Bunching for Free Electricity - pier.or.th · Marginal and average cost Introduction Bunching for...
Transcript of Bunching for Free Electricity - pier.or.th · Marginal and average cost Introduction Bunching for...
Bunching for Free Electricity
Tosapol Apaitan1, Thiti Tosborvorn2, Wichsinee Wibulpolprasert3
April 26, 2018
1Bank of Thailand, Email: [email protected] of Thailand, Email: [email protected] Development Research Institute, Email: [email protected].
Challenges in studying households’ electricity consumption
• Residential sector is the third largest consumer of electricity inThailand
• Yet, electricity consumption behaviors of the Thai households arerarely studied
• Understanding the factors that affect household electricityconsumption will help inform policy on the following dimensions:
1 Welfare impact of electricity price change2 Program design to encourage energy efficiency and conservation
Introduction Bunching for Free Electricity
This paper
1 Characterize stylized facts on residential electricity consumption
2 Explore the consumption response to exogenous price change createdby the Thailand’s Free Basic Electricity program
• Free if usage is under threshold, pay for everything if above• Very large exogenous price jump (“notch”)
• Research questions:• How do consumer respond to a financial notch created by the FBE?• What can the behavioral response reveal about price elasticity?• Policy implications?
Introduction Bunching for Free Electricity
A quick stylized facts on residential electricity consumptionin Thailand
1 In most region of Thailand, residential consumption constitute amajor share of total electricity consumption
2 Residential consumption is highly seasonal with a peak in May
3 Inequality in residential consumption is also seasonal with a peak inMay
4 Residential consumption is highly correlated with householdexpenditure
5 Residential electricity consumption is relatively unresponsive to pricechanges
Introduction Bunching for Free Electricity
Stylized fact #1: Residential electricity consumptiondominates in most of Thailand
(a) RES consumption share (b) LGS consumption share
Introduction Bunching for Free Electricity
Stylized fact #2: Residential consumption is highlyseasonal with a peak in May
500
600
700
800
900
GW
h
2002
m1
2004
m1
2006
m1
2008
m1
2010
m1
2012
m1
2014
m1
2016
m1
2018
m1
YM
Small RES
1,00
01,
500
2,00
02,
500
3,00
03,
500
GW
h
2002
m1
2004
m1
2006
m1
2008
m1
2010
m1
2012
m1
2014
m1
2016
m1
2018
m1
YM
Large RES
Introduction Bunching for Free Electricity
Stylized fact #3: Inequality in residential consumption isalso seasonal with a peak in May
Figure: Ratio between 90th percentile and 10th percentile consumption
• Consumption in May reflects inequality in electrical appliances(implies wealth inequality)
Introduction Bunching for Free Electricity
Stylized fact #4: Residential consumption is highlycorrelated with household expenditure
(a) Electricity expenditure (THB) (b) Electricity consumption (kWh)
Introduction Bunching for Free Electricity
Stylized fact #5: Residential consumption is relativelyunresponsive to price change
1 Price elasticity estimates from a panel data regression: -0.03 to -0.06
2 Price elasticity estimates from the observed bunching response areeven lower: -0.01 to -0.03
The remaining of the presentation will focus on consumers’response to price changes.
Introduction Bunching for Free Electricity
Data
• Meter-level monthly billing data from the Provincial ElectricityAuthority (PEA)
• PEA account for more than 70% of total consumption in Thailand
• Analysis period is from January 2012 to December 2015
Introduction Bunching for Free Electricity
Thailand’s Free Basic Electricity (FBE) Program
• Main objective of the FBE: subsidize cost of living for lower incomehouseholds
• Full subsidy for those who are eligible, none for others
2008 2009 2010 2011 2012 2013 2014 2015 2016 2017
Aug ’08 - 80 units, 50% discount for 81-150 units
Feb ’09 - 90 units Jul ’11 - 90 units
Jun ’12 - 50 units Jan ’16 - 50 units
