RESEARCH DOSSIER

Increasing Returns and Coordination Failures

Abstract. Increasing returns can make the viability of productive activity depend on market size, complementary investments, technological adoption, and the spatial distribution of economic activity. This research monograph investigates the conditions under which such relationships create coordination failures. It distinguishes nonconvex production from external economies, strategic complementarities from equilibrium multiplicity, and equilibrium selection from social welfare. The study integrates established theoretical literature, explicitly derived numerical examples, and comparative empirical evidence on the Tennessee Valley Authority and South Korea's industrial policy. It assesses financing, network compatibility, institutional commitment, identification, and the incidence of policy gains. The findings establish that coordinated intervention requires separate evidence of a feasible superior allocation and of a mechanism capable of achieving it at an acceptable social cost. The analytical synthesis does not claim a novel theorem or original empirical estimates.

Keywords: Increasing returns; economies of scale; nonconvex production; coordination failure; strategic complementarity; multiple equilibria; industrialization; network externalities; path dependence; agglomeration; industrial policy; welfare economics

Introduction

Economic development, technological diffusion, industrial location, and market structure are connected by a fundamental problem: the conditions under which an economically advantageous activity becomes individually worthwhile may depend on the decisions of other participants. A factory needs customers, a supplier needs factories, a worker needs employment opportunities, and a network user needs compatible participants. The opportunity available to one decision maker is partly produced by the decisions of others. When these interactions are sufficiently strong, economic coordination becomes an object of analysis in its own right.

The study concerns two distinct properties of economic organization. Increasing returns describes how output, average cost, or the value of participation changes with a specified scale of activity. Coordination failure describes a situation in which individually appropriate choices sustain an outcome inferior to a feasible alternative requiring compatible changes in behavior. A demonstration of either property is insufficient to establish the other. The distinction determines both the explanation of a low-activity equilibrium and the legitimacy of interventions designed to change it.

The research question is whether the economy can organize profitable complementary activity using its existing institutions and given resources, and what evidence would support a different answer. The inquiry proceeds from production technology to individual incentives, then to decentralized equilibrium, dynamics, measurement, and policy. The mathematical examples are pedagogical constructions using hypothetical units rather than new estimated models. The empirical discussion evaluates published research; it does not claim replication of underlying microdata. The international scope is analytical: evidence from the United States and South Korea is examined alongside general mechanisms that could operate elsewhere, without treating two national experiences as representative of every economy.

Historical development and theoretical foundations

The idea of coordinated industrialization developed in response to the problem of establishing modern industries in markets where initial purchasing power was limited. Murphy, Shleifer, and Vishny (1989), in their analysis of the big push, explicitly trace this problem to Rosenstein-Rodan's 1943 work and examine it through imperfect competition. Their contribution makes the profitability of industrial investment conditional on the pattern of investment in other sectors, rather than treating industrial expansion as the independent addition of firms.

A complementary tradition approaches economic organization through the existence of market equilibrium. Arrow and Debreu's (1954) competitive equilibrium existence analysis specifies assumptions on consumers and producers within an integrated production and exchange economy. The results concern existence under stated conditions. In particular, the production-side convexity assumptions are relevant when studying indivisibilities and scale-intensive production. A failure of those assumptions removes this particular guarantee; it does not constitute a universal theorem of market nonexistence.

Dixit and Stiglitz (1977) redirect attention to optimum product diversity. Product differentiation gives firms room to recover setup costs while serving a market at prices above marginal cost. The number of varieties, the size of each operation, and the resulting consumption possibilities become simultaneous welfare questions. This tradition reveals that scale economies can be accommodated through imperfect competition even when a familiar competitive benchmark is unsuitable.

The interaction of independent decisions becomes explicit in Cooper and John's (1988) coordination failure analysis. Their framework brings production, matching, and demand into the study of interconnected incentives. The subsequent adoption and geography literature examines how these interactions unfold over time and across locations. These approaches share a concern with endogenous conditions for profitable behavior, but each retains a different economic mechanism and different conditions for identifying failure.

Increasing returns and nonconvex production

Technological scale and cost

Let output be q = f(x), where x is a vector of rival inputs. Increasing returns to scale at x means that f(tx) exceeds t f(x) for a specified t greater than one when all components of x are scaled proportionally. Global increasing returns require the inequality throughout the relevant domain; a local scale elasticity above one is a different, differential statement. For a differentiable production function the elasticity of f(tx) with respect to t at t = 1 is the sum of the individual input elasticities. Measurement requires an explicit boundary for inputs: the productivity of doubling a factory's workforce while leaving a shared electricity network unchanged does not reveal what happens when the entire economy doubles all resources.

