J Ecolecon 2017 07 018
J Ecolecon 2017 07 018
J Ecolecon 2017 07 018
Ecological Economics
journal homepage: www.elsevier.com/locate/ecolecon
A R T I C L E I N F O A B S T R A C T
Article history: The world is banking on a major increase in food production, if the dietary needs and food preferences of an
Received 9 March 2017 increasing, and increasingly rich, population are to be met. This requires the further expansion of modern
Received in revised form 12 July 2017 agriculture, but modern agriculture rests on a small number of highly productive crops and its expansion has
Accepted 16 July 2017 led to a significant loss of global biodiversity. Ecologists have shown that biodiversity loss results in lower
Available online xxxx
plant productivity, while agricultural economists have linked biodiversity loss on farms with increasing
variability of crop yields, and sometimes lower mean yields. In this paper we consider the macro-economic
JEL classification:
consequences of the continued expansion of particular forms of intensive, modern agriculture, with a focus
N10
on how the loss of biodiversity affects food production. We employ a quantitative, structurally estimated
N50
O31 model of the global economy, which jointly determines economic growth, population and food demand,
O44 agricultural innovations and land conversion. We show that even small effects of agricultural expansion on
Q15 productivity via biodiversity loss might be sufficient to warrant a moratorium on further land conversion.
Q16 © 2017 Elsevier B.V. All rights reserved.
Q57
Keywords:
Agricultural productivity
Biodiversity
Endogenous growth
Food security
Land conversion
Population
http://dx.doi.org/10.1016/j.ecolecon.2017.07.018
0921-8009/© 2017 Elsevier B.V. All rights reserved.
B. Lanz et al. / Ecological Economics 144 (2018) 260–277 261
But expanding the modern agricultural system will have important more than 20% of species are lost, the effects may rival other drivers
implications for ecological systems and creates a number of chal- of global environmental change such as planetary warming, ozone
lenges for global management of the commons. Our objective in and acidification (Cardinale et al., 2012; Hooper et al., 2012). While
this paper is to study what this expansion might imply for global the negative relationship between biodiversity loss and plant pro-
food supply, taking into account the feedbacks between agricultural ductivity is reliably found in natural ecosystems such as grasslands, it
intensification, extensification, biodiversity loss and agricultural pro- is clearly possible for intensively managed monocultures to be highly
ductivity. productive. Nonetheless a recurrent finding from empirical studies
An increase in agricultural output can be achieved in various ways by economists is that genetic and species diversity on farms reduces
and the great increases seen in the second half of the twentieth cen- the variability of crop yields and sometimes increases the mean yield
tury came mainly from intensification and corresponding increases (see notably Di Falco, 2012; Tilman et al., 2005). This has been found
in yields (FAOSTAT; Klein Goldewijk et al., 2011). Nonetheless the not only in low-intensity, biologically diverse farming systems such
clear consensus from global land-use models is that some of the addi- as Ethiopia (Di Falco and Chavas, 2009; Di Falco et al., 2010) and Pak-
tional future production will come from expanding the agricultural istan (Smale et al., 1998), but also in high-intensity, low-biodiversity
land area. According to the Agricultural Model Intercomparison and farming systems such as the East of England (Omer et al., 2007).
Improvement Project or AgMIP, the area of world cropland in 2050 There are several reasons why more biologically diverse farming
will be between 10 and 25% larger than today, under a reference sce- systems would display lower yield variability, and sometimes higher
nario in which world food production rises by 43 to 99% (von Lampe mean yields. These include symbiotic interactions and resource-use
et al., 2014; Schmitz et al., 2014). complementarities between species, as well as statistical averaging
The expansion of modern agriculture through a combination between species that respond differently to exogenous shocks such
of intensification and extensification has managed to sustain the as extreme weather, pests and pathogens (Tilman, 1999). This is a
world population explosion that began with the industrial revolu- portfolio effect (Baumgärtner, 2007) that is also provided within crop
tion and accelerated in the early to mid twentieth century (United species by genetic diversity. In the ecological literature, there is a
Nations, 2015). For example, the prevalence of undernourishment particular emphasis on how biologically diverse farming systems can
has declined globally (Fogel, 1997; World Bank, 2016), while the real be less vulnerable to pests and pathogens thanks to these kinds of
prices of agricultural commodities fell quite significantly between mechanism. Pests and pathogens have a very significant impact on
1950 and 2000 (Alston and Pardey, 2014).2 However, the expansion global crop yields, with direct losses estimated to be in the range of
of modern agriculture has had other, less desirable consequences. 20 to 40% (Oerke, 2006; Savary et al., 2012). A famous example is the
Both agricultural intensification – of the prevailing, non- potato famine of 1845-8 that contributed to 1.5 million deaths in Ire-
ecological or unsustainable variety (cf. Bommarco et al., 2013; God- land. Furthermore, expanding the agricultural land area reduces the
fray and Garnett, 2014) – and extensification have been primary extent of natural reserve lands, so that the pool of genetic material
causes of a historically unprecedented loss of global biodiversity. that can potentially be used as an input to agricultural R&D activities
According to the Millennium Ecosystem Assessment (2005), the cur- decreases (Simpson et al., 1996; Rausser and Small, 2000).
rent global rate of species extinction is up to 1000 times higher than In this work we explore the implications of global biodiversity
the background rate that has been estimated from the fossil record. loss, caused by the expansion of modern agriculture, for agricultural
A broader index of global biodiversity has been in decline since 1970 production itself. At the extensive margin, the conversion of natu-
(the first year for which data are available) and there is no statistical ral lands into modern agriculture is undertaken with the intention of
indication that the rate of decline is slowing (Butchart et al., 2010). increasing the production of food, but at the same time the evidence
Local species richness is estimated to have declined by over 10% in we have just presented suggests that it imposes a risk on agricul-
the last 200 years, globally on average (Newbold et al., 2015). tural productivity, through the loss of biodiversity on natural and
In terms of agricultural biodiversity, Khoury et al. (2014) have agricultural land.3 The creation of this risk to global agricultural pro-
documented how global crop production has become less diverse in ductivity results from individual decisions. Profit-maximizing farm-
the last 50 years, in the sense that it has become more dominated ers clear land and plant it with small numbers of high-yielding crop
by a small number of crops. Using data from the Food and Agri- varieties, leading to the loss of biodiversity at the local and global
culture Organization of the United Nations (FAO), they show that, scales. In this process, farmers only partially take into account their
while national food supplies have come to rely on a more diverse set marginal impact on biodiversity, and in turn on agricultural produc-
of crops on average, the opposite is true of the global food supply. tivity (Bowman and Zilberman, 2013; Heal et al., 2004; Weitzman,
Although this might seem paradoxical at first, it is in fact because 2000). Decisions at the individual level about land conversion and
the same crops have been driving both greater diversity in most crop selection thus cause an externality with respect to aggregate
countries (particularly developing countries) and greater similarity production.
globally: wheat, rice, soybeans and oil crops such as palm oil and sun- To study the socially optimal expansion of agricultural land in this
flowers. These are precisely the crops we would expect to become setting, we employ a quantitative two-sector endogenous growth
more prevalent as diets change with rising incomes (Poleman and model of the global economy. This model was first presented in a
Thomas, 1995). In addition to inter-specific diversity of crops, nearly companion paper (Lanz et al., 2017a) and provides an integrated
all countries reporting to the FAO’s global stocktake of crop genetic framework to study the future evolution of global population, eco-
diversity documented the erosion of genetic diversity, with the most nomic growth and food production.4 In the present paper, we extend
commonly identified causes being respectively the replacement of the model to include a biodiversity externality by means of a global-
local varieties as part of the modernization of production systems, level hazard or damage function, which links cropland conversion
and land clearing (FAO, 1996, 2010).
Following a proliferation of research in the last 25 years, there is
now a consensus that biodiversity at different levels increases the 3
From the evidence presented by Khoury et al. (2014), for instance, we might
plant productivity of natural ecosystems, as well as reducing the assume that agriculture at the extensive margin displays lower-than-average crop and
variability of plant productivity (Cardinale et al., 2012). Plant produc- genetic diversity.
4
tivity decreases more than proportionally as biodiversity is lost. Once More specifically, in Lanz et al. (2017a) we set out our framework for integrated
modeling of global population, economic growth and food production, and used the
model to make ‘baseline’ projections. See also Lanz et al. (2017b) where we study the
impact of exogeneous, stochastic shocks to agricultural TFP. However, those papers do
2
Since 2000 they have been on a slightly increasing trend. not include any specific consideration of the effects of biodiversity loss.
262 B. Lanz et al. / Ecological Economics 144 (2018) 260–277
with depreciation of agricultural productivity. This is in the spirit of biodiversity hazard function. The following four sections report our
damage functions in integrated assessment models of climate change results. Section 4 analyzes the socially optimal use of cropland in
(e.g. Nordhaus, 1992; Nordhaus and Boyer, 2000), which, despite the presence of the biodiversity externality, compared with a base-
legitimate concerns about their predictive capabilities, have proved line scenario without the externality. Section 5 then considers the
fundamental to our understanding of that problem. Our growth scenario in which land-conversion and fertility decisions are decen-
model has a number of other features which enable us to derive tralized to households. Section 6 analyzes the effect of assuming
insights about the interactions between global economic growth and lower substitutability of land, and Section 7 briefly reports some
the impact of declining biodiversity on food production. In order to further sensitivity analysis. Section 8 concludes.
characterize agriculture’s role in producing food and sustaining pop-
ulation, the model distinguishes agriculture from other economic
2. The Model
activities. Demand for food in the model is proportional to the size
of the population and it increases with per-capita income. The world This section provides a succinct description of all the compo-
population is endogenously determined by households’ preferences nents of the model and estimation procedure. A more comprehensive
for fertility (Barro and Becker, 1989), as well as the opportunity cost motivation for the structure of the model, selection and estimation
of raising children, which is itself linked with technical progress in of the parameters, baseline projections from the model, as well as
the economy (Galor and Weil, 2000). Technical progress in agricul- sensitivity analysis, is reported in Lanz et al. (2017a).5 The present
ture and the rest of the economy is therefore key and it is modeled as paper rather deploys the model described therein, but adds the haz-
Schumpeterian innovation (Aghion and Howitt, 1992), whereby total ard function that captures the external cost of land conversion to
factor productivity (TFP) growth requires labor as an input to R&D agricultural productivity.
activities. The model is structurally estimated to fit 1960–2010 data
on world GDP, population, TFP growth and cropland area, providing
2.1. The Economy
an empirical framework to study the socially optimal allocation of
land to meet the growing demand for food in the future.
