RESEARCH DOSSIER
Cournot Bertrand and Oligopoly Equilibria
Abstract. Oligopoly equilibrium is the outcome of mutual optimization among a finite number of economically significant decision makers whose feasible strategies, information, and payoffs depend on one another. This monograph reconstructs the classical comparison between quantity competition and price competition, derives their characteristic outcomes, and investigates the institutional and mathematical reasons why neither outcome supplies a universal prediction. A common demand and cost environment permits controlled comparisons among monopoly, symmetric and asymmetric Cournot competition, homogeneous Bertrand pricing, price competition with committed capacity, and differentiated-product competition. The analysis distinguishes existence from uniqueness, equilibrium from adjustment, price-cost margins from demonstrated collusion, and structural estimation from causal evidence. It examines historical primary texts, modern equilibrium results, and selected international evidence from automobiles, electricity markets, and retail gasoline, with explicit attention to bid schedules, financial commitments, product substitution, and algorithmic pricing. The original analytical contribution consists of conditional comparative derivations, numerical sensitivity exercises, and a synthesis connecting market institutions to the empirical identification of equilibrium mechanisms. The most important conclusion is that a market's equilibrium is determined by the complete strategic environment, including what firms can actually commit to and how demand is allocated; observed prices and the number of sellers alone are generally insufficient to recover that environment. All numerical illustrations are constructed examples. Published empirical findings are attributed to their original investigators and interpreted within their documented scope.
Keywords: Oligopoly; Cournot equilibrium; Bertrand equilibrium; Nash equilibrium; price competition; quantity competition; capacity constraints; product differentiation; strategic interaction; equilibrium selection; market power; structural identification; industrial organization; algorithmic pricing
Object and Method of Oligopoly Analysis
An oligopoly is a market environment in which a limited number of decision makers internalize economically consequential effects of their actions on market outcomes. The relevant unit is ordinarily the economic firm or pricing decision center, which can control several products, plants, or legal entities. Counting brands can therefore misrepresent strategic concentration. The geographical and temporal boundaries of a market also matter: a generator may be pivotal in a congested region for an hour, although it has a small annual share across the national system.
The analytical object is a game consisting of players, feasible actions, the timing of decisions, the information available at each decision, a rule mapping actions into transactions, and payoffs. This description is economically substantive. A quantity offer is a commitment to supply, a price quote may be merely an invitation to trade, a bid schedule is a contingent offer, and a production capacity is a constraint on future fulfillment. The same notation for market demand can coexist with several different institutional mechanisms.
Let the firms be indexed by i and write their strategies as . Their profits are . A pure-strategy Nash equilibrium satisfies
The comparison holds rivals' strategies fixed when evaluating unilateral deviations; it does not assert that competitors never respond to observed decisions in real time. In a dynamic game the strategy includes contingent responses, and the equilibrium comparison holds those complete plans fixed. The finite-game result in Nash's equilibrium paper (1950) supplies the general strategic conception; its existence result permits mixed strategies and therefore does not establish pure equilibria in arbitrary continuous oligopoly games.
Industrial organization studies such games jointly with the institutions that sustain them. The disciplinary setting is represented by Tirole's (1988) The Theory of Industrial Organization. The specific object of this dossier is narrower: quantity and price competition and the conditions selecting their equilibria. Firm production duality, a full theory of entry, and independent econometric measurement of markups belong to related research programs. They are addressed here only where they alter the oligopoly mechanism.
The investigation uses three levels of evidence. Conditional analytical results follow from explicitly stated demand, cost, and action assumptions. Numerical examples are generated from those equations to clarify magnitudes and failure modes; they are not observations. Historical and empirical evidence is drawn from identified primary texts, journal papers, and institutional records, with access limitations respected. A study-specific empirical claim is retained only within the meaning supported by its source. The distinction matters because an exact derivation under a hypothetical linear demand curve and an estimated effect from an electricity market are different forms of knowledge.
Historical Formation of the Cournot and Bertrand Questions
Cournot and the interdependence of production
Augustin Cournot's 1838 work made the determination of price and output through interacting producers a mathematical research problem. Chapter VII of his Researches on the Theory of Wealth develops producer competition; the historical source consulted here is the reprint of the Bacon translation published in 2008 from the 1971 edition. The intellectual achievement concerns the mutual dependence of individually selected outputs through a common price.
The modern reconstruction assumes that each supplier can choose its output and evaluates the marginal revenue from an additional unit at the total market quantity. Increasing production increases own sales but also depresses the price on units already supplied. When the other firm's output enters this marginal calculation, the profit-maximizing decision becomes a function of the rival's action. Market equilibrium requires simultaneous consistency of these choices.
The familiar use of linear demand and identical constant marginal costs is a pedagogical specialization, not a complete transcription of Cournot's work. Cournot's historical contribution survives the replacement of linear demand by an economically admissible nonlinear schedule, although the strongest existence and uniqueness results require additional restrictions.
Bertrand and customers' allocation of purchases
Joseph Bertrand's 1883 review of Cournot and Walras in the Journal des Savants criticized the treatment of competitors' sold quantities as independently variable and emphasized the attraction of purchasers to a lower-priced supplier. His objection, in the original French text, concerns the economic mechanism that assigns sales when one seller alone reduces its price. It is especially sharp when purchasers regard the competing commodities as equivalent.
The modern Bertrand game formalizes that insight by making prices the independent actions and sales the resulting allocation. With unlimited fulfillment, homogeneous goods, common constant marginal cost, and continuous prices, a common price above cost is vulnerable to slight undercutting. The historical criticism predates the terminology of Nash equilibrium and the exact modern assumptions about tie-breaking, capacity, and feasible prices.
These two developments marked a distinction between choosing production before the determination of a clearing price and selecting transaction prices before the allocation of market demand. A seller can face both types of decisions at different moments. The historical comparison thus opens a question about the sequencing and enforceability of commitments.
From classical examples to game forms
The twentieth-century strategic formulation introduced a language in which both examples can be evaluated by the same unilateral-deviation criterion. Subsequent research expanded the comparison into differentiated products, capacity commitments, contingent schedules, repeated games, and empirical tests. The continuity is methodological: each extension specifies more of the actual institution and then asks what incentive constraints follow.
A quantitative result that differs from a classical prediction may reflect different institutional premises, mistaken behavioral assumptions, inaccurate measurement, or failure of an equilibrium-selection rule. These possibilities must be separated before any historical model is judged against market evidence.
