Kung 2015

THE JOURNAL OF FINANCE • VOL. LXX, NO.
3 • JUNE 2015
Innovation, Growth, and Asset Prices

HOWARD KUNG and LUKAS SCHMID∗
ABSTRACT
We examine the asset pricing implications of a production economy whose long-term
growth prospects are endogenously determined by innovation and R&D. In equilib-
rium, R&D endogenously drives a small, persistent component in productivity that
generates long-run uncertainty about economic growth. With recursive preferences,
households fear that persistent downturns in economic growth are accompanied by
low asset valuations and command high-risk premia in asset markets. Empirically,
we find substantial evidence for innovation-driven low-frequency movements in ag-
gregate growth rates and asset market valuations. In short, equilibrium growth is
risky.
AN ECONOMY’S LONG-TERM GROWTH prospects reflect its innovative potential.

At a fundamental level, innovation is a key source of sustained growth in
aggregate productivity. Empirical measures of innovation, such as research
and development (R&D) expenditures, tend to be volatile and quite persistent.
Such movements affect the dynamics of growth. Indeed, in U.S. post-war data,
productivity growth exhibits long and persistent swings.1 Similarly, innovation-
driven growth waves associated with the arrival of new technologies, such
as telecommunication, computers, and the internet, to name a few, are well
documented.2 Stock prices reflect such changes in growth prospects. Moreover,
if agents fear that a persistent slowdown in economic growth will lower asset
prices, these movements will give rise to high-risk premia in asset markets.
In this paper, we develop a general equilibrium model of innovation and
R&D to link asset prices and aggregate risk premia to endogenous movements
∗ Howard Kung is at the London Business School. Lukas Schmid is at Duke University and the
University of California, Los Angeles. We thank the Acting Editor Dave Backus and two anonymous
referees as well as Hengjie Ai, Ravi Bansal, Gian Luca Clementi, John Coleman, Diego Comin,
Max Croce, Ian Dew-Becker, Bernard Dumas, Lorenzo Garlappi, João Gomes, Francois Gourio,
Leonid Kogan, Lars Lochstoer, Pascal Maenhout, Stavros Panageas, Dimitris Papanikolaou, Adri-
ano Rampini, Norman Schurhoff, Tom Tallarini, Amir Yaron, and Lu Zhang for helpful discussions.
We have also benefited from the comments of seminar participants at Bocconi, Boston University,
Carnegie Mellon University, Cornell University, Duke University, Federal Reserve Bank of San
Francisco, Imperial College, INSEAD, London Business School, London School of Economics, MIT,
Paris School of Economics, UCLA, and Universitat Pompeu Fabra, as well as from conference par-
ticipants at Western Finance Association, Society for Economic Dynamics, CEPR Macro Finance
Conference London, European Finance Association, Tel Aviv Finance Conference, Econometric
Society, American Finance Association, and NBER Spring Asset Pricing Meeting.
1 See, for example, Gordon (2010) and Jermann and Quadrini (2004).
2 For example, see Helpman (1998) and Jovanovic and Rousseau (2005).
DOI: 10.1111/jofi.12241
1001
1002 The Journal of FinanceR
in long-term growth prospects. Our setup has two distinguishing features.

First, we embed a stochastic model of endogenous growth based on industrial
innovation3 into an otherwise standard production economy. In this model, pro-
ductivity growth is endogenous and sustained by the creation of new patented
technologies through R&D. Patents represent an endogenous stock of intangi-
ble capital. Second, we assume that households have recursive preferences, so
that they care about uncertainty regarding long-term growth prospects.
When calibrated to match empirical evidence on innovation and long-run
economic growth, our model can quantitatively replicate key features of asset
returns in the data. In particular, our model rationalizes a sizeable equity pre-
mium and a low and stable risk-free interest rate. Moreover, our model predicts
a sizeable spread between the returns on physical capital and intangible cap-
ital, which is related to the value premium in the data. In short, we find that
equilibrium growth is risky.
We first show that in the model innovation and R&D endogenously drive a
small but persistent component in the growth rate of productivity. In our gen-
eral equilibrium setting, these low-frequency movements in productivity trig-
ger long and persistent swings in aggregate growth rates, such as consumption
and output, which we label growth cycles. Intuitively, shocks affect the incen-
tives to innovate, which in turn impact long-term growth prospects. Notably,
transitory shocks in this setting have long-lasting permanent effects through
the innovation channel and generate endogenous persistence in growth rates.
Thus, a bad temporary shock not only lowers the level of consumption and
cash flows today, but also depresses long-term growth rates. When agents have
recursive preferences, they are sensitive to both short-run and long-run uncer-
tainty about consumption growth. Growth cycles help rationalize sizeable risk
premia in asset markets, as agents fear that such prolonged slumps in economic
growth coincide with low asset valuations. Similarly, agents save for extended
low growth episodes, driving down the real interest rate. Furthermore, in the
model, physical capital is endogenously more exposed to predictable variation
in growth than intangible capital, which generates a sizeable value spread.
An innovation-driven persistent component in productivity growth provides
an equilibrium foundation of long-run risks in the spirit of Bansal and Yaron
(2004). More precisely, in our model, long-run productivity risks, in the sense
of Croce (2014), arise naturally in equilibrium. Furthermore, persistent move-
ments in expected productivity affect all aggregate growth rates and therefore
give rise to equilibrium long-run consumption risks and cash flow risks.
The model helps to identify economic sources of long-run risks in the data. In
particular, the model predicts that R&D and innovation are equilibrium deter-
minants of productivity growth. In line with the predictions of the model, we
provide novel empirical evidence that measures of innovation have significant
predictive power for aggregate growth rates including productivity, consump-
tion, and output growth at horizons of one to five years.
3 Here, we build on the seminal work of Romer (1990) and Grossman and Helpman (1991).
Innovation, Growth, and Asset Prices 1003
While predictability in growth rates is at the core of the long-run risk model,
empirical evidence regarding this channel is still limited. The model provides
novel theoretical and empirical support for the notion that movements in long-
term growth prospects are a significant source of priced risk in asset markets.
Moreover, our results suggest that extending macroeconomic models to account
for the endogeneity of innovation and long-term growth can make progress
toward an environment that jointly captures the dynamics of aggregate quan-
tities and asset markets. We therefore view stochastic models of endogenous
growth as a useful tool for macrofinance.4
Our paper is related to several strands of literature in asset pricing, economic
growth, and macroeconomics. The economic mechanisms driving the asset pric-
ing implications are similar to those in the consumption-based long-run risks
model of Bansal and Yaron (2004). We contribute to this literature by showing
that predictable movements in growth prospects are an equilibrium outcome
of stochastic models of endogenous growth and by providing novel empirical
evidence identifying economic sources of long-run risks.
A number of recent papers examine the link between technological growth
and asset prices. Garleanu, Panageas, and Yu (2012) model technological
progress as the arrival of large, infrequent technological innovations and
show that the endogenous adoption of these innovations leads to predictable
movements in consumption growth and expected excess returns. Garleanu,
Kogan, and Panageas (2012) examine the implications of the arrival of new
technologies for existing firms and their workers, and show that, in an
overlapping-generations model, innovation creates a systematic risk factor
labeled displacement risk. The asset pricing implications of displacement risk
are further examined in a model of heterogeneous workers and firms in Kogan,
Papanikolaou, and Stoffman (2012). Pástor and Veronesi (2009) explain bubble-
like behavior of stock markets in the 1990s by the arrival of new technologies.
While our model has implications for consumption dynamics and asset re-
turns that are related to these models, our approach is different but complemen-
tary. In the above models of technology adoption, the arrival of new technolo-
gies is assumed to be exogenous. In contrast, we examine the asset pricing and
growth implications of the endogenous creation of new technologies through
R&D, which leads to a distinct set of empirical predictions. Moreover, by em-
bedding a model of endogenous technological progress into a real business cycle
model, our paper provides a straightforward extension of the workhorse model
of modern macroeconomics.
In this respect, the paper is closer to recent attempts to address asset pric-
ing puzzles within versions of the canonical real business cycle model. Start-
ing with the habit-based models of Jermann (1998) and Boldrin, Christiano,
and Fisher (2001), recent examples, such as Tallarini (2000), Campanale, Cas-
tro, and Clementi (2008), Kuehn (2008), Kaltenbrunner and Lochstoer (2010),
4 In a companion paper, Kung (2015) shows that a similar mechanism coupled with imperfect
price adjustment quantitatively rationalizes many aspects of the term structure of interest rates
in a production economy.
and Papanikolaou (2011) explore endogenous long-run consumption risks in

real business cycle models with recursive preferences, while Gourio (2012,
2013) examines disaster risks. Particularly closely related are recent papers
by Croce (2014), Backus, Routledge, and Zin (2007, 2010), Gomes, Kogan,
and Yogo (2009), and Favilukis and Lin (2013a, 2013b), who examine the im-
plications of long-run productivity risk with recursive preferences for equity
market returns. However, while they specify long-run productivity risk exoge-
nously, our model shows how such risk arises endogenously through innovation.
Our cross-sectional return implications are related to Gala (2010), Kogan and
Papanikolaou (2010), and Lin (2012), who examine the effects of technological
progress on the cross-section of returns.
Methodologically, our paper extends recent work by Comin and Gertler (2006)
and Comin, Gertler, and Santacreu (2009). Building on the seminal work of
Romer (1990) and Grossman and Helpman (1991), these papers integrate in-
novation and the adoption of new technologies into a real business cycle model
and show that the resulting stochastic endogenous growth model features rich
movements at a lower-than-business-cycle frequency, which they label medium-
term business cycles. We contribute to this literature by linking medium-term
cycles to long-run risks and aggregate risk premia.
The paper is structured as follows. In Section I, we describe our benchmark
model. In Section II, we qualitatively explore the equilibrium growth and pro-
ductivity processes and relate them to a canonical real business cycle model.
In Section III, we quantitatively examine the asset pricing implications of our
benchmark model and detail a number of empirical tests. Section IV concludes.
I. Model
In our baseline framework, we embed a model of industrial innovation in
the tradition of Romer (1990) into a fairly standard macroeconomic model with
convex adjustment costs and Epstein-Zin preferences. In the model, rather than
assuming exogenous technological progress, sustained growth arises through
the accumulation of patented intermediate goods (henceforth, patents) that
facilitate the production of a final consumption good. New patents are created
through innovation, which requires investment in R&D, and can be stored.
Therefore, patents in this model represent an endogenous stock of intangible
capital.
We start by describing in detail the production sector and the innovation
process in our economy. We then present the household sector and define the
general equilibrium.
A. Production
The production process involves three sectors. The final consumption good is
produced in a perfectly competitive sector, namely, the final goods sector, using
physical capital, labor, and patents. Stationary shocks drive stochastic fluctua-
tions in the production of the final consumption good. Patents are produced in
the intangible goods sector, where firms have monopoly power due to product
differentiation. New patents are created by innovation through R&D in the
innovation sector, which is also perfectly competitive.
Absent patents, decreasing returns to physical capital in the production func-
tion would imply that growth ceases in the long run without an exogenous trend
component in the level of technology, which is a standard result from the Solow
growth model. In contrast, in our setup, acquiring patents from the intangible
sector facilitates production for a given stock of physical capital, allowing the
final goods firm to grow and thereby creating demand for individual patents.
That demand is met by the intangible sector, which earns profits from selling
patents to the final goods firm by charging a markup over its marginal costs.
Monopoly power is important as the associated profits provide rents for creat-
ing new patents. The innovation sector sells a newly developed patent to the
intangible sector at the competitive price, which in equilibrium is equal to the
present value of its profits.
Sustained growth is obtained in this economy because the demand for patents
of the final goods firm creates new profit opportunities in the intangible sec-
tor and thus raises the incentives to create new patents through innovation.
These new patents increase the efficiency of physical capital and thereby boost
investment, creating even more demand for patents.
A.1. Final Good Sector

