---
id: VOLT-HOME-WP-031
title: "What round-trip efficiency makes day-ahead battery arbitrage worthwhile?"
slug: what-round-trip-efficiency-makes-day-ahead-battery-arbitrage-worthwhile
description: "A normalized BESS-v2 screening result translates captured day-ahead spreads into an implied efficiency threshold while preserving the limits of perfect foresight and wholesale-only accounting."
published: 2026-08-30
cluster: "Home batteries"
status: measured
evidence_url: /research-data/home-papers/what-round-trip-efficiency-makes-day-ahead-battery-arbitrage-worthwhile.json
figure_url: /research-media/home-papers/what-round-trip-efficiency-makes-day-ahead-battery-arbitrage-worthwhile.webp
figure_alt: "Chart for What round-trip efficiency makes day-ahead battery arbitrage worthwhile?: median implied round-trip loss tolerance, 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

This paper asks how much round-trip energy loss a battery could absorb before the gross day-ahead spread represented by Volt BESS-v2 disappears. The frozen evidence reports a median implied round-trip loss tolerance of 28.19470715689887% over 16,368 index observations. As an algebraic translation of that evidence value, the corresponding retained-energy level is about 71.81%; it is not a separately estimated engineering threshold. The result is descriptive and applies to a normalized, wholesale-energy-only, perfect-foresight index. It excludes retail taxes, network charges, device-specific controls, standby consumption, degradation, installation costs, financing, export restrictions, and forecast error.

The useful conclusion is therefore conditional. Within the BESS-v2 abstraction, captured spreads were large enough at the median observation to cover the stated conversion loss. That does not mean every device above roughly 72% round-trip efficiency is economic. A real decision must also clear degradation and all other costs, and it cannot know the realized day-ahead curve before scheduling. The evidence object calls the threshold a screening statistic rather than a device warranty; this paper keeps that distinction throughout.

## Plain-language answer

On this index, the central loss-tolerance estimate is 28.19%. Put differently, retaining about 71.81% of charged energy would exhaust the loss allowance implied by the estimate before any other cost is counted. A battery with higher efficiency loses less energy, so more of the captured price spread remains. A battery with lower efficiency loses more, so the same buy-low, sell-high pair may no longer cover the energy that disappears during charging and discharging.

That is only the first filter. “Worthwhile” cannot be decided from efficiency alone. BESS-v2 is normalized per unit of power and uses the realized day-ahead curve with perfect foresight. It is not a household bill simulation or a measured device cashflow. It does not deduct tariff mark-ups, taxes, grid fees, fixed charges, cycling wear, inverter clipping, auxiliary use, financing, or installation. It also assumes the best intervals can be identified with information that is unavailable when an actual schedule is set. The 28.19% value is best read as a gross spread budget: all efficiency losses and omitted costs must fit inside that budget.

## Research question

The narrow research question is: given the spreads captured by the frozen BESS-v2 day-ahead index, what round-trip loss would consume the gross indexed value? This framing is deliberately different from asking which battery to buy, what payback period to expect, or whether a particular tariff permits profitable export. Those questions require asset, tariff, and control data absent from the evidence.

Efficiency matters because charging purchases more energy than discharging returns. If a schedule buys in a low-price interval and sells or offsets consumption in a high-price interval, conversion losses raise the effective acquisition cost of each delivered unit. The break-even loss depends on the selected spread, not just the efficiency label on a datasheet. The study therefore derives an implied tolerance from observed indexed spreads instead of imposing a universal engineering cut-off. The registered method family covers dispatch optimization with efficiency, degradation, foresight, concentration, and emissions sensitivities; this paper isolates the efficiency component and does not borrow findings from the other sensitivities.

## Data and provenance

The evidence was generated from a single SELECT-only production snapshot with SHA-256 `7e97489fc8528c8cc8c38830e05b48d949ce1f67b98575dff26f5d7c321e4c67`. The transaction was read-only and used a 180-second statement timeout. The frozen source contracts are `day_ahead_prices`, `bess_index_daily`, `bess_forecast_daily`, `generation_mix`, and `capture_stats`. Their inclusion defines the permitted corpus; it does not imply that every table contributes directly to this one scalar result.

Three windows are disclosed. Daily prices cover 2021-01-01 through 2026-08-29. Detailed intervals cover 2025-10-01 through 2026-08-29. The long-history boundary is 2015-01-01 through 2026-08-29. The publication cut-off is 2026-08-30T00:00:00Z. The primary result has sample size 16,368 and unit percent. The evidence, figure, protocol, registry, and analysis code are frozen separately; the analysis-code hash is `57c57de79cdab2b5b6d6c54c485cb5162598c5ba0b0bfe995da40d75e6c52ba9`, the protocol hash is `adb36bf6b447af9f96339249b8becaefc20422499cca1977242866347a97bd4b`, the registry hash is `7bcb91d7476d0a69fe9fa75a5c7782f8117e0153f82f9112b7e1d307d3943717`, and the source-registry hash is `07949550ac443ff673fda5c0209b99f137544f3ffecf6775f109bb9d09663bd6`.

## Method

Volt BESS-v2 is a normalized day-ahead arbitrage index. “Normalized” means values describe a common asset convention per unit of power rather than the economics of a named installation. This supports comparisons across zones, dates, and durations without pretending that a 5 kW home system, a larger commercial unit, and a market-scale asset have identical constraints. Scaling an index by nameplate power is arithmetic; it does not add site-specific operating rules or make the scaled output observed cash.