The Cournot Quantity Game
Demand, profit, and best responses
Consider n firms producing a homogeneous commodity. Let aggregate output be , inverse demand be on its positive-price region, with and , and each firm's marginal cost be the constant c. Fixed costs are absent, output is continuously divisible, there are no capacity constraints, and all firms observe the model parameters. For global feasibility one can truncate demand at zero and restrict attention to positive-demand profitable production; the relevant equilibrium is interior for the active symmetric firms.
Each firm maximizes
Write . The best response on the relevant linear segment is
Where residual willingness to pay exceeds cost, the second derivative of profit is , ensuring a strict maximum. If other producers already supply at least , the best response includes zero. When c is zero and demand is truncated, excess free production can generate degenerate zero-payoff equilibria; the unique economically relevant result below is the equilibrium on the positive-price branch, or the globally unique equilibrium under a strictly positive unit cost and these assumptions.
At the symmetric interior equilibrium,
Each firm earns , market operating profit is , and the price-cost wedge decreases with the number of identical unconstrained firms.
Uniqueness in the positive-demand region follows directly from the common price condition: subtracting any two active first-order conditions yields equality of quantities. An inactive firm could profitably enter with a small positive output at an equilibrium price above c, so all n firms must be active. The symmetric solution then determines the sole positive-price equilibrium in this specification.
Welfare and comparative statics
With quasilinear preferences and no externality, efficient output is . Consumer surplus at the Cournot outcome is . Total surplus equals the integral of willingness to pay above marginal cost:
The deadweight loss is
For n equal firms the markup falls at rate relative to A, while the lost surplus falls at its square. A large number of firms is therefore associated with the competitive limit in this particular sequence of environments. The statement treats demand, cost, product homogeneity, and the number of independently optimizing sellers in a specific way. It does not describe arbitrary industry expansion in which product differentiation, entry quality, or capacity scarcity varies with n.
A uniform per-unit tax t can be represented by replacing c with , provided firms remain active and the effective cost lies below a. Price increases by times the tax, while aggregate output declines by per unit of tax. In this simple model an incomplete pass-through follows from linear demand and strategic quantity choice. For nonlinear demand and cost curves, pass-through can differ substantially.
Quantity Competition Beyond Symmetry
Heterogeneous costs and active sets
With constant but distinct marginal costs , an active firm satisfies
If m firms are active and their cost sum is C, solving the market clearing identity gives
For inactive firms the condition is . The correct solution is therefore an active-set problem: sort candidate marginal costs, solve for a candidate price with the chosen active set, and verify both activity and inactivity inequalities. Counting high-cost producers that are incapable of profitably serving the market artificially increases the number of economic competitors.
For example, take a=100 and b=1, with costs 10, 30, and 80. The first two firms form the active set. The price is 140/3, approximately 46.67, and outputs are 36.67 and 16.67. The third firm correctly produces zero. The nominal three-firm industry behaves as a two-active-firm market at this demand level, though the third firm could become relevant after sufficiently large demand or cost changes.
Firm size is endogenous in this example. An inexpensive firm has a larger output and higher share; its higher share alone cannot demonstrate greater discretionary anticompetitive conduct. A counterfactual ownership change requires recomputing the optimization problem, not simply relocating the existing shares.
General inverse demand and marginal costs
For smooth inverse demand and differentiable cost , interior optimality requires
Concavity or other global optimization conditions are required for first-order equations to characterize best responses. In the interior the residual-demand elasticity and market share produce a useful implication. Define the positive market elasticity and share . Under differentiability and invertibility,
This is a conditional Cournot first-order relationship, not an accounting identity valid for all markets. Multiplying by shares and summing gives a share-weighted markup equal to , where HHI is the sum of squared shares expressed on the zero-to-one scale. If reported HHI uses percentage shares, divide by 10,000 before applying this equation.
The relationship specifies precisely why concentration is economically informative in a particular model and why it cannot independently establish the model. Differences in measured concentration may coincide with changes in market elasticity, cost heterogeneity, product boundaries, ownership, or capacity.
Complementarity conditions and capacity
Suppose quantities are bounded by installed capacity . Define the marginal profit . A necessary optimality system has at interior quantities, at zero, and at a strictly positive upper bound. Under concavity in own quantity these conditions are sufficient for a best response.
The upper-bound sign has economic content. A constrained supplier can have strictly positive marginal profit and still be optimizing. Observing high price relative to short-run marginal operating cost may then reflect scarcity and the opportunity cost of capacity. Interpreting the whole wedge as discretionary market power without considering the binding constraints confuses physical feasibility with conduct.
The Homogeneous Bertrand Price Game
Deriving the benchmark
Consider at least two firms, identical constant marginal cost c, a homogeneous good, continuous posted prices, and the ability of every firm to meet the entire market demand at its price. Buyers purchase from the lowest-priced eligible seller; ties are allocated equally. Demand is positive near c, continuous as a function of price, and falls to zero at a finite choke price greater than c. Production occurs as needed after orders arrive. There are no fixed costs or purchase frictions.
At for every firm, each earns zero operating profit. Raising one's price causes sales to disappear, while lowering it generates a loss on positive sales. The profile is a Nash equilibrium. A common lowest price strictly above c with multiple firms serving positive demand cannot be an equilibrium: a tied seller can undercut slightly and acquire the whole market, gaining approximately the profit of the whole market in place of its fraction.
If one firm alone charges the lowest positive-margin price and the second-lowest price is strictly higher, a profitable small undercut by a losing rival can capture the market. A lowest price below c gives a seller negative operating profit and an incentive to avoid those sales. A profile with no positive demand can be defeated by offering a profitable price below the choke price. These deviation arguments establish that the minimum transaction price must equal c in equilibrium under the assumptions.
For two firms the pure equilibrium is the common price c. For more than two, some firms may quote prices above c and receive no sales, provided at least two firms quote c. The transaction outcome is competitive, although the complete posted-price profile is not necessarily unique. This subtle distinction between outcome uniqueness and strategy uniqueness is important when connecting observed prices to mathematical equilibrium.
Tick sizes, changing costs, and discontinuous payoffs
The slight-undercutting argument relies on continuous prices. On a discrete grid, undercutting requires a finite tick. Let prices be nonnegative integer monetary units, cost c=20, and demand for p between zero and 100. If two firms quote 21, each earns 39.5, while undercutting to 20 earns zero and raising price loses demand. The common price 21 is consequently also an equilibrium. The common price 20 remains an equilibrium. This example illustrates the sensitivity of uniqueness to feasible price actions, even when goods remain homogeneous.