A simple cost example isolates the indivisibility. Suppose total cost is zero at q = 0 and C(q) = F + cq for positive q, where F > 0 is an avoidable setup cost and c > 0 is constant marginal cost. With F = 100 and c = 5, production of 10 units costs 150, or 15 per unit. Production of 20 costs 200, or 10 per unit. Doubling the output raises cost by only one third. For positive output average cost is c + F/q, which falls as q grows, while marginal cost remains c.

At q = 0 the firm avoids F, so average cost at zero is undefined. The distinction matters: a sunk F already incurred does not enter the forward-looking decision to continue production in the same way that an avoidable cost enters new investment. Fixed capital may also deteriorate, need replacement, or have a resale value. A proper profitability calculation assigns these costs to the decision horizon in which they can change.

Nonconvexity and pricing

Represent feasible output-cost pairs by points (q,z) with z at least C(q). The midpoint between (0,0) and (20,200) is (10,100). It is infeasible because producing 10 costs 150. Consequently the feasible set in this example is nonconvex.

At a unit price p = c, a new producer earns revenue cq and incurs F + cq; profit is minus F whenever q is positive. At p greater than c the firm requires sufficient demand to cover F. If q is limited by installed capacity, the profit-maximizing positive quantity may be at the capacity limit; with unlimited linear sales at a fixed p greater than c, profit has no finite maximizer. This is a reminder that the cost example alone is not a complete equilibrium model.

With two-part tariffs, a producer may finance the fixed cost by an access charge and price usage at marginal cost. Such an arrangement can be efficient in a stylized setting with known preferences, no participation distortion, enforceability, and a suitable cost-sharing rule. Access charges can exclude low-demand consumers; uniform per-unit markups may instead distort intensive use. Efficient provision therefore requires attention to both adoption and utilization.

Internal, external, and dynamic returns

Internal scale economies operate within a producer's own cost or production function. External scale economies arise because the scale of a cluster, industry, or network changes opportunities available to individual participants. Learning by doing is an intertemporal mechanism in which accumulated experience changes future costs. Network effects operate through the benefit of participation conditional on other users. None of these mechanisms can be assumed merely because a product has a large market.

Romer's (1990) endogenous technological change illustrates a further distinction: knowledge can be used repeatedly without being physically depleted by each use, while access can remain restricted. The combination influences the possibility of decentralized financing of innovation. Rival research labor and equipment remain scarce even when an idea is reusable. Thus the cost of obtaining a design, producing copies of it, and integrating it into a productive system must be distinguished.

Strategic complementarities and equilibrium selection

The logic of complementary decisions

Let two firms independently decide whether to make an investment, denoted I, or refrain, denoted N. Normalize the payoff from N to zero regardless of the other's behavior. Let the payoff from I be -3 when the other chooses N and +5 when the other chooses I, measured in identical hypothetical monetary units. This is a standard illustrative binary coordination game, not a proposed original theory or calibrated economic model.

Firm A / Firm B N I
N (0, 0) (0, -3)
I (-3, 0) (5, 5)

Each firm chooses I when the other has chosen I and N when the other has chosen N. The two diagonal profiles are strict Nash equilibria. Joint installation yields a payoff of five to each participant, whereas joint abstention yields zero to each. Under the stipulated payoffs and absence of affected third parties, joint installation Pareto-dominates joint abstention. Individual maximization can therefore sustain a collectively inferior result without either participant making an error.

With probability r that the other invests, expected payoff from investment is 5r - 3(1-r) = 8r - 3. The indifference threshold is r = 3/8. Independent randomization at this probability is also a mixed-strategy equilibrium; its existence does not make all coordinated outcomes equally likely. The welfare ranking of the pure profiles is separate from any criterion of equilibrium selection.

For example, with r = 0.25, expected investment payoff is -1. With r = 0.50 it is +1. A change in expectations can alter action holding production opportunities and prices implicit in these payoffs fixed. Yet if the change in expectations is unsupported by actual follow-through, the investment may lose money. Credible coordination requires more than optimism.