“Perfect foresight” means interval selection is evaluated with the realized day-ahead price curve already known. That creates a ceiling-like benchmark: the optimizer can select ex post low and high intervals that a live controller would have to identify from a pre-decision forecast. The method then relates the captured spread to round-trip energy loss and reports the median implied loss tolerance. The registered assumptions restrict the accounting to wholesale energy and exclude retail taxes and network charges. No inferential significance claim is made. The evidence says within-family Holm control applies to inferential claims, but this result is descriptive.

## Results

The evidence value is 28.19470715689887%, which this paper rounds to 28.19% when fewer digits improve readability. Subtracting that recorded loss fraction from 100% gives the approximate 71.81% retained-energy translation. The empirical quantity is the loss tolerance; the efficiency figure is a direct arithmetic restatement, not a second model output. The evidence interpretation says the threshold is derived from observed captured spreads and is a screening statistic, not a device warranty.

The practical shape of the result is asymmetric. Improving round-trip efficiency above the translated threshold preserves more gross indexed value, but it does not guarantee positive net value because omitted costs can consume the remainder. Falling below it means conversion loss alone would use more than the median allowance under the index convention. Even that is not a universal failure statement: the distribution varies by zone, date, duration, and spread opportunity, while the reported statistic is a median across 16,368 observations.

The regenerated evidence reports `bootstrap_95_interval: null` and identifies the interval method as “not reported for this estimand.” No confidence interval is therefore attached to the 28.19470715689887% median loss tolerance. The figure publishes P10 = 10.147169936627577%, median = 28.19470715689887%, and P90 = 59.378523910798364%; these are distribution quantiles, not uncertainty bounds around the median.

## Robustness and placebo checks

The corpus protocol controls family-level inference with Holm adjustment, but no unadjusted significance claim is made here. That is appropriate because the question is a descriptive break-even transformation, not a test of a treatment effect. Robust interpretation comes first from respecting the frozen assumptions: wholesale-only prices, normalized BESS-v2 operation, and perfect foresight. Changing any of those assumptions changes the economic meaning even if the arithmetic code is unchanged.

Several useful placebo-style comparisons are conceptual rather than reported empirical results. A zero-loss calculation would show the gross spread before efficiency cost; adding degradation would test whether the remaining allowance survives cycling wear; replacing perfect foresight with a frozen pre-decision forecast would test schedule attainability. The current JSON does not provide those counterfactual values for this paper, so none is invented. Because no estimator-valid interval is reported, the paper makes no precision claim beyond the descriptive figure quantiles.

## Limitations

The largest limitation is information timing. Perfect foresight can choose intervals after their realized prices are known, whereas a deployable schedule must be fixed from a forecast or updated under explicit market and device rules. Forecast error can choose the wrong charge or discharge interval, compress the spread, or reverse it. BESS-v2 therefore describes an opportunity ceiling, not a promise of attainable operation.

The accounting boundary is equally important. Retail taxes and network charges are excluded, as are contract-specific import and export prices. Round-trip efficiency may vary with power, temperature, state of charge, age, and inverter operating point; a single number cannot capture that surface. The evidence contains no installation cost, financing, warranty, cycle-life curve, standby load, reserve requirement, backup-energy constraint, or local export cap. No interval is reported for the estimand. These omissions mean the result should screen assumptions, not close an investment decision.

## Practical implication

For an analyst, the 28.19% loss tolerance is a transparent first-stage stress test. Start with the indexed gross spread, apply the device’s efficiency convention carefully, then deduct degradation, auxiliary use, tariff spreads, taxes, network charges, and any export discount. Replace perfect foresight with the actual forecast and decision clock. If the case works only when all omitted frictions are set to zero, it is not robust.

For a home-energy controller, efficiency should be treated as one constraint in a full schedule rather than a universal enable/disable threshold. A controller should preserve native market intervals, enforce state-of-charge and power limits, and decline marginal cycles when the forecasted spread does not cover explicit costs. The evidence does not authorize autonomous market action and does not identify a product recommendation. It offers a reproducible normalized benchmark against which more realistic schedules can be compared.

## Reproducibility

Retrieve the public evidence JSON at `/research-data/home-papers/what-round-trip-efficiency-makes-day-ahead-battery-arbitrage-worthwhile.json` and verify that `paper_id`, slug, title, status, assumptions, windows, source tables, result, and provenance match this paper. The evidence status is `measured`. Verify the evidence file against SHA-256 `fe3ec994e02f5d475648d904cc91dc72cf5aa3401b26037948d62744993ee786` and the figure against `a93bb49f3b5b83245f2e2565345e484f0171ea7e79c76616aa63f3e1981e822e` in the evidence manifest.

A clean reproduction should use the protocol and registry hashes above, the exact snapshot hash, read-only access, the publication cut-off, and the registered source contracts. It should report the primary metric as loss tolerance in percent and clearly label any efficiency threshold as derived arithmetic. Reproduction should fail closed if the snapshot, code, protocol, or registry identity differs. Licensing is governed by `/legal/data-licensing` and [Voltcast data licensing and redistribution](https://github.com/ossedk/voltcast/blob/main/docs/voltcast/LICENSING.md); only the frozen public aggregates are presented here.

## Disclosure

Analysis and drafting were model-assisted. Sources, code identity, assumptions, evidence hashes, and limitations are disclosed so readers can separate recorded output from interpretation. This working paper is not peer reviewed. It is educational research, not trading advice, financial advice, an investment recommendation, or a device warranty. Voltcast has no live traders or live capital, and this home-battery index does not authorize any strategy or capital action.

## 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.