Sherwin - Center for Climate and Energy Decision Making

USING SMART-METER DATA TO UNDERSTAND THE
CONSEQUENCES OF LOW INCOME ELECTRICITY RATES
By Evan D. Sherwin
PhD Student, Engineering and Public Policy
MS Student, Machine Learning
Carnegie Mellon University
With: Inês M. L. Azevedo
LOAD SHAPE EFFECTS
2
LOAD SHAPE EFFECTS
• Electricity must enter the grid at the moment it is
consumed
3
LOAD SHAPE EFFECTS
• Electricity must enter the grid at the moment it is
consumed
• Changes in peak consumption dictate need for
electricity transmission, distribution, and peak
generation capacity
4
LOAD SHAPE EFFECTS
• Electricity must enter the grid at the moment it is
consumed
• Changes in peak consumption dictate need for
electricity transmission, distribution, and peak
generation capacity
• A better understanding load shape can inform
allocation of infrastructure resources
5
LOAD SHAPE EFFECTS
• Electricity must enter the grid at the moment it is
consumed
• Changes in peak consumption dictate need for
electricity transmission, distribution, and peak
generation capacity
• A better understanding load shape can inform
allocation of infrastructure resources
• Some evaluation of load shape effects of energy
efficiency (1) and behavioral and price (2) interventions
6
LOAD SHAPE EFFECTS
• Electricity must enter the grid at the moment it is
consumed
• Changes in peak consumption dictate need for
electricity transmission, distribution, and peak
generation capacity
• A better understanding load shape can inform
allocation of infrastructure resources
• Some evaluation of load shape effects of energy
efficiency (1) and behavioral and price (2) interventions
• Numerous customer-facing utility programs likely have
load shape effects, intended or not
7
LOW-INCOME ELECTRICITY SUBSIDIES
8
LOW-INCOME ELECTRICITY SUBSIDIES
• Millions of households receive low-income energy
assistance
9
LOW-INCOME ELECTRICITY SUBSIDIES
• Millions of households receive low-income energy
assistance
• California Alternate Rates for Energy (CARE) provides
~4.5 million households (~1/3 of households) with an
average electric bill discount of 33% (3)
10
LOW-INCOME ELECTRICITY SUBSIDIES
• Millions of households receive low-income energy
assistance
• California Alternate Rates for Energy (CARE) provides
~4.5 million households (~1/3 of households) with an
average electric bill discount of 33% (3)
• Average discount is 42% in Pacific Gas and Electric territory
11
LOW-INCOME ELECTRICITY SUBSIDIES
• Millions of households receive low-income energy
assistance
• California Alternate Rates for Energy (CARE) provides
~4.5 million households (~1/3 of households) with an
average electric bill discount of 33% (3)
• Average discount is 42% in Pacific Gas and Electric territory
• Directly subsidizes electricity and gas prices for
households within 200% of the poverty line
• Aims to ensure energy services are affordable
12
HOW DO SUBSIDES AFFECT LOAD SHAPE?
13
HOW DO SUBSIDES AFFECT LOAD SHAPE?
• How does enrollment in CARE affect overall
electricity consumption and load shape?
14
HOW DO SUBSIDES AFFECT LOAD SHAPE?
• How does enrollment in CARE affect overall
electricity consumption and load shape?
• How do these results vary by region?
15
HOW DO SUBSIDES AFFECT LOAD SHAPE?
• How does enrollment in CARE affect overall
electricity consumption and load shape?
• How do these results vary by region?
