{"id":"2d04250b-ca4f-4e9f-ab32-836e3e420d25","arxiv_id":"1908.03167","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":4,"one_line_summary":"In a Western European case study, utility-scale storage investments are smaller and located differently when producers wield Cournot market power, with consumers capturing most of the welfare gains.","lead":"This paper models where and how much utility-scale battery storage a profit-seeking merchant or a welfare-maximizing investor would build, in either a competitive or an oligopolistic European power market. It finds that the market structure matters more than the investor's goal, and that consumers usually gain more from storage than the investors do.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The 0/100 MWh investment grid can mask investor-objective differences; the claim that market structure dominates may be an artifact of discretization.","rationale":"The reader's weakest assumption identifies the coarse 0/100 MWh grid and the limited week-count robustness for the Cournot and merchant cases. My review agrees: the most load-bearing concern is that the coarse grid can mask real differences between investor objectives, which directly threatens the central claim that market structure matters more than objectives. The paper is otherwise transparent and the exhaustive search is a reasonable workaround given the stated MIQCQP solver failures. The analysis is a case study, not a theorem, so the appropriate verdict is conditional on robustness checks. I would not change the reader's CONDITIONAL verdict, but the concrete test proposed here should be run before the quantitative results are used for policy. Credit is due for the three-node verification, the central-planning week robustness check, and the thorough sensitivity analysis on investment costs and transmission limits, all of which support the paper's credibility. However, the base-case conclusion that SW and M coincide under both market structures is exactly the kind of result that discrete approximation can manufacture, so it must be tested with a finer grid before the headline is accepted.","tokens_in":24733,"tokens_out":7661,"duration_ms":92064,"concrete_test":"Re-solve cases 4 (SW-CO) and 5 (M-CO), and also cases 2 and 3 (PC), using a finer discrete investment grid, e.g., 50 MWh (or 25 MWh) increments at each of the seven nodes, with the same four representative weeks and the base amortized cost of e50/MWh. Check whether SW-CO and M-CO still choose identical investments at the same locations. If a finer grid causes the investor types to diverge in total size or location, or shrinks the PC-to-CO difference, then the headline claim that market structure dominates investor objectives is not robust to discretization.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim, stated in the abstract and Section 4, is that market competition affects storage investment sizes, locations, and profitability more than the investor's objectives. In the base case (e50/MWh), the reported result is stark: SW-PC and M-PC both invest 100 MWh at n2, n3, n6, while SW-CO and M-CO both invest 100 MWh at n1. Thus the investor objective has zero effect. But this result is obtained by exhaustive search over a grid with only two options per node: 0 or 100 MWh (Section 3.2.2). With only two discrete levels, any divergence between welfare-maximizing and profit-maximizing investment that does not cross the 100 MWh threshold is invisible. For instance, if the merchant's optimal continuous investment under Cournot were 80 MWh at n1 and the welfare-maximizer's were 150 MWh, both would be forced to the same 100 MWh choice, hiding the objective effect. The paper's own sensitivity analysis shows that investor objectives do matter at some cost levels (e.g., at e15/MWh under CO, merchant invests 600 MWh vs. welfare-maximizer's 500 MWh, Section 3.3.3). This suggests the base-case equivalence between SW and M may be an artifact of the coarse grid rather than a structural property. Because the headline comparison depends on the absence of investor-objective differences, the coarseness of the investment grid is the most load-bearing assumption. The week-count robustness check (footnote 10) is reported only for central planning, so it does not address whether the Cournot/merchant results, which are driven by thin arbitrage margins, are stable under different representative weeks.