2.1.1. Production and Capital Accumulation
In the absence of the biodiversity externality, we project that
The model comprises two sectors: a manufacturing sector that
world population increases by about 40% by 2050, world GDP dou-
produces the traditional consumption good in one-sector models,
bles and cropland area increases by about 7%. In 2100 the world
and an agricultural sector that produces food to sustain contempora-
population reaches about 12.4 billion. In the presence of the exter-
neous population. In manufacturing, aggregate output is represented
nality, the global-level hazard function links cropland conversion
by a standard Cobb-Douglas production function:
with depreciation of agricultural TFP. This means agricultural inno-
vation is a contest between man-made R&D on the one hand and nat-
z 1−z
ural depreciation from biodiversity loss on the other hand (Goeschl Yt,mn = At,mn Kt,mn Nt,mn , (1)
and Swanson, 2003). To illustrate the implications of the hazard
function for intertemporal resource allocation, we first consider the where Yt,mn is real manufacturing output at time t, At,mn is an index
socially optimal response from a global perspective. Our model sug- of productivity in manufacturing, Kt,mn is capital allocated to man-
gests that a social planner would respond by immediately preventing ufacturing, and Nt,mn is the manufacturing workforce. We assume
the further expansion of cropland, as well as allocating more labor to that technology is Hicks-neutral, so that the Cobb-Douglas functional
agricultural production and agricultural R&D, with negligible effects form is consistent with long-term empirical evidence (Antràs, 2004).
on fertility and population. Given that the planner can select the We use a standard value of 0.3 for the share of capital (see, for
least-cost management strategy to tackle the biodiversity external- example, Gollin, 2002).
ity, and that land is assumed to be relatively easily substituted in Agricultural production requires land services Xt as an input and,
agricultural production (the elasticity of substitution between land following Kawagoe et al. (1986) and Ashraf et al. (2008), we employ
and other factors is 0.6, which is taken from Wilde, 2013), we find a nested constant-elasticity-of-substitution (CES) function to rep-
that the welfare cost is small. resent substitution possibilities between land and a capital-labor
We then consider situations in which managing the external- composite:6
ity might impose greater social costs. First, we solve the model
under the assumption that fertility and land-conversion choices can- s
s−1 s−1 s−1
hK 1−h s
not feasibly be controlled by a global social planner. Instead, these Yt,ag = At,ag (1 − hX ) Kt,ag Nt,ag K + hX X ts
, (2)
choices are made in a decentralized manner by atomistic farming
households, who do not take into account the biodiversity exter-
nality. In this scenario, the planner increases the share of labor in where s is the elasticity of substitution between the capital-labor
agricultural production and agricultural R&D, in order to counter composite and agricultural land. We set s = 0.6 based on long-run
TFP depreciation, but the biodiversity externality still generates a empirical evidence reported in Wilde (2013).7 We come back to the
substantial welfare cost, suggesting that a global policy to limit crop- importance of this parameter for our results in Section 6 below. We
land expansion might have significant value. Second, we consider
lower substitutability of land in agricultural production (the elastic-
ity of substitution between land and other factors is lowered from 5
The GAMS code for the model, replicating the baseline runs reported here, is
0.6 to 0.2) and how this affects the optimal response of the global available on Bruno Lanz’s website.
6
social planner. In this scenario, the reduction of cropland is sig- A Cobb-Douglas function is often used for agriculture, notably in Mundlak (2000)
nificantly lower, and, despite allocating more labor to agricultural and Hansen and Prescott (2002). However, this implies that, in the limit, land is not an
essential input to agriculture, with drawbacks that have been extensively discussed in
production and agricultural R&D, the welfare cost is again larger. relation to the scarcity of non-renewable resources (see Dasgupta and Heal, 1979, for
Finally, we report a sensitivity analysis covering a larger set of model a seminal contribution).
parameters and document how the land use response varies across 7
The estimate by Wilde (2013) is based on 550 years of data from pre-industrial
specifications. England, thus reflecting long-term substitution possibilities, and is estimated in a way
that is consistent with our CES functional form (2). However, external validity may be
The remainder of the paper is structured as follows. Section 2
an issue, in particular when applying results for pre-industrial England to developing
describes the model and its estimation on historical data, while countries with rapidly growing population. In the discussion of the results below, we
Section 3 focuses on the specification and parameterization of the also consider the case of s = 0.2.
B. Lanz et al. / Ecological Economics 144 (2018) 260–277 263
further set hX = 0.25 and hK = 0.3, consistent with data reported in In the model, the cost of children, or marginal cost of effective labor
Hertel et al. (2012). units, is a key driver of fertility decisions and population growth. We
postulate an increasing relationship between the time cost of child-
2.1.2. Innovations and Technological Progress rearing/education and the level of technology:
As in the Schumpeterian model of Aghion and Howitt (1992),
sectoral TFP evolves according to w̄t = wNt,N /Ay
f−1
t ,
At+1,j = At,j • (1 + qt,j S), j ∈ {mn, ag}, (3) where w > 0 is a productivity parameter, f ∈ (0, 1) is an elasticity
representing scarce factors required in child-rearing, At is the index
where S is the maximum growth rate of TFP each period and qt,j ∈ of technology and y > 0 measures how the cost of children increases
[0, 1] is the arrival rate of innovations each period. Actual sectoral with the level of technology.
TFP growth is then the share of the maximum feasible TFP growth This formulation implies that, as the stock of knowledge in
(we set S = 0.05 based on Fuglie, 2012) and depends on the number the economy grows, additions to the stock of effective labor units
of innovations arriving within each time period.8 The rate at which become increasingly costly. In other words the parameter y cap-
innovations arrive in each sector is a function of labor allocated to tures a well-documented complementarity between technology and
sectoral R&D: skills (Goldin and Katz, 1998). An implication of this setup is that
lj labor productivity in fertility and child-rearing activities declines
Nt,Aj with technology. This captures the more detailed mechanism in
qt,j = kj , j ∈ {mn, ag},
Nt Galor and Weil (2000), whereby education decisions respond to the
demand for human capital, itself derived from the prevailing level
where Nt,Aj is labor employed in R&D in sector j, kj > 0 is a pro- of technology (see Lanz et al., 2017a). Growing education require-
ductivity parameter and l j ∈ (0, 1) is an elasticity. This formulation ments raise the cost of each individual child, which implies that
implies that TFP growth increases with the share of labor allocated to technological progress induces a transition from a situation with a
the R&D sector, rather than the absolute amount. This forumulation, large number of children with low human capital (and low edu-
discussed in more details by Chu et al. (2013), avoids the so-called cation cost), to one where households have a smaller number of
‘population scale effect’, which is not supported by data (see Jones, children who possess higher human capital. Therefore, the ‘quality’
1995).9 Furthermore, our representation of R&D implies decreasing of children required to keep up with technology will be favored over
returns to labor in R&D through the parameter l j , which captures the the quantity, and a demographic transition will occur as education
duplication of ideas among researchers (Jones and Williams, 2000). requirements increase over time (Galor and Weil, 2000). Further,
The parameter kj is normalized to unity to ensure that TFP growth f captures the fact that the cost of child-rearing over a period of
is bounded between 0 and S, and the parameters l mn and l ag are time increases more than proportionally with the number of children
estimated as described below. (see Barro and Sala-i Martin, 2004, p.412, and Bretschger, 2013). The
parameters determining the cost of fertility and how it evolves over
2.1.3. Labor and Population Dynamics time (w, f and y) are estimated from the data, as described below.
In each period, the change in world population is given by the Population dynamics are further affected by food availability, as
difference between fertility nt and the rate at which population exits measured by agricultural output. Specifically, in each period agricul-
the labor force dN : tural production is consumed entirely to sustain contemporaneous
population:10
Nt+1 = Nt (1 − dN ) + nt Nt , N0 given. (4)
Yt,ag = Nt f̄t ,
Since population equals the total labor force in our model, dN is
the inverse of the expected working lifetime, which we set to 45 where f̄t is per-capita food demand, i.e. the quantity of food required
years (hence the ‘working mortality rate’ is dN = 0.022). to maintain an individual in a given society. We further specify per-
Fertility derives from the allocation of labor to child-rearing capita demand for food as a concave function of per-capita income:
activities, so that child-rearing competes with other labor-market j
activities: Yt,mn
f̄ = n • ,
Nt
nt Nt = w̄t • Nt,N ,
where n is a scale parameter and j > 0 is the income elasticity of
food consumption. Food demand thus captures both physiological
where Nt,N is labor allocated to child-rearing activities and w̄t is an
requirements (e.g. minimum per-capita caloric intake) and a posi-
inverse measure of the time cost of producing effective labor units.
tive relationship between income and the consumption of calories.
