1 s2.0 S1094202522000266 Main

JID:YREDY AID:1136 /FLA [m3G; v1.316] P.
1 (1-24)
Review of Economic Dynamics ••• (••••) •••–•••
Contents lists available at ScienceDirect
Review of Economic Dynamics

www.elsevier.com/locate/red
Product innovation, diffusion and endogenous growth

Michael A. Klein a,∗ , Fuat Şener b
a
Rensselaer Polytechnic Institute, Troy, NY, USA
b
Union College, Schenectady, NY, USA
a r t i c l e i n f o a b s t r a c t
Article history: We develop a model of Schumpeterian growth featuring a stochastic diffusion process
Received 28 June 2021 where the rate of commercial success of product innovations is endogenously determined
Received in revised form 20 April 2022 by advertising intensity. We consider both informative advertising, which young techno-
Available online xxxx
logical leaders use to increase the probability of diffusion, and defensive advertising, which
JEL classification:
incumbents use to prevent the diffusion of competing products. Economic growth depends
O31 positively on the arrival rate of product innovations and the diffusion rate of innovations
O33 into the mainstream market. We show that R&D subsidies shift relative investment in-
M30 centives towards innovation and away from diffusion. This creates an inverted U-shaped
relationship between R&D subsidies and both economic growth and welfare as innovations
Keywords: arrive more frequently, but fewer commercialize successfully. We find that advertising sub-
Innovation sidies increase diffusion, growth, and welfare when advertising is purely informative. In the
Diffusion presence of defensive advertising, advertising subsidies lead to socially wasteful increases
Commercialization in resources devoted to advertising without large increases in diffusion, reducing growth
Advertising
and welfare.
Marketing
R&D subsidies
© 2022 Elsevier Inc. All rights reserved.
“There’s two big frustrations of being an inventor. The first is when you can’t solve a problem ... The second one [is
when] you can’t get the world to adopt it.”
- Nathan Myhrvold1
1. Introduction
Schumpeterian endogenous growth models routinely assume that successful innovators costlessly and instantaneously
capture their entire potential market share. This implies that all consumers immediately adopt newly improved products
and that incumbent firms are fully displaced as soon as innovation occurs. Although analytically convenient, this assump-
tion is clearly counterfactual. Research on the marketing and diffusion of innovations has long emphasized that technical
and functional superiority does not guarantee a new product’s commercial success.2 Overall, empirical estimates suggest
*
Corresponding author.
E-mail addresses: kleinm5@rpi.edu (M.A. Klein), senerm@union.edu (F. Şener).
1
Former Chief Technology Officer at Microsoft and co-founder of the firm Intellectual Ventures, speaking on the podcast “People I (Mostly) Admire” in
2020.
2
In his foundational textbook treatment, Rogers (1962) notes, “many technologists think that advantageous innovations will sell themselves, that the
obvious benefits of a new idea will be widely realized by potential adopters, and that the innovation will therefore diffuse rapidly. Unfortunately, this is
very seldom the case.”
https://doi.org/10.1016/j.red.2022.05.001
1094-2025/© 2022 Elsevier Inc. All rights reserved.
JID:YREDY AID:1136 /FLA [m3G; v1.316] P.2 (1-24)
M.A. Klein and F. Şener Review of Economic Dynamics ••• (••••) •••–•••
that 40-50% of new product launches fail within their first four years.3 Even in cases where commercialization is ultimately
successful, it is typically a slow, uncertain process. Most successful new products experience an initial period of low pene-
tration and slow growth followed eventually by a sharp sales increase or “takeoff” to its market share as a mature product
(Agarwal and Bayus, 2002; Golder and Tellis, 2004). Furthermore, “the time to sales takeoff can vary considerably across
product innovations; some quickly achieve sales takeoff after commercialization, whereas others languish for years with low
sales” (Agarwal and Bayus, 2002).
To be sure, firms expend substantial resources to build and maintain market share. Marketing expenditures as a whole
are estimated to comprise as much as 8% of GDP, with advertising alone accounting for over 2% (Gourio and Rudanko, 2014;
Cavenaile and Roldan-Blanco, 2021). Especially when launching a new product, advertising is often necessary to establish
product awareness and inform consumers of a new product’s advantages (Goeree, 2008; Eliaz and Spiegler, 2011). New
product advertising also plays a fundamental role in persuading hesitant consumers to change their status quo consumption
behavior by weakening incumbent brand loyalty and reducing perceived switching costs (Shum, 2004; Gourville, 2006;
Bagwell, 2007). As argued by Yohn (2019), “innovation alone may be enough to initiate the adoption life cycle, but marketing
remains the bridge necessary to cross the chasm between early adopters to the wider group of people who will form a
viable, valuable customer base.”
In this paper, we develop a novel theoretical framework to examine the dynamic interaction between product innovation,
commercialization, and economic growth. As in standard Schumpeterian quality ladder models, entrepreneurial firms invest
in R&D to innovate higher quality products across a fixed (measure one) set of industries and new innovations arrive
according to a stochastic Poisson process. However, we introduce an endogenous commercialization process that takes place
in two phases. In the first phase, successful innovators instantaneously capture a small share of the market comprised of
consumers who immediately recognize the superiority of innovative products. Borrowing marketing terminology, we refer
to this subset of consumers as early adopters. In the second phase, innovators must invest in costly advertising in order to
convince the remaining mainstream consumers to recognize their innovative product’s quality advantage.
To capture the uncertain nature of sales takeoff, we model this diffusion of innovations into the mainstream market as
a stochastic Poisson process whose arrival rate depends upon advertising intensity. We consider two distinct formulations
for the relationship between advertising and the probability of diffusion. First, we assume advertising is purely informative;
only young innovators invest in advertising to expand their market share by communicating their product’s advantages to
potential consumers. Second, we allow for advertising to be combative; incumbent firms also invest in defensive advertising
to protect their existing market share against new entrants. In this formulation, the rate of new product diffusion depends
upon the advertising contest between young and old firms endogenously battling for consumers through advertising expen-
diture.4
In our framework, firms endogenously cycle through distinct life stages of stochastic length as new innovations arrive
and either commercialize successfully or fail. Each new innovator begins life as a young technology leader, serves only early
adopters, and invests in advertising to increase its chances of diffusing its product into the mainstream. If a subsequent
innovation arrives in the industry before a young firm captures the mainstream market, its product fails. If the young firm
instead successfully diffuses into the mainstream prior to the arrival of the next competing innovation, it fully replaces
the existing incumbent and begins its tenure as an adult technological leader. Once the next innovation occurs, the now
incumbent adult firm transitions to its final stage where it is no longer the technological leader, and early adopters abandon
the old product for the newest iteration. However, these old firms still retain a sizable market share until they are fully
displaced by the next young firm that diffuses its product successfully. In our combative advertising formulation, old firms
also endogenously invest in defensive advertising to protect their market share and prolong their final stage of life.
As in traditional models, economic growth is driven by the incorporation of higher quality products into households’
consumption bundles. This implies that the growth rate depends positively on both the rate of innovation and the rate of
product diffusion since only innovations that commercialize successfully are adopted by mainstream consumers. We show
that this relationship provides novel insights into the role of R&D subsidies in promoting economic growth. Unlike the
traditional models in which R&D subsidies always promote growth, we find that R&D subsidies can have a non-monotonic
effect on growth. This is because young firms’ incentives to invest in advertising in order to diffuse their product depend
upon the expected length of their reign of market dominance as an adult firm. The more frequently new innovations arrive,
the faster technology leaders transition to their old firm stage, and the smaller the incentive to invest in advertising. Thus,
although the traditional growth-promoting effect of R&D subsidies of stimulating innovation is present in our model, we
3
New product failure is defined in the empirical literature as either a total removal from the market or a sufficiently large underperformance relative to a
pre-specified sales target. See Asplund and Sandin (1999), Castellion and Markham (2013) and Victory et al. (2021) for a summary of empirical research on
new product failure. See Gourville (2006) and Chiesa and Frattini (2011) for an extensive discussion of products that incorporated innovative technologies
yet failed to achieve commercial success. For example, Gourville (2006) notes that “although TiVo’s digital video recorder (DVR) has garnered rave reviews
since the late 1990s from both industry experts and product adopters, the company had amassed $600 million in operating losses by 2005 because demand
trailed expectations.” Indeed, while the DVR technology that TiVo pioneered is now standard, TiVo’s innovative products never enjoyed mainstream success.
4
This combative advertising formulation reflects empirical evidence that “advertising is often characterized over time by reciprocal cancellation ... new
entrants advertise to gain market share and thereby induce increased advertising by incumbents ... in order to limit the sales of new entrants” (Bagwell,
2007).
2
identify a novel, competing growth-reducing effect as a smaller proportion of innovations commercialize successfully. We
show that this fundamental relationship holds in both the informative and combative advertising versions of the model.
Using numerical simulations, we find that R&D subsidies exhibit an inverted U-shaped relationship with both economic
growth and welfare. In our benchmark case of a 25% product failure rate in the initial equilibrium, an R&D subsidy rate
of 15.5% maximizes growth and a subsidy rate of 9.2% maximizes welfare. Furthermore, as the initial product failure rate
increases (i.e. less frequent successful diffusion in the baseline equilibrium), the case for R&D subsidies becomes weaker.
Specifically, the scope for using R&D subsidies to improve welfare diminishes with the failure rate because the social benefit
of stimulating innovation is lower when a smaller proportion of new innovations succeed. Indeed, we find that the optimal
R&D policy shifts to a tax when the initial failure rate is high but still within an empirically plausible range. Hence, our
results suggest that standard endogenous growth models that assume instantaneous innovation diffusion may overstate the
case for large R&D subsidies.5
Finally, we use the model to examine the economic impact of advertising policy. When advertising is informative, adver-
tising subsidies lead to faster product diffusion, which increases economic growth and welfare. However, when advertising
is combative, advertising subsidies also stimulate defensive advertising by existing incumbents. Since advertising is char-
acterized by reciprocal cancellation in this formulation, the primary effect is a socially wasteful increase in the resources
devoted to advertising, without the dynamic benefit of a substantial increase in product diffusion. In this case, economic
growth and welfare both fall when advertising is subsidized. Thus, our analysis suggests that the welfare impact of adver-
tising policy depends critically on whether advertising is informative or combative; a finding that echoes conclusions from
the long-standing literature on the economic effects of advertising (Butters, 1977; Dixit and Norman, 1978; Stegeman, 1991;
Grossman and Shapiro, 1984; Dinlersoz and Yorukoglu, 2012; Cavenaile et al., 2021). We show that this pattern of results
remains present in a fully dynamic setting where advertising and R&D investment decisions interact and together influence
the process of economic growth.
1.1. Related literature
Fundamentally, our paper belongs to the expansive literature that builds on the canonical Schumpeterian growth frame-
work developed by Segerstrom et al. (1990), Grossman and Helpman (1991) and Aghion and Howitt (1992). We present
the standard components of the Schumpeterian growth model closely following later refinements by Segerstrom (1998) and
Dinopoulos and Segerstrom (1999).6 Our main departure from the standard model is to introduce an endogenous diffusion
process that is driven by the deliberate advertising efforts of successful innovators.
Our paper is related to several strands of the Schumpeterian growth literature including analyses that incorporate non-
instantaneous new product diffusion, defensive behavior of market incumbents, and intangible advertising investment. Both
Dinopoulos and Waldo (2005) and Dinopoulos et al. (2021) develop Schumpeterian growth models in which innovator
market share evolves over time. Dinopoulos and Waldo (2005) impose an exogenous diffusion process that mimics a sales
takeoff based S-curve. They use the model to explore the relationship between gradual product diffusion and the dynamics
of asset prices, but the relationship between R&D and advertising incentives is absent by construction. In several respects,
our methodological approach is most similar to Dinopoulos et al. (2021), who allow for endogenous changes to innovator
market share over a firm life cycle. However, they focus on employment frictions as young firms expand and use the model
to examine the relationship between unemployment and economic growth.
Similarly, Chu and Furukawa (2013) and Cozzi and Galli (2014) develop models featuring a two stage innovation process
that requires both basic and applied research to bring a product to market. These papers analyze the role of patent policy in
shaping incentives through the profit division between distinct basic and applied research firms. In contrast, we focus on the
relationship between separate innovation and commercialization stages of product development within a single firm. Several
papers including Dinopoulos and Syropoulos (2007), Davis and Şener (2012), and Klein (2020) have analyzed the defensive
actions, or rent protection activities, of market incumbents in a Schumpeterian growth framework. In all cases, these papers
consider defensive actions that increase the effective cost of rival innovation, such as patent infringement litigation. By
incorporating defensive advertising, our model advances a distinct form of rent protection activities that targets competitors
after they have entered the market.
To our knowledge, only Grossmann (2008), Cavenaile and Roldan-Blanco (2021), and Cavenaile et al. (2021) have con-
sidered the interactions between advertising and innovation in a Schumpeterian growth framework. In these models,
advertising acts as a demand shifter that increases consumers’ perceived quality of existing incumbent products with an
established market position. While advertising impacts innovator profits, and therefore R&D incentives, in these models, it
does not impact the dynamics of innovation diffusion into the mainstream market by construction. We contribute to this ex-
isting work by focusing on the distinct role of advertising investment in establishing brand awareness, acquiring customers
5
Examples of Schumpeterian analyses that make the case for optimal R&D subsidies include Segerstrom (2007), Şener (2008), Impullitti (2010), and
Minniti et al. (2013).
6
See Dinopoulos and Sener (2007) and Aghion et al. (2014) for overviews of more recent developments in Schumpeterian growth theory.
3
and building market share. Within this framework, we analyze the interaction between the resulting endogenous diffusion
process, R&D incentives and economic growth.7
In this sense, our work follows a stream of recent research that examines the role of customer acquisition in firm dy-
namics (Arkolakis, 2010; Gourio and Rudanko, 2014; Perla, 2019; Einav et al., 2021; Rachel, 2021). We follow this literature
in modeling advertising expenditure as fundamental to the process of building market share; a relationship that has re-
cently been corroborated in firm-level econometric analyses (Fitzgerald et al., 2016; Argente et al., 2021). We contribute to
this literature by embedding customer acquisition into a Schumpeterian growth framework in a way that is consistent with
core findings from research on the marketing and diffusion of innovations. Specifically, our model incorporates heterogenous
preferences towards adopting new products within a population, a sales takeoff based diffusion process and the effect of
advertising investment in determining the commercial success and failure of innovative new products.
In our framework, investment in R&D and advertising advance firms through distinct life stages and work in tandem to
increase profits. Nonetheless, our model is consistent with empirical evidence that R&D and advertising act as substitutes at
the firm level (Cavenaile and Roldan-Blanco, 2021; Cavenaile et al., 2021). Specifically, we find that R&D subsidies stimulate
innovation but reduce advertising investment by young innovative firms. Furthermore, since this directly implies that a
