RESEARCH DOSSIER

Knowledge Nonrivalry and Endogenous Growth

Abstract. This monograph examines the role of reusable productive knowledge in endogenous economic growth. It connects the neoclassical accumulation benchmark, the economics of invention, models of horizontal design expansion and vertical creative destruction, semi-endogenous growth, and empirical research on innovation productivity and diffusion. The study distinguishes knowledge as a nonrival input from the scarcity of research labor, access rights, absorptive capability, and physical implementation. Explicit teaching derivations compare balanced-growth implications, stochastic technological advances, research productivity, and adoption dynamics. Evidence is evaluated according to its measurement units, sample boundaries and identification assumptions. The original analytical synthesis establishes how differences among creation, access and utilization explain the coexistence of knowledge spillovers, private innovation rents, declining measured research productivity and persistent international productivity disparities. The work is a bounded synthesis of existing literature with illustrative calculations; it claims neither a new named theory nor independently estimated empirical results.

Keywords: Knowledge nonrivalry; endogenous growth; innovation; research productivity; intellectual property; creative destruction; semi-endogenous growth; technology diffusion; absorptive capacity; productivity measurement

The economic object of knowledge

An economy grows through changes in what its participants can accomplish with available resources. Some changes expand the quantity of labor or equipment; others alter the feasible uses of existing resources. A method for making a stronger alloy, a reproducible algorithm for scheduling cargo, and a verified protocol for preventing contamination are examples of the latter. Each can be represented as a rule that connects conditions, actions, and outcomes. A rule can remain available after an application has taken place. That feature is the economic starting point of this monograph.

Nonrivalry is a property of simultaneous use: one person's use of an idea leaves the same idea available for another person's use. Excludability is the practical or legally supported ability to prevent access or use. A published mathematical identity is both nonrival and difficult to exclude from use. A confidential manufacturing instruction can be nonrival and strongly excludable. A laboratory instrument is rival when its capacity is fully occupied, whether it belongs to a company, a university, or the public sector. Distinguishing the properties of these objects prevents the substitution of ownership categories for technological characteristics.

Knowledge can be embodied in persons, equipment, organizational routines, or codified records. A scientist's time is rival even when the proposition communicated is nonrival. A reproducible design can be widely available while its operation requires training and constrained inputs. Tacit knowledge describes capabilities that are difficult to transmit through explicit instructions; it creates a transfer problem, not necessarily a depletion problem. A firm's confidential process may be hard to copy because the recipient lacks the corresponding skills or machinery. The resulting scarcity of implementation belongs to a different analytical margin from the scarcity of access.

The stock of ideas in a growth model is a theoretical aggregate. Its unit may be a count of designs, an index of attainable quality, or an efficiency shifter. Empirically these measures answer different questions. One thousand small improvements and one major discovery have no generally valid common conversion factor. A count of records can increase even when economically useful capabilities stagnate. Consequently an investigator must define the unit of knowledge before interpreting growth, accumulation, or depreciation. This dossier considers productive knowledge with a reproducible contribution to feasible production, while recognizing that much valuable knowledge has effects outside measured output.

The neoclassical benchmark

Solow's (1956) growth model supplies the benchmark for explaining why a stock of ordinary capital alone need not sustain per-worker growth. A familiar modernized Cobb-Douglas illustration uses output Y, capital K, labor L, labor-augmenting technology A, a savings fraction s, a depreciation rate d, and constant labor and technology growth rates n and g. This formulation is a teaching specialization of the neoclassical tradition rather than a transcription of the original equations.

Y=Kα(AL)1−α,0<α<1,K˙=sY−dK.Y=K^{\alpha}(AL)^{1-\alpha},\quad 0<\alpha<1,\qquad \dot K=sY-dK.

With k = K/(AL), output per effective worker is k raised to the power alpha. Differentiation gives

k˙=skα−(n+g+d)k.\dot k=sk^{\alpha}-(n+g+d)k.

For positive s and n+g+d, the positive steady state is

k∗=(sn+g+d)1/(1−α).k^*=\left(\frac{s}{n+g+d}\right)^{1/(1-\alpha)}.

At that state, Y/L grows at g. A permanent increase in s changes the steady-state level of capital and output per effective worker, and creates transition growth while k adjusts. It leaves the stipulated g unchanged. This is a conditional result: if technical progress is absent and other conditions hold, diminishing returns to reproducible physical capital terminate sustained growth of output per worker. The model isolates the variable whose explanation lies beyond the accumulation of machines.

For an illustrative economy with alpha = 1/3, s = 0.2 and combined effective depreciation of 0.06 per year, k-star is (0.2/0.06) raised to 1.5, approximately 6.09 in the model's normalized units. A change in s to 0.3 raises k-star to about 11.18. The ratio of steady-state output per effective worker is the cube root of 11.18/6.09, about 1.225. This calculation is an illustration, not an estimate of any country. It makes the distinction between a 22.5 percent level increase and a change in annual long-run growth concrete.