Complementarity and unique equilibrium

Let a decision maker choose an interior level of activity a_i with payoff a_i(b + gamma a_j) - a_i squared/2, where b is positive and gamma lies between zero and one. The unconstrained best response is a_i = b + gamma a_j. In a symmetric interior equilibrium activity is b/(1-gamma). With b = 2 and gamma = 0.5, equilibrium is 4 for each agent. Substituting the responses gives a unique pair; under synchronous best-response adjustment deviations contract by the factor gamma. Complementarity is present even though multiplicity is absent.

The example also introduces a conventional multiplier: holding gamma fixed, changing b by one changes equilibrium activity by 1/(1-gamma), or two units with gamma = 0.5. This multiplier is a property of the specified system and not an estimate of a fiscal multiplier. Whether an activity increase improves welfare depends on the full payoff structure, resource constraints, and external costs. Cooper and John (1988) analyze related strategic complementarities in a broader class of overlapping macroeconomic mechanisms.

External effects and coordination failure

An increase in one participant's action can raise another's payoff without necessarily increasing the other's best response. Conversely, increasing differences in a payoff function can arise while limiting aggregate welfare through third-party harm. Strategic complementarity concerns marginal incentives, whereas positive external effects concern levels of benefits or costs. A policy analysis must establish both the relevant behavioral response and all material welfare consequences.

Pure multiplicity is also compatible with several equally efficient equilibria, for example two equally good compatible standards. Selection has practical importance there but no inherent Pareto improvement within the simple example. The welfare question requires knowledge of the feasible alternatives and the opportunity cost of switching.

Market size, product diversity, and industrial organization

The financing of capacity

For a firm with fixed cost F and unit contribution margin m = p - c greater than zero, break-even volume is F/m. With F = 120 and m = 6, the threshold is 20 units. An isolated market producing sales of 12 gives a loss of 48; a jointly expanded market supporting sales of 30 gives profit of 60. These calculations assume the margin and costs remain constant across scenarios. They establish arithmetic viability conditional on demand, not why demand grows or how resources are financed.

Aggregate market size can change with participation through wages, purchases of intermediate inputs, the availability of complementary services, or lower logistical costs. A full general equilibrium mechanism requires tracing where the income and purchasing power originate and where they go. Counting new wages as entirely additional demand without subtracting displaced income or alternative uses double counts resources. Moreover, income spent on imports or saved may produce a different domestic demand response than income spent on locally manufactured goods.

The big-push analysis of Murphy, Shleifer, and Vishny (1989) supplies a formal demand-spillover mechanism. The fact that several activities can break even together when each would lose money alone is its central relevance to this study. The stylized payoff example above isolates the conditional-profit logic and leaves the distribution of income and expenditure to the substantive sectoral analysis.

A textbook differentiated-product calculation

With constant elasticity of substitution sigma > 1, a firm facing residual demand with constant own-price elasticity sigma chooses price p = sigma c/(sigma-1) under the standard monopolistic-competition approximation. This follows from marginal revenue p(1-1/sigma) = c. With free entry, identical firms, and zero economic profit, contribution margins cover F: q = F/(p-c) = F(sigma-1)/c.

For sigma = 4, c = 3, and F = 60, price is 4, margin is 1, and break-even output is 60. Revenue of 240 equals cost 60 + 180. The calculation concerns a symmetric textbook case with constant-elasticity demand, a fixed aggregate-price environment for an individual firm, and no binding capacity constraint. Free entry also needs market demand sufficient to sustain the implied set of firms.

Dixit and Stiglitz's (1977) fixed-cost and variety analysis motivates the welfare issue behind this exercise. A larger available market can support more differentiated products, while prices above marginal cost can restrict consumption. The welfare optimum cannot be inferred from the markup alone; the household's preferences over variety, cost of additional product introduction, and possible distortions in entry each matter.

Competition and bargaining

Internal scale economies can favor concentration because a larger producer spreads overhead across more units. Market share by itself does not measure the size of scale economies: product quality, acquisitions, entry restrictions, and exclusionary strategies can also produce concentration. Even with genuine cost advantages, the producer's ability to withhold output, price discriminate, or foreclose rivals determines distribution and welfare effects.

A coordinated buyer commitment may allow an entrant to finance scale-sensitive investment without public subsidies. A long-term procurement contract can reduce demand uncertainty, but its duration and exclusivity may entrench a technology that later becomes inferior. Competition among alternative financing and procurement arrangements is therefore part of the institutional comparison, rather than an assumption that scale demands a single permanent supplier.