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper analyzes how utility-scale battery storage investment is affected by the investor's objective (welfare maximization vs. merchant profit maximization) and by the competitiveness of the wholesale market (perfect competition vs. Cournot oligopoly). The authors formulate bi-level optimization models, reformulate them as MIQCQPs via an MPPDC approach, and then, for a Western European case study, replace the intractable MIQCQP with an exhaustive enumeration of 128 discrete investment combinations (0 or 100 MWh at seven nodes) and solve the lower-level QP for each combination. In the base case, both investor types invest in the same nodes under each market structure (n2, n3, n6 under perfect competition; n1 under Cournot), leading the authors to conclude that market competition affects storage investment more than investor objectives do. Sensitivity analyses over investment cost and transmission capacity show some investor-type divergences. The paper argues that policy makers need to anticipate producer market power when designing storage support.","tokens_in":25093,"tokens_out":14735,"duration_ms":172458,"significance":"If the headline finding is robust, the paper provides a valuable policy insight: storage-support policies evaluated under perfect competition may be misleading because market power changes whether and where storage is built. The paper's strengths include a careful MPPDC derivation, an explicit and transparent discussion of solver failures, an exact enumeration of the chosen discrete investment set, and a broad set of sensitivity analyses. The central claim, however, rests on a coarse 0/100 MWh investment grid and on four representative weeks, and the paper's own sensitivity results show investor-objective effects at other cost levels. The conclusion is therefore plausible but not yet established at the level of generality claimed in the abstract.","major_comments":[{"comment":"The headline finding that investor objectives do not matter in the base case is an artifact of the extremely coarse investment grid. With only two discrete options per node (0 or 100 MWh), any divergence between the welfare-maximizing size and the profit-maximizing size that does not cross the 100 MWh threshold is invisible: both are forced to the same discrete choice. The paper's own sensitivity analysis (Tables 7 and 8) shows investor-objective effects at other cost levels, e.g., at e15/MWh under Cournot the merchant invests 600 MWh while the welfare-maximizer invests 500 MWh, and at e55/MWh the welfare-maximizer invests while the merchant does not. To support the claim that investor objectives matter less than market structure, the authors should either solve a finer or continuous investment grid for the base case and the diverging cost levels, or otherwise demonstrate that the base-case equivalence is invariant to grid refinement. As it stands, the central comparison may reflect the grid rather than an economic property.","section":"Section 3.2.3 and Section 3.3.1"},{"comment":"The representative-week robustness check is reported only for the central-planning variant. The paper states that with 4, 6, 8, 15, 20, and 25 weeks, social welfare and storage investment size remain similar under central planning, but the headline results concerning Cournot and merchant cases are computed with only four representative weeks. Since the number of weeks affects the temporal price spread and hence the profitability of arbitrage, the authors should repeat the iterative QP enumeration for the SW-CO and M-CO models with, say, 8 and 15 weeks to verify that investment locations, sizes, and the equivalence between welfare-maximizing and merchant investors are stable. Without this check, the claim that market structure dominates investor objectives is not robust to the temporal representation.","section":"Footnote 10 and Section 3.2.3"},{"comment":"The abstract's statement that market competition affects storage investment sizes, locations, and profitability 'more than the investor's objectives' is stronger than the evidence presented. The base case shows exact equality between SW-PC and M-PC and between SW-CO and M-CO, but the sensitivity analysis contains several counterexamples: at e65/MWh under perfect competition the merchant invests less than the welfare-maximizer; at e55/MWh under Cournot the welfare-maximizer invests while the merchant does not; and at e15/MWh under Cournot the merchant invests more than the welfare-maximizer. The paper should either qualify the headline claim to the base case or provide a systematic comparison across cost levels and model dimensions that actually measures the relative influence of market structure versus investor objective.","section":"Abstract and Section 4"}],"minor_comments":[{"comment":"The sentence 'we make an exhaustive search and now that the optimal investments are unique' appears to contain a typo ('now' for 'know'); please also clarify how ties among investment combinations are handled, since multiple optima would make the reported investment pattern non-unique.","section":"Section 3.2.2"},{"comment":"After Eq. (26), the phrase 'The right-hand side of Eq. (26)' is confusing because the bilinear terms appear on the right-hand side; please rephrase to refer to the products of lower-level dual variables and the investment decision in the last four terms of Eq. (26).","section":"Section 2.3.2"},{"comment":"The footnotes