Online Data Appendix for "Headwinds and Tailwinds: Implications of Inefficient Retail Energy Pricing for Energy Substitution" by Severin Borenstein and James B Bushnell, 2021

This appendix describes the details of the calculations that were done to create the data on which the tables and figures in the paper were based. Data sets for each are also available as described below.

Section II. Estimation of Existing Pricing Distortions

A. Electricity

For details of construction of these data, please see Borenstein and Bushnell, "Do Two Electricity Pricing Wrongs Make a Right? Cost Recovery, Externalities, and Efficiency" (2021) https://haas.berkeley.edu/wp-content/uploads/WP294.pdf

Dataset is annual_elec_by_utilstate.csv

Variables:
eia_id_d: distribution utility EIA ID
eia_id_e: energy utility EIA ID
year
state
eia_name_d: distribution utility name
eia_name_e: energy utility name
varcharge: variable price in cents/kwh. calculated as (annual residential revenue - 12*monthly fixed charge*number of residential customers)/(annual residential sales).
pmc: private marginal cost in cents/kwh.
emc: external marginal cost in cents/kwh with SCC=$50/ton.
emc100: external marginal cost in cents/kwh with SCC=$100/ton.
smc: social marginal cost in cents/kwh. equal to pmc + emc.
smc100: social marginal cost in cents/kwh with SCC=$100/ton. equal to pmc + emc100.
p_min_smc_kwh: price minus social marginal cost in cents/kwh. equal to varcharge - smc.
p_min_smc_kwh100: price minus social marginal cost in cents/kwh with SCC=$100/ton. equal to varcharge - smc100.
res_sales: annual residential sales in kwh.


B. Natural Gas

We used revenue, sales, and customer data from EIA form 176 and manually collected fixed charge data to determine the variable price for residential natural gas utilities. We received fixed charge data from Catherine Hausman that covered 91% of residential sales volume. We then manually collected additional fixed charges to eventually cover 96% of residential sales volume and 95% of residential customers. We calculated the average variable price as (residential revenue - 12*number of residential customers*monthly fixed charge) / residential volume. Prices are in $/MMBtu.

Some utilities in Georgia and Ohio did not have residential sales volume, only residential transported volume, due to retail choice. For Atlanta Gas & Light, it is the only utility in Georgia that has transported volume and no sales volume. We use a combination of sources - fixed charges from Georgia PUC historical rate documents, average prices from EIA, and volume and customers from EIA - to calculate variable prices.  For Ohio, the majority of utilities have a retail component.  Based on EIA data, the average prices for LDCs are similar to those of marketers. So we used the variable price of the LDC as an approximation of the variable price that all customers face.

We used LAUF (Lost and Unaccounted for Gas) data from EIA. For the utilities missing LAUF, we first replaced LAUF with the previous (or following) year's LAUF if the utility had LAUF reported for another year. If the utility was missing all three years, we replaced it with the national sales-weighted average  for that year. 

For PMC, annual state level citygate natural gas prices are used and assigned to each utility in a given state. 

EMC is created using county level damages data from AP3. We first converted damages from $/short ton to $/metric ton by dividing by 0.907184. We then converted to $/MMBTU. For NOx damages, we assumed 94 lbs per 10^6 scf (EPA AP42) and 4.536x10^-4 metric tons/lb so that our conversion becomes ($/metric ton * 4.536x10^-4 metric tons/lb * 94 lbs per 10^6 scf * 0.001 10^6 scf per MMBTU). This is the NOx damages based on furnace emission factor (from EPA AP42) and excludes NOx damages from water heaters. For CO2, we assumed 0.0543 metric tons per MMBTU).  We then aggregated county level damages to the utility level by taking a population-weighted average of the damages of each county in a given utility service territory. The weight is the proportion of the county that is in the utility service territorry multiplied by the population of that county. Finally, we included LAUF damages using the estimate ($27/MMBTU) from Hausman and Muehlenbachs (2018) but adjusting to use $50/ton SCC instead of $41/ton SCC, giving an estimate of $33.34/MMBTU. We then multiplied this estimate by the estimate of LAUF per MMBTU from EIA 176.

