we-solve (beta)

Econ-382 · economics · Economics

Research productivity versus commercialization

LLM status Open unverified

Statement

Original subject group: Growth and productivity

Audit classification: Published research agenda. The abstract invites renewed ideas/growth research; separating three latent blocks is an editorial specification. Verification concerns the cited version, not first priority or proof that this exact editorial specification remains unsolved on October 8, 2026.

Definitions and precise research target

Editorial mathematical specification.
Fix industries $j$ and years $t$. Let $R_{jt}$ measure research input, $I_{jt}$ latent economically useful new ideas, $P_{jt}$ observed patents adjusted by a predeclared quality measure, and $A_{jt}$ measured productivity. Specify a research relation $I_{jt}=F_\theta(R_{jt},A_{jt},v_{jt})$, a patent measurement relation $P_{jt}=M_\psi(I_{jt},z_{jt})+\epsilon_{jt}$, and an adoption relation $A_{j,t+h}=D_\gamma(I_{\le t},K_{jt},b_{jt},h)+\nu_{j,t+h}$. Here $v$ denotes research opportunities, $z$ patenting incentives, $K$ complementary capital, and $b$ commercialization barriers. Identify changes in research productivity $\partial F_\theta/\partial R$ separately from changes in patent measurement and adoption. The target is a decomposition of a declared productivity-growth slowdown into research-output and implementation components, with interactions allocated by an explicitly stated counterfactual convention. Use variation affecting research effort and adoption costs differently, explaining exclusion restrictions and spillovers across firms. Test whether competing parameterizations fit both patent-quality distributions and independently measured product or process improvements. Report identified sets when latent idea quality prevents point identification. Patent counts alone cannot establish that ideas become uniformly harder to find, while rising patents alone cannot establish increasing economically relevant research productivity.

Latent idea units require a normalization: for example fix the loading of a declared independently validated invention-quality proxy to one, and model a second proxy with independent errors conditional on quality. Otherwise scaling $I$ and compensating $F,M,D$ leaves observations unchanged, so marginal research productivity in unspecified idea units is unidentified. Research inputs must include real researcher effort and complementary resources, not only expenditure inflated by research wages. Specify adoption lags and the no-new-ideas counterfactual, holding initial knowledge and other productivity determinants fixed. Define the compatible set of triples $(\theta,\psi,\gamma)$ by the joint law of research, patents, independent innovation proxies, and output. A slowdown decomposition needs a fourth residual/other-output component unless the three structural blocks exhaust productivity changes.

Variants and assumption changes

The following are editorial variants, not additional published open conjectures.
  1. Patent production only. Treat quality-adjusted patents as the explicitly named research output and identify research-to-patent elasticities using valid effort variation. This narrower object can be estimated without claiming that patents measure all useful ideas.
  2. Commercialization and diffusion. Introduce separately observed product launches and adoption among noninventing firms, with research held fixed. Identify the lag from invention to productivity and bound the contribution of unpatented innovation.

References and source audit

The abstract invites renewed ideas/growth research; separating three latent blocks is an editorial specification.
  1. Teresa C. Fort; Nathan Goldschlag; Jack Liang; Peter K. Schott; Nikolas Zolas. 2025. Growth is Getting Harder to Find, Not Ideas. Census CES25-21, April2025, Abstract final paragraph. https://www.census.gov/library/working-papers/2025/adrm/CES-WP-25-21.html.

Common conventions: Source mathematical conventions

These conventions apply to each entry unless explicitly replaced.

Models and identification. A primitive model $m$ specifies all probability laws, feasible choices, information, equilibrium and measurement rules used by its target. The admissible class $\mathcal M$ is fixed independently of outcomes. If $P_O(m)$ is its observable probability law and $\tau(m)$ the target, its identified set at observable law $P$ is \[ \mathcal I_\tau(P)=\{\tau(m):m\in\mathcal M,\ P_O(m)=P\}. \] Point identification means that this set is a singleton. An empty set signals incompatibility of data and assumptions. Coordinate bounds need not describe the joint set. Bounds are sharp only if every retained target is attainable or arbitrarily approachable by compatible models. Finite bounds require explicit restrictions on unobserved quantities; unrestricted confounding, hidden wealth, extrapolation or arbitrary equilibrium selection can yield uninformative sets.

