---
id: VOLT-HOME-WP-036
title: "Can price quantiles produce a safer battery schedule?"
slug: can-price-quantiles-produce-a-safer-battery-schedule
description: "Observed BESS-v2 dispersion measures a 74.3295 EUR/MW-day monthly P90–P10 width but does not test a forecast-quantile dispatch policy."
published: 2026-08-30
cluster: "Home batteries"
status: measured
evidence_url: /research-data/home-papers/can-price-quantiles-produce-a-safer-battery-schedule.json
figure_url: /research-media/home-papers/can-price-quantiles-produce-a-safer-battery-schedule.webp
figure_alt: "Chart for Can price quantiles produce a safer battery schedule?: median monthly BESS index P90–P10 width, shown as P10, Median, P90."
source_ids:
  - energy-informatics-storage
  - acer-retail-2025
  - iea-electricity-2026
  - dynamic-tariff-viability
  - volt-research-content
peer_reviewed: false
---
## Abstract

The evidence reports a median monthly BESS index P90–P10 width of 74.3295 EUR/MW-day across 16,368 observations. This measures dispersion in the frozen BESS-v2 index. The evidence interpretation explicitly says the observed dispersion is a transparent risk band, not a forecast quantile schedule. Therefore, the result does not establish that selecting dispatch from price quantiles makes a battery schedule safer.

The distinction between outcome dispersion and predictive quantiles is central. An observed P10–P90 range describes how indexed values varied in the sample. A forecast P10 or P90 must be issued before delivery, calibrated against later outcomes, and incorporated into a decision rule. BESS-v2 is normalized per unit of power and uses perfect foresight of the realized day-ahead curve. It is wholesale-energy-only and excludes retail taxes and network charges. The 74.3295 EUR/MW-day width is useful for showing that central indexed value hides substantial variation, but safety requires a separately frozen, forecast-clock-valid policy and a stated loss function.

## Plain-language answer

Price quantiles can help express uncertainty, but this evidence does not test that claim directly. It shows that the middle 80% span of monthly BESS-v2 index outcomes has a median width of €74.3295 per normalized MW-day. A wide band warns that a single average or median is not enough to describe the range of indexed outcomes. It does not tell a controller which intervals to choose tomorrow.

A safer schedule might avoid charging unless a low-price quantile is sufficiently below a high-price quantile after efficiency and wear, or it might preserve reserve energy when downside risk is high. To validate such a rule, the quantiles must be forecasts available at the scheduling clock and must have demonstrated coverage. Here the index uses realized prices with perfect foresight, while the P10–P90 width is observed dispersion. The direct answer is thus: potentially in principle, but not demonstrated by this result. What is demonstrated is material variation in normalized monthly index value.

## Research question

The operational question is whether probabilistic prices improve a battery decision relative to a point forecast. “Safer” must be defined: fewer negative-value cycles, lower downside tail, more stable state of charge, better reserve protection, or reduced schedule churn are different objectives. A quantile policy should be scored on its declared objective and compared with simple controls using identical issue times and realized settlement rows.

The published metric asks a prior descriptive question: how wide is the P90–P10 range of monthly BESS-v2 values? That range can motivate probabilistic scheduling but cannot validate it. A historical distribution is not automatically a calibrated forecast, and an outcome quantile is not a command. This paper holds the evidentiary boundary at that distinction rather than translating observed dispersion into an unmeasured risk reduction.

## Data and provenance

The source snapshot is SHA-256 `7e97489fc8528c8cc8c38830e05b48d949ce1f67b98575dff26f5d7c321e4c67`, read through a SELECT-only transaction marked read-only with a 180-second statement timeout. Permitted contracts are `day_ahead_prices`, `bess_index_daily`, `bess_forecast_daily`, `generation_mix`, and `capture_stats`. Daily prices cover 2021-01-01 to 2026-08-29, detailed intervals cover 2025-10-01 to 2026-08-29, and the long-history boundary covers 2015-01-01 to 2026-08-29. The publication cut-off is 2026-08-30T00:00:00Z.