Endogenizing technology means supplying an economic explanation for the trajectory of A. It requires identifying which resources enter the production of ideas, why they are allocated there, and how existing ideas change the opportunities available to later research. Endogeneity in this sense is a property of model closure. It does not demand that every aspect of discovery, science, or demography be determined inside a single economic model.

Invention, incentives, and the economics of information

A reproducible instruction can have a costly first creation and inexpensive subsequent copying. The separation between discovery cost and replication cost changes the revenue problem. Let F denote an up-front discovery cost and c the marginal cost of providing a copy or service. Competitive pricing at c recovers variable expenses but leaves no contribution to F unless payment occurs through another channel. Subscription fees, bundled complements, licensing contracts, collective financing, grants, and prizes can establish different divisions between financing and access.

Arrow (1962) identifies appropriability, uncertainty, and indivisibility as central obstacles to efficient inventive investment. The analysis supplies a welfare-economic bridge from information characteristics to institutional arrangements. For example, potential customers must be able to evaluate a discovery before paying for it, yet complete disclosure can weaken the seller's ability to secure payment. Secrecy, certification, enforceable contracts, staged disclosure and complementary services can mitigate the transaction difficulty. Their effectiveness depends on what recipients need to learn and what legal claims the parties can enforce.

Suppose an invention has two potential users whose gross benefits are 70 and 50 units, with negligible dissemination cost and an invention cost of 90. The joint gain from invention is 30 units. If only the first user can be charged and the second receives the same idea without compensating its creator, the inventor cannot recover the cost even though social investment would be worthwhile. A feasible contract collecting 55 and 40 units finances discovery and leaves both users better off. The hypothetical arithmetic shows how incomplete appropriation can discourage efficient research; it also shows how exclusion can finance investment. It provides no universal case for a particular patent duration or subsidy.

Uncertainty makes investment depend on the distribution of outcomes, not their realized values. A portfolio of ten unsuccessful experiments and one successful experiment can be economical ex ante. Ex post accounting that assigns all costs to the success can misrepresent both risk and the opportunity cost of abandoned research. A social evaluation must include unsuccessful paths, dissemination benefits, and the alternative uses of the resources committed. The welfare comparison is therefore intertemporal, contingent, and institution-specific.

Nonrival designs and the production frontier

Romer's (1990) Endogenous Technological Change connects purposeful innovative investment to the economic organization of production. The treatment of designs as reusable inputs makes it possible for a producer's private cost and a society's aggregate productive capability to follow different scaling relationships. The example below isolates the replication argument through an illustrative variety technology.

Let A measure the number of available differentiated inputs, x(i) their physical quantity, L labor, and alpha a parameter strictly between zero and one:

Y=L1−α∫0Ax(i)α di.Y=L^{1-\alpha}\int_0^A x(i)^\alpha\,di.

Assume total intermediate input use X is fixed, units have equal resource costs, and symmetric quantities maximize output. Then x(i)=X/A and

Y=L1−αXαA1−α.Y=L^{1-\alpha}X^\alpha A^{1-\alpha}.

Holding A fixed, proportional scaling of L and X scales output proportionally. Doubling A with fixed L and X raises output by a factor of 2 to the power 1-alpha. The gain is created by access to more productive varieties under the specified aggregation technology. The example is not a claim that merely duplicating identical files, patents, or machine names creates output. It requires productively distinct inputs or methods and the stated curvature.

The private production frontier takes available designs as given. The aggregate frontier includes the possibility of research that creates additional designs. Paying each input its competitive marginal product under constant returns in rival factors exhausts corresponding output; paying a further positive scarcity rent to a freely available nonrival factor raises a financing difficulty. Many arrangements can resolve it, including government financing and private control of particular applications. The model's intellectual contribution is to place these questions in the same allocation problem as capital and labor.

If designs become freely usable everywhere, additional applications need complementary labor and materials but do not require recreating the designs in every region. The social value of discovery may thus extend across firms and borders, depending on actual use. The appropriate unit of duplication is central: production processes can be replicated, scientific knowledge can be transmitted, and the physical resources supporting both activities are scarce. This is why the geometry of a knowledge-intensive production frontier differs from the geometry of an economy composed entirely of rival reproducible factors.

Research allocation and balanced growth

A stylized version of an idea-production function connects the stock of usable designs A to the amount of effective research labor R. Let eta denote positive research efficiency:

A˙=ηRA,gA=A˙A=ηR.\dot A=\eta RA,\qquad g_A=\frac{\dot A}{A}=\eta R.

The equation says that existing ideas proportionally increase the rate at which a given research effort produces new ones. Holding R fixed and positive produces exponential growth. A larger constant R raises the proportional rate permanently. This is a strong assumption about research opportunities. The physical nonrivalry of designs alone cannot establish it: nonrival use is consistent with declining, constant, or increasing marginal effectiveness in generating new discoveries.