• Hypothesis:
• Modest increase in electricity consumption
• Largest increase after ~6pm
16
DATA FROM PG&E
From Pacific Gas and Electric, via the Wharton Customer Analytics Initiative:
Source: (4)
17
DATA FROM PG&E
From Pacific Gas and Electric, via the Wharton Customer Analytics Initiative:
1. Smart-meter
data:
~30,000 customers
from 2008 to 2011
with 15-min interval
readings
Source: (4)
18
DATA FROM PG&E
From Pacific Gas and Electric, via the Wharton Customer Analytics Initiative:
1. Smart-meter
data:
~30,000 customers
from 2008 to 2011
with 15-min interval
readings
2. Program
participation:
Household
participation in
programs & energy
efficiency rebates
Source: (4)
19
DATA FROM PG&E
From Pacific Gas and Electric, via the Wharton Customer Analytics Initiative:
1. Smart-meter
data:
~30,000 customers
from 2008 to 2011
with 15-min interval
readings
2. Program
participation:
Household
participation in
programs & energy
efficiency rebates
3. Household
location defined by
Census Block
Source: (4)
20
DATA FROM PG&E
From Pacific Gas and Electric, via the Wharton Customer Analytics Initiative:
1. Smart-meter
data:
~30,000 customers
from 2008 to 2011
with 15-min interval
readings
2. Program
participation:
Household
participation in
programs & energy
efficiency rebates
3. Household
location defined by
Census Block
We complement this dataset with:
5. Demographic information
at the census block (2010)
demographic information
(median income, median home
value, % renters, % poor,
education)
Source: (4)
21
DATA FROM PG&E
From Pacific Gas and Electric, via the Wharton Customer Analytics Initiative:
1. Smart-meter
data:
~30,000 customers
from 2008 to 2011
with 15-min interval
readings
2. Program
participation:
Household
participation in
programs & energy
efficiency rebates
3. Household
location defined by
Census Block
We complement this dataset with:
5. Demographic information
at the census block (2010)
demographic information
(median income, median home
value, % renters, % poor,
education)
6. Weather data
Daily temperature data from
NOAA (daily high and low
temperatures)
Source: (4)
22
DATA FROM PG&E
From Pacific Gas and Electric, via the Wharton Customer Analytics Initiative:
1. Smart-meter
data:
~30,000 customers
from 2008 to 2011
with 15-min interval
readings
2. Program
participation:
Household
participation in
programs & energy
efficiency rebates
3. Household
location defined by
Census Block
4. Natural gas
consumption (daily)
We complement this dataset with:
5. Demographic information
at the census block (2010)
demographic information
(median income, median home
value, % renters, % poor,
education)
6. Weather data
Daily temperature data from
NOAA (daily high and low
temperatures)
Source: (4)
23
SMART METER ROLLOUT
Coast
Inland Hills
Central Valley
Climate Zones
Coast
Inland Hills
Central Valley
Source: (4, 5, 6)
24
SMART METER ROLLOUT
Central Valley
Inland Hills
Median
income
52,000
79,000
Median %
poor
12
6
% with AC
~95%
~60%
# of
households
8,597
Central
11,391
63,000
Coast
9
Inland Hills
Climate Zones
Coast
Inland Hills
Central Valley
Source: (4, 5, 6)
25
Coast
~15%
Valley
10,217
QUASI-EXPERIMENTAL APPROACH
26
QUASI-EXPERIMENTAL APPROACH
• Treatment group: Active CARE participants (N ≅ 10,000)
27
QUASI-EXPERIMENTAL APPROACH
• Treatment group: Active CARE participants (N ≅ 10,000)
• Control group: Everyone not actively in CARE (N ≅ 25,000)
28
QUASI-EXPERIMENTAL APPROACH
• Treatment group: Active CARE participants (N ≅ 10,000)
• Control group: Everyone not actively in CARE (N ≅ 25,000)
• Flat time trend -> Fixed effects equivalent to diff-in-diff
29
QUASI-EXPERIMENTAL APPROACH
•
•
•
•
Treatment group: Active CARE participants (N ≅ 10,000)
Control group: Everyone not actively in CARE (N ≅ 25,000)
Flat time trend -> Fixed effects equivalent to diff-in-diff
Hourly regressions:
ln 𝑘𝑊ℎ𝑖,𝑡,ℎ = 𝛼 + 𝛽𝑗 𝑇𝑒𝑚𝑝𝑖,𝑡
𝑗
+ 𝛾 𝐶𝐴𝑅𝐸𝑖,𝑡 + 𝛿𝑘 𝑇𝑖𝑚𝑒𝑡
𝜁 𝑇𝑖𝑚𝑒𝑇𝑟𝑒𝑛𝑑𝑡 + 𝜑𝑞 𝑃𝑟𝑜𝑔𝑟𝑎𝑚𝑖,𝑡
30
𝑞
+ 𝑢𝑖 + 𝜀𝑖,𝑡
𝑘
+
LIMITATIONS
• Selection on omitted variables
31
LIMITATIONS
• Selection on omitted variables
• CARE eligibility criteria depend on income and household size
32
LIMITATIONS
• Selection on omitted variables
• CARE eligibility criteria depend on income and household size
• Eligible households can get free energy efficiency measures
through the Energy Saving Assistance program
33
LIMITATIONS
• Selection on omitted variables
• CARE eligibility criteria depend on income and household size
• Eligible households can get free energy efficiency measures
through the Energy Saving Assistance program
• Time period coincident with Financial Crisis and Great
Recession
34
LIMITATIONS