The primary metric is median monthly BESS index P90–P10 width, unit EUR/MW-day, sample size 16,368. Provenance hashes are analysis code `57c57de79cdab2b5b6d6c54c485cb5162598c5ba0b0bfe995da40d75e6c52ba9`, protocol `adb36bf6b447af9f96339249b8becaefc20422499cca1977242866347a97bd4b`, registry `7bcb91d7476d0a69fe9fa75a5c7782f8117e0153f82f9112b7e1d307d3943717`, and source registry `07949550ac443ff673fda5c0209b99f137544f3ffecf6775f109bb9d09663bd6`. The common identities allow readers to audit this paper against the same corpus used by the battery family.

## Method

BESS-v2 is a normalized day-ahead index per MW of power. Its perfect-foresight convention selects opportunities using the realized price curve, making the output a ceiling-like benchmark rather than an attainable live schedule. The paper-level analysis groups indexed observations into monthly distributions and summarizes the distance between the 90th and 10th percentiles. The median of those widths is the primary output.

That computation differs from producing predictive price quantiles. No model issue time, lead horizon, calibration curve, pinball loss, coverage test, or quantile-driven dispatch policy is specified in the evidence. Wholesale-only accounting omits retail taxes and network charges. The result is descriptive; the family’s Holm correction applies to inferential claims, but no unadjusted significance claim is made. The method therefore characterizes spread in indexed outcomes without claiming a reduction in decision risk.

## Results

The frozen point estimate is 74.3295 EUR/MW-day. It says the median monthly distance between P90 and P10 BESS-v2 index outcomes is €74.3295 per normalized MW-day under the registered corpus. A larger width means more dispersion; it does not by itself identify upside, downside, skew, or forecastability. Nor does it establish that the same band would cover future outcomes at an 80% rate.

The regenerated JSON reports `bootstrap_95_interval: null` and no interval method for this estimand. The figure publishes P10 = 17.2335, median = 74.3295, and P90 = 300.8197 EUR/MW-day. These values describe the distribution of monthly widths and are not uncertainty bounds around the median. The defensible empirical record is the point value, unit, sample size, windows, assumptions, figure values, and evidence interpretation.

No secondary results are reported. In particular, the evidence contains no frequency of loss, no expected shortfall, no coverage score, and no comparison between quantile and point schedules. Those omissions prevent a “safer” verdict but do not erase the measured dispersion.

## Robustness and placebo checks

A proper quantile-schedule test should freeze P10, P50, and P90 price curves before the day-ahead decision, verify monotonicity, and measure empirical coverage by horizon and zone. It should then compare a declared risk-aware policy with a P50 policy and simple conservative controls. Outcomes might include gross and net value, downside quantiles, negative-cycle frequency, state-of-charge violations, and schedule changes. The current evidence reports none of these comparisons.

Placebos should include quantiles shuffled across dates, unconditional historical quantiles with no current information, and widened bands that appear conservative but are uncalibrated. A policy should not earn a safety claim merely by cycling less; it must be compared on the stated objective and opportunity cost. These are future protocol requirements, not results. The frozen sample, transparent interpretation, and refusal to invent an interval where none is reported are the present checks against overstatement.

## Limitations

Observed monthly dispersion is retrospective. It mixes market regimes, zones, and durations and may not represent a conditional distribution available at a specific issue time. A P90–P10 width loses information about skew, multimodality, tails beyond P10 and P90, and dependence between intervals. It does not show whether low and high battery-index outcomes are predictable.

Perfect foresight separates BESS-v2 from a deployable schedule. Wholesale-only accounting excludes retail taxes, network charges, tariff asymmetry, degradation, and auxiliary consumption. Normalization per MW does not represent a named battery or its reserve needs. The evidence gives no policy, calibration, safety endpoint, or uncertainty interval for the estimand. Thus the result is a risk-description input, not proof that price quantiles improve control.

Safety also requires calibration under the same conditions in which the policy operates. An unconditional band can have correct overall coverage while failing during price spikes, negative-price episodes, or particular horizons that drive battery decisions. Coverage should therefore be reported by lead time and relevant regime, with misses shown on the same public record as successes. Quantile crossing should be corrected without silently changing the issued forecast, and every revision should preserve its original timestamp.

