RESEARCH DOSSIER

Empirical Market Power and Markup Measurement

Abstract. Empirical market power is inferred through the relation between observed transactions and a counterfactual competitive or constrained equilibrium. The price–marginal-cost wedge is a central measure of static seller power, yet its observability requires economic restrictions that differ fundamentally across demand, conduct, production, and direct-cost approaches. This monograph reconstructs those restrictions, identifies conditions under which the resulting estimands coincide, and examines how accounting conventions, endogenous prices, heterogeneous substitution, input adjustment, market boundaries, and aggregation change empirical interpretations. It develops transparent numerical counterexamples and compares the literature on U.S. manufacturing and automobiles, Indian trade reform, German retail gasoline, U.S. brewing, and international firm-level markup trends. The synthesis establishes that credible market-power inference depends on identifying a particular wedge at a stated decision margin and horizon, then connecting it to strategic alternatives and welfare with separately justified assumptions. The research is an interpretive and methodological synthesis of scholarly and institutional evidence, with conditional derivations and illustrative calculations; it does not present newly estimated firm data or a new named economic theory.

Keywords: Market power; price–cost margin; markup; marginal cost; Lerner index; differentiated demand; industrial organization; conduct identification; production functions; output elasticity; pass-through; concentration; aggregation; competition policy; econometrics

The Economic Object and Its Boundaries

Market power describes the capacity of an economic actor to influence the terms of exchange relative to a specified alternative in which the relevant competitive constraint is stronger. For sellers, the central static object is the ability to maintain price above the incremental economic cost of supplying an additional unit. For buyers, the counterpart can be the ability to reduce the compensation of an input supplier below its marginal contribution to the buyer's revenue. These capacities are related through the structure of transactions, but a seller markup and a buyer markdown measure different margins of decision.

Three analytical objects therefore require separation. A realized wedge measures the distance between price and marginal cost at an observed allocation. A conduct parameter describes how decision makers anticipate, internalize, or respond to the effects of their own actions on market outcomes. A competitive counterfactual describes the allocation that would arise if specified constraints, ownership links, strategic interactions, technologies, or institutions were altered. A single observed wedge may be compatible with several conduct assumptions and several competitive counterfactuals. The wedge is economically meaningful while its causal interpretation remains conditional.

The scientific unit is an economically coherent transaction. A plant, establishment, consolidated enterprise, brand, product, platform, location, and legal entity may supply different observations. A company selling a portfolio of differentiated products generally has several prices and several marginal costs. Its accounting margin is an aggregated outcome; the relevant pricing decisions can be product-specific. Vertical integration changes the meaning of an upstream purchase and a downstream sale. Measurement begins with a definition of the unit, ownership boundary, market, timing, currency, and price net of rebates and transaction taxes as appropriate.

This inquiry belongs to Microeconomics through its analysis of strategic market structure and to Econometrics through its concern with identification. Production duality is used only where it establishes a markup estimator. The general theory of production, comprehensive merger simulations, labor-market monopsony, and platform governance have their own research domains. The present scientific question is how the numerical measurement of seller power is constructed and how its evidentiary content changes when the underlying restrictions change.

Historical Formation of Empirical Measurement

The marginalist tradition made price and opportunity cost comparable at the margin rather than treating monopoly profit as a sufficient statistic for competition. Lerner's measurement of monopoly power (1934) is a historical point of reference for the normalized price–marginal-cost gap. The continuing significance of this object is its comparability across units of currency and across transactions with different price levels, provided that the underlying economic costs are defined consistently.

The subsequent challenge was that prices and quantities are typically observed jointly at an equilibrium, whereas the schedules that generated the equilibrium are not. An industry with high price and low observed production could have high demand, high marginal cost, restricted capacity, or strategic output restriction. Cross-sectional correlations between structural indicators and accounting profitability do not by themselves discriminate among these explanations.

Bresnahan's conduct identification argument (1982) made the distinction between identifying an equilibrium supply relation and identifying its strategic content explicit. The observable response to exogenous variation can distinguish alternative conduct restrictions if that variation changes demand curvature or slope in a way that cannot be absorbed into the marginal-cost specification.

Hall (1988) approached the problem from measured changes in resource use. His U.S. industry cost study inferred the marginal-cost component of changes in output from annual observations rather than assuming that accounting average cost equaled marginal cost. Differentiated-product empirical industrial organization subsequently integrated consumer preference heterogeneity and equilibrium pricing. Berry, Levinsohn, and Pakes (1995) developed an influential automobile-market framework in which observed product characteristics, market shares, prices, and economic restrictions jointly inform demand and cost estimates.

Production-side methods later enabled comparisons across large firm panels where detailed demand systems were unavailable. This expansion in coverage transformed the empirical agenda: researchers could investigate the distribution of markups, exporter behavior, trade reform, and aggregate macroeconomic developments. The expansion also exposed a new identification problem. Financial statements principally record expenditures and receipts. Recovering a physical-output elasticity and economically flexible input expenditure from these records requires additional assumptions. The historical development of the field is thus a movement from a general wedge concept toward competing strategies for determining its otherwise latent components.