• Selection on omitted variables
• CARE eligibility criteria depend on income and household size
• Eligible households can get free energy efficiency measures
through the Energy Saving Assistance program
• Time period coincident with Financial Crisis and Great
Recession
• Deployment of smart meters may be non-random,
conditional on region, in important ways
• The fraction of CARE enrollees over time matches independent
population statistics
• Unbalanced panel concerns
35
HOURLY EFFECTS OF CARE
Off-peak
36
Partial
peak
On-peak
Partial
peak
Offpeak
HOURLY EFFECTS OF CARE
Off-peak
37
Partial
peak
On-peak
Partial
peak
Offpeak
Off-peak
Source: (7)
Partial
peak
On-peak
Partial
peak
Offpeak
HOURLY EFFECTS OF CARE
Off-peak
38
Partial
peak
On-peak
Partial
peak
Offpeak
Off-peak
Source: (7)
Partial
peak
On-peak
Partial
peak
Offpeak
HOURLY EFFECTS OF CARE
Off-peak
39
Partial
peak
On-peak
Partial
peak
Offpeak
Off-peak
Source: (7)
Partial
peak
On-peak
Partial
peak
Offpeak
HOURLY EFFECTS OF CARE
Off-peak
40
Partial
peak
On-peak
Partial
peak
Offpeak
Off-peak
Source: (7)
Partial
peak
On-peak
Partial
peak
Offpeak
EFFECTS ACROSS TIME AND SPACE
41
EFFECTS ACROSS TIME AND SPACE
• 40% bill savings -> 13% increase in demand
42
EFFECTS ACROSS TIME AND SPACE
• 40% bill savings -> 13% increase in demand
• Implicit price elasticity of demand of -0.33 is close to literature
average for short-run, -0.35, well below long-run average, -0.85 (8)
43
EFFECTS ACROSS TIME AND SPACE
• 40% bill savings -> 13% increase in demand
• Implicit price elasticity of demand of -0.33 is close to literature
average for short-run, -0.35, well below long-run average, -0.85 (8)
• No pair of hours statistically distinguishable
44
EFFECTS ACROSS TIME AND SPACE
• 40% bill savings -> 13% increase in demand
• Implicit price elasticity of demand of -0.33 is close to literature
average for short-run, -0.35, well below long-run average, -0.85 (8)
• No pair of hours statistically distinguishable
• Constant elasticity throughout the day reasonable
45
EFFECTS ACROSS TIME AND SPACE
• 40% bill savings -> 13% increase in demand
• Implicit price elasticity of demand of -0.33 is close to literature
average for short-run, -0.35, well below long-run average, -0.85 (8)
• No pair of hours statistically distinguishable
• Constant elasticity throughout the day reasonable
• Hourly differences likely resolvable with more data
46
EFFECTS ACROSS TIME AND SPACE
• 40% bill savings -> 13% increase in demand
• Implicit price elasticity of demand of -0.33 is close to literature
average for short-run, -0.35, well below long-run average, -0.85 (8)
• No pair of hours statistically distinguishable
• Constant elasticity throughout the day reasonable
• Hourly differences likely resolvable with more data
• Magnitudes of our coefficients suggest this could be nontrivial
47
EFFECTS ACROSS TIME AND SPACE
• 40% bill savings -> 13% increase in demand
• Implicit price elasticity of demand of -0.33 is close to literature
average for short-run, -0.35, well below long-run average, -0.85 (8)
• No pair of hours statistically distinguishable
• Constant elasticity throughout the day reasonable
• Hourly differences likely resolvable with more data
• Magnitudes of our coefficients suggest this could be nontrivial
• Suggestive evidence that:
• Greatest increases are in the morning and evening, not on peak
48
EFFECTS ACROSS TIME AND SPACE
• 40% bill savings -> 13% increase in demand
• Implicit price elasticity of demand of -0.33 is close to literature
average for short-run, -0.35, well below long-run average, -0.85 (8)
• No pair of hours statistically distinguishable
• Constant elasticity throughout the day reasonable
• Hourly differences likely resolvable with more data
• Magnitudes of our coefficients suggest this could be nontrivial
• Suggestive evidence that:
• Greatest increases are in the morning and evening, not on peak
• Highest effect in Inland Hills: Air conditioning is on the margin
49
EFFECTS ACROSS TIME AND SPACE
• 40% bill savings -> 13% increase in demand
• Implicit price elasticity of demand of -0.33 is close to literature
average for short-run, -0.35, well below long-run average, -0.85 (8)
• No pair of hours statistically distinguishable
• Constant elasticity throughout the day reasonable
• Hourly differences likely resolvable with more data
• Magnitudes of our coefficients suggest this could be nontrivial
• Suggestive evidence that:
• Greatest increases are in the morning and evening, not on peak
• Highest effect in Inland Hills: Air conditioning is on the margin
• This is just the beginning
50
WORKS CITED
1.
2.
3.
4.
5.
6.
7.
8.