A policy comparison must account for abstention. A conservative quantile rule may appear safer simply because it cycles less often. That can be appropriate, but the foregone indexed value is part of the trade-off. Report downside, expected value, number of actions, and equivalent throughput together. The observed 74.3295 EUR/MW-day width can motivate this evaluation, yet it cannot choose the risk tolerance or establish that any particular abstention rule is optimal.

A separate choice is whether quantiles describe prices, spreads, or final battery value. Interval-price quantiles do not automatically combine into a valid quantile for a multi-interval dispatch because forecast errors are dependent across time. Optimizing each marginal quantile independently can create an internally incoherent curve or an infeasible state path. Scenario-based optimization can preserve dependence, but it introduces its own calibration and model-risk requirements. The frozen monthly BESS-v2 width is downstream outcome dispersion and avoids pretending to solve that dependence problem.

For public evaluation, “safer” should be tied to a measurable threshold before results are read. Examples include limiting the probability of a negative net cycle or bounding lower-tail monthly value while retaining a stated fraction of the point-policy opportunity. The threshold and comparator must be frozen together. Otherwise, safety can be claimed after selecting whichever metric happens to improve. No such threshold is included here, so the paper remains descriptive.

## Practical implication

Do not optimize a battery from a single expected value when the relevant index distribution is wide. Present a range, identify the forecast issue clock, and state what downside the controller is designed to limit. A risk-aware rule might require spread to remain positive under conservative charge and discharge quantiles after explicit costs. That rule should be validated prospectively and compared with a simple P50 schedule.

Use the 74.3295 EUR/MW-day width only as a benchmark for historical dispersion under BESS-v2. Do not substitute it for a predictive interval or assume it has future coverage. A transparent application should show forecast calibration, the realized schedule result, and the perfect-foresight ceiling separately. It should also make clear when conservatism reduces expected opportunity as the price of downside protection.

## Reproducibility

Retrieve `/research-data/home-papers/can-price-quantiles-produce-a-safer-battery-schedule.json`. Verify status `measured`, value 74.3295 EUR/MW-day, sample size 16,368, figure values, absence of an interval, interpretation, source contracts, and assumptions. The evidence manifest records JSON SHA-256 `0499fb8fe666d630959bc1e1b194ac2dd73e38b7b42139f2ee8cf7d94957278c` and figure SHA-256 `83317cb3c80644540a07b5b57440419c45c315c243c987cea7cc8922f2e19764`.

A reproduction should use the exact snapshot, analysis code, protocol, registry, windows, and publication cut-off. It must calculate retrospective monthly P10–P90 index widths, not forecast quantiles. Any quantile-policy extension needs separate issue-time evidence and scoring. Licensing is governed by `/legal/data-licensing` and [Voltcast data licensing and redistribution](https://github.com/ossedk/voltcast/blob/main/docs/voltcast/LICENSING.md).

## Disclosure

Analysis and drafting were model-assisted. The difference between observed dispersion and a predictive quantile schedule is disclosed explicitly. This working paper is not peer reviewed. It is not trading advice, financial advice, an investment recommendation, or operational authorization. Voltcast has no live traders or live capital, and no battery action is recommended by the measured width.

## References

- `energy-informatics-storage` — Energy Informatics. [Risk and reward: evaluating household energy storage for optimizing demand-side flexibility under dynamic tariffs](https://doi.org/10.1186/s42162-025-00602-9). Kind: peer-reviewed.
- `acer-retail-2025` — 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). Kind: official.
- `iea-electricity-2026` — International Energy Agency. [Electricity 2026](https://www.iea.org/reports/electricity-2026). Kind: official.
- `dynamic-tariff-viability` — Advances in Applied Energy. [Assessing the conditions for economic viability of dynamic electricity retail tariffs for households](https://doi.org/10.1016/j.adapen.2024.100174). Kind: peer-reviewed.
- `volt-research-content` — Voltcast. [Voltcast Research Content Plan](https://github.com/ossedk/voltcast/blob/main/docs/voltcast/RESEARCH-CONTENT-PLAN.md). Kind: canonical.