Start of FBE Program
Residential Small ResidentialSmall ResidentialNon-business
2018
Data coverage (2012-2015)
Introduction Bunching for Free Electricity
FBE creates a large marginal price jump at the threshold
130.7326
050
100
150
THB/
kWh
0 100 200 300 400 500kWh
marginal cost average costmarginal cost at 51st unit
Marginal and average cost
Introduction Bunching for Free Electricity
The FBE incentive should lead to two types of responses
1 Bunching from below: everybody who used to consume below q̄should move up to q̄
• unable to quantify with the current data• likely to be small since electricity consumption is constrained by the
stock of appliance
2 Bunching from above: those between q̄ and “marginal buncher”who used to consume at q∗ should now consume q̄
Introduction Bunching for Free Electricity
Descriptive evidence #1: FBE beneficiaries areconcentrated in the North and Northeast region and isseasonal
(a) Fraction of FBE-eligible meterthat received benefit Jan 2015 (b) Seasonality of FBE benefits
0.2
.4.6
.8Fr
actio
n re
ceiv
ed F
BE
01
23
4N
umbe
r rec
eive
d FB
E (m
illion
s)
2012m1 2013m1 2014m1 2015m1 2016m1year-month
Number received FBE Fraction received FBE
Data Bunching for Free Electricity
Descriptive evidence #2: Obvious bunching and missingmass
(a) 2012 (90 units free)
010
020
030
0N
umbe
r of m
eter
s (th
ousa
nds)
0 50 100 150Consumption bin
(b) 2013 (50 units free)
010
020
030
0N
umbe
r of m
eter
s (th
ousa
nds)
0 50 100 150Consumption bin
(c) 2014 (50 units free)
010
020
030
0N
umbe
r of m
eter
s (th
ousa
nds)
0 50 100 150Consumption bin
(d) 2015 (50 units free)
010
020
030
0N
umbe
r of m
eter
s (th
ousa
nds)
0 50 100 150Consumption bin
Data Bunching for Free Electricity
Descriptive evidence #3: Bunching is difficultDistribution of the number of months when a meter consumes at thenotch point or receives free electricity in 2013
Number of months Exact bunching Free electricityin 2013 No. meters Percent No. meters Percent
0 3,718,881 86.71 2,001,171 46.661 465,009 10.84 353,102 8.232 85,392 1.99 253,213 5.93 14,978 0.35 193,318 4.514 2,941 0.07 157,233 3.675 889 0.02 131,847 3.076 391 0.01 120,215 2.87 213 0 116,329 2.718 139 0 117,376 2.749 91 0 124,703 2.91
10 52 0 142,400 3.3211 45 0 175,133 4.0812 33 0 403,014 9.4
Total 4,289,054 100 4,289,054 100
Data Bunching for Free Electricity
Descriptive evidence #4: Bunching increases with financialincentive and FBE program awareness
(a) 90 units free (2012)
010
020
030
0N
umbe
r of m
eter
s (th
ousa
nds)
0 50 100 150Consumption bin
(b) 50 units free (2013)
010
020
030
0N
umbe
r of m
eter
s (th
ousa
nds)
0 50 100 150Consumption bin
(c) Rural area (2013)
050
100
150
200
250
300
Num
ber o
f met
ers
(thou
sand
s)
0 50 100 150Consumption bin
(d) Urban area (2013)
050
100
150
200
250
300
Num
ber o
f met
ers
(thou
sand
s)
0 50 100 150Consumption bin
Data Bunching for Free Electricity
Descriptive evidence #5: Bunching varies across monthand exhibits learning
020
000
4000
060
000
8000
0N
umbe
r of m
eter
s
0 50 100 150consumption (kWh)
1, 2012 1, 20131, 2014 1, 2015
Data Bunching for Free Electricity
Descriptive evidence #5: Bunching varies across monthand exhibits learning
020
000
4000
060
000
8000
0N
umbe
r of m
eter
s
0 50 100 150consumption (kWh)
2, 2012 2, 20132, 2014 2, 2015
Data Bunching for Free Electricity
Descriptive evidence #5: Bunching varies across monthand exhibits learning
020
000
4000
060
000
8000
0N
umbe
r of m
eter
s
0 50 100 150consumption (kWh)
3, 2012 3, 20133, 2014 3, 2015
Data Bunching for Free Electricity
Descriptive evidence #5: Bunching varies across monthand exhibits learning
020
000
4000
060
000
8000
0N
umbe
r of m
eter
s
0 50 100 150consumption (kWh)
4, 2012 4, 20134, 2014 4, 2015
Data Bunching for Free Electricity
Descriptive evidence #5: Bunching varies across monthand exhibits learning
020
000
4000
060
000
8000
0N
umbe
r of m
eter
s
0 50 100 150consumption (kWh)
5, 2012 5, 20135, 2014 5, 2015
Data Bunching for Free Electricity
Descriptive evidence #5: Bunching varies across monthand exhibits learning