We combined all datasets to get a dataset at the utility-state-year level. Prices vary by utility-state-year. PMC varies by state-year. EMC varies by utility-state for NOx and by utility-state-year for LAUF. Damages from CO2 do not vary. We calculate P-SMC as the variable price minus citygate natural gas price (PMC) minus EMC which consists of NOx, CO2, and LAUF damages. We also calculate P-SMC in cents per kwh using 293.071 kwh/MMBTU). We also keep a county level dataset and average across years to create a utility-state level dataset.  We converted all dollar values to 2016 dollars.

Dataset is annual_ng_by_utilstate.csv

Variables: 
eia176_id: utility EIA ID
state
year
utility_name
varcharge: variable price in $/MMBtu. calculated as (annual residential revenue - 12*monthly fixed charge*number of residential customers)/(annual residential sales).
pmc: private marginal cost in $/MMBtu.
emc: external marginal cost in $/MMBtu.
emc100: external marginal cost in $/MMBtu with SCC=$100/ton.
smc: social marginal cost in $/MMBtu. equal to pmc + emc.
smc100: social marginal cost in $/MMBtu with SCC=$100/ton. equal to pmc + emc100.
p_min_smc_mmbtu: price minus social marginal cost in $/MMBtu. equal to varcharge - smc.
p_min_smc_mmbtu100: price minus social marginal cost in $/MMBtu with SCC=$100/ton. equal to varcharge - smc100.
p_min_smc_kwh: price minus social marginal cost in cents/kwh. 
p_min_smc_kwh100: price minus social marginal cost in cents/kwh with SCC=$100/ton.
res_sales_mmbtu: annual residential sales in MMBtu.


C. Gasoline

We obtained monthly gasoline retail prices by city scraped from AAA website. We took the unweighted average over the cities in a state to get a monthly state average retail price. We then used monthly sales of regular gasoline by state from EIA to take a sales-weighted average of the monthly prices to get a state annual average retail price. We also obtained monthly state average rack prices based on DTN Energy data. In most states, there is only conventional or reformulated gasoline. AZ, IL, IN, KY, MD, ME, MO, NH, NY, PA, TX, VA, WI have both conventional and reformulated. For these states, we used monthly sales of conventional and reformulated gasoline in the state from EIA to get the proportion of sales that are conventional versus reformulated gasoline. We then took a weighted average of the prices to get a single monthly average rack price. Then we used the regular gasoline sales data to take a weighted average across months to get a state annual average rack price (the same method as for the retail prices). 

There are some missing data in the conventional and reformulated sales data: In AZ, many months are missing data, so the average proportion of reformulated/conventional in a given year is used (since there is more variation in the proportions across states than within a state across time). In TX, NH, and ME, single months are occasionally missing so the proportions from the two surrounding months are averaged and used for that month (e.g., if May 2015 is missing, the average of April 2015 and June 2015 is used). There are also some missing data in rack prices: NM, CO, NH, DC, WY all have missing data. For these states, we used rack prices from neighboring states since there is substantial time variation in prices across months. For NM we use AZ conventional since NM only has conventional. For CO we use UT. For WY we use MT. For NH we use VT conventional and MA reformulated (since NH has both, but neighboring states do not have both). For DC we use MD reformulated. 

To construct PMC, we used the state level annual rack price and made some adjustments. In California, we subtracted the cap and trade price, because the full \$50/ton is included in the EMC. In all states, we add a credit card fee which is equal to 2.5% of the retail price. Finally, we add 10 cents for other marginal costs, such as labor, delivery, etc., as discussed in the paper.

To construct EMC, we converted the county level damages from HMMY per mile to damages per gallon. We assumed a vehicle that gets 40 MPG. We also changed the SCC used in HMMY ($41/ton CO2) to $50 per ton CO2. We assumed 0.008887 ton of CO2 per gallon of gasoline (following HMMY). This ignores the fact that gasoline is typically 10% ethanol, so for 10% of the fuel one should use the CO2 emissions of ethanol. Abiding strictly by our approach of considering only tailpipe emissions,  ethanol tailpipe emissions of CO2 are approximately 2/3 of gasoline. However, those emissions are based on combustion of a fuel that has sequestered the CO2 upstream in the previous few years.  Then again, production of corn and refining into ethanol create significant GHG emissions.  We chose to use the HMMY numbers for simplicity.  Using a 10% lower number would not significantly change the results.