There is a representative firm that uses capital Kt , labor Lt , and a composite
of patents Gt to produce final (consumption) goods according to the production
technology
ξ
Yt = (Ktα (t Lt )1−α )1−ξ Gt , (1)
where the composite Gt is defined as

1ν
Nt
ν
Gt ≡ Xi,t di (2)
0
and Xi,t is the quantity of patent i ∈ [0, Nt ]. Also, Nt is the measure of patents
in use at date t, α is the physical capital share, ξ is the intangible capital
share, and 1−ν 1
is the elasticity of substitution between patents with ν < 1. We
interpret Nt as the stock of intangible capital.
We introduce uncertainty into the model by means of an exogenous stochastic
process t affecting the level of output. Importantly, t is assumed to follow a
stationary Markov process by specifying that t = eat and at = ρat−1 + t , with
t ∼ N(0, σ 2 ) and ρ < 1. While t resembles labor-augmenting technology, it
does not represent measured productivity in our setting, as we discuss in more
detail below. Because of the stationarity of the forcing process, sustained growth
arises endogenously from the development of new patents. We describe how
new patents are developed by innovation below.
The firm’s objective is to maximize shareholder value. This can be formally

stated as
∞

max E0 Mt Dt , (3)
{It ,Lt ,Kt+1 ,Xi,t }t≥0,i∈[0,Nt ]
t=0
where the firm’s dividends are

Nt
Dt = Yt − It − Wt Lt − Pi,t Xi,t di. (4)
0
Here, Mt is the stochastic discount factor, It is investment in physical capital,

Wt is the wage rate, and Pi,t is the price per unit of patent i. The last term
captures the costs of buying patents at time t. Prices Pi,t are set by patent
producers in the intangible sector, while the stochastic discount factor and the
wage rate are determined in general equilibrium and are both taken as given
by the final goods firm.
In line with the literature on production-based asset pricing, we assume that
investment is subject to convex capital adjustment costs, so that the physical
capital stock evolves as

It
Kt+1 = (1 − δ)Kt +
Kt , (5)
Kt
where δ is the depreciation rate of physical capital and
(·) the capital adjust-
ment cost function.5
A.2. Intangible Goods Sector

Patents are produced in the intangible goods sector. Patent producers have
monopoly power. Given the demand schedules set by the final goods firm, mo-
nopolists producing the patents set the prices Pi,t in order to maximize their
profits i,t . Patent producers transform one unit of the final good into one unit
of their patented good. This fixes the marginal cost of producing one patent at
unity.
Formally, monopolists solve the following static profit maximization problem
each period
i,t ≡ max Pi,t · Xi,t (Pi,t ) − Xi,t (Pi,t ). (6)
Pi,t
The value Vi,t of owning exclusive rights to produce patent i is equal to the
present discounted value of the current and future monopoly profits, so that
Vi,t = i,t + (1 − φ)Et [Mt+1 Vi,t+1 ], (7)
ζ
It α1 It 1
5 We specify
(·) as in Jermann (1998),
Kt ≡ ζ Kt + α2 . Here, 1−ζ represents the elas-
ticity of the investment rate with respect to Tobin’s Q. The parameters α1 and α2 are set so that
there are no adjustment costs in the deterministic steady state.
where φ is the probability that a patent becomes obsolete. This asset price
is important in our model, as it provides the payoff to creating new patents
through innovation as we describe next. This highlights the importance of
monopoly power, as the associated profits provide the rents to innovation.
A.3. Innovation Sector

Innovators develop new patents used in the production of final output. They
do so by conducting R&D, using the final good as input at unit cost. These
newly developed patents can be sold to patent producers. Assuming free entry
and perfect competition, the price of a new patent equals its value to the patent
producer, namely, Vi,t .
We link the evolution of the intangible capital stock Nt to innovation as
follows:
Nt+1 = ϑt St + (1 − φ)Nt , (8)
where St denotes R&D expenditures (in terms of the final goods) and ϑt rep-
resents the productivity of the innovation sector, which is taken as exogenous
by the R&D sector. In a similar spirit as Comin and Gertler (2006), we assume
that this technology coefficient involves an externality effect
χ · Nt
ϑt = 1−η η
, (9)
St Nt
where χ > 0 is a scale parameter and η ∈ [0, 1] is the elasticity of new patents
with respect to R&D. This specification posits that concepts already discovered
make it easier to come up with new ideas, ∂ϑ/∂ N > 0, thus capturing posi-
tive spillovers of the aggregate stock of intangible capital as in Romer (1990),
and that R&D investment has decreasing marginal returns, ∂ϑ/∂ S < 0, cap-
turing a congestion effect that raises the cost of developing new products as the
aggregate level of R&D increases.6
B. Household
The household sector is standard. The representative household has Epstein-
Zin preferences defined over consumption:

θ
1θ
Ut = (1 − β)Ctθ + β(Et [Ut+1 ]) 1−γ
1−γ
, (10)
6 Similarly, this congestion externality can be thought of as giving rise to adjustment costs to
investment in intangible capital, that is, R&D. Below, we will see that the optimality condition for
R&D is ϑ1t = Et [Mt+1 Vt+1 ], equating the marginal cost of creating a new patent with its marginal
benefit. Absent the congestion externality, this boils down to 1 = Et [Mt+1 Vt+1 ], a result analogous
to Q-theory, in which case the absence of adjustment cost fixes marginal Q at unity.
where γ is the coefficient of relative risk aversion and ψ ≡ 1−θ

1
is the elasticity of
intertemporal substitution. When ψ = γ , the agent cares about news regarding
1
long-run growth prospects. We assume that ψ > γ1 , so that the agent has a
preference for early resolution of uncertainty and dislikes uncertainty about
long-run growth rates.
The household maximizes utility by participating in financial markets and
by supplying labor. Specifically, the household can take positions Zt in the
stock market, which pays an aggregate dividend Dt , and in the bond market
Bt . Accordingly, the budget constraint of the household is
Ct + Qt Zt+1 + Bt+1 = Wt Lt + (Qt + Dt )Zt + (1 + r f ,t )Bt , (11)
where Qt is the stock price, r f ,t is the risk-free rate, Wt is the wage, and Lt
denotes hours worked.
We assume that stocks are claims to all the production sectors, namely, the
final good sector, the intangible sector, and the innovation sector. Accordingly,
we define the aggregate dividend as the net payout from the production sector,
Nt
Dt = Dt + i,t di − St . (12)
0
C. Equilibrium and Asset Prices

We define an equilibrium for our economy in a standard way. In our setup,
there is one exogenous state variable, t , and two endogenous state variables,
the physical capital stock Kt and the intangible capital stock Nt . Given an
initial condition {0 , K0 , N0 } and the law of motion for the exogenous state
variable t , an equilibrium is a set of sequences of quantities and prices such
that (i) quantities solve producers’ and the household’s optimization problems
and (ii) prices clear markets. Moreover, we focus on a symmetric equilibrium
in which all patent producers make identical decisions. In the following, we
describe the most important equilibrium conditions; we defer the complete list
of all relationships characterizing the equilibrium to Appendix A.
The stochastic discount factor in the economy is given by
⎛ ⎞1−γ −θ
θ−1
Ct+1 ⎜ Ut+1 ⎟
Mt+1 = β ⎝ 1 ⎠
1−γ , (13)
Ct 1−γ
Et Ut+1
where the second term, involving continuation utilities, captures preferences

concerning uncertainty about long-run growth prospects. Optimality implies
the following asset pricing conditions:
Qt = Et [Mt+1 (Qt+1 + Dt+1 )], (14)

1
= Et [Mt+1 ]. (15)
1 + rt
In equilibrium, the representative agent holds the entire supply of equities,
while bonds are in zero net supply. The former is normalized to one (i.e.,
Zt = 1 ∀t).
Since the agent has no disutility for labor, she will supply her entire endow-
ment, which we normalize to unity, so that Lt ≡ 1.
The final good firm’s optimality conditions are mostly standard. Denoting
by qt =
1 the shadow value of physical capital, the first-order condition for
t
investment in physical capital is

1 Yt+1 It+1
1 = Et Mt+1 α(1 − ξ ) + qt+1 (1 − δ) − + qt+1
t+1 . (16)
qt Kt+1 Kt+1
On the other hand, the final goods firm’s demand for patent i is determined by
ξν −1
1−ξ Nt
ν−1
Pi,t = Ktα (t Lt )1−α ξ ν
Xi,t di Xi,t , (17)
0
where it takes the price Pi,t as given. In fact, Pi,t is set by the monopolistically
competitive producer of patent i. In a symmetric equilibrium, the monopolisti-
cally competitive characterization of the intangible goods sector à la Dixit and
Stiglitz (1977) implies
1
Xi,t ≡ Xt and Pi,t ≡ Pt = . (18)
ν
That is, each patent producer charges a markup 1ν > 1 over unit marginal cost,
so that its profits are

1
i,t ≡ t = − 1 Xt , (19)
ν
1−ξ ξν −1 1−ξ
1
with Xt = ξ ν Ktα (t Lt )1−α Nt . Profits depend positively on the ag-

gregate productivity shock t and thus are procyclical.
Discounted future profits on patents are the payoff to innovation. Thus, since
the R&D sector is competitive, the optimality condition for R&D investment
becomes
Et [Mt+1 Vt+1 ](Nt+1 − (1 − φ)Nt ) = St , (20)
which says that the expected sales revenues equal costs, or equivalently, at the
margin, ϑ1t = Et [Mt+1 Vt+1 ]. By pinning down the amount of R&D investment,
this condition is crucial in this model, as it ultimately determines the equilib-
rium growth rate of the economy. Importantly, the procyclicality of profits will
be reflected in the dynamics of R&D.
C.1. Resource Constraint

Final output is used for consumption, investment in physical capital, factor
inputs used in the production of patents, and R&D investment:
Yt = Ct + It + Nt Xt + St , (21)
1− 1ν
= Ct + It + Nt Gt + St , (22)
1− 1
where the last equality exploits the optimality conditions and the term Nt ν Gt
captures the costs of patent production. Given that ν < 1, reflecting monopolis-
tic competition, it follows that a growing intangible capital stock increases the
efficiency of patent production, as the costs fall with Nt growing.
C.2. Stock Market