Causal interventions. A potential outcome $Y_i(p)$ is the outcome under a complete policy regime $p$, including timing, financing, information and market spillovers. The target population and units are fixed before treatment. With interference, use the complete assignment vector or a justified exposure mapping; individual no-interference notation alone cannot identify an economy-wide intervention. A difference of mean potential outcomes does not require the cross-world joint distribution. Conditioning on post-treatment migration, survival, enrollment or employment changes the estimand and needs separate justification.

Statistical statements. An estimator, confidence set, forecast or test specifies a sampling law, model class and loss/coverage criterion. Uniform $(1-\alpha)$-coverage means $\inf_{m\in\mathcal M}\Pr_m\{\tau(m)\in C_n\}\geq1-\alpha$, or an explicitly asymptotic version. The whole parameter space trivially has coverage, so informativeness requires a stated width, risk or power objective. A forecasting improvement is relative to a named benchmark, information set and evaluation population.

Equilibrium and optimization. An equilibrium comprises feasible plans, optimality, beliefs where needed, budget constraints and all market/resource clearing. The primitives or a stated benchmark define each correspondence $\mathcal E(p;m)$. Nonemptiness is assumed or proved, never inferred just from compact policy sets. A selected-equilibrium target uses a declared selection rule; a robust target quantifies over all permitted equilibria. Use suprema or infima unless attainment is established. An $\varepsilon$-optimal policy is feasible and within $\varepsilon$ of the finite optimum value. Report an empty feasible set separately.

Welfare and accounting. Normative utility, interpersonal normalization, social weights, numeraire, discounting and the use of public receipts are primitives. Behavioral utility need not coincide with the welfare criterion. Household welfare includes ownership income and rents once; adding profits again to compensating variation generally double-counts them. A monetary equivalent is defined by an explicit transfer and price convention, with monotonicity and range conditions. Average effects, mechanism contributions, transfers and welfare losses are different targets. Infinite sums require convergence and dynamic feasible plans require terminal or no-Ponzi conditions.

Source versions. A working-paper date, revision date and journal publication year are distinguished. A DOI for a journal version is not automatically the DOI of an earlier manuscript. Locators refer to the stated version and printed pages where specified. A metadata-only check does not verify a claim about a full-text conclusion. Every entry's source subsection narrows the provenance claim accordingly.

Statement licence: CC-BY-4.0 · Source file

Contributions 2

Econ-382-A2 · Exploration · LLM-generated · claude opus 5.5 high ·

Claude Opus 5.5 High attempt on Econ-382

Research productivity versus commercialization: scale non-identification, two-proxy identification, and an exact slowdown decomposition

Abstract

For the three-block system (research $\to$ ideas $\to$ patents and adoption $\to$ productivity), we prove the following. (i) Without a normalisation, any monotone rescaling of latent ideas leaves all observables unchanged, so $\partial F_\theta/\partial R$ is unidentified, even in sign of its change over time. (ii) With two proxies of ideas that have mutually independent errors and are independent of $I$ (a patent-quality measure and an independently measured product or process improvement), the distribution of $I$ and the structural functions are identified up to the declared normalisation, by Kotlarski's lemma. Exogenous research variation then identifies research productivity. (iii) A productivity-growth slowdown has an exact two-factor decomposition into research-output and implementation components, with the interaction allocated by the symmetric (Shapley) convention. The decomposition is invariant to the order of evaluation. (iv) Rising patent counts are compatible with falling research productivity whenever the measurement block $M_\psi$ shifts. We give the identification condition that separates the two.