smaller share of innovations commercialize successfully, our model is also consistent with recent evidence that suggests
R&D subsidies are associated with a lower average market return of new patents and products. For example, Svensson
(2013) finds that firms that receive subsidized R&D loans have a significantly lower renewal rate of new patents. Similarly,
Czarnitzki et al. (2011) find that R&D subsidies increase the average number of new products introduced by firms, but do
not improve general firm performance indicators such as profitability or market share. Indeed, as a possible explanation of
their findings, Czarnitzki et al. (2011) suggest that “the reduced cost of R&D funds may shift firms’ allocation of funding
for innovation activities away from necessary complementary activities such as marketing.” Our analysis formalizes this
intuition.
The remainder of this paper is organized as follows. In Section 2, we develop the informative advertising version of the
model. We explore the welfare properties of the model in Section 3. In Section 4, we examine the impact of R&D subsidies
on economic growth and welfare. In Section 5, we discuss the calibration of the model and analyze optimal R&D policies
numerically. Section 6 develops the combative advertising extension of the model. We investigate the impact of advertising
policy under both advertising specifications in Section 7. Section 8 concludes.
2. The model
2.1. Households and perceived quality
The economy is populated by a unit continuum of households indexed by i ∈ [0, 1]. Each household i maximizes dis-
counted utility
∞
Ui = e −ρ t ln(u i (t ))dt , (2.1)
0
where ρ > 0 is the subjective discount rate. Household i’s sub-utility at time t is defined as
1
ln(u i (t )) = ln q̃i (k, ω, t ) zi (k, ω, t ) dω, (2.2)
0 k
where zi (k, ω, t ) denotes household i’s quantity consumed of a product that has experienced k successful innovations in
industry ω ∈ [0, 1] at time t, and q̃i (k, ω, t ) denotes household i’s perceived quality of the associated product. Each household
inelastically supplies one unit of labor and maximizes (2.1) by allocating individual consumption expenditure c i (t ) given
prices at time t.8 Adjusted for perceived quality, products within each industry are perfect substitutes and each household
optimally purchases only the product with the lowest perceived quality adjusted price. Products enter utility symmetrically,
so households evenly spread consumption expenditure across industries. Establishing notation, household i’s demand for
the good with the lowest perceived quality adjusted price in a typical industry is
c i (t )
zi (t ) = , (2.3)
p i (t )
7
As argued by Agarwal and Bayus (2002), incorporating the “timing and causes of sales takeoff is critically important ... because they have serious short-
and long-term resource implications for research and development, product development, marketing, and manufacturing.”
8
Since there is no population growth in the model, the aggregate supply of labor is fixed at L (t ) = 1. As in Chu and Cozzi (2018), Yang (2018) and Chu
et al. (2019), fixing the supply of labor allows us to avoid the issue of scale-effects that is present in the Schumpeterian growth framework. We discuss
how population growth can be incorporated into our model in Section S.4.1 of our supplementary material.
4
where p i (t ) is the market price of the good for which perceived quality adjusted price is lowest for consumer i. Maximizing
(2.1) subject to the standard intertemporal budget constraint yields
ċ i (t )
= r (t ) − ρ , (2.4)
c i (t )
where r (t ) is the market interest rate.
In each industry ω and time t, there exists a single firm that produces the current state of the art product that all house-
holds perceive to be of quality q̃(k, ω, t ) = λk(ω,t ) , where λ > 1 is the constant step size of the innovation quality ladder.
Since all households share this quality perception, this product represents the definitive quality standard at level “k.” Similar
to traditional models, we assume that challenger firms within each industry may produce products using the previous k − 1
quality standard, which all consumers perceive to be one step down the λ quality ladder. Challengers invest resources in
R&D to innovate new versions of the product that represent possible candidates or prototypes for the k + 1 quality standard.
Each k + 1 prototype eventually either succeeds or fails. A prototype succeeds if it establishes itself as the definitive k + 1
quality standard by convincing all consumers to perceive it to be of quality λk(ω,t )+1 . A prototype fails if all consumers
ultimately reject its attempted quality improvement.
There exist two types of households that are differentiated by how they perceive the quality of k + 1 prototypes. Follow-
ing the traditional consumer classifications in marketing, we assume that a constant proportion of households, φ ∈ (0, 1),
are early adopters who immediately consider the latest prototype to be superior to all other existing options. Specifically,
these consumers exhibit novelty seeking preferences; they always perceive the most recently developed k + 1 prototype to
be a λ size quality improvement over the current k level quality standard, but view all previously offered k + 1 prototypes
to be of quality level k or lower.9 In contrast, the remaining 1 − φ proportion of households are mainstream consumers who
initially do not consider prototypes to be viable alternatives to the current k level quality standard. Instead, mainstream
consumers continue to perceive the standard product as quality superior to all other available options until they are per-
suaded, through endogenous advertising efforts detailed in the following section, that a particular prototype constitutes an
actual quality improvement. In other words, the introduction of a new prototype does not initially affect the current incum-
bent’s perceived quality lead in the eyes of mainstream consumers. Broadly, this assumption encompasses well-established
consumer characteristics such as (1) imperfect consumer awareness of new products and their attributes (Eliaz and Spiegler,
2011; Perla, 2019), and (2) consumer inertia for reasons other than tangible product attributes such as initial skepticism of
new products, brand loyalty and reluctance to change consumption behavior (Gourville, 2006; Bornstein et al., 2018).
2.2. Innovation, diffusion, and industry structure
The potential for different perceived quality among consumer types implies that industries and firms endogenously cycle
through distinct stages as new prototypes are innovated and either succeed or fail to diffuse into the mainstream market. To
see this, consider the evolution of a typical industry up the quality ladder from the kth to the kth + 1 step. Since industries
are structurally identical, we omit the ω index to avoid clutter. The industry achieves the kth step on the quality ladder
when a firm successfully establishes its prototype as the definitive k quality standard. Until the first k + 1 prototype arrives,
this firm enjoys a λ size perceived quality advantage over all other firms that can only produce inferior k − 1 level products.
Standard limit pricing implies that this technology leader captures the industry’s entire market share. We refer to such a
firm as an adult firm and all such industries served by an adult firm as an A industry.
Challenger firms invest in R&D to innovate new k + 1 prototypes. As is standard, we model innovation as a stochas-
tic Poisson process that depends on the intensity of R&D investment by challengers within each industry. Specifically, a
challenger j that employs L I , j (t ) units of labor in R&D innovates a new k + 1 prototype with instantaneous probability
I j (t ) = L I , j (t )/κ I , where κ I > 0 is a parameter that captures the difficulty of innovation. That is, I j (t )dt is the probabil-
ity that challenger j innovates by time t + dt conditional on not having innovated by time t, where dt is an infinitesimal
interval of time. The industry wide instantaneous probability of innovative success at time t is determined by total R&D
employment within the industry. We define10
L I (t )
I (t ) ≡ , where L I (t ) ≡ L I , j (t ). (2.5)
κI j
9
Within marketing, consumers with this type of novelty seeking preference structure are categorized as “innovative” consumers, which are defined by
their “predisposition to buy new and different products and brands rather than remain with previous choices and consumption patterns” (Steenkamp et
al., 1999). Empirical evidence shows that “novelty seeking plays an essential role in the early stages of consumer adoption of new products” (Tellis et al.,
2009). To avoid confusion with the concept of new product innovation, we choose to label these consumers under the broader related term, early adopters.
See also Furukawa et al. (2019, 2020) for recent analyses of the role of novelty seeking preferences in creating demand for innovative products.
10
As in Segerstrom (1998) and Dinopoulos and Segerstrom (1999), the presence of constant returns to scale in R&D employment implies that the number
of challenger firms that conduct R&D within an industry is indeterminate. However, the industry wide instantaneous probability of innovative success is
well defined in terms of total R&D labor employed within an industry.
5
As soon as the first k + 1 prototype is innovated, the φ ∈ (0, 1) measure of early adopters immediately perceive the
prototype to be a λ size quality improvement over the k quality standard. Under limit pricing, the innovative, or young,
firm’s prototype captures this portion of the market and partially displaces the incumbent adult firm’s product. This forces
the incumbent into its old firm stage where it is no longer the undisputed technology leader. However, since the old firm
initially maintains its perceived quality lead with mainstream consumers, it continues to serve this 1 − φ portion of the
market as before. We label all industries where a young and old firm are both active as a B industry.
In B industries, each young firm either succeeds or fails to establish its prototype as the definitive k + 1 quality rung by
persuading mainstream consumers that its prototype offers λ quality improvement over the k quality standard. If successful,
the prototype enjoys a λ perceived quality advantage with all consumers. With limit pricing, the young firm captures the
entire industry’s market share, fully displaces the previous incumbent’s k quality standard product, and takes its place as an
adult technology leader. Note that this implies that the industry transitions back to an A type industry at quality level k + 1,
and the process begins anew with the search for k + 2 prototypes.
To capture the notion of uncertain new product sales takeoff, we model this prototype diffusion into the mainstream
market as a stochastic Poisson process that depends upon advertising intensity. Specifically, a young firm in a typical B
industry that employs L δ, y (t ) units of labor in advertising successfully diffuses its prototype with instantaneous probability
L δ, y (t )
δ(t ) ≡ , (2.6)
X δ (t )
where X δ (t ) captures the difficulty of diffusion. In our primary specification, we assume that advertising is purely infor-
mative and set X δ (t ) = κδ , with κδ > 0. That is, we assume that the probability of diffusion depends only on young firm
advertising as they build product awareness and communicate product advantages to mainstream consumers. In Section 6,
we consider the case of combative advertising in which old firms hire labor to block the diffusion efforts of young firms.
In particular, we consider an alternate specification of X δ (t ) = κδ L δ,o (t ), where L δ,o (t ) denotes the endogenous defensive
advertising employment by the old firm in the same industry. Thus, the alternate specification frames the diffusion pro-
cess as a marketing contest between young and old firms who battle for the favor of mainstream consumers. Under either
specification, until a second k + 1 prototype is introduced into the industry, the first k + 1 prototype has an endogenous
probability of successfully diffusing per time interval dt equal to δ(t )dt.
However, innovative efforts continue in B industries while prototypes struggle to win over mainstream consumers. Should
a second k + 1 prototype be introduced into the market prior to successful diffusion of the first prototype into the main-
stream, early adopters no longer view the now outdated first prototype favorably. Recall that early adopters are novelty
seeking; they attach a λ quality advantage to the latest prototype only, while viewing all previous prototypes as quality
inferior. Limit pricing implies that the firm with the second prototype serves all early adopters and fully displaces the pre-
vious young firm from the market. Thus, from the perspective of challengers conducting R&D, the reward for innovating a
new prototype in A and B industries is identical. New innovators always serve early adopters and attempt to diffuse their
prototype as a young firm. As we will see, this implies that the model is consistent with a balanced growth equilibrium in
which A and B industries share a common innovation rate. Finally, note that the introduction of additional prototypes into
an existing B industry does not impact the quality perception of mainstream consumers, and therefore does not impact the
old firm that serves them.
Let n A (t ) and n B (t ) = 1 − n A (t ) denote the proportion of A and B industries in the economy at time t respectively. An A
industry cycles to a B industry when an innovation occurs. Over an interval of time dt, the transition rate of A industries
to B industries is n A (t ) I (t )dt. Similarly, a B industry switches to an A industry when a prototype diffuses successfully. The
associated flow into A industries is (1 − n A (t ))δ(t )dt. This implies that the proportion of A industries in the economy evolves
endogenously according to,
ṅ A = (1 − n A (t ))δ(t ) − n A (t ) I (t ). (2.7)
We define the diffusion failure rate as the proportion of prototypes that exit the market without successfully diffusing into
the mainstream. Over an interval of time dt, each existing prototype has a δ(t )dt probability of diffusion success and an
I (t )dt probability of failure. Therefore, the endogenous rate of diffusion failure for the mass of prototypes is given by,
I (t )
f (t ) ≡ . (2.8)
I (t ) + δ(t )
2.3. Labor and production
Labor is used for three separate tasks: advertising, R&D, and the production of consumption goods. Households supply
labor inelastically and labor is freely mobile across industries and tasks. We normalize the wage rate common to all labor
to unity and assume that one unit of labor produces one unit of the consumption good in each industry. As discussed in
the previous sections, each type of firm (young, adult, old) enjoys a λ size perceived quality advantage over its nearest
competitor with a fixed proportion of the population (φ , 1, 1 − φ respectively). Thus, each firm type optimally captures its
respective market share with limit pricing at a common price of p (t ) = λ.
6
In A industries, a single adult firm serves the entire market with corresponding quantity sold equal to za (t ) = c (t )/λ,
where c (t ) denotes per capita consumption expenditure common to all households. Thus, each adult firm earns flow profits
equal to
(λ − 1)
πa (t ) = p (t )za (t ) − za (t ) = c (t ) . (2.9)
λ
The situation is identical in B industries after accounting for the proportion of consumers served by young and old
firms. Young firms serve the economy’s φ early adopters, sell z y = c (t )φ/λ units, and earn flow profits (gross of advertising
expenditure) equal to π y (t ) = c (t )φ(1 − λ−1 ) = φ πa (t ). Old firms serve the economy’s (1 − φ) mainstream consumers, sell
zo = c (t )(1 − φ)/λ units, and earn flow profits equal to πo (t ) = (1 − φ)πa (t ). Note that total employment in production is
common across A and B industries. For the economy as a whole, labor used in the production of consumption goods is
c (t )
L c (t ) = . (2.10)
λ
Finally, note that advertising is conducted only by young firms in B industries and challengers target their R&D efforts at
all industries. Using (2.5) and (2.6), total employment in R&D and advertising respectively are given by
L I (t ) = κ I I (t ), L δ (t ) = n B (t ) L δ, y (t ) = n B (t )κδ δ(t ). (2.11)
2.4. Stock market valuations and optimal advertising
Each challenger j chooses R&D employment L I , j (t ) to maximize its expected discounted profits. Free-entry into R&D
implies that the expected return to R&D must exactly offset its cost in every industry with positive research expenditure.
Let V k (t ) denote the value of a firm of type k in a typical industry and σ I < 1 denote the subsidy rate for research expen-
ditures.11 In a symmetric equilibrium with I (ω, t ) = I (t ) > 0, free-entry requires that challengers in each industry choose
R&D employment so that I j (t ) V y (t ) = (1 − σ I ) L I , j (t ). Given (2.5), the free-entry condition can be rewritten as
V y (t ) = (1 − σ I )κ I , (2.12)
which relates the value of a young firm to the cost of innovating a new prototype. Note that (2.12) directly implies that the
value of a young firm will be constant in equilibrium.
The value of a young firm V y (t ) is determined by a no-arbitrage condition that equates the expected return on stocks
issued by the firm to the risk-free market interest rate r (t ). Over a time interval dt, the young firm earns flow profit φ πa (t )dt
and incurs a cost of advertising (1 − σδ ) L δ, y (t )dt, where σδ < 1 denotes the subsidy rate for advertising expenditures. With
probability δ(t )dt, the firm successfully diffuses its product and enjoys a capital gain of V a (t ) − V y (t ). With probability I (t )dt,
a new prototype arrives and displaces the young firm creating a capital loss of V y (t ). With probability (1 − δ(t )dt − I (t )dt ),
the firm retains its position as a young firm and there is an associated change in valuation of V̇ y (t )dt. Combining terms, the
corresponding no-arbitrage condition is
r (t ) V y (t )dt = φ πa (t )dt − (1 − σδ ) L δ, y (t )dt + δ(t )( V a (t ) − V y (t ))dt −