Given our definition of stocks as claims to the net payout of all production
sectors, in the symmetric equilibrium the aggregate dividend becomes
Dt = Dt + t Nt − St . (23)
Defining the stock market value to be the discounted sum of future aggre-
gate dividends and exploiting the optimality conditions, this value can be
rewritten as
Qt = qt Kt+1 + Nt (Vt − t )
∞

+ Et Mt+i+1 Vt+i+1 (Nt+i+1 − (1 − φ)Nt+i ) − St+i+1 , (24)
i=0
similar to Comin, Gertler, and Santacreu (2009). Thus, the stock market value
consists of (i) the current market value of the installed capital stock (first
term), (ii) the market value of currently used patents (second term), and (iii)
the market value of patents to be developed in the future (third term). There-
fore, the stock market values intangible capital and the option value of future
intangibles in addition to the tangible capital stock.
II. Equilibrium Growth Risk

In our benchmark model, sustained growth is an equilibrium phenomenon re-
sulting from agents’ decisions. In contrast to variants of the workhorse stochas-
tic growth model of dynamic macroeconomics, trend growth arises endoge-
nously from the accumulation of patents rather than from an exogenous drift
in productivity. In this section, we qualitatively examine the determinants of
equilibrium growth and its dynamics. Most importantly, we document that in
the model, movements in innovative activity generate predictable variation in
growth rates. In the context of Bansal and Yaron (2004), the model identifies a
novel source of long-run risks in the economy. In Section III, we quantitatively

evaluate the ability of the model to rationalize aggregate asset risk premia and
provide empirical evidence supporting this channel.
A. Endogenous Productivity
To start, it is convenient to represent the aggregate production function in
our benchmark model in a form that permits straightforward comparison with
specifications commonly used in macroeconomic models. Using the equilibrium
conditions, final output can be rewritten as
ξ
ξ ν −ξ
Yt = (ξ ν ) 1−ξ Ktα (t Lt )1−α Nt 1−ξ . (25)
To obtain sustained growth in this setting, we need to impose a parametric

restriction. Technically, balanced growth requires the aggregate production
function to be homogeneous of degree one in the accumulating factors Kt and
ξ
−ξ
Nt . In the following, we thus impose the restriction that α + 1−ξ
ν
= 1.7 In this
case, we obtain a standard neoclassical production function of the form Yt =
Zt Ktα L1−α
t , where
Zt ≡ A(t Nt )1−α (26)

ξ
is the Solow residual, or productivity, with A ≡ (ξ ν ) (1−ξ ) > 0.8 The equilibrium
productivity process thus contains a component driven by the exogenous forcing
process, t , and an endogenous component reflecting the intangible capital
stock, Nt . Importantly, while t is strictly stationary, productivity Zt grows at
an endogenous rate in equilibrium through the accumulation of patents. In this
sense, technological progress is endogenous in our model.
B. Growth Cycles
Expression (26) highlights the importance of the accumulation of patents for
productivity growth in the economy. The payoff to the creation of a new patent
is its value, Vt . Naturally, the growth rate of intangible capital Nt+1 ≡ NNt+1
t
reflects patent values and thus in equilibrium we have
η
Nt+1 = (1 − φ) + Et χ Mt+1 Vt+1 1−η . (27)
This relationship, linking the growth rate of intangible capital to patent values,
is central for the model and helps illustrate the main mechanisms determining
7 Without that restriction, the economy will exhibit either decreasing or increasing returns to
scale, so that growth rates either go to zero or will diverge in the long run. While positive growth
would still obtain along the transition path in both cases, we impose the restriction to obtain
positive growth in a stationary environment as does the bulk of the endogenous growth literature.
8 Similar decompositions can be found in Ethier (1982), Comin and Gertler (2006), and others.
equilibrium growth. In particular, iterating this expression forward, we obtain
⎡ ⎤ 1−η
η
∞

Nt+1 = (1 − φ) + Et ⎣χ η Mt+ j|t (1 − φ) j−1 t+ j ⎦
1
, (28)
j=1

where Mt+ j|t ≡ sj Mt+s|t is the j-step ahead stochastic discount factor and
Mt|t ≡ 1. The equilibrium growth rate is therefore tied to discounted future
profits on patents. We thus suggestively identify two major channels driving
equilibrium growth, a profit channel and a discount factor channel.
The profit channel implies that the dynamics of the growth rate of the
intangible capital stock reflect the dynamics of profits. Given t = 1ν −
1−ξ ξν −1 1−ξ
1
1 ξ ν Ktα (t Lt )1−α Nt , profit dynamics mirror the level of the forcing
process t , and hence are procyclical and persistent. Accordingly, expression
(28) suggests that the growth rate of intangible capital is procyclical and per-
sistent. We label these cyclical movements of the equilibrium growth rate as
growth cycles.
Figure 1 illustrates these dynamics. The figure displays the impulse re-
sponses of quantities in the intangible sector to a productivity shock. Impor-
tantly, after a positive shock, monopoly profits rise persistently. Intuitively, a
positive shock in the final goods sector raises the demand Xt for patents, which
translates directly into higher profits in the intangible sector. This raises the
value of a patent, which triggers a prolonged increase in R&D and the growth
rate of intangible capital. Finally, persistent R&D dynamics lead to a persistent
expected productivity growth component.
The discount factor channel, on the other hand, points to the relevance of
risk considerations for equilibrium growth dynamics, which can be seen in
equation (17). With recursive preferences, the stochastic discount factor is sen-
sitive to the entire intertemporal distribution of risk. Thus, a higher degree of
predictability in consumption growth translates into a more volatile stochas-
tic discount factor, and a more variable discount factor is reflected in a more
volatile equilibrium growth rate. This feedback channel between the discount
factor and endogenous growth dynamics implies a long-run growth amplifica-
tion mechanism that generates quantitatively significant long-run risks.
C. Equilibrium Long-Run Productivity Risk

In our economy, persistent variation in the growth rate of the intangible
stock is reflected in an endogenous persistent component in the dynamics of
productivity Zt by virtue of expression (26). Following Croce (2014), recent work
in equilibrium asset pricing with production specifies productivity growth to
contain a small persistent component, or, in other words, to exhibit long-run
productivity risk (Gomes, Kogan, and Yogo (2009), Backus, Routledge, and
Figure 1. Endogenous growth mechanism. This figure plots impulse response functions in
the benchmark model for the exogenous component of technology a, monopoly profits π , market
value of patents v, R&D intensity s − n, R&D stock growth n, and expected productivity growth
E[z] to a positive productivity shock ( > 0).
Zin (2007, 2010), Favilukis and Lin (2013a, 2013b)). Our model thus gives an
equilibrium interpretation to this channel.
Given the mean-reversion of t , the growth rate is negatively autocorrelated.
However, if ρ is sufficiently close to one, the growth rate of t is close to i.i.d.
and we may write
η
St
Et log Zt+1 ≈ (1 − α)χ . (29)
Nt
Any persistent movement in R&D thus provides an equilibrium source of long-

run productivity risk. More precisely, such long-run productivity risk is driven
St
by the dynamics of N t
, a ratio that we refer to as R&D intensity.
Quantitatively, the strength of this channel depends on the dynamics of R&D
intensity. Figure 2 depicts a series of an empirical counterpart of R&D intensity,
measured as the ratio of private business R&D investment (supplied by the
National Science Foundation) to the stock of R&D (supplied by the Bureau of
Labor Statistics) in the United States between 1953 and 2008. We discuss our
data sources in more detail in the next section. Casual inspection suggests that
Figure 2. Empirical R&D intensity. This figure plots the annual R&D intensity, defined as the
ratio of private business R&D investment to the stock of R&D, from 1953 to 2008. The data are
from the National Science Foundation and the Bureau of Labor Statistics.
this driver of productivity growth expectations is volatile and highly persistent.

In fact, in the data, the autocorrelation is 0.93. In our quantitative work,
we calibrate our model carefully to be consistent with observed properties of
innovation, and specifically R&D intensity. Moreover, we show through model
simulations that exposure to such equilibrium long-run productivity risk helps
in rationalizing sizeable risk premia in asset markets. This is because the
ensuing predictable variation in productivity growth leads to both substantial
long-run consumption risks as well as long-run cash flow risks. Thus, our model
allows us to identify economic sources of long-run risks.
D. Growth Cycles versus Business Cycles

Most of the literature in equilibrium asset pricing with production operates
in versions of the workhorse real business cycle model in which trend growth is
specified exogenously. In our quantitative work, we contrast the implications of
our benchmark endogenous growth model with those of a nested standard real
business cycle model with exogenous growth. The real business cycle model
we consider is a version of our benchmark model with constant R&D intensity.
Specifying R&D intensity exogenously is equivalent to specifying an exogenous
trend growth component in productivity, as in standard real business cycle
models. We thus contrast the asset pricing implications of growth cycles and
business cycles.
One way to interpret the differences between our model and the real business
cycle framework is that the trend component of the productivity process, Nt ,
is endogenous and fluctuates in our setup, while it is exogenous and typically
deterministic in real business cycle models. Through this channel, transitory
shocks have permanent and persistent effects in our model. This is important
from an asset pricing perspective, as Alvarez and Jermann (2005) show that,
in an economy exclusively driven by transitory shocks, the term premium is
the highest risk premium.
III. Quantitative Implications

In this section, we calibrate our model and explore its ability to replicate
key moments of both macroeconomic quantities and asset returns. We view our
endogenous growth model as a theory of long-run movements and therefore,
rather than match standard business cycle moments, we parameterize it to
be quantitatively consistent with long-run growth cycles by isolating the low-
frequency components of growth rates using a bandpass filter. On the other
hand, we find it instructive to compare our benchmark model with a version in
which trend growth is given exogenously, in the spirit of a business cycle model.
In the following, we refer to the benchmark endogenous growth model as the
growth cycle model and the exogenous growth counterpart as the business cycle
model.9
The models are calibrated at a quarterly frequency. The empirical moments
correspond to the U.S. sample from 1953 to 2008. We focus on this particular
period as R&D data become available only in 1953. The model is solved using
second-order perturbation methods.
A. Calibration
Our benchmark model requires that we specify 13 parameters: three for
preferences, seven relating to the final goods production technology, and three
for the innovation technology. We focus on long-run growth cycles to help us
determine key parameters. Note that we measure long-run growth cycles as
movements at frequencies between 100 and 200 quarters that we isolate using a
bandpass filter. In particular, we target the average growth rate of the economy
and the growth cycle volatilities σ GC of output, consumption, investment, and
R&D intensity. Table I summarizes our parameter choices.
To construct our targets, we need to find empirical measures of innovation
S
and R&D intensity, N . Our empirical series for St measures private business
9 While there is no exactly corresponding model with exogenous growth, we find our choice
natural as it facilitates comparison. The main conclusions are robust across a broad spectrum of
exogenous growth models. Extensive robustness checks with other exogenous growth specifications
are available from the authors on request.
Table I
Calibration
This table reports the benchmark quarterly calibration used for the benchmark growth cycle and
business cycle models.
Parameter Description Growth Cycle Business Cycle
β4 Subjective discount factor 0.984 0.984