To see the allocation problem, divide a fixed effective labor endowment H between production H-Y and research H-R, with H-Y+H-R=H and R=H-R. Moving one unit into research sacrifices some capacity to produce current output and increases the flow of new ideas. The benefit depends on future application, appropriation, discounting and the effectiveness of research. A competitive researcher compares the expected private value of successful discoveries with wage and equipment costs. A social planner evaluates the discounted product and welfare effects on other users, as well as on the innovator.

A larger stock of designs increases productive opportunities in the variety illustration. At constant output elasticity gamma with respect to A, the contribution of idea growth to output growth equals gamma times g-A. It is generally incorrect to equate the growth of a count of designs with growth in GDP. The relation depends on aggregation, capital adjustment, and complementary factors. Any balanced-growth solution requires that expenditure, factor allocation, and asset valuation remain mutually consistent along the proposed trajectory.

The strong-scale form also gives a comparative prediction. Two otherwise identical integrated economies with twice as many researchers as one isolated economy would produce a higher idea growth rate if research productivity remained unchanged and the formula continued to hold. That prediction depends on the definition of an integrated research market and the absence of congestion, duplication, and diminishing opportunity. It is a testable property of the model rather than an immediate implication of the existence of knowledge.

Creative destruction and the value of innovation

Aghion and Howitt (1992) analyze growth through creative destruction in which improvements displace previous leaders. Their contribution is especially important for welfare analysis because the owner of an innovation can lose future rents when the next successful innovation arrives. The following elementary stochastic illustration clarifies the mechanism without reproducing the complete equilibrium.

Let each technological advance multiply a quality index Q by q>1. Let N(t) be a Poisson counting process with constant intensity lambda. Then Q(t)=Q(0)q^{N(t)}, and

E[log⁡Q(t)−log⁡Q(0)]=λtlog⁡q.E[\log Q(t)-\log Q(0)]=\lambda t\log q.

Expected logarithmic growth per unit time is lambda log q. By contrast, expected Q(t)/Q(0) is exp(lambda t(q-1)). The two objects are different because the logarithm is nonlinear. For q=1.1 and lambda=0.2 per year, expected log growth is approximately 0.01906 per year, or 1.906 percent; the level expectation grows exponentially at 0.02 per year under these assumptions. Neither is an observed economic statistic. The distinction matters when reporting average growth under uncertainty.

Consider an incumbent receiving a constant profit flow pi until the next discovery destroys its monopoly position. With constant discount rate r, hazard lambda, and no terminal payment, its expected discounted value is

V=∫0∞πe−(r+λ)t dt=πr+λ.V=\int_0^\infty \pi e^{-(r+\lambda)t}\,dt=\frac{\pi}{r+\lambda}.

Assumptions r+lambda>0 and constant hazard make the integral finite. Increasing the threat of displacement reduces V. At pi=100 per year, r=0.05 and lambda=0.15 per year, V=500. Raising lambda to 0.25 reduces V to about 333.33. In a real economy both profit and research intensity respond to competition; the exercise holds them apart to identify one channel.

An innovation can confer consumer gains, cost reductions and knowledge spillovers, while destroying the incumbent's revenue and potentially stranding specialized assets. A research race may devote scarce resources to the first arrival even when multiple research teams pursue nearly the same objective. Market selection may reward genuine quality improvements or primarily redistributive strategies. Therefore the mere observation of displacement does not quantify the social gain. The relevant comparison includes the quality of successive products, foregone alternatives, transition costs, and knowledge available to subsequent researchers.

Scale effects and semi-endogenous growth

Jones (1995) challenges strong scale effects on the basis of industrial-country time series. The general issue is whether higher levels of research effort support proportionately higher long-run growth rates. A transparent teaching equation generalizes the earlier idea-production relation:

A˙=ηRλAϕ,η>0,λ>0.\dot A=\eta R^{\lambda} A^{\phi},\quad \eta>0,\quad \lambda>0.

For phi<1, dividing by A gives g-A=eta R to the power lambda times A to the power phi-1. If R grows at a constant rate n-R, taking logarithmic derivatives gives

g˙AgA=λnR−(1−ϕ)gA.\frac{\dot g_A}{g_A}=\lambda n_R-(1-\phi)g_A.

On a positive balanced-growth trajectory the left-hand side is zero, implying g-A-star=lambda n-R/(1-phi). With stable research effort, positive growth generally decelerates toward zero under phi<1; the exact solution for constant R follows by integrating A to the power minus phi. A permanent increase in the level of R changes the transition and the long-run level relative to the previous path, but not the balanced-growth rate when n-R is unchanged. The latter statement requires the maintained parameters, positive initial knowledge and an exogenous or otherwise fixed long-run researcher growth rate.