Price, Marginal Cost, and Economic Margins

Let p denote a strictly positive unit price and c denote strictly positive marginal economic cost at the observed output. The gross markup ratio is mu = p/c. The markup over cost is mu - 1. The normalized price–cost margin is L = (p-c)/p = 1 - 1/mu. A price of 120 and marginal cost of 80 yield mu = 1.5, a markup over cost of 50 percent, and L = 1/3. Reporting all three figures without names would introduce a substantial ambiguity into international comparisons.

Economic marginal cost is the incremental opportunity cost of the relevant output decision, holding the intended state variables fixed or allowing them to adjust as specified. In a smooth one-output cost function C(q), it is dC/dq. With co-produced outputs, it is a partial derivative holding other outputs fixed unless a joint adjustment is part of the estimand. When additional production requires crossing a discrete capacity threshold, a finite incremental cost and the capacity horizon may be more relevant than a smooth derivative. Under binding capacity, the opportunity cost of scarce capacity belongs to the marginal decision, even where the associated shadow value is absent from the accounts.

Consider total cost C(q) = F + cq, with fixed cost F. Economic profit before financing conventions is (p-c)q-F. If p = 12, c = 8, q = 100, and F = 300, the operating surplus in this illustrative economic model is 100, or 8.33 percent of revenue, even though the markup over marginal cost is 50 percent. Raising F to 400 leaves the markup unchanged and makes surplus zero. Thus a zero-profit equilibrium can accommodate substantial price–cost wedges when firms incur setup costs. The condition for zero profit, p = average cost, differs from p = marginal cost except under specific cost structures.

At an interior static optimum for a single-product price setter with demand q(p), differentiating (p-c)q(p) for constant c gives q + (p-c)q' = 0. More generally the same local first-order condition uses C'(q). If the magnitude of the own-price elasticity is e = -pq'/q, then L = 1/e. The formula requires differentiability, interior optimality, positive sales, and the relevant residual demand curve. It does not apply mechanically to a regulated price, corner solution, multiproduct portfolio, or dynamically chosen penetration price.

A realized price below measured marginal cost can reflect an intentional loss leader, a temporary subsidy, demand learning, liquidation of inventory, a measurement error, or a genuinely different optimization horizon. The algebra alone cannot classify the cause. When marginal cost is zero, the ratio p/c is undefined; a finite normalized margin can still be calculated for positive price, but the interpretation of scarce resources and long-run supply requires scrutiny. For subsidized or zero-price services, even a conventional normalized seller margin may be inappropriate.

Demand-Based Identification and Product Substitution

Demand-side measurement reconstructs consumers' substitution opportunities and combines them with an explicit pricing mechanism. If two brands are close alternatives, raising one price transfers sales to the other. If both brands belong to the same profit-maximizing firm, that substitution is partially internalized; if they belong to different firms, the transfer ordinarily strengthens the incentive to keep the first price low. An elasticity matrix must therefore be interpreted together with ownership.

For products indexed by j, let q_j(p) be demand as a function of all prices, c_j incremental production cost, and F(j) the set of products belonging to the owner of product j. Under simultaneous, unconstrained differentiated-product Bertrand pricing with independently maximizing firms and smooth costs, the first-order condition for price j is

q_j + sum over k in F(j) of (p_k - c_k) times partial q_k / partial p_j = 0.

Define A_jk = minus the partial derivative of q_k with respect to p_j if j and k have common ownership, and zero otherwise. In vector notation A(p-c)=q. If A is invertible, the candidate margin vector is A inverse times q. The construction requires consistent signs, ownership links, price units, share-to-quantity conversion, and an equilibrium satisfying the assumed conditions. A numerical inversion of A does not establish that the solution maximizes profit globally or that it is the unique market equilibrium.

A two-product illustration makes the strategic content explicit. Suppose the two products each sell 100 units and their own demand derivatives are -20 units per dollar, while cross derivatives are +10 units per dollar. Independent owners have margins of 100/20 = 5 dollars under the stated pricing conditions. Common ownership yields a symmetric system 20m - 10m = 100, giving margins of 10 dollars conditional on those same local demand levels and derivatives. These are two different implied-cost calculations at an identical hypothetical observation; they are not an equilibrium merger forecast because the demand levels and derivatives would generally change with equilibrium prices.

The demand framework in Berry, Levinsohn, and Pakes supplies a systematic example of heterogeneous substitution and market equilibrium. Simple logit demand imposes restrictive substitution patterns, while random coefficients allow correlations between price sensitivity and product characteristics. Both require sufficient variation and an outside-good convention. Market size changes the inferred outside share, which may alter elasticity estimates and therefore inferred marginal costs.

Price is endogenous whenever unobserved quality or anticipated demand affects pricing. In a structural demand system, instruments must shift equilibrium prices or rival characteristics while satisfying the specified demand-error exclusion. An instrument's statistical strength alone cannot establish that it is excluded from consumer utility. Rival characteristics can themselves respond to local tastes; wholesale costs can correlate with quality changes; policy changes can alter demand directly. The research design needs an institutional justification, formal identifying conditions, and sensitivity to competing instrument sets.