(2.13)
I (t ) V y (t )dt + (1 − δ(t )dt − I (t )dt ) V̇ y (t )dt .
Each young firm chooses L δ, y (t ) in order to maximize its valuation. This is equivalent to maximizing its expected stock
return as given by the right hand side of (2.13). Using the definition of δ(t ) from (2.6) and taking limits as dt → 0, the
first order condition associated with optimal advertising can be written δ(t )( V a − V y ) = (1 − σδ ) L δ, y (t ), which equates the
expected return and cost of advertising expenditures. Noting that (2.12) implies that V̇ y (t ) = 0, substituting this condition
into (2.13) yields the following expression for the value of a young firm
φ πa (t )
V y (t ) = . (2.14)
I (t ) + r (t )
Much like firms in traditional endogenous growth models, young firms discount flow profits by the rate of replacement
I (t ) and the interest rate r (t ). After incorporating optimal advertising employment, the valuation of a young firm does not
directly depend on the expected value of diffusing into the mainstream since it is exactly offset by advertising expenditure.
Finally, again using (2.6), note that the optimal advertising condition is equivalent to
V a (t ) − V y (t ) = (1 − σδ )κδ , (2.15)
11
Following Segerstrom (1998) and Dinopoulos and Segerstrom (1999), we assume that subsidies are financed through non-distortionary, lump-sum taxes
imposed on households. This ensures that subsidies do not influence households’ dynamic optimization as captured by equation (2.4).
7
which provides a useful expression for the equilibrium value of successful diffusion.
The expected return of holding V a (t ) of stock in an adult firm over interval dt includes the profit flow πa (t )dt, minus
the I (t )dt probability that a new prototype will be innovated and force the firm into old age with associated capital loss
V a (t ) − V o (t ). With probability (1 − I (t )dt ), no such innovation occurs and the value of the adult firm changes by V̇ a (t )dt = 0.
Rearranging the associated no-arbitrage condition yields an expression for value of an adult firm
πa (t ) + I (t ) V o (t )
V a (t ) = . (2.16)
I (t ) + r (t )
Finally, an old firm generates a profit flow (1 − φ)π dt and faces a capital loss of V o (t )dt if the young firm in the
industry displaces it by successfully diffusing a new prototype with probability δ(t )dt. If diffusion does not occur, the old
firm experiences a change in valuation of V̇ o (t )dt = 0. Once again using the associated no-arbitrage condition, we obtain an
expression for the value of an old firm
(1 − φ)πa (t )
V o (t ) = . (2.17)
δ(t ) + r (t )
2.5. Equilibrium
We now solve for a steady state equilibrium in which { I (t ), δ(t ), c (t ), f (t ), na (t ), πa (t ), V y (t ), V a (t ), V o (t )} are constant
and the following three equilibrium conditions are met: 1) the free-entry condition of (2.12) holds, 2) young firms choose
advertising expenditure to maximize their value according to (2.15), and 3) the labor market clears. Henceforth, we drop
the time index for all variables that are constant in equilibrium.
Imposing ṅ A = 0 in (2.7) yields an expression for the equilibrium proportion of A and B type industries in terms of the
rate of innovation and diffusion,
δ I
nA = , nB = 1 − n A = . (2.18)
δ+I δ+I
Observe from (2.8) that the economy’s endogenous failure rate f of new innovations is equal to the proportion of B type
industries n B in equilibrium. This is because industries cycle from their B to A configurations if and only if a prototype
diffuses successfully. In traditional models of endogenous growth with instantaneous new product diffusion, all industries
exhibit an A type structure and innovations never fail. In our framework, this corresponds to the limit case of δ → ∞.
Next, rewrite the free entry condition (2.12) using the value of a young firm (2.14), noting that r = ρ in equilibrium.
This provides the following equilibrium relationship between I and c based on the cost and reward from innovating a new
prototype,
c φ(λ − 1)
(1 − σ I )κ I = . (2.19)
λ( I + ρ )
As in traditional models, a greater level of per capita consumption c implies a greater profit incentive to conduct R&D and
a corresponding higher rate of innovation. Similarly, rewrite the optimal advertising condition (2.15) using the expressions
for the value of each firm type (2.14), (2.16), and (2.17). This provides an equilibrium relationship between δ , I and c based
on the cost and reward from successfully diffusing a prototype into the mainstream,
c (1 − φ)(λ − 1) δ+I +ρ
(1 − σδ )κδ = . (2.20)
λ ( I + ρ )(δ + ρ )
Once again, a greater c implies young firms have a greater incentive to advertise since capturing the entire industry’s market
share as an adult firm becomes more profitable. On the other hand, a greater rate of new prototype innovation I reduces
the incentive to invest in advertising since the expected duration of market dominance as an adult firm is shortened.
Finally, labor market clearing requires that the aggregate supply of labor, L = 1, equals aggregate demand for labor across
production, R&D, and advertising, given by (2.10) and (2.11) respectively. Combining terms, we have
c Iδ
1= + κ I I + κδ . (2.21)
λ I +δ
Equation (2.21) captures the trade-off inherent to allocating finite labor resources across the three market activities that
require labor. Note that the labor requirement of increasing δ depends on the cost of diffusion in a particular industry κδ
and the proportion of industries with young firms using resources to advertise their prototypes n B = I /( I + δ).
Free-entry (2.19), optimal advertising (2.20) and market clearing (2.21) comprise the model’s three equilibrium condi-
tions in terms of our three central endogenous outcomes c, δ , and I . To facilitate our analytical results, we now combine
these conditions to derive two equilibrium relationships in terms of δ and I only. Substituting (2.20) into (2.19) to elimi-
nate c yields our first equilibrium relationship in (δ, I ) space, which captures the relative incentive to invest in R&D over
advertising,
8
φ(1 − σδ )κδ
I = (ρ + δ)
, where
≡ −1 [RDAC] (2.22)
(1 − φ)(1 − σ I )κ I
We refer to equation (2.22) as the “R&D - advertising curve” (RDAC). Clearly, in order to be consistent with an equilibrium
with a positive rate of innovation and diffusion, the parameters of the model must produce
> 0. This condition will be
met when the benefit of R&D (an initial φ market share as a young firm) relative to the cost of R&D (the (1 − σ I )κ I term)
is sufficiently high compared to the benefit of diffusion (capturing an additional 1 − φ market share) relative to the cost of
diffusion (the (1 − σδ )κδ term). Henceforth, we assume the following,
Assumption 1.
> 0, as defined in (2.22).
With this parameter restriction in place, the RDAC specifies a linear and upward sloping relationship when graphed in (δ ,
I ) space. To understand why, first recall that the incentives to invest in R&D and advertising both scale proportionally with
c. Thus, the c term vanishes in (2.22). Second, although R&D incentives do not directly depend upon the rate of diffusion,
advertising incentives are strictly decreasing in the economy’s diffusion rate. This is because a higher overall diffusion rate
decreases a firm’s expected tenure as an old firm through equation (2.17), thereby decreasing the value of adult firm through
equation (2.16). For a given level of I , a larger δ implies a greater relative incentive to invest in R&D over advertising. To
restore the RDAC, I must change to realign relative incentives. It follows from (2.19) and (2.20) that, although a greater I
decreases the reward from both R&D and advertising, the effect is stronger for R&D. Thus, a greater I decreases the relative
incentive to invest in R&D and restores equilibrium.
To derive our second equilibrium relationship in (δ, I ) space, we use the relationship between I , δ and c from the
optimal advertising condition (2.20) to substitute for c in equation (2.21). This allows us to express the labor market clearing
condition (LMCC) in its final form in terms of δ and I ,
(1 − σδ )κδ (ρ + I )(ρ + δ) Iδ
1= + κ I I + κδ [LMCC] (2.23)
(1 − φ)(λ − 1) (ρ + I + δ) I +δ
Since both R&D and advertising require labor resources, the LMCC is strictly downward sloping in (δ , I ) space.
We determine the model’s steady state equilibrium by solving the RDAC given by (2.22) and the LMCC given by (2.23)
for δ and I . Note that the LMCC provides an upper bound on the rate of innovation the economy’s resources can support.
Specifically, when δ → 0, the LMCC implies that I → I max , where
1− (1 − σδ )κδ ρ
I max = , and = . (2.24)
κI (1 − φ)(λ − 1)
Intuitively, I max represents the rate of innovation in the economy if no resources are devoted to advertising and all firms
remain in their young stage until they are displaced by subsequent innovation. Similarly, the RDAC provides a lower bound
on the rate of innovation that is consistent with relative R&D and advertising incentives that produce a positive rate of
diffusion in equilibrium. That is, when δ → 0, the RDAC implies that I → I min , where
I min = ρ
. (2.25)
Fig. 1 depicts the model’s equilibrium by graphing the RDAC and LMCC in (δ , I ) space. As illustrated in the figure, the
following additional parameter restriction is necessary and sufficient to guarantee a unique steady state equilibrium.
Assumption 2. I max > I min , as defined in (2.24) and (2.25) respectively.
After establishing δ and I , all other endogenous variables can be determined. We follow standard practice in quality
ladder models and define economic growth as the growth rate of per capita sub-utility ln(u (t )). As shown in the Appendix,
we can decompose per capita sub-utility into the following three terms,
ln(u (t )) = ln(c /λ) + n B φln(λ) + ln(λ)(1 − f ) It . (2.26)