For an illustrative n-R of 1 percent annually, lambda=1 and phi=0.5, balanced idea growth is 2 percent annually. If phi=0, it is 1 percent. If phi approaches one from below while n-R remains positive, the expression diverges. That mathematical sensitivity signals a singular boundary: the phi=1 case must be analyzed separately; it cannot be obtained by plugging into the denominator. For phi=1, g-A=eta R to the power lambda, so R growing without bound gives accelerating idea growth in this stylization. Resource and opportunity constraints can invalidate that extrapolation.

The production of ideas can exhibit both cumulative opportunities and exhaustion of accessible problems. A method learned yesterday can enable a discovery tomorrow, while the easiest remaining improvements may become scarce. The net exponent in a reduced-form equation combines several effects: what researchers count as a discovery, how they search, and which problems become economically relevant. Estimating phi without modeling changes in these margins can attach a structural interpretation to a changing statistical relationship.

Semi-endogenous growth shifts the policy question. A subsidy that permanently increases the fraction of workers engaged in research can improve the productivity level and transition path even when the eventual growth rate remains determined by researcher growth and research parameters. A reform that durably improves the research process itself may alter other parameters, but the existence and permanence of that change must be established empirically. The distinction between a level effect, a transitional growth effect and a permanent trend change is indispensable.

Research productivity and empirical discipline

Bloom, Jones, Van Reenen, and Webb (2020) examine research productivity across multiple domains. Their reported semiconductor example requires more than eighteen times the research effort of the early 1970s to maintain the same approximate rate of chip-density improvement in the comparison used in the study. That observation illustrates a deterioration in measured research productivity for a specified technological outcome during a specified historical sample. It is not a November 2026 update, a measure of all scientific discovery, or a causal estimate of the effect of employing additional scientists.

Define a research productivity measure as an outcome growth rate divided by a measure of effective R&D effort. Suppose the chosen outcome grows at 2 percent per year and the effort index increases from one to four. Output growth per effort-index unit falls from 0.02 to 0.005 per year, a fourfold reduction. This numerical identity does not establish why. Researchers may tackle harder targets, measured inputs may become more expensive, outcome quality may shift, or successful innovations may generate unmeasured services. If the research effort index is constructed from spending, care is needed to separate resource quantities from rising researcher compensation.

A second comparison illustrates the hazard of choosing the outcome metric. If two innovations increase computation speed by 10 percent and reduce energy per operation by 20 percent, measuring only speed records no contribution from the second achievement. A price or welfare index can incorporate both, but requires a defensible valuation of attributes. Patent counts, scientific articles, engineering performance and aggregate productivity growth are complementary indicators with distinct coverage.

An empirical test should specify the geographical boundary of the knowledge stock, the industrial perimeter of R&D, the time between expenditure and deployment, and the assumed depreciation of useful knowledge. Long-run estimates should report whether they rely on trends, differences, instruments, or structural restrictions. A stationary relationship derived from levels may fail if the series contain unit roots. A regression of growth on research spending can yield a strong correlation even if demand expectations cause both variables. The equation used in theory is not itself an identification strategy.

Evidence should be read at its actual resolution. The empirical study mentioned above is used here for its published headline result, not as a replication of the underlying datasets. A robust conclusion is that observed effort and performance can diverge markedly; the general magnitude and causal mechanism of that divergence require domain-specific investigation.

Diffusion, absorptive capacity, and international growth

An idea that has been created may enter a firm's production set only after verification, training, adaptation and investment. Cohen and Levinthal's (1990) absorptive capacity concept directs attention to prior knowledge as an enabling condition for using external research. An engineering team with experience in a technology can more readily evaluate a relevant publication and connect it to a production process. A recipient that lacks the necessary skills may face a substantial implementation project despite zero access price.

Let A-F be a frontier technology index and A-D a domestic adoption index, in common units. A descriptive catch-up equation is

A˙D=κ(AF−AD),κ>0.\dot A_D=\kappa(A_F-A_D),\qquad \kappa>0.

For a constant frontier, the gap closes exponentially at rate kappa. If the frontier grows at positive rate g-F, define z=A-D/A-F. Then

z˙=κ−(κ+gF)z,z∗=κκ+gF.\dot z=\kappa-(\kappa+g_F)z,\quad z^*=\frac{\kappa}{\kappa+g_F}.

For kappa=0.1 per year and g-F=0.02 per year, the long-run ratio is 5/6, approximately 0.833. The domestic index grows at the frontier rate in the steady ratio while remaining below the level of the frontier. This is an illustrative diffusion equation, not an estimate or newly proposed structural law. It makes precise why convergence of growth rates does not require equality of levels.

Keller's (2004) international technology diffusion survey places cross-border knowledge transmission within the explanation of productivity and income differences. Importing advanced capital goods, working with multinational firms, accessing technical publications, training abroad, or participating in international standards can enable transfer. The magnitude of each channel depends on industrial structure, trade relations, complementary investments and the nature of the knowledge.

Comparative research should distinguish invention at the global frontier, adaptation to local environments and diffusion across domestic firms. A small economy might achieve rapid productivity growth through adoption with modest frontier research expenditure. A large frontier economy can spend extensively on R&D while experiencing slow aggregate productivity growth during a period of costly reorganization. Ranking these economies by contemporaneous R&D-to-GDP ratios alone confuses different stages of development.