ψ Elasticity of intertemporal substitution 1.85 1.85
γ Risk aversion 10 10
ξ Patent share 0.5 –
ν Inverse markup 0.6 –
α Capital share 0.35 0.35
ρ4 Autocorrelation of 0.95 0.95
χ Scale parameter 0.332 –
φ Patent obsolescence rate 0.0375 –
η Elasticity of new patents with respect to R&D 0.83 –
δ Depreciation rate of capital stock 0.02 0.02
σ Volatility of exogenous shock 1.75% 0.97%
ζ Investment adjustment cost parameter 3.3 3.3
μ∗4 Trend growth rate – 1.90%
R&D investment and comes from the National Science Foundation. The Bu-
reau of Labor Statistics (BLS) constructs the R&D stock by accumulating R&D
expenditures and allowing for depreciation, much in the same way as the phys-
ical capital stock is constructed. We thus use the R&D stock as our empirical
counterpart for Nt to be consistent with the accumulation process in (8). For
consistency, we use the same depreciation rate φ in our calibration as the BLS
uses in its calculations. The R&D stock can be interpreted as measuring the
economic benefits of R&D that spill over from the innovating firm to other
firms. We provide further details on the data sources in Appendix A.
We start by discussing the less standard parameters. We set χ , which is a
pure scaling parameter, to match the average growth rate of the U.S. economy
in our sample. We pick η, the elasticity of new patents with respect to R&D,
to match the growth cycle volatility of R&D intensity. This parameter can be
thought of as an adjustment cost parameter for R&D. Furthermore, our choice
of η is within the range of panel and cross-sectional estimates from Griliches
(1990). Analogously, we set ζ to match the growth cycle volatility of investment.
We choose σ , the volatility of the exogenous component of productivity, to match
the growth cycle volatility of output. Finally, we choose ρ to be consistent
with the autocorrelation of R&D intensity. This puts further discipline on the
importance of movements in innovation as a source of long-run productivity
risk, as becomes apparent in expression (29).
The choices of the remaining parameters follow the literature. Recursive
preferences have been used extensively in recent work in asset pricing. We
follow this literature and set preference parameters to standard values that
are also supported empirically. The elasticity of intertemporal substitution ψ
is set to 1.85 and the coefficient of relative risk aversion γ is set to 10.10 An
elasticity of intertemporal substitution larger than one is consistent with the
notion that an increase in uncertainty lowers the price-dividend ratio. Note
that, in this parametrization, ψ > γ1 , which implies that the agent dislikes
shocks to expected growth rates. The subjective discount factor β is set to an
annualized value of 0.984 so as to be consistent with the level of the risk-free
rate.
In the final goods sector, α is set to 0.35 to match the average capital share
and the depreciation rate of capital δ is set to 0.02 to match the average capital
investment rate, which are standard in the macroeconomic literature. The
parameter ξ is set to 0.5 to accord with the choice in Comin and Gertler (2006).
The inverse markup parameter in the intangible sector ν is set to 0.6 to be
consistent with the balanced growth restriction. While markups are generally
difficult to measure, and especially so on intangible capital, varying the param-
eter around a reasonable range does not change our key quantitative results.11
Since we interpret the variety of patents as the stock of R&D, as discussed, we
can interpret φ as the depreciation rate of the R&D stock. Hence, we set φ to
0.0375, which corresponds to an annualized depreciation rate of 15%, which is
a standard value and assumed by the BLS in the R&D stock calculations.
We calibrate the real business cycle model to facilitate direct comparison
with our benchmark model. To do so, we set a trend growth parameter μ
equal to 1.90% to match average output growth, and we adjust the volatility
of the forcing process to match the volatility of consumption growth in the
benchmark model. The remaining parameter choices are identical to those of
the benchmark model.
A.1. Implications for Growth and Cycles

To assess our benchmark calibration, we start by discussing its implications
for steady-state growth. We then explore economic fluctuations at higher (busi-
ness cycle) and lower (growth cycle) frequencies. The nature of fluctuations in
the model will be a key determinant of asset prices.
A.1.1. Steady-State Growth

Since trend growth is an endogenous variable in our model, the determinis-
tic steady-state growth rate is a function of the deep parameters of the model.
While closed-form expressions for the steady-state growth rate are not avail-
able, Table II illustrates the relationship between the model parameters and
trend growth through comparative statics analysis via our numerical solution.
10 The choice of the elasticity of intertemporal substitution is consistent with the estimation
evidence in Fernandez-Villaverde et al. (2012), while Bansal, Kiku, and Yaron (2013) use Euler
conditions and a GMM estimator to provide empirical support for these parameter values.
11 Competition has an effect on the equity premium in the model, as reducing the markup
lowers the volatility of cash flows, which leads to a smaller risk premium. On the other hand, the
quantitative effects in the present specification are small.
Table II
Determinants of Steady-State Growth
This table reports comparative statics analysis of the deterministic steady-state growth rate yss
from the benchmark growth cycle model. Panel A reports the impact of varying preferences pa-
rameters (around the benchmark calibration) on the steady-state growth rate. Panel B reports
the impact of varying technological parameters (around the benchmark calibration) on the steady-
steady growth rate.
Panel A: Comparative Statics with Preference Parameters
β 0.9945 0.995 0.9955 0.996 0.9965

yss 0.22% 0.30% 0.38% 0.45% 0.53%
ψ 1.70 1.75 1.80 1.85 1.90
yss 0.42% 0.43% 0.44% 0.45% 0.46%
γ 5 10 15 20 25
yss 0.45% 0.45% 0.45% 0.45% 0.45%
Panel B: Comparative Statics with Technological Parameters
ξ 0.496 0.498 0.500 0.502 0.504

yss 0.62% 0.54% 0.45% 0.37% 0.29%
ν 1.63 1.64 1.65 1.66 1.67
yss 0.69% 0.57% 0.45% 0.34% 0.24%
χ 0.3314 0.3317 0.3320 0.3323 0.3326
yss 0.445% 0.449% 0.453% 0.457% 0.461%
η 0.8290 0.8295 0.8300 0.8305 0.8310
yss 0.464% 0.459% 0.453% 0.448% 0.443%
φ 0.9615 0.9620 0.9625 0.9630 0.9635
yss 0.36% 0.40% 0.45% 0.50% 0.55%
δ 0.0190 0.0195 0.0200 0.0205 0.0210
yss 0.53% 0.49% 0.45% 0.42% 0.38%
ζ 0.05 0.06 0.07 0.08 0.09
yss 0.45% 0.45% 0.45% 0.45% 0.45%
We begin with the preference parameters (top panel). A higher value for the
time-preference parameter β implies that the agent values the future more rel-
ative to the present. Hence, the agent is willing to defer consumption and invest
more, which leads to higher growth. An increase in the intertemporal elasticity
of substitution ψ means that the agent is less concerned about smoothing the
consumption path and therefore leads to higher growth. Since we are analyzing
the deterministic steady-state, the coefficient of relative risk aversion does not
affect trend growth.12
On the technology side, increasing the patent share ξ leads to a reallocation of
resources from physical capital to intangible capital. As intangible capital has
a higher depreciation rate than physical capital, this reduces production effi-
ciency and therefore lowers growth. Increasing the parameter that determines
the average markup ν has two opposing effects. First, increasing the markup,
holding all else equal, increases monopoly profits in the intermediate sector.
12 Note that, in the stochastic steady-state, higher risk aversion will increase the precautionary
savings motive of the agent and raise the average growth rate.
Table III
Macro Moments
This table presents annualized macroeconomic moments from the data, the benchmark growth
cycle model and the business cycle model. Panel A reports the average long-run growth rate and
volatilities of low-frequency components of output growth y, consumption growth c, physical
investment growth i, and R&D intensity S/N. The low-frequency components are obtained using
the bandpass filter from Christiano and Fitzgerald (2003) and isolating frequencies between 100
and 200 quarters. Panel B reports short-run volatilities of output growth y, consumption growth
c, physical investment growth i, R&D expenditures growth s, and productivity growth z.
Panel C reports first autocorrelations for these macro growth rates and for Tobin’s Q. The model
statistics correspond to population moments.
Data Growth Cycle Business Cycle
Panel A: Growth Cycle Statistics
E[y] 1.90% 1.90% 1.90%

σy
GC
0.24% 0.22% 0.13%
σc
GC
0.28% 0.24% 0.15%
σi
GC
0.18% 0.17% 0.09%
σ GC (S/N) 0.71% 0.72% –
Panel B: Business Cycle Statistics
σc 1.42% 1.42% 1.42%

σc /σy 0.61 0.61 1.13
σi /σc 4.38 2.23 0.79
σs /σy 2.10 1.64 –
σz /σy 1.22 1.52 1.54
Panel C: Autocorrelation
AC1(z) 0.09 0.11 −0.020

AC1(c) 0.40 0.39 −0.002
AC1(y) 0.37 0.21 0.001
AC1(i) 0.25 0.14 0.012
AC1(Q) 0.95 0.96 0.89
Second, a higher markup depresses the demand for intermediate goods inputs,
which reduces monopoly profits. In our benchmark calibration, the second effect
dominates, and therefore a higher average markup lowers steady-state growth.
A higher scale parameter χ directly raises the level of productivity in the R&D
sector and therefore increases growth. A higher η increases the marginal re-
turns to R&D, which raises growth. Increasing the obsolescence rate of the
R&D stock φ reduces the returns to R&D and therefore growth declines. Anal-
ogously, a higher depreciation rate of physical capital stock δ lowers growth.
A.1.2. Growth and Business Cycles

Table III reports the main macro moments of the benchmark model and
the corresponding business cycle model. As targeted, the benchmark model is
quantitatively in line with the average growth rate of the economy and the
growth cycle volatilities σ GC of output, consumption, investment, and R&D

intensity in the data. In contrast, while similarly calibrated, the business cycle
counterpart generates quantitatively insignificant growth cycles.
Panel B reports standard business cycle statistics from simulations. The
model is also reasonably consistent with basic business cycle properties of the
U.S. economy. In particular, our benchmark model does just about as well as
the business cycle model in explaining short-run fluctuations. Both specifi-
cations match the low volatility of consumption growth of the post-war era
in the United States. On the other hand, they all predict investment to be too
smooth. This is because the benchmark model is calibrated to generate realistic
growth cycles, which, in sharp contrast to business cycle fluctuations in invest-
ment, are actually significantly smoother than the corresponding movements
in consumption. This suggests that the pronounced movements of investment
at business cycle frequencies are driven by a different set of shocks than the
long-run movements our model readily captures.13
Moving beyond business cycle volatilities to autocorrelations of growth rates,
as reported in Panel C, reveals a striking difference between the benchmark
model and the business cycle model. While the former qualitatively and some-
times quantitatively captures the substantial autocorrelation of most macro
variables in the data, the corresponding persistence implied by the business
cycle model is virtually zero, and sometimes slightly negative. This is remark-
able as the exogenous stochastic driver of productivity is the same across model
specifications. The lack of persistence in growth rates has long been identified
as a weakness of the real business cycle model (e.g., Cogley and Nason (1995)).
In stark contrast, endogenous movements in R&D induce a strong propagation
mechanism in our benchmark model that translates persistence in transitory
shocks to (i) persistence in the levels of macro variables (around the trend) and
(ii) persistence in the time-varying trend growth rate. Intuitively, a good shock
encourages innovation and thereby boosts growth in the long run even further.
In sum, both model specifications exhibit a similar amount of variation at
business cycle frequencies. In an asset pricing context, we refer to such move-
ments as short-run risk. However, in sharp contrast to the business cycle model,
our benchmark model exhibits significant persistence at lower frequencies,
namely, growth cycles. In an asset pricing context, we relate such movements
to long-run risks in the sense of Bansal and Yaron (2004), in a way made pre-
cise below. The presence of growth cycles in our economy leads to substantially
different asset pricing implications, as we explore in the next section.
B. Asset Pricing Implications

Table IV reports the asset pricing implications of the benchmark model
and alternative specifications. To understand these results, it is instructive to
compare the asset pricing implications of the benchmark model with those of
13 Similarly, we abstract from endogenous movements in the labor supply, as those drive a large
proportion of the fluctuations at business cycle frequencies.