51
Boomhower, Judson; Davis, Lucas W. “Do Energy Efficiency Investments
Deliver at the Right Time?” 2016. Working paper.
https://ei.haas.berkeley.edu/research/papers/WP271.pdf
Jessoe, Katrina, and David Rapson. 2014. Knowledge Is (Less) Power:
Experimental Evidence from Residential Energy Use. American Economic
Review 104(4): 1417–1438.
Evergreen Economics. “Needs Assessment for the Energy Savings Assistance
and the California Alternate Rates for Energy Programs.” December 16, 2013.
http://www.calmac.org/publications/LINA_report_-_Volume_1_-_final.pdf
Meyer, Russell M.; Sherwin, Evan D.; Azevedo, Inês M. L. “Energy
Consumption and Energy Efficiency Rebate Programs.” Working paper.
Wharton Consumer Analytics Initiative Informational Materials, 2012
Palmgren, Claire; Stevens, Noel; Goldberg, Miriam; Rothkin, Karen. 2009
California Residential Appliance Saturation Study. California Energy
Commission 2009
PG&E, Find out if Peak Day Pricing is right for your business. Pac. Gas Electr.
(2017), (available at https://www.pge.com/en_US/business/rate-plans/rateplans/peak-day-pricing/peak-day-pricing.page).
Espey, James A.; Espey, Molly. Turning on the Lights: A Meta-Analysis of
Residential Electricity Demand Elasticities. Journal of Agricultural and Applied
Economics. V39-1, p65-81. April, 2004.
WORKS CITED
•
•
•
•
•
•
52
LIHEAP Clearinghouse. “Low-income energy programs funding history 19772013.” https://liheapch.acf.hhs.gov/Funding/lhemhist.htm
Personal communication with Carol Edwards of Southern California Edison
Personal communication with Brock Glasgo
Gainesville Green. “Your home energy tracking system.” 2016. gainesvillegreen.com/
OpenEEMeter. openeemeter.org
Pacific Gas and Electric. “Submission of High Opportunity Projects and
Programs (HOPPs) Proposal - Residential Pay-for-Performance Program.”
Advice submitted to the Public Utilities Commission of the State of California.
March 25, 2016 http://www.pge.com/nots/rates/tariffs/tm2/pdf/GAS_3698G.pdf
THE INCREASE IS HIGHER DURING SUMMER
53
DAILY REGRESSION TABLE
lkWh |
Coef. Std. Err.
P>|t| [95% Conf. Interval]
----------------+------------------------------------------------------------CARE| .104
.013 0.000 .079
.128
Rebate| .071
.034
0.037 .004
.137
BPP | .067
.028
0.015 .013
.121
ClimateSmart | -.575
.197 0.003 -.961
-.189
DirAccess | .073
.039 0.064 -.004
.150
SmartAC | .055
.043
0.201 -.029
.140
SmartRate | -.018
.037
0.624 -.091
.055
Time | 4.43e-5 1.36e-5 0.001 1.77e-5 7.09e-5
tmax15 | -.005
8.06e-5 0.000 -.005
-.005
tmin15 | -.002
1.77e-4 0.000 -.002
-.001
tmax15sq | 2.97e-5 3.21e-7 0.000 2.91e-5 3.03e-5
tmin15sq | 7.24e-6 9.92e-7 0.000 5.29e-6 9.18e-06
intercept | 2.56
.016
0.000 2.52
2.59
Time controls | Included
54
DAILY REGRESSION TABLE
lkWh |
Coef. Std. Err.
t P>|t| [95% Conf. Interval]
-------------+---------------------------------------------------------------mon | -.0158479 .0012796 0.000 -.0183562 -.0133396
tue | -.0288065 .0014081 0.000 -.0315666 -.0260465
wed | -.0322126 .0014079 0.000 -.0349724 -.0294528
thu | -.029273 .0014006 0.000 -.0320185 -.0265275
fri | -.0341331 .0013651 0.000 -.036809 -.0314572
sat | -.0095719 .0009768 0.000 -.0114866 -.0076573
jan | -.0465839 .0044764 0.000 -.0553585 -.0378093
feb | -.0821803 .0052182 0.000 -.092409 -.0719516
mar | -.1163857 .00567 0.000 -.1274999 -.1052714
apr | -.1477856 .005866 0.000 -.1592841 -.136287
may | -.1381218 .006241 0.000 -.1503554 -.1258882
jun | -.0777472 .0068655 0.000 -.0912049 -.0642896
jul | -.0331107 .0073513 0.000 -.0475207 -.0187008
aug | -.043679 .0070718 0.000 -.0575412 -.0298169
sep | -.0833405 .006869 0.000 -.0968052 -.0698758
oct | -.1104138 .0051853 0.000 -.1205779 -.1002496
nov | -.085717 .0033219 0.000 -.0922285 -.0792054
55