020
000
4000
060
000
8000
0N
umbe
r of m
eter
s
0 50 100 150consumption (kWh)
6, 2012 6, 20136, 2014 6, 2015
Data Bunching for Free Electricity
Descriptive evidence #5: Bunching varies across monthand exhibits learning
020
000
4000
060
000
8000
0N
umbe
r of m
eter
s
0 50 100 150consumption (kWh)
7, 2012 7, 20137, 2014 7, 2015
Data Bunching for Free Electricity
Descriptive evidence #5: Bunching varies across monthand exhibits learning
020
000
4000
060
000
8000
0N
umbe
r of m
eter
s
0 50 100 150consumption (kWh)
8, 2012 8, 20138, 2014 8, 2015
Data Bunching for Free Electricity
Descriptive evidence #5: Bunching varies across monthand exhibits learning
020
000
4000
060
000
8000
0N
umbe
r of m
eter
s
0 50 100 150consumption (kWh)
9, 2012 9, 20139, 2014 9, 2015
Data Bunching for Free Electricity
Descriptive evidence #5: Bunching varies across monthand exhibits learning
020
000
4000
060
000
8000
0N
umbe
r of m
eter
s
0 50 100 150consumption (kWh)
10, 2012 10, 201310, 2014 10, 2015
Data Bunching for Free Electricity
Descriptive evidence #5: Bunching varies across monthand exhibits learning
020
000
4000
060
000
8000
0N
umbe
r of m
eter
s
0 50 100 150consumption (kWh)
11, 2012 11, 201311, 2014 11, 2015
Data Bunching for Free Electricity
Descriptive evidence #5: Bunching varies across monthand exhibits learning
020
000
4000
060
000
8000
0N
umbe
r of m
eter
s
0 50 100 150consumption (kWh)
12, 2012 12, 201312, 2014 12, 2015
Data Bunching for Free Electricity
Descriptive evidence #6: Bunching comes from above andbelow the threshold
Data Bunching for Free Electricity
Linking the observed bunching to price elasticityThe FBE creates a notch in the consumer’s budget set
Model and Estimation Bunching for Free Electricity
Linking the observed bunching to price elasticity
• Two goods: electricity (qi ) and numeraire (ηi )
• Parameters: taste for electricity (αi ) and price elasticity (e)
• Isoelastic quasi-linear utility function
U(qi ) =αi
1 + 1/e
(qiαi
)1+1/e
+ ηi
• Budget constraint
I ≥
{pqi + ηi if qi > q̄,
ηi if qi ≤ q̄
Model and Estimation Bunching for Free Electricity
Linking the observed bunching to price elasticity (cont.)
• Interior solution: qi = αipe
• Marginal buncher is indifferent between consuming q∗ and q̄
• This gives us the relationship between consumption response andprice elasticity:
(−e)e
e+1 =q∗
q̄
Intuition: bunching end point (q∗) reflects price elasticity
Model and Estimation Bunching for Free Electricity
Estimating bunching and price elasticity
Use the method outlined in Kleven and Waseem (2013) to estimate q∗
and excess bunching
• Pick an excluded region [zl , q∗]
• Approximate the distribution using a polynomial of degree r over thenon-excluded region
Nj =r∑
i=0
βi (zj)i +
q∗∑i=zl
γi I [zj = i ] + νj (1)
• Calculate counterfactual (predicted) distribution:
N̂j =r∑
i=0
β̂i (zj)i . (2)
Model and Estimation Bunching for Free Electricity
Estimating bunching and price elasticity
• Solve for the ending bin q∗ using iterative process that equate theexcess mass with the missing mass
• Calculate structural elasticity e using expression from earlier
1000
0015
0000
2000
0025
0000
3000
00
30 40 50 60 70 80b
actual frequency predicted frequency
• q∗ represents the consumption response of the highest-elasticityconsumer → the upper bound of the average elasticity
Model and Estimation Bunching for Free Electricity
Estimation results: Overall
1 “End Ponit” = q∗ or the largest consumption response observed inthe data
2 “Excess Bunching” = #actual−#counterfactual#counterfactual
Results and Discussions Bunching for Free Electricity
Estimation results: Overall
Year End Point Excess Bunching Estimated Elasticity
2012 95.35∗ 0.115∗ −0.013∗
(1.48) (0.004) (0.005)
2013 53.90∗ 0.056∗ −0.018(2.52) (0.003) (0.018)
2014 56.05∗ 0.061∗ −0.032∗
(1.36) (0.003) (0.010)
2015 54.70∗ 0.060∗ −0.023(1.97) (0.004) (0.014)
1 Standard errors are in parentheses. Standard errors are calculatedusing bootstrapping with 500 replications. Data is limited to January–May of every year.