Dynamics, lock-in, and expectations

Adoption paths

Consider two technologies with benefits that depend on installed users. If users of a new standard need counterparties using the same standard, its value to any one user depends partly on expected adoption. Adoption may increase installed capacity for support, spare parts, training, and complementary software. Switching can require replacing accumulated assets or coordinating with business partners, and these costs affect both entrants and existing users.

Arthur's (1989) model of technological lock-in studies adoption feedback and the possibility that early events influence lasting technology choices. An observed dominant technology could be efficient, inferior, or ambiguous depending on the counterfactual evolution of each technology. Dominance is an observation; inefficiency requires additional evidence.

A simple dynamic illustration makes the threshold logic explicit. Suppose a population revises adoption each period under a deterministic update rule: the share x moves to 0.1 + 0.8x when x is below 0.5, and to 0.3 + 0.6x when x is at least 0.5. The lower fixed point solves x = 0.1 + 0.8x, yielding 0.5, which is outside the strictly lower domain. The upper fixed point solves x = 0.3 + 0.6x, yielding 0.75, which belongs to its domain. For an initial x = 0.25 the lower sequence approaches 0.5 from below without reaching it in finite exact arithmetic. This deliberately chosen example demonstrates that apparent threshold behavior need not yield two fixed points. A discontinuity and domain restriction must be checked separately from the algebraic fixed-point equation.

Changing the lower rule's intercept to 0.04 yields a valid lower fixed point at 0.2 and retains the upper fixed point at 0.75. Starting below 0.5 now leads toward 0.2; starting at or above 0.5 leads toward 0.75. A one-time shift past 0.5 can change the trajectory, but this conclusion depends entirely on the stipulated update rule. No welfare ordering follows from adoption shares alone. Both calculations are hypothetical illustrations of standard dynamical reasoning.

Commitment and information

A coordination agreement can address pessimistic expectations by conditioning a participant's investment on credible commitments from others. Conditional orders, escrow, verifiable milestones, and shared infrastructure contracts make expectations partly enforceable. An unconditional statement of intent cannot substitute for financing, legal rights, or delivered components.

Sponsorship changes the allocation of incentives. Katz and Shapiro (1986), in their network adoption analysis, identify the importance of property rights and promotion incentives when technologies compete. Compatibility decisions can be socially significant in both directions: insufficient interconnection can fragment a network, while universal standardization can sacrifice valuable differentiation. A transition away from an established standard should compare the present cost of migration with future gains, including risk of disruption for users who would otherwise remain served.

In empirical investigations, persistence is consistent with feedback but also with durable infrastructure, long-lived contracts, persistent policy transfers, stable geography, and fixed preferences. The analyst needs a counterfactual for the path that would have prevailed without the initial event. Historical dependence becomes scientifically informative when the candidate mechanisms can be separated by variation in timing, exposure, or switching costs.

Spatial concentration and productive networks

Location as a coordination decision

When firms choose locations, the cost of delivering goods, hiring specialized workers, purchasing inputs, and cooperating with other establishments may depend on local concentrations of activity. A producer can gain access to a larger customer base by entering a cluster, and its entry may enlarge the local base of demand. The existence of increasing returns within firms and the geographic distribution of expenditure together affect the pattern of specialization.

Krugman's (1991) economic geography model gives a formal illustration of endogenous spatial differentiation based on manufacturing scale, transport costs, and demand location. In this setting an initial difference between regions may reinforce itself. The theoretical result is conditional on the model's assumptions and does not justify predicting a single metropolitan outcome for every transport improvement.

A firm-centred accounting measure of lower unit costs must be adjusted when the cluster raises land rents or transport congestion. Workers may benefit from access to a wider labor market, while their housing costs rise. Landowners may capitalize part of the benefit. Producer output gains and resident welfare gains therefore require separate measures. This distinction becomes especially relevant when the objective of a regional intervention is improving the opportunities of existing households rather than attracting additional establishments.

Production linkages and capacity constraints

In a production network, firms use one another's outputs. A steel producer requires sufficient demand from manufacturers; manufacturers may in turn require reliable intermediates of suitable quality and cost. Complementarity can propagate upstream or downstream. A bottleneck at one stage may reduce the return to investment throughout the system.

