Pith. sign in

REVIEW 4 major objections 3 minor

Shared battery-bidding algorithms cause rivals to internalise one another's profits, forgoing profitable dispatch and raising prices once a provider's near-margin share exceeds ~30%.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

T0 review · grok-4.5

2026-07-15 01:38 UTC pith:S5DYVQLD

load-bearing objection Abstract-only: clean empirical claim of provider-layer soft collusion in NEM batteries, but the load-bearing separation of joint maximisation from shared information is uncheckable without methods. the 4 major comments →

arxiv 2607.13002 v2 pith:S5DYVQLD submitted 2026-07-14 econ.GN q-fin.EC

Shared Bidding Algorithms and Competition: Evidence from Electricity Markets

classification econ.GN q-fin.EC
keywords shared algorithmsalgorithmic collusionelectricity marketsbattery storageautobiddingmarket powerAustralian NEMnear-margin capacity
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

This paper asks whether competitors who outsource bidding to the same third-party algorithm end up acting as if they care about one another's profits. Using high-frequency bid data from the Australian National Electricity Market, where every battery can be linked to an identifiable autobidding provider, the authors show that bids from the same provider co-move more strongly once a disclosure reform made the common scarcity state easier to observe. To isolate joint profit maximisation from mere common information, they estimate each battery's dynamic value of stored energy and reclear the market under counterfactual bids. Owner-level profits alone cannot explain the data: batteries forgo profitable dispatch precisely when that dispatch would depress the prices earned by same-provider batteries owned by rival firms, and the estimated weight placed on those rivals' profits is close to one. The pattern appears only when a provider controls more than roughly 30% of near-margin battery capacity (about 20% of installed capacity). On the current fleet the conduct costs consumers an annualised $5.5 million, and it is organised at the algorithm-provider layer rather than the asset-owner layer that conventional concentration screens examine.

Core claim

Owner-level profits cannot rationalise observed bidding: batteries systematically forgo profitable dispatch when that dispatch would lower prices for same-provider batteries owned by rivals, and the estimated weight on those rivals' profits is close to one. The conduct appears only once a provider's share of near-margin capacity exceeds roughly 30% and costs consumers $5.5 million a year.

What carries the argument

Dynamic value-of-stored-energy estimates combined with market re-clearing under counterfactual bids: these isolate the residual pattern of forgone profitable dispatch that identifies a weight near one on same-provider rivals' profits, separating joint maximisation from co-movement caused by shared information.

Load-bearing premise

That estimating each battery's dynamic value of stored energy and re-clearing the market under counterfactual bids cleanly separates joint profit maximisation from remaining confounds such as shared information or unobserved costs.

What would settle it

Re-estimate the weight on same-provider rivals' profits after an independent recalculation of opportunity costs of stored energy (or after a natural experiment that breaks the shared-algorithm channel while holding ownership fixed); a weight near zero would falsify the joint-maximisation claim.

Watch this falsifier — get emailed when new claim-graph text bears on it.

If this is right

  • Ownership-based concentration screens miss the relevant market-power layer: algorithm-provider market share, not asset ownership, organises the conduct.
  • The same information that improves efficient arbitrage also enables synchronised bidding among competitors who share a provider.
  • Conduct is thresholded: it appears only above ~30% provider share of near-margin capacity (~20% installed), giving a concrete screen for intervention.
  • On the present Australian battery fleet the annualised consumer cost is $5.5 million, a lower bound that scales with fleet size.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • Analogous shared-algorithm effects could appear in other continuous-auction settings (advertising, cloud, ride-hail) once a single provider's near-margin share crosses a similar threshold.
  • Disclosure reforms that improve common-state observability may have an unintended side-effect of tightening algorithmic collusion when providers are shared.
  • Regulators could require algorithmic-provider registration and near-margin share reporting as a complement to traditional ownership screens.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit.

