---
id: "VOLT-HOME-WP-072"
title: "What is the carbon cost of always charging an EV overnight?"
slug: "what-is-the-carbon-cost-of-always-charging-an-ev-overnight"
description: "A bounded alignment diagnostic relating overnight day-ahead prices to daily low-carbon generation share."
published: "2026-08-30"
cluster: "Carbon-aware household flexibility"
status: "measured"
evidence: "/research-data/home-papers/what-is-the-carbon-cost-of-always-charging-an-ev-overnight.json"
figure: "/research-media/home-papers/what-is-the-carbon-cost-of-always-charging-an-ev-overnight.webp"
figure_alt: "Chart for What is the carbon cost of always charging an EV overnight?: association between overnight price and daily low-carbon share, shown as overnight price, low-carbon share."
source_ids:
  - "iea-electricity-2026"
  - "acer-retail-2025"
  - "iea-demand-flexibility"
  - "ec-retail-flexibility-2026"
  - "volt-licensing"
peer_reviewed: false
---

## Abstract

The title asks for the carbon cost of a fixed overnight EV-charging rule, but the frozen evidence supports a much narrower answer. It reports a Pearson association of 0.3666390116420147 between an overnight day-ahead-price sequence and a daily low-carbon-generation-share sequence, using 3,340 observations for the primary result. That is an alignment diagnostic, not a carbon-cost estimate. No EV energy requirement, charging power, arrival time, departure time, interval-specific generation mix, marginal generator, or measured household load is included in the reported outcome.

Daily low-carbon generation share is an operational proxy. It is not lifecycle marginal emissions and cannot identify the emissions caused by adding EV demand overnight. Imports are not fully allocated, and a daily average cannot identify an overnight carbon intensity. The evidence itself states this limitation. Accordingly, this paper does not report grams of carbon, avoided emissions, or a causal effect. It explains what the measured association says, why it does not answer the title literally, and what evidence a future charging-emissions study would need.

## Plain-language answer

This study cannot calculate the carbon cost of always charging an EV overnight. It finds that the archived overnight-price sequence and daily low-carbon-share sequence have a positive Pearson correlation of 0.3666390116420147 in the frozen calculation. That tells us the two recorded series were not independent in a simple linear descriptive sense. It does not tell us how much carbon an EV caused, whether overnight charging was cleaner than daytime charging, or whether changing the schedule would reduce emissions.

The environmental measure is a daily operational generation-share proxy. It is not an interval-specific, consumption-based, lifecycle, or marginal-emissions measure. The price series is not itself a carbon measure. A positive association between price and low-carbon share also does not imply that expensive periods are causally cleaner. The practical lesson is that a fixed “overnight equals clean” rule is not justified by this evidence; neither is the opposite rule.

## Research question

The registered question is intentionally household-facing: what carbon cost follows from always charging an EV overnight? The measured research question is narrower: how does an overnight day-ahead-price summary align with daily low-carbon generation share in the available corpus? The primary metric is explicitly named “association between overnight price and daily low-carbon share.”

This distinction is not semantic housekeeping. A carbon-cost question requires an energy quantity and an emissions factor appropriate to the increment of demand. The frozen outcome has neither. It therefore cannot estimate a charging footprint or compare charging schedules on measured emissions. It can only test whether an overnight market-price summary and a daily average generation-mix proxy move together in the archived calculation.

The paper preserves the registered title because it is the question readers may ask, while rejecting an answer that the data do not identify. This is an observational study and makes no causal claim.

## Data and provenance

The public evidence JSON linked above is the canonical aggregate for this paper. It lists the registered contracts `generation_mix`, `day_ahead_prices`, `capture_stats`, `res_accuracy`, and `zone_temp_weighted`. The primary diagnostic draws on price-day records and generation-mix records. No household meter, vehicle, charger, mobility, retail-bill, or customer dataset is listed.