A linear accounting example uses the identity x = Ax + d, where x denotes gross production, A a matrix of constant intermediate input coefficients, and d final demand. If the spectral radius of A is less than one, the solution is x = (I - A) inverse times d. In this conventional fixed-coefficient formulation the propagation of final demand is an accounting implication under fixed coefficients. It is not proof of strategic complementarity or increasing returns. Capacities, input substitution, prices, inventories, trade and financing are outside this linear calculation unless explicitly incorporated.

An apparent network multiplier found in gross revenue data may largely represent transactions counted along multiple stages. Estimation of welfare gains must focus on value added and opportunity costs rather than adding every sale across the supply chain. A meaningful diagnosis also separates scarcity caused by one supplier's market power from a genuinely indivisible shared investment that several suppliers cannot independently finance.

Macroeconomic coordination and development

The relationship between activity in separate markets motivates a macroeconomic account of low utilization. A business may defer hiring when sales are uncertain; households may defer purchases because job prospects are weak; financial institutions may tighten lending when firms are likely to fail. These linked responses can amplify an initiating shock. The amplification mechanism remains compatible with a unique equilibrium if the feedback is insufficient to generate multiplicity.

A high-saving or low-investment observation cannot, by itself, identify a coordination failure. If production technology is genuinely weak, investors lack managerial capabilities, credit is unavailable because repayment is implausible, or real resources are destroyed, the low-activity outcome may reflect fundamentals. A low equilibrium must be shown to be inferior to a feasible alternative at the relevant resource constraints and institutional capabilities. Otherwise a demand expansion may raise prices or imports rather than establish sustainable domestic capacity.

Complementary investments may have different maturation dates. A railway project can require substantial funding before generating any revenue, while users may need years to establish businesses. Even a positive net present value for the combined investments need not make each participant's private project sufficiently profitable. The financing problem concerns the distribution and timing of returns in addition to their sum. Long-term arrangements may help if commitments can be monitored and liabilities credibly assigned. Debt guarantees can shift default exposure onto taxpayers and must be treated as real contingent costs.

The macroeconomic implication is not that every economy requires simultaneous investment in every sector. The relevant set of activities consists of those whose feasible scales or timing are tightly coupled. Access to international suppliers and customers can break some local dependencies; it can also expose producers to foreign competition and foreign demand shocks. Which linkages matter most is a research question dependent on tradedness, infrastructure, skill availability, and commercial institutions.

Empirical identification and measurement

Distinguishing mechanisms

The central empirical problem is separating endogenous feedback from common causes. For firms A and B observed over time, a positive correlation between investment decisions might arise because A affects B, B affects A, both respond to a common price shock, both face changing credit supply, or the data select only surviving firms. A regression of one investment on the other will generally not identify a causal complementarity without additional identifying assumptions.

Increases in productivity associated with establishment size may reflect selection: more productive establishments grow larger. Revenue per worker can rise because of prices or market power even if physical productivity remains constant. For internal economies, engineering estimates of cost functions and plant-level physical outputs can help; for external economies, geographically or commercially relevant exposure must be defined. Data on input prices, quality, product mix, utilization, and establishment births and exits are needed to interpret average costs.

A credible design isolates variation in one participant's capacity or adoption that affects another participant's returns through the mechanism of interest. The direct effect of a public project and spillover effects through neighboring firms are distinct estimands. In a network treatment setting, untreated units can receive exposure through their connections; conventional treatment and control labels alone may no longer define an unexposed counterfactual.

Dynamic and spatial evidence

A difference-in-differences estimate compares changes in outcomes for more-exposed and less-exposed units. Its causal interpretation requires appropriate assumptions about counterfactual trends, treatment timing, other simultaneous shocks, composition, and spillovers. Event-study dynamics can reveal anticipatory movement and delayed adjustments, but a lack of significant pretrends does not establish that all confounding is absent. A place-based program can change migration, employment location, and prices outside the treated area. Estimating treatment effects on the initial population requires tracking that population rather than examining only the post-treatment residents.

Multiple equilibria pose an especially difficult identification problem. A discontinuous reaction near a threshold may be caused by a binding regulatory rule or credit constraint rather than equilibrium switching. Persistent post-treatment activity can reflect installed capital whose depreciation is slow. A claim that temporary coordination moved an economy to another self-sustaining equilibrium requires evidence that the relevant fundamentals and ongoing support are sufficiently accounted for, and that the observed transition mechanism matches the theory.

