{"id":"c9453d1e-0e8d-4423-800f-afb4f166e9c3","arxiv_id":"2505.05153","paper_version":3,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":5.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":1,"one_line_summary":"Under a one-price balancing market, wind farms should cap how far their day-ahead bid deviates from the production forecast, because extreme bids can shift balancing prices and erase profits.","lead":"This paper studies how an offshore wind farm should bid in European day-ahead electricity markets when imbalances are settled at one price. It derives a risk-limited bidding rule and shows, using Belgian market data, that extreme bids can move the balancing price and cause losses.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Equation (11) double-counts the wind farm's historical open position: the counterfactual system imbalance should be ψ_t + 4(y_hist_t − y_t), not ψ_t + 4(E_t − y_t), so the price-impact results in Figs. 3b and 4 are not yet supported.","rationale":"The analytical contribution in Section III is clean: the objective is linear in y_t, and the variance constraint (1b) correctly yields the interval in Eq. (8), so the binary decision in Eq. (10) follows under the stated price-taker assumptions. The parameter count is one (α) and the derivation is essentially parameter-free. The weakest point is the ex-post price-impact evaluation. I checked the counterfactual accounting in Eq. (11) and the reader's concern is exactly right: for a total system imbalance series, Eq. (11) double-counts the farm's realized imbalance. This is not a disagreement with market conventions or a matter of style; it is an internal accounting inconsistency in the evaluation methodology. The correction requires the farm's historical day-ahead schedules, which the paper does not provide, so the empirical claims are currently unverified. That said, the error is localized to Section IV-B, and the methodological core—the analytical bidding rule and the merit-order framework—can still stand if the counterfactual is corrected. A corrected analysis could plausibly preserve or overturn the qualitative conclusions, so the appropriate verdict remains CONDITIONAL rather than ACCEPT or REJECT. No additional load-bearing concern emerged from the analytical derivations, the assumptions, or the data description.","tokens_in":10552,"tokens_out":7985,"duration_ms":84062,"concrete_test":"Obtain the farm's historical day-ahead bid schedule y_hist_t, recompute the counterfactual system imbalance as ψ_t^α = ψ_t + 4(y_hist_t − y_t), rebuild the balancing prices from the historical merit-order curves, and regenerate Figs. 3b and 4. If the all-or-nothing strategy still shows sustained long-term losses and the risk-constrained strategy still beats point-forecast bidding, the empirical conclusion survives; if the ordering or the sign of the profits changes, the headline claim is an artefact of double-counting the farm's historical open position.","verdict_should_be":"UNCHANGED","load_bearing_attack":"Equation (11) defines the counterfactual system imbalance under strategy y_t as ψ_t^α = ψ_t + 4(E_t − y_t), where ψ_t is the historical system imbalance. If ψ_t is the actual total imbalance recorded by the TSO, it already contains the wind farm's realized imbalance, which is 4(E_t − y_hist_t), with y_hist_t the farm's historical day-ahead contracted volume. Replacing the historical bid by the strategy bid y_t changes the total imbalance by 4(y_hist_t − y_t), not by 4(E_t − y_t). Equation (11) therefore adds an extra term 4(E_t − y_hist_t), counting the farm's realized open position twice. This error is largest exactly in the hours where the strategy is aggressive, so it can flip the sign and magnitude of the simulated system imbalance and materially change the merit-order re-pricing. Since this equation drives all price-impact results, including the claims that the all-or-nothing strategy causes long-term losses and that the risk-constrained strategy remains profitable, those empirical conclusions are unsupported unless the authors either demonstrate that ψ_t excludes the farm's own imbalance or re-run the evaluation with the corrected formula. The paper provides no historical schedule data and does not define y_hist_t, so the correction cannot be checked from the current manuscript.