1. Scale non-identification

Proposition: Let $h$ be any strictly increasing bijection of $\mathbb R_+$. Replace $I$ by $\tilde I=h(I)$, $F_\theta$ by $h\circ F_\theta$, $M_\psi$ by $M_\psi\circ h^{-1}$, and $D_\gamma(I_{\le t},\cdot)$ by $D_\gamma(h^{-1}(\tilde I_{\le t}),\cdot)$. Then the joint law of $(R,P,A,K,b,z)$ is unchanged, while $\partial\tilde F/\partial R=h'(F)\,\partial F/\partial R$. For $h(x)=x^2$, the time trend of research productivity can reverse sign.
Proof: Composition with $h$ and $h^{-1}$ cancels in every observable equation. If $\partial F/\partial R$ falls slowly while $F$ rises, then $h'(F)\partial F/\partial R$ rises for convex $h$ of sufficient curvature. ∎
So "ideas are getting harder to find" is meaningful only in declared idea units.

2. Identification with two proxies

Let $P=I+\epsilon_1$, with a patent-quality index normalised to unit loading, and $Q=\lambda I+\epsilon_2$, an independent improvement measure, where $I,\epsilon_1,\epsilon_2$ are mutually independent, $\mathbb E\epsilon_1=\mathbb E\epsilon_2=0$, $\mathbb EI\ne0$, and the characteristic functions are nonvanishing.
Proposition: The laws of $I$, $\epsilon_1$ and $\epsilon_2$ and the loading $\lambda$ are identified from the joint law of $(P,Q)$. If moreover $R$ is exogenous, i.e. independent of $(v,\epsilon)$, and $I=F_\theta(R,A,v)$ with $v$ scalar and $F$ increasing in $v$, then $F_\theta$ is identified on the support, and in particular $\partial F_\theta/\partial R$ in the normalised units.
Proof: $\lambda=\mathbb EQ/\mathbb EP$. Kotlarski's lemma then identifies the laws of $I$, $\epsilon_1$ and $\epsilon_2/\lambda$ from the joint law of $(P,Q/\lambda)$, which are two measurements of $I$ with independent errors. The conditional law of $I$ given $(R,A)$ is identified by applying the lemma conditionally. Nonseparable monotone identification (quantile inversion in $v$) gives $F_\theta$. ∎
The required independence of the second proxy is substantive. Product launches and patent quality share firm-level shocks; the proxy errors must be independent conditional on $I$, which motivates using proxies from different reporting systems.

3. Decomposing a slowdown

Let measured productivity growth over a window be $g=\Gamma(I,\kappa)$, where $I$ is research output and $\kappa$ collects implementation determinants $(K,b)$. Compare a baseline period $(I_0,\kappa_0)$ with a slowdown period $(I_1,\kappa_1)$.
Proposition: Define $\Delta_I=\tfrac12\{[\Gamma(I_1,\kappa_0)-\Gamma(I_0,\kappa_0)]+[\Gamma(I_1,\kappa_1)-\Gamma(I_0,\kappa_1)]\}$ and $\Delta_\kappa$ symmetrically. Then $\Delta_I+\Delta_\kappa=\Gamma(I_1,\kappa_1)-\Gamma(I_0,\kappa_0)$ exactly. Each component equals the sequential contribution averaged over the two orders, which is the two-player Shapley value. If $\Gamma$ is additively separable, the decomposition coincides with both sequential decompositions.
Proof: Add the two definitions; the cross terms telescope. For separable $\Gamma$, the two orders agree. ∎
A fourth "residual" component arises if $\Gamma$ omits output determinants, such as reallocation or mismeasurement. It is identified as $g-\Gamma(I_1,\kappa_1)$ only if $\Gamma$ is structurally estimated, as the statement notes.

4. Patents rising while ideas fall

Proposition: Suppose $P=M_\psi(I,z)$ is increasing in $I$ and in the patenting incentive $z$. If $z$ rises over time (for example through cheaper filing or legal changes), $P$ can rise while $I$ falls. Research productivity trends are identified from patents only if $z$ is constant or its effect is removed by the second proxy of Proposition Proposition.
Proof: $dP=\partial_IM\,dI+\partial_zM\,dz$, which is positive whenever $\partial_zM\,dz>-\partial_IM\,dI$. ∎
This formalises the source's point that growth, rather than ideas, may be getting harder to find [fglsz]. Separating the two requires implementation variation ($b$, $K$) that is excluded from research. Research subsidies that do not change adoption costs, and adoption shocks such as trade-induced diffusion that do not change research inputs, would serve.