Unequal costs introduce further qualifications. With the efficient producer may be able to earn a positive margin without a viable competing offer below the rival's cost. If the low-cost firm's monopoly-optimal price is strictly below , that price can be sustained with the rival charging at least . If the optimal unconstrained price is higher than , existence and precise price depend on tie-breaking, tick size, restrictions on weakly dominated below-cost offers, and whether limits to undercutting are attained. A formula claiming that the price always equals the second-lowest cost is too strong for a continuous-price game without those details.
Market demand is smooth in price in the benchmark; individual firm demand is discontinuous when priority switches. Differentiating a firm's profit as if customers were allocated continuously across identical sellers misses precisely the strategic discontinuity driving the classical result.
Capacity, Rationing, and the Timing of Commitment
Price competition after production commitments
Production technologies routinely restrict how quickly firms can respond to demand. In this case a lower-priced supplier may exhaust capacity before high-priced sellers lose all sales. Residual demand for a higher-priced firm then depends on both the competitor's capacity and the rationing rule.
Under efficient rationing, customers with the highest willingness to pay obtain the available units of the lower-priced supplier first. For two firms with the same homogeneous product, when the rival charges less and has capacity , residual demand at price is
The high-priced firm sells the smaller of this residual and its own capacity. Under proportional rationing, available lower-price output instead goes to a representative fraction of that price's demand; the resulting high-price residual generally differs. These are distinct games, even if demand and capacities are identical.
Take over the positive-demand region, c=20, and two capacities of 20. At the capacity-clearing common price 60, each sells 20 and earns 800. Lowering price leaves the firm's volume unchanged and reduces its profit. Raising price above 60 under efficient rationing leaves demand , less than 20. Profit is decreasing for prices above 60 because its unconstrained maximum occurs at 50. A price at 60 is therefore a mutual best response. The positive wedge of 40 coexists with price competition.
Now replace the capacities by 35. The capacity-clearing price becomes 30 and each firm earns 350. Raising price to 42.5 leaves residual demand 22.5 after the other firm supplies 35, and yields profit 506.25. The capacity-clearing price is no longer an equilibrium. The calculation establishes a failure of that candidate, not a complete characterization of every possible pure or mixed equilibrium of the new game.
Endogenous capacity and sequential rationality
Kreps and Scheinkman (1983) showed in Quantity Precommitment and Bertrand Competition that a symmetric two-stage capacity-then-price game produces the Cournot outcome under their conditions. The ability to commit capacity before price competition changes the second-stage opportunities. Their result also depends on how limited production is allocated among customers and on the modeled cost and demand conditions.
The economic significance extends beyond comparison of the two textbook games. Posted prices can implement an outcome rooted in earlier quantity commitments. An empirical observer who sees firms posting prices may infer Bertrand-type competition even where the binding strategic decisions were made substantially earlier. Identifying the last visible price decision is insufficient; the timing of the practically irreversible commitment must be investigated.
A sequential game calls for equilibrium strategies in all relevant continuation games. A threatened price response after an off-path capacity choice is credible only if it is optimal in the resulting subgame. Subgame perfection removes equilibria sustained solely by noncredible continuation behavior. The conditional Cournot result does not imply that every real industry with capacity has an exact Cournot equilibrium.
Differentiated Products and Price Competition
A common demand foundation
Product differentiation makes the price-setting firm's residual demand continuous over relevant ranges. Consumers can prefer one good because of location, quality, compatibility, reliability, or other attributes even when it carries a higher money price. Let inverse demands for two goods be
with constant common marginal cost c and . The demand system follows from a quasilinear quadratic utility with own quadratic terms and cross term . The matrix with diagonal b and off-diagonal d is positive definite. Positive d describes substitute goods in this system; negative d describes complementary goods. Results below refer to the symmetric interior equilibria, with nonnegative quantities and the corresponding economically feasible demand schedule.
The symmetric Cournot quantities and margins satisfy
Invert the two inverse demand equations. The partial derivative of own demand with respect to own price is and with respect to the other price is in the interior. A single-product price-setting firm's first-order condition is
At the symmetric Bertrand solution,
Own-profit second derivatives are negative in the interior for both games. Corner regimes must be examined when extending the linear demand system outside its positive-quantity region. For the stated symmetric solutions, the piecewise demand from quadratic utility supports the interior optimum: in the substitutes case, a corner where the rival's demand vanishes can occur only at lower own prices, where the residual profit derivative at the boundary does not reverse the interior maximization; a corner with own demand zero yields zero profit; for complements, feasible single-good demand can be checked against the joint-consumption branch. The comparative welfare result is asserted for the standard interior framework, not for arbitrary demand truncation or multi-product ownership.
Quantities, markups, profits, and welfare
The efficient symmetric quantity is . Comparing the two interior solutions,
Both equilibrium quantities are below the efficient level. For substitutes, the price equilibrium supports lower prices and higher output than quantity competition. For complements, the price equilibrium also increases symmetric output in this model. At d=0 each firm is effectively a monopoly over an independent product; the two formulations yield the same result. As d approaches b from below, the goods approach perfect substitution, the Bertrand markup tends to zero, and the Cournot markup remains positive. The demand matrix is singular at d=b, so the homogeneous price game must be formulated separately rather than evaluated by inverting that matrix exactly at the boundary.
The primary literature by Singh and Vives (1984), Price and Quantity Competition, gives the broader linear differentiated-duopoly comparison, including the welfare and profit differences and the role of binding price versus quantity contracts. The special-case derivation here isolates the effect of the interaction parameter d and supplies transparent numerical counterfactuals. It is an exposition of established economic results, not a newly proposed model.
For two products sold by different firms under the symmetric setup, total surplus is
The term is the quadratic scarcity and substitution cost from the utility function. Since the derivative is positive for q less than the efficient quantity, the larger Bertrand quantity raises total surplus when d is nonzero. Consumer surplus at a symmetric market-clearing quantity is . Operating profits aggregate to . For substitutes the Cournot profit is greater; for complements the comparison reverses in the standard symmetric linear model. The profit ranking need not extend to asymmetric costs or alternative demand systems.