The evidence records daily prices from 2021-01-01 through 2026-08-29, detailed intervals from 2025-10-01 through 2026-08-29, and a broader long-history range from 2015-01-01 through 2026-08-29. The cutoff is 2026-08-30T00:00:00Z. Extraction was read-only and used a statement timeout of 180 seconds.

The frozen snapshot SHA-256 is `f77e3ae328f93916e53b1bab7516e1d0ac740a0dbf424cd2b73c81fee2559318`. The analysis-code hash is `57c57de79cdab2b5b6d6c54c485cb5162598c5ba0b0bfe995da40d75e6c52ba9`, the protocol hash is `adb36bf6b447af9f96339249b8becaefc20422499cca1977242866347a97bd4b`, the paper-registry hash is `7bcb91d7476d0a69fe9fa75a5c7782f8117e0153f82f9112b7e1d307d3943717`, and the source-registry hash is `07949550ac443ff673fda5c0209b99f137544f3ffecf6775f109bb9d09663bd6`. These bind the publication to a declared computation. They do not cure the difference between an alignment proxy and a charging-emissions estimand.

## Method

The archived analysis constructs overnight price summaries from available price-day slots. “Overnight” is implemented with the frozen local-time cutoff in the analysis and by taking the selected slots’ mean. It then computes a Pearson correlation against the available daily low-carbon-share sequence, limited to the common sequence length. The primary result records 3,340 observations.

Pearson correlation describes linear co-movement on its own scale. It is not a regression coefficient in emissions units. It does not state how much the low-carbon share changes for a price change, and it provides no conversion from either series to a household carbon quantity. Correlation is also symmetric and does not identify a direction of influence.

The environmental side is daily, while the charging label is overnight. That resolution mismatch is decisive. A daily low-carbon average can remain unchanged while the overnight interval mix varies. Even a perfectly aligned daily proxy could not reveal the marginal generator responding to a charger. Low-carbon generation share remains an operational proxy, not lifecycle marginal emissions.

The calculation also does not simulate a vehicle. There is no charging duration, energy target, charger capacity, availability window, efficiency loss, or alternative schedule. It is therefore not a counterfactual scheduling study despite the title’s practical framing. Its defensible role is preliminary signal alignment.

## Results

The reported Pearson r is 0.3666390116420147, with a primary-result sample size of 3,340. The frozen interpretation states: “Daily low-carbon share cannot identify an overnight carbon intensity; the result is an alignment diagnostic.” That interpretation is the governing reading of the result.

The positive sign indicates that the two sequences tended to move in the same direction in a linear descriptive sense. It should not be translated into a claim that higher prices cause higher low-carbon output, that overnight periods are cleaner, or that an EV incurred a particular carbon cost. The magnitude is not an emissions factor.

The result also cannot rank an overnight rule against a price-optimized or carbon-aware schedule. No alternative charging schedule is measured. Because the dependent environmental concept is a daily average proxy, the study does not observe the carbon intensity of the selected overnight slots. Most importantly, it does not measure marginal or lifecycle emissions and does not measure household emissions.

The figure labels are “overnight price” and “low-carbon share,” with values 87.11981173466559 and 59.38346189343282, rendered as 87.1 and 59.4. They are summaries on different scales, not the Pearson coefficient. The evidence reports no bootstrap interval and labels the interval method “not reported for this estimand.”

## Robustness and placebo checks

The family protocol specifies Holm control for inferential claims and blocked resampling for serially dependent data. The public evidence labels this particular result descriptive and says it makes no unadjusted significance claim. It records no secondary results, and its `assumptions` array is empty. The empty array means no paper-specific assumptions were encoded in that field; it does not make the diagnostic assumption-free. The evidence also records no paper-specific placebo comparison. We therefore do not claim statistical significance, robustness across specifications, or successful placebo tests.

Useful but unperformed checks would include key-aligned zone-day pairing, alternative overnight windows, local-time versus UTC construction, zone-specific correlations, seasonal splits, and interval-level generation measures. A genuine charging study would compare the fixed overnight schedule with declared alternatives under the same vehicle constraints. None of those results is present in the evidence JSON, so they are not reported here.

