---
id: VOLT-HOME-WP-066
title: "How fast do day-ahead price shocks propagate between zones?"
slug: how-fast-do-day-ahead-price-shocks-propagate-between-zones
description: "Daily signed neighboring-zone spreads show substantial one-day persistence, but daily lag correlation cannot resolve the speed or direction of quarter-hour shock propagation."
published: 2026-08-30
cluster: "Grid coupling, flows, and outages"
status: measured
evidence: "/research-data/home-papers/how-fast-do-day-ahead-price-shocks-propagate-between-zones.json"
figure: "/research-media/home-papers/how-fast-do-day-ahead-price-shocks-propagate-between-zones.webp"
figure_alt: "Chart for How fast do day-ahead price shocks propagate between zones?: median one-day neighbor spread persistence, shown as P10, Median, P90."
source_ids:
  - entsoe-sdac
  - ec-sdac-15m
  - iea-electricity-2026
  - acer-retail-2025
  - volt-architecture
peer_reviewed: false
---

## Abstract

This paper asks how rapidly day-ahead price shocks move between neighboring European bidding zones. The frozen analysis cannot observe propagation at an intraday horizon. Instead, it measures one-day persistence in the signed daily mean-price difference for each directed border link. For links with enough observations, the procedure correlates today’s signed spread with the next observed row’s signed spread and reports the median correlation across links.

The evidence contains 192 link-level lag correlations. Their median is 0.5020288289, and the regenerated evidence reports no uncertainty interval for this estimand. The primary value indicates moderate persistence: the sign and size of a link’s daily mean-price difference tend to carry information into its next observation. It does not estimate how many minutes or hours a shock takes to cross a border, identify which endpoint leads, or prove transmission through cross-border flow.

Daily aggregation, possible missing dates, common regional shocks, coupled-auction simultaneity, and stable structural differences all limit interpretation. The correct conclusion is about one-day spread persistence, not propagation speed. The result can inform expectations about spatial price regimes, but it is not a causal network model, a production forecast promotion, or trading advice.

## Plain-language answer

Neighboring-zone daily price differences tend to persist from one observation to the next. The middle link-level lag correlation is about 0.502, where one would indicate perfect same-pattern persistence and zero would indicate no linear relationship.

That number does not answer “how fast” in clock time. The calculation uses one daily mean price per zone and compares adjacent rows. A shock could be reflected in both zones within the same day, could take several settlement intervals, or could be a shared response to weather or fuel conditions. All of those pathways can produce daily persistence.

For a household, the useful message is that a local-versus-neighbor price difference may not disappear immediately at the next daily observation. Still, the household should use its local curve and a validated forecast. A neighboring price is contextual information, not proof of a leading signal or a guaranteed predictor.

## Research question

The registered question contains three concepts: a price shock, propagation between zones, and speed. The implemented statistic directly captures none of them at fine resolution. It evaluates whether a signed daily mean-price spread on a border is linearly related to that border’s signed spread in the next sorted observation.

The estimand is therefore “median one-step persistence across links.” A positive link-level coefficient means that relatively positive source-minus-destination spreads tend to be followed by relatively positive spreads, and negative ones by negative ones. It does not locate an initiating shock, identify source and receiver, or distinguish propagation from shared persistence.

A genuine speed study would need a clear event threshold, native interval timestamps, synchronized publication and delivery clocks, directional lead-lag models, and placebo links. The current result can place a daily-scale bound on what the evidence resolves: persistent relationships remain visible at one-step daily aggregation, while subdaily speed remains unknown.

## Data and provenance

The aggregate is published at `/research-data/home-papers/how-fast-do-day-ahead-price-shocks-propagate-between-zones.json`. Status is `measured`; cutoff is `2026-08-30T00:00:00Z`. The recorded daily-price window is 2021-01-01 through 2026-08-29. The corpus has detailed intervals from 2025-10-01 and longer price history from 2015-01-01, but the current metric uses daily mean-price spreads.

The allowed grid-family contracts are `border_flows`, `day_ahead_prices`, `outage_events`, `generation_mix`, and `zone_load`. Directional flow rows define available border identities in the common matched panel, while endpoint daily prices generate the signed spread. The published lag statistic does not adjust for outage, mix, or load conditions.

The production snapshot was read in a SELECT-only transaction with a 180-second timeout. Its SHA-256 is `7e97489fc8528c8cc8c38830e05b48d949ce1f67b98575dff26f5d7c321e4c67`. Analysis code is bound by `57c57de79cdab2b5b6d6c54c485cb5162598c5ba0b0bfe995da40d75e6c52ba9`; protocol by `adb36bf6b447af9f96339249b8becaefc20422499cca1977242866347a97bd4b`; the paper registry by `7bcb91d7476d0a69fe9fa75a5c7782f8117e0153f82f9112b7e1d307d3943717`; and the source registry by `07949550ac443ff673fda5c0209b99f137544f3ffecf6775f109bb9d09663bd6`.

ENTSO-E explains the coupled day-ahead market and its common algorithm. The European Commission source documents the transition to 15-minute day-ahead market time units, underscoring the resolution that a daily study cannot exploit. The IEA supplies broader flexibility and grid context. ACER and CEER connect market conditions with consumer contract exposure. Voltcast’s architecture describes UTC timestamps and native-resolution price storage. The measured persistence is generated from the frozen aggregate, not borrowed from these references.

## Method

The shared panel begins with directional flow records keyed by source zone, destination zone, and UTC date. Endpoint daily mean prices are matched to each key. The signed spread is calculated as source-zone mean price minus destination-zone mean price.