When estimating differentiated demand, researchers should examine elasticity signs and magnitudes, substitution toward the outside option, diversion patterns, implied cost plausibility, and specification stability. An estimated negative marginal cost in an ordinary physical product is an immediate diagnostic for a misspecified demand system, pricing rule, margin definition, or tax adjustment, although special joint-production contexts may have unusual incremental cost structures. Demand-based estimates are especially informative when the research objective is a product-level policy counterfactual and observable substitution is rich enough to support the selected demand specification.

Conduct Restrictions and Equilibrium Identification

Conduct-based analysis asks whether observed equilibrium responses support a particular pattern of strategic behavior. For a homogeneous product with inverse demand p(Q,Z), use a simplified industry supply relation

p = c(Q,W) - lambda Q times partial p(Q,Z) / partial Q,

where lambda indexes an assumed local degree of internalization. For the standard symmetric benchmark, lambda = 0 corresponds to competitive price taking, lambda = 1/n to symmetric Cournot with n equal firms, and lambda = 1 to joint output maximization under suitable assumptions. Other conduct models need different restrictions. The parameter should not be interpreted as a universal quantitative measure of collusion.

Suppose p = a + z - bQ and marginal cost is d + gQ, where b > 0. The supply relation becomes p = d + (g + lambda b)Q. Even with perfect observation of p, Q, z and demand slope b, the slope of the estimated supply relation identifies g + lambda b rather than g and lambda separately. For b = 1, an observed supply slope of 2 is compatible with g = 2 and lambda = 0, or g = 1.5 and lambda = 0.5, or g = 1 and lambda = 1. Each combination has a different implied competitive allocation. Additional observations from identical-slope parallel demand shifts do not resolve this equivalence.

If b instead varies with a demand shifter that is excluded from marginal cost, observed changes in the supply relation can provide separate leverage on lambda. For example, slopes of 2 and 2.5 when b changes from 1 to 2 identify lambda = 0.5 and g = 1.5 in this simple exact model. This calculation reconstructs the logic of Bresnahan's identification example. In empirical applications, the credibility of the rotation, the stability of conduct, and the adequacy of the cost restriction all require evidence.

Repeated-game pricing, capacity commitments, wholesale contracts, negotiated prices, and inventory decisions can generate supply relations that differ from a static simultaneous game. Observationally equivalent conduct may persist when the researcher only observes transactions from a narrow range of demand and cost states. A conduct estimate should be accompanied by the equilibrium restriction, the instruments or natural variation that identify it, and the observational alternatives that remain possible.

A concentration statistic can be used to select settings for further investigation. It can also enter a theoretically derived relationship under restrictive assumptions. For homogeneous Cournot with downward-sloping market demand, firm i's interior first-order condition yields L_i = s_i/E, where s_i is its quantity share and E is the magnitude of market demand elasticity. The quantity-share-weighted Lerner index then equals the fractional HHI divided by E. A change in concentration has no universally determined margin effect when E, capacity, ownership, product differentiation, or conduct vary simultaneously.

Production-Based Markup Measurement

The production approach starts with constrained cost minimization rather than a particular product-market game. Let q = F(v,k,omega) represent physical output with variable input v, predetermined inputs k, and productivity omega. Given positive input price w and a target quantity, minimize wv plus other appropriately specified costs subject to producing q. With interior input choice and differentiability, the multiplier on the output constraint is marginal cost c and the condition for v is w = c F_v.

Define theta_v = F_v v/q as the output elasticity and alpha_v = wv/(pq) as expenditure on the selected input relative to revenue. Rearrangement gives alpha_v = theta_v c/p, hence mu = theta_v/alpha_v. This is a conditional identity. The contribution of De Loecker and Warzynski's production-based markup estimator (2012) is an empirical strategy that exploits this relation with production-function estimation rather than requiring an explicit product-market pricing game.

For example, assume an estimated physical-output elasticity of 0.60 and a measured variable-input expenditure share of 0.40. The implied markup is 1.50. Its Lerner transformation is one-third. If a second input is truly flexible and affects output independently of demand, its ratio should target the same markup when evaluated at the same margin. Disagreement between these ratios can reveal adjustment costs, measurement mismatches, different input-market conditions, or misspecified technology. It is an empirical diagnostic, not an automatic test with a single interpretation.

Production-function identification is demanding because productive firms choose inputs in response to information unavailable to the researcher. Simply regressing log output on log inputs can confound productivity with technology parameters. Fixed assets, material use, labor hours, energy, and intermediate expenditure mix decisions made at different dates. Proxy-variable methods and dynamic moment restrictions potentially address some simultaneity, but their validity depends on timing, monotonicity, state transition assumptions, input prices, and demand variation. The estimated elasticity is the numerator of a ratio, so its error propagates directly to the markup.

The economic meaning of flexible input differs from classification as a variable accounting expense. Salaried labor can be contracted in advance. Materials can be drawn from inventory rather than purchased contemporaneously. Energy contracts may have take-or-pay provisions. Cloud computing may exhibit committed-use discounts. The input cost used in the condition must correspond to the marginal effective expenditure for the input decision being modeled. A flexible-input method is most credible where technological output data, unit input prices, and the adjustment horizon can be documented.