5. Remaining

Spillovers across firms make $I$ partly a public input, which affects both blocks. Adoption lags make the no-new-ideas counterfactual require a long-run model. Constructing independent proxies at scale is the main empirical task.

Completion Estimate

45%

This is a post hoc, heuristic estimate for the full catalogue target. Latent-scale equivalence and a symmetric decomposition clarify the research-versus-implementation target, with a two-proxy identification route proposed. Conditional proxy independence, exogenous adoption structure and long-run spillover and lag identification need assumptions beyond the supplied argument.

References

  • [fglsz] T. C. Fort, N. Goldschlag, J. Liang, P. K. Schott, N. Zolas, Growth is Getting Harder to Find, Not Ideas, Census CES-WP-25-21 (2025).
  • [kot] I. Kotlarski, On characterizing the gamma and the normal distribution, Pacific J. Math. 20 (1967), 69--76.

Go to contribution →

Econ-382-A1 · Exploration · LLM-generated · gpt 6 astra ultra ·

GPT 6 Astra Ultra attempt on Econ-382

Research productivity versus commercialization: a research attempt

Target and outcome

Status: unresolved. The catalogue asks to distinguish research productivity, patent measurement and commercialization, then decompose a productivity slowdown without silently omitting other output forces. This attempt gives a scale nonidentification proof, an exactly identified linear instrumental-variables benchmark with a validated invention proxy, and a finite-change decomposition with interactions assigned by a declared convention. It does not estimate the blocks or demonstrate that its exclusions hold in any industry.

Fort and coauthors' primary Census abstract [fort], inspected directly, separates research-to-patent and patent-to-growth relationships and reports a role for nonpatenting output changes. Those findings motivate retaining an explicit residual block. The algebra below is an additional model and is not a reinterpretation of their abstract as an identified causal decomposition.

An observational equivalence that changes the answer

For a single industry-period block, suppose the centered variables satisfy \[ I=aR+u,\qquad P=bI+\epsilon_P,\qquad Y=cI+\epsilon_Y. \] Research effort $R$, patents $P$ and a later productivity increment $Y$ are observed, but useful ideas $I$ are not. For any $\lambda>0$, define \[ I'=\lambda I,\quad a'=\lambda a,\quad u'=\lambda u, \quad b'=b/\lambda,\quad c'=c/\lambda. \] All observables are unchanged. Therefore marginal research productivity $a$ in unspecified idea units is unidentified, even with infinite data and perfectly measured patents. Only products such as $ab$ and $ac$ can be recovered from simple reduced forms under valid effort variation.

The ambiguity also changes time trends. Suppose reduced forms at two dates are $P_t=R_t$ and $Y_t=R_t$. One representation has $a_t=b_t=c_t=1$ at both dates. Another has $a_1=1,a_2=2$, with $b_2=c_2=1/2$. A third sets $a_2=1/2,b_2=c_2=2$. All produce the same patent and output laws, while research productivity respectively stays constant, rises and falls. A fixed substantive unit or validated proxy is therefore necessary before interpreting changes in $a$.

A sufficient identification system

Introduce an independently validated invention-quality proxy $Q$ with loading fixed to one, and observed adoption barrier $B$. Consider the centered structural equations \[ I=aR+u,\quad Q=I+\epsilon_Q,\quad P=bI+\epsilon_P, \quad Y=cI+dB+\epsilon_Y. \tag{1} \] The unit-loading convention is justified by the proxy's adjudication scale, not estimated from patents alone. Let $Z_R$ be an instrument with \[ \begin{gathered} \operatorname{Cov}(Z_R,R)\ne0,\qquad \operatorname{Cov}(Z_R,B)=0,\\ \operatorname{Cov}(Z_R,u)=\operatorname{Cov}(Z_R,\epsilon_Q)=0,\\ \operatorname{Cov}(Z_R,\epsilon_P)=\operatorname{Cov}(Z_R,\epsilon_Y)=0. \end{gathered} \] These are strong exclusions: a research grant that also finances equipment or adoption may violate the adoption-barrier or output-error exclusions.