We set the income elasticity of food demand j = 0.25, which is
consistent with evidence across countries and over time reported in
8
In the original work of Aghion and Howitt (1992), time is continuous and the Subramanian and Deaton (1996), Beatty and LaFrance (2005), and
arrival of innovations is modeled as a Poisson process. Our representation is qualita-
tively equivalent, but somewhat simpler, as qt,j implicitly uses the law of large number
Logan (2009). The parameter measuring food consumption for uni-
to smooth out the random nature of innovations over discrete time periods. tary income (n) is calibrated such that the demand for food in 1960
9
Scaling the labor force in R&D by Nt is consistent with micro-foundations of represents about 15% of world GDP, which corresponds to the GDP
more recent representations of technological change, such as Dinopoulos and Thomp-
son (1998), Peretto (1998) and Young (1998), where R&D activities simultaneously
develop new products and improve existing ones. An implication is that the number
10
of products grows with population, thereby diluting R&D inputs and avoiding the pop- Note that food consumption only indirectly contributes to social welfare. In par-
ulation scale effect. Another strategy to address the scale effect involves postulating ticular, the level of population enters the social welfare criterion (together with the
a negative relationship between labor productivity in R&D and the existing level of utility of per-capita consumption of the manufacturing good), so that food availability
technology, giving rise to “semi-endogenous” growth models (Jones, 1995, 2001). In will affect social welfare through the impact of the subsistence requirement on pop-
this setup, however, long-run growth is only driven by population growth, which is ulation. For a similar treatment, see Strulik and Weisdorf (2008), Vollrath (2011) and
also at odds with the data (Ha and Howitt, 2007). Sharp et al. (2012).
264 B. Lanz et al. / Ecological Economics 144 (2018) 260–277
share of agriculture reported in Echevarria (1997). This implies n = where dK is the per-period depreciation rate. Because we solve for
0.4. the social planner solution of the problem (see below), savings-
cum-investment decisions mirror those of a one-sector economy. In
2.1.4. Land other words, only manufacturing output can be used to invest in the
As a primary factor, the land input to agriculture has to be con- stock of physical capital (see Ngai and Pissarides, 2007, for a simi-
verted from a total stock of available land X̄ by applying labor.11 Over lar treatment of savings in a multi-sector growth model), whereas
time the stock of land used in agriculture develops according to: agricultural output only produces food that is directly consumed to
sustain contemporaneous population.
e
Xt+1 = Xt (1 − dX ) + x • Nt,X , X0 given, Xt ≤ X̄, (5)
2.2. Solution and Optimal Control Problem
where Nt,X is labor allocated to land-clearing activities, x > 0 mea-
sures labor productivity in land-clearing activities, e ∈ (0, 1) is an We consider the planner’s problem of selecting the allocation of
elasticity, and the depreciation rate dX measures how fast converted labor and capital as well as the saving rate to maximize the utility of
land reverts back to natural land. We assume that the period of a representative dynastic household.12 Specifically, a representative
regeneration of natural land is 50 years, so that dX = 0.02. The household chooses paths for Nt,j , Kt,j , and Ct by maximizing Eq. (6)
parameters x and e are estimated from the data as described below. subject to technological constraints (1), (2), (3), (4), (5), (7), (8) and
The overall constraint on land X̄ is set to 3 billion hectares, a num- resource allocation constraints for capital and labor:
ber reported in Alexandratos and Bruinsma (2012), although the
constraint does not bind in the numerical simulations. Kt = Kt,mn + Kt,ag , Nt = Nt,mn + Nt,ag + Nt,Amn + Nt,Aag + Nt,N + Nt,X .
2.1.5. Preferences and Savings The numerical model is solved as a constrained non-linear opti-
The utility function of a representative household is defined mization problem and thus mimics the welfare maximization pro-
over own consumption of the manufactured good ct , fertility nt and gram by directly searching for a local optimum of the objective
the utility its children will experience in the future Ui,t+1 . More function (the discounted sum of utility) subject to the require-
specifically, we represent household preferences with the recursive ment of maintaining feasibility as defined by the constraints of the
formulation of Barro and Becker (1989): problem.13
A schematic representation of the model is provided in Fig. 1.
1−c
ct − 1 1−g This shows interlinkages between quantities that are endogenously
Ut = + bnt Ui,t+1 , determined by the model. Let us highlight the different elements of
1−c
the planner’s response to manage the externality. Agricultural land
where c is the elasticity of marginal utility of consumption of the serves as a basic input to the production of food, which sustains the
manufactured good or in other words the inverse of the intertem- population/labor supply. Substituting the land input with capital and
poral elasticity of substitution, b is the discount factor and g is an labor is mainly driven by the opportunity cost of not using these
elasticity determining how the utility of parents changes with nt . inputs for other purposes, as well as the elasticity of substitution s.
The objective function is given by the utility function of the dynastic Moreover, a decline in agricultural land area can be compensated
head and obtained by successive substitution of the recursive utility through R&D, allocating labor towards the growth of agricultural TFP.
function (see Lanz et al., 2017a, for a detailed derivation): Another strategy to cope with the externality is to reduce the
demand for food. In the context of our model, there are two ways
∞
to achieve this. First, the planner can reduce population by reducing
1−g (Ct /Nt )1−c − 1
U0 = bt Nt , (6) optimal fertility. Therefore an important feature of our model is that,
1−c
t=0 instead of modeling mortality as a function of food supply, popula-
tion can respond to the availability of food through a reduced birth
where Ct = ct Nt is aggregate consumption at t.
rate relative to the baseline. Second, since the demand for food is
The parameterization of the objective function is as follows. First,
proportional to per-capita manufacturing consumption, the planner
the elasticity of intertemporal substitution is set to 0.5 in line with
can reduce manufacturing consumption. This can be done either by
estimates by Guvenen (2006). In the model, this corresponds to
reducing factor inputs to manufacturing output (including manufac-
c = 2. Second, we set g = 0.001 so that altruism towards the wel-
turing R&D), or by increasing the saving rate. The relative magnitude
fare of children remains almost constant as the number of children
of these different responses is determined by the parameters we use
increases, while at the same time ensuring concavity of the objec-
in the model, which are summarized in Table 1.
tive function. This implies that the objective function is very close to
being Classical Utilitarian. Third, we set the discount factor to 0.99,
2.3. Estimation of the Model
which corresponds to a pure rate of time preference of 1% per year.
Aggregate consumption derives from manufacturing output, As mentioned above, parameters determining the cost of fertility
which can alternatively be invested into a stock of capital: (w, f, y), labor productivity in R&D (l mn , l ag ) and labor productiv-
ity in land conversion (x, e) are estimated by fitting the model to
Yt,mn = Ct + It , (7) 1960–2010 trajectories for world GDP (Maddison, 1995; Bolt and
van Zanden, 2013), population (United Nations, 1999, 2013), crop
where Ct and It are aggregate consumption and investment respec-
tively. The accumulation of capital is then given by:
12
The fact that we solve the model as a social planner problem simplifies the
Kt+1 = Kt (1 − dK ) + It , K0 given, (8) notation and allows us to exploit efficient solvers for constrained non-linear opti-
mization. However, it abstracts from externalities that would arise in a decentralized
equilibrium. As discussed below, however, market imperfections prevailing over the
estimation period will be reflected in the parameters that we estimate from observed
11
Note that land productivity can potentially vary after conversion, as typically crop- trajectories.
13
land on recently cleared forest has relatively high productivity. Instead, our model The numerical problem is formulated in GAMS and solved with KNITRO (Byrd et
captures long-run soil productivity, and the associate incentives to convert it. al., 1999, 2006) specialized software for constrained non-linear programs.
B. Lanz et al. / Ecological Economics 144 (2018) 260–277 265
land area (Goldewijk, 2001; Alexandratos and Bruinsma, 2012) and in Appendix A. First, we impose specific parameter values for a num-
sectoral TFP (Martin and Mitra, 2001; Fuglie, 2012). The estimation ber of quantities that are standard in the literature (see discussion
procedure includes three main steps, and is discussed in more detail above). Second, we calibrate values for the state variables to initialize
the model in 1960. Third, we define a minimum distance criterion for
GDP (Yt,mn +Yt,ag ), population (Nt ), crop land (Xt ), and TFP (At,mn , At,ag )
Table 1
as a way to select the vector of parameters that best fits the observed
Benchmark parameter values used for the quantitative results.
trajectories. Using simulation methods to select the vector of param-
eters that minimizes our criterion, we find that the model closely fits
Imposed parameters: the targeted data (see goodness-of-fit measures in Appendix A).
Share of capital in manufacturing z = 0. 3
Share of capital in capital-labor composite for agriculture hK = 0.3
Share of land in agriculture hX = 0.25 3. Specification of the Biodiversity Externality
Substitutability between land and the capital-labor s = 0.6
composite in agriculture As described above, innovation in agriculture is driven by allocat-
Yearly rate of capital depreciation dK = 0.1
ing labor to R&D activities, while land acts as an input to agricultural
Maximum increase in TFP each year S = 0.05
Labour productivity parameter in R&D kmn = kag = 1 production. We now introduce the critical second role of land allo-
Inverse of the intertemporal elasticity of substitution c=2 cation, which is to modulate biodiversity loss and in turn the depre-
Elasticity of altruism towards future members of the dynasty g = 0.001 ciation of agricultural TFP. This is because the rate of depreciation
Income elasticity of food demand j = 0.25 of agricultural TFP is endogenous and depends on the size of the
Discount factor b = 0.99
agricultural system in terms of global cropland area.
Initial values and calibrated parameters: Following Goeschl and Swanson (2003), we assume that agricul-
Initial value for population N0 = 3.03 tural R&D proceeds in a similar fashion to R&D in the manufacturing
Initial the stock of converted land X0 = 1.35 sector, but is subject to depreciation shocks as a result of biodi-
Initial value for TFP in manufacturing A0,mn = 4.7
versity loss.14 At each point in time, there is a hazard rate that a
Initial value for TFP in agriculture A0,ag = 1.3
Initial value for capital stock K0 = 20.5 negative shock of a given size will occur, which is modeled as a
Food consumption for unitary income n = 0.4 Poisson process. As we work in discrete time, integrating this over
Exogenous mortality rate dN = 0.022 unit time-steps yields a process for the evolution of agricultural
Rate of natural land reconversion dX = 0.02 TFP that extends our representation of Aghion and Howitt’s (1992)
Schumpeterian model of creative destruction set out in Section 2.1.2.