refer to 'bold font' to distinguish Cournot terms from perfect-competition terms; since this formatting may be lost in some rendering or printing environments, please label the Cournot terms explicitly in the equations or add a note identifying them by equation number.","section":"Footnotes 1 and 2"},{"comment":"The method for clustering representative weeks relies on a reference listed as 'Submitted for publication' (Reichenberg and Hedenus, 2019); please provide a published version or a more accessible citation, since the temporal aggregation is central to the case study.","section":"References"},{"comment":"The solver name 'BONMINH 1.8' appears to be a typo for 'BONMIN'.","section":"Section 3.2.1"}],"recommendation":"major_revision","confidential_remarks":"The paper is a competent and transparent modeling exercise, and the exhaustive-search workaround for the MIQCQP is clearly described. My main reservation is that the central policy conclusion is built on a very coarse investment grid and a four-week temporal sample, and the authors' own sensitivity analyses show that investor objectives do matter in other cases. These are fixable by additional computation and more careful qualification of the claims, so I recommend major revision rather than rejection. There are no citation or novelty concerns; the self-citations to Virasjoki et al. [2016] and Siddiqui et al. [2019] are appropriate."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"First, the genuinely new thing: this paper numerically reverses Siddiqui et al. [2019]'s analytical result that a merchant overinvests relative to a welfare-maximizer under perfect competition. In a spatial, seasonal Western European setting, the welfare-maximizer invests at least as much, and both investors under Cournot invest less and in different locations. That is a useful counterpoint to the analytical literature.\n\nThe model is carefully built: bi-level, with lower-level PC or Cournot, upper-level welfare-maximizer or merchant, transmission constraints, seasonal VRES. The MPPDC derivation is laid out in detail, and the authors are unusually honest about the MIQCQP failing at scale and their fallback to exhaustive QP enumeration over 2^7=128 size-location combinations. The sensitivity analysis on investment cost and transmission capacity is a genuine strength, as is the welfare decomposition showing consumers usually gain most.\n\nNow the soft spots, and they are real. The exhaustive enumeration uses only two storage options per node: 0 or 100 MWh. With two discrete levels, any divergence between a welfare-maximizer and a merchant that does not cross the 100 MWh threshold is invisible. The paper's own Table 8 shows that at e15/MWh under Cournot, the merchant invests 600 MWh vs the welfare-maximizer's 500 MWh—so investor objectives do matter at some cost levels. The base-case equivalence at e50/MWh may therefore be an artifact of the coarse grid rather than a structural property. The week-count check (footnote 10) is reported only for central planning; the Cournot and merchant cases, which drive the headline comparison, are not checked against more than four weeks. And no code or data are shipped, so the results cannot be independently reproduced.\n\nAre these fatal? No. The qualitative point that market structure (PC vs Cournot) changes investment sizes, locations, and the cost threshold is robust across the sensitivity runs. But the stronger claim that market competition affects outcomes 'more than the investor's objectives' is not fully established, because the grid cannot resolve objective-driven differences except at extremes. The paper's own sensitivity runs undermine the base-case symmetry.\n\nWho should read this: power-system economists and regulators thinking about storage policy in oligopolistic markets, and modelers working on bi-level storage investment. It is a case-study-level contribution, not a general theorem, and the authors acknowledge many limitations (day-ahead only, stylized network).\n\nRecommendation: yes, send it to peer review. A serious referee should push for a finer investment grid and for week-count robustness across all market regimes, but the manuscript is substantial, honest, and worth engaging with.","headline":"A transparent, useful bi-level study of storage investment; the headline claim about market structure vs investor objectives is plausible but rests on a coarse 0/100 MWh grid and unchecked week-count robustness.","tokens_in":25610,"tokens_out":2774,"would_cite":true,"duration_ms":25707,"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":"Market power, not investor goals, decides where utility-scale storage is built.","keywords":["energy storage","variable renewable energy","power market modeling","market power","Cournot oligopoly","bi-level optimization","storage arbitrage","welfare analysis"],"falsifier":"Solve the same exhaustive-search model under Cournot oligopoly with a