The first term captures the standard static effect of per capita consumption given limit pricing. This term is common
across all industries since all firms charge the same price of p = λ. On the other hand, in B industries, early adopters and
mainstream consumers purchase products of different perceived quality. The second term captures this additional static
effect by accounting for the n B industries in which a φ proportion of early adopter consumers purchase products that they
perceive to be one λ step up the quality ladder from the current quality standard. The final term captures the dynamic effect
of the arrival of new innovations. In our framework however, only innovations that successfully diffuse into the mainstream
push the economy up the quality ladder.12
12
Recall that transition from a B industry to an A industry occurs if and only if a prototype diffuses successfully. Thus, the aggregate rate of successful
Iδ
diffusion is δnb = I +δ , which is equal to (1 − f ) I with f defined as in (2.8).
9
Fig. 1. Equilibrium.
Differentiating (2.26) with respect to time yields an expression for the rate of economic growth, g,
Iδ
g = ln(λ)(1 − f ) I = ln(λ). (2.27)
I +δ
Note that innovation and diffusion have a complementary impact on economic growth; the marginal increase in economic
growth associated with an increase in innovation (diffusion) depends positively on the rate of diffusion (innovation). As we
discuss in the following sections, this relationship will drive the growth impact of policy changes that move the innovation
and diffusion rate in opposite directions.
3. Welfare
In this section, we evaluate the optimal allocation of labor across its three possible uses: advertising, R&D and the
manufacture of consumption goods. We consider a social planner who chooses the associated first-best levels of I , δ and
c to maximize social welfare, subject to the aggregate resource constraint (2.21). As is standard, we assume that the social
planner cannot impact firm pricing decisions. Using our expression for per capita sub-utility (2.26), welfare discounted to
time zero can be written as
ln(λ) I δ
ρU = + ln(c ) − ln(λ) 1 − φn B . (3.1)
ρ ( I + δ)
The Lagrangian associated with the social planner’s optimization problem and resulting first-best (FB) allocation is
c
L F B (c , I , δ, μ) = ρ U + μ 1 − − κ I I − κδ δn B , (3.2)
λ
where μ is the Lagrange multiplier and ρ U is given by (3.1). The welfare maximizing condition for I is obtained from the
following first order condition
∂nB c ln(λ)δ ∂ n B c ∂nB
L IF B = 0 ⇒ κ I + κδ δ = + φln(λ) , (3.3)
∂I λ ρ ∂I λ ∂I
which equates the social cost and return to R&D at the margin. The analogous expression for the market cost and return to
R&D is given by (2.19).
Comparing these expressions, we see several reasons for the market and socially optimal levels of R&D to differ. To
begin with, we capture two standard effects that are well established in the Schumpeterian growth literature. The first is
the monopoly distortion effect. The social planner considers the utility benefit of a marginal innovation measured by ln(λ),
whereas firms consider the monopoly mark-up rate of λ − 1. Since λ − 1 > ln(λ), this effect implies the market overinvests
in R&D. The second is the intertemporal spillover effect. The social planner discounts the benefits of innovation by ρ , while
the effective market discount rate of I + ρ incorporates the expected capital loss due to replacement. With I > 0, this effect
implies that the market underinvests in R&D.
Furthermore, our modeling of endogenous diffusion generates three novel effects.
1. Diffusion resource burden of innovation. A greater innovation rate raises the proportion of industries, n B , in which young
firms invest resources in advertising to diffuse their newly innovated prototypes. At a given diffusion rate δ , this
reduces the resources available for R&D and production. The corresponding social cost of R&D is captured by the
κδ δ(∂ n B /∂ I ) > 0 term on the left hand side of (3.3). Since private firms do not consider this additional resource burden
on the aggregate economy, this effect implies that the market overinvests in R&D.
10
2. Dynamic effect of stochastic diffusion. The social planner scales down the dynamic impact of a marginal innovation since
prototype diffusion is uncertain (prototypes fail to diffuse at rate f ). The first term on the right hand side of (3.3)
captures this scaling factor of δ(∂ n B /∂ I ) = (1 − f )2 ∈ (0, 1). Thus, the social planner recognizes that only prototypes
that diffuse successfully push the economy up the quality ladder and generate economic growth. Although private R&D
incentives also incorporate the effect of stochastic diffusion, firms consider only the internal cost and benefits of their
efforts to use advertising to gain market share in the value of an innovation. Private firms scale the entire dynamic
benefit of innovation by φ ∈ (0, 1) to reflect their initial market share of early adopters, but market R&D incentives
do not directly incorporate the economy’s failure rate. Holding all else constant, this effect implies that the market
underinvests in R&D if φ < (1 − f )2 .
3. Early adopter market effect. The social planner considers the one-time utility gain of early adopters from innovation. A
higher innovation rate raises the aggregate mass of the early-adopter market φn B and generates a social gain that is
independent of the dynamic effect of prototype diffusion. This utility gain is captured by (φ cln(λ)/λ)(∂ n B /∂ I ) > 0, the
last term on the right hand side of (3.3). Since private R&D firms do not consider this social gain associated with early
adopters, this effect implies that the market underinvests in R&D.
We now turn to the welfare maximizing condition for δ , which is obtained from the first order condition
cln(λ) φ cln(λ)
LδF B = 0 ⇒ κδ = − , (3.4)
ρλ λI
and equates the social cost and benefit of diffusion promoting advertising at the margin. The analogous expression for the
market based cost and return to advertising is given by (2.20). In addition to the monopoly distortion effect, we identify
three novel differences between the market-determined and socially optimal rates of advertising.
1. Dynamic effect of life-cycle replacement. The social planner recognizes that prototype diffusion always pushes the economy
up the quality ladder and generates economic growth. Thus, the social planner applies a ρ discount factor to the welfare
benefit of diffusion. In contrast, young firms discount the benefit of diffusion based on the different rate of replacement
they face over the firm life-cycle. That is, firms enjoy the adult status they gain from diffusion only until an innovation
occurs and they transition to their old stage. Once old, they are displaced from the market entirely at the rate of
diffusion. The corresponding market discount rate is captured by the second term in brackets in (2.20). For all δ > 0
and I > 0, this effect implies that the market underinvests in diffusion promoting advertising.
2. Effect of early adopters on the benefit of diffusion. The social planner recognizes that early adopters do not immediately
benefit from the diffusion of new prototypes since they already purchase the prototype prior to diffusion. Instead, early
adopters benefit only after the next innovation arrives, which enables them to consume a prototype at the next quality
ladder step. The social planner accounts for this effect by subtracting φ cln(λ)/(λ I ) from the overall benefit of diffusion
in (3.4). Young firms engaged in advertising do not take this effect into consideration. For any positive and finite I , this
effect implies that the market overinvests in advertising.
3. Mainstream market effect. Since young firms already serve the early adopter market, they only consider the benefit of the
additional 1 − φ market share gained by capturing the mainstream market. In contrast, the social planner considers the
dynamic full impact of diffusion through the aggregate growth rate. This effect implies that the market underinvests in
advertising.
4. R&D subsidies
As the preceding discussion makes clear, the model’s market equilibrium features two potential sources of dynamic
inefficiency: the traditional allocative inefficiency from over or underinvestment in R&D and a novel source of inefficiency
from over or under investment in diffusion promoting advertising. Theoretically, the first-best allocation can be achieved by
correcting both of these dynamic distortions through the use of two distinct policy instruments: one targeted at R&D and
the other targeted at advertising. In this section however, we consider the use of R&D policy as a single policy instrument
to achieve a second-best outcome. This focus aligns with a large endogenous growth literature that has emphasized the use
of R&D policy to correct dynamic inefficiency, and typically makes the case for large R&D subsidies. We wish to investigate
how incorporating an endogenous diffusion process can impact these results. We return to a discussion of advertising policy
in Section 7.
We begin with a comparative statics exercise to examine how the R&D subsidy rate (σ I ) impacts the equilibrium rate
of innovation, diffusion, product failure, and economic growth. Recall that the RDAC reflects the relative incentive to invest
in R&D and advertising according to their relative cost and reward. Implementing an R&D subsidy directly reduces the cost
of conducting R&D relative to advertising. As shown in equation (2.19), free-entry into R&D implies that challenger firms
respond to the decrease in R&D costs by increasing R&D investment at any given expected reward for innovating successfully
(V y ). Since the reward associated with advertising investment (V a − V y ) is not directly impacted by the R&D subsidy, this
implies a greater rate of innovation I at any given diffusion rate δ . Thus, the RDAC shifts leftward as depicted in Fig. 2. This
shift generates movement along the downward sloping LMCC as more labor resources are devoted to R&D. As a result, the
equilibrium rate of innovation increases when an R&D subsidy is implemented, but the rate of diffusion always falls.
11
Fig. 2. R&D subsidy.
This immediately implies an increase in the equilibrium product failure rate, f = I /( I + δ). Since only prototypes that
diffuse successfully push the economy up the quality ladder, the net effect on the equilibrium rate of growth, g = ln(λ) I (1 −
f ), is determined by the competing effects of the increased arrival of new prototypes through innovation against their
increased rate of diffusion failure. In the general case, the overall change to economic growth is ambiguous. These findings
are summarized in the following proposition,
Proposition 1. Subsidizing R&D investment decreases the diffusion rate, increases the innovation rate, increases the product failure
∂I ∂f ∂g
rate, and has an ambiguous effect on economic growth. That is, ∂∂δ
σ < 0, ∂ σ > 0, ∂ σ > 0, and ∂ σ >< 0.
I I I I
Proposition 1 establishes that our model is consistent with empirical evidence that R&D and advertising act as substitutes
at the firm level (Cavenaile and Roldan-Blanco, 2021; Cavenaile et al., 2021). That is, we find that R&D subsidies stimulate
R&D investment by challenger firms but decrease each successful innovator’s investment in diffusion promoting advertising.
This is because R&D subsidies reduce the reward from successful diffusion by shortening each young firm’s expected period
of market dominance as an adult firm. In other words, it is precisely because R&D subsidies successfully stimulate innovation
that they also reduce equilibrium advertising incentives. The greater the rate of innovation, the more frequently adult firms
are partially replaced by innovative young firms, and the weaker the incentive to invest in diffusion promoting advertising.
For this reason, economic growth may increase or decrease when R&D is subsidized.13
In terms of aggregate expenditures however, R&D and advertising may behave as substitutes or complements. To see this,
note that aggregate expenditure on R&D is equal to L I = κ I I , which strictly increases when R&D is subsidized. Although each
individual young firm decreases advertising investment, aggregate advertising expenditure also depends on the equilibrium
proportion of industries in which young firms seek to diffuse a prototype into the mainstream market, n B = I /( I + δ).
Proposition 1 directly implies that n B increases when R&D is subsidized. Since L δ = n B κδ δ = lnκ(λ)
δ
g, aggregate advertising
expenditure is proportional to the growth rate, which may increase or decrease in response to changes in σ I .
4.1. Welfare effects of R&D subsidies
As stated in Proposition 1, subsidizing R&D always increases the equilibrium rate of innovation and decreases the rate
of diffusion. This implies that the net welfare impact of subsidizing R&D depends on its effect on both dynamic distortions
present in the model. To understand the forces at work, we consider an augmented social planner problem in which market-
determined advertising incentives enter as an additional constraint. This ensures that the social planner takes into account
how her allocation of labor resources affects the dynamic inefficiency associated with equilibrium over or under investment
in diffusion promoting advertising. Therefore, this augmented social planner’s problem directly corresponds to a second-best
allocation that may be achieved with R&D policy as the sole policy instrument available.
Formally, the Lagrangian associated with this social planner’s problem and resulting second-best (SB) allocation is
c
L S B (c , I , δ, μ, η) = ρ U + μ 1 − − κ I I − κδ δn B + η H (c , I ) − δ , where (4.1)
λ
( I + ρ )[c (1 − φ)(λ − 1) − ρκδ λ]
δ = H (c , I ) ≡ ,
( I + ρ )κδ λ − c (1 − φ)(λ − 1)
13
In this paper’s supplementary material (Section S.4.2), we show that this result continues to hold in an alternative specification of the model that
removes the link between I , δ , and c from the labor market clearing condition. This exercise illustrates that the equilibrium decrease in δ is driven by the
impact of the increase in the innovation rate on advertising incentives, rather than crowding-out effects from the reallocation of labor resources.
12
∂ H (c , I ) ∂ H (c , I )
∂c > 0,< 0, and η is the Lagrange multiplier associated with this new constraint. The expression for δ follows
∂I
directly from (2.20), the optimal advertising condition. The optimality condition for the second-best level of I is obtained
from the first order condition L IS B = 0. In the Appendix, we show that this condition is equivalent to