Estimated parameters:
Labour productivity parameter in child-rearing w = 0.153 Formally:
Elasticity of labor in child-rearing f = 0.427
Elasticity of labor productivity in child-rearing w.r.t. y = 0.089
At+1,ag = At,ag • (1 + qt,ag S − 0t S). (9)
technology
Elasticity of labor in manufacturing R&D l mn = 0.581
Elasticity of labor in agricultural R&D l ag = 0.537
Labour productivity in land conversion x = 0.079 14
Consistent with the various lines of evidence presented in the Introduction, this
Elasticity of labor in land conversion e = 0.251
captures biodiversity loss at multiple levels, including genetic, species and ecosystem
Notes: This table reports the full set of benchmark parameters used for the simulations. diversity.
266 B. Lanz et al. / Ecological Economics 144 (2018) 260–277
0.001
0.0005
Man-made R&D depreciates at the rate 0t , which is a function of the interpreted as the expected value of TFP shocks, which is consistent
amount of cultivatable land allocated to crop production: with evidence that biodiversity loss on and off farms reduces mean
crop yields. It could also be interpreted as the certainty-equivalent
0t = kD (Xt )lD , (10) value of TFP shocks, which would mean that mean-preserving
spreads in crop yields as a result of biodiversity loss are themselves
sufficient to increase agricultural TFP depreciation in the model. The
where kD ≥ 0 and l D > 1, which is consistent with the relationship certainty-equivalent interpretation is consistent with a risk-averse
between biodiversity and plant productivity in natural ecosystems planner and is similar to the treatment of climate risks in Nordhaus’
(Hooper et al., 2012). The expected growth rate of agricultural TFP work with the DICE and RICE family of integrated assessment models
is then the net result of the flow of innovations from man-made (Nordhaus and Boyer, 2000).
R&D (or moves up the TFP ladder) and the arrival of biological
hazards associated with the scale of agriculture (moves down the
technological ladder). 4. Results: Optimal Management of the Biodiversity Externality
While the ecological processes underpinning Eq. (10) are well
documented (see the Introduction), there exists as yet no empirical In this section we consider the socially optimal response under
evidence that can directly guide the parameterization of an agri- the two hazard schedules, as well as in a baseline scenario where the
cultural biodiversity hazard function at the global level. Separately biodiversity externality is absent. What is shown is the outcome of a
identifying man-made innovation and natural depreciation of agri- representative household cum social planner controlling all the vari-
cultural TFP is challenging in empirical data.15 The large body of ables of the problem in order to maximize welfare. In the presence of
ecological experiments is strongly suggestive of a negative relation- the biodiversity externality, this implies that the social planner can
ship between species diversity and plant productivity that holds internalize the negative impact of land conversion on agricultural
globally (Cardinale et al., 2012; Hooper et al., 2012), but many of TFP by conserving natural land and reducing fertility, among other
these experiments have been conducted outside cropping systems. things. In the presence of the externality, it is indeed appropriate to
The empirical evidence from agricultural economics (Di Falco, 2012) interpret the representative agent as a social planner rather than a
is directly relevant, but it is simply more sparse at this point. We household.16 To begin with, we are particularly interested in how the
therefore select the parameters determining the scale of the exter- social planner would optimally manage the externality. That is why
nality, kD and l D in Eq. (10), in order to simply illustrate the processes we do not impose the externality on an economy that is constrained
at play and how these impact the macro-economy. in how it can respond. We return to this issue in the next section.
We run with two specifications, low and high, as shown in Fig. 2. In the following, we first report aggregate impacts for popula-
In the low hazard function, kD = 1.9e − 5 and l D = 10. In the tion, land, technology and agricultural output. We then turn to the
high hazard function, kD = 6.65e − 5 and again l D = 10. Both implied reallocation of labor across sectors in response to the exter-
functions imply that the TFP depreciation rate is almost insignificant nality. Finally, we quantify the welfare cost of the optimal policy in
for the area of cropland up to and including 2010. Under the low terms of consumption of the manufactured good and food.
hazard function, the impact on TFP depreciation of additional land
conversion beyond the 2010 area remains small, but under the high
4.1. Aggregate Impacts
hazard function it rises more sharply. Along our baseline scenario
(see below), the global area of cropland reaches around 1.8 billion Fig. 3 displays trajectories for world population (panel a) and
hectares in 2100 and the high hazard function associates this with a global cropland (panel b). The baseline scenario delivers a world pop-
TFP depreciation rate of around 0.1%. This compares with TFP growth ulation of 9.8 billion in 2050 and 12.4 bn. in 2100. This is towards
from man-made innovation of around one per cent in 2010, declining
over time to 0.5%.
The hazard function illustrates the impact of cropland expansion
on agricultural TFP. Since our model is deterministic, it could be 16
In the presence of externalities, the social planner’s solution will generally dif-
fer from the decentralized allocation. The Schumpeterian model of growth that we
use incorporates positive externalities to R&D activities. However, by estimating the
model on the past 50 years of data, we rationalize observed outcomes ‘as if’ they
15
However, because the model is fitted to the last 50 years of data on agricul- resulted from the decisions of a social planner, so the concepts of a representative
tural TFP growth, estimates of labor productivity in agricultural R&D will reflect the household and a social planner are equivalent in the baseline scenario. They are not
occurrence of biological hazards in the past. equivalent when the new land-use externality is introduced, however.
B. Lanz et al. / Ecological Economics 144 (2018) 260–277 267
14
12 1.75
10
8 1.5
6
4 1.25
Baseline without externality Baseline without externality
2 Low hazard Low hazard
High hazard High hazard
0 1
2010 2050 2100 2010 2050 2100
Years Years
0.015
-0.0005
0.01
-0.001
0.005
Baseline without externality
Low hazard
High hazard
-0.0015 0
2010 2050 2100 2010 2050 2100
Years Years
(e) Growth rate of agricultural TFP (f) Growth rate of agricultural yield
0.01 0.02
Baseline without externality Baseline without externality
Low hazard Low hazard
High hazard High hazard
0.005 0.01
0 0
2010 2050 2100 2010 2050 2100
Years Years
Fig. 3. Projections for population, land conversion and agricultural technology, 2010–2100.
the upper end of the 20–80 % confidence interval of the latest UN Thus our results conform with the widespread expectation that the
projections (United Nations, 2015). The area of cropland reaches long-standing processes of global population growth and cropland
1.73 billion hectares in 2050 and stabilizes at 1.78 bn. ha. towards the conversion are in decline. This reflects a shift from a quantity-
end of the century, which is in the middle of the range of projections based economy with rapid population growth and associated land
from the AgMIP model comparison exercise under a broadly com- conversion, towards a quality-based economy with investments in
parable baseline scenario (Schmitz et al., 2014). Therefore we find technology and education, and lower fertility.
that a steady state in land conversion is consistent with sustained Turning to the impact of the biodiversity externality on agricul-
growth in agricultural output, even though we are only mildly opti- tural TFP, it is plain to see that population is almost unaffected (panel
mistic about future technological progress: agricultural TFP growth a); under the high hazard function it is just 0.025 bn. lower in 2100.
in 2010 is around one per cent per year and it declines thereafter So the optimal response of the planner only involves a small reduc-
(see Fig. 3, panel e). An important feature of these projections is tion in fertility. By contrast – and even though the rate of cropland
that the growth rates of the variables decline towards a balanced conversion is in decline in the baseline scenario – the externality has
growth path, where population, land and capital reach a steady state. a large impact on socially optimal land conversion (panel b). Under
268 B. Lanz et al. / Ecological Economics 144 (2018) 260–277
(a) Agricultural output index (1960 = 100) (b) Growth rate of agricultural output
700 0.025
Baseline without externality
Low hazard
High hazard
600 0.02
500
0.015
400
0.01
300
0.005
200 Baseline without externality
Low hazard
High hazard
100 0
2010 2050 2100 2010 2050 2100
Years Years
the high hazard function, cropland immediately declines and reaches 4.2. Changes in Sectoral Labor Allocation
a steady state at around 4% below the 2010 level. This corresponds to
a reduction of cropland of 60 million ha. relative to 2010, as opposed Fig. 5 reports on how the allocation of labor differs across scenar-
to an increase of around 150 mn. ha. under the baseline scenario. ios. This is a potentially important lever for the planner. The baseline
The planner puts an immediate halt to further land conversion and scenario without the externality is used as the benchmark and we
moreover leaves a portion of global cropland that is in use in 2010 report percentage points differences in labor shares allocated to dif-
to be reclaimed by nature, in order to mitigate the negative impact ferent activities under the low and high hazard schedules. In the
of the scale of the cropping system on productivity via biodiversity presence of the biodiversity externality, we have already seen that
loss. Under the low hazard function, cropland stays close to its 2010 the social planner significantly reduces cropland area relative to the
value throughout the century, increasing very slightly in the first few baseline. Panels (a) and (c) show that food supply is maintained by
decades. increasing the shares of labor devoted to agricultural production and
The impact of land conversion on agricultural TFP growth is agricultural R&D. This substitutes for the reduced land input and
illustrated in panels (c) and (d). Under the high hazard function, deals with residual natural TFP depreciation due to the scale of the
TFP depreciation peaks immediately at about 0.04%, but then the cropping system. In 2010, the share of labor allocated to agricultural
decline in cropland area brings depreciation back down towards a R&D is about 0.5 percentage points higher under the high hazard
steady-state rate of about 0.027%. Under the low hazard function, the function than it is in the baseline, which roughly corresponds to a
long-term rate of TFP depreciation is roughly half of that, at 0.015%. 10% increase in the workforce allocated to agricultural R&D.18
In response to natural TFP depreciation, the planner increases the Panel (b) shows that the share of labor allocated to manufactur-
level of man-made agricultural R&D, the result of which is shown in ing is initially higher in the presence of the externality, as labor that
panel (d). Consequently overall growth in agricultural TFP is actu- would have been employed in land conversion is reallocated, but it
ally higher in the presence of the externality than it is in the baseline quickly converges to a share very close to the baseline. The share
scenario (panel e), compensating for the smaller area of land used in of labor allocated to manufacturing R&D is very similar in all three
agriculture. scenarios.