finer investment grid (e.g., 50 MWh increments at all seven nodes) and with, say, 25 representative weeks, then check whether the optimal location is still Germany and whether the merchant still invests nothing at 50 e/MWh; if the location moves to France or Belgium, or investment becomes positive at the base cost, the paper's headline comparison fails.","tokens_in":24567,"feed_emoji":"🔋","tokens_out":6240,"duration_ms":58971,"temperature":0.7,"pith_summary":"This paper asks whether utility-scale battery storage is built, and where, depends more on whether power producers can exert market power than on whether the investor is a welfare-maximizer or a profit-seeking merchant. Using a bi-level model with a Western European test network, the authors find that under perfect competition the welfare-maximizer (equivalent to central planning) invests the most, in nuclear-dominated France and Belgium, while under Cournot oligopoly investments shrink, may disappear entirely at realistic costs, and the primary location shifts to Germany. Because storage flattens price differences, producers with market power prefer flatter but higher prices, which erodes the arbitrage revenue that storage needs. The upshot for policy is that storage support schemes cannot be designed from a competitive-market model alone; anticipating producer market power changes both the size and the site of socially desirable storage.","feed_headline":"Market power, not investor goals, decides where storage is built","feed_subtitle":"Under Cournot oligopoly, battery sites shift from France and Belgium to Germany, or vanish entirely.","key_machinery":"The load-bearing machinery is a bi-level optimization model: an upper-level investor (welfare-maximizer or standalone merchant) chooses discrete storage sizes at network nodes, while a lower-level independent system operator clears the market under either perfect competition or Cournot oligopoly, with the Cournot case represented by a quadratic objective with an added conjectural-variation term. Because the resulting mixed-integer quadratically constrained quadratic program could not be solved at realistic scale, the authors reformulate the lower level through primal-dual (MPPDC) strong-duality conditions for small instances and, for the Western European case, replace the full model by an exhaustive search: they solve the lower-level QP for every combination of 0 or 100 MWh at the seven demand nodes and pick the investor-optimal outcome. The argument's transmission mechanism is the set of equilibrium nodal price differences - temporal spreads for arbitrage and spatial spreads for congestion - which Cournot competition flattens even as it raises the average price level. This price-flattening is what reverses the location ranking and can eliminate investment altogether.","core_discovery":"The paper's central claim is that the state of competition in the wholesale market is the dominant determinant of storage investment outcomes, outweighing the investor's objective. A welfare-maximizing investor under perfect competition reproduces the central-planning outcome and buys 300 MWh of batteries (100 MWh at each of the French and two Belgian nodes), while an otherwise identical welfare-maximizer facing Cournot producers buys only 100 MWh at the German node, and a profit-maximizing merchant under Cournot also invests only there. At the base amortized cost of 50 e/MWh, the merchant under Cournot does not invest at all; investments under Cournot become profitable only below roughly 15-25 e/MWh, compared with 65-80 e/MWh under perfect competition. Price spreads are the mechanism: Cournot producers raise average prices but flatten their temporal variation, destroying the arbitrage value that storage captures. Consumers, not investors, are the main beneficiaries in every scenario, and a welfare-maximizer may invest even when the storage asset itself loses money, because the system-wide gains exceed the investor's loss.","pith_inferences":["If storage could also earn revenue from ancillary services or reserve markets, the Cournot profitability thresholds would likely drop further and possibly revive merchant investment at locations the day-ahead-only model finds unattractive, changing the policy conclusions.","The exhaustive-search grid (0 or 100 MWh at seven nodes) may hide intermediate optimal sizes; a finer grid under Cournot could reveal that the socially optimal investment is not 'smaller but at the same node' but different in kind, so the headline 'location shifts to Germany' should be tested before being used prescriptively.","The paper's four-week sample is validated only for the central-planning case; applying the same week-count robustness check to the Cournot and merchant cases is a cheap, concrete test of whether the market-power comparison is stable.","Because the merchant is modeled as a price-taker with a minor position, the analysis excludes the case of a strategic