A licensing restriction can slow diffusion even when the knowledge is technologically reusable. Absence of restrictions can leave diffusion slow when infrastructure and skills are scarce. These counterfactuals identify different intervention targets. The first concerns access institutions; the second concerns capabilities. The potential returns from easier access rise when recipients can use it, and the benefits of capability investment rise when useful knowledge is available.

Spillovers, competition, and identification

A firm's innovation may improve the opportunities of firms in related technologies, reduce the value of rival products, and change demand for suppliers. A single distance measure based on industry membership can merge these effects. Bloom, Schankerman, and Van Reenen (2013) distinguish technology spillovers and product rivalry using US firm data and R&D tax-incentive changes. Their reported gross social return exceeds the private return substantially, by at least a factor of two in their study's framework. This is a specific estimate, not a universal return on public subsidies or a measure of net welfare gains in every industry.

A research design can define technological exposure as a weighted sum of other firms' research using a technology-proximity matrix. A separate product-market matrix captures competitive overlap. The weights need justification: patent classes, cited research, skill composition, product similarity and transaction networks each represent different relationships. If the matrices are nearly collinear, disentangling their effects becomes difficult regardless of sample size.

Suppose a new process increases the producer's value by 15 units, decreases competitors' value by 10, and increases purchasers' surplus by 25. The change in these three components is 30 units before other resource costs, under consistent nonoverlapping accounting. If a research cost of 20 has already been deducted in the producer value, subtracting it again would double-count the cost. If it has not been included, the net change is 10. The example emphasizes the boundary of the welfare ledger; firm valuation changes alone are incomplete.

A causal estimate must address anticipation, nonrandom research location, knowledge-source selection and simultaneous shocks. Tax-credit variation can provide useful variation in R&D costs, but a valid exclusion restriction requires defending how the policy affects outcomes through the posited research channel rather than through financing, employment, relocation or other contemporaneous changes. International comparisons face analogous problems when reform adoption correlates with expected growth.

Evidence of a positive spillover motivates assessment of its social reach and the cost of supporting the marginal project. The case for intervention depends on the change in research induced by the intervention, not the entire volume of research expenditure observed after adoption. It also depends on whether the additional budget displaces other projects of greater social value.

Firms, research portfolios, and aggregate change

Innovation is heterogeneous within firms as well as between firms. Akcigit and Kerr (2018) examine heterogeneous innovations through improvements to firms' existing products and innovations that expand their product portfolios. This distinction is useful because a firm can be large through the accumulation of many successful product lines, through the scale of each line, or through both.

Consider two stylized firms with equal research budgets. One spends its entire budget on reducing defects in a single production process, while another divides its budget among prototypes for new markets. The first may produce a sequence of small quality improvements with an immediate commercial payoff. The second may incur losses on unsuccessful prototypes and obtain value from one new line. Counting the number of projects incorrectly treats the two portfolios as equivalent. Counting only successful patents omits the experimental resources and organizational learning underlying them.

Capital markets affect which portfolios are feasible. Expected research rents may be difficult to pledge because they depend on uncertain future discoveries and personnel. An incumbent can support research from current earnings, whereas an entrant may require outside equity, contractual commitments, grants or staged financing. The asset's technological nonrivalry does not make the required funding unlimited. Financial frictions and economic knowledge interact through the organization of research.

An aggregate innovation indicator must also distinguish growth within continuing firms, market-share reallocation, firm entry and exit. Let aggregate labor productivity be the employment-share-weighted average of firm productivity for a sector. Its change reflects both changing firm productivity and changing weights. Entry and exit alter the population being measured. A stable average can conceal rapid improvements among leading firms when employment expands elsewhere; a rising average can reflect selection even without within-firm improvement. Such decompositions are accounting exercises whose interpretation requires causal evidence.

Research policy should be evaluated against its effect on the composition and quality of innovative activity. A program that induces more patent applications may finance genuinely additional inventions or mainly alter filing strategies. A procurement contract can foster useful production capability without increasing academic publication counts. The economic criterion is the incremental contribution to valued outcomes, together with the opportunity costs and distribution of gains.

Intellectual property and the financing of knowledge

The institution of exclusion can create a stream of income that finances research. An exclusive right can permit a price above the marginal cost of serving another user, thereby enabling recovery of risky discovery costs. Its social cost depends on the applications postponed or abandoned because of the price or the need for permission. The interaction is especially important when one protected idea becomes an essential input to a later invention.

Consider two sequential discoveries. The first costs 80 units and creates a private return of 90 when exclusivity is strong. The second costs 50 and would produce 100 units of gross value if it could use the first discovery. If a license fee of 60 is required, the second project's private net return before other effects is -10, so it will not proceed in this simple example. A fee of 30 makes its return 20. The first inventor's total expected receipts and cost recovery must be assessed jointly; the second developer's valuation and outside options also matter. These numbers are hypothetical, not an empirical patent assessment.