∂ H (c , I ) ∂ n B
∂I ∂δ − λ κ I + κδ δ ∂∂nIB ∂ H∂(cc, I ) ∂∂δ
nB
L IS B = L IF B + sδ · LδF B = 0, where sδ ≡ . (4.2)

λκδ ∂ H∂(cc, I ) ∂∂δ
nB
− 1I
Just as in Section 3, L IF B expresses the difference between the social benefit and social cost of a marginal increase in
R&D from the first-best solution. Thus, it embeds all static and dynamic welfare effects associated with allocating additional
labor resources to R&D as discussed in Section 3. Indeed, it is exactly equal to the sum of all five previously defined effects
that give rise to the wedge between the market equilibrium and first-best levels of R&D.14 Similarly, LδF B expresses the
difference between the social benefit and cost of a marginal increase in diffusion promoting advertising, and is exactly equal
to the sum of all three welfare effects that determine the size of the dynamic distortion in advertising investment. Equation
(4.2) thus implies that the optimality condition for R&D investment associated with the second-best allocation is a weighted
combination of the two dynamic distortions present in the model.
To clarify the intuition, suppose that the initial market equilibrium exhibits underinvestment in both R&D and adver-
tising.15 Necessarily in this case, L IF B > 0 and LδF B > 0 since the social benefit of both R&D and advertising investment
exceed their social cost. If feasible, a policy maker would be inclined to subsidize both activities, eliminate both investment
distortions and achieve the first-best solution with L IF B = 0 and LδF B = 0. When the policy maker is constrained and has
R&D policy as the sole instrument available, it is no longer possible to eliminate both distortions. Instead, she must weigh
the welfare gain from reducing underinvestment in R&D against the welfare cost of magnifying underinvestment in advertis-
ing. The sδ term in (4.2) is a scaling factor that captures the magnitude of the general equilibrium effects that are associated
with the decrease in advertising investment when R&D is subsidized. Since ∂ H∂(cc , I ) > 0, ∂ H∂(cI , I ) < 0, and ∂∂δ
nB
< 0, this scaling
factor is always negative, sδ < 0. When there is underinvestment in advertising, the multiplicative term sδ · LδF B < 0 cap-
tures the total welfare loss specifically arising from the general equilibrium trade-off between innovation and diffusion. In
the following section, we use the welfare decomposition of equation (4.2) to explore the welfare effects of R&D subsidies
and characterize the optimal policy through numerical analysis.
5. Numerical analysis
The model features the following parameters {ρ , λ, φ, κ I , κδ }. As is standard practice, we set the discount factor ρ and
innovation size λ to reflect estimates of the long-run real interest rate and gross mark-up over marginal cost in the US
respectively. We set ρ = 0.06, which is an intermediate value in the range of conventional estimates of 0.05 - 0.07 (Chu and
Cozzi, 2018; Chu et al., 2019; Klein, 2020). Similarly, we set λ = 1.25 to reflect mark-up estimates in the range of 1.05 - 1.4
(Segerstrom, 2007; Davis and Şener, 2012; Klein, 2020).
The parameter φ determines the proportion of households that comprise our two broad consumer categories, early
adopters (φ ) and mainstream consumers (1 − φ ). These two categories represent a simplified version of the classic partition
of consumers used in the innovation diffusion marketing literature into five groups: innovators, early adopters, early ma-
jority, late majority, and laggards (Rogers, 1962). We choose φ = 0.20 to align with traditional estimates of the combined
size of the innovator and early adopter marketing categories of 12.5% to 23% of the market and the corresponding size of
the remaining three categories of 77% to 87.5% (Mahajan et al., 1990). We note that φ = 0.20 is also consistent with the
estimated 15-32% profit reduction when firms enter the “slowdown” phase of their life cycle, which corresponds to their
old firm stage in our model (Chandrasekaran and Tellis, 2007).
The remaining parameters κ I and κδ determine the relative difficulty of innovation and diffusion. We jointly calibrate
κ I and κδ to match target rates of economic growth and prototype failure. We choose a conventional growth rate target
of g 0 = 1.5%. As noted in the introduction, the rate new product failure is typically estimated in the range of 40-50%. In
this empirical literature, the definition of product failure often includes products that exhibit a sufficiently large under-
performance relative to sales targets, in addition to products that are removed from the market entirely. Furthermore, the
literature does not attempt to estimate the quality of new products relative to existing offerings. In our model however, a
new prototype fails only when it is fully displaced from the market and each new prototype has the potential to be viewed
as a sizable quality improvement over the current state of the art product. To account for these differences, we target a
conservative benchmark value of f 0 = 0.25. This yields calibrated values of κ I = 0.292 and κδ = 1.486. We analyze the
impact of alternate failure rate targets in Section 5.2.
Together with the preset value of λ = 1.25, our g 0 and f 0 targets also pin down the associated equilibrium rates of
innovation and diffusion, since g = I (1 − f )ln(λ) and f = I /( I + δ). Specifically, our benchmark targets imply I 0 = 0.090
14
For details, see our Section S.1 of our supplementary material.
15
This is the case in the numerical simulations we examine and corresponds to the typical finding in endogenous growth models that the market
equilibrium exhibits underinvestment in growth promoting activity. In traditional models, this underinvestment corresponds to R&D only by construction.
In our model, both R&D and diffusion promoting advertising contribute positively to growth.
13
and δ0 = 0.269. This corresponds to an expected 1/ I 0 = 11.11 years between subsequent innovations within an industry.
This reflects an intermediate value within the broad range used in the endogenous growth literature of about eight years
on the low end (Chu and Cozzi, 2018; Yang, 2018) and about twenty years on the high end (Segerstrom, 2007; Klein, 2020).
The expected 1/δ0 = 3.72 years before successful diffusion is consistent with empirical estimates of the average time from
product introduction to sales takeoff in developed economies of between two and six years (Chandrasekaran and Tellis,
2007).
5.1. Optimal R&D subsidies
We begin by analyzing the impact of subsidizing R&D in the benchmark equilibrium and report results in Fig. 3 and
Table 1. Panels (a) and (b) of Fig. 3 illustrate the main results of Proposition 1. Subsidizing R&D always increases the equi-
librium innovation rate and reduces the equilibrium rate of diffusion. Consequently, a greater proportion of newly innovated
prototypes fail to diffuse into the mainstream market. Since g = ln(λ) I (1 − f ) as in (2.27), the change in equilibrium growth
rate is determined by the competing effects of the increased arrival rate of new prototypes through innovation against the
increased diffusion failure rate of existing prototypes. At low levels of the R&D subsidy rate σ I , the failure rate is sufficiently
low so that the innovation effect dominates and economic growth increases with the subsidy. At higher levels of σ I , the
effect of the relatively large failure rate dominates and economic growth decreases with the subsidy. This generates an
inverted U-shaped relationship between R&D subsidies and economic growth. In this case, growth is maximized at an R&D
subsidy rate of σ I, g = 0.155.
Fig. 3. Impact of R&D subsidies.
14
Table 1
Optimal R&D subsidy.
g (%) I δ c f U L IS B L IF B sδ LδF B σ I, g σ I,U

σI = 0 1.500 0.090 0.269 1.092 0.250 2.106 0.170 0.782 -0.612 0.155 0.092
σ I,U = 0.092 1.520 0.104 0.199 1.086 0.342 2.127 0.000 0.545 -0.545 – –
Table 1 presents equilibrium values of the model’s endogenous variables when σ I = 0 and when R&D subsidized at the
welfare maximizing level (σ I,U ).
Panels (c) and (d) of Fig. 3 illustrate the welfare effects of an R&D subsidy. First, observe that L IF B > 0 and sδ LδF B < 0 in
the initial equilibrium. Since sδ is always negative, we have LδF B > 0, and the initial equilibrium exhibits underinvestment
in both R&D and advertising. However, L IS B ≡ L IF B + sδ LδF B > 0 in the initial equilibrium, which implies that the social
benefit of using an R&D subsidy to reduce underinvestment in R&D exceeds the social cost of magnifying underinvestment
in advertising.
As R&D is subsidized more heavily, the equilibrium rate of innovation increases and the magnitude of underinvestment
in R&D decreases. Consequently, the social benefit of further stimulating innovation with an R&D subsidy, L IF B , decreases
in the subsidy level. Interestingly, the total social cost of further exacerbating underinvestment in advertising, |sδ | · LδF B ,
also decreases in the subsidy level. This is because the net change to this social cost depends on two competing forces.
On the one hand, subsidizing R&D magnifies underinvestment in advertising as the diffusion rate falls. This implies that
LδF B increases in the subsidy level. On the other hand, the rate at which advertising investment further decreases with
the subsidy declines in the subsidy level. This is illustrated in panel (a), where the rate of decline in δ is clearly sharpest
at lower levels of the subsidy. This corresponds to a decrease in the scaling factor |sδ | as R&D is subsidized more heavily,
which dominates the associated increase in LδF B .
Nevertheless, the net welfare benefit of subsidizing R&D, L IS B , strictly decreases in the subsidy level, as shown in panel
(c). This gives rise to an inverted-U shaped relationship between R&D subsidies and welfare. The welfare maximizing R&D
subsidy is such that L IS B = 0 at σ I,U = 0.092. In contrast to traditional endogenous growth models that consider only the
distortion in R&D investment, observe that the optimal 9.2% R&D subsidy does not eliminate equilibrium underinvestment in
R&D. Indeed, as seen in panel (c), a much higher subsidy of σ I = 37.9% is required to achieve L IF B = 0. Rather, the optimal
subsidy balances the welfare gain of stimulating R&D against the welfare loss of reducing diffusion promoting advertising
incentives. This analysis necessarily reflects the static and dynamic welfare effects of reallocating labor resources across
R&D, advertising and production as emphasized in Section 3 because these effects are directly embedded in L IF B and LδF B .
The optimal subsidy of 9.2% generates a dynamic welfare gain by increasing the growth rate from 1.5% to 1.52% despite the
decline in the diffusion rate. At the same time, the R&D subsidy increases the fraction of early adopter markets from 0.25
to 0.342 since n B = f , producing a static welfare gain. These welfare gains dominate the static loss associated with the fall
in equilibrium consumption expenditure from 1.092 to 1.086.
5.2. Alternate initial failure rates
In this section, we examine how the optimal R&D subsidy depends on the failure rate in the initial equilibrium. We
consider a low initial failure rate case with f 0 = 0.01 and a high initial failure rate case with f 0 = 0.45. In both cases, we
recalibrate κ I and κδ such that economic growth remains at its initial targeted level of g = 1.5%.16 We report numerical
results in Table 2.
Table 2
Alternate failure rates: optimal R&D subsidies.
(a): f 0 = 0.01 g (%) I δ c f U L IS B L IF B sδ LδF B σ I, g σ I,U