Multiproduct enterprises create an additional challenge. The same labor, design system, electricity, or production line may serve many goods. Revenue aggregates different output prices and product mixes, while a product-level cost derivative concerns a specific output. An estimator at the consolidated firm level cannot generally be read as each product's markup. Product-level price and quantity data, a defensible aggregation structure, or restrictions on cost allocation are needed. The researcher should report what level the estimator actually identifies rather than treating an available financial reporting entity as the economic decision unit.

Revenue Elasticities and Input-Cost Distortions

A central identification dispute concerns the difference between physical output and revenue. Let single-product revenue be R(v) = p(q(v))q(v). Its input elasticity is

theta_R = theta_q times (1 + d log p / d log q).

Under the single-product static optimum, marginal revenue equals marginal cost, so 1 + d log p / d log q = c/p. Consequently theta_R = theta_q/mu. Substituting this revenue elasticity into the production ratio yields theta_R/alpha_v = 1 under these conditions, regardless of the actual markup. Bond, Hashemi, Kaplan, and Zoch (2021) develop this issue in Some Unpleasant Markup Arithmetic, including the distinct problems of identification and estimation when output prices are unavailable.

Consider the previous example in which the true output elasticity is 0.60, the revenue share is 0.40, and mu = 1.50. The associated revenue elasticity is 0.40 under the stated assumptions. Replacing theta_q with theta_R returns one rather than 1.50. An estimated revenue elasticity above 0.40 may produce a ratio above one, but the result by itself does not establish recovery of the physical-output elasticity. A common industry deflator removes a common price index; it leaves variation in relative prices, quality, and product composition.

Several other disturbances can enter the same ratio. With input adjustment costs, observed purchases do not necessarily equal the effective shadow price multiplied by use. With an input employed in both production and customer acquisition, total expenditure can include resources whose incremental contribution operates through demand rather than physical supply. With input-market power, the marginal expenditure associated with one extra input unit may exceed its average observed purchase price. These are different economic mechanisms, and combining them into a single unexplained residual conceals the interpretation of the estimated wedge.

Accounting classifications compound the problem. Cost of goods sold, inventory movements, sales and marketing, research, administrative expense, and depreciation describe reporting categories. Their content is not determined by the derivative of an economic cost function. Reallocating an expense across accounting categories can change a measured expenditure share without changing price or marginal cost. In the illustrative ratio theta = 0.60, changing alpha from 0.40 to 0.45 lowers the estimate from 1.50 to 1.333. If the additional spending is economically flexible production expenditure, the alternative may correct an omission; if it represents fixed organizational expenditure, including it may distort the selected first-order condition.

A credible analysis must preserve the distinction between an estimand and an estimator. The identity describes what would be learned from correctly observed quantities and expenditures satisfying the optimization conditions. An empirical regression estimates quantities under statistical assumptions. The final reported ratio adds accounting and measurement conventions. Agreement with an expected magnitude cannot validate all three levels simultaneously.

Direct Cost Evidence and Pass-Through

Direct cost measurement includes engineering estimates, marginal resource requirements, procurement prices, operating logs, and firm records. Its attractiveness is the possibility of observing components of opportunity cost without inferring all of them from equilibrium demand. Yet actual unit cost accounts commonly mix incremental costs with allocation rules for joint or fixed resources. A detailed engineering estimate may exclude congestion, capacity shadow costs, product-specific service commitments, or the opportunity value of inventories.

Hall's industry-level approach illustrates the use of changes in cost and output. In a simplified one-input environment with physical output q, input v, and price w, holding other state variables and productivity fixed, incremental expenditure is w dv. The local incremental cost is w dv/dq. If productivity changes simultaneously, the observed ratio of expenditure change to output change no longer isolates cost. Exogenous demand variation that changes utilization without directly shifting technology may help. Input price variation, capital adjustment, intermediate inputs, and aggregation require their own controls.

Pass-through offers complementary evidence. Suppose the firm's local pricing decision has constant marginal cost c and demand D(p). Write the first-order condition D + (p-c)D' = 0. Differentiating with respect to c under the interior optimum gives

dp/dc = D' / (2D' + (p-c)D'').

With linear demand and constant marginal cost, pass-through is one-half. With constant-elasticity demand D(p) = Ap raised to -e, e > 1, the optimal price is e c/(e-1), so pass-through equals e/(e-1), greater than one. Both environments can possess market power. Therefore a pass-through coefficient above or below one is not, on its own, an index of stronger or weaker conduct.

For explicit arithmetic, let marginal cost fall from 100 to 90 and price fall from 150 to 145. The absolute pass-through of the cost change is 0.5. The markup rises from 1.50 to 145/90 = 1.611, and the normalized margin rises from 33.33 percent to approximately 37.93 percent. Consumers pay less while the wedge rises. Conversely, if costs increase faster than prices, the ratio can decline despite consumer price increases. Cost and price changes must be tracked together.