Referee Report

4 major / 3 minor

Summary. The paper studies whether shared third-party bidding algorithms cause competing battery operators in the Australian National Electricity Market to internalise one another’s profits. Using 5-minute bid data linked to identifiable autobidding providers, it documents co-movement of same-provider bids that strengthens after a disclosure reform improved observability of the common scarcity state. To separate shared-information co-movement from joint profit maximisation, the authors estimate each battery’s dynamic value of stored energy and reclear the market under counterfactual bids. They conclude that owner-level profits cannot rationalise observed bidding: batteries forgo profitable dispatch when it would depress prices earned by same-provider batteries owned by rivals, with an estimated weight on those rivals’ profits close to one. The conduct appears only where a provider’s share of near-margin battery capacity exceeds roughly 30% (installed share ~20%), costs consumers an annualised $5.5 million on the current fleet, and arises at the algorithm-provider layer rather than the asset-owner layer.

Significance. If the identification strategy holds, the result is policy-relevant: shared algorithms can generate effective multi-firm internalisation that ownership-based concentration screens miss. The design combines high-frequency bid-level data, a disclosure-reform natural experiment, structural estimation of storage opportunity cost, and market reclearing—tools that, when correctly implemented, can speak to algorithmic competition beyond electricity. The explicit claim that conduct is provider- rather than owner-level is a useful contribution to the antitrust literature on common software and third-party intermediaries.

major comments (4)
  1. The central identification claim—that estimating each battery’s dynamic value of stored energy and reclearing under counterfactual bids cleanly separates shared-information co-movement from joint profit maximisation—is load-bearing for the headline weight near one, the ~30% near-margin threshold, and the $5.5M consumer cost. From the abstract alone it is not possible to verify that residual forgone dispatch is free of confounds (misspecified opportunity cost of storage, unmodelled physical or market constraints, common shocks not absorbed by the dynamic value, or provider-level information that is not pure joint maximisation). This step must be fully specified, with transparent assumptions and robustness, before the structural magnitudes can be accepted.
  2. The claim that ‘owner-level profits cannot rationalise observed bidding’ and that the estimated weight on same-provider rivals’ profits is ‘close to one’ is a fitted structural parameter whose credibility rests on the counterfactual reclearing design. Without the full methods, objective function, and estimation details, it is unclear whether the weight is identified separately from remaining co-movement or is partly mechanical. The paper needs to show that alternative opportunity-cost specifications and information sets do not drive the weight to one by construction.
  3. The finding that conduct arises only above a provider near-margin capacity share of roughly 30% (installed ~20%) is presented as a sharp threshold. Thresholds of this kind are sensitive to how ‘near-margin’ capacity is defined and to sample composition. The manuscript should document the definition, report continuous share specifications, and show that the threshold is not an artefact of binning or of a small number of high-share provider-hours.
  4. The annualised consumer cost of $5.5 million is a key policy number. It depends on the counterfactual reclearing and on the fleet and period used for annualisation. The paper should report the exact counterfactual (what bids are replaced, whose profits are maximised, how prices and quantities are recomputed), confidence intervals or sensitivity ranges, and whether the cost is concentrated in a few events or is diffuse.
minor comments (3)
  1. Abstract only: notation for the weight on rivals’ profits, the precise definition of ‘near-margin’ capacity, and the mapping from installed share to near-margin share should be stated clearly when the full text is available.
  2. Abstract only: the disclosure reform is used as evidence of shared-information co-movement; the full paper should date the reform, describe what became observable, and show pre/post co-movement with appropriate controls so that the reform is not confounded with concurrent market changes.
  3. When full text is available, figures linking provider share to the estimated weight and to forgone dispatch would help readers assess the 30% threshold claim.