Table IV
Asset Pricing Implications
This table reports the asset pricing implications for the benchmark growth cycle model, a high
volatility calibration of the growth cycle model, and the business cycle model. More specifically,
the volatility calibration corresponds to calibrating the volatility parameter σ in the growth cycle
model to match consumption volatility in the post–Great Depression sample (1930 to 2008). Also,
the scale parameter χ and the subjective discount factor β are adjusted to match the average
output growth rate and risk-free rate from the benchmark model, while all of the other parameters
are kept the same as the benchmark calibration. Panel A reports the means of the risk-free rate r f ,
the risk premium on the aggregate stock market E[rm ∗ − r ], the risk premium on physical capital
f
E[rd∗ − r f ], and the spread between physical capital and intangible capital E[rd∗ − ric∗ ]. Panel B
reports the standard deviation of these returns. The risk premiums are levered following Boldrin,
Christiano, and Fisher (2001). The model statistics correspond to population moments.
Growth Cycle—
Growth Cycle Volatility Calibration Business Cycle
Panel A: Mean
E[r f ] 1.21% 1.21% 2.61%

E[rm∗ −r ] 2.92% 5.76% 0.12%
f
E[rd∗ − r f ] 4.10% 8.33% 0.12%
E[rd∗ − ric
∗] 3.27% 6.89% –
Panel B: Standard Deviation
σc 1.42% 2.72% 1.42%

σr f 0.30% 0.38% 0.05%
σrm∗ −r f 4.86% 6.73% 2.27%
σrd∗ −r f 7.08% 9.49% 2.27%
σrd∗ −ric∗ 5.13% 7.81% –
the business cycle specification, which are radically different. These differences
are inherently linked to the presence of growth cycles in our benchmark econ-
omy. Growth cycles and persistent uncertainty about expected growth prospects
are akin to long-run risks, and give rise to both long-run consumption risks and
long-run cash flow risks.
Our benchmark model generates a low and smooth risk-free rate, while in
the business cycle model the risk-free rate is counterfactually high. Intuitively,
in the presence of long-run consumption risks, agents with a preference for
early resolution of uncertainty save for persistent low-growth episodes, low-
ering the equilibrium interest rate. In contrast, in the business cycle model
expected growth prospects are roughly constant, so that agents want to borrow
against their future income, which can only be prevented by a prohibitively
high equilibrium interest rate as documented in the table.
The presence of long-run cash flow risks, on the other hand, renders stocks
risky. Consistent with the multisector structure of our benchmark model, the
stock market is a claim to the net payout from production. Quantitatively, the
benchmark model generates a sizeable excess return on the aggregate stock
market of close to 3%, with a volatility close to 5%.
To facilitate comparison with the business cycle model, it is useful to decom-

pose the aggregate stock return into its components. Equation (24) provides a
decomposition of the value of this claim into the value of physical capital and
patents (intangible capital). Accordingly, we can separately define the return
on physical capital, the return on intangible capital, and the spread between
the two.
The excess returns on physical capital in the benchmark and business cy-
cle models are radically different. While in the presence of growth cycles the
premium on physical capital is in excess of 4%, it is close to zero in the busi-
ness cycle model and only a tiny fraction of the historical equity premium. In
this case, in the presence of growth cycles, agents with a preference for early
resolution of uncertainty fear that persistent slowdowns in cash flow growth co-
incide with a decrease in asset prices. In equilibrium, households thus require
a sizeable risk premium on capital.
We suggestively relate the spread between physical and intangible capital to
the value premium, which is defined as the return spread between high book-
to-market stocks (value stocks) and low book-to-market stocks (growth stocks).
Under this interpretation, the benchmark model generates a value spread
close to the excess return on the aggregate stock market, with considerable
volatility. The link is more suggestive, as growth firms in the data are likely
intangibles intensive but also hold physical capital, while in our model they do
not. Similarly, in our one-factor economy, a conditional CAPM holds. In other
words, value firms in our model have higher expected returns because they
have a higher conditional β. In this respect, this rationalization of a value
spread follows the arguments in Gomes, Kogan, and Zhang (2003), Carlson,
Fisher, and Giammarino (2004) and Zhang (2005).
While sizeable, the premia and volatilities of returns in the model do not
rationalize their empirical counterparts entirely. In line with our interpreta-
tion of the benchmark model as a model of long-run growth cycles, we view
the model-implied risk premia and volatilities as the components reflecting
uncertainty about long-term growth prospects and productivity. As discussed
earlier, our economy thus does not give a complete account of the relevant
short-run risks, which are not likely to be entirely productivity driven. In-
deed, Ai, Croce, and Li (2013) report that, empirically, the productivity-driven
fraction of return volatility is around 6%, which is roughly consistent with
our quantitative finding. On the other hand, the table also reports the asset
pricing implications of a version of the endogenous growth model that is cali-
brated to match short-run consumption risks in a long sample starting from the
Great Depression, as is customary in the literature. This calibration produces
an overall equity premium of close to 6%, and a value premium of a similar
magnitude.
Ultimately, in our model the dynamics of consumption and cash flows reflect
endogenous movements in productivity, that is, long-run productivity risks. We
now examine and quantify exposure to those risks, and relate them to their
determinant, R&D intensity.
Table V
Properties of Consumption Growth and the Stochastic Discount
Factor (SDF)
This table reports statistics for consumption growth dynamics (Panel A), expected consumption
dynamics (Panel B), and SDF dynamics (Panel C). Panel A reports the annualized volatility and
first autocorrelation of consumption growth and the annualized volatility of expected consumption
growth for the data, the benchmark growth cycle model, and the business cycle model. In Panel B,
we fit the the expected consumption growth process Et [ct+1 ] from the growth cycle and business
cycle models to an AR(1) process xt = ρx xt−1 + σx x,t , where x,t ∼ N(0, 1), and compare it to the
exogenous expected consumption growth component from Bansal and Yaron (2004). We report the
persistence parameter ρx and the annualized volatility parameter σ˜x from the fitted AR(1) process,
the relative volatility of expected consumption growth and realized consumption growth, and
the correlation between expected consumption growth and realized consumption growth. Panel C
reports the maximal Sharpe ratio and the one-period mean entropy as defined in Backus, Chernov,
and Zin (2014) for the growth cycle and business cycle models. The model statistics correspond to
population moments.
Panel A: Consumption Dynamics
σc 1.42% 1.42% 1.42%

AC1(c) 0.40 0.39 −0.002
σ (Et [ct+1 ]) – 0.51% 0.09%
Panel B: Expected Consumption Dynamics
Bansal-Yaron Growth Cycle Business Cycle
ρx 0.979 0.981 0.990

σ˜x 0.12% 0.10% 0.03%
σ (E[c)])/σ (c) 0.345 0.359 0.065
corr(E[c)], c) 0.344 0.526 −0.041
Panel C: SDF Dynamics
Growth Cycle Business Cycle
σ (M)/E(M) 0.326 0.015

I(1) 0.0520 0.0001
B.1. Long-Run Consumption Risks

Table V documents basic properties of consumption growth in the model.
Panel A shows that the benchmark model matches the volatility and the annual
autocorrelation of consumption growth in the data. While such persistence
points to predictable variation in consumption growth in the benchmark model,
the table also shows that the conditional mean of consumption growth Et [ct+1 ]
is quite volatile.
Panel B shows that this uncertainty about growth prospects in consumption
is also very persistent. We quantify the persistence by fitting an AR(1) process
to the expected consumption growth process from the benchmark model, in
the spirit of Bansal and Yaron (2004). This procedure reveals that expected
consumption growth is highly persistent and quite volatile in our model, close
to the exogenous parameterization in Bansal and Yaron (2004). In particular,
our process is slightly more persistent but slightly less volatile than their
specification. In other words, there is a fair amount of persistent uncertainty
about growth prospects in consumption, or long-run risks.
Our benchmark model thus exhibits quantitatively significant endogenous
long-run risks in consumption. This is in sharp contrast to the companion
business cycle specification, which, as noted before, counterfactually exhibits
near-i.i.d. consumption growth and therefore minimal time-variation in ex-
pected consumption growth. On the other hand, both model specifications ex-
hibit the same amount of short-run risks, as measured by the volatility of
realized consumption growth. Note that, in contrast to the exogenous specifi-
cation in Bansal and Yaron (2004), where innovations to consumption growth
and expected consumption growth are uncorrelated, in our model long- and
short-run risks are endogenously positively correlated. Bad news for the short
run are thus bad news for the long run, reinforcing the endogenous risks in our
model.
The presence of endogenous long-run risks in consumption has important
implications for the stochastic discount factor, and especially measures of its
volatility and dispersion. In Panel C, we report results for two such mea-
sures, namely, the maximal Sharpe ratio, σ (Mt )/E(Mt ), and the mean entropy,
E log Et Mt+1 − E log Mt+1 , following Backus, Chernov, and Zin (2014). Rela-
tive to both measures, the benchmark model generates much higher volatility
and dispersion than the business cycle specification. In terms of the maxi-
mal Sharpe ratio, sometimes referred to as the price of risk, it is instructive
to keep the implied value with power utility in mind. In that case, with the
calibrated consumption volatility and risk aversion, the Sharpe ratio would
be 0.145 10 × 1.45%. Deviations from that value reflect the dynamics of
expected consumption growth captured by the continuation utility term in
Epstein-Zin preferences. In the business cycle model, consumption growth
exhibits slight negative autocorrelation, so that innovations to consumption
growth and expected consumption growth tend to hedge each other, leading
to a low price of risk. The opposite obtains in the benchmark model, resulting
in a higher price of risk than the business cycle model. More importantly, the
volatility of expected consumption growth is much lower in the real business
cycle model.
The presence of long-run risks can also be characterized by inspecting the
autocorrelation function for consumption growth and observing persistence.
Figure 3 plots the first 10 autocorrelations for annual consumption growth from
the benchmark model, the business cycle model, and the data. The benchmark
model generates a sizeable first autocorrelation as in the data, but slightly
more persistence at longer horizons. Importantly, the consumption dynamics
from our model are broadly consistent with the data. In contrast, the business
cycle model generates virtually no persistence in consumption growth.
Figure 3. Consumption growth autocorrelations. This figure plots the first 10 autocorrela-
tions of consumption growth. The line with circular markers plots the autocorrelations from the
data for the 1953 to 2008 sample period. The line with the triangular markers plots the short-
sample autocorrelations from the growth cycle model. The line with the square-like markers plots
the short-sample autocorrelations from the business cycle model. From the models, we average
across 100 simulations that are equivalent in length to the data sample. The dashed lines repre-
sent the lower and upper boundaries of the 95% confidence interval.
B.2. Long-Run Cash Flow Risks