σ =0 1.500 0.068 6.636 1.104 0.010 2.097 0.735 1.512 -0.777 0.316 0.240
σ I,U = 0.240 1.652 0.104 0.257 1.076 0.289 2.306 0.000 0.645 -0.645 – –
(b): f 0 = 0.45 g (%) I δ c f U L IS B L IF B sδ LδF B σ I, g σ I,U
σ =0 1.500 0.122 0.149 1.087 0.450 2.167 -0.080 0.361 -0.441 -0.035 -0.077
σ I,U = −0.077 1.499 0.110 0.173 1.090 0.388 2.177 0.000 0.495 -0.495 – –
Table 1 presents equilibrium results when σ I = 0 and when R&D subsidized at the welfare maximizing level (σ I,U ) in our low
and high failure rate cases with f = 0.01 and f = 0.45 respectively.
Our central result is that the optimal R&D policy σ I,U can be a subsidy or a tax depending on the initial diffusion
failure rate f 0 . In our low failure case with f 0 = 0.01, we find a large optimal subsidy of σ I,U = 24.0%. This is because
the welfare gain from reducing the substantial underinvestment in R&D dominates the welfare loss from exacerbating the
16
In the low failure rate case, these parameter values are κ I = 0.345 and κδ = 1.394. In the high failure rate case, they are κ I = 0.239 and κδ = 1.511.
15
market’s relatively less severe underinvestment in advertising. Indeed, the relatively high initial value of δ = 6.636 implies
that underinvestment in advertising in the initial equilibrium is minor (LδF B > 0 is small). This makes the total welfare cost
of reducing advertising investment with an R&D subsidy relatively low, despite the substantial decline in the equilibrium
value of δ and the associated large scaling factor |sδ |. Overall, subsidizing R&D is welfare improving since L IF B > |sδ |LδF B
in the initial equilibrium. The large optimal R&D subsidy of 24% promotes innovation and raises the economic growth rate
from 1.5% to 1.65%, even though the failure rate increases substantially from 0.01 to 0.28. While there is a small decline in
the level of consumption per capita as fewer resources are devoted to manufacturing, the dynamic welfare gain from higher
growth and the static welfare gains due to the increased share of early adopter markets dominate the static losses from
lower consumption.
In contrast, when the failure rate is relatively high with f 0 = 0.45, the size of equilibrium underinvestment in diffusion
is sufficiently large relative to underinvestment in R&D that it becomes optimal to tax R&D, with σ I,U = −7.8%. As a result
of the tax, the innovation rate falls and the diffusion rate increases, but the growth rate g = ln(λ) I (1 − f ) remains almost
unchanged. Hence, the increase in diffusion following the optimal R&D tax almost completely offsets the negative growth
impact of reducing innovation. The social cost associated with the minor decreases in economic growth and the share
of early adopter markets is dominated by the static benefit of greater consumption as more resources are available for
manufacturing.
This simulation exercise provides two main insights about welfare-maximizing R&D policies. First, our model demon-
strates that the optimal R&D policy is highly sensitive to the initial failure rate. We find that the case for promoting R&D
policy becomes weaker as the initial failure rate increases, even though economic growth is held constant across cases.
This is because a higher failure rate decreases the welfare benefit from reducing the relatively small equilibrium underin-
vestment in R&D and increases the welfare cost of magnifying the relatively large underinvestment in advertising. Indeed,
we find that the optimal R&D policy switches from a subsidy to a tax at the high, but empirically plausible, initial failure
rate of f 0 = 0.45. Our model thus frames the typical finding of large optimal R&D subsidies in Schumpeterian models as
conditional on the implicit assumption of instantaneous diffusion and failure rate of zero.
Second, even in cases where R&D subsidies are welfare improving, our model reveals that the relationship between R&D
subsidies and the endogenous rate of diffusion plays a crucial role in determining the optimal subsidy level. This is because
the equilibrium failure rate always increases with the subsidy. As the subsidy rate increases, the negative growth and welfare
effects of increasing the failure rate and exacerbating underinvestment in advertising, coupled with the negative effect of
lower consumption, eventually dominate the positive effects of stimulating innovation. As illustrated in Fig. 4, this dynamic
implies that the relationship between the R&D subsidy level and welfare and that between the R&D subsidy and growth are
both characterized by similar inverted U-shaped curves for each initial equilibrium failure rate.
Fig. 4 plots equilibrium values of growth g and welfare against the R&D subsidy rate σ I for three distinct initial values of
the failure rate f 0 . Panel (b) plots U as specified in equation (3.1) after normalizing its value to one when σ I = 0.
Fig. 4. Optimal R&D Subsidies.
6. Combative advertising
In this section, we consider an extension to the baseline model to incorporate defensive advertising by old firms. We
now define the instantaneous probability of prototype diffusion in a typical B industry as,
L δ, y (t )
δ(t ) = , (6.1)
κδ L δ,o (t )
16
where L δ,o (t ) denotes defensive advertising employment of the old firm in the industry. Under this specification, the rate
of diffusion is determined by an advertising contest between young and old firms battling for market share. Each old firm
chooses L δ,o (t ) to maximize its value. This yields an optimality condition of (1 − σδ ) L δ,o (t ) = δ(t ) V o (t ), which equates the
total cost of defensive advertising to its benefit in the form of expected prevention of capital loss. Note that the advertising
subsidy, σδ , also applies to defensive advertising. That is, we assume that the policy maker can not specifically target a
particular form of advertising to subsidize or tax. The benefit of defensive advertising increases in the value of serving the
market as an old firm, V o (t ), and the rate that old firms are replaced by young firms through successful prototype diffusion,
δ(t ). The expression for the value of an old firm is now
(1 − φ)πa (t )
V o (t ) = , (6.2)
2δ(t ) + r (t )
where the 2δ(t ) term in the denominator reflects the added cost of defensive advertising expenditure.17
As before, we solve for the model’s steady state equilibrium by deriving two equilibrium conditions in I and δ . Labor
market clearing provides our first equilibrium condition. In equilibrium, the aggregate supply of labor L = 1 must equal
total employment in advertising, R&D and production. Total employment in advertising is given by L δ = n B ( L δ, y + L δ,o ).
Note that, since the combative advertising formulation accommodates reciprocal cancellation, advertising has the potential
to be directly socially wasteful. That is, proportional increases in young and old advertising draw labor resources away from
production and R&D without increasing the diffusion rate. Using the definition of δ from (6.1), we have L δ = n B [ L δ,o (κδ δ +
1)]. Substituting (6.2) into the optimality condition for defensive advertising gives
δ(1 − φ)(λ − 1)c
L δ,o (c , δ) = , (6.3)
(1 − σδ )λ(2δ + ρ )
∂L (c ,δ) ∂L (c ,δ)
where δ,∂o c > 0 and δ,∂δo
> 0.
Finally, we use the equilibrium expression for n B from (2.18), L δ,o from (6.3), and the relationship between I and c from
the R&D condition (2.19) to express the labor market clearing condition in terms of δ and I ,