Cost-shifting policy provides identification only within its institutional setting. An import tariff may alter sourcing costs, competitor entry, product availability, quality, exchange-rate exposure, and expectations. A tax levied on sellers is not automatically equivalent to a change in resource marginal cost from the perspective of welfare. Researchers must distinguish consumer gross price, producer net receipts, taxes, subsidies, and imported inputs. The relevant counterfactual should state which of these components changes and why.

Data Architecture and Estimation Discipline

Research design begins with a data dictionary that maps variables to their economic role. Minimum useful fields include an observation identifier, product and firm identifiers, ownership date, market and calendar period, physical output and sales quantities, net transaction price, revenues, expenditure categories, variable-input quantities and prices, capital or capacity indicators, product characteristics, and geographic and institutional controls. Demand instruments and cost shifters require their own provenance. Missing physical quantities should be recorded explicitly instead of silently replaced by deflated revenues as if the variables were identical.

At the transaction level, rebates, bundling, discounts, taxes, returns, shipping, and currency conversion affect comparability. At the firm level, consolidated sales may include foreign subsidiaries while input expenses refer to domestic production. Sales weights can count intermediate goods multiple times in value chains. Entry and exit can alter both the population and its composition. The economic definitions of quantity, price, geographic market, and industry classification must remain consistent across time or be linked by documented concordances.

The choice of panel design depends on the research question. An unbalanced panel may accurately represent entry and exit while introducing changing composition in the observed average. A balanced panel facilitates within-firm comparisons while excluding firms whose absence may itself be economically important. Researchers can report estimates for both, decomposing entry, exit, and reweighting rather than assuming one sample is inherently superior.

For uncertainty quantification, an elasticity divided by a cost share is a nonlinear statistic. Differentiating mu = theta/alpha gives dmu approximately dtheta/alpha minus theta dalpha/alpha squared. The first-order variance approximation includes theta uncertainty, alpha uncertainty, and their covariance. Treating the estimated elasticity as known is inappropriate if it is itself a generated regressor. Ratio tails can become unstable when alpha is small; trimming must be transparent because high-markup observations may be removed selectively.

A bootstrap should replicate the actual sequence of estimation, sampling, input and output cleaning, first-stage estimation, and aggregation at the appropriate clustering level. Where firms share markets or common shocks, independent resampling of individual observations can be misleading. Sensitivity to instruments, functional form, accounting convention, sample weights, and treatment of implausible estimates remains essential even after a narrow statistical confidence interval is obtained. Sampling uncertainty conditional on a model is distinct from uncertainty about the model.

A reproducible study should provide data-vintage identification, population restrictions, code, unit harmonization, table definitions, estimation objective, convergence diagnostics, and disclosure restrictions. Where proprietary microdata cannot be redistributed, synthetic examples establish code structure but do not constitute replication. This monograph uses only published evidence and fully specified illustrative values; it makes no claim to have obtained the underlying confidential datasets.

Aggregation, Concentration, and Distribution

A population of firms has multiple defensible summary statistics. Let revenue weights be s_i = R_i/sum R_i and firm markups mu_i. The arithmetic revenue-weighted mean is sum s_i mu_i. If output prices and marginal costs refer to comparable unit-level margins and q_i is the appropriate quantity, the aggregate ratio of total revenue to the sum of marginal costs times quantities is 1/(sum s_i/mu_i). This harmonic expression weights inverse ratios and is a different estimand. It is not automatically a general-equilibrium aggregate markup.

For two firms with markups 1.10 and 2.00 and equal revenue shares, the arithmetic mean is 1.55; the harmonic expression is approximately 1.419. If the high-markup firm's revenue share rises to 0.80 while both firm markups remain fixed, the arithmetic mean becomes 1.82 and the harmonic expression becomes approximately 1.719. Neither firm changed its individual wedge. The aggregate rose entirely through reallocation. A median of firm-level ratios need not change when high-markup firms gain weight or when upper-tail ratios rise.

For a continuing set of firms with weights and markups at two dates, the change in the arithmetic mean can be written exactly as a sum of midpoint weights times changes in markups plus midpoint markups times changes in weights. This is a transparent arithmetic decomposition, not an economic causal decomposition. Entry and exit require a consistent extension, such as separately defined contributions using the union of firms and a stated treatment of absent observations. The identification of reallocation does not identify the mechanism generating it.

Market concentration measures the distribution of shares across competing entities within a chosen market boundary. With percentage shares, the HHI is the sum of squared shares; with fractional shares it is scaled from zero toward one. Four equal firms with percentage shares of 25 each produce 2,500. If two combine and all shares otherwise remain fixed, the new shares 50, 25, 25 produce 3,750, a mechanical increase of 1,250 points. These values make no demand or cost assumption and predict no particular new price.

The same measured HHI can describe markets with or without imports, product differentiation, geographic isolation, rapid potential entry, switching costs, or excess capacity. National industrial statistics may group firms that do not compete locally, whereas local-market boundaries can omit online or cross-border alternatives. Concentration should be interpreted as structural evidence conditioned on boundary definition. The concentration–markup link is a separate economic claim.