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper studies day-ahead bidding for a wind farm under a one-price balancing market. It first shows that, without risk constraints, the optimal bid is an all-or-nothing strategy: bid the full installed capacity when the expected day-ahead price exceeds the expected balancing price, and bid zero otherwise. To control the risk of large open positions, the authors add a constraint on the expected squared deviation between production and bid, governed by a risk certificate α, and derive in Section III the analytical solution in Eq. (10): the optimal bid is the upper or lower endpoint of a prediction interval around the point forecast, capped by capacity, depending on the sign of the expected price difference. The proposal is evaluated on Belgian market data and an offshore wind farm for the first half of 2024, comparing a no-price-impact evaluation using historical balancing prices with a price-impact evaluation that recalculates balancing prices from a shifted system imbalance (Eq. (11)) and a merit-order curve built from historical balancing volume bids. Under the price-impact evaluation, the all-or-nothing strategy yields long-term losses, while a risk-constrained strategy with α̃=25% still outperforms point-forecast bidding. The paper also compares profit distributions to show that ignoring price impact underestimates risk.","tokens_in":10894,"tokens_out":8364,"duration_ms":84125,"significance":"The analytical derivation is clean and correct under the stated assumptions, and the risk-constrained closed form is a useful, interpretable extension of the point-forecast baseline. The empirical price-impact framework is a valuable step toward realistic backtesting, and the use of real Belgian market data together with an offshore wind farm adds practical relevance. The paper is generally transparent about its assumptions and provides an appendix for the unequal-resolution case. However, the central empirical claims—that the all-or-nothing strategy produces long-term losses and that the α̃=25% strategy remains profitable under price impact—depend critically on the counterfactual system imbalance in Eq. (11), which appears to double-count the wind farm's historical open position. Until this is resolved, these conclusions are not supported.","major_comments":[{"comment":"The counterfactual system imbalance ψ_t^α = ψ_t + 4(E_t − y_t) appears to double-count the wind farm's historical open position. If ψ_t is the total system imbalance reported by the TSO, it already contains the wind farm's actual imbalance, which in the same units is 4(E_t − y_t^hist), where y_t^hist is the farm's historical day-ahead volume. Replacing the historical bid by the strategy bid y_t should change the total imbalance by 4(y_t^hist − y_t), not by 4(E_t − y_t). The paper never defines y_t^hist or demonstrates that ψ_t excludes this farm's imbalance. Because Eq. (11) drives the recalculation of balancing prices in Figs. 3b and 4, the claims that the all-or-nothing strategy causes long-term losses and that the α̃=25% strategy remains profitable under price impact are not supported without either a justification that ψ_t excludes the farm's own deviation or a re-evaluation using the corrected formula. This is a load-bearing issue for the paper's main empirical conclusions.","section":"Section IV-B, Eq. (11)"},{"comment":"The evaluation does not report the historical day-ahead contracted volume y_t^hist of the wind farm, which is needed both to assess the magnitude of the double-counting in Eq. (11) and to reproduce the price-impact results. Please provide this data or explicitly state the assumption that the historical system imbalance excludes the farm's open position; without this, the empirical claims cannot be independently verified.","section":"Section IV-A and IV-B"}],"minor_comments":[{"comment":"The normalization of the risk certificate α̃ is not defined precisely; please specify the exact mapping between α and α̃ (e.g., α̃ = (α − Var(E_t))/(α_all − Var(E_t))) so that the reported values can be translated back to constraint (1b).","section":"Section IV-A"},{"comment":"The y-axis label 'Frequency [.]' is uninformative; if these are probability densities or relative frequencies, please state the estimator and label the axis accordingly.","section":"Figure 4"},{"comment":"The merit-order price calculation is described only in words; please add a step-by-step algorithm that explains how the balancing price is selected from the sorted bids, including the treatment of ties and the simultaneous activation of aFRR and mFRR volumes.","section":"Section IV-B"},{"comment":"The currency symbol appears as 'C/MWh' in several places, which is likely a rendering error; if the intended unit is euros, please use '€/MWh' consistently.","section":"Throughout"}],"recommendation":"major_revision","confidential_remarks":"The main issue is the counterfactual imbalance formula in Eq. (11). I would ask the authors to provide the historical day-ahead schedule of the test wind farm or otherwise justify that ψ_t excludes its imbalance; if they cannot, the price-impact results in Figs. 3b and 4 should be re-run with the corrected formula or removed. The analytical contribution is sound and could stand on its own, so a revision is appropriate rather than rejection."