In panel (f) of Fig. 3, we report the growth rate of agricultural
yields, i.e. the rate of change of agricultural output per unit of land 4.3. Per-capita Consumption and Social Welfare
area. This is an important measure of the resources allocated to
agriculture for a given area of cropland. Under the baseline sce- Fig. 6 shows the implications of the externality for consumption
nario, agricultural yield growth starts at around 1.2% per year in per capita of the two goods. We report the paths taken by consump-
2010 and declines towards 0.5% per year in 2100. In the presence of tion per capita of the manufactured good (panel a) and food (panel
the externality, yield growth is initially higher, but then converges b) under the two hazard schedules, as a percentage of the baseline
to the baseline growth rate by 2100. This suggests an important run. In the presence of the biodiversity externality, which is after
reallocation of factors to increase agricultural yields early on, but a all a supply-side shock to the agricultural sector, consumption of
convergence of the three paths in the long run. both goods is lower. Given our representation of preferences in Eq.
Fig. 4 shows an index of agricultural output (panel a) and its (6), consumption per capita of the manufactured good is in fact a
associated growth rate (panel b).17 It demonstrates that agricultural measure of equivalent variation (as a percentage of baseline con-
output is virtually the same in all three scenarios, despite differ- sumption). Thus the welfare cost of the externality under the high
ences in the composition of inputs (recall that population is almost hazard schedule, optimally managed, rises to about 0.3% in the sec-
identical in all three scenarios too). Our model suggests an increase ond half of this century. Under the low hazard schedule this welfare
of agricultural output of 67% between 2010 and 2050, in line with cost is closer to 0.1%. Because consumption of the manufactured good
other projections (Alexandratos and Bruinsma, 2012; Robinson et al., and of food are complementary in our model (which comes from the
2014), and a doubling of agricultural output by 2100.
18
While the proportion of labor employed in agricultural R&D (around five per cent)
17
The base year of the index is 1960, suggesting that agricultural output increases may appear high, it should be noted that it includes any labor time dedicated to
by a factor of 2.8 over the estimation period. While this figure is not directly targeted improving factor productivity. This includes many informal activities taking place in
by the estimation, it is very close to the factor of 2.7 reported to apply over the same developing economies in particular, such as seed selection or improving irrigation
period in Alexandratos and Bruinsma (2012). practices.
B. Lanz et al. / Ecological Economics 144 (2018) 260–277 269
0.1
0.1
0.4
0.2
0 -0.1
2010 2050 2100 2010 2050 2100
Years Years
positive relationship between household income and food prefer- further expansion of cropland (even under the low hazard sched-
ences), food consumption follows a similar trajectory and is quickly ule, cropland barely expands after 2010), as well as changes in the
lower in the presence of the externality, though the difference is very allocation of labor to different uses, in particular a shift towards
small, in line with the empirical evidence on income elasticity of food agricultural production and agricultural R&D. However, in reality
demand. the mechanisms promoting centralized choice of land conversion
and fertility are at best weak and at worst non-existent, espe-
cially at the global level. Therefore it is of interest to analyze a
5. Results: Fixed Population and Land scenario, in which the biodiversity externality is present, but indi-
vidual farms/households do not internalize it in their fertility and
The previous section focused on how a global social plan- land-conversion decisions.
ner would optimally manage the biodiversity externality. On the In this section, we hence evaluate a scenario in which the high
assumption that agriculture continues to employ ecologically dam- hazard schedule applies, but the trajectories for population and crop-
aging practices, our results broadly suggested a moratorium on land area are constrained to follow the baseline scenario without
-0.2
-0.4 -0.2
2010 2050 2100 2010 2050 2100
Years Years
0.01
-0.001
0.005
Baseline without externality
High hazard: social optimum
High hazard under baseline population and land
-0.0015 0
2010 2050 2100 2010 2050 2100
Years Years
(c) Growth rate of agricultural TFP (d) Growth rate of agricultural yield
0.01 0.02
Baseline without externality
High hazard: social optimum
High hazard under baseline population and land
0.005 0.01
Fig. 7. Projections for population, land conversion and agricultural technology, 2010–2100.
the externality (from above). We still posit a social planner in this agricultural production is quite different to the unconstrained social
scenario, but the planner’s response to the externality is curtailed optimum. Because land conversion continues along its baseline tra-
and may only involve changing the share of labor allocated to differ- jectory, there is not the immediate need to boost labor in agricultural
ent uses, the share of capital allocated to the two sectors, as well as production that exists under the unconstrained social optimum.
the overall consumption/savings margin. For comparison, we include However, because continued land conversion puts more and more
both the baseline without the externality and the social optimum in downward pressure on agricultural productivity, the share of labor
the presence of the high hazard schedule, as explored in the previous in agriculture steadily increases and it overtakes the unconstrained
section. social optimum by the end of the century. For the same reason, the
In the following, we again focus on three different model out- share of labor allocated to agricultural R&D is initially lower under
comes: (i) aggregate impacts; (ii) reallocation of labor across sectors; decentralized fertility and land-conversion choices than it is at the
and (iii) welfare costs. unconstrained social optimum, but the former overtakes the latter
just after 2050 and by the end of the century it is markedly higher.
5.1. Aggregate Impacts Importantly, the constrained optimal solution implies that labor is
diverted away from manufacturing production (panel b) and manu-
Fig. 7 displays trajectories for agricultural TFP (panels a–c) and facturing R&D (d), which will induce large welfare costs, as we shall
yield growth (d); population and land are omitted since these tra- now see.
jectories are fixed to their baseline levels. Panel (a) shows that
decentralized fertility and land conversion choices lead to a much
larger rate of agricultural TFP depreciation, which increases to about
0.1% by the end of the century, around four times higher than 5.3. Per-capita Consumption and Social Welfare
under the socially optimal response to the externality. This carries
through to the overall growth rate of agricultural TFP (panel c); Fig. 9 (panel a) provides estimates of these welfare costs, in the
rather than being higher than the baseline to compensate for the form of the path of consumption per capita of the manufactured
reduced land input, under the scenario of decentralized fertility and good. Compared with the unconstrained social optimum, consump-
land-conversion decisions it is lower and, as a consequence, yields tion per capita is much lower, starting at 0.8% below the baseline and
track the baseline (panel d). falling approximately linearly to 2.9 % below the baseline by the end
of the century. So, while changing the share of labor devoted to dif-
5.2. Changes in Sectoral Labor Allocation ferent uses is able to match food supply and demand and sustain a
world population of more than 12 bn. by 2100, this comes at a sub-
Fig. 8 reports on labor shares in the two planning scenarios, stantial cost in terms of living standards (as proxied by consumption
relative once again to the baseline without the externality. Panel per capita of the manufactured good). The complementarity between
(a) shows that the trajectory of the share of labor employed in living standards and food consumption explains why per-capita food
B. Lanz et al. / Ecological Economics 144 (2018) 260–277 271
0.1
0.1
0
-0.1
0
-0.2
High hazard: social optimum High hazard: social optimum
High hazard under baseline population and land High hazard under baseline population and land
-0.1 -0.3
2010 2050 2100 2010 2050 2100
Years Years
0.4 0
0.2 -0.1
0 -0.2
2010 2050 2100 2010 2050 2100
Years Years
consumption also falls approximately linearly over the course of the against agricultural TFP depreciation. One key assumption under-
century (panel b). By 2100 it is more than 0.6% less than in the lying this result is the substitutability of land in agricultural pro-
baseline. duction. Our simulations so far have assumed s = 0.6, which is
based on empirical evidence reported in Wilde (2013). There is, how-
6. Sensitivity Analysis: Land Use ever, considerable uncertainty about the value of this parameter,
particularly when considering the global aggregate. Here we con-
The results of Section 4 demonstrated that the optimal global sider an alternative scenario in which s = 0.2, meaning it is more
management of the biodiversity externality involves a very differ- difficult to substitute land with capital and labor. To get a consis-
ent future for agricultural land-use management. The social planner tent baseline, it is necessary to re-estimate the parameters of the
immediately protects remaining natural land reserves as a buffer model. We therefore run both the baseline scenario without the
-1 -0.2
-1.5 -0.4
-2
-0.6
-2.5
-3 -0.8
2010 2050 2100 2010 2050 2100
Years Years
externality and the social optimum under the high hazard sched- falling immediately, it increases fractionally over several decades.
ule when s = 0.2, and we do the same when s = 0.6. We Because land is more integral to food production when s = 0.2,
start with aggregate results, then move to labor shares and finally panel (c) shows that the social planner must tolerate a relatively high
welfare. level of natural depreciation of agricultural TFP, reaching about 0.05%
later in the century. The planner compensates for this by allocating
more resources to agricultural R&D, the results of which can be seen
6.1. Aggregate Impacts in panel (d).
As a consequence of low substitutability of land, the planner
Fig. 10 reports trajectories for population (panel a) and cropland drives up the rate of agricultural TFP innovation with or without
(panel b). As we would expect, when there is less scope to substitute the externality. In the presence of the externality, the rate of agri-
capital and labor for land in agricultural production, baseline crop- cultural TFP innovation is around 1.2% initially, which is about a
land expansion is higher. In 2050 it reaches 1.77 bn. ha., on its way to fifth higher than when s = 0.6. Panel (e) shows that this in
a steady state of around 1.82 bn. ha. towards the end of the century. turn results in higher net agricultural TFP growth, while panel (f)
However, the population trajectory is almost identical. shows that, with relatively more cropland in use, yields are lower
Under low substitutability, the social planner responds to the bio- when substitutability is lower. Finally, once again agricultural out-
diversity externality by slowing cropland expansion, but not as fast put is virtually the same in all the scenarios (and hence is not
or as far as when s = 0.6. Instead of the area of global cropland reported).