1 1 I δ (1 − φ)[κδ δ + 1]
= I + (1 − σ I )(ρ + I ) + [LMCC] (6.4)
κI φ(λ − 1) I + δ φ(1 − σδ )(2δ + ρ )
We establish in the Appendix that, as long as diffusion is sufficiently difficult (κδ sufficiently large), (6.4) is strictly down-
ward sloping in (δ, I ) space. Thus, an increase to the diffusion rate always requires more total labor resources devoted to
advertising, and the labor market clearing condition under combative advertising continues to reflect a trade-off between
the resources allocated towards innovation and diffusion.
Our second equilibrium condition is derived from the equilibrium determination of the diffusion rate based on the
optimized advertising expenditures of old and young firms. As before, the optimality condition for young firm expenditure
on diffusion promoting advertising is (1 − σδ ) L δ, y = δ( V a − V y ). Combining this with the optimality condition for defensive
advertising, (1 − σδ ) L δ,o = δ V o , using (2.14), (2.16), (6.1) and (6.2) yields the following equilibrium diffusion curve (DC) that
captures relative advertising incentives,
ρ + δ(2 − ρκδ )
I= [DC] (6.5)
κδ δ − 1
When graphed in (δ , I ) space, this diffusion curve is strictly downward sloping. This is because a greater innovation rate
does not directly impact old firms’ incentives to invest in defensive advertising. However, young firms’ advertising incentives
remain strictly decreasing in I since a greater rate of innovation still reduces the reward from successful diffusion (V a − V y ).
Note that the equilibrium determination of δ under combative advertising is now entirely independent of the equilibrium
consumption level, c. Although greater equilibrium consumption increases the reward from successful diffusion, thereby
incentivizing young firms to increase their advertising expenditure, it generates a proportional increase in the value of
serving the market as an old firm. The corresponding increase in defensive advertising exactly offsets the effect of increased
advertising by young firms. The net effect is an increase in total advertising volume, without changing the diffusion rate.
In the following proposition, we establish the existence and uniqueness of the model’s balanced growth equilibrium.
The equilibrium is illustrated in Fig. 5. Furthermore, we show that the inclusion of combative advertising does not change
the model’s predictions for the effect of subsidies to R&D. Just as in the informative advertising specification, subsidizing
R&D impacts advertising incentives only indirectly through the general equilibrium effect of the associated increase in
the innovation rate. The higher replacement rate of adult firms implied by the increase in innovation decreases young firm
advertising incentives. Since defensive advertising incentives do not depend on the innovation rate, this results in a decrease
17
To see the complete derivation of (6.2), first note the no-arbitrage condition for the old firm: r (t ) V o (t )dt = (1 − φ)πa (t )dt − δ V o (t )dt + (1 − δ(t )dt ) V˙o dt −
(1 − σδ ) L δ,o (t )dt. Substituting for δ using (6.1) and maximizing the right hand side of the resulting expression with respect to L δ,o gives the optimality
condition for old firm advertising, (1 − σδ ) L δ,o (t ) = δ(t ) V o (t ). Substituting this expression back into the no-arbitrage condition for the old firm gives (6.2).
17
Fig. 5. Combative advertising equilibrium.
in the equilibrium diffusion rate. As in the informative advertising specification, subsidizing R&D has an ambiguous effect
on economic growth when advertising is combative.18
Proposition 2. The model with combative advertising has a unique equilibrium with I > 0 and δ > 0 under the following parameter
restriction,
2 φ(λ − 1)
<ρ < .
κδ (1 − σ I )κ I
Furthermore, subsidizing R&D investment decreases the diffusion rate, increases the innovation rate, increases the product failure rate,
∂I ∂f ∂g
and has an ambiguous effect on economic growth. That is, ∂∂δ
σ < 0, ∂ σ > 0, ∂ σ > 0, and ∂ σ >< 0.
I I I I
Proof. See Appendix.
7. Advertising policy
In this section, we examine the economic impact of advertising policy in both the informative and combative advertising
versions of the model. We focus on the use of an advertising subsidy or tax as the sole policy instrument available in order
to directly compare our findings with the analysis of R&D policy in Sections 4 and 5.19
In the informative advertising version of the model, subsidizing advertising expenditure impacts the relative incentive
to invest in R&D and advertising without changing the labor resource requirement of either type of investment. To see
this more formally, recall that the labor market clearing condition of (2.21) equates the fixed supply of labor to total labor
demand across the three market activities that require labor. In order to express this condition in (δ, I ) space, we used the
optimal advertising condition (2.20) to eliminate c. The resulting LMCC of (2.23) depends upon the advertising subsidy only
because it embeds the condition for optimal advertising. Equivalently, we can instead use the free-entry condition, equation
(2.19), to eliminate c from (2.21). This results in the following alternative expression of the LMCC in (δ, I ) space.
(1 − σ I )κ I Iδ
1= (ρ + I ) + κ I I + κδ [Alt LMCC] (7.1)
φ(λ − 1) I +δ
The alternate LMCC remains strictly downward sloping in (δ, I ) space, but no longer directly depends upon σδ . Using (7.1)
allows us to neatly illustrate the impact of an advertising subsidy in panel (a) of Fig. 6. Subsidizing advertising expenditure
reduces the cost of advertising relative to R&D, and shifts relative investment incentives away from R&D towards advertising.
Consequently, the RDAC shifts rightward. This induces movement along the alternate LMCC, as fewer labor resources are
devoted to R&D. The end result is an increase in the rate of diffusion, a decrease in the innovation rate, and a decrease in
the rate of product failure. Since the change to I and δ have opposite signs, and the overall change to economic growth g
is ambiguous in general.
In the combative advertising version of the model, the subsidy reduces the cost of both diffusion promoting and defensive
advertising proportionally. Consequently, there is no direct change to the relative advertising incentives of young and old
18
In this paper’s supplementary material (Section S.2), we derive the welfare properties of the model and examine the optimal R&D policy numerically.
Numerical simulations confirm that the quantitative effects of R&D subsidies are very similar in the informative and combative advertising frameworks.
19
We explore the joint use of advertising and R&D policy in Section S.3 of our paper’s supplementary material.
18
Fig. 6. Advertising subsidy.
firms captured by the DC, equation (6.5). However, defensive advertising intensity determines the difficulty of diffusion in
the combative advertising framework. Since the cost reduction stimulates defensive advertising, the labor resources required
to maintain a constant rate of diffusion δ increase. As depicted in panel (b) of Fig. 6, this effect generates a leftward shift in
the LMCC, equation (6.4), as fewer labor resources are available for R&D at any δ . This effectively tightens the labor resource
constraint in (δ , I ) space. It is in this sense that the primary effect of an advertising subsidy in the combative advertising
model is a socially wasteful increase in resources devoted to advertising. However, this resource reallocation does indirectly
increase the relative advertising incentives of young firms since the associated decrease in I implies a longer expected
reign as an adult firm if they diffuse successfully. This general equilibrium effect is captured by the movement along the
downward sloping DC curve as the LMCC shifts left in response to the subsidy in panel (b). Thus, advertising subsidies
have the same qualitative impact on the model’s endogenous variables in both the informative and combative advertising
specifications. This result is summarized in the following proposition,
Proposition 3. In both the informative and combative advertising models, subsidizing advertising expenditure increases the diffusion
rate, decreases the innovation rate, decreases the product failure rate, and has an ambiguous effect on economic growth. That is,
∂δ ∂I ∂f ∂g
∂ σ > 0, ∂ σ < 0, ∂ σ < 0, and ∂ σ >< 0.
δ δ δ δ
Despite this qualitative equivalence, the growth and welfare effects of subsidizing advertising can be markedly different
when advertising is informative or combative. To illustrate, we turn to numerical simulations and examine the effects
of advertising policy in our benchmark equilibrium with f = 0.25 and g = 1.5% in the absence of an advertising policy
(σδ = 0).20 We begin by examining the impact of implementing an advertising policy that ranges from a 15% tax to a
15% subsidy and report results in Fig. 7. Since both model versions are calibrated to a common benchmark equilibrium,
they exhibit the same values of δ , I , f and g when σδ = 0. Initial consumption expenditure is lower when advertising is
combative, since more total labor resources are devoted to advertising in the presence of defensive advertising by old firms.
Consequently, initial welfare is also lower in the combative advertising model.
As stated in Proposition 3, subsidizing advertising expenditure increases the equilibrium diffusion rate, decreases the
innovation rate, and decreases the failure rate in both versions of the model. However, we find that economic growth and
welfare are monotonically increasing in the subsidy when advertising is purely informative, and monotonically decreasing
in the subsidy when advertising is combative. The primary reason for this difference is apparent from panel (a); subsidizing
advertising is far more effective at increasing the diffusion rate in the informative advertising version of the model. In this
case, the subsidy stimulates diffusion promoting advertising by young firms and the difficulty of diffusion is constant by
construction. As a result, the subsidy generates a large increase in the diffusion rate that overcomes the decrease in I and
economic growth increases. Although there is static welfare loss from a reduced proportion of labor in manufacturing and
lower consumption, this is dominated by the dynamic welfare gain of increased economic growth.
In the combative advertising model, the subsidy also stimulates defensive advertising by old firms. This increases the
difficulty of diffusion and exactly offsets the decrease in the expected cost of diffusion from the subsidy. That is, there is
20
Note that both versions of the model feature the same set of parameters, {ρ , λ, φ, κ I , κδ }. To calibrate the combative advertising model to the bench-
mark equilibrium, we continue to set ρ = 0.06, λ = 1.25 and φ = 0.2 as described in Section 4, and jointly calibrate κ I and κδ to match the f = 0.25 and
g = 1.5% targets. This results in κ I = 0.289 and κδ = 17.09.
19
Fig. 7. Effect of advertising subsidy/tax.
no change to relative advertising incentives as captured by the DC in equation (6.5). The diffusion rate increases modestly
only because the amount of labor resources available for R&D decreases with the subsidy and the associated decrease in
the innovation rate increases the relative advertising incentives of young firms. However, this small increase in diffusion is
insufficient to offset the decrease in innovation, and economic growth declines. This dynamic welfare loss is coupled with a
static welfare loss from lower consumption.
Indeed, we generally find that welfare is monotonically increasing (decreasing) in the subsidy in the informative (com-
bative) advertising models. That is, this relationship holds across the entire feasible range of advertising policy in the
benchmark equilibrium and more broadly across plausible alternative initial equilibria. As a result, we find that the optimal
advertising policy is the maximum subsidy (tax) available to the policy maker if advertising is informative (combative).21 To
better understand the source of this finding, we examine the impact of advertising policy at these corner solutions in the
benchmark equilibrium. Results are reported in Table 3.
In the informative advertising case, the optimal advertising subsidy approaches the maximum subsidy consistent with a
positive and finite I and δ . The associated upper bound of σδ follows directly from relative R&D and advertising incentives
as captured by the RDAC in equation (2.22). More specifically, the parameter restriction of Assumption 1 implicitly defines
this upper bound as
φ κδ − (1 − φ)(1 − σ I )κ I
σδsup = , where lim δ = ∞. (7.2)
φ κδ σδ →σδsup
Approaching this bound is possible in equilibrium with finite labor resources because the proportion of industries in which
young firms use resources for advertising (n B ) also approaches zero. That is, L δ = n B κδ δ = κδ I δ/( I + δ), which implies
21
We note that initial equilibria do exist where it becomes optimal to tax informative advertising and subsidize combative advertising. Specifically, because
advertising subsidies always decrease the innovation rate, we find that it can be optimal to tax informative advertising when the model’s parameters are
set such that the initial innovation rate is extremely low. Similarly, it can be optimal to subsidize combative advertising when the initial innovation rate is
extremely high. However, our numerical analysis indicates that these results do not arise under plausible initial conditions.
20
Table 3
Optimal advertising policy.
(a): Informative g I δ c f U Lδ
σδ = 0 1.500 0.090 0.269 1.092 0.250 2.106 0.100
σδ → σδsup = 0.214 1.902 0.085 ∞ 1.061 0.000 2.544 0.127
(b): Combative g I δ c f U Lδ
σδ = 0 1.500 0.090 0.269 1.082 0.250 1.940 0.109
σδ → −∞ 1.545 0.108 0.194 1.211 0.356 4.029 0.000
that L δ → I κδ as δ → ∞. As seen in Table 3, panel (a), an advertising subsidy at this level increases the labor resources
used in advertising modestly and the static cost of the resulting decrease in consumption expenditure is dominated by the
substantial dynamic gain from the large increase in economic growth.
In the combative case, the optimal advertising policy approaches as infinitely large tax, σδ → −∞. Although the policy
makes advertising prohibitively costly, recall that relative advertising incentives of old and young firms, as captured by the
DC, do not depend on advertising policy. Since δ is defined in (6.1) in terms of the ratio of young to old firms advertising,
the diffusion rate does not go to zero as a result of the advertising tax. Intuitively, even though the cost of employing
labor resources in diffusion promoting advertising becomes infinite with the tax, the expected labor required to successfully
diffuse a prototype goes to zero as old firms cease defensive advertising. As seen in Table 3, panel (b), the result of the
tax is that the total labor resources used in advertising approaches zero with only a relatively modest decrease in the
diffusion rate. By eliminating the resources used in combative advertising, the tax frees up resources for innovation and the
manufacture of consumption goods, increasing both I and c. The end result is both a static welfare gain from the increase
in consumption and a dynamic gain from increased growth.
To be sure, these corner solutions for the optimal advertising policy should be interpreted cautiously. They necessarily
depend on the model’s stylized representation of advertising and the resulting limit properties of the model. Nevertheless,
our general result that the welfare implications of advertising policy hinge on whether advertising is informative or com-
bative is consistent with conclusions from the long-standing literature on the economic effects of advertising. Our analysis
shows that this same pattern of results arises in a dynamic endogenous growth framework where advertising and R&D
investment decisions interact.
8. Conclusion
In this paper, we analyze the dynamic interaction between product innovation, diffusion, and economic growth. We con-
tribute to the Schumpeterian growth literature by introducing a stochastic diffusion process in which the rate of commercial
success of product innovations is determined by advertising intensity. Through this mechanism, firms endogenously cycle
through distinct life stages of stochastic length as new innovations arrive and either commercialize successfully or fail. Un-
like traditional quality ladder models, our framework is consistent with empirical evidence that (1) firms devote substantial
resources to advertising innovative products, (2) a large proportion of product launches fail despite these advertising efforts,
and (3) even when ultimately successful, most new products experience an initial period of low market penetration before
an eventual sales takeoff to their mature market share.
In this framework, economic growth depends positively on both the rate of innovation and the rate of product diffusion
since only product innovations that commercialize successfully enter the mainstream market. In contrast to traditional mod-
els where R&D subsidies always promote growth, we find that R&D subsidies can have a non-monotonic effect on growth
by shifting incentives towards R&D but away from the complementary advertising investment needed for innovation dif-
fusion. In particular, we show that this effect gives rise to an inverted U-shaped relationship between R&D subsidies and
both economic growth and welfare. We find that this relationship holds both when advertising is purely informative and in
the presence of defensive advertising by incumbent firms. Furthermore, our simulation exercises illustrate that the optimal
R&D policy is highly sensitive to the initial failure rate of newly innovated products, even though the growth rate is held
constant. While we find that a large subsidy is optimal when the initial failure rate is very low, the optimal policy shifts to
a modest tax when the initial failure rate is high, but still within the empirically plausible range. In general, the scope for
using R&D subsidies to improve welfare diminishes with the failure rate because the social benefit of stimulating innovation
is lower when a smaller proportion of new innovations succeed. We argue that these results suggest that standard endoge-
nous growth models that assume costless and instantaneous innovation diffusion may overstate the case for large subsidies
to R&D.
In addition, we use the model to investigate the economic impact of a targeted advertising policy. When advertising
is purely informative, we show that subsidizing advertising expenditure generates a large increase in the diffusion rate,
which reduces the rate of new product failure, boosts economic growth and improves welfare. However, when advertising is
combative, much of the increase in diffusion promoting advertising by young technological leaders is offset by the increase
in defensive advertising by incumbent firms. The net result is a decrease in economic growth and welfare as there is
a substantial increase in resources devoted to advertising, without the expected large increase in diffusion. Our analysis
21
highlights the importance of understanding the responsiveness of different types of advertising to changes in advertising
policy.
Appendix A
A.1. Growth rate derivation
At any point in time, an n A proportion of industries are served by an adult firm whose product enjoys a perceived
quality of k(ω, t ) with all consumers. In the remaining (1 − n A ) proportion of B industries, old firms serve 1 − φ proportion
of mainstream consumers that purchase a product of perceived quality at the k(ω, t ) quality standard, while young firms
serve φ proportion of early adopters that purchase a prototype they perceive to be one step up the λ quality ladder. All
products have a common price of p = λ. This gives the following expression for instantaneous per capita sub-utility,
c (t )λk(ω,t ) c (t )λk(ω,t ) c (t )λk(ω,t )+1
ln(u (t )) = ln dω + ln dω + ln dω (A.1)
λ λ λ
nA (1−n A )(1−φ) (1−n A )φ
1
= ln(c (t )) + ln λk(ω,t )−1 dw + lnλdω
0 (1−n A )φ
1
= ln(c (t )) − ln(λ) + (1 − n A )φln(λ) + ln(λk(ω,t ) )dω
0
= ln(c /λ) + (1 − n A )φln(λ) + ln(λ)(1 − f ) It (A.2)
where the last line follows since I (1 − f ) is the expected aggregate rate of progress up the quality ladder. Differentiating
ln(u (t )) gives the rate of utility growth
u̇ Iδ
g≡ = ln(λ) I (1 − f ) = ln(λ) (A.3)
u I +δ
A.2. Proof of Proposition 2
First, we prove the existence and uniqueness of an equilibrium with I > 0 and δ > 0 by establishing a single crossing of
the LMCC displayed in (6.4) and the DC displayed in (6.5). As noted in the main text the DC is strictly downward sloping in
(δ, I ) space. Plugging I = 0 into the DC gives its horizontal intercept, δmax , where
ρ 1
δmax = > > 0, (A.4)
ρκδ − 2 κδ
since the parameter restriction in Proposition 2 implies that ρκδ > 2. Next, note that the DC curve implies an asymptote of
ρ
I → ∞ as δ → 1/κδ from the right. Thus, the DC is well defined with 0 < I < ∞ for δ ∈ ( κ1 , ρκ −2 ).
δ δ
∂ h(δ, I ) ∂ h(δ, I )
Let h(δ, I ) denote the right hand side of the LMCC as written in (6.4). Clearly, ∂ I > 0. This implies that, if ∂δ > 0,
then the LMCC is strictly downward sloping in (δ, I ) space. Differentiating, we have
∂ h(δ, I ) (1 − σ I )(1 − φ)(ρ + I ) I

= I (2δ + ρ )(κδ δ + 1) + δ( I + δ)(κδ ρ − 2) (A.5)
∂δ (1 − σδ )φ( I + δ) (2δ + ρ )
2 2
Since the parameter restriction in Proposition 2 ensures that ρκδ > 2, we have that ∂ h∂δ
(δ, I )
> 0, and the LMCC is strictly
downward sloping. To establish the single crossing, we show that the rate of innovation implicitly defined by the LMCC is
positive and finite over the domain δ ∈ (0, ∞). When δ → 0, the LMCC gives
1 (1 − σ I )(ρ + I max )
= I max + , (A.6)
κI φ(λ − 1)
and when δ → ∞, we have that
1 1 (1 − φ)κδ
= I min + (1 − σ I )(ρ + I min ) + . (A.7)
κI φ(λ − 1) 2φ(1 − σδ )
φ(λ−1)
The parameter restriction in Proposition 2 ensures that ρ< (1−σ I )κ I , which implies that 0 < I min < I max < ∞. Therefore,
single crossing obtains as illustrated in Fig. 5.
22
To obtain the rest of Proposition 2, observe that σ I enters only the LMCC. This is because the LMCC represents the
resource trade-off between I and δ , after taking into account the requisite resources devoted to consumption to align R&D
incentives. Through the free-entry condition, subsidizing R&D implies that a smaller c is associated with any level of I in
equilibrium. Thus, an increase to σ I effectively slackens the labor resource constraint in (δ, I ) space. This shifts the LMCC to
the right in Fig. 5, generating movement along the downward sloping DC as young firms respond to the higher innovation
rate by decreasing advertising expenditure. As a result, the innovation rate increases and the diffusion rate decreases, just
as in the informative advertising case.
A.3. Decomposing the welfare effects of an R&D subsidy
In this section, we derive the expression for the decomposition of welfare effects from an R&D subsidy displayed in
equation (4.2) of the main text. As discussed in the main text, we consider an augmented social planner problem in which
market advertising incentives enter as an additional constraint. The Lagrangian associated with the augmented social plan-
ner’s problem is
c
L S B (c , I , δ, μ, η) = ρ U + μ 1 − − κ I I − κδ δn B + η H (c , I ) − δ , (A.8)
λ
where η is the Lagrange multiplier associated with the advertising incentives constraint and H (c , I ) is defined in equation
(4.1). First order conditions with respect to I , δ and c respectively provide
∂ n B ln(λ) φln(λ) ∂nB ∂ H (c , I )