International and Sectoral Evidence

The empirical record is heterogeneous in both observation units and identification methods. The following cases illustrate distinct questions rather than a statistically pooled worldwide estimate.

U.S. differentiated products and brewing

The 1995 automobile study of Berry, Levinsohn, and Pakes analyzes a product market with substantial observable differentiation. Its methodological object is substitution and supply-side equilibrium, not the deduction of a universal markup from an industry's accounting return. Product characteristics help discipline which buyers substitute between vehicles, while ownership determines which substitution flows enter a firm's pricing incentives.

The U.S. brewing industry offers an especially useful comparative design because production and demand approaches can be examined on related economic activity. De Loecker and Scott's Census working paper (2017) explicitly compares their different restrictions. Their later brewing manuscript (June 17, 2026; subsequently published online in September 2026) reports compatible average estimates from the two approaches and examines wholesale-retail conduct. In the author version, the methods draw on different but overlapping data: brewery production records for 1972–2007 and retail demand observations, with overlap in 2007. Retail data omit some outlets and consumption away from home; domestic production data include exports and omit direct imports. The authors also replace selected demand instruments with a moment derived from production-side markups. This comparison illustrates independent identifying information, while differing coverage and the treatment of retail costs require explicit alignment. The cited text is the accessible author version; journal-publication metadata were confirmed in the indexed publisher record.

Trade reform and India

De Loecker, Goldberg, Khandelwal, and Pavcnik's Prices, Markups, and Trade Reform (2016) uses India's liberalization experience and product-level information to separate prices, marginal costs, and markups. The authors report that lower input tariffs contributed to cost reductions exceeding the corresponding price reductions. Markup adjustment therefore affected the distribution of the gains from reform.

This result illustrates two simultaneous margins of adjustment. Lower output tariffs expose firms to competitive pressure; lower input tariffs reduce the cost of sourcing resources. These forces can operate in different directions on the price–cost wedge. A fall in price is compatible with an increase in markup if the reduction in marginal cost is proportionally larger. The welfare incidence of the reform cannot be recovered from the price change alone because productive efficiency, varieties, entry, and distribution of rents also matter.

German retail gasoline

Assad, Clark, Ershov, and Xu (2024) investigate the German gasoline pricing algorithms introduced widely around 2017. Their empirical design infers adoption from changes in pricing indicators and uses headquarters-level adoption as an instrument for station-level behavior. The authors report that effects on measured margins depended on local competitive conditions, particularly the adoption of competing stations.

The evidentiary distinction is important. A margin response following technology adoption measures a change in an observed price–cost proxy in a particular market setting. It does not directly reveal every station's full opportunity cost or establish a legal agreement. The causal interpretation depends on the credibility of the adoption measure, instrument exclusion, and market-level exposure definition. The external applicability of the result to online marketplaces, hotels, freight services, or grocery retail requires separate evidence about adjustment frequency, strategic interaction, and cost variation.

Aggregate U.S. and cross-country trends

De Loecker, Eeckhout, and Unger (2020) report in The Rise of Market Power an increase in their aggregate markup from 21 percent above marginal cost in 1980 to 61 percent at the sample endpoint. The authors attribute much of the movement to increases in the upper tail and shifts of sales toward high-markup firms, while the median remained comparatively stable. These figures describe the study's historical series and estimator, not a current-year measurement or a global average.

Traina's 2018 working paper, Is Aggregate Market Power Increasing?, reaches more modest conclusions when accounting for a broader cost convention and considering the representativeness of listed firms. It responds to the earlier research version of the rising-markup literature rather than directly replicating the 2020 final article. The competing findings demonstrate why cost classification and sample coverage are material components of the estimand.

The IMF's April 2019 World Economic Outlook, chapter 2, reports a moderate rise in markups in advanced economies, close to 8 percent since 2000 in the studied period, with a much smaller change in the emerging-market group covered. A relative percent change is not a change of eight percentage points in the Lerner index, nor is the sample a census of every economy. The institutional report broadens geographic coverage but shares the general measurement challenges that affect firm financial data. The evidence supports studying country and industry heterogeneity instead of treating one aggregate trajectory as a worldwide structural fact.

These studies collectively reveal the role of context. Sectoral methods exploit rich strategic information but cover narrower markets. Broad panels extend coverage at a cost in direct observability of economic quantities. Policy changes can identify causal effects under specific restrictions but may not establish long-run responses across other institutions. An informative scientific comparison begins with the question each study was capable of answering.

Competition Policy and Welfare Interpretation

In the elementary one-market model with linear inverse demand p = 100 - q and constant marginal cost 20, competitive quantity is 80 and price is 20. A single profit-maximizing seller chooses quantity 40 and price 60. The gross markup is 3 and the normalized margin is 2/3. Under quasilinear welfare, absence of externalities, and fixed product quality, the efficiency loss is the triangular area of 800 monetary units: one-half times the quantity shortfall of 40 times a wedge of 40.

The example demonstrates the welfare consequence of a wedge relative to a defined feasible allocation, not a formula for measuring actual welfare from a markup alone. With fixed costs, a competitive price may fail to support entry; with product differentiation, consumers value variety; with research costs, future product quality may depend on expected returns; with externalities, private marginal cost differs from social marginal cost. Welfare questions require explicit choices of time horizon and valuation of the consequences for consumers, workers, suppliers, entrants, and owners.

