Explaining forest transitions: The role of governance

Ecological Economics 119 (2015) 252–261

Contents lists available at ScienceDirect

Ecological Economics journal homepage: www.elsevier.com/locate/ecolecon

Analysis

Explaining forest transitions: The role of governance☆ Edward B. Barbier a,⁎, Anteneh Tesfaw b a b

Department of Economics and Finance, University of Wyoming, Laramie, WY 82071, United States Moore Center for Science & Oceans, Conservation International, 2011 Crystal Drive, Arlington, VA 22202, United States

a r t i c l e

i n f o

Article history: Received 3 April 2015 Received in revised form 18 September 2015 Accepted 20 September 2015 Available online 29 September 2015 JEL classification: Q23 Q24 O13

a b s t r a c t We analyze how governance may influence competing land uses for forests, and thus the occurrence of forest transitions, across different low and middle-income countries. We develop a model of competing land uses that allows for governance to impact the risk of future versus current agricultural and forested land allocations. The resulting hypothesis on the relationship between governance and the likelihood of a forest transition is then tested using cross-country data. The empirical analysis offers strong support for the competing land use framework, and indicates that rule of law, forest policy and regulatory quality influence forest transitions. These findings inform not only the ongoing debate on forest transitions but also policy options for managing such transitions in developing economies. © 2015 Elsevier B.V. All rights reserved.

Keywords: Developing countries Forest transition Land use Probit estimation Cross-country regression

1. Introduction Although globally forest conversion still remains pervasive, forest recovery has occurred for decades in developed regions, notably Western Europe and North America, and most recently in some developing countries, such as Bangladesh, China, Costa Rica, Dominican Republic, India, Morocco and Vietnam (Bray, 2010; Hansen et al., 2013; Hosonuma et al., 2012; Mather, 2007; Meyfroidt and Lambin, 2011; Rudel et al., 2005). Mather (1990 and 1992) coined the term forest transition to describe this observed turnaround from deforestation to recovery. As shown in Fig. 1, the long-run trend suggests a “U-shaped curve” for forest land with respect to time: a prolonged decline in country's forest cover in the early stages of economic development followed by a partial recovery through conserving remaining primary forest, plantations and reforestation. Thus, an important advantage of the forest transition is that it offers a long-term perspective on land-use management; a country that is deforesting today may not necessarily continue to convert forest land in the future but eventually transition to a stage of forest recovery.

This pattern of forest cover change from deforestation to recovery occurs at different scales across countries or even across regions within a country, which can be explained by changes in the overall allocation of land (Hansen et al., 2013; Mather, 1992; Meyfroidt and Lambin, 2011; Meyfroidt et al. 2010; Pfaff and Walker, 2010; Rudel et al., 2005). Land use in turn depends on the rate of return, or rent, obtained from forest land compared to its competing uses (Angelsen and Rudel, 2013; Barbier et al., 2010). Consequently, the analysis of the forest transition lends itself to the standard competing land use model in economics, which has been used extensively to analyze the conversion of forest land to agriculture and other activities as well as forest recovery (Amacher et al., 2008 and 2009; Angelsen (2007); Barbier and Burgess, 1997; Barbier et al., 2010; Hartwick et al., 2001; Ollivier, 2012; Delacote and Garcia, 2015). However, governance may also be an important factor affecting forest transition outcomes.1 There are several ways in which this may occur (Barbier et al., 2010). First, although some evidence suggests that forest recovery is more likely in countries with democratic political institutions, for many low and middle-income countries political

☆ Paper for the Thematic Session “The Economic Analysis of the Forest Transition” at the World Congress of Environmental and Resource Economists (WCERE), Istanbul, Turkey, June 28–July 2, 2014. This research was funded by a research grant on Sustainable Business Practices and a faculty research grant, provided by the College of Business, University of Wyoming. ⁎ Corresponding author. E-mail address: [email protected] (E.B. Barbier).

1 Throughout this paper, we use the term governance as a short-hand for economywide governance, which Kaufmann et al. (2009, p. 5) define as “the traditions and institutions by which authority in a country is exercised. This includes the process by which governments are selected, monitored and replaced; the capacity of the government to effectively formulate and implement sound policies; and the respect of citizens and the state for the institutions that govern economic and social interactions among them.”

http://dx.doi.org/10.1016/j.ecolecon.2015.09.010 0921-8009/© 2015 Elsevier B.V. All rights reserved.