L IS B = 0 ⇒ δ + − μ κ I + κδ δ +η = 0, (A.9)
∂I ρ δ ∂I ∂I
∂ n B ln(λ) φln(λ) ∂n
B
LδS B = 0 ⇒ I − − μ κδ I − η = 0, (A.10)
∂δ ρ I ∂δ
1 ∂ H (c , I ) 1
LcS B = 0 ⇒ +η − μ = 0. (A.11)
c ∂c λ
Using (A.10) and (A.11), we have
cln(λ) c φln(λ)
λρ − λI − κδ
η= . (A.12)
c ∂ H (c , I ) 1
λ κδ λ ∂ c − ∂n
I ∂δB
Note that the numerator is exactly equal to the social benefit minus the social cost of a marginal increase in advertising
investment from the first-best solution, LδF B as displayed in equation (3.4) in Section 3. Substituting (A.11) and (A.12) into
(A.9) yields,
∂ n B c δln(λ) c φln(λ) ∂nB

L IS B = + − κ I + κδ δ
∂I ρλ λ ∂I

L IF B

∂ H (c , I ) ∂ n B nB
∂I ∂δ − λ κ I + κδ δ ∂∂nIB ∂ H∂(cc, I ) ∂∂δ cln(λ) c φln(λ)
+ · − − κδ . (A.13)
λκδ ∂ H∂(cc, I ) ∂∂δ
nB
− 1I λρ λI

L FB
δ
sδ
Appendix B. Supplementary material
Supplementary material related to this article can be found online at https://doi.org/10.1016/j.red.2022.05.001.
References
Agarwal, R., Bayus, B.L., 2002. The market evolution and sales takeoff of product innovations. Management Science 48 (8), 1024–1041.
Aghion, P., Akcigit, U., Howitt, P., 2014. What do We Learn from Schumpeterian Growth Theory? Handbook of Economic Growth, vol. 2. Elsevier, pp. 515–563.
Aghion, P., Howitt, P., 1992. A model of growth through creative destruction. Econometrica 60 (2), 323–351.
Argente, D., Fitzgerald, D., Moreira, S., Priolo, A., 2021. How do firms build market share? Working Paper.
Arkolakis, C., 2010. Market penetration costs and the new consumers margin in international trade. Journal of Political Economy 118 (6), 1151–1199.
Asplund, M., Sandin, R., 1999. The survival of new products. Review of Industrial Organization 15 (3), 219–237.
Bagwell, K., 2007. The economic analysis of advertising. In: Handbook of Industrial Organization, vol. 3, pp. 1701–1844.
Bornstein, G., et al., 2018. Entry and profits in an aging economy: the role of consumer inertia. Working Paper.
Butters, G.R., 1977. Equilibrium distributions of sales and advertising prices. The Review of Economic Studies 44 (3), 465–491.
23
Castellion, G., Markham, S.K., 2013. Perspective: new product failure rates: influence of argumentum ad populum and self-interest. The Journal of Product
Innovation Management 30 (5), 976–979.
Cavenaile, L., Celik, M.A., Roldan-Blanco, P., Tian, X., 2021. Style over substance? Advertising, innovation, and endogenous market structure. Working Paper.
Cavenaile, L., Roldan-Blanco, P., 2021. Advertising, innovation and economic growth. American Economic Journal: Microeconomics 13 (3), 251–303.
Chandrasekaran, D., Tellis, G.J., 2007. A critical review of marketing research on diffusion of new products. In: Review of Marketing Research. Emerald Group
Publishing, Limited, pp. 39–80.
Chiesa, V., Frattini, F., 2011. Commercializing technological innovation: learning from failures in high-tech markets. The Journal of Product Innovation
Management 28 (4), 437–454.
Chu, A.C., Cozzi, G., 2018. Effects of patents versus R&D subsidies on income inequality. Review of Economic Dynamics 29, 68–84.
Chu, A.C., Cozzi, G., Fan, H., Furukawa, Y., Liao, C.-H., 2019. Innovation and inequality in a monetary Schumpeterian model with heterogeneous households
and firms. Review of Economic Dynamics 34, 141–164.
Chu, A.C., Furukawa, Y., 2013. Patentability and knowledge spillovers of basic R&D. Southern Economic Journal 79 (4), 928–945.
Cozzi, G., Galli, S., 2014. Sequential R&D and blocking patents in the dynamics of growth. Journal of Economic Growth 19 (2), 183–219.
Czarnitzki, D., Hanel, P., Rosa, J.M., 2011. Evaluating the impact of R&D tax credits on innovation: a microeconometric study on Canadian firms. Research
Policy 40 (2), 217–229.
Davis, L.S., Şener, F., 2012. Private patent protection in the theory of Schumpeterian growth. European Economic Review 56 (7), 1446–1460.
Dinlersoz, E.M., Yorukoglu, M., 2012. Information and industry dynamics. The American Economic Review 102 (2), 884–913.
Dinopoulos, E., Grieben, W.-H., Şener, F., 2021. The impact of economic policies on long-run unemployment and Schumpeterian growth. Working Paper.
Dinopoulos, E., Segerstrom, P., 1999. A Schumpeterian model of protection and relative wages. The American Economic Review 89 (3), 450–472.
Dinopoulos, E., Sener, F., 2007. New directions in Schumpeterian growth theory. In: Hanush, H., Pyka, A. (Eds.), Elgar Companion to Neo-Schumpeterian
Economics. Elgar Press.
Dinopoulos, E., Syropoulos, C., 2007. Rent protection as a barrier to innovation and growth. Economic Theory 32 (2), 309–332.
Dinopoulos, E., Waldo, D., 2005. Gradual product replacement, intangible-asset prices and Schumpeterian growth. Journal of Economic Growth 10 (2),
135–157.
Dixit, A., Norman, V., 1978. Advertising and welfare. Bell Journal of Economics, 1–17.
Einav, L., Klenow, P.J., Levin, J.D., Murciano-Goroff, R., 2021. Customers and retail growth. Working Paper.
Eliaz, K., Spiegler, R., 2011. Consideration sets and competitive marketing. The Review of Economic Studies 78 (1), 235–262.
Fitzgerald, D., Haller, S., Yedid-Levi, Y., 2016. How exporters grow. Working Paper.
Furukawa, Y., Lai, T.-k., Sato, K., 2019. Love of novelty: a source of innovation-based growth... or underdevelopment traps? Working Paper.
Furukawa, Y., Lai, T.-K., Sato, K., 2020. Novelty-seeking traits and applied research activities. Applied Economics Letters 27 (11), 945–950.
Goeree, M.S., 2008. Limited information and advertising in the US personal computer industry. Econometrica 76 (5), 1017–1074.
Golder, P.N., Tellis, G.J., 2004. Growing, growing, gone: cascades, diffusion, and turning points in the product life cycle. Marketing Science 23 (2), 207–218.
Gourio, F., Rudanko, L., 2014. Customer capital. The Review of Economic Studies 81 (3), 1102–1136.
Gourville, J.T., 2006. Eager sellers & stony buyers. Harvard Business Review 84 (6), 98–106.
Grossman, G.M., Helpman, E., 1991. Quality ladders in the theory of growth. The Review of Economic Studies 58 (1), 43–61.
Grossman, G.M., Shapiro, C., 1984. Informative advertising with differentiated products. The Review of Economic Studies 51 (1), 63–81.
Grossmann, V., 2008. Advertising, in-house R&D, and growth. Oxford Economic Papers 60 (1), 168–191.
Impullitti, G., 2010. International competition and US R&D subsidies: a quantitative welfare analysis. International Economic Review 51 (4), 1127–1158.
Klein, M.A., 2020. Secrecy, the patent puzzle and endogenous growth. European Economic Review 126, 103445.
Mahajan, V., Muller, E., Srivastava, R.K., 1990. Determination of adopter categories by using innovation diffusion models. Journal of Marketing Research 27
(1), 37–50.
Minniti, A., Parello, C.P., Segerstrom, P.S., 2013. A Schumpeterian growth model with random quality improvements. Economic Theory 52 (2), 755–791.
Perla, J., 2019. A model of product awareness and industry life cycles. University of British Columbia. Unpublished Paper.
Rachel, L., 2021. Leisure-enhancing technological change. Working Paper.
Rogers, E.M., 1962. Diffusion of Innovations. Free Press of Glencoe, New York.
Segerstrom, P.S., 1998. Endogenous growth without scale effects. The American Economic Review, 1290–1310.
Segerstrom, P.S., 2007. Intel economics. International Economic Review 48 (1), 247–280.
Segerstrom, P.S., Anant, T.C., Dinopoulos, E., 1990. A Schumpeterian model of the product life cycle. The American Economic Review, 1077–1091.
Şener, F., 2008. R&D policies, endogenous growth and scale effects. Journal of Economic Dynamics and Control 32 (12), 3895–3916.
Shum, M., 2004. Does advertising overcome brand loyalty? Evidence from the breakfast-cereals market. Journal of Economics & Management Strategy 13
(2), 241–272.
Steenkamp, J.-B.E., Ter Hofstede, F., Wedel, M., 1999. A cross-national investigation into the individual and national cultural antecedents of consumer
innovativeness. Journal of Marketing 63 (2), 55–69.
Stegeman, M., 1991. Advertising in competitive markets. The American Economic Review, 210–223.
Svensson, R., 2013. Publicly-funded R&D programs and survival of patents. Applied Economics 45 (10), 1343–1358.
Tellis, G.J., Yin, E., Bell, S., 2009. Global consumer innovativeness: cross-country differences and demographic commonalities. Journal of International Mar-
keting 17 (2), 1–22.
Victory, K., Nenycz-Thiel, M., Dawes, J., Tanusondjaja, A., Corsi, A.M., 2021. How common is new product failure and when does it vary? Marketing Letters 32
(1), 17–32.
Yang, Y., 2018. On the optimality of IPR protection with blocking patents. Review of Economic Dynamics 27, 205–230.
Yohn, D.L., 2019. Why great innovation needs great marketing. Harvard Business Review.
24

1 s2.0 S1094202522000266 Main

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

1 s2.0 S1094202522000266 Main

Uploaded by

Copyright:

Available Formats

JID:YREDY AID:1136 /FLA [m3G; v1.316] P.

Contents lists available at ScienceDirect

Review of Economic Dynamics

Product innovation, diffusion and endogenous growth

1.1. Related literature

2.1. Households and perceived quality

2.2. Innovation, diffusion, and industry structure

2.3. Labor and production

L I (t ) = κ I I (t ), L δ (t ) = n B (t ) L δ, y (t ) = n B (t )κδ δ(t ). (2.11)

2.4. Stock market valuations and optimal advertising

r (t ) V y (t )dt = φ πa (t )dt − (1 − σδ ) L δ, y (t )dt + δ(t )( V a (t ) − V y (t ))dt −

Assumption 2. I max > I min , as deﬁned in (2.24) and (2.25) respectively.

ln(u (t )) = ln(c /λ) + n B φln(λ) + ln(λ)(1 − f ) It . (2.26)

Fig. 2. R&D subsidy.

4.1. Welfare effects of R&D subsidies

L IS B = L IF B + sδ · LδF B = 0, where sδ ≡  . (4.2)

5.1. Optimal R&D subsidies

Fig. 3. Impact of R&D subsidies.

g (%) I δ c f U L IS B L IF B sδ LδF B σ I, g σ I,U

5.2. Alternate initial failure rates

(a): f 0 = 0.01 g (%) I δ c f U L IS B L IF B sδ LδF B σ I, g σ I,U

Fig. 4. Optimal R&D Subsidies.

Fig. 5. Combative advertising equilibrium.

Proof. See Appendix.

Fig. 6. Advertising subsidy.

Fig. 7. Effect of advertising subsidy/tax.

A.1. Growth rate derivation

A.2. Proof of Proposition 2

∂ h(δ, I ) (1 − σ I )(1 − φ)(ρ + I ) I 

A.3. Decomposing the welfare effects of an R&D subsidy

∂ n B ln(λ) φln(λ)  ∂nB  ∂ H (c , I )

∂ n B c δln(λ) c φln(λ)  ∂nB 

Appendix B. Supplementary material

Supplementary material related to this article can be found online at https://doi.org/10.1016/j.red.2022.05.001.

You might also like

L IS B = L IF B + sδ · LδF B = 0, where sδ ≡ . (4.2)

g (%) I δ c f U L IS B L IF B sδ LδF B σ I, g σ I,U

(a): f 0 = 0.01 g (%) I δ c f U L IS B L IF B sδ LδF B σ I, g σ I,U

∂ h(δ, I ) (1 − σ I )(1 − φ)(ρ + I ) I

∂ n B ln(λ) φln(λ) ∂nB ∂ H (c , I )

∂ n B c δln(λ) c φln(λ) ∂nB