The provenance hashes are the completed integrity check: they allow the precise archived computation to be distinguished from future variants. The figure is a presentation of the registered two-series diagnostic, not an independent validation.

## Limitations

The first limitation is identification. Average low-carbon generation share is an operational proxy, not lifecycle marginal emissions. It does not fully allocate imports and does not identify the unit responding to additional demand. The study cannot infer caused or avoided emissions from that proxy.

The second is temporal resolution. A daily generation-share value is not an overnight intensity. The title refers to an EV action within a portion of the day, while the environmental sequence summarizes the day. That mismatch prevents a carbon-cost interpretation even before vehicle behavior is considered.

The third is missing household physics. No energy requirement, charging curve, efficiency, charger power, plug-in time, departure deadline, or mobility uncertainty is modeled. No taxes, network tariffs, supplier charges, or retail pass-through are modeled either. This is not a household bill study and cannot estimate savings or costs on a customer invoice.

The fourth is observational dependence. Market price and generation mix share common drivers and may be serially, seasonally, and spatially structured. A simple pooled Pearson correlation does not isolate those drivers. No causal claim follows, and no measured household emissions appear in the source.

## Practical implication

An overnight default should be treated as a convenience rule, not automatically as a carbon rule. A controller using this evidence could warn that daily average generation mix does not identify the environmental consequence of a specific charging interval. If users can choose between price and environmental objectives, the data and method behind each objective should be named separately.

For carbon-aware charging, the next requirement is an interval-matched environmental signal with a clearly stated accounting boundary. If the claim concerns causal consequences of moving demand, an average operational share is insufficient; the method would need an appropriately validated marginal response estimate. Lifecycle claims would require a lifecycle boundary rather than this proxy.

Nothing in the result supports a household bill estimate, a claim about measured emissions, or a trade. It is not trading advice. It is a warning against overinterpreting a convenient schedule label.

## Reproducibility

Start from the evidence path in the frontmatter and confirm `paper_id` is `VOLT-HOME-WP-072`, status is `measured`, and the metric is “association between overnight price and daily low-carbon share.” Verify the publication cutoff and all five provenance hashes before attempting to reproduce the value.

Using the frozen code identity, form mean prices from the slots selected by the registered overnight cutoff, obtain the low-carbon-share sequence, limit the inputs to their common length, and compute Pearson correlation. The expected archived output is 0.3666390116420147 with sample size 3,340. A key-aligned join, another time window, a zone weighting, an interval environmental measure, or an EV simulation would be a new analysis and should receive a new evidence record rather than overwrite this one.

The evidence and figure are aggregate publication artifacts. The licensing link governs use and redistribution of underlying data.

## Disclosure

Analysis and drafting were model-assisted. This working paper is not peer reviewed. It is not a household bill study, not a charging-footprint calculation, not a causal study, and not trading advice. Volt has no live traders or live capital. No measured household emissions are claimed. Low-carbon generation share is an operational proxy and is not lifecycle marginal emissions.

## References

- [International Energy Agency — Electricity 2026](https://www.iea.org/reports/electricity-2026)
- [ACER and CEER — Rewarding flexibility: How retail contract choice can help unlock consumer flexibility](https://www.ceer.eu/wp-content/uploads/2025/11/ACER-CEER-2025-Retail-monitoring.pdf)
- [International Energy Agency — Scaling Up Demand Flexibility](https://www.iea.org/reports/scaling-up-demand-flexibility)
- [European Commission — Communication from the Commission on the Citizens Energy Package](https://eur-lex.europa.eu/legal-content/EN/TXT/?uri=CELEX%3A52026DC0115)
- [Voltcast — Voltcast data licensing and redistribution](https://github.com/ossedk/voltcast/blob/main/docs/voltcast/LICENSING.md)