The comparison of institutions includes fixed licensing fees, per-use royalties, nonexclusive licensing, research exemptions where legally available, open licensing, time-limited rights, and collective purchasing arrangements. Each allocates risks, costs and authority differently. An exclusivity rule broad enough to cover unrelated improvements may give the original innovator bargaining power over a diverse set of downstream opportunities. A narrow rule can preserve those opportunities while making recovery of initial expenses more difficult. Legal scope, time horizon, enforcement and market structure are decisive.

Public grants and prizes can finance research without a price on every subsequent application. Their design depends on selecting relevant problems, recognizing achievement and evaluating results. A prize works most transparently when the desired outcome is observable and independently testable. Grants are better suited to exploratory projects whose results are uncertain and cannot be specified in advance. Both use scarce funds and expose allocation decisions to error. Private contracts, universities and public agencies can coexist in a research system; a meaningful welfare assessment compares feasible institutional alternatives, including their administrative costs.

A policy that maximizes the current rate of patenting may not maximize cumulative knowledge or welfare. The proper evaluation follows the sequence of creation, financing, access, adoption and further invention. A change in any stage can alter the effect of incentives at later stages. An exclusive right is a tool governing some of these transitions, rather than a direct measure of the economic value of the knowledge itself.

Measurement and the evidence ledger

The Frascati Manual 2015 (OECD, 2015) establishes an international framework for collecting and classifying research and development statistics. R&D expenditure describes resources devoted to specified activities; it does not directly count economically useful ideas. The Oslo Manual 2018 (OECD and Eurostat, 2018) addresses a broader measurement domain encompassing innovation activity and outcomes. The two measurement tasks are related but should be kept distinct.

An R&D expenditure series can be combined with an assumed knowledge-investment depreciation schedule to construct a stock. In a generic perpetual inventory calculation, existing stock S evolves as

St+1=(1−dS)St+It.S_{t+1}=(1-d_S)S_t+I_t.

Here I denotes real knowledge investment under the chosen accounting convention and d-S the assumed depreciation rate. A positive depreciation rate represents obsolescence, loss of relevant capability, or reduced expected usefulness, even when the historical record remains stored. An unchanged formula may retain truth indefinitely while the machinery for which it was designed disappears from use. Therefore the estimated productive stock depends on its economic purpose.

For a hypothetical S=100, d-S=0.1 and I=20, the next stock is 110. If the assumed depreciation is 0.2, it is 100. The example is a bookkeeping identity, not a measured stock of inventions or an estimate of depreciation. Empirical applications must explain initial-stock construction, deflators, the classification of inputs, treatment of unsuccessful projects, and whether expenditure performed abroad is included.

A growth-accounting residual is another indicator. For an explicitly specified aggregate production function Y=A K^alpha L^(1-alpha), logarithmic differentiation yields g-A=g-Y-alpha g-K-(1-alpha)g-L with fixed alpha. This A is Hicks-neutral productivity in this expression, unlike the labor-augmenting A in the earlier neoclassical illustration. The residual can absorb unobserved input quality, omitted organizational capital, changes in capacity use, measurement error, and departures from competitive assumptions. Interpreting the entire residual as discovery is unjustified.

Patent indicators capture a subset of inventive activities subject to legal and strategic selection. Citations and renewal decisions can help distinguish economic significance but also reflect institutional rules, time since filing, and field-specific citation practices. Scientific publication counts measure recorded research output within indexed domains. Engineering performance measures permit closer connection to specific technical changes, yet limited coverage makes cross-industry aggregation difficult.

A practical evidence ledger uses several indicators at once: research labor and expenditure, documented outputs, engineering improvements, realized adoption, sector productivity, and outcome measures relevant to users. For each it records units, jurisdiction, sector, period, vintage, missing values, and whether the causal claim depends on an identifying assumption. This monograph reports published literature findings and analytic examples; it does not present new official national estimates or independently replicated microdata results.

Testing alternative mechanisms

The main theories considered in this dossier imply distinguishable patterns only under their maintained assumptions. Strong scale effects imply a sustained positive relationship between the level of effective research effort and proportional idea growth. Semi-endogenous specifications instead relate the balanced-growth rate to the continuing growth of research inputs, given stable parameters. Creative-destruction mechanisms predict a link between displacement risk, innovative rents and research incentives. Diffusion models imply that growth may be rapid away from the frontier when absorptive capabilities permit a gap to close.

These mechanisms can operate simultaneously. A successful innovation may increase the technological opportunities of other researchers, increase demand for specialized employees, and reduce a rival's expected rent. A study that identifies only the total effect on patenting cannot infer which mechanism dominates. Researchers should specify observables that differ across alternatives and make a disciplined attempt to measure them. A rise in researcher employment alongside stable technical progress is informative about measured research productivity, but changes in research composition and targets must be investigated before treating it as a structural estimate.