Rows are grouped by exact directed link. Within each group, they are sorted by date. Groups with more than three rows are eligible. For each eligible link, Pearson’s correlation is calculated between the sequence of signed spreads excluding its final observation and the same sequence shifted forward by one row. This produces one lag-one correlation per link.

The primary estimate is the median of those link-level coefficients, giving each eligible link one value in the final distribution regardless of its history length. The evidence sample size of 192 refers to eligible link coefficients, not days. The current `bootstrap_95_interval` is null and the interval method is “not reported for this estimand,” so no confidence interval is attached to the median.

Adjacent sorted observations can be separated by more than one calendar day if rows are missing. No check enforces one-day continuity. The method also uses signed spread persistence, not an event-selected “shock” response. These choices are part of the frozen implementation and define the limits of exact reproduction.

## Results

Across 192 eligible directed links, the median lag-one Pearson correlation of signed daily mean-price spreads is 0.5020288289441167. Rounded, the median persistence coefficient is 0.502.

The coefficient is materially above zero and below one. Descriptively, signed spatial price differences have appreciable continuity at the next observed daily step. Because the statistic is a median, it summarizes the middle eligible link rather than pooling every day into one coefficient.

The figure shows the distribution of link coefficients using labels “P10,” “Median,” and “P90.” Their values are 0.31871158108229103, 0.5020288289441167, and 0.7591819892366298, rendered as 0.319, 0.502, and 0.759. These quantiles are descriptive; the evidence reports no interval for the median.

No secondary results are published. There are no endpoint-leading coefficients, minute-level lags, shock thresholds, border-specific values, event counts, flow-conditioned estimates, or household forecast scores. The study cannot rank propagation speeds or say that one named zone consistently leads another.

## Robustness and placebo checks

Computing a separate correlation for each eligible link prevents long-history links from entirely dominating a single pooled time-series coefficient. Taking the median limits the influence of extreme link values. Keeping the spread signed preserves orientation and regime direction that absolute spreads would discard.

Those design choices do not eliminate serial dependence or common shocks. They also do not verify calendar continuity. A robustness rerun should require exactly adjacent dates, compare absolute and signed spreads under a frozen hypothesis, cluster by link and date, and stratify around prespecified shock events. A placebos programme could pair non-neighboring zones, shuffle event dates within seasonal blocks, or reverse lead and lag to test directional claims.

The family protocol refers to blocked intervals and Holm correction for inferential claims. This paper remains descriptive; no paper-specific placebo output, interval, or multiplicity-adjusted significance test is in its evidence. Therefore the robust claim is limited to deterministic reproduction of a moderate median lag association under the stated routine.

## Limitations

Daily means are too coarse to estimate speed measured in minutes or hours. Finer 15-minute market intervals can contain sequences that cancel in the daily average. The method also lacks price-publication timing and does not determine when information became available to either zone.

Lag correlation is not shock propagation. Stable fuel, weather, hydrology, generation, load, market boundaries, and transfer constraints can make a spread persist without a discrete shock moving across a border. Coupled day-ahead prices are jointly formed, which complicates any leader-follower interpretation.

The panel starts from links represented in daily flow data. Coverage can be unbalanced; reciprocal directed links may both exist; and adjacent rows need not be adjacent calendar days. A link needs only more than three rows, so coefficient precision can vary greatly. The final median gives sparse and dense links equal weight.

The evidence does not control for outages, generation mix, load, flow magnitude, capacity, season, or crisis regime. It reports no causal identification and no out-of-sample forecast comparison. A correlation of 0.502 cannot be converted into EUR/MWh, household savings, or forecast accuracy.

Retail contracts further separate wholesale price dynamics from bills. Taxes, margins, network charges, caps, and hedges are not represented. No customer behavior or measured saving is part of the result.

## Practical implication

If a household-facing forecast uses neighboring-zone information, it should be evaluated against a local persistence baseline and on the actual decision clock. This paper alone does not validate such a feature. It merely shows that daily signed spreads possess persistence worth testing carefully.

Public explanations should avoid phrases such as “shocks cross the border in one day.” The data say that one-step daily spread relationships persist, not that transmission takes a day. Native-interval event studies are required before making a speed claim.

## Reproducibility

Using the frozen snapshot, match each directional flow-link date to both endpoint daily mean prices. Compute the signed source-minus-destination spread, group by directed link, sort by ISO date, and retain groups with more than three observations. Correlate each link’s spread sequence at rows `0..n-2` with rows `1..n-1`, then take the median of valid correlations. Expected output is 192 coefficients with median 0.5020288289441167.

An exact rerun should preserve UTC dates, directed link identity, missing-row behavior, and the code’s Pearson formula. A calendar-contiguous, interval-level, event-selected, or weighted analysis is a new specification. Reproducers should preserve the reported null interval rather than manufacture uncertainty on the median. Licensing terms appear at `/legal/data-licensing` and the canonical licensing document.

## Disclosure

The analysis and writing were model-assisted. This public working paper is not peer reviewed. It does not upgrade daily persistence into a causal or subdaily propagation claim, and it does not invent an uncertainty interval absent from the regenerated evidence. Volt has no live traders or live capital. This is not trading advice, an investment signal, or a promise of household savings.

## References

- [ENTSO-E — Single Day-ahead Coupling (SDAC)](https://www.entsoe.eu/network_codes/cacm/implementation/sdac/)
- [European Commission — EU electricity trading in the day-ahead markets becomes more dynamic](https://energy.ec.europa.eu/news/eu-electricity-trading-day-ahead-markets-becomes-more-dynamic-2025-10-01_en)
- [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)
- [Voltcast — Voltcast Architecture](https://github.com/ossedk/voltcast/blob/main/docs/voltcast/ARCHITECTURE.md)