[WesIngwersen]

Closes #901. Part of the plan in #902.

Adds bedrock/transform/eeio/electricity_price.py: the annual electricity price per kWh delivered, by BEA 2017 detail manufacturing industry, for 2017-2024. Nothing reads it yet, so no model output changes. #902 will divide bills by it.

How the price is built

price_i(t) = relative_i(t) x level(t)
level(t)   = EIA industrial price(t) x MECS-to-EIA ratio(t)

• Survey prices: MECS Table 7.10 dollars over Table 7.7 kWh. Both count purchases from any supplier, so they include delivery. A row with no total uses its local-utility values. A NAICS with no published row of its own takes what its parent row has left after the published children.
• Relative prices: anchored on MECS 2022, moving geometrically to MECS 2018 and held to 2017. An industry whose surveys publish different rows stays at its 2022 relative.
• Level: matches the industries' average MECS price in both surveys, using a ratio to EIA's industrial price that moves between them and holds outside.

Buyers outside manufacturing take their EIA class price (class_price). Prices are per kWh delivered, as published. The allocator's class kWh are scaled to eGRID generation, so #902 has to convert.

Why not MECS's 31-33 row for the level. In MECS 2022 the three-digit rows sum to $2.1bn less than that row. Scaling to it moved every 2022 price about 0.16 ¢/kWh above its published value.

Results on the real tables

Industries priced	232 of 232
Cornerstone manufacturing commodities without a price	0
2022 against the published price	exact
2018 against the published price	median gap 0.02 ¢/kWh; above 0.05 for the 4 industries on different rows, at most 1.06
Industries on the same rows in both surveys	228 of 232
Rank correlation of relative prices, 2018 and 2022	0.91
kWh-weighted mean absolute change in relative price	8.5%
Relative price moving more than 20%	2 industries, 1.5% of kWh

The two movers are 331313 alumina and primary aluminum (+36%) and 331314 secondary aluminum (-37%). They are left unsmoothed, because smelter closures make a real change plausible.

¢/kWh delivered	2017	2018	2020	2021	2022	2023	2024
EIA industrial	6.88	6.92	6.67	7.18	8.32	8.04	8.13
Lowest industry	4.05	4.07	4.60	4.88	5.42	5.24	5.30
Median industry	8.56	8.61	8.05	8.46	9.96	9.63	9.73
Highest industry	11.73	11.80	11.46	12.39	14.40	13.92	14.07

Limits

• Survey grade: the anchor follows the plan in Split electricity bills into generation and a buyer-specific delivery margin (42ELEC) #902. I haven't used EIA's relative standard errors for these tables.
• Shared rows: 3-digit fallbacks give several industries one price. Furniture (337) has no published total, so it uses the local-utility price.
• 331314: this BEA code is priced here but folds into a Cornerstone commodity. Split electricity bills into generation and a buyer-specific delivery margin (42ELEC) #902 applies that mapping.
• Withheld cells: MECS shows these as zero, and they are treated as missing.

Checks

Unit tests are synthetic and need no data. black, ruff and mypy are clean on the new files. The real-table results above came from a local run. The suite is left to CI.

🤖 Generated with Claude Code