Welfare accounting

A useful comparison takes the full discounted social benefit of a feasible transition, subtracts real resource costs and opportunity costs, and identifies distribution separately. A grant is a transfer between parties, but administering and financing it can create resource and distortion costs. Any increase in producer profit due to higher prices paid by consumers requires accounting for consumer effects. Congestion, emissions, and resource scarcity must be included when consequential.

For a purely illustrative five-year project, assume annual net benefit 30 after real operating costs, up-front resource cost 100, and the same consumption discount rate of 5 percent each year. The annuity present value is 30 times (1 minus 1.05 to the power -5) divided by 0.05, approximately 129.88. Net present value is approximately 29.88. If annual net benefit falls to 20, the net present value becomes approximately -13.41. These are hypothetical monetary quantities, not empirical policy evaluation. Different beneficiary groups may have distinct welfare weights, and risk or a changing policy environment introduces additional assumptions. The calculation establishes how easily a seemingly profitable transition depends on assumptions about persistent benefits.

Comparative evidence from industrial and regional policy

The Tennessee Valley Authority

Kline and Moretti (2014) study the Tennessee Valley Authority using proposed but unrealized regional authorities as comparison areas. They find contrasting persistence patterns across sectors and distinguish the infrastructure's direct productivity contribution from the spatial allocation of agglomeration effects. Their analysis estimates an increase of about 0.3 percent in national manufacturing productivity from direct productivity effects; the estimated indirect gains within the TVA area are offset elsewhere in the country.

This is a model-based empirical estimate rather than a universal relationship between public investment and national income. The comparison identifies effects if the proposed authorities provide a credible counterfactual conditional on the research design. Its national interpretation additionally depends on labor mobility, spatial adjustment, and estimated agglomeration mechanisms. The study provides an important warning against identifying regional employment growth with aggregate welfare gains. Direct improvements in productive infrastructure may be beneficial even when local agglomeration gains principally redistribute activity.

South Korea's heavy and chemical industry drive

Lane's (2025) study of Korean industrialization analyzes the heavy and chemical industry initiative of 1973–1979 and follows industries after its end. The research uses dynamic difference-in-differences comparisons of targeted and other industries and incorporates exposure to the policy through intermediate input relationships. The reported expansion of targeted activities and downstream benefits persists beyond the immediate intervention.

Several distinct interpretations are possible. Lower-priced or improved domestic intermediates can raise productivity in downstream sectors. An initial shift into activities with greater potential for learning can change later comparative advantage. Additional physical capacity can outlast the policy even without an equilibrium selection mechanism. The reported patterns support durable sectoral and network effects; they do not, by themselves, establish that every initially unprofitable industry had a superior equilibrium available, or that net national welfare rose by a known amount.

This distinction has implications for comparative research. South Korea's sectoral exposure and timing differ from the TVA study's geographic intervention. The unit of comparison, outcome, institutional administration, and counterfactual differ. Combining their results by averaging numerical effects would produce an incoherent estimand. Their common contribution here is to demonstrate how direct intervention, propagation through connected activities, and persistence can be investigated empirically.

Transferability and research design

An investigator evaluating industrial coordination in another economy would first document the infrastructure base, financing conditions, domestic demand and access to export markets. Large differences in fiscal capacity, administrative competence, property institutions, and the preexisting organization of firms change the feasible intervention set. A port or power network may be a binding complement in one country; skills, credit enforcement, or common standards may be binding elsewhere.

Cross-national comparison should identify comparable economic mechanisms rather than assuming national outcomes are a response to a single policy label. Evidence from the two case studies provides a basis for testing questions about persistence and spillovers; global extrapolation remains a separate exercise requiring new observations. No comprehensive world panel or new causal estimates are provided here.

Institutional alternatives and policy design

Assigning the instrument to the problem

When a project is profitable only with complementary private investments, a coordination contract may allocate expected gains and losses more directly than a general subsidy. The appropriate tool depends on whether uncertainty concerns counterparties' commitments, technology quality, demand, financing costs, or effects on third parties. An agreement that promises to purchase output if a factory opens can address the first problem; it cannot directly improve poor engineering or resolve an unrelated foreign exchange constraint.