14
12 1.75
10
8 1.5
0.01
-0.001
0.005
Baseline with σ = 0.6
High hazard: social optimum with σ = 0.6
High hazard: social optimum with σ = 0.2
-0.0015 0
2010 2050 2100 2010 2050 2100
Years Years
(e) Growth rate of agricultural TFP ( ) (f) Growth rate of agricultural yield
0.015 0.02
Baseline with σ = 0.6 Baseline with σ = 0.6
High hazard: social optimum with σ = 0.6 High hazard: social optimum with σ = 0.6
Baseline with σ = 0.2 Baseline with σ = 0.2
High hazard: social optimum with σ = 0.2 High hazard: social optimum with σ = 0.2
0.01
0.01
0.005
0 0
2010 2050 2100 2010 2050 2100
Years Years
Fig. 10. Projections for population, land conversion and agricultural technology, 2010–2100.
B. Lanz et al. / Ecological Economics 144 (2018) 260–277 273
0.1
0.1
0
-0.1
High hazard: social optimum with σ = 0.6 High hazard: social optimum with σ = 0.6
High hazard: social optimum with σ = 0.2 High hazard: social optimum with σ = 0.2
-0.1 -0.2
2010 2050 2100 2010 2050 2100
Years Years
0.8
0
0.6
0.4
0.2
0 -0.1
2010 2050 2100 2010 2050 2100
Years Years
Fig. 11. Sectoral labor shares relative to Baseline (in percentage points).
6.2. Changes in Sectoral Labor Allocation much more labor to agricultural R&D (panel c), initially more than
twice as much as when s = 0.6. In fact, s = 0.2 implies a higher
Fig. 11 compares the allocation of labor with and without the complemetarity between primary factors in agricultural production.
externality, for the two different elasticities of substitution of land Hence as land becomes more costly to use (because of the external-
with capital-labor. That is, we plot the difference between the social ity), the planner uses less of all the inputs, and instead focuses on
optimum under the high hazard and the baseline for s = 0.2 and we increasing agricultural TFP. Note also that, in addition, because crop-
do the same for s = 0.6. Given that the baselines are not far apart land area is larger with s = 0.2, more labor is allocated to land
from each other, we can also compare the two social optima to a first conversion.
approximation. We can see that, in the presence of an externality,
lower substitutability of land leads to a reduction of the share of labor 6.3. Per capita Consumption and Social Welfare
in agricultural production (panel a), manufacturing (b) and manufac-
turing R&D (d). Instead, to manage the biodiversity externality under Fig. 12 shows that lower substitutability of land leads to a consid-
conditions of lower substitutability of land, the planner allocates erably higher reduction of per capita consumption, and hence higher
-0.2 0
-0.4
-0.6 -0.2
-0.8
-1 -0.4
2010 2050 2100 2010 2050 2100
Years Years
welfare cost of the biodiversity externality. Consumption per capita by the balance of a growing body of empirical evidence from ecol-
of the manufactured good reaches a minimum of about 0.8% lower ogy and agricultural/ecological economics. Furthermore it is easy
than the baseline in the second half of the century. Food consumption to reconcile the empirical evidence with theory. The third part of
also declines, although to a lesser extent, being around 0.3% lower the argument is that reductions in agricultural productivity that are
than in the baseline (no externality) situation. due to the expansion of modern, intensive agriculture are partly
externalized by the farmers that cause them. This follows logically
7. Further Sensitivity Analysis from the types of ecological mechanism at play, for example how
As well as the substitutability of land in agriculture, a number of pests and pathogens operate across landscapes comprising many
other imposed model parameters (Table 1) are candidates for sensi- farms. Together these suggest the expansion of modern, intensive
tivity analysis, in particular: the income elasticity of food demand j, agriculture not only increases agricultural production through the
the discount factor b and the mortality rate dN . Since the set of results cultivation of more land, it also has a negative effect on agricultural
we could present here is enormous, we focus on the sensitivity of productivity. They also suggest growth in agricultural productivity
land conversion to the parameter variations (Table 2). Note that for can be conceptualized as a contest between man-made innovations
each variation of parameters we re-estimate the model over the 1960 and biological hazards (Goeschl and Swanson, 2003).
to 2010 period in order to obtain comparable baseline trajectories. We explore the implications of this contest using a quantita-
Lanz et al. (2017a) showed that baseline projections with the model, tive macro-economic model, which owes its intellectual debts to
especially of population and land trajectories, are robust to variations the micro-economic theory of fertility choice (e.g. Barro and Becker,
in parameters. However parameter variations may still determine 1989), especially the trade-off between the ‘quality’ and quantity
very different responses to the land conversion externality of children emphasized by unified growth theory (e.g. Galor and
It is immediately clear that our conclusion that optimal man- Weil, 2000), and endogenous growth theory (e.g. Aghion and Howitt,
agement of the biodiversity externality involves reduced expansion 1992). To the best of our knowledge this is the first such modeling
of global cropland is robust to these parameter variations, albeit it exercise. Our model is novel in how it integrates these different com-
varies in extent. Starting with higher substitutability of land (s = 1), ponents, as well as treating agricultural/natural land as a stock and
the reduction in converted cropland area relative to the baseline is incorporating the biodiversity externality.
beyond our benchmark case studied above, but only very slightly. Running the model without the biodiversity externality, we
The reduction in converted cropland area is also higher if the mortal- project large increases in world population and agricultural output,
ity rate is lower. This is due to the fact that individuals in the model which is based on continued expansion of the area of cropland, albeit
live longer, extending the productivity of workers in the model, so growth in the scale of the economy and food system gradually slows
that labor is available to substitute for land. The effect is economi- down. In the presence of the externality, population and agricul-
cally significant, as the decline in agricultural land is five percentage tural output are virtually unaffected, but the policy that maximizes
points larger as compared to our benchmark results. social welfare from the global point of view is one that entirely –
Conversely, the reduction in cropland area relative to the bench- or almost entirely, depending on the parameterization of the hazard
mark results is lower if the income elasticity of food demand (j) function – curtails further conversion of land into cropping. Together
is zero. This can be interpreted as a case where food consumption with diverting labor towards agricultural production and agricultural
reflects subsistence needs, so that per-capita demand for food cannot R&D, this policy is able to limit the welfare cost of the biodiversity
adjust downward by reducing manufacturing output. In turn, since externality to a low level. Of course, this result and in particular the
food has to be produced in proportion to population, the decline in role of land-use management does depend on our ‘business-as-usual’
agricultural output is lower, and more land is required as an input assumption about the damage inflicted by farming practices on bio-
to food production. Similarly, if the discount factor is lower (i.e. the diversity. In recent years attention has focused on how agricultural
utility discount rate is higher) the adjustment of agricultural land is practices might be made more sustainable (Bommarco et al., 2013;
smaller than in the baseline. This reflects the smaller weight given to Godfray and Garnett, 2014; Tilman et al., 2011). This would provide
future outcomes, including external costs from land conversion, and an alternative response option that we do not explore in this paper.
therefore implies smaller incentives to conserve land. The choice has been neatly summarized as being between ‘land spar-
ing’ and ‘land sharing’ (Phalan et al., 2011). Land sparing has been
8. Concluding Comments the traditional focus of policies and measures to protect natural land,
while land sharing is often the aim of agri-environment schemes,
One might summarize the premise of this paper as an argument
including payments for ecosystem services.
in three parts. The first part of the argument is that the expansion
In any case the feasibility of globally managed population growth
of modern, intensive agriculture causes biodiversity loss, unless it is
and cropland expansion can be questioned, to put it mildly. Therefore
practised in a sustainable way that is still rare in reality. This first
we also run the model assuming baseline, decentralized choices on
part of the argument seems incontrovertible. The second part is that
fertility and agricultural land conversion. Without these tools in the
biodiversity loss reduces agricultural productivity. This part of the
box, we find that, while our social planner still attempts to mitigate
argument does not perhaps have the same status, but it is backed
the externality by diverting labor towards agricultural production
Table 2 and R&D, the welfare cost of the externality is much higher. There-
Converted cropland relative to baseline without externality (in %). fore there could be much value in being able to manage land use
Parameter 2025 2050 2075 2100 at the global scale. We go on to evaluate the important assumption
Benchmark parameters −5.88 −10.26 −12.05 −12.67
about the substitutability of land in agriculture, and other sensitivi-
s = 0 .2 −3.56 −6.81 −8.42 −9.10 ties. If land is less substitutable, then the social planner is less able to
s =1 −6.19 −10.72 −12.41 −12.85 manage the externality by limiting cropland expansion and again a
dN = 0.015 −7.35 −13.39 −16.30 −17.77 larger welfare cost ensues.
j=0 −4.29 −6.92 −7.66 −7.68
Although our model is structurally estimated on 50 years of data,
b = 0.97 −2.91 −5.33 −6.25 −6.34
the lack of evidence that can directly inform the parameterization of
Notes: This table reports socially optimal agricultural land under the high hazard the biodiversity hazard function is a notable problem. For this reason
schedule relative to the baseline scenario without the externality, displayed as a per-
we specify two functions, both of which should be regarded as illus-
centage change. The first line reports our results for the benchmark set of parameters
corresponding to Fig. 3. In subsequent lines, we individually change each parameter, trative. In view of the weight of relevant evidence, obtaining better
re-estimating the model each time to obtain a consistent baseline. estimates of this function is a research priority.
