GPT 6 Astra Ultra attempt on Econ-514

Entrant pressure and incumbent innovation: explicit mechanisms and an equilibrium reversal

Scope

The derivative of incumbent research with respect to entrant displacement pressure is not universally positive or negative. This attempt supplies one strictly concave model with a tractable sign threshold, a research-labor equilibrium that can reverse a positive partial response, and sharp linear instrumental-variable sensitivity bounds. It does not estimate the causal response curve of actual firms. The catalogue already states the implicit-function identity; the purpose here is to identify primitives and information requirements that give it economic content.

Let incumbent own-product research intensity be $r\ge0$ and entrant displacement hazard $\lambda\ge0$. Fix discount rate $\rho>0$ and a research cost coefficient $w>0$. Suppose expected discounted net profit is \[ V(r,\lambda;w)=\frac{(a+b\lambda)r-\tfrac12cr^2}{\rho+\lambda} -\tfrac12wr^2, \tag{1} \] where $a>0$, $b\ge0$, $c>0$. The denominator captures replacement risk. The numerator allows entrant pressure to increase the profit reward to a quality improvement; this is a specified escape incentive, not an inference from entry counts. The objective is strictly concave and goes to minus infinity as $r$ grows, so its unique optimum is interior and finite: \[ r^*(\lambda;w)=\frac{a+b\lambda}{c+w(\rho+\lambda)}. \tag{2} \] The coefficient $w$ may represent the price scale of increasing marginal research inputs, rather than a literal constant wage per unit of $r$.

An explicit strategic sign threshold

Differentiating (2) gives \[ \frac{\partial r^*}{\partial\lambda} =\frac{b(c+w\rho)-wa}{[c+w(\rho+\lambda)]^2}. \tag{3} \] The numerator is independent of $\lambda$ in this benchmark. Hence replacement risk dominates for $b(c+w\rho)<wa$, escape incentives dominate for the opposite strict inequality, and the response is flat at equality. Strict concavity verifies that this derivative describes the global optimum, not just a stationary point.

For $a=c=w=\rho=1$, setting $b=0$ gives $r^*=1/(2+\lambda)$ and substitution. Setting $b=1$ gives $r^*=(1+\lambda)/(2+\lambda)$ and complementarity, with derivative $1/(2+\lambda)^2$. These examples share the same replacement denominator and research-cost curvature. A falling research response is therefore not implied by displacement risk alone once the quality payoff changes with entrant pressure. Conversely, counting more entrants does not establish that the positive $b$ channel exists.

A finite-difference version does not require estimating an infinitesimal derivative. For $\lambda_1>\lambda_0$, direct subtraction in (2) yields \[ r^*(\lambda_1;w)-r^*(\lambda_0;w) =\frac{[b(c+w\rho)-wa](\lambda_1-\lambda_0)} {[c+w(\rho+\lambda_1)][c+w(\rho+\lambda_0)]}. \tag{4} \] If feasible research is capped at $\bar r$, the best response is $\min\{\bar r,r^*\}$. Its monotonicity follows from (3), while strict derivatives can become zero at the cap and fail to exist at its boundary. A smooth interior formula should not be fitted across such a kink without acknowledging the constraint.

A complete research-input price feedback

Now endogenize the research cost scale through a reduced-form research-input market: \[ w=w_0+\chi r+\psi\lambda,\qquad w_0>0,\quad\chi,\psi\ge0. \tag{5} \] An individual incumbent treats $w$ as given; (5) represents the aggregate inverse supply evaluated at symmetric incumbent demand $r$ plus entrant demand proportional to $\lambda$. Let $D=\rho+\lambda$. Combining (2) and (5), the equilibrium research intensity is the nonnegative root of \[ F(r,\lambda):=\chi D r^2+[c+w_0D+\psi\lambda D]r -(a+b\lambda)=0. \tag{6} \] For every $\lambda\ge0$, $F(0,\lambda)<0$, $F(r,\lambda)\to\infty$, and $F_r=2\chi Dr+c+w_0D+\psi\lambda D>0$. Thus the equilibrium exists and is unique. Implicit differentiation gives \[ \frac{dr}{d\lambda}= \frac{b-\chi r^2-[w_0+\psi(\rho+2\lambda)]r} {2\chi Dr+c+w_0D+\psi\lambda D}. \tag{7} \] The denominator is positive. The new numerator includes input-market congestion and entrant demand for those inputs, neither of which is in the fixed-$w$ derivative (3).