2 The ∗ indicates statistical significance at the 5% level.
Results and Discussions Bunching for Free Electricity
Estimation results: Urban vs. Rural
Group Year End Point Excess Bunching Estimated Elasticity
Rural Area 2013 52.52∗ 0.019∗ −0.011(2.15) (0.003) (0.015)
2014 52.83∗ 0.022∗ −0.012(2.78) (0.003) (0.019)
2015 52.99∗ 0.020 −0.013∗
(3.11) (0.046) (0.021)
Urban Area 2013 53.03∗ 0.077∗ −0.013∗
(0.37) (0.004) (0.002)2014 53.29∗ 0.081∗ −0.015∗
(0.99) (0.004) (0.007)2015 53.01∗ 0.068∗ −0.013∗
(0.79) (0.005) (0.005)
1 Standard errors are in parentheses. Standard errors are calculated usingbootstrapping with 500 replications. Data is limited to January–May ofevery year.
2 The ∗ indicates statistical significance at the 5% level.
Results and Discussions Bunching for Free Electricity
Summary
• The FBE triggers obvious excess bunching
• However, bunching implies small underlying price elasticities
• Small excess bunching and elasticity are due to various sources offrictions:
• Consumers face uncertainty on consumption shocks and/or cannotkeep track of their cumulative consumption
• Consumers are not aware of the FBE incentive and/or ways to saveenergy
Next steps:
1 How to quantify bunching from below?
2 How to quantify or explicitly model the sources of frictions?
3 Using learning dynamic to confirm the role of consumer’s awareness?
Results and Discussions Bunching for Free Electricity
Policy implications
From the Perspective of FBE Program Administrator:
• FBE aims to redistribute wealth to the low-income households
• Bunching response represents a “leakage” of the subsidy→ Bad for the policy
• Although evidence shows bunching, the degree is very small
• Advantage of using necessity good as a threshold
Results and Discussions Bunching for Free Electricity
Policy implications
From the Perspective of Electricity Pricing Design:
• Using financial incentive to encourage conservation might not bepractical because would require an unrealistically large financialincentive
• Roles for other interventions to remove frictions...• Information campaign on how to save energy• Smart app/meter that lets people keep track of their energy usage• Behavioral nudges to make consumers more aware of their energy
usage
Results and Discussions Bunching for Free Electricity
Thank you
Backup Slides
Estimation Results: OverallActual and counterfactual densities (January–December of each year)
(a) 2012 (50 units free)10
0000
1500
0020
0000
2500
0030
0000
60 70 80 90 100 110b
actual frequency predicted frequency
(b) 2013 (50 units free)
1000
0015
0000
2000
0025
0000
3000
00
30 40 50 60 70 80b
actual frequency predicted frequency
(c) 2014 (50 units free)
1500
0020
0000
2500
0030
0000
30 40 50 60 70 80b
actual frequency predicted frequency
(d) 2015 (50 units free)
1500
0020
0000
2500
0030
0000
30 40 50 60 70 80b
actual frequency predicted frequency
Backup Slides Bunching for Free Electricity
Estimation Results: Overall
Year End Point Excess Bunching Estimated Elasticity
January – December
2013 53.81∗ 0.058∗ −0.018(2.06) (0.003) (0.014)
2014 54.10∗ 0.060∗ −0.020∗
(1.41) (0.004) (0.010)
2015 53.87∗ 0.052∗ −0.018(2.39) (0.005) (0.017)
1 Standard errors are in parentheses. Standard errors are calculated usingbootstrapping with 500 replications. Data is limited to January–Decemberof every year.
2 The ∗ indicates statistical significance at the 5% level.
Backup Slides Bunching for Free Electricity
Estimation Results: by urbanization status
Bunching is more obvious among consumers in the urban area, who arebetter informed about the FBE program
(a) Rural area (2013)
1000
0015
0000
2000
0025
0000
30 40 50 60 70 80b
actual frequency predicted frequency
(b) Urban area (2013)
1500
0020
0000
2500
0030
0000
30 40 50 60 70 80b
actual frequency predicted frequency
Backup Slides Bunching for Free Electricity
Discussions
• Without adjustment cost, there should not be any consumers belowthe threshold (i.e. “strictly dominated region”)
• In reality, we still observe a mass of consumers under the threshold→ the presence of optimization friction or adjustment cost
• However, we cannot quantify the friction since the strictly dominatedregion is so wide that we cannot credibly estimate the counterfactualdistribution using the cross-sectional data
• Moreover, we lack the pre-FBE data so we cannot use that as acounterfactual distribution either
Backup Slides Bunching for Free Electricity
Manipulated by meter surveyors?
• Unlikely
• Meter surveys are done by noting the cumulative usage, so need tonote last month’s number as well
• Meter surveyors are rotated every few months, unlikely to colludewith households
• Since meter surveys are done cumulatively, if there’s collusion, it’smore likely that consumption in the month after consuming at thethreshold would be higher
Backup Slides Bunching for Free Electricity
020
4060
8010
0q_
0 20 40 60 80 100q_before
Backup Slides Bunching for Free Electricity