We merged all of these data together to create an annual county level dataset for 2014-2016. Note the only variation by county comes from the EMC, but this does not vary across time. Price and PMC vary by year but do not vary within a state. We converted all dollar values to 2016 dollars. We calculated SMC as PMC + EMC and created variables for retail P-SMC both in $/gallon and cents/kwh (multiplied by 3 for vehicle efficiency).  To convert to cents per kwh we used 0.120286 MMBtu per gallon (https://www.eia.gov/energyexplained/units-and-calculators/) and 293.071 kwh per MMBtu. 

Dataset is annual_gas_by_county.csv

Variables:
FIPS: county fips
year
state
retail_price: annual average retail price by state in $/gallon.
pmc: private marginal cost in $/gallon. equal to rack price + 2.5% of retail price for credit card fees + 10 cents for other marginal costs.
emc_per_gal: external marginal cost in $/gallon with SCC=$50/ton.
emc_per_gal100: external marginal cost in $/gallon with SCC=$100/ton.
smc: social marginal cost in $/gallon. equal to pmc + emc_per_gal.
smc100: social marginal cost in $/gallon with SCC=$100/ton. equal to pmc + emc_per_gal100.
p_min_smc_gal: price minus social marginal cost in $/gallon. equal to retail_price - smc.
p_min_smc_gal100: price minus social marginal cost in $/gallon with SCC=$100/ton. equal to retail_price - smc100.
p_min_smc_kwh: price minus social marginal cost in cents/kwh. 
p_min_smc_kwh100: price minus social marginal cost in cents/kwh with SCC=$100/ton. 
population: county population


Section III. Implications for Building and Transportation Electrification

A. Space Heating 

We brought together RASS survey appliance usage and age data, EnergyStar new appliance efficiencies, E3 new appliance efficiencies, and electricity/natural gas prices and SMCs in order to calculate the distribution of savings if people use an electric or natural gas appliance.

In the RASS, we kept only single-family homes and townhomes. We used the natural gas and electric utility listed in the survey. If there was no utility listed, we matched the household's zip code to the utility service territory. For space heating, we required that households have a square footage listed and that they pay for heating. Finally, we only kept households with the following types of space heating: natural gas central forced-air furnace, natural gas floor or wall heater/furnace, electric central forced air furnace, electric central heat pump, or electric through the wall heat pump.  We assigned appliance efficiencies of the appliance in use using the age of the appliance. For each age bin, we took a weighted average of the federal minimum efficiency and EnergyStar efficiency where the weights are the EnergyStar national market shares in that given age bin (no California-specific shares are available). 

For price and SMC, we used the 2016 data from section II for the relevant CA utilities (electric: PGE, LADWP, SCE, SDGE, SMUD; gas: LBGO, PGE, SDGE, SCG, SWGAS). RASS reports estimated appliance usage (UEC) in kwh or therms. We used these data and the assumed efficiency of the appliance given its age to determine how much "energy services out" a given household consumes. For furnace/heat pump fans for forced air space heating, we calculated the ratio of energy used by the fan to heat out and include that energy in our calculations. The ratio for heat pumps is 0.067 and for gas furnaces is 0.03. We can then determine how much energy the household would need to consume to get the same energy services out using either a new electric or new gas appliance. The efficiencies for new appliances are the current EnergyStar guideline efficiencies. We also examined efficiencies assumed in a recent E3 study (https://www.ethree.com/wp-content/uploads/2019/04/E3_Residential_Building_Electrification_in_California_April_2019.pdf). We then calculated a household's savings from a new electric appliance vs a new gas appliance at price and at SMC.