When indivisible infrastructure serves many participants, joint financing may spread setup costs and support efficient usage. Access rules determine whether entrants can use the infrastructure without duplicating it. Public provision may be feasible when charging for individual uses is difficult, but its opportunity cost remains part of social profitability. Where demand is measurable, competitive tenders and enforceable access commitments can constrain overbuilding and exclusion.

Network compatibility can be promoted through open interfaces, portability, interoperability agreements, or standards developed with participation from affected actors. The choice requires evaluation of security, maintenance, quality, and switching costs. An imposed uniform standard may restrict useful experimentation. Conversely, fragmented incompatible systems can increase duplication of fixed development cost and reduce the value of participation.

Temporary incentives and exit conditions

Return to the hypothetical two-firm game. A subsidy s paid per investing participant increases the private payoff from investing when the counterpart does not invest from -3 to -3+s. If s is strictly greater than 3, both players have a strictly dominant investment action in the given game. At s = 3 the participant is indifferent when the other abstains, so the original low profile may remain a weak equilibrium. The difference between equality and strict inequality is decisive in an instrument intended to remove the lower equilibrium.

With s = 4 and payment to both investors, budget outlay in the high profile is 8. That expenditure is not automatically a social loss of 8: it is received by investors. The public financing burden, administrative costs, distributional effects, and any forgone alternative expenditure must be evaluated. The hypothetical equilibrium is self-sustaining after the subsidy ends only if both participants continue to receive the high-profile private payoff under the subsequent decision structure. If investments are reversible, beliefs and switching conditions determine future actions. If costs change irreversibly, a new decision model is required.

An exit rule may tie support to independently measured volume, completed common infrastructure, verified supply contracts, or declining unit costs. End dates should be established against realistic investment horizons. Permanent support can conceal insufficient private viability and attract efforts to renew political privileges. Abrupt termination may also destroy a beneficial investment if support was explicitly contracted to cover a long maturation period. Evaluation must therefore distinguish an agreed time path from retrospective discretionary rescue.

Government and collective-action failures

A public coordinator has its own information and incentive problems. It may select projects using forecasts that favor incumbents, promise support without fiscal resources, or overlook the displaced productive activities. A government might coordinate a high-output equilibrium that is environmentally damaging or redistributes income from vulnerable groups. An administrative intervention is itself an institutional alternative with costs, risks, and accountability requirements.

Private collective arrangements face related difficulties: free riding, strategic misreporting, hold-up after complementary investment, and agreements that become barriers to entry. The economically relevant comparison is among feasible institutions, considering contractual enforcement and information constraints. A proof of failure under one decentralized mechanism is not a proof of success under any specified alternative.

Analytical synthesis and open research questions

Separating technological, behavioral, and welfare claims

The literature supports a disciplined sequence of analytical questions. The first identifies a scale-dependent opportunity: what changes when a specified activity expands, and over what domain? The second examines which participants control the complementary actions and what information or contracts connect their choices. The third tests the equilibrium properties under those institutions, with explicit attention to the strength of feedback, the possibility of a unique solution, and the range of feasible outcomes. The fourth establishes welfare effects under a common resource and distributional accounting boundary. The fifth studies whether an available institutional change actually improves the resulting allocation.

This sequence is an editorial synthesis of established economics, not a newly asserted theory or doctrine. Its contribution is to prevent several common errors from entering the same inference. Scale economies can be present with a viable monopolistically competitive market; strategic complements can support a unique equilibrium; multiple equilibria can be equally valuable; a local intervention can move activity between regions; and a persistent policy effect can have a technological explanation other than self-fulfilling expectations. The examples developed in this monograph demonstrate each distinction directly or identify what evidence would be required to do so.

The boundary of coordinated activity

The economically significant boundary is determined by actual complementarities and feasible substitution. If a regional manufacturer can source components reliably abroad, its local supplier network may be less essential than its access to ports, power, and foreign buyers. Conversely, a specialized in-person workforce may make local concentration particularly important for some activities. Such variation supports careful project-specific coordination rather than maximizing the number of participants under a single program.

There is no reason to equate the size of a coordination coalition with the scale of the economy. A small coalition can solve a sharply defined interoperability problem. A large infrastructure project may require several layers of commitments and public financing. The social objective should be articulated before comparing coalition sizes. Small coalitions may be easier to enforce yet exclude valuable participants, while large ones may achieve scale but magnify bargaining and monitoring costs.