B. Lanz et al. / Ecological Economics 144 (2018) 260–277 275
As detailed in Lanz et al. (2017a), the seven parameters {l mn , l ag , w, f, y, x, e} are estimated using simulation-based structural methods. The
moments we target are taken from observed trajectories over the period 1960 to 2010 of world GDP (Maddison, 1995; Bolt and van Zanden,
2013), world population (United Nations, 1999, 2013), crop land area (Goldewijk, 2001; Alexandratos and Bruinsma, 2012) and sectoral TFP
(Martin and Mitra, 2001; Fuglie, 2012).19 In the model these correspond respectively to Yt,mn + Yt,ag , Nt , Xt , At,mn and At,ag . We target one data
point for each 5-year interval, yielding 11 data points for the targeted quantity (55 points in total), and use these to formulate a minimum
distance estimator.
Specifically, the parameters minimise the value of the following expression:
∗
(Zk,t − Zk,t )2 / Zk,t , (A1)
k t t
6
0.02
0.01
2
0 0
1960 1970 1980 1990 2000 2010 1960 1970 1980 1990 2000 2010
Years Years
Converted agricultural land in billions hectares (X) World GDP in trillions 1990 intl. dollars (Ymn + Yag)
2
Observed Observed
Simulated Simulated
50
1.5
40
1 30
20
0.5
10
0 0
1960 1970 1980 1990 2000 2010 1960 1970 1980 1990 2000 2010
Years Years
Total factor productivity in agriculture (Aag) Total factor productivity in manufacturing (Amn)
3 12
Observed Observed
Simulated Simulated
10
2 8
1 4
0 0
1960 1970 1980 1990 2000 2010 1960 1970 1980 1990 2000 2010
Years Years
Fig. A1. Estimation of the model 1960–2010 (source: Lanz et al., 2017a).
19
Data on TFP is derived from TFP growth estimates and are thus subject to some uncertainty. Nevertheless, a robust finding of the literature is that the growth rate of TFP
economy-wide and in agriculture is on average around 1.5–2% per year. To remain conservative about the pace of future technological progress, we assume it declines from 1.5%
between 1960 and 1980 to 1.2% from 1980 to 2000, and then stays at 1% over the last decade.
276 B. Lanz et al. / Ecological Economics 144 (2018) 260–277
∗
where Zk,t denotes the observed quantity k at time t and Zk,t is the corresponding value simulated from the model. For each parameter to be
estimated from the data, we start by specifying bounds of a uniform distribution. For elasticity parameters, these bounds are 0.1 and 0.9 and
for the labor productivity parameters we use 0.03 and 0.3. We then solve the model for 10,000 randomly drawn vectors of parameters and
evaluate the error between the simulated trajectories and those observed. By gradually refining the bounds of the distribution, this procedure
converges to the vector of parameters that minimizes goodness-of-fit objective. This procedure converges and the vector of estimates reported
in Table 1.20
The resulting fit of the model is reported in Fig. A1, which compares trajectories that were observed over the period from 1960 to 2010 with
the trajectories simulated from the model. As the charts show, the estimated model provides a very good fit of recent history, and the relative
squared error (A1) across all variables is 3.52%. The size of the error is mainly driven by the error on output (3.3%), followed by land (0.1%) and
population (0.03%). Fig. A1 also reports the growth rate of population, which is not directly targeted by the estimation procedure, showing that
the simulated trajectory closely fits the observed dynamics of population growth.
References Di Falco, S., Bezabih, M., Yesuf, M., 2010. Seeds for livelihood: crop biodiversity and
food production in Ethiopia. Ecol. Econ. 69 (8), 1695–1702.
Aghion, P., Howitt, P., 1992. A model of growth through creative destruction. Econo- Dinopoulos, E., Thompson, P., 1998. Schumpeterian growth without scale effects. J.
metrica 60 (2), 323–351. Econ. Growth 3, 313–335.
Alexandratos, N., Bruinsma, J., 2012. World agriculture towards 2030/2050, The 2012 Echevarria, C., 1997. Changes in sectoral composition associated with economic
revision. ESA Working Paper No. 12-03. growth. Int. Econ. Rev. 38 (2), 431–452.
Alston, J.M., Pardey, P.G., 2014. Agriculture in the global economy. J. Econ. Perspect. 28 FAO, 1996. The state of the world’s plant genetic resources for food and agriculture.
(1), 121–146. Technical report. Food and Agriculture Organization of the United Nations, Rome.
Antràs, P., 2004. Is the U.S. aggregate production function Cobb-Douglas? New esti- FAO, 2010. The state of the world’s plant genetic resources for food and agriculture.
mates of the elasticity of substitution. Contrib. Macroecon. 4 (1), 1–34. Technical report. Food and Agriculture Organization of the United Nations, Rome.
Ashraf, Q.H., Lester, A., Weil, D., 2008. When does improving health raise GDP? In: Fogel, R.W., 1997. New findings on secular trends in nutrition and mortality: some
Acemoglu, D., Rogoff, K., Woodford, M. (Eds.), NBER Macroeconomics Annual. implications for population theory. In: Rosenzweig, M.R., Stark, O. (Eds.), Hand-
University of Chicago Press., pp. 157–204. book of Population and Family Economics. 1A. pp. 433–481. Amsterdam: North
Barro, R.J., Becker, G.S., 1989. Fertility choice in a model of economic growth. Econo- Holland, Chap. 9.
metrica 57, 481–501. Fuglie, K.O., 2012. Productivity growth and technology capital in the global agricul-
Barro, R., Sala-i Martin, X., 2004. Economic Growth - Second Edition. Cambridge MA: tural economy. In: Fuglie, K.O., Wang, S.L., Ball, V.E. (Eds.), Productivity Growth in
MIT Press. Agriculture: An International Perspective. CAB International, Wallingford, U.K.,
Baumgärtner, S., 2007. The insurance value of biodiversity in the provision of ecosys- pp. 335–368.
tem services. Nat. Resour. Model. 20 (1), 87–127. Galor, O., Weil, D.N., 2000. Population, technology, and growth: from Malthusian
Beatty, T.K.M., LaFrance, J.T., 2005. United States demand for food and nutrition in the stagnation to the demographic transition and beyond. Am. Econ. Rev. 90 (4),
twentieth century. Am. J. Agric. Econ. 87 (5), 1159–1166. 806–828.
Bolt, J., van Zanden, J.L., 2013. The first update of the Maddison project; re-estimating Godfray, H.C.J., Garnett, T., 2014. Food security and sustainable intensification. Philos.
growth before 1820. Maddison Project Working Paper 4. Trans. R. Soc. B Biol. Sci. 369 (1639), 20120273. April.
Bommarco, R., Kleijn, D., Potts, S.G., 2013. Ecological intensification: harnessing Goeschl, T., Swanson, T., 2003. Pests, plagues, and patents. J. Eur. Econ. Assoc. 1 (2),
ecosystem services for food security. Trends Ecol. Evol. 28 (4), 230–238. 561–575.
Bowman, M.S., Zilberman, D., 2013. Economic factors affecting diversified farming Goldewijk, K., 2001. Estimating global land use change over the past 300 years: the
systems. Ecol. Soc. 18 (1), 33. hyde database. Glob. Biogeochem. Cycles 15 (2), 417–433.
Bretschger, L., 2013. Population growth and natural resource scarcity: long-run Goldin, C., Katz, L., 1998. The origins of technology-skill complementarity. Q. J. Econ.
development under seemingly unfavourable conditions. Scand. J. Econ. 115 (3), 113, 693–732.
722–755. Gollin, D., 2002. Getting income shares right. J. Polit. Econ. 110 (2), 458–474.
Butchart, S.H., Walpole, M., Collen, B., van Strien, A., Scharlemann, J.P., Almond, R.E., Gourieroux, C., Monfort, A., 1996. Simulation-Based Econometric Methods. Oxford
Baillie, J.E., Bomhard, B., Brown, C., Bruno, J., et al. 2010. Global biodiversity: University Press.
indicators of recent declines. Science 328 (5982), 1164–1168. Guvenen, F., 2006. Reconciling conflicting evidence on the elasticity of intertemporal
Byrd, R., Hribar, M.E., Nocedal, J., 1999. An interior point method for large scale substitution: a macroeconomic perspective. J. Monet. Econ. 53, 1451–1472.
nonlinear programming. SIAM J. Optim. 9 (4), 877–900. Ha, J., Howitt, P., 2007. Accounting for trends in productivity and R&D: a Schumpete-
Byrd, R.H., Nocedal, J., Waltz, R.A., 2006. KNITRO: an integrated package for nonlinear rian critique of semi-endogenous growth theory. J. Money, Credit, Bank. 39 (4),
optimization. In: di Pillo, G., Roma, M. (Eds.), Large-Scale Nonlinear Optimization. 733–774.
Springer-Verlag., pp. 35–59. Hansen, G.D., Prescott, E.C., 2002. Malthus to Solow. Am. Econ. Rev. 92 (4), 1204–1217.