The risk premium on equity is a reflection of both consumption and cash flow
risks. Panel A of Table VI documents properties of dividend growth in the model.
In line with most of the extant general equilibrium asset pricing literature, our
model does not adequately capture the dynamics of stock market dividends
obtained from Compustat data.14 On the other hand, our model rationalizes
some of the risks inherent in macroeconomic dividends as measured by the BLS.
These dividends are measured as net corporate dividends from both publicly
and privately held firms paid out to U.S. investors and arguably form a closer
empirical counterpart to the notion of dividends entertained in the model. We
provide further details on the data source in Appendix B.
The table reveals that our benchmark model endogenously generates more
short-run risk as well as long-run risk in dividend growth than the busi-
ness cycle specification. Several statistics indicate persistent uncertainty about
14 In a recent paper, Favilukis and Lin (2013b) make progress on this front by explicitly account-
ing for wage rigidity, leverage, as well as nonconvexities in investment adjustment costs.
Table VI
Properties of Dividend Growth and the P/D Ratio
This table reports statistics for dividend growth dynamics (Panel A), expected dividend dynamics
(Panel B), and price-dividend ratio dynamics (Panel C). Panel A reports the annualized volatility
and first autocorrelation of dividend growth and the annualized volatility of expected dividend
growth for the data, the benchmark growth cycle model, and the business cycle model. In Panel
B, we fit the expected dividend growth process Et [ct+1 ] from the growth cycle and business
cycle models to an AR(1) process xt = ρx xt−1 + σx x,t , where x,t ∼ N(0, 1), and compare it to the
exogenous expected dividend growth component from Bansal and Yaron (2004). We report the
persistence parameter ρx and the annualized volatility parameter σ˜x from the fitted AR(1) process,
the relative volatility of expected dividend growth and realized dividend growth, and the correlation
between expected dividend growth and realized dividend growth. Panel C reports the annualized
volatility of the log price-dividend ratio, the annualized volatility of the low-frequency component
of the log price-dividend ratio, and the first autocorrelation of the log price-dividend ratio for the
data and the growth cycle and business cycle models. The low-frequency component is obtained
using the bandpass filter from Christiano and Fitzgerald (2003) and isolating frequencies between
100 and 200 quarters. The model statistics correspond to population moments.
Panel A: Dividend Growth Dynamics
σ (d) 6.72% 3.21% 2.48%

σ GC (d) 0.93% 0.77% 0.48%
σ (E[d]) – 0.78% 0.18%
AC1(d) 0.02 0.11 −0.01
Panel B: Expected Dividend Growth Dynamics
Bansal-Yaron Growth Cycle Business Cycle
ρ 0.979 0.971 0.990

σ̃xd 0.36% 0.17% 0.05%
σ (E[d)])/σ (d) 0.239 0.257 0.072
corr(E[d)], d) 0.236 0.101 −0.070
Panel C: P/D Ratio Dynamics
σ ( p − d) 41.54% 23.26% 13.21%

σ GC ( p − d) 25.86% 7.65% 4.40%
AC1( p − d) 0.89 0.90 0.94
dividend growth in the benchmark model. In contrast to the business cycle

model, and qualitatively in line with the data, dividend growth exhibits posi-
tive autocorrelation. Moreover, the conditional mean of dividend growth is quite
volatile, pointing to persistent uncertainty about cash flow growth. Similarly,
dividends exhibit fairly volatile growth cycles, as measured by σ GC .
One difficulty that often arises in general equilibrium asset pricing models
with production is that they predict dividends to be countercyclical. In these
models, companies find it optimal to cut dividends in order to take advantage
of productive investment opportunities, so that cash flows effectively end up
Table VII
Properties of Innovation and Productivity
This table reports statistics for variables pertaining to innovation and productivity from the data
and the benchmark growth cycle model. Panel A reports volatilities and first autocorrelations
of innovation-related measures (R&D expenditure growth s, R&D stock growth n, and R&D
intensity S/N) and productivity growth z. Panel B reports the annual persistence and standard
deviation of the expected growth rate component of productivity growth. The data estimates are
taken from Croce (2014), where the expected growth rate component of productivity x̃t−1 is a latent
variable that is assumed to follow an AR(1). In contrast, in the growth cycle model the expected
growth rate component is the growth rate of the variety of intermediate goods nt , an endogenous
structural variable of the model. The model statistics correspond to population moments.
Data Growth Cycle
Panel A: Innovation and Productivity Dynamics
σs 4.89% 3.82%

AC1(s) 0.21 0.06
AC1(n) 0.90 0.94
AC1(S/N) 0.93 0.93
AC1(z) 0.09 0.11
σ (Et [zt+1 ]) 0.38%
Panel B: Expected Productivity Dynamics
ρx̃ 0.93 0.95

σ (x̃) 1.10% 1.20%
hedging consumption risks and reduce risk premia on equity. This effect is
alleviated in our model due to strongly procyclical aggregate profits Nt t on
patents. Moreover, in the model, aggregate profit growth exhibits a substantial
amount of low-frequency variation itself, with the volatility of its conditional
mean equal to 0.42%. Accounting for profits thus helps the benchmark model
capture more realistic cash flow risks. However, while the model generates sub-
stantial low-frequency variation in profits, it underestimates the total volatility
of profits. We view matching cash flow dynamics accurately as an interesting
and important extension for future work.
More realistic cash flow dynamics also affect valuation ratios, as documented
in Panel C of Table VI. In particular, endogenous long-run risks capture roughly
half of the empirical volatility of price-dividend ratios.
B.3. Innovation and Long-Run Productivity Risks

Ultimately, in the benchmark model, long-run risks in both consumption and
cash flows reflect innovation-driven movements in endogenous productivity, or,
in other words, endogenous long-run productivity risks. As equation (29) above
highlights, the significance of this channel depends crucially on the empirical
S
properties of R&D intensity, N . Table VII documents properties of innovation
and productivity in the benchmark model and in the data.
Panel A documents that the model is broadly consistent with volatilities and
autocorrelations of R&D investment, the stock of R&D, and R&D intensity
in the data. Crucially, in line with its empirical counterpart, R&D intensity
is a persistent process and we match its annual autocorrelation of 0.93. Not
surprisingly, such persistence is reflected in a positive autocorrelation of pro-
ductivity growth in the model, close to the value in the data. Moreover, the
model predicts considerable uncertainty about future productivity growth as
measured by the volatility of its conditional mean.
These results suggest quantitatively significant long-run productivity risks
driven by empirically plausible movements in R&D intensity. More formally,
uncovering the persistent component in productivity growth as a latent variable
in the data (as in Croce (2014)) yields an annual persistence coefficient of 0.93
for the expected growth rate of productivity, while our model closely matches
this number with a persistence coefficient of 0.95. Moreover, the volatilities of
expected productivity growth rates in the data and in the model roughly match.
Note that, in contrast to our benchmark model, the business cycle specification
implies that productivity growth is essentially i.i.d., which is inconsistent with
empirical evidence.
Qualitatively, the model predicts that R&D intensity should closely track
productivity growth. Figure 4 depicts these patterns in the model, using a
simulated sample path, as well as in the data. The plots highlight the small
but persistent component in productivity growth induced by equilibrium R&D
activity. In our general equilibrium model, this persistent component shows
up in consumption growth as well, as the figure also illustrates. Empirically,
therefore, we expect R&D intensity to forecast productivity and consumption
growth. We test this prediction in Section III.G.
B.4. Comovement
Our model also has realistic implications for comovement between prices
and quantities at lower frequencies. In the following, we identify low-frequency
movements in growth rates using a bandpass filter that isolates movements at
frequencies between 100 and 200 quarters.
Figure 5 reveals the close match between the price-dividend ratio and pro-
ductivity growth in the data and the benchmark model at low frequencies.
This evidence strongly suggests the presence of slow productivity-driven move-
ments in asset market valuations in the data. In the model, these movements
are driven by variation in expected cash flows, induced by time-variation in
R&D intensity. This is because the benchmark model generates little time-
variation in risk premia. While there is evidence for time-variation in expected
cash flows as discussed above, time-variation in price-dividend ratios is often
related to time-varying risk premia. In an extension, we augment our model
with stochastic volatility in the exogenous shock and find that it replicates the
predictability evidence well.
At lower frequencies we also find strong cross-correlations between stock
returns and consumption growth. This is displayed in Figure 6, indicating the
Figure 4. Growth rates and R&D intensity. The top left panel plots demeaned log consumption
S
growth ct (thin line) and R&D intensity Nt−1 (thick line) from the growth cycle model, while the
t−1
bottom left panel shows the same plot for the data. The top right panel plots demeaned log output
S
growth yt (thin line) and R&D intensity Nt−1 (thick line) from the growth cycle model, while the
t−1
bottom right panel shows the same plot for the data.
lag-lead structure between returns and consumption growth. In the data and at
low frequencies, returns lead consumption growth by several quarters and the
lead correlations die away more slowly relative to the lag correlations. In other
words, lower-frequency movements in returns contain important information
regarding long-run movements in future growth. In contrast to the business
cycle specification, the benchmark model replicates this feature of the data
quite well. This important divergence between the two models is due to the
endogenous predictable component in productivity growth, which is absent in
the business cycle model. In sum, the benchmark model is able reconcile to the
long-term relationship between returns and growth quite well.
C. Empirical Evidence
Time-varying growth prospects in consumption are at the core of the long-
run risks literature following Bansal and Yaron (2004). However, the empirical
Figure 5. Low-frequency component of productivity growth and price-dividend