Multi-product ownership
A firm controlling many products chooses a vector of prices. For product j owned by firm f, the first-order condition takes the form
The sum embodies internalized diversion between products. A retail brand merger can change incentives even if the number of physical products, their marginal costs, and observed characteristics remain unchanged. If two owned products are substitutes, an increase in one's price can divert some sales to the other owned product. The economic firm is therefore the relevant ownership unit.
As with single-product competition, these first-order conditions require interior solutions and global best-response or suitable concavity checks. In multi-product markets, the demand Jacobian, ownership matrix, strategic variables, and common constraints must be specified before a price-cost margin can be recovered from equations.
Existence, Uniqueness, Stability, and Selection
Existence and mathematical regularity
The existence question asks whether at least one strategy profile satisfies the equilibrium inequalities. For continuous quantity games, compact convex action sets, continuous profits, and quasiconcavity in each own action are familiar sufficient conditions for a fixed-point argument. The feasibility restrictions that make action sets compact should be economically justified, for example through physical capacities or an output bound above relevant demand.
Novshek's (1985) existence analysis for Cournot equilibrium establishes a more particular result for homogeneous-product competition using restrictions on firms' marginal revenues as rivals' aggregate output increases. The paper also identifies cases of nonexistence when relevant restrictions fail. This matters because arbitrary cost and demand functions need not sustain a smooth concave best-response problem.
In homogeneous price games, a price tie can produce discontinuous payoffs. One cannot simply reuse a fixed-point theorem requiring continuous payoffs. Finite price grids allow application of finite-game mixed-equilibrium existence, but the existence of mixtures on a finite grid is not an automatic proof of convergence to a pure equilibrium under a continuous-price limit. The tie-allocation rule and boundary conditions remain relevant to that limit.
Uniqueness and stability answer separate questions
Uniqueness is a property of the equilibrium set; stability is a property of a specified process for updating strategies. A static game has no inherent adjustment law. For an interior symmetric linear Cournot game, consider simultaneous naive best-response updating:
Around the interior solution, its Jacobian has zero diagonal and off-diagonal elements -1/2. The eigenvalue associated with the common-output direction is , while each of the n-1 orthogonal difference directions has eigenvalue 1/2. Thus this particular process is locally asymptotically stable for n=2, has a unit-modulus common eigenvalue for n=3, and is locally unstable for n greater than three. The underlying positive-price equilibrium is still unique for all these n.
If firms update only a fraction of the gap between their current output and best response, the Jacobian eigenvalues become and . For , local asymptotic stability requires . With n=5 and , these eigenvalues are -0.5 and 0.75; the damped rule is stable even though full updating is not. The exercise demonstrates mathematically that an equilibrium condition, uniqueness, and an adjustment mechanism are distinct objects.
The process is an illustration, not a claim that firms literally apply these update equations. Real firms learn with delays, receive noisy demand signals, and revise decisions at different frequencies. A fitted adjustment model must be compared with observations of those actual processes.
Selection in multiple equilibria
A selection criterion may depend on initial conditions, expectations, risk dominance in an appropriate game, institutional sequencing, learning dynamics, or coordination history. Such criteria add substance beyond mutual optimization. An algorithm that converges to one fixed point from one starting value establishes neither that the equilibrium is unique nor that it will be chosen by actual participants.
A computational investigation should compare multiple feasible initializations, detect boundary solutions, calculate residuals in best-response inequalities, and distinguish numerical convergence from economic optimality. In games with discontinuities, local gradients alone are inadequate. Deviations across ties or capacity thresholds can dominate apparently stationary candidates.
Sequential Decisions and Supply Schedules
Quantity leadership and commitment observability
In a two-firm homogeneous market with linear demand and common marginal cost c, suppose one firm commits its output before the second responds; the second observes that commitment and optimizes. Its interior response is . Substitution into the leader's profit function yields leader output and follower output . The price is , compared with the simultaneous Cournot price .
The leader earns , the follower , and consumer surplus is . Total output is higher than under simultaneous choices. The advantage depends upon the ability to commit and upon credible observability. Publishing a target that can be revised instantly may not establish the strategic advantage attributed to a fixed first-stage decision.
The follower's optimization problem changes when the leader is constrained, unobservable, or uncertain. A commitment investment may be costly; an optimized static leadership profit cannot be compared directly with an unconstrained one without counting commitment costs.
Bids as contingent strategies
A bid schedule relates the offered quantity to a potential market-clearing price. Under demand uncertainty a schedule can preserve flexibility while committing the firm to a conditional action. Klemperer and Meyer's (1989) supply function equilibria analyze conditions for existence and uniqueness of Nash equilibria when firms select such schedules. This is an alternative strategic space, rather than an intermediate price computed by averaging the Cournot and Bertrand formulas.
Electricity markets illustrate the economic relevance. Generators often submit multi-unit supply offers, face uncertain demand, and possess portfolios of plants with heterogeneous costs. Settlement rules, dispatch constraints, and market information determine which bids are feasible and how they translate into profits. Vickrey's (1961) auctions and competitive tenders provides an earlier point of reference in the development of strategic bidding research. A uniform-price multi-unit auction is not automatically equivalent to a single-unit auction with a dominant-strategy payment rule.
Financial contracts can separate the price received for part of output from the real-time spot price. An elementary fixed-price hedge with quantity h and contracted price K gives the firm a stylized cash flow . The derivative with respect to spot price holding q fixed is q-h. This shows why the net price exposure, rather than gross production alone, enters strategic incentives. Wolak's (2001) hedge contracts and bidding develops the relationship in a model of wholesale electricity. The algebraic illustration here abstracts from margin requirements, counterparty risk, and the particulars of financial settlement.
Repeated Interaction and Computational Pricing
Dynamic incentives
The one-period game is a useful baseline because it isolates the immediate gain from an action. Firms in a repeated market also face future consequences. For a conceptual infinite-horizon comparison with a common discount factor between zero and one, let be a symmetric high-profit period payoff, the greatest one-period deviation payoff, and the payoff under an equilibrium continuation. If a deviation switches future periods permanently to the latter continuation, the no-deviation inequality is
When , the resulting bound is . This inequality is a textbook repeated-game incentive calculation. Its interpretation depends on monitoring, credible continuation strategies, persistence, and the information structure. It is neither evidence of a real agreement nor a procedure to coordinate actual firms.
A one-period Nash equilibrium does not exhaust the possible equilibria of repeated competition. Conversely, stable high margins do not identify which repeated-game equilibrium, if any, has been selected. Market-wide shocks, capacity bottlenecks, and product differentiation also generate persistent prices above marginal operating cost.