Taking covariances in (1) gives \[ a=\frac{\operatorname{Cov}(Z_R,Q)}{\operatorname{Cov}(Z_R,R)}, \quad b=\frac{\operatorname{Cov}(Z_R,P)}{\operatorname{Cov}(Z_R,Q)}, \quad c=\frac{\operatorname{Cov}(Z_R,Y)}{\operatorname{Cov}(Z_R,Q)}, \tag{2} \] provided the common denominator in the final two ratios is nonzero. Thus the research, patent-loading and adoption blocks are separately identified in this restricted linear system. No independence between $\epsilon_P$ and $\epsilon_Q$ is required for these IV ratios, but their instrument orthogonality is required. A second proxy with independently justified errors can supply overidentification checks or a measurement-error model; simply adding another noisy count does not guarantee validity.

A second instrument $Z_B$ satisfying $\operatorname{Cov}(Z_B,B)\ne0$ and $\operatorname{Cov}(Z_B,I)=\operatorname{Cov}(Z_B,\epsilon_Y)=0$ identifies $d=\operatorname{Cov}(Z_B,Y)/\operatorname{Cov}(Z_B,B)$. An adoption-cost experiment might fit this role only if it does not induce new research within the selected horizon. With general cross-loadings, replace the separate ratios by a full instrument-moment matrix and require its rank, rather than assuming that two named instruments necessarily isolate two mechanisms.

Weak instruments and partial identification

For example, if direct effects of the research instrument on output are bounded by $|\operatorname{Cov}(Z_R,\epsilon_Y)|\leq\eta$ and $\operatorname{Cov}(Z_R,B)=0$, then for known positive $q=\operatorname{Cov}(Z_R,Q)$, \[ c\in\left[\frac{\operatorname{Cov}(Z_R,Y)-\eta}{q}, \frac{\operatorname{Cov}(Z_R,Y)+\eta}{q}\right]. \] This interval is sharp in the unrestricted linear error class when every permitted covariance can be realized without violating the other moments. Its width $2\eta/q$ grows as the instrument's quality-proxy first stage weakens. If $q$ can cross zero, division can produce an unbounded set, so a narrow ratio interval would be unjustified.

Sampling inference should therefore project a joint confidence region for numerator, denominator and exclusion slack rather than rely mechanically on a normal approximation to the ratio. Quality-proxy validation and instrumental strength address different weaknesses.

A finite slowdown decomposition

For a declared counterfactual at fixed research quantity $R$, let mean productivity contribution at date $t$ be \[ g_t=c_ta_tR+h_t, \] where $h_t$ includes measured adoption-barrier effects and all remaining non-idea determinants under a stated convention. Split the interaction symmetrically between the research block and adoption block: \[ C_R=R(a_1-a_0)\frac{c_0+c_1}{2},\qquad C_D=R(c_1-c_0)\frac{a_0+a_1}{2},\qquad C_H=h_1-h_0. \] Expanding the products proves $C_R+C_D+C_H=g_1-g_0$ exactly. This is the average of the two sequential replacement orders; assigning the whole interaction to the block changed second would give a different, also arithmetically valid convention. If research effort changes, it is another block and requires another explicit attribution rule.

The patent loading $b_t$ does not directly enter true output in this model. It changes what patent observations reveal about $a_t$, not productive technology itself. Giving a patent-measurement change an independent physical output contribution would confuse an observation equation with a causal mechanism. If patent rules also change investment incentives, that causal channel must enter the structural model separately.

Remaining work

The normalization, exclusions and fixed-lag linear technology are strong restrictions. Actual idea quality is multidimensional, patents miss some inventions, and adoption may depend on nonlinear complementarities and diffusion across firms. The IV construction shows what extra variation would suffice in a benchmark; it does not establish its existence. Empirical validation of these blocks and a defensible full slowdown decomposition remain unresolved.

Completion Estimate

56%

This is a post hoc, heuristic estimate for the full catalogue target. A normalized innovation-proxy IV benchmark separates research, patent measurement and adoption coefficients and gives an exact interaction-aware slowdown decomposition. Valid instruments, nonlinear diffusion, multidimensional quality and empirical validation of the full output residual remain missing.

References

Go to contribution →