Cardinale, B.J., Duffy, J.E., Gonzalez, A., Hooper, D.U., Perrings, C., Venail, P., Narwani, A., Heal, G., Walker, B., Levin, S., Arrow, K., Dasgupta, P., Daily, G., Ehrlich, P., Maler, K.-G.,
Mace, G.M., Tilman, D., Wardle, D.A., Kinzig, A.P., Daily, G.C., Loreau, M., Grace, J.B., Kautsky, N., Lubchenco, J., et al. 2004. Genetic diversity and interdependent crop
Larigauderie, A., Srivastava, D.S., Naeem, S., 2012. Biodiversity loss and its impact choices in agriculture. Resour. Energy Econ. 26 (2), 175–184.
on humanity. Nature 486 (7401), 59–67. June. Hertel, T., Tsigas, M., Narayanan, B., 2012. Chapter 12.a: Primary factor shares. In:
Chu, A., Cozzi, G., Liao, C.-H., 2013. Endogenous fertility and human capital in a Narayanan, B., Aguiar, A., McDougall, R. (Eds.), Global Trade, Assistance, and
Schumpeterian growth model. J. Popul. Econ. 26, 181–202. Production: The GTAP 8 Data Base. Center for Global Trade Analysis, Purdue
Clarke, L., Jiang, K., Akimoto, K., et al. 2014. Assessing transformation pathways. In: University.,
Edenhofer, O., Pichs-Madruga, R., Sokona, Y., Climate Change 2014: Mitigation of Hooper, D.U., Adair, E.C., Cardinale, B.J., Byrnes, J.E., Hungate, B.A., Matulich, K.L., Gon-
Climate Change. Contribution of Working Group III to the Fifth Assessment Report zalez, A., Duffy, J.E., Gamfeldt, L., O’Connor, M.I., 2012. A global synthesis reveals
of the Intergovernmental Panel on Climate Change. Cambridge University Press, biodiversity loss as a major driver of ecosystem change. Nature 486 (7401),
Cambridge, UK and New York, NY, USA. 105–108.
Dasgupta, P., Heal, G., 1979. Economic Theory and Exhaustible Resources. Cambridge Jones, C.I., 1995. R&D-based model of economic growth. J. Polit. Econ. 103 (4),
University Press, Cambridge, UK. 759–784.
Di Falco, S., 2012. On the value of agricultural biodiversity. Annu. Rev. Resour. Econ. 4 Jones, C.I., 2001. Was an industrial revolution inevitable? Economic growth over the
(1), 207–223. very long run. B.E. J. Macroecon. 1 (2), 1–45.
Di Falco, S., Chavas, J., 2009. On crop biodiversity, risk exposure, and food security in Jones, C., Williams, J., 2000. Too much of a good thing? The economics of investment
the highlands of Ethiopia. Am. J. Agric. Econ. 91 (3), 599–611. in R&D. J. Econ. Growth 5, 65–85.
Kawagoe, T., Otsuka, K., Hayami, Y., 1986. Induced bias of technical change in
agriculture: The United States and Japan, 1880-1980. J. Polit. Econ. 94 (3),
523–544.
20 Khoury, C.K., Bjorkman, A.D., Dempewolf, H., Ramirez-Villegas, J., Guarino, L., Jarvis, A.,
As for other simulation-based estimation procedures involving highly non-linear
Rieseberg, L.H., Struik, P.C., 2014. Increasing homogeneity in global food supplies
models, the uniqueness of the solution cannot be formally proved (see Gourieroux and
and the implications for food security. Proc. Natl. Acad. Sci. 201313490.March.
Monfort, 1996). Our experience with the model suggests however that the solution
Klein Goldewijk, K., Beusen, A., van Drecht, G., de Vos, M., 2011. The HYDE 3.1 spatially
is unique, with no significantly different vector of parameters providing a compara- explicit database of human-induced global land-use change over the past 12,000
ble goodness-of-fit objective. In other words, estimates reported in Table 1 provide years. Glob. Ecol. Biogeogr. 20 (1), 73–86.
a global solution to the estimation objective. This is due to the fact that we target a von Lampe, M., Willenbockel, D., Ahammad, H., Blanc, E., Cai, Y., Calvin, K., Fujimori,
large number of data points for several variables, and that changing one parameter S., Hasegawa, T., Havlik, P., Heyhoe, E., et al. 2014. Why do global long-term sce-
will impact trajectories across all variables in the model, which makes the selection narios for agriculture differ? An overview of the agmip global economic model
criteria for parameters very demanding. intercomparison. Agric. Econ. 45 (1), 3–20.
B. Lanz et al. / Ecological Economics 144 (2018) 260–277 277
Lanz, B., Dietz, S., Swanson, T., 2017a. Global population growth, technology, Savary, S., Ficke, A., Aubertot, J.-N., Hollier, C., 2012. Crop losses due to diseases and
and land conversion: a quantitative growth theoretic perspective. Int. Econ. their implications for global food production losses and food security. Food Sec. 4
Rev. forthcoming. (4), 519–537. December.
Lanz, B., Dietz, S., Swanson, T., 2017b. Global economic growth and agricultural land Schmitz, C., van Meijl, H., Kyle, P., Nelson, G.C., Fujimori, S., Gurgel, A., Havlik, P.,
conversion under uncertain productivity improvements in agriculture. MIT Joint Heyhoe, E., d’Croz, D.M., Popp, A., et al. 2014. Land-use change trajectories up to
Program Report 313. 2050: insights from a global agro-economic model comparison. Agric. Econ. 45
Logan, T., 2009. The transformation of hunger: the demand for calories past and (1), 69–84.
present. J. Econ. Hist. 69 (2), 388–408. Sharp, P., Strulik, H., Weisdorf, J., 2012. The determinants of income in a Malthusian
Maddison, A., 1995. Monitoring the world economy 1820–1992. Organization for equilibrium. J. Dev. Econ. 97, 112–117.
Economic Cooperation and Development, Paris. Simpson, R., Sedjo, R., Reid, J., 1996. Valuing biodiversity for use in pharmaceutical
Martin, W., Mitra, D., 2001. Productivity growth and convergence in agriculture versus research. J. Polit. Econ. 104 (1), 163–185.
manufacturing. Econ. Dev. Cult. Chang. 49 (2), 403–422. Smale, M., Hartell, J., Heisey, P.W., Senauer, B., 1998. The contribution of genetic
Assessment, Millennium Ecosystem, 2005. Ecosystems and Human Well-Being: Syn- resources and diversity to wheat production in the Punjab of Pakistan. Am. J.
thesis. Island Press, Washington, DC. Agric. Econ. 80 (3), 482–493.
Mundlak, Y., 2000. Agriculture and economic growth: theory and measurement. Strulik, H., Weisdorf, J., 2008. Population, food, and knowledge: a simple unified
Harvard University Press, Cambridge M.A.. growth theory. J. Econ. Growth 13, 195–216.
Newbold, T., Hudson, L.N., Hill, S.L., Contu, S., Lysenko, I., Senior, R.A., Börger, L., Subramanian, S., Deaton, A., 1996. The demand for food and calories. J. Polit. Econ. 104
Bennett, D.J., Choimes, A., Collen, B., et al. 2015. Global effects of land use on local (1), 132–162.
terrestrial biodiversity. Nature 520 (7545), 45–50. Tilman, D., 1999. The ecological consequences of changes in biodiversity: a search for
Ngai, L.R., Pissarides, C.A., 2007. Structural change in a multisector model of growth. general principles. Ecology 80 (5), 1455–1474.
Am. Econ. Rev. 97 (1), 429–443. Tilman, D., Balzer, C., Hill, J., Befort, B.L., 2011. Global food demand and the sustainable
Nordhaus, W.D., 1992. An optimal transition path for controlling greenhouse gases. intensification of agriculture. Proc. Natl. Acad. Sci. 108 (50), 20260–20264.
Science 258, 1315–1315. Tilman, D., Polasky, S., Lehman, C., 2005. Diversity, productivity and temporal stability
Nordhaus, W.D., Boyer, J., 2000. Warming the World: Economic Models of Global in the economies of humans and nature. J. Environ. Econ. Manag. 49, 405–426.
Warming. MIT Press (MA). Nations, United, 1999. The world at six billion. Department of Economic and Social
Oerke, E.-C., 2006. Crop losses to pests. J. Agric. Sci. 144, 31–43. Affairs, Population Division.
Omer, A., Pascual, U., Russell, N.P., 2007. Biodiversity conservation and productivity in Nations, United, 2013. World Population Prospects: The 2012 Revision. Department of
intensive agricultural systems. J. Agric. Econ. 58 (2), 308–329. Economic and Social Affairs, Population Division, New York.
Peretto, P.F., 1998. Technological change and population growth. J. Econ. Growth 3, Nations, United, 2015. World Population Prospects: The 2015 Revision, DVD edition.
283–311. United Nations, Department of Economic and Social Affairs, Population Division.
Phalan, B., Onial, M., Balmford, A., Green, R.E., 2011. Reconciling food production and Vollrath, D., 2011. The agricultural basis of comparative development. J. Econ. Growth
biodiversity conservation: land sharing and land sparing compared. Science 333 16, 343–370.
(6047), 1289–1291. Weitzman, M., 2000. Economic profitability versus ecological entropy. Q. J. Econ. 115
Poleman, T.T., Thomas, L.T., 1995. Income and dietary change: international com- (1), 237–263.
parisons using purchasing-power-parity conversions. Food Policy 20 (2), Wilde, J., 2013. How substitutable are fixed factors in production? Evidence from
149–159. pre-industrial England. Working paper 0113. University of South Florida, Depart-
Rausser, G., Small, A., 2000. Value research leads: bioprospecting and the conservation ment of Economics.,
of genetic resources. J. Polit. Econ. 108 (1), 173–206. World Bank, 2016. World Development Indicators. World Bank, Washington, DC.
Robinson, S., Meijl, H., Willenbockel, D., Valin, H., Fujimori, S., Masui, T., Sands, R., Young, A., 1998. Growth without scale effects. J. Polit. Econ. 106, 41–63.
Wise, M., Calvin, K., Havlik, P., et al. 2014. Comparing supply-side specifications
in models of global agriculture and the food system. Agric. Econ. 45 (1), 21–35.