Open empirical problems

An important research direction is the joint measurement of scale economies, expectation formation, and actual contract enforceability. Firm-level surveys about other participants' intentions can be linked to observed investments and cost data; the causal interpretation of responses would still require exogenous or otherwise credible variation and controls for common shocks. A second direction concerns the decomposition of post-intervention persistence into long-lived assets, organizational learning, network adoption, and continued political support. Combining establishment histories, contractual records, and industry transaction data could help discriminate between these mechanisms.

A third question concerns the geographic and social incidence of coordination policies. Where gains are capitalized into land, an increase in local production may benefit property owners more than workers. Where a new network displaces nonparticipants, wider exposure and transition costs must be measured. Data and methods should follow people and firms beyond the initial treated area and explicitly account for displacement and opportunity costs.

These research questions are proposals for future empirical inquiry, rather than claims that the necessary data have already been assembled. The evidence reviewed here supports conditional mechanisms and careful comparisons. It does not establish a universally optimal form of industrial organization or policy.

Conclusion

Increasing returns alter the relationship between economic scale and private profitability. When the payoff to participation depends on complementary investments, decentralized choices can sustain low levels of activity even though jointly compatible actions would be feasible and mutually advantageous. The formal possibility of such an outcome is established through explicit incentives and equilibrium conditions, not inferred solely from the presence of fixed costs.

The distinction between scale, complementarity, multiplicity, selection, and welfare provides the central analytical result of this monograph. The illustrative cost function demonstrates nonconvexity and the difficulty of financing fixed costs through marginal-cost pricing. The coordination game demonstrates Pareto-ranked equilibria and the sensitivity of investment to expected participation. The continuous-choice example demonstrates complementarity with a unique equilibrium, while the dynamic examples show why fixed points must satisfy their relevant domains and why persistence does not imply inefficiency.

The reviewed empirical research underscores the distinction between changes within an industry or region and aggregate benefits. Productive infrastructure, sectoral linkages, and the duration of policy effects matter, but the evidence must be interpreted relative to each design's counterfactual and institutional environment. An efficient institutional response is one that resolves a demonstrated constraint at a social cost consistent with its benefits, including the costs of displaced activity, uncertainty, administration, and implementation.

The study makes an original analytical synthesis of established research and hypothetical calculations. It claims neither a new named theoretical system nor novel empirical estimates. The remaining research agenda is to identify the actual sources and strength of complementarities, to measure counterfactual outcomes at the relevant scale, and to evaluate feasible mechanisms of coordination with comparable standards of evidence.

References

  • Arrow, K. J., & Debreu, G. (1954). Existence of an equilibrium for a competitive economy. Econometrica, 22(3), 265–290. DOI: 10.2307/1907353. Research source

  • Arthur, W. B. (1989). Competing technologies, increasing returns, and lock-in by historical events. Economic Journal, 99(394), 116–131. DOI: 10.2307/2234208. Research source

  • Cooper, R., & John, A. (1988). Coordinating coordination failures in Keynesian models. Quarterly Journal of Economics, 103(3), 441–463. DOI: 10.2307/1885539. Research source

  • Dixit, A. K., & Stiglitz, J. E. (1977). Monopolistic competition and optimum product diversity. American Economic Review, 67(3), 297–308. Research source

  • Katz, M. L., & Shapiro, C. (1986). Technology adoption in the presence of network externalities. Journal of Political Economy, 94(4), 822–841. DOI: 10.1086/261409. Research source

  • Kline, P., & Moretti, E. (2014). Local economic development, agglomeration economies, and the big push: 100 years of evidence from the Tennessee Valley Authority. Quarterly Journal of Economics, 129(1), 275–331. DOI: 10.1093/qje/qjt034. Research source

  • Krugman, P. (1991). Increasing returns and economic geography. Journal of Political Economy, 99(3), 483–499. DOI: 10.1086/261763. Research source

  • Lane, N. (2025). Manufacturing revolutions: Industrial policy and industrialization in South Korea. Quarterly Journal of Economics, 140(3), 1683–1741. DOI: 10.1093/qje/qjaf025. Research source

  • Murphy, K. M., Shleifer, A., & Vishny, R. W. (1989). Industrialization and the big push. Journal of Political Economy, 97(5), 1003–1026. DOI: 10.1086/261641. Research source

  • Romer, P. M. (1990). Endogenous technological change. Journal of Political Economy, 98(5, Part 2), S71–S102. DOI: 10.1086/261725. Research source