Circularity Check

0 steps flagged

No circularity detectable from the abstract: claims rest on external NEM bid data, a disclosure reform, and a structural counterfactual exercise rather than definitional or self-citation reductions.

full rationale

Abstract-only review finds no load-bearing step that reduces by construction to its own inputs. The paper reports co-movement of same-provider bids (stronger after a disclosure reform), then uses estimated dynamic values of stored energy and market reclearing under counterfactual bids to argue that owner-level profits cannot rationalise observed dispatch and that the weight on same-provider rivals' profits is near one above a ~30% near-margin share, with an associated $5.5M annualised consumer cost. These are empirical/structural claims grounded in external market data and a counterfactual exercise; the estimated weight is a fitted structural parameter of interest (standard in structural IO), not a quantity forced by construction from a prior fit of the same object. No equations, self-citations, uniqueness theorems, or ansatzes appear in the abstract that would allow a reduction of the form Eq. X = Eq. Y by definition or fitted-input-as-prediction. Residual identification risk (whether dynamic-value estimation cleanly separates shared information from joint maximisation) is a correctness concern, not circularity. Per the default and hard rules, score 0 with empty steps is the warranted finding.

Axiom & Free-Parameter Ledger

2 free parameters · 3 axioms · 0 invented entities

Abstract-only: free parameters and axioms are those implied by the identification design. The weight on rivals' profits and the 30% threshold are estimated objects, not free inputs. Domain assumptions about market clearing, battery dynamics, and the mapping from bids to dispatch are required for the counterfactual reclearing step. No new physical entities are invented.

free parameters (2)
  • weight on same-provider rivals' profits
    Structural parameter recovered from the gap between observed bids and owner-only optimal bids; reported as 'close to one' and is the central quantitative claim rather than an input, but its estimation embeds modeling choices not visible in the abstract.
  • near-margin capacity share threshold (~30%)
    Empirical cutoff above which the joint-profit conduct is found; may be data-driven rather than pre-specified, which matters for inference but is not a free scale factor fitted to force the result.
axioms (3)
  • domain assumption Market reclearing under counterfactual bids recovers the dispatch and prices that would have obtained under owner-only profit maximisation.
    Load-bearing for the claim that owner-level profits cannot rationalise observed bids; invoked in the abstract's identification paragraph.
  • domain assumption Each battery's dynamic value of stored energy can be estimated well enough to define profitable vs forgone dispatch.
    Required to label bids as forgoing profitable dispatch; abstract states this estimation is performed but does not detail the dynamic program.
  • domain assumption Co-movement after the disclosure reform and residual forgone-profit patterns identify internalisation of rivals' profits rather than remaining common shocks.
    The separation strategy rests on this; abstract presents it as the design purpose.

pith-pipeline@v1.1.0-grok45 · 6142 in / 2634 out tokens · 28579 ms · 2026-07-15T01:38:55.867583+00:00 · methodology

0 comments
read the original abstract

Competing firms increasingly delegate pricing and bidding decisions to algorithms supplied by the same third-party providers. We study whether a shared algorithm leads competitors to internalise one another's profits, using data from the Australian National Electricity Market, where every battery's bids are observed at 5-minute frequency and can be linked to an identifiable autobidding provider. Bids constructed by the same provider co-move, and do so more strongly after a disclosure reform made the common scarcity state easier to observe: the same information that steers batteries towards efficient arbitrage also synchronises the bids of competitors who share a provider. To separate co-movement due to shared information from joint profit maximisation, we estimate each battery's dynamic value of stored energy and reclear the market under counterfactual bids. Owner-level profits cannot rationalise observed bidding: batteries forgo profitable dispatch where it would depress the prices earned by same-provider batteries owned by rival firms, and the estimated weight on those rivals' profits is close to one. We find evidence of this conduct only where a provider's share of near-margin battery capacity exceeds roughly 30%, corresponding to an installed share of roughly 20%. The identified conduct costs consumers an annualised $5.5 million on the current fleet, and it arises at the level of the algorithm provider rather than the asset owner, a layer that ownership-based concentration screens do not capture.

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.