In a market with heterogeneous markups, resource misallocation can arise when marginal social benefits and opportunity costs vary across uses in ways that strategic pricing distorts. However, an efficient reallocation toward a superior producer can increase concentration and an arithmetic mean of firm markups at the same time. The sign of a welfare change depends on the joint movement of costs, output, quality, variety, resources, and entry, not simply on whether a weighted wedge rises.

Competition policy typically combines structural screens with evidence on substitution, entry, capacity, buyer power, conduct, and efficiencies. As a concrete dated institutional example, Guideline 1 of the 2023 Merger Guidelines of the U.S. Department of Justice and Federal Trade Commission describes a structural presumption when post-merger HHI exceeds 1,800 and its increase exceeds 100 points. This is a rebuttable screening rule in that document, not a universal economic theorem or a statement about the legality of an individual transaction in every jurisdiction. Applying any current legal standard requires verification of the relevant law, date, and forum.

When assessing a transaction, the analyst can compare inferred demand-side margins with product-level accounting and engineering evidence. Simulated price effects must be distinguished from observed historical effects. Alternative assumptions about common ownership, pricing conduct, efficiency changes, and entry may materially alter results. The analyst should communicate the distribution of plausible counterfactuals instead of presenting an apparently precise estimate generated by one untested strategic model.

Comparative Robustness and Analytical Synthesis

The original analytical contribution of this dossier is a synthesis of the correspondence between empirical measures and the restrictions that give them economic meaning. It does not claim discovery of a new theorem or introduce a named economic theory. The primary result is methodological: a measured wedge is interpretable when its decision unit, cost horizon, identifying variation, aggregation rule, and intended counterfactual can be made explicit and mutually consistent.

Consider an illustrative industry observation with price 150, marginal cost 100, market sales of 1,000 units, and a large fixed expenditure of 45,000. The gross markup is 1.50; the normalized margin is one-third; economic surplus after the specified fixed expenditure is 5,000, or 3.33 percent of revenue. A single-product demand elasticity of magnitude three is compatible with that margin under static interior monopoly pricing. A physical output elasticity of 0.60 and flexible-input revenue share of 0.40 are also compatible under cost minimization. These three independent calculations target the same wedge by construction.

Now alter one restriction at a time. Under a common-owner multiproduct pricing condition, a demand elasticity of magnitude three for a single product need not characterize the full pricing incentive. Under the production restriction, substituting revenue elasticity 0.40 for output elasticity 0.60 returns a misleading ratio of one. Reclassifying a fixed expense as variable can change the measured expenditure share and thereby the estimated ratio, even when the stipulated economic wedge has not changed. Measuring the surplus after fixed cost instead yields 3.33 percent, which describes a different economic object. Agreement in the initial case therefore comes from compatible definitions rather than a general interchangeability of estimators.

A practical research sequence begins by specifying the estimand. Next, record which data are observed and which objects are inferred. Then articulate the economic restrictions necessary for the estimate, test their observable implications where possible, and examine whether an alternative mechanism fits the same observations. Finally, compare compatible estimates and report the contribution of model variation, sampling variation and measurement variation separately.

The strongest triangulation uses evidence with distinct sources of identifying information. Two estimates derived from the same revenue measure and the same assumption about unobserved prices can agree while sharing the same bias. A direct engineering estimate can provide independent support, although its horizon may differ. Demand and production estimates for the same product can be compared only when their ownership boundary, price convention, cost definition, and decision horizon coincide. Strategic conduct and welfare remain additional questions even after agreement about a wedge.

The three provisional research expectations in the scientific passport receive conditional support. First, the formal examples demonstrate that the economic interpretation changes when conduct, cost, and input restrictions change. Second, the comparative cases show institution-specific responses in pricing, trade exposure and technology adoption. Third, the aggregation examples prove that averages, accounting returns and product-level markups may move differently. These conclusions concern logical conditions and the reviewed case evidence; they do not constitute newly calculated statistical tests of a global population.

The literature's disagreements are scientifically useful because each exposes a boundary of inference. A demand model makes substitution explicit but may impose contestable supply assumptions. A production ratio can scale across many firms but may rely on an unobserved elasticity. A direct-cost estimate makes resource requirements more visible but can omit nonaccounting opportunity costs. A structural screen is easy to reproduce but is not itself a causal estimate. No single observed series can resolve every empirical question about competition.

Research Limits and Future Directions

Several important research directions follow from the preceding analysis. First, product-level quantities and transaction prices should be linked to actual production establishments while preserving ownership relationships over time. The resulting data would make it easier to separate changes in product composition, output prices, productivity, and variable-input use. Second, longitudinal observation of contractual costs and capacity constraints can distinguish short-run incremental production decisions from long-run capacity decisions.