Learning and changes in the decision technology
The introduction of software alters the effective speed, frequency, and possibly conditionality of prices. Reactions can become contingent on competitors' prices before any human manager intervenes. The institution must then be described at the relevant frequency of decision making, with the distinction between the action implemented and the rule that implements it.
Calvano and colleagues (2020), in algorithmic pricing and collusion, use controlled computational experiments in a repeated-price setting with Q-learning. That methodological design investigates outcomes under specified software and market environments. It cannot directly establish the prevalence of corresponding behavior in actual markets. Different learning rules, action grids, memory, exploration, costs, and entry conditions may select different outcomes.
The OECD's (2023) Algorithmic Competition supplies institutional policy context for this emerging decision technology. A measured effect of automation on margins is an economic result whose interpretation depends on the identification strategy. Legal assessment requires separate facts and applicable law.
Measurement and Empirical Identification
Defining the observed variables
An empirical oligopoly analysis begins with an explicit data dictionary. The market-time unit must be defined before aggregating transactions. Price should distinguish posted from realized prices, gross from net amounts, and nominal from deflated values. Quantity should use consistent physical units or quality-adjustment conventions. Sales may differ from production, capacity, bids, and shipments through inventories and trade flows. Revenue divided by quantity can conceal price discrimination and changes in product composition.
Marginal cost is the cost of supplying an incremental unit under a defined operating regime. It is not generally equal to accounting average cost, an allocated overhead measure, or the purchase price of one input. In a capacity-constrained industry the relevant economic incremental cost may depend on scarcity, startup costs, foregone sales in other markets, intertemporal production limits, and uncertainty. Both the observed cost series and the modeled counterfactual require explicit conventions.
Ownership must be identified for each market-time unit. A change in the definition of a firm can change observed concentration without any change in the number of plants. Product characteristics and destination must describe feasible substitution; legal boundaries alone may not coincide with economic markets. In particular, a local price gradient, an electricity transmission constraint, and a national automobile differentiation pattern create different economically relevant choice sets.
Demand estimation requires own-price and cross-price derivatives or other sufficient information about substitution. Price is often endogenous because unobserved demand shocks alter both consumer choices and profit-maximizing prices. A credible empirical design needs exogenous variation or explicit assumptions that separate the demand disturbance from the price equation. Instrument relevance and exclusion are distinct requirements.
The identification problem
Consider an illustrative homogeneous industry conduct equation
associated with linear inverse demand , a common marginal cost c, and an aggregate behavioral parameter . Symmetric Cournot gives , price-taking gives zero, and monopoly gives one. These associations hold within this particular definition of the conduct equation. A fitted value outside that range is not automatically proof of an impossible market outcome: measurement error, changing cost, or a misspecified behavioral equation may produce it.
Given a single observation of p and Q with known b but unobserved c, there are several compatible cost-conduct pairs. At p=60, Q=40 and b=1, c=60 is compatible with price taking; c=40 with symmetric two-firm Cournot; and c=20 with the monopoly conduct value. Each pair satisfies the observation. Observational equivalence is broken only by additional economically valid information.
Bresnahan's (1982) identification of the oligopoly solution concept established a foundational econometric question: under what structural conditions can market data separately recover conduct? The simple numerical example illustrates the logic of underidentification without replicating that paper's exact model. In a richer setting, variation that changes demand curvature or elasticity independently of cost can help identify conduct. The exclusions allowing this comparison must be evaluated rather than assumed from a statistical correlation.
Structural pricing with differentiated goods
Berry, Levinsohn, and Pakes (1995), in Automobile Prices in Market Equilibrium, integrate consumer choice, differentiated products, and oligopoly supply in a structural framework applied to US automobiles. The demand component estimates substitution among products and the supply component uses an assumed pricing institution to relate prices to costs. Their empirical approach demonstrates how product characteristics and aggregate sales information can enter an equilibrium model, while the resulting cost inferences remain conditional on the structural assumptions.
A practical sequence is to define products and markets, estimate demand with a credible treatment of price endogeneity, determine the relevant ownership and capacity structure, recover marginal costs under competing conduct assumptions, and test implications against independent cost or price variation. A model that fits historical prices by allowing unconstrained cost residuals has limited power to distinguish alternative conduct mechanisms.
Merger simulations and tax or capacity counterfactuals generally require equilibrium recomputation. Standard errors around demand estimates do not exhaust uncertainty about market definition, alternative game form, equilibrium multiplicity, or strategic adaptation following the intervention. A robust analysis reports sensitivity to those alternatives separately.
Data scope and reproducibility
This monograph does not download or estimate original firm-level datasets. It uses published empirical findings as comparative evidence, and constructed numerical examples as deductive checks. This is a substantive distinction: no original causal estimates, standard errors, or statistically representative worldwide market ranking are claimed.
A full replication project would require source licenses, versioned raw data, firm and product identifiers, unambiguous market-time definitions, the estimation code and instruments, a record of exclusions and missing values, and documented software versions. Proprietary market data may prevent independent replication by a reader unless an access mechanism is available. Public metadata alone cannot supply the missing transactions. Cross-market comparisons in this dossier therefore compare the mechanisms and research designs, not nominal price levels across different years and currencies.
International Empirical Evidence
United States automobiles
The US automobile industry provides a setting where product differentiation, multi-product ownership, and consumer substitution are central. The structural approach of Berry, Levinsohn, and Pakes (1995) draws on market-level product and consumer information rather than treating automobiles as one homogeneous commodity. Comparing an industry-wide count of manufacturers to the homogeneous Cournot formula would discard product-specific substitution and the overlap of ownership portfolios.
The most important measurement implication is that the cross-price demand derivative between two models need not be the same for every competing model. A money-price change on a small vehicle may induce a different switching pattern from the same change on a premium vehicle. Demand substitution and ownership therefore determine the relevant price incentives jointly. An investigation of changes in market power over decades would additionally need harmonized characteristics, ownership histories, quality definitions, and costs.
The study supplies an empirical methodology and a specific application; the present dossier does not reestimate its parameters. Any claim that a particular manufacturer had a given markup or consumer surplus at an exact date would require reading and reproducing the corresponding tables and definitions, which are outside the empirical scope adopted here.
California electricity
In Measuring Market Inefficiencies in California, Borenstein, Bushnell, and Wolak (2002) decompose wholesale electricity expenditures relative to a competitive benchmark using data from June 1998 through October 2000. They report expenditure rising from $2.04 billion in summer 1999 to $8.98 billion in summer 2000. Of that increase their analysis attributes 21 percent to production costs, 20 percent to competitive rents, and 59 percent to market power.