An empirical program can begin with a panel of industries or countries that includes real research inputs, measures of technical outcomes and adoption, and suitable factor-quality measures. It can test whether frontier and follower observations exhibit different patterns and whether external knowledge exposure adds predictive information beyond domestic research. Cross-country comparisons need harmonization of research definitions, prices and population series. Industry studies can exploit engineering measures at higher resolution while acknowledging limited external validity.

A policy evaluation should define the treated margin, the counterfactual and the outcome horizon. An increase in research credits may raise reported expenditure without much change in the quantity of researchers if wages adjust. A change in ownership rights may improve the return to upstream discovery and reduce downstream experimentation. A training program may mainly improve adoption of existing technologies. The observed effect on output over two years can differ from the eventual effect on new knowledge and consumption.

Researchers can challenge causal interpretations through pretrend checks, placebo outcomes, alternative exposure measures, bounds for unobserved confounding, and sensitivity to lags and sample selection. For structural work, units and equilibrium restrictions must be explicit. A successful fit to aggregate historical growth does not establish that a model reproduces the firm's innovative decisions, and a successful fit to firm data does not establish its aggregate counterfactual. Both dimensions should be examined when the substantive conclusion spans microeconomic incentives and macroeconomic growth.

Author's analytical synthesis

The literature establishes that knowledge has economic effects at several linked margins. The contribution of this dossier is a comparative interpretation of those margins and the consequences of confusing them. It develops existing economic reasoning rather than introducing a named theory or asserting a novel discovery.

The first analytical result is the separation of reusable knowledge from accessible knowledge and from productive application. A public idea can remain economically dormant for want of complementary capability. A private idea can generate widespread production effects through licensing and transfer. A heavily funded scientific program can produce valuable research without measurable near-term output growth. These outcomes become intelligible when the investigator follows the distinct steps of discovery, access and implementation instead of treating research spending as an immediate increment in GDP.

The second result is the conditional character of scale effects. Nonrivalry creates a potential for multiplication through reuse. Exponential growth of a knowledge stock requires a further assumption about the effectiveness of existing ideas in producing new ones. A greater number of researchers can raise the flow of discovery without permanently raising its proportional growth rate. This distinction reconciles the technological feature at the center of the Romer tradition with empirical evidence of declining research productivity; the latter challenges particular production-function assumptions rather than the reusability of ideas.

The third result concerns ownership and welfare. The revenue necessary to finance discovery and the access required for broad diffusion can interact at every stage of cumulative innovation. A monopoly value captures some of the expected private gain from a discovery and may change as subsequent innovators arrive. Social value extends to buyers, knowledge recipients and future investigators, while accounting for displaced capabilities, duplicated expenditure and opportunity cost. The evaluation of intellectual property or research subsidies must therefore be performed at the margin of the institution being considered and under the actual distribution of technological opportunities.

The fourth result concerns the appropriate geographical level of explanation. In a connected world, the source of a country's growth and the location of the research that generated its productive potential can differ. Domestic innovation indicators and global knowledge indicators describe different phenomena. Catch-up can raise the growth rate of a follower for an extended interval without requiring that it become the original source of each technology. Investment in absorptive capacity can increase the domestic value of both domestic and external research.

The fifth result is a measurement discipline. Inputs, intermediate records, technological capabilities and realized welfare are connected but noninterchangeable. Research intensity is a useful resource measure; patenting is a selected output measure; engineering advances are technology-specific performance measures; productivity is an aggregate outcome affected by many forces. A credible argument about endogenous growth should establish what changes when each indicator changes and make its assertion at the level actually measured.

Together these results support a coherent interpretation of growth as an interaction between incentives for discovery, cumulative opportunities, productive use, and institutional selection. No single coefficient in the illustrative equations can describe all research systems. The appropriate application of an endogenous-growth model follows from the question asked, the mechanisms included, and the empirical conditions under which those mechanisms can be distinguished.

Limits and open scientific problems

A central empirical difficulty is the construction of a usable measure of the knowledge frontier. A scientific result may be important because it removes a false hypothesis, opens a previously inaccessible research path, or provides a capability whose immediate market demand is small. Such contributions are difficult to value using counts of patents or current revenue. The challenge increases when combining ideas across disciplines: the usefulness of a design can depend on several other results arriving in a particular sequence.

The interpretation of reduced-form research productivity also remains open. A falling ratio of outcome growth to research input can reflect harder scientific problems, new safety requirements, defensive duplication, shifts toward other performance dimensions, resource prices, or the movement of effort into new domains. These mechanisms imply different interventions. Distinguishing them requires better links between project-level inputs and outcomes, with unsuccessful and discontinued efforts represented alongside successes.