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Colleague,\n\nThe paper is worth a look for the analytical part: under a one-price balancing scheme, the unconstrained optimum is all-or-nothing, and with a variance constraint on the open position you get a transparent formula that clamps the bid to the point forecast plus or minus a deviation term. That derivation is correct, clearly presented, and extends Browell's risk-constrained idea in a useful way. The writing is crisp and the authors are upfront about the price-taker assumption and about using realized prices, so the results are an upper bound.\n\nThe genuinely new piece is the ex-post price-impact evaluation: they propose reconstructing the balancing price from historical merit-order bid curves after shifting the system imbalance by the strategy's open position. That is a sensible idea and, to my knowledge, not in the prior literature.\n\nThe soft spot is in that very shift. Equation (11) defines the counterfactual imbalance as ψ_t + 4(E_t − y_t), where ψ_t is the historical system imbalance. If ψ_t comes from the TSO, it already includes the farm's actual imbalance, E_t − y_hist_t. Replacing the historical bid with the strategy bid changes the farm's contribution by y_hist_t − y_t, not by E_t − y_t. So the formula adds the farm's realized open position a second time. The paper never defines y_hist_t or argues that ψ_t excludes the farm, so the price-impact results in Figures 3b and 4 are not supported as written. This is not a minor detail; it drives the headline claim that the all-or-nothing strategy produces long-term losses.\n\nThere are smaller quibbles: the risk certificate is normalized and α̃ = 25% is selected after a sweep, which is mild post-hoc but not damning; no code is provided, though data sources are public. The analytical contribution stands independently of the empirical error, but the empirical story needs correction or a defensible justification for the imbalance shift.\n\nWho gets value: a wind-energy trading specialist, or an editor deciding whether to send it out. I'd give it a serious referee, but only with the expectation of heavy revision. If the authors can fix the counterfactual or show that ψ_t excludes the farm, the paper becomes a solid contribution. As it stands, it is not publishable in its current form.","headline":"A clean analytical bidding rule undermined by an empirical counterfactual that double-counts the wind farm's open position; the paper deserves a referee but not publication as is.","tokens_in":11342,"tokens_out":2797,"would_cite":false,"duration_ms":26467,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"Under a one-price balancing scheme, the expected-profit-maximizing day-ahead bid for a wind farm is all-or-nothing; adding a risk constraint on balancing positions makes the optimal bid the point forecast shifted by sqrt(alpha…","keywords":["day-ahead bidding","one-price balancing scheme","wind farm","stochastic optimization","risk certificate","balancing price","price impact","offshore wind"],"falsifier":"Compare Eq. (11)'s counterfactual prices with an alternative shift that subtracts the farm's historical position, e.g. $\\psi_t+4(y_t^{\\mathrm{hist}}-y_t)$, on the same Belgian data; if the all-or-nothing strategy's long-run loss disappears or reverses, the price-impact result depends on double counting the farm's imbalance. A second check: rerun the evaluation on a smaller farm whose open positions are negligible relative to system imbalance, where the paper's own prediction is that the no-price-impact and price-impact profit curves should nearly coincide.","tokens_in":10388,"feed_emoji":"⚡","tokens_out":6358,"duration_ms":59519,"temperature":0.7,"pith_summary":"This paper tries to establish what a wind-farm operator should bid in the day-ahead market when imbalance settlement uses a one-price scheme, the common European