The increase is $6.94 billion. Applying the rounded published percentage shares gives approximate contributions of $1.46 billion, $1.39 billion, and $4.09 billion. These are arithmetic translations of the reported proportions, not independent estimates. The shares refer to the increase in expenditure, not to the total level of expenditure in 2000 or the share of generation physically withheld.
The research design matters because some increase in wholesale electricity payments could arise competitively when marginal plants become expensive or scarcity rents increase. A simple subtraction of the previous year's expenditure from current expenditure would incorrectly attribute every change to strategic conduct. The competitive benchmark isolates an economically specified comparison subject to its modeling and data assumptions.
British electricity and Texas auctions
Wolfram (1999), British electricity duopoly power, reports British spot price-cost margins below the magnitudes predicted by many theoretical models even though prices exceeded marginal costs. This evidence is especially instructive when compared with California: a positive wedge alone does not tell us whether a textbook oligopoly model overpredicts or underpredicts market performance.
Hortaçsu and Puller (2008), strategic bidding in Texas, study hourly supply schedules in the Texas electricity spot market and compare behavior with a static optimization benchmark using firm-level cost and bid information. Large-stake firms approached that benchmark; smaller firms submitted schedules steeper than the benchmark suggested. Firm heterogeneity can therefore affect both the strategic optimization and how closely a stylized solution describes actual choices.
The institutional differences among California, Britain, and Texas concern the relevant settlements, technologies, market histories, and financial positions. Their estimates cannot be pooled as if they were observations of a single standardized oligopoly parameter. Electricity is a particularly useful comparative domain precisely because physical constraints and clearing rules are economically consequential.
Germany's retail gasoline market
Assad, Clark, Ershov, and Xu (2024), evidence from German gasoline markets, study the diffusion of algorithmic pricing software that became widely available in 2017. Adoption timing is inferred from breaks in pricing-pattern indicators; the authors use headquarter adoption as an instrument to address endogenous station adoption. They report margin increases for nonmonopoly stations, and in duopoly and triopoly markets increases only when all stations adopt.
These are conditional empirical findings. An instrument based on organizational adoption must affect local margins through the modeled station adoption channel rather than a correlated direct change in brand strategy or cost. The study's identification approach is therefore central to interpreting its finding; a comparison of adopters' and nonadopters' mean margins would answer a weaker question. No station-level coefficient, confidence interval, or exact adoption rate is supplied here because the full restricted empirical tables were not inspected.
The relationship to Cournot and Bertrand lies in changes to effective strategic timing and information: software can alter the reaction process of price-setting firms without altering the physical good. Retail locations and local market boundaries remain important. A national average margin cannot directly identify outcomes in local duopolies.
Controlled laboratory evidence
Huck, Normann, and Oechssler (2004), experimental Cournot oligopolies, investigate number effects using controlled experiments and a literature review. Their findings suggest that cooperation sometimes arises in duopoly and becomes harder in markets with more firms. Laboratory variation in the number of players is valuable because other aspects of the game can be held fixed more successfully than in many observational market comparisons.
The controlled result and the field evidence identify different relationships. Experiments can isolate the number of decision makers in a prescribed environment; real-world variation in firm count often arrives with entry, demand shifts, capacity changes, and ownership changes. The external application of experimental results requires considering those differences. The evidence supports disciplined use of strategic benchmarks rather than a universal numerical cutoff for the number of competitors.
Welfare, Market Structure, and Institutional Design
Surplus, payments, and cost
Consumer surplus measures willingness to pay above the transaction payment under specified preference assumptions. Producer surplus measures receipts above relevant variable costs; economic profits additionally subtract opportunity and fixed costs as appropriate. Total surplus counts resources and benefits rather than double-counting payments transferred from purchasers to suppliers.
The static homogeneous linear example provides a clean demonstration. At the Cournot output, buyers pay a positive markup and the economy foregoes transactions whose willingness to pay exceeds marginal cost. The markup on units still traded is largely a transfer in the simple quasi-linear welfare calculation. The deadweight loss is the value of the omitted mutually beneficial transactions. If real resources are expended to maintain market power, if production is inefficiently allocated among suppliers, or if there are externalities, welfare losses can exceed this simple triangle.
Differentiated-product welfare includes substitution and product variety. A reduction in the number of available variants can harm some consumers even when mean prices are lower. A credible policy counterfactual must include both price and product changes, including entry and exit if those are plausible responses.
An electricity benchmark must distinguish the cost of supplying demand from payments to generators. Competitive scarcity rents can support investment in capacity; high payments do not necessarily constitute real resource destruction. Conversely, misallocation of dispatch can raise real costs even where a payment decomposition looks modest. Time-dependent reliability and investment incentives complicate the static welfare ranking.
Concentration and comparative changes
Concentration metrics summarize an ownership distribution. The standard HHI formula is ; for symmetric n it equals . Its interpretation is strongest when the assumptions connecting concentration to market power can be assessed. Even in homogeneous Cournot competition the elasticity is essential. Under differentiated Bertrand pricing, diversion between products and ownership enter the pricing conditions through a different structure.
A counterfactual increase in the number of firms under fixed demand and marginal cost has a transparent result in the classical quantity game. Actual entry can also introduce capacity, a distinct product, or new constraints. Similarly, exit of a nominal supplier that never served demand may have little immediate effect despite a changed firm count, while withdrawal of a pivotal supplier in a constrained hour may have an enormous effect.
Policy design must specify what the intervention changes: the number of economic decision makers, the cost of entry, capacity access, the information environment, the settlement mechanism, or the ability to commit. An intervention that lowers cost and increases output may simultaneously alter the distribution of profits. The choice of welfare criterion and the treatment of transfers should be explicit.
This dossier provides economic analysis of those mechanisms. It does not assess any named company's legal conduct or establish the applicability of competition statutes in a particular jurisdiction. Institutional facts and legal standards have separate evidentiary requirements.