merchant whose storage operations themselves influence prices; such a merchant would likely invest less to preserve spreads, reinforcing the market-power result."],"forward_implications":["If the central claim is right, regulatory assessment of storage support must model producer market power; a competitive-market model will overstate storage capacity and point to the wrong nodes (France and Belgium instead of Germany).","Under Cournot oligopoly, the profitability threshold for merchant storage drops dramatically (from roughly 65-80 e/MWh under perfect competition to below about 25 e/MWh), so storage support policies must anticipate that merchant entry needs much cheaper batteries or additional revenue streams.","Consumers are the main beneficiaries of storage investment in all modeled regimes, so consumer-side benefits justify considering storage as a public-good investment even when investor surplus is negative.","A welfare-maximizing investor under perfect competition is equivalent to central planning and thus can serve as a benchmark for evaluating market-based storage deployment.","Reducing transmission capacity under perfect competition leaves the investment pattern unchanged, while removing transmission limits entirely kills storage investment - implying congestion is a necessary driver of storage value in this system."],"supporting_citations":[{"why":"Analytical two-period model showing a merchant invests more than a welfare-maximizer under perfect competition but less under Cournot; the current paper's results are contrasted with this benchmark.","marker":"Siddiqui et al. [2019]"},{"why":"Establishes that producers underuse storage due to the price-smoothing effect, the basis for expecting market power to distort storage investment.","marker":"Schill and Kemfert [2011]"},{"why":"Shows merchant storage ownership can enhance welfare and that producer-owned storage is underused, framing the ownership-structure question.","marker":"Sioshansi [2010]"},{"why":"Prior single-level complementarity model of storage operations, which this paper extends to bi-level investment decisions.","marker":"Virasjoki et al. [2016]"},{"why":"Provides the strong-duality-as-inequality technique used to reformulate the bi-level problem into a single-level MPPDC.","marker":"Huppmann and Egerer [2015]"},{"why":"Supplies the hierarchical clustering method that produces the representative weeks used in the case study.","marker":"Reichenberg and Hedenus [2019]"},{"why":"Introduces the Western European test network used for the case study.","marker":"Neuhoff et al. [2005]"},{"why":"Extends the test network data and demonstrates the discretely-constrained MPEC solution approach adapted in this paper.","marker":"Gabriel and Leuthold [2010]"},{"why":"Co-planning of transmission and merchant storage, providing the comparison point for storage location and profitability findings.","marker":"Dvorkin et al. [2018]"}],"fun_headline_variants":["Market power beats investor goals in storage siting","Competition, not objectives, steers battery investments","Consumers win most, but market power picks storage spots","Storage location driven by competition, not investor aims","How market power silently decides where storage goes"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The results rest on a coarse decision grid - only 0 or 100 MWh at each of seven nodes - and on four representative weeks, with week-count robustness checked only for the central-planning variant; if a finer grid or more representative weeks change which locations are optimal or which investor invests, the comparison between market structure and investor objectives weakens.","fun_headline_variants_meta":{"raw":{"variants":["Market power beats investor goals in storage siting","Competition, not objectives, steers battery investments","Consumers win most, but market power picks storage spots","Storage location driven by competition, not investor aims","How market power silently decides where storage goes"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000268,"raw_usage":{"total_tokens":1639,"prompt_tokens":986,"completion_tokens":653,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":602,"completion_tokens_details":{"reasoning_tokens":580}},"tokens_in":602,"tokens_out":653,"duration_ms":7119,"temperature":1.0,"reasoning_tokens":580,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-14T14:21:18.142551+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Solve the same exhaustive-search model under Cournot oligopoly with a finer investment grid (e.g., 50 MWh increments at all seven nodes) and with, say, 25 representative weeks, then check whether the optimal location is still Germany and whether the merchant still invests nothing at 50 e/MWh; if the location moves to France or Belgium, or investment becomes positive at the base cost, the paper's headline comparison fails.","supporting_citations":[],"review_version":1}