Dataset is space_heat_savings.csv

Variables:
hh_id: household ID in RASS.
servzip: household zip code.
wt: household weight.
heat_type: current type of space heating.
GFURNACE_wtd: calculated efficiency of current gas furnace based on appliance age and historical efficiencies/market shares. only applies to households with heat_type="Gas". Electric Furnace households have assumed efficiency = 1. efficiency measure is "Annual Fuel Utilization Efficiency" (AFUE),  which is the percentage of heat in the incoming fuel that is converted to space heat.
ASHP_heat_wtd: calculated efficiency of current heat pump based on appliance age and historical efficiencies/market shares. only applies to households with heat_type="Electric Heat Pump". efficiency measure is "Heating Seasonal Performance Factor" (HSPF) which is the ratio of heat output in BTU to electricity used in watt-hours.
ght_uec: unit energy consumption (uec) of gas furnace as calculated by RASS in therms.
eht_uec: unit energy consumption (uec) of electric furnace as calculated by RASS in kwh.
ehp_uec: unit energy consumption (uec) of electric heat pump as calculated by RASS in kwh.
el_p: electric price in $/kwh.
el_smc: electric smc in $/kwh.
ng_p: natural gas price in $/therm.
ng_smc: natural gas smc in $/therm.
saving_p: fuel savings in $ from switching from a new gas furnace to a new electric heat pump at price.
saving_smc: fuel savings in $ from switching from a new gas furnace to a new electric heat pump at smc.


B.  Water Heating

Same as for space heating except:
For water heating, we only included homes that pay for water heating and that included the number of rooms in the home. We only kept households with the following types of water heating: natural gas standard tank, electric standard tank, electric heat pump. We assigned current appliance efficiencies using the age of the appliance. For each age bin, we took a weighted average of the federal minimum efficiency and EnergyStar efficiency where the weights are the EnergyStar national market shares in that given age bin (no California-specific shares available). 

Dataset is water_savings.csv

Variables:
hh_id: household ID in RASS.
servzip: household zip code.
wt: household weight.
heat_type: current type of water heating.
GWH_wtd: calculated efficiency of current gas water heater based on appliance age and historical efficiencies/market shares. only applies to households with heat_type="Gas". efficiency measure is "Energy Factor" (EF),  which is the ratio of useful energy output from the water heater to the total amount of energy delivered to the heater.
ASHP_heat_wtd: calculated efficiency of current electric water heater based on appliance age and historical efficiencies/market shares. only applies to households with heat_type="Electric". efficiency measure is "Energy Factor" (EF),  which is the ratio of useful energy output from the water heater to the total amount of energy delivered to the heater.
gwh_uec: unit energy consumption (uec) of gas water heater as calculated by RASS in therms.
wht_uec: unit energy consumption (uec) of electric water heater as calculated by RASS in kwh.
el_p: electric price in $/kwh.
el_smc: electric smc in $/kwh.
ng_p: natural gas price in $/therm.
ng_smc: natural gas smc in $/therm.
saving_p: fuel savings in $ from switching from a new gas water heater to a new electric heat pump water heater at price.
saving_smc: fuel savings in $ from switching from a new gas water heater to a new electric heat pump water heater at smc.

C. Vehicles

We brought together vehicle usage data from NHTS, gasoline and electricity prices and SMCs, and car efficiency information. For price and SMC, we used the 2016 electricity and gasoline prices. We used current gas mileage/efficiency for Nissan Leaf S Plus, Nissan Versa, BMW 740i, Tesla Model S, Tesla Model 3, and Prius Prime. We used households' zip codes to map them to a given electric utility. If a zip code maps to more than one utility, we took a sales-weighted average of the utilities to determine price and SMC. We also used the household's county to match to the correct gasoline EMC. We only have state level gasoline prices so all households have the same gasoline price and PMC. Using the prices and efficiencies and the estimate of VMT from NHTS, we then calculated a distribution of the savings across households for either gasoline or electric cars. We compared Nissan Leaf S Plus to Nissan Versa and BMW 740i to Tesla Model S.

Dataset is car_savings.csv

Variables:
sampno: household ID in NHTS.
vehno: vehicle number within a household.
wthhfin: household weight.
odmiles: annual miles traveled.
emean_p: electricity price in $/kwh.
emean_smc: electricity social marginal cost in $/kwh.
gmean_p: gasoline price in $/gallon.
gmean_smc: gasoline social marginal cost in $/gallon.
savings_nissan_p: fuel savings in $ from switching from Nissan Versa to Nissan Leaf S Plus at price.
savings_nissan_smc: fuel savings in $ from switching from Nissan Versa to Nissan Leaf S Plus at smc.
savings_bmwtesla_p: fuel savings in $ from switching from BMW 740i to Tesla Model S at price.
savings_bmwtesla_smc: fuel savings in $ from switching from BMW 740i to Tesla Model S at smc.