Comparative Numerical Investigations
Homogeneous market comparison
Set a=100, c=20, and b=1 with no fixed costs. Efficient output is 80 and total efficient surplus is 3,200 monetary units. All examples in this chapter are calculated from the stated equations, not collected data. Rounding below is to two decimal places where appropriate.
| Game | Aggregate quantity | Price | Consumer surplus | Aggregate operating profit | Deadweight loss |
|---|---|---|---|---|---|
| Monopoly | 40.00 | 60.00 | 800.00 | 1,600.00 | 800.00 |
| Cournot, two firms | 53.33 | 46.67 | 1,422.22 | 1,422.22 | 355.56 |
| Cournot, three firms | 60.00 | 40.00 | 1,800.00 | 1,200.00 | 200.00 |
| Cournot, five firms | 66.67 | 33.33 | 2,222.22 | 888.89 | 88.89 |
| Cournot, ten firms | 72.73 | 27.27 | 2,644.63 | 528.93 | 26.45 |
| Sequential quantity duopoly | 60.00 | 40.00 | 1,800.00 | 1,200.00 | 200.00 |
| Homogeneous Bertrand | 80.00 | 20.00 | 3,200.00 | 0.00 | 0.00 |
The sequential quantity duopoly and simultaneous three-firm Cournot produce identical total quantity and price in this special linear example. Their ownership distributions and profits by firm are different. Thus even without measurement noise one observed market-level price and aggregate quantity can be consistent with different strategic games. This is a direct demonstration of the empirical identification problem, rather than a proposed new theory.
The outcomes also show why the comparison of Cournot and Bertrand requires the same underlying demand and cost assumptions. If a price game is compared with a quantity game that simultaneously has lower marginal cost, the strategic effect is confounded with the cost effect.
Differentiation sensitivity
Take a=100, c=20, b=1 and vary d over a subset of the admissible interval (-1,1). The following entries refer to price and output per product, together with total surplus across both products. The formulas are evaluated at the interior symmetric solutions.
| Cross-demand coefficient d | q Cournot | p Cournot | q Bertrand | p Bertrand | Total surplus Cournot | Total surplus Bertrand |
|---|---|---|---|---|---|---|
| -0.50 | 53.33 | 73.33 | 64.00 | 68.00 | 7111.11 | 8192.00 |
| 0.00 | 40.00 | 60.00 | 40.00 | 60.00 | 4800.00 | 4800.00 |
| 0.25 | 35.56 | 55.56 | 36.57 | 54.29 | 4108.64 | 4179.59 |
| 0.50 | 32.00 | 52.00 | 35.56 | 46.67 | 3584.00 | 3792.59 |
| 0.90 | 27.59 | 47.59 | 38.28 | 27.27 | 2967.90 | 3340.58 |
At d=0, output, price, and welfare coincide. For d>0 the price-setting outcome increases output and reduces margins. At d=-0.5 the same welfare ranking obtains in this quadratic special case, despite the economic distinction of complements. The limiting movement as d approaches one is consistent with the transition to highly substitutable goods; the singular case d=1 is handled by the separate homogeneous analysis.
Each row changes the product-demand system. Comparisons between Cournot and Bertrand within one row hold utility parameters fixed, while comparisons across rows do not. Consequently, the absolute surplus across rows is not a causal effect of strategic mode. It captures the assumed utility change as well. This distinction between within-environment comparisons and cross-environment comparisons is essential when transporting conclusions across heterogeneous real markets.
Capacity and adjustment sensitivity
The two capacity experiments give a complementary comparison. When each capacity is 20, the clearing price of 60 is supported by unilateral profit comparisons under efficient rationing. When each capacity is 35, the clearing price of 30 fails because the proposed higher-price deviation gives 506.25 rather than 350. The difference is attributable to the changed feasibility and residual demand. It is not evidence that one firm has become less rational or that its objectives changed.
The adjustment experiment likewise changes only the speed of moving toward a calculated best response. For five firms, undamped common-direction deviations are locally magnified by a factor of two and reverse sign. Half-adjustment instead multiplies the same deviations by -0.5, while differences among firms decay with factor 0.75. Both processes concern the same unique static positive-price equilibrium. A researcher observing persistent oscillation cannot infer multiplicity without further evidence about the adjustment law.
These exercises provide a check on three distinct sources of comparison failure: strategic timing, the feasible action set, and behavioral updating. Each produces testable differences if the necessary institution-specific observations are available.
Analytical Synthesis and Research Boundaries
The strategic mechanism as the object of inference
The separate studies and calculations support a unified interpretive conclusion: market equilibrium is a property of the full game as institutionally specified. It is the mutual compatibility of independent optimal choices under the modeled constraints and information. The term price competition identifies a type of decision, not the entire process producing observed market transactions. The term quantity competition identifies an alternative commitment, which may itself be implemented through a price-clearing mechanism.
The strongest comparative inference starts with the variable that a firm can credibly control at the relevant horizon. Price-setting after a predetermined physical or contractual capacity can produce a result closer to quantity competition; an unconstrained supplier of homogeneous goods facing immediate substitution has a much more elastic effective residual demand. Where products are differentiated, the substitution matrix characterizes that residual demand alongside the ownership structure. A supply schedule allows participants to make state-contingent offers, and learning or repeated decisions introduce further strategic state variables.
This interpretation is anchored in established literature, particularly the capacity game of Kreps and Scheinkman and the differentiated-product comparison of Singh and Vives. The authorial contribution here is the integration of these mechanisms with the mathematical distinctions among outcome uniqueness, stability, and identification, and the controlled comparisons showing how market-level observations can coincide despite different underlying games. No priority claim is made for the established equations; no new named theoretical system is introduced.
Evidence that distinguishes mechanisms
Quantity and price data measured at the moment of settlement are insufficient in many applications. Records of capacity decisions, orders that could not be filled, delivery lead times, bid schedules, and the timing of revisions can help separate feasible-game descriptions. Ownership relationships are essential when firms control several products. Estimates of substitution and marginal cost test whether a candidate pricing rule is consistent with the observed margins.
A useful comparison evaluates multiple plausible market models against independent predictions rather than merely selecting the best in-sample fit. A demand increase with fixed capacity may alter the scarcity constraint; a change in the market-clearing rules may alter the marginal return to a bid; an ownership change can alter internalized diversion. The prediction that follows from each counterfactual is meaningful only when the intervention and selection mechanism are explicit.
A high price may fit a differentiated single-period price game, a homogeneous quantity game, constrained price competition, a repeated-game equilibrium, or an efficient scarcity allocation. Additional evidence determines which explanation is scientifically credible. The same principle holds for low prices: they can reflect cost advantages, aggressive competition, a demand decline, regulated price constraints, or a temporary strategic adjustment.