The value of new computational tools in research provides a live methodological question for future empirical study. Faster search, simulation, experimental design and documentation may reduce the resource requirements of some investigative tasks. Whether they permanently raise economy-wide trend growth depends on the breadth of effective adoption, bottlenecks in physical experimentation, the reliability of generated results, and the allocation of human and material complements. The possibility of acceleration is analytically relevant, but this dossier makes no numerical prediction about the present or future effect of artificial intelligence.

Distributional analysis presents another unresolved issue. Nonrival benefits can coexist with concentrated ownership, geographically clustered high-skill employment and unequal access to costly complements. Average output growth reveals little about who receives the gains or bears the costs of creative destruction. Integrating firm-level ownership, labor-market adjustment and international diffusion into credible quantitative analysis remains an important research objective.

Finally, the research corpus used here has unequal levels of access: several foundational papers were inspected in original-text reproductions, while a number of later studies and statistical manuals were checked through authoritative abstracts and bibliographic records. The arguments attributed to these sources are restricted accordingly. The numerical examples are constructed for explanation; no dataset was downloaded or independently re-estimated. A more comprehensive empirical investigation would need harmonized microdata, explicit code and reproducible estimates. These boundaries constrain the kind of scientific claim made in this work.

Conclusion

Knowledge nonrivalry makes cumulative production possible through the reuse of existing ideas across applications. Endogenous growth theory explains the economic allocation of scarce resources to the expansion of such capabilities. Its key explanatory challenge is to connect the financing of discovery to research effectiveness, institutional access, diffusion, and the realized productive use of knowledge.

The neoclassical benchmark isolates the importance of technical progress for sustained per-worker growth. Models of intentionally produced designs and stochastic quality improvements explain how economic incentives can generate technical change. Their differing assumptions about the research process, competitive displacement and spillovers determine the implications for long-run growth and welfare. The semi-endogenous response clarifies the conditions under which rising research effort yields level improvements and transition growth instead of permanently higher proportional growth.

The empirical literature supports careful distinctions among research inputs, innovation outcomes and aggregate productivity, and shows the importance of separating knowledge transmission from product-market rivalry. The source-specific results inform empirical investigation without providing universal policy coefficients. International diffusion further demonstrates that the location of invention and the beneficiaries of innovation need not coincide.

The analytical synthesis identifies the separate roles of nonrival availability, institutional access and scarce implementation. This separation gives precision to the economic analysis of research support, competition, intellectual property and the accumulation of capabilities. Knowledge can expand productive possibilities repeatedly; whether societies realize those possibilities depends on research allocation, the quality of institutions, availability of complements, and the rate at which opportunities become usable. These are the questions that connect theories of ideas to the observed history of economic growth.

References

  • Aghion, Philippe, and Peter Howitt (1992). A Model of Growth through Creative Destruction. Econometrica 60(2): 323-351. DOI: 10.2307/2951599. Source

  • Akcigit, Ufuk, and William R. Kerr (2018). Growth through Heterogeneous Innovations. Journal of Political Economy 126(4): 1374-1443. DOI: 10.1086/697901. Source

  • Arrow, Kenneth J. (1962). Economic Welfare and the Allocation of Resources for Invention. In The Rate and Direction of Inventive Activity: Economic and Social Factors, 609-626. Princeton University Press for NBER. Source

  • Bloom, Nicholas, Charles I. Jones, John Van Reenen, and Michael Webb (2020). Are Ideas Getting Harder to Find? American Economic Review 110(4): 1104-1144. DOI: 10.1257/aer.20180338. Source

  • Bloom, Nicholas, Mark Schankerman, and John Van Reenen (2013). Identifying Technology Spillovers and Product Market Rivalry. Econometrica 81(4): 1347-1393. DOI: 10.3982/ECTA9466. Source

  • Cohen, Wesley M., and Daniel A. Levinthal (1990). Absorptive Capacity: A New Perspective on Learning and Innovation. Administrative Science Quarterly 35(1): 128-152. DOI: 10.2307/2393553. Source

  • Jones, Charles I. (1995). R & D-Based Models of Economic Growth. Journal of Political Economy 103(4): 759-784. DOI: 10.1086/262002. Source

  • Keller, Wolfgang (2004). International Technology Diffusion. Journal of Economic Literature 42(3): 752-782. DOI: 10.1257/0022051042177685. Source

  • OECD (2015). Frascati Manual 2015: Guidelines for Collecting and Reporting Data on Research and Experimental Development. Paris: OECD Publishing. DOI: 10.1787/9789264239012-en. Source

  • OECD and Eurostat (2018). Oslo Manual 2018: Guidelines for Collecting, Reporting and Using Data on Innovation. 4th ed. Paris/Luxembourg: OECD Publishing/Eurostat. DOI: 10.1787/9789264304604-en. Source

  • Romer, Paul M. (1990). Endogenous Technological Change. Journal of Political Economy 98(5, Part 2): S71-S102. DOI: 10.1086/261725. Source

  • Solow, Robert M. (1956). A Contribution to the Theory of Economic Growth. Quarterly Journal of Economics 70(1): 65-94. DOI: 10.2307/1884513. Source