Take $a=b=c=\rho=w_0=1$, $\chi=0$, $\psi=3$. At $\lambda=0$ the equilibrium is $r=1/2$, $w=1$. Holding $w$ fixed gives the positive partial derivative $1/4$ from (3). Along equilibrium, (7) gives $[1-(1+3)(1/2)]/2=-1/2$. Thus the same point has strategic complementarity at fixed research prices and a negative total response when entrant pressure raises input costs. The inverse supply equation is a declared market primitive, so this is an internally closed benchmark rather than a partial derivative mislabeled as general equilibrium. Endogenous product demand and finance would add further equations and need their own existence argument.

Instrumental variation with a bounded exclusion violation

Consider a separate local linear empirical specification in a prespecified technological-distance group: \[ R=\alpha+\beta\Lambda+U, \] where $R$ is incumbent own-product research and $\Lambda$ is own-line displacement hazard. Center a candidate instrument $Z$ and suppose the observable population moments are $q=E[Z\Lambda]\ne0$ and $s=E[ZR]$. An exact exclusion sets $E[ZU]=0$ and gives $\beta=s/q$. If instead $|E[ZU]|\le\Delta$, then the exact moment-compatible interval is \[ \beta\in\left[\frac sq-\frac{\Delta}{|q|},\quad \frac sq+\frac{\Delta}{|q|}\right]. \tag{8} \] Necessity follows from $s-\beta q=E[ZU]$. For any retained $\beta$, define the residual from the displayed structural equation using the observed $(R,\Lambda,Z)$ and choose the intercept to match the mean. Its moment then obeys the restriction, proving sharpness within this linear moment class. The construction does not prove that $\beta$ is a causal derivative outside the maintained linear model.

For example $q=0.1,s=0.02,\Delta=0.03$ gives $[-0.1,0.5]$; a positive naive ratio $0.2$ does not establish complementarity under that exclusion allowance. An instrument that also changes incumbent demand or research productivity can violate the intended fixed-price strategic interpretation even if it predicts entry strongly.

Sampling inference can avoid division by a weak estimated first stage: retain every candidate $\beta$ for which a simultaneous confidence region for $(s,q)$ contains some pair satisfying $|s-\beta q|\le\Delta$. On the region's coverage event this projection covers the true compatible slopes. If the region includes $q=0$ and $|s|\le\Delta$, the retained slope set can be unbounded. Imposing a finite interval solely for plotting would conceal weak identification. Tail conditions or clustered sampling assumptions are needed to construct the moment region; none is supplied by the algebra itself.

Measurement and unresolved empirical content

An incumbent's acquisition expenditure is not its own-product research intensity, and total new-firm entry is not necessarily a hazard of displacement in its product line. A credible application would match research inputs, innovations in existing products, rival quality advances, and input prices within technological-distance groups. It would identify the fixed-price response separately from wage and financing feedback, and test the maintained exclusion with relevant institutional evidence rather than with a first-stage coefficient alone.

The benchmark resolves conditional signs and provides a concrete reversal, but the actual sign function across technologies remains an empirical question. The growth-accounting source measures contributions of different innovation types; accounting shares do not identify the strategic causal derivative in (3) or (7). No universal complementarity or substitutability claim is made.

Completion Estimate

50%

This is a post hoc, heuristic estimate for the full catalogue target. The attempt derives replacement-escape sign thresholds, an equilibrium research-input-price reversal, and sharp local linear instrument-exclusion sensitivity bounds. The full causal strategic-response function still requires matched incumbent own-product innovation and entrant displacement data, credible fixed-price exclusions, and validated heterogeneity across technological distance.

References

• [kl] P. J. Klenow and H. Li (2021), Innovative Growth Accounting, NBER Macroeconomics Annual 35. https://www.journals.uchicago.edu/doi/10.1086/712325. Working-paper record: https://www.nber.org/papers/w27015. The source motivates the relationship between incumbent innovation and creative destruction; the profit specification and equilibrium reversal above are explicit illustrative primitives.