ratio. This figure plots the low-frequency components for productivity growth (thick line) and
for the price-dividend ratio (thin line). The left panel corresponds to a sample simulation from the
growth cycle model and the right panel corresponds to the data. The low-frequency component is
obtained by applying the bandpass filter from Christiano and Fitzgerald (2003) and selecting a
bandwidth of 100 to 200 quarters. The correlation between the two series is 0.46 in the data and
0.67 in the model.
Figure 6. Low-frequency cross-correlation of returns and consumption growth. The left

panel plots cross-correlations of the medium-frequency component of the equity return and the
low-frequency component of consumption growth for the growth cycle (thick line) and business
cycle (dashed line) models: corr(rd,t , ct+k). The right panel plots the same cross-correlations from
the data. The low-frequency component is obtained using the bandpass filter from Christiano and
Fitzgerald (2003) and selecting a bandwidth of 100 to 200 quarters.
evidence regarding this channel is still controversial. In particular, few instru-

ments have been shown to successfully predict consumption growth over longer
horizons. Our benchmark model implies that R&D intensity should predict con-
sumption growth. We now present empirical evidence supporting this channel,
based on quarterly data from 1953 to 2008. Innovation-related measures are
thus economically meaningful predictors of aggregate growth rates.
Panel A of Table VIII documents the results from projecting future consump-
tion growth for horizons of one to five years on log R&D intensity, both in the
data and in model simulations. Empirically, the slope coefficients are positive,
Table VIII
Consumption Growth Forecasts
This table presents short-sample annual consumption growth forecasting regressions from the
data and from the benchmark growth cycle model for horizons (k) of one to five years. In
Panel A, log consumption growth is projected on log R&D intensity, ct,t+1 + · · · + ct+k−1,t+k =
α + β(s − n)t + νt,t+k. In Panel B, log consumption growth is projected on log R&D stock growth,
ct,t+1 + · · · + ct+k−1,t+k = α + βnt + νt,t+k. The regressions are estimated via OLS with Newey-
West standard errors with k − 1 lags and overlapping annual observations. The estimates from the
model regression are averaged across 100 simulations that are equivalent in length to the data
sample.
Horizon (Years)
1 2 3 4 5
Panel A: Forecasts with R&D Intensity
β (Data) 0.017 0.034 0.048 0.062 0.077

S.E. (Data) 0.006 0.012 0.017 0.023 0.030
R2 (Data) 0.070 0.105 0.131 0.163 0.202
β (Growth cycle) 0.068 0.118 0.168 0.200 0.224
S.E. (Growth cycle) 0.028 0.052 0.073 0.095 0.116
R2 (Growth cycle) 0.141 0.161 0.179 0.175 0.168
Panel B: Forecasts with R&D Growth
β (Data) 0.217 0.395 0.540 0.703 0.842

S.E. (Data) 0.084 0.178 0.276 0.347 0.401
R2 (Data) 0.094 0.115 0.132 0.168 0.198
β (Growth cycle) 0.573 1.012 1.437 1.750 1.993
S.E. (Growth cycle) 0.189 0.356 0.526 0.704 0.878
R2 (Growth cycle) 0.158 0.189 0.207 0.203 0.193
increasing with horizon, and statistically significant. The R2 s are between 0.07
and 0.2, and are monotonically increasing with horizon. Less surprisingly, we
find a similar pattern in our model simulations. For completeness, Panel B re-
ports results from projecting future consumption growth on a related measure
of innovation, namely, the growth rate of the log R&D stock, log Nt+1 − log Nt .
A similar pattern obtains. In the data, R&D stock growth forecasts consump-
tion growth over horizons of one to five years with statistically significant and
positive slope coefficients and sizeable R2 s, in line with our model. These regres-
sions give empirical support for the notion of innovation-driven low-frequency
variation in consumption growth, consistent with the implications of our bench-
mark model.
In the model, just as changing growth expectations in consumption reflect
movements in innovative activity, so too do changing productivity and output
growth expectations. Therefore, we expect measures of innovation to forecast
productivity and output growth. In Table IX, we provide empirical evidence
supporting these predictions. The table documents that in the data both R&D
intensity and R&D stock growth forecast productivity and output growth over
Table IX
Output and Productivity Growth Forecasts
This table presents annual output and productivity growth forecasting regressions from the data
and from the benchmark growth cycle model for horizons (k) of one to five years. In the top two
panels, log output growth is projected on log R&D intensity, yt,t+1 + · · · + yt+k−1,t+k = α + β(s −
n)t + νt,t+k (Panel A) and on log R&D stock growth, yt,t+1 + · · · + yt+k−1,t+k = α + βnt + νt,t+k
(Panel B). In the bottom two panels, log productivity growth is projected on log R&D intensity,
zt,t+1 + · · · + zt+k−1,t+k = α + β(s − n)t + νt,t+k (Panel C) and on log R&D stock growth, zt,t+1 +
· · · + zt+k−1,t+k = α + βnt + νt,t+k (Panel D). The regressions are estimated via OLS with Newey-
West standard errors with k − 1 lags and overlapping annual observations. The estimates from the
model regression correspond to population estimates.
Horizon (Years)
1 2 3 4 5
Panel A: Output Forecasts with R&D Intensity
β (Data) 0.020 0.046 0.068 0.089 0.114

S.E. (Data) 0.013 0.022 0.029 0.041 0.051
R2 (Data) 0.040 0.084 0.119 0.158 0.210
β (Growth cycle) 0.085 0.163 0.236 0.306 0.372
R2 (Growth cycle) 0.105 0.161 0.195 0.217 0.231
Panel B: Output Forecasts with R&D Growth
β (Data) 0.267 0.453 0.572 0.763 0.940

S.E. (Data) 0.130 0.261 0.387 0.457 0.499
R2 (Data) 0.061 0.067 0.073 0.113 0.159
β (Growth cycle) 0.635 1.230 1.780 2.307 2.792
R2 (Growth cycle) 0.120 0.159 0.193 0.212 0.222
Panel C: Productivity Forecasts with R&D Intensity
β (Data) 0.014 0.031 0.049 0.069 0.091

S.E. (Data) 0.009 0.015 0.024 0.032 0.041
R2 (Data) 0.031 0.080 0.120 0.174 0.232
β (Growth cycle) 0.075 0.142 0.204 0.261 0.314
R2 (Growth cycle) 0.039 0.062 0.077 0.088 0.095
Panel D: Productivity Forecasts with R&D Growth
β (Data) 0.431 0.820 1.230 1.707 2.092

S.E. (Data) 0.190 0.315 0.452 0.522 0.599
R2 (Data) 0.113 0.192 0.262 0.376 0.444
β (Growth cycle) 0.560 1.070 1.533 1.948 2.322
R2 (Growth cycle) 0.037 0.060 0.076 0.084 0.090
several years significantly, with R2 s that are increasing with horizon. Qualita-
tively, the model replicates these patterns well.
Taken together, we find evidence that aggregate growth rates, including
consumption, are indeed time-varying and predictable by innovation-related
measures over longer horizons, just as predicted by the benchmark model.
The model thus helps identify economic sources of long-run risks in the
data.
IV. Conclusion
This paper provides a quantitative analysis of a production economy whose
long-term growth prospects are endogenously determined by innovation and
R&D. By integrating innovation and R&D into a real business cycle model
with recursive preferences, our model constitutes a straightforward extension
of the workhorse model of modern macroeconomics. In sharp contrast to the
latter, however, our baseline model jointly rationalizes key features of asset
returns and long-run macroeconomic performance in the data.
In the model, favorable economic conditions boost innovation and the de-
velopment of new technologies. Since technological progress fosters long-run
economic growth, endogenous innovation generates a powerful propagation
mechanism for shocks reflected in persistent variation in long-term growth
prospects. With recursive preferences, innovations to expected growth are
priced and lead to high-risk premia in asset markets, as agents fear that per-
sistent slowdowns in growth coincide with low asset valuations. Formally, we
show that R&D drives an endogenous predictable component in measured pro-
ductivity, which gives an innovation-based explanation of long-run productivity
risk in the data.
Our model thus allows us to empirically identify economic sources of long-
run risks. Indeed, we document novel empirical evidence that measures of
innovation have significant predictive power for aggregate growth rates at
longer horizons.
Initial submission: April 22, 2012; Final version received: September 1, 2014
Acting Editor: David Backus
Appendix A: Equilibrium
In this section, we collect all the equations that determine the symmetric
equilibrium in our economy.
A symmetric equilibrium in the model is defined as an exogenous
stochastic sequence, {t = exp(at )}∞ t=0 , an initial condition {K0 , N0 } for
the endogenous state variables, a sequence of endogenous variables,
{Ct , Ut , Mt , Yt , Wt , qt , It ,
t , Xt , t , Vt , St }∞ ∞
t=0 , and laws of motion {Kt+1 , Nt+1 }t=0
such that
a. the state variables {Kt , Nt }∞

t=0 satisfy their laws of motions,
b. the endogenous variables solve the producers’ and the consumers’ prob-
lems, and
c. the aggregate resource constraint is satisfied.
d. prices are set such that markets clear.
The equilibrium conditions of the model are summarized by the following 18

equations:

1−γ θ 1θ
Ut = (1 − β)Ctθ + β Et Ut+1 1−γ ,
θ−1 1−γ −θ
Ct Ut
Mt = β 1
,
Ct−1 Et (Ut
1−γ
) 1−γ
Yt = Ktα Zt1−α ,
Zt = Aeat Nt ,
at = ρat−1 + t ,
Wt = (1 − α)(1 − ξ )Yt ,
1
qt =

t

1 Yt+1 It+1
1 = Et Mt+1 α(1 − ξ ) + qt+1 (1 − δ) − + qt+1
t+1 ,
qt Kt+1 Kt+1
Kt+1 = (1 − δ)Kt +
t Kt
ξ
1
−1 1−ξ
Xt = ξ νeat (1−α)(1−ξ ) Ktα(1−ξ ) Ntν ,

α1 It ζ

t = + α2 ,
ζ Kt
ζ −1
It

t = α1 ,
Kt

1
t = − 1 Xt ,
ν
Vt = t + (1 − φ)Et [Mt+1 Vt+1 ],
Nt+1 = ϑt St + (1 − φ)Nt ,
χ · Nt
ϑt = 1−η η
,
St Nt
St = Et [Mt+1 Vt+1 ](Nt+1 − (1 − φ)Nt ),
Ct = Yt − It − Nt Xt − St .
We solve the model in dynare++4.2.1 using a second-order approximation.