Third, researchers should investigate how marketing and digital infrastructure contribute both to production and to demand. Data-center utilization, software development, distribution networks, and customer acquisition do not fit a uniform distinction between physical variable costs and fixed expenditures. A zero-price user-facing service may derive revenue from transactions with advertisers or other sides, making a single-user-price markup ill-defined. Platform research must identify the relevant vector of prices and cross-side effects before selecting an estimator.

Fourth, international research should harmonize currency, industry taxonomy, production unit, firm coverage, and price conventions. Extending a panel to more jurisdictions is scientifically valuable only when the interpretation of its variables remains intelligible. Cross-country differences in market institutions and reporting incentives can generate real heterogeneity as well as measurement heterogeneity. Both deserve analysis.

Fifth, empirical competition studies should release replication materials consistent with confidentiality protections and should state exactly which raw datasets cannot be independently acquired. Transparent limitations allow later investigations to distinguish a reproducible method from a fully replicated empirical result. This dossier has checked selected original texts, primary publisher abstracts and official reports; access to subscription full texts and some direct links was incomplete. It has not independently replicated any confidential estimates and does not present an exhaustive database of all market-power research through November 2026.

Conclusion

Empirical measurement of market power consists of the disciplined inference of an economic wedge and the evaluation of the institutional and strategic conditions that sustain it. A markup ratio, a normalized margin, a financial profit rate, a conduct parameter, a concentration measure and a welfare estimate answer distinct questions. Their interrelations depend on explicitly specified assumptions.

Demand-based estimation connects price choice to own- and cross-product substitution. Conduct analysis uses variations in equilibrium responses to identify strategic restrictions. Production-based estimation connects variable input decisions to marginal cost through a physical-output elasticity. Direct-cost and pass-through evidence provide complementary information about incremental resource requirements and adjustment. Each method becomes scientifically informative when its target, data, and identifying assumptions correspond.

The comparative record supports heterogeneous rather than universal claims about changes in market power. Trade reform may reduce prices while raising a price–cost ratio. Algorithmic pricing may alter margins differently across market structures. Aggregate markups can rise through concentration of revenue in the upper tail while individual firm markups remain stable. Disputes over cost classification and revenue elasticities affect the interpretation of broad panels. These mechanisms establish why changes in the measured wedge cannot be converted mechanically into conclusions about collusion, innovation, or total welfare.

The appropriate standard of evidence is therefore the alignment of economically meaningful measures, empirically defensible restrictions, and honestly delimited counterfactuals. The scientific value of a markup estimate lies in the precision with which its meaning and evidentiary conditions are established.

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. Original source.
  • Berry, Steven, James Levinsohn, and Ariel Pakes. (1995). Automobile prices in market equilibrium. Econometrica, 63(4), 841–890. DOI: 10.2307/2171802. Original source.
  • Bond, Steve, Arshia Hashemi, Greg Kaplan, and Piotr Zoch. (2021). Some unpleasant markup arithmetic: Production function elasticities and their estimation from production data. Journal of Monetary Economics, 121, 1–14. DOI: 10.1016/j.jmoneco.2021.05.004. Original 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. Original source.
  • De Loecker, Jan, and Frederic Warzynski. (2012). Markups and firm-level export status. American Economic Review, 102(6), 2437–2471. DOI: 10.1257/aer.102.6.2437. Original source.
  • De Loecker, Jan, and Paul T. Scott. (2017). Estimating market power: Evidence from the US brewing industry. Center for Economic Studies Working Paper 17-06, U.S. Census Bureau. Original source.
  • De Loecker, Jan, and Paul T. Scott. (2026). Markup estimation using production and demand data: An application to the US brewing industry. Author manuscript, June 17, 2026. Subsequent journal publication: Review of Economic Studies, advance online publication, September 22, 2026. DOI: 10.1093/restud/rdag109. Author manuscript.
  • De Loecker, Jan, Jan Eeckhout, and Gabriel Unger. (2020). The rise of market power and the macroeconomic implications. Quarterly Journal of Economics, 135(2), 561–644. DOI: 10.1093/qje/qjz041. Original source.
  • De Loecker, Jan, Pinelopi K. Goldberg, Amit K. Khandelwal, and Nina Pavcnik. (2016). Prices, markups, and trade reform. Econometrica, 84(2), 445–510. DOI: 10.3982/ECTA11042. Original source.
  • Hall, Robert E. (1988). The relation between price and marginal cost in U.S. industry. Journal of Political Economy, 96(5), 921–947. DOI: 10.1086/261570. Original source.
  • International Monetary Fund. (2019). The rise of corporate market power and its macroeconomic effects. In World Economic Outlook, April 2019: Growth Slowdown, Precarious Recovery, chapter 2. Washington, DC: IMF. Original source.
  • Lerner, A. P. (1934). The concept of monopoly and the measurement of monopoly power. Review of Economic Studies, 1(3), 157–175. DOI: 10.2307/2967480. Original source.
  • Traina, James. (2018). Is aggregate market power increasing? Production trends using financial statements. Stigler Center New Working Paper Series No. 17, University of Chicago Booth School of Business, February. Original source.
  • U.S. Department of Justice and Federal Trade Commission. (2023). 2023 Merger Guidelines, guideline 1, section 2.1. Washington, DC. Original source.