. Dholakia, H. H., Purohit, P., Rao, S., and Garg, A. 2013. "Impact of current policies
on future air quality and health outcomes in Delhi, India". Atmospheric Environment 75,
241-248.
5. Draxler, R.R., and G.D. Hess. 1997. "Description of the HYSPLIT4 modeling system".
NOAA Tech. Memo. ERL ARL-224, NOAA Air Resources Laboratory, Silver Spring, MD,
24 pp.
6. Draxler, R.R., and G.D. Hess. 1998. "An overview of the HYSPLIT4 modeling system
of trajectories, dispersion, and deposition". Aust. Meteor. Mag., 47, 295-308.
7. Draxler, R.R 1999. "HYSPLIT4 user’s guide". NOAA Tech. Memo. ERL ARL-230, NOAA
Air Resources Laboratory, Silver Spring, MD.
8. Du, Xiaodong and Li, Shanjun. March 23, 2015. "Flexible-Fuel Vehicle Adoption and the
U.S. Biofuel Market", https://ssrn.com/abstract=2583808 or http://dx.doi.org/10.2139/ssrn.2583808
9. Energy Policy and Planning Office 2001. "Natural Gas Price Manual". http://www.eppo.go.th/
petro/pt-KPC2544-01.html.
10. Greene, David L. 1997. "Survey Evidence on the Importance of Fuel Availability to Choice
of Alternative Fuels and Vehicles" Oak Ridge National Laboratory
11. Houde, S. 2008. "Spatial differentiation in retail markets for gasoline". American Economic
Review. 102(5), 1-47.
18
12. Krupnick, A. J. 2011. "Will Natural Gas Vehicles Be in Our Future?" Retrieved April 5,
2015, from http://www.rff.org/Publications/Resources/Pages/181-Natural-Gas-Vehicles.aspx
13. Kumar, N., and Foster, A. D. 2007. "Have NGV Regulations in Delhi Done Their Job?"
Economic and Political Weekly. 42(51), 48-58. Retrieved from http://www.jstor.org/stable/40276877
14. Li, S., Xing, J., Tong, L., and Yiyi Z. 2015. "The Market for Electric Vehicles: Indirect
Network Effects and Policy Design". Working paper
15. Manuszak, Mark D. and Moul, Charles C. November 2009. "How Far for a Buck?
Tax Differences and the Location of Retail Gasoline Activity in Southeast Chicagoland" The
Review of Economics and Statistics, 59 (2010) 91(4): 744-765
16. Narain, Urvashi, and Alan Krupnick 2007. "The Impact of Delhi’s CNG Program on
Air Pollution". Resources for the Future Discussion Paper
c
17. Ravindra, Khaiwal, Eric Wauters, Sushil K. Tyagi, Suman Mor, and RenÃ
Van Grieken 2006. "Assessment of Air Quality after the Implementation of Compressed
Natural Gas (CNG) as Fuel in Public Transport in Delhi, India". Environmental Monitoring
and Assessment
18. Schifter, I., Diaz, L., Lopez-Salinas, E., and Avalos, S. 2000. "Potential Impacts of
Compressed Natural Gas in the Vehicular Fleet of Mexico City". Environmental Science &
Technology 34(11), 2100-2104.
19. Schlenker, Wolfram and Walker, Reed W. 2016. "Airports, Air Pollution and Contemporaneous Health" Review of Economic Studies 83 (2016), 768-809
20. Sullivan, Daniel M. 2016. "Residential Sorting and the Incidence of Local Public Goods:
Theory and Evidence from Air Pollution" Resources for the Future Working Paper
21. Suthawaree, Jeeranut, Helena Akhter Sikder, Charlotte Emily Jones, Shungo
Kato, Hitoshi Kunimi, Abu Naser Mohammed Hamidul Kabir, and Yoshizumi
Kajii "Influence of Extensive Compressed Natural Gas (CNG) Usage on Air Quality". Atmospheric Environment
22. Wadud, Z., and Khan, T. 2013. "Air Quality and Climate Impacts Due to NGV Conversion
of Motor Vehicles in Dhaka , Bangladesh".Environmental Science & Technology
19
23. Yusuf, A. A., and Resosudarmo, B. P. 2010. "Is Reducing Subsidies on Vehicle Fuel
Equitable? A Lesson from Indonesian Reform Experience". In Fuel Taxes and the Poor:
The Distributional Effects of Gasoline Taxation and Their Implications for Climate Policy
(pp.171-180).
20