Research boundaries and extensions
Three questions remain substantive research programs. First, in industries with rapidly changing feasible actions, how can production, pricing, and investment decisions be jointly reconstructed from high-frequency operational data? Second, when several equilibria exist, which observable histories and institutional features discriminate among alternative adjustment models? Third, how can model uncertainty about demand and conduct be propagated into welfare estimates without confusing it with ordinary sampling uncertainty?
An extended literature review could address asymmetric mixed pricing equilibria, general rationing mechanisms, stochastic dynamic oligopoly, entry and product positioning, and nonparametric inference. These objects require separate methodological treatments. They are recognized here as boundaries of the specific dossier, whose scientific center is the connection between price and quantity competition and equilibrium selection.
The historical and empirical source access in this research is selective: primary-text excerpts and publisher or institutional abstracts support bounded claims; several full journal articles, underlying datasets, and image-based scans were unavailable for independent review. The formal benchmark algebra is reproducible from the equations, and the calculations do not depend on inaccessible empirical parameters. No external peer review or original field-data replication is claimed.
Conclusion
Cournot and Bertrand competition reveal how the strategic form of markets determines the relation between firms' independent actions and collective economic outcomes. In the symmetric homogeneous benchmark, a quantity commitment produces a positive price-cost margin that declines with the number of active firms, whereas continuous price competition with unrestricted fulfillment leads to marginal-cost transaction prices. These results follow from different feasible deviations and transaction-allocation mechanisms.
Capacity restrictions, unequal costs, discrete prices, differentiated products, ownership portfolios, bid schedules, and repeated interaction change those deviations and their consequences. A market featuring posted prices can sustain a quantity-commitment outcome when production is fixed beforehand; even a homogeneous price game can have positive margins when capacity is scarce or prices move on a discrete grid. A unique static equilibrium can coexist with unstable adjustment. Conversely, distinct games can generate the same aggregate observations.
The research surveyed across automobiles, electricity, gasoline, and controlled experiments illustrates the importance of institution-specific demand, cost, capacity, and decision data. The scholarly task is to identify the economic game that actually structures the market, verify that its strategic conditions fit the evidence, and interpret welfare within the corresponding counterfactual. The strongest conclusion is conditional precision: oligopoly analysis gains explanatory power by specifying what is committed, what is observed, what can be fulfilled, and how transactions arise from the available actions.
References
-
Assad, Stephanie, Robert Clark, Daniel Ershov, and Lei Xu. 2024. Algorithmic pricing and competition: Empirical evidence from the German retail gasoline market. Journal of Political Economy 132(3): 723–771. DOI: 10.1086/726906. Source
-
Berry, Steven, James Levinsohn, and Ariel Pakes. 1995. Automobile prices in market equilibrium. Econometrica 63(4): 841–890. DOI: 10.2307/2171802. Source
-
Bertrand, Joseph. 1883. Théorie des richesses. Review of Léon Walras and Augustin Cournot. Journal des Savants, September, 499–508. Source
-
Borenstein, Severin, James B. Bushnell, and Frank A. Wolak. 2002. Measuring market inefficiencies in California's restructured wholesale electricity market. American Economic Review 92(5): 1376–1405. DOI: 10.1257/000282802762024557. Source
-
Bresnahan, Timothy F. 1982. The oligopoly solution concept is identified. Economics Letters 10(1–2): 87–92. DOI: 10.1016/0165-1765(82)90121-5. Source
-
Calvano, Emilio, Giacomo Calzolari, Vincenzo Denicolò, and Sergio Pastorello. 2020. Artificial intelligence, algorithmic pricing, and collusion. American Economic Review 110(10): 3267–3297. DOI: 10.1257/aer.20190623. Source
-
Cournot, Augustin. 1838. Researches into the Mathematical Principles of the Theory of Wealth. Translated by Nathaniel T. Bacon. Chapters VII and IX, reprinted from the 1971 Augustus M. Kelley edition in Competition Policy International 4(1), Spring 2008. Source
-
Hortaçsu, Ali, and Steven L. Puller. 2008. Understanding strategic bidding in multi-unit auctions: A case study of the Texas electricity spot market. RAND Journal of Economics 39(1): 86–114. DOI: 10.1111/j.0741-6261.2008.00005.x. Source
-
Huck, Steffen, Hans-Theo Normann, and Jörg Oechssler. 2004. Two are few and four are many: Number effects in experimental oligopolies. Journal of Economic Behavior & Organization 53(4): 435–446. DOI: 10.1016/S0167-2681(03)00138-0. Source
-
Klemperer, Paul D., and Margaret A. Meyer. 1989. Supply function equilibria in oligopoly under uncertainty. Econometrica 57(6): 1243–1277. DOI: 10.2307/1913707. Source
-
Kreps, David M., and José A. Scheinkman. 1983. Quantity precommitment and Bertrand competition yield Cournot outcomes. Bell Journal of Economics 14(2): 326–337. DOI: 10.2307/3003636. Source
-
Nash, John F., Jr. 1950. Equilibrium points in n-person games. Proceedings of the National Academy of Sciences 36(1): 48–49. DOI: 10.1073/pnas.36.1.48. Source
-
Novshek, William. 1985. On the existence of Cournot equilibrium. Review of Economic Studies 52(1): 85–98. DOI: 10.2307/2297471. Source
-
OECD. 2023. Algorithmic Competition. OECD Roundtables on Competition Policy Papers. Paris: OECD Publishing. DOI: 10.1787/cb3b2075-en. Source
-
Singh, Nirvikar, and Xavier Vives. 1984. Price and quantity competition in a differentiated duopoly. RAND Journal of Economics 15(4): 546–554. DOI: 10.2307/2555525. Source
-
Tirole, Jean. 1988. The Theory of Industrial Organization. Cambridge, MA: MIT Press. ISBN 9780262200714. Source
-
Vickrey, William. 1961. Counterspeculation, auctions, and competitive sealed tenders. Journal of Finance 16(1): 8–37. DOI: 10.1111/j.1540-6261.1961.tb02789.x. Source
-
Wolak, Frank A. 2001. An empirical analysis of the impact of hedge contracts on bidding behavior in a competitive electricity market. NBER Working Paper 8212. DOI: 10.3386/w8212. Journal version published in International Economic Journal 14(2), 2000, 1–39. Source
-
Wolfram, Catherine D. 1999. Measuring duopoly power in the British electricity spot market. American Economic Review 89(4): 805–826. DOI: 10.1257/aer.89.4.805. Source