E.B. Barbier, A. Tesfaw / Ecological Economics 119 (2015) 252–261

Fig. 1. Land use change and the forest transition. Although timber harvesting and fuelwood consumption may play a role, the initial loss of natural forest cover is mainly the result of rapid loss agricultural area expansion in response to rising demand for food and other commodities as economic development proceeds and populations grow. As agricultural land expansion slows down, so does the decrease in primary forest area. Increased environmental protection of remaining primary forest also stabilizes its size. However, as an economy develops further, the increased demand for wood products and non-market ecosystem services from forested land may lead to recovery in the total forest area, with protection of remaining primary forest, reforestation and plantations all playing a role. The time period when the long-run decline in forest area is superseded by forest recovery is defined as the forest transition (Mather, 1990 and 1992). Source: Adapted from Barbier et al. (2010).

stability may be a more important influence (Grainger, 2004; Grainger and Malayang, 2006; Mather and Needle, 1999). Second, regulatory quality may also matter; for example, countries with very different political regimes, such as China, India and Vietnam, have all developed effective regulatory institutions that have provided incentives to restore degraded forests, promote replanting by landowners and induce afforestation (Mather, 2007). Finally, the rule of law and the protection of property rights could also matter, especially given deforestation in developing countries is caused less by state-funded enterprises and large-scale settlement investments and more through the decentralized decision-making by farmers, land speculators, Agri-business enterprises and ranchers (Chomitz et al., 2007; Gibbs et al., 2010; Lambin and Meyfroidt, 2011; Rudel, 2007).2 There is also extensive literature on the impact of institutions on global deforestation, which parallels the concerns about how governance may influence the forest transition of developing countries. Some of this literature focuses on forest governance, notably forest property rights and tenure security (Agrawal, 2007; Agrawal et al., 2008; Alix-Garcia et al., 2004; Alston et al., 1996; Chhatre and Agrawal, 2008; Liscow, 2013).3 Other studies link deforestation with economywide governance, such as political stability, ownership security, corruption and rule of law (Barbier et al., 2005; Barbier and Burgess, 2001; Bhattarai and Hammig, 2004; Bohn and Deacon, 2000; Damatte and Delacote, 2012; Deacon, 1994; Ferreira and Vincent, 2010; Galinato and Galinato, 2012 and Galinato and Galinato, 2013; López and Galinato, 2005; Nguyen-Van and Azomahou, 2007). Most studies find positive correlation between lower governance indices and higher deforestation rates in developing countries, although Damatte and Delacote (2012) suggest that the institutional influences may have been over-estimated in previous studies relative to the impacts of timber harvesting, economic growth, and possibly macroeconomic shocks. By exploring the impact of governance on forest transitions in developing countries, our paper makes two principal contributions. 2

Angelsen and Rudel (2013, p. 105) describe the resulting implications for competing forest land uses: As land-use decisions become increasingly enterprise driven, “property right determine the extent to which forest users capture the different forest rents”; in contrast, “in an open access situation, where forest clearing and agricultural uses provide some land rights, there are limited incentive for farmers to factor forest rents into their decisions.” 3 Forest governance refers to how decisions are made about the management, use, and transfer of forest lands and resources (Agrawal, 2007)

