Full citation

Aghion, P., & Howitt, P. (1992). A model of growth through creative destruction. Econometrica, 60(2), 323–351.^1

Summary

The paper presents a tractable model of endogenous growth in which productivity increases stem from a sequence of “vertical” product innovations that improve the quality of an intermediate good used in final output. Each successful innovation gives its inventor a temporary monopoly but simultaneously destroys the rents of the previous innovator, formalising Schumpeter’s idea of growth through creative destruction. The model treats the time between innovations as random, with a Poisson arrival rate determined by the amount of research effort, and assumes a representative household with linear utility, three types of labour, and a monopoly intermediate sector.^1

The authors ask how equilibrium research effort and growth emerge when forward-looking research firms choose R&D given the threat that future innovations will erode their monopoly rents. They derive a forward-looking difference equation in which current research depends negatively on expected future research because more future R&D both accelerates creative destruction and bids up skilled wages, reducing the value of current innovation. They characterise stationary equilibria where research employment is constant and output follows a random walk with positive drift, show how parameters such as innovation size, research productivity, skilled-labour endowment, and market power affect the average growth rate and its variance, and discuss the possibility of cyclical equilibria and “no-growth traps”.^1

Normatively, the paper compares the decentralised equilibrium with a social planner’s allocation and identifies four key distortions: intertemporal knowledge spillovers, imperfect appropriability of innovation benefits, business-stealing, and monopoly distortions in labour allocation. The first two push laissez-faire growth below the social optimum, while the business-stealing and monopoly-distortion effects can make research and growth excessive, so the laissez-faire growth rate can be either too low or too high. Extensions show that when innovators can choose the size as well as the arrival rate of innovations, laissez-faire leads to innovations that are too small (a form of under-investment in quality), and the results are generalised to nondrastic innovations, strategic monopsony behaviour by monopolists, and models with multiple intermediate sectors.^1

Key topics by page

Quotes

"This paper examines a channel that has received little attention in the endogenous growth literature, namely that of industrial innovations which improve the quality of products." (p. 323)^1

"The intertemporal-spillover and appropriability effects tend to make the laissez-faire average growth rate less than optimal, whereas the business-stealing and monopoly-distortion effects tend to make it greater than optimal." (p. 338)^1

Key insights

• On the core mechanism of growth: The model treats growth as a sequence of vertical innovations that improve the productivity of an intermediate input by a constant factor $\gamma > 1$, so each innovation permanently raises the level of output but only temporarily yields monopoly rents until displaced by the next innovation. Time between innovations follows a Poisson process with arrival rate increasing in research employment, making the expected growth rate proportional to both the innovation size and equilibrium research effort.^1
• On equilibrium research and dynamics: Equilibrium research employment is pinned down by a forward-looking condition equating the marginal cost of research today to the discounted marginal benefit of winning the next monopoly, which itself depends on expected future research. Higher expected future R&D both raises the future skilled wage and increases the hazard of creative destruction, so current R&D is a decreasing function of expected future R&D, generating a unique stationary equilibrium and, under some parameter values, cyclical equilibria and “no-growth traps” where research periodically collapses to zero.^1
• On determinants of average growth and volatility: In stationary equilibrium, real output grows as a random walk with drift: the average growth rate equals the innovation arrival rate times $\ln \gamma$, and the variance of growth is proportional to the same arrival rate. Both the mean and variance of growth rise with the size of innovations, the productivity of research (the Poisson parameter), the size of the skilled labour force, and the degree of monopoly power, and fall with the representative agent’s rate of time preference.^1
• On welfare and distortion channels: Comparing decentralised equilibrium to a social planner’s allocation reveals four distortions: (i) intertemporal knowledge spillovers (innovators do not internalise the effect of their innovations on future research productivity), (ii) imperfect appropriability (innovators capture only monopoly profits, not the full social surplus), (iii) business-stealing (innovators ignore the destruction of incumbents’ rents), and (iv) monopoly-induced labour misallocation. The spillover and appropriability effects push equilibrium research below the optimum, while business-stealing and monopoly distortions push it above, so laissez-faire growth can be either too low or too high depending on parameters such as the size of innovations and the degree of market power.^1
• On the size of innovations and business-stealing: When research firms can choose both the frequency and size of innovations, private firms maximise the product of the arrival rate and the gross size of innovations, whereas a planner maximises the arrival rate times the net benefit ($\gamma - 1$), leading to equilibrium innovations that are too small in welfare terms. This under-sizing of innovations is another manifestation of business-stealing: firms ignore the negative externality that making a bigger step up the quality ladder more completely wipes out the previous vintage’s rents.^1
• On structural extensions and robustness: Allowing innovations to be nondrastic, introducing many intermediate sectors, or letting research productivity vary across states leaves the core comparative statics largely intact but shows richer implications, such as lower aggregate volatility via cross-sector averaging and the possibility that improving research productivity in one state can reduce average growth by raising the expected intensity of creative destruction. When a monopolist recognises its influence on the skilled wage and research effort (strategic monopsony), it may choose higher employment to slow future creative destruction, altering the balance between business-stealing and underinvestment but not eliminating the fundamental welfare ambiguities.^1