The policies are centered about a fixed point that takes into account the effects
of volatility on decision rules.
Appendix B: Data
Annual and quarterly data for consumption, capital investment, and GDP
are from the Bureau of Economic Analysis. Annual data on private business
R&D investment are from the survey conducted by The National Science Foun-
dation. Annual data on the stock of private business R&D are from the BLS.
Annual productivity data come from the BLS and are measured as multifactor
productivity in the private nonfarm business sector. Quarterly data on divi-
dends are obtained from the BLS. The sample period is 1953 to 2008, since
R&D data are only available during that time period. Consumption is mea-
sured as expenditures on nondurable goods and services. Capital investment is
measured as private fixed investment. Output is measured as GDP. Dividends
are measured as net corporate dividends. The nominal variables are converted
to real terms using the consumer price index (CPI), which is obtained from the
Center for Research in Security Prices (CRSP).
Monthly nominal return and yield data are from CRSP. The real market
return is constructed by taking the nominal value-weighted return on the New
York Stock Exchange and American Stock Exchange and deflating it using the
CPI. The real risk-free rate is constructed by using the nominal average one-
month yields on Treasury bills and taking out expected inflation.15 Aggregate
market and book values of assets are from the Flow of Funds account. Price-
dividend ratio data are from Robert Shiller’s webpage:
http://www.econ.yale.edu/∼shiller/data.htm.
REFERENCES
Ai, Hengjie, Massimiliano Croce, and Kai Li, 2013, Towards a quantitative equilibrium model of
intangible capital, Review of Financial Studies 26 491–538.
Alvarez, Fernando, and Urban Jermann, 2005, Using asset prices to measure the persistence of
the marginal utility of wealth, Econometrica 73, 1977–2016.
Backus, David, Mikhail Chernov, and Stanley Zin, 2014, Sources of entropy in representative agent
models, Journal of Finance 69, 51–99.
Backus, David, Bryan Routledge, and Stanley Zin, 2007, Asset prices in business cycle analysis,
Working paper, New York University.
Backus, David, Bryan Routledge, and Stanley Zin, 2010, Cyclical component of U.S. asset returns,
Working paper, New York University.
Bansal, Ravi, Dana Kiku, and Amir Yaron, 2013, Risks for the long run: Estimation with time
aggregation, Working paper, Duke University.
Bansal, Ravi, and Amir Yaron, 2004, Risks for the long run: A potential resolution of asset pricing
puzzles, Journal of Finance 59, 1639–1672.
Boldrin, Michele, Lawrence Christiano, and Jonas Fisher, 2001, Habit persistence, asset returns
and the business cycle, American Economic Review 91, 149–166.
Campanale, Claudio, Rui Castro, and Gian Luca Clementi, 2008, Asset pricing in a production
economy with Chew-Dekel preferences, Review of Economic Dynamics 13, 379–402.
Carlson, Murray, Adlai Fisher, and Ron Giammarino, 2004, Corporate investment and asset price
dynamics: Implications for the cross-section of returns, Journal of Finance 59, 2577–2603.
Christiano, Lawrence, and Terry Fitzgerald, 2003, The band-pass filter, International Economic
Review 44, 435–465.
15 We model the monthly time series process for inflation using an AR(4).
Cogley, Tim, and James Nason, 1995, Output dynamics in real business cycle models, American
Economic Review 85, 492–511.
Comin, Diego, and Mark Gertler, 2006, Medium term business cycles, American Economic Review
96, 523–551.
Comin, Diego, Mark Gertler, and Ana Maria Santacreu, 2009, Technology innovation and diffusion
as sources of output and asset price fluctuations, Working paper, Harvard University.
Croce, Massimiliano, 2014, Long-run productivity risk: A new hope for production-based asset
pricing, Journal of Monetary Economics 66, 13–31.
Dixit, Avinash, and Joseph Stiglitz, 1977, Monopolistic competition and optimum product diversity,
American Economic Review 67, 297–308.
Ethier, Wilfried, 1982, National and international returns to scale in the modern theory of inter-
national trade, American Economic Review 72, 389–405.
Favilukis, Jack, and Xiaoji Lin, 2013a, Long run productivity risk and aggregate investment,
Journal of Monetary Economics 60, 737–751.
Favilukis, Jack, and Xiaoji Lin, 2013b, Wage rigidity: A quantitative solution to several asset
pricing puzzles, Working paper, London School of Economics.
Fernandez-Villaverde, Jesus, Ralph Koijen, Juan Rubio-Ramirez, and Jules van Binsbergen, 2012,
The term structure of interest rates in a DSGE model with recursive preferences, Journal of
Monetary Economics 59, 634–648.
Gala, Vito, 2010, Irreversible investment and the cross-section of stock returns in general equilib-
rium, Working paper, London Business School.
Garleanu, Nicolae, Leonid Kogan, and Stavros Panageas, 2012, The demographics of innovation
and asset returns, Journal of Financial Economics 105, 491–510.
Garleanu, Nicolae, Stavros Panageas, and Jianfeng Yu, 2012, Technological growth and asset
pricing, Journal of Finance 67, 1265–1292.
Gomes, Joao F., Leonid Kogan, and Motohiro Yogo, 2009, Durability of output and expected stock
returns, Journal of Political Economy 117, 941–986.
Gomes, Joao F., Leonid Kogan, and Lu Zhang, 2003, Equilibrium cross-section of returns, Journal
of Political Economy, 111, 693–731.
Gordon, Robert J., 2010, Revisiting U.S. productivity growth over the past century with a view of
the future, Working paper, Northwestern University.
Gourio, Francois, 2012, Disaster risk and business cycles, American Economic Review 102, 2734–
2766.
Gourio, Francois, 2013, Credit risk and disaster risk, American Economic Journal: Macroeconomics
5, 1–34.
Griliches, Zvi, 1990, Patent statistics as economic indicators: A survey, Journal of Economic Liter-
ature 28, 1661–1707.
Grossman, Gene, and Elhanan Helpman, 1991, Innovation and Growth in the Global Economy
MIT Press, Cambridge, MA.
Helpman, Elhanan, 1998, General Purpose Technologies and Economic Growth MIT Press, Cam-
bridge, MA.
Jermann, Urban, 1998, Asset pricing in production economies, Journal of Monetary Economics 41,
257–275.
Jermann, Urban, and Vincenzo Quadrini, 2007, Stock market boom and the productivity gains of
the 1990s, Journal of Monetary Economics 54, 413–432.
Jovanovic, Boyan, and Peter Rousseau, 2005 Vol. 1, Part B, General purpose technologies in
P. Aghion and S.N. Durlauf, eds.: Handbook of Economic Growth (Elsevier, North Holland,
Amsterdam).
Kaltenbrunner, Georg, and Lars Lochstoer, 2010, Long run risk through consumption smoothing,
Review of Financial Studies 23, 3190–3224.
Kogan, Leonid, and Dimitris Papanikolaou, 2010, Growth opportunities and technology shocks,
American Economic Review 100, 532–536.
Kogan, Leonid, Dimitris Papanikolaou, and Noah Stoffman, 2012, Technological innovation: Win-
ners and losers, Working paper, MIT.
Kuehn, Lars-Alexander, 2008, Asset pricing implications of real investment commitment, Working
paper, Carnegie Mellon University.
Kung, Howard, 2015, Macroeconomic linkages between monetory policy and the term structure of
interest rates, Journal of Financial Economics 115, 41–57.
Lin, Xiaoji, 2012, Endogenous technological progress and the cross-section of stock returns, Journal
of Financial Economics 103, 411–428.
Papanikolaou, Dimitris, 2011, Investment shocks and asset prices, Journal of Political Economy
119, 639–685.
Pástor, Lǔboš, and Pietro Veronesi, 2009, Technological revolutions and stock prices, American
Economic Review 99, 1451–1483.
Romer, Paul, 1990, Endogenous technological change, Journal of Political Economy 98, 71–102.
Tallarini, Thomas, 2000, Risk-sensitive real business cycles, Journal of Monetary Economics 45,
507–532.
Zhang, Lu, 2005, The value premium, Journal of Finance 60, 67–103.

Kung 2015

Uploaded by

Copyright:

Available Formats

Kung 2015

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

Kung 2015

Uploaded by

Copyright:

Available Formats

THE JOURNAL OF FINANCE • VOL. LXX, NO.

Innovation, Growth, and Asset Prices

AN ECONOMY’S LONG-TERM GROWTH prospects reflect its innovative potential.

in long-term growth prospects. Our setup has two distinguishing features.

and Papanikolaou (2011) explore endogenous long-run consumption risks in

A.1. Final Good Sector

where the composite Gt is defined as

The firm’s objective is to maximize shareholder value. This can be formally

where the firm’s dividends are

Here, Mt is the stochastic discount factor, It is investment in physical capital,

A.2. Intangible Goods Sector

A.3. Innovation Sector

Nt+1 = ϑt St + (1 − φ)Nt , (8)

where γ is the coefficient of relative risk aversion and ψ ≡ 1−θ

Ct + Qt Zt+1 + Bt+1 = Wt Lt + (Qt + Dt )Zt + (1 + r f ,t )Bt , (11)

C. Equilibrium and Asset Prices

where the second term, involving continuation utilities, captures preferences

Qt = Et [Mt+1 (Qt+1 + Dt+1 )], (14)

with Xt = ξ ν Ktα (t Lt )1−α Nt . Profits depend positively on the ag-

Et [Mt+1 Vt+1 ](Nt+1 − (1 − φ)Nt ) = St , (20)

C.1. Resource Constraint

C.2. Stock Market

II. Equilibrium Growth Risk

novel source of long-run risks in the economy. In Section III, we quantitatively

To obtain sustained growth in this setting, we need to impose a parametric

Zt ≡ A(t Nt )1−α (26)

equilibrium growth. In particular, iterating this expression forward, we obtain

C. Equilibrium Long-Run Productivity Risk

Any persistent movement in R&D thus provides an equilibrium source of long-

this driver of productivity growth expectations is volatile and highly persistent.

D. Growth Cycles versus Business Cycles

III. Quantitative Implications

Parameter Description Growth Cycle Business Cycle

β4 Subjective discount factor 0.984 0.984

A.1. Implications for Growth and Cycles

A.1.1. Steady-State Growth

Panel A: Comparative Statics with Preference Parameters

β 0.9945 0.995 0.9955 0.996 0.9965

Panel B: Comparative Statics with Technological Parameters

ξ 0.496 0.498 0.500 0.502 0.504

Data Growth Cycle Business Cycle

Panel A: Growth Cycle Statistics

E[y] 1.90% 1.90% 1.90%

Panel B: Business Cycle Statistics

σc 1.42% 1.42% 1.42%

AC1(z) 0.09 0.11 −0.020

A.1.2. Growth and Business Cycles

growth cycle volatilities σ GC of output, consumption, investment, and R&D

B. Asset Pricing Implications

proportion of the fluctuations at business cycle frequencies.

E[r f ] 1.21% 1.21% 2.61%

Panel B: Standard Deviation

σc 1.42% 2.72% 1.42%

To facilitate comparison with the business cycle model, it is useful to decom-

Panel A: Consumption Dynamics

Data Growth Cycle Business Cycle

σc 1.42% 1.42% 1.42%

Panel B: Expected Consumption Dynamics

Bansal-Yaron Growth Cycle Business Cycle

E[y] 1.90% 1.90% 1.90%

σc 1.42% 1.42% 1.42%

AC1(z) 0.09 0.11 −0.020

σc 1.42% 2.72% 1.42%

σc 1.42% 1.42% 1.42%

σ (d) 6.72% 3.21% 2.48%

σs 4.89% 3.82%