rule in which an open position that helps the system is rewarded and one that hurts it is penalized. It shows that the expected-profit-maximizing bid is an all-or-nothing choice, zero or the installed capacity, because the objective is linear in the bid. Since that extreme strategy is risky when balancing prices are hard to forecast and large open positions can flip the sign of the system imbalance, the paper adds a risk certificate limiting the expected squared open position and derives a closed-form optimum: the point forecast shifted by $\\sqrt{\\alpha-\\mathrm{Var}(E_t)}$, capped at zero and capacity. Using Belgian market data and an offshore wind farm, it then shows that the all-or-nothing strategy loses money in the long run once its price impact on the balancing market is modeled, while the risk-constrained strategy still beats point-forecast bidding. If right, the paper gives operators a transparent, analytically solved bidding rule and a warning that backtests ignoring price impact overstate returns and understate risk.","feed_headline":"One-price balancing turns wind bids all-or-nothing","feed_subtitle":"A risk constraint restores finite bids; on Belgian 2024 data, the extreme strategy loses once its price impact is counted.","key_machinery":"The load-bearing object is the linearized profit expression in Eq. (6): $\\mathbb{E}[\\Lambda_t^{\\mathrm{DA}} y_t + \\Lambda_t^{\\mathrm{B}}(E_t-y_t)] = (\\mathbb{E}[\\Lambda_t^{\\mathrm{DA}}]-\\mathbb{E}[\\Lambda_t^{\\mathrm{B}}]) y_t + \\mathbb{E}[\\Lambda_t^{\\mathrm{B}} E_t]$, which turns bidding into a sign test on the price gap. The risk certificate $\\alpha$ constrains the expected squared open position and, through the variance decomposition $\\mathrm{Var}(X)=\\mathbb{E}[X^2]-\\mathbb{E}[X]^2$, becomes a symmetric interval $\\pm\\sqrt{\\alpha-\\mathrm{Var}(E_t)}$ around the point forecast. The ex-post price-impact analysis uses the counterfactual system imbalance $\\psi_t^\\alpha=\\psi_t+4(E_t-y_t)$ (the factor 4 converts MWh over a 15-minute window to MW) plus the merit-order curve built from historical balancing volume bids to recompute balancing prices.","core_discovery":"The paper's central discovery is that, under a one-price balancing scheme, the day-ahead bidding problem has a boundary solution unless risk is constrained explicitly. Rewriting expected profit as $\\left(\\mathbb{E}[\\Lambda_t^{\\mathrm{DA}}] - \\mathbb{E}[\\Lambda_t^{\\mathrm{B}}]\\right) y_t + \\mathrm{const}$ makes the objective linear in $y_t$, so the optimizer bids $\\beta$ when $\\mathbb{E}[\\Lambda_t^{\\mathrm{DA}}] > \\mathbb{E}[\\Lambda_t^{\\mathrm{B}}]$ and $0$ otherwise. With the risk constraint $\\mathbb{E}[(E_t-y_t)^2]\\le\\alpha$, variance decomposition bounds the feasible bid by $E[E_t]\\pm\\sqrt{\\alpha-\\mathrm{Var}(E_t)}$, and the same price comparison picks the upper or lower endpoint: $y_t^*=\\min\\{E[E_t]+\\Delta_t;\\beta\\}$ or $y_t^*=\\max\\{E[E_t]-\\Delta_t;0\\}$ where $\\Delta_t=\\sqrt{\\alpha-\\mathrm{Var}(E_t)}$. The paper further claims that when the bid is large enough to alter the system imbalance's direction, the all-or-nothing strategy incurs long-term losses, whereas a moderate risk certificate still improves operational profit over point-forecast bidding.","pith_inferences":["Beyond the paper: the all-or-nothing result is purely a consequence of linearity, so it likely extends to any price-taker whose balancing-market reward has the same sign structure, such as storage or flexible demand, whenever they can bid at a price of zero.","Beyond the paper: the expression $\\Delta_t=\\sqrt{\\alpha-\\mathrm{Var}(E_t)}$ says better production forecasts (lower variance) allow larger profitable open positions under the same risk budget, suggesting a direct economic value-of-forecasts calculation that the paper does not perform.","Beyond the paper: the paper notes that price impact makes the optimization bilevel; a testable extension is to learn the optimal bid directly from market data, treating the balancing-price response as an unknown function, and compare with the closed-form rule on the same Belgian dataset."],"forward_implications":["A risk certificate set to $\\alpha=\\mathrm{Var}(E_t)$ reduces the rule exactly to point-forecast bidding, while a sufficiently large $\\alpha$ makes the constraint non-binding and recovers the all-or-nothing strategy.","Any moderate $\\alpha$ between those extremes yields the binary rule $\\min\\{E[E_t]+\\Delta_t;\\beta\\}$ or $\\max\\{E[E_t]-\\Delta_t;0\\}$ for each hour, so the operator's decision reduces to a price-gap forecast plus forecasts of the mean and variance of production.","If the price-impact model is right, backtesting a large offshore wind farm's bidding strategies on historical balancing prices without counterfactual price shifts will overstate cumulative profit and understate left-tail losses.","Across the first half of 2024 in Belgium, the all-or-nothing strategy turns from the most profitable to a long-term loss once price impact is included, while a moderate risk certificate remains superior to point-forecast bidding."],"supporting_citations":[{"why":"Supplies the newsvendor-based two-price bidding baseline that this paper contrasts with the one-price all-or-nothing solution.","marker":"[16]"},{"why":"Proposed constraining the deviation from the point forecast under a single-price balancing market, the approach the risk constraint extends.","marker":"[2]"},{"why":"Provides the Incremental Quantile Functions model used to obtain the conditional mean and variance of wind power production.","marker":"[13]"},{"why":"Motivates the risk-aware treatment of unknown balancing prices and the distributionally robust line of work this paper builds on.","marker":"[15]"},{"why":"Source of the Belgian balancing volume bids used to reconstruct merit-order curves and counterfactual balancing prices.","marker":"[19]"},{"why":"Source of the day-ahead and balancing market data used in the empirical evaluation.","marker":"[11]"}],"fun_headline_variants":["One-price balancing drives wind bidding to all-or-nothing","Risk constraints rescue wind profits from extreme bids","Wind farm losses from all-or-nothing bids in Belgian market","Moderate wind bids beat aggressive strategy under one-price","Price impact dooms extreme wind bids, risk cap wins"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing assumption is that the historical system imbalance data does not already include the wind farm's own past deviations; if it does, the balancing-price impact and the reported losses change.","fun_headline_variants_meta":{"raw":{"variants":["One-price balancing drives wind bidding to all-or-nothing","Risk constraints rescue wind profits from extreme bids","Wind farm losses from all-or-nothing bids in Belgian market","Moderate wind bids beat aggressive strategy under one-price","Price impact dooms extreme wind bids, risk cap wins"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000298,"raw_usage":{"total_tokens":1776,"prompt_tokens":1050,"completion_tokens":726,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":666,"completion_tokens_details":{"reasoning_tokens":648}},"tokens_in":666,"tokens_out":726,"duration_ms":7193,"temperature":1.0,"reasoning_tokens":648,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-15T23:11:16.012356+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Compare Eq. (11)'s counterfactual prices with an alternative shift that subtracts the farm's historical position, e.g. $\\psi_t+4(y_t^{\\mathrm{hist}}-y_t)$, on the same Belgian data; if the all-or-nothing strategy's long-run loss disappears or reverses, the price-impact result depends on double counting the farm's imbalance. A second check: rerun the evaluation on a smaller farm whose open positions are negligible relative to system imbalance, where the paper's own prediction is that the no-price-impact and price-impact profit curves should nearly coincide.","supporting_citations":[{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies the newsvendor-based two-price bidding baseline that this paper contrasts with the one-price all-or-nothing solution."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Proposed constraining the deviation from the point forecast under a single-price balancing market, the approach the risk constraint extends."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Provides the Incremental Quantile Functions model used to obtain the conditional mean and variance of wind power production."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Motivates the risk-aware treatment of unknown balancing prices and the distributionally robust line of work this paper builds on."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Source of the Belgian balancing volume bids used to reconstruct merit-order curves and counterfactual balancing prices."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Source of the day-ahead and balancing market data used in the empirical evaluation."}],"review_version":1}