253

First, we develop a competing land use model as our framework for analyzing a forest transition. As Fig. 1 indicates, this transition connects two distinct phases, one in which forest area continuously declines followed by forest recovery. To analyze land use decisions across both phases, we adopt a two-stage optimal control model, which is a standard approach to modeling a dynamic problem with multiple phases (Amit, 1986; Makris, 2001; Tomiyama, 1985). The key proposition to emerge from this model is that worse (better) governance delays (hastens) the time when a forest transition occurs. The result is that the wedge between the returns from agricultural and forested land widens, and the forest transition is postponed. Inadequate governance therefore implies that the likelihood of attaining a forest transition is lower. In addition, this proposition also suggests a testable hypothesis. Better (worse) governance ceteris paribus increases (decreases) the likelihood of a forest transition occurring. Thus, the second contribution of this paper is to test this hypothesis empirically for 132 developing countries, of which 27 have shown evidence of a forest transition – a shift from net deforestation to net reforestation – by 2010 (Bray, 2010; Food and Agriculture Organization of the United Nations (FAO), 2010; Hosonuma et al., 2012; Meyfroidt and Lambin, 2011; Rudel et al., 2005). Although reliable forest governance indicators across developing countries is generally lacking (Agrawal, 2007; Agrawal et al., 2008; Chhatre and Agrawal, 2008), we employ a wide variety of economy-wide measures in our analysis, including forest policy and ownership variables, governance indicators, economic policy indicators, country risk classifications, and country lending premiums. For nearly all these indicators, the hypothesis that governance influences the probability of a forest transition in developing countries is rejected. The exceptions are the presence of forest policy, rule of law and regulatory quality, which are statistically significant. Whereas forest policy and the rule of law increase the likelihood of a forest transition, regulatory quality reduces this probability. The latter result is surprising, but could be evidence that an improved regulatory climate in the overall economy may actually facilitate enterprise-driven deforestation more than forest recovery (Angelsen and Rudel, 2013; Rudel, 2007; Liscow, 2013). Because the estimation approach is based on a competing land use model of forest change in developing countries, it differs from other analyses of the possible causes of the forest transition across countries (Hosonuma et al., 2012; Köthke et al., 2013; Rudel et al., 2005). To our knowledge, this is the first paper that develops an economic model of the forest transition, including explicitly the influence of governance, and then uses the analysis to inform an empirical cross-country examination of the resulting hypothesis. We hope that this approach inspires further economic analyses of the forest transition and its most likely causes, including the role of economy-wide and forest governance as well as market and policy failures more generally. The paper is organized as follows. The next section develops a twostage optimal control problem of competing land uses for analyzing the forest transition and the potential influence of governance. The following section tests the hypothesis derived from our model that better (worse) governance increases (decreases) the likelihood of a forest transition through empirical analysis across developing countries. We conclude by summarizing our key findings, and identifying the policy implications and issues for further research. 2. A forest transition model. Assume an initial stock of forest land F(0) = F0 subject to agricultural land conversion. Let L(0) = L0 denote the initial stock of agricultural land. Given the initial stock of land L0 + F0, at each time t the social planner determines the most valuable land use allocation between agriculture L(t)and forest land F(t). We assume that forest land is initially abundant, possibly due to low initial population pressure relative to the amount of agricultural land available. However, as economic development continues and population grows, increased demand for

254

E.B. Barbier, A. Tesfaw / Ecological Economics 119 (2015) 252–261

agricultural goods results in more forest land conversion. If c(t) is the amount of primary forest converted to agricultural land at time t, the changes in land use are respectively. Z t ∂F F ðt Þ ¼ F 0 − cðsÞds; ¼ F ¼ −cðt Þ ∂t 0 Z

t

Lðt Þ ¼ L0 þ

cðsÞds;

0

∂L ¼ L ¼ cðt Þ : ∂t

ð1aÞ

ð1bÞ

Agricultural output Q is assumed to be a function of total agricultural land, i.e.,Q(t) = q(L(t)) = q(L0 + F0 − F(t)) and q′(L(t)) N 0 , q″(L(t)) b 0. We normalize agricultural prices to one, so that Q also represents the value of agricultural production. The remaining standing forest produces a flow timber production and environmental benefits.4 Let B(F(t)) denote the periodic aggregate flow of these benefits, which is increasing with the forest land, albeit at a decreasing rate, i.e. B′(F(t)) N 0 , B″(F(t)) b 0. Finally, conversion is costly, and these costs are convex, C1(c(t)), C1′ N 0 , C″1 N 0 and C1(c(0))=C1′(c(0)) = 0. However, at some future time0 ≤ t1 ≤ ∞, the planner may find it optimal to undertake forest restoration. That is, at time t1 the planner may switch from a regime of forest conversion (henceforth Regime I) active over0 ≤ t ≤ t1 to a regime of forest restoration (henceforth Regime II) active over t1 ≤ t ≤ ∞. In essence the switching time t 1 is a choice variable. Let F(t1) = F1denote the remaining primary forest at the switching time between regimes, and L(t 1 ) = L 1 the corresponding stock of agricultural land. If g(t) represents restoration at time t ≥ t1, then the state equations corresponding to Regime II are. Z F ðt Þ ¼ F 1 þ

t t1

Z Lðt Þ ¼ L1 −

t t1

g ðsÞds; F ¼ g ðt Þ; t 1 ≤t ≤∞

g ðsÞds; L ¼ g ðt Þ; t 1 ≤t ≤∞:

ð2aÞ

ð2bÞ

To emphasize the possibility that benefits from restored land via reforestation and plantations can be different from that provided by natural forests, for Regime II we identify two distinct benefit functions corresponding to standing forest. The first is the benefit of remaining natural forest B(F1 ). The second is the periodic benefit of restored land R(F(t) − F 1 ) = R(L1 − L(t)), which is assumed to be concave R′ N 0 , R″ b 0. However, restoration is costly and is denoted by C2(g(t)) , C2′ N 0 , C″2 N 0 and C2(g(0)) = C2′(g(0)) = 0. As discussed in the introduction, forest and economy-wide governance is likely to influence the forest planner's land allocation decision concerning both forest conversion and restoration. The most likely effect is to alter the planner's perception of the risk of future versus current gains from any allocation. For example, a planner's perceived risk is greater when legal contracts are absent or weakly enforced (Deacon, 1994), ownership risk is high (Agrawal, 2007; Araujo et al., 2009; Bohn and Deacon, 2000; Chhatre and Agrawal, 2008), political instability is pervasive (Ferreira and Vincent, 2010; Galinato and Galinato, 2012 and Galinato and Galinato, 2013), and comprehensive forest policy and regulations are absent (Agrawal 4

The World Bank (2011, pp. 146) states that “the predominant economic use of forests has been as a source of timber”; thus, timber production must be considered an important value of a country's forest, even if timber production itself may lead to substantial damages or loss of this stock if production if unsustainably managed. As noted by Barbier et al. (2010, p. 101), the environmental benefits of forests “are wide ranging, and may include various benefits, such as harvested wood and other products collected by local populations, which may be consumed by them or sold for cash income; ecosystem services such as watershed protection, habitat for forest-dwelling species, and so forth; and tourism, recreational and amenity benefits associated with relatively undisturbed, primary forests.” In addition, given the recent emphasis of reducing the carbon emissions from deforestation and forest degradation, carbon sequestration is another important environmental benefit of forests that may influence land use allocation decisions leading to a forest transition (Angelsen and Rudel, 2013; Köthke et al., 2013).

et al., 2008; Food and Agriculture Organization of the United Nations (FAO), 2010). Given one or more of these governance difficulties, a planner is likely to place a higher risk premium on the future returns to any forest land use allocation compared to current benefits. To capture this effect, we introduce a governance risk parameter 0≤θ ≤θ. From a mathematical perspective, θ is similar to a discount factor; lower values of θ favor preserving the forest while higher values favor current exploitation.5 For example, when political instability or tenure insecurity are pervasive (θ approaches or equals θ), the planner's perceived risk is greater and thus favor forest conversion. Conversely, lower values of θ indicate better governance, with θ = 0 denoting fully effective governance. The problem faced by the planner can be formulated as a two-stage optimal control problem (Amit, 1986; Makris, 2001; Tomiyama, 1985).6 For a given θ, the planner seeks to determine the optimal forest conversion path c(t) ≥ 0 for t ∈ [0, t1), the optimal switching instant t1, and the optimal forest restoration path g(t) ≥ 0 for t ∈ (t1, ∞], such that the planner's objective function J is maximized over the entire planning horizon. That is, given the social discount rate r and suppressing time arguments to simplify notation, the planner's problem is to Z Max J ¼

t1

e−ðrþθÞt ðBð F Þ þ qðLÞ−C 1 ðcÞÞdt

0

Z þ

∞

ð3Þ e−ðrþθÞt ðBð F 1 Þ þ Rð F− F 1 Þ þ qðLÞ−C 2 ðg ÞÞdt

t1

subject to  F¼

−c; 0 ≤t ≤t 1 g; t 1 ≤t ≤∞

 ;

ð4Þ

by choice of c(t) ≥ 0 and g(t) ≥ 0, where F(0) = F0is fixed, and t1 and F(t1) = F1 are free. Let μi(t) , i = 1 , 2 denote the current shadow value associated with the forest stock F(t) in Regime j = I , II, respectively. The current-value H 0≤t ≤t 1 Hamiltonian is H ¼ f 1 g where H2 t 1 ≤t ≤∞ H 1 ¼ Bð F Þ þ qðLÞ−C 1 ðcÞ−μ 1 c

ð5Þ

H 2 ¼ Bð F 1 Þ þ Rð F− F 1 Þ þ qðLÞ−C 2 ðg Þ þ μ 2 g :

ð6Þ

The first-order conditions include μ 1 ¼ −C 0 ðcÞ1 ≤0; 0≤t ≤t 1

ð7Þ

μ 1 ¼ ðr þ θÞμ 1 −B0 ð F Þ þ q0 ðLÞ; 0≤t ≤t 1

ð8Þ

μ 2 ¼ C 02 ðg Þ≥0; t 1 ≤t ≤∞

ð9Þ

μ 2 ¼ ðr þ θÞμ 2 −R0 ð F−F 1 Þ þ q0 ðLÞ; t 1 ≤t ≤∞:

ð10Þ

Combining conditions (Eqs. (7) and (8)), as well as conditions (Eqs. (9) and (10)), yields q0 ðLÞ μ 1 þ B0 ð F Þ − ¼ −μ 1 ¼ C 01 ðcÞ≥0; 0≤t ≤t 1 rþθ rþθ

ð11Þ

5 Our modeling best fits governance variables such as, political stability, rule of law, forest tenure type, and forest tenure security, that alter a planner's perception of the risk of future versus current gains from any land allocation; but in no way should this be interpreted as implying that these are the only governance variables influencing a planner's land allocation nor do we consider our approach appropriate to model variables such as regulatory quality and forest policy. 6 See Delacote and Garcia (2015) for an alternative modeling approach, where the authors assume a concurrent occurrence of deforestation and reforestation. We chose not to follow a similar approach as it wouldn't have allowed us to solve for the time of the forest transition, t1, and perform comparative statics of interest.

E.B. Barbier, A. Tesfaw / Ecological Economics 119 (2015) 252–261 :

μ 2 þ R0 ð F þ F 1 Þ q0 ðLÞ − ¼ μ 2 ¼ C 02 ðg Þ≥0; t 1 ≤t ≤∞: rþθ rþθ

ð12Þ

Condition (11) states that, along the optimal conversion path during Regime I, the difference between the capitalized marginal value of agricultural land as opposed to primary forest just equals the marginal cost of conversion. Note that, for this phase preceding the forest transition, primary forest is serving mainly as a “reservoir” for agricultural land, and thus the shadow value of the remaining forest is negative, i.e.μ1 b 0. Forests still produce valuable services (B′(F) N 0), but because they are relatively abundant compared to agricultural land, during Regime I their capitalized value, or “price”, is less than that of agricultural land. Consequently, the negative shadow value μ1 reflects that forest land is more valuable if it is converted to land for agricultural production than retained as primary forest. This explains the continuous conversion of primary forest to agriculture during the pre-forest transition phase (see Fig. 1). Finally, better or worse governance clearly influences this land allocation decision through the effective discount rater + θ. For example, worse governance raises the discount rate, which means that the opportunity cost of holding onto primary forest is higher. This leads to the following proposition. Proposition 1. . During the phase preceding the forest transition, worse (better) governance leads to a higher (lower) forest land to agriculture ∂c N0. conversion rate, i.e. ∂θ The proof is straightforward. Taking the time derivative of Eq. (7) and substituting it into Eq. (8) yields an expression for the change over time in the optimal conversion of forest land to agriculture : −ðr þ θÞC 01 ðcÞ−B0 ð F Þ þ q0 ðLÞ c¼ −C ″1 ðcÞ

ð12’Þ

:

C 0 ðcÞ

∂c Differentiating (12′) with respect to θ yields ∂θ ¼ C 1″ ðcÞ N 0, suggesting 1

that higher θ leads to a greater conversion rate (i.e., an increase in the change over time in optimal land conversion). Intuitively, bad governance implies that the perceived riskiness of holding onto this land unit as primary forest and thus the effective discount rate has increased. The result is a higher conversion rate of primary forest land to agriculture. It can be easily seen that a lower θ (better governance) leads to the opposite outcome of reducing the riskiness of holding onto land as primary forest and thus a lower conversion rate. Condition (12) states that, along the optimal restoration path during Regime II, the difference between the capitalized marginal value of forested as opposed to agricultural land just equals the marginal cost of restoration. For this post-forest transition phase, agricultural land is less valuable than forest, and so the latter's shadow value μ2 is positive. As a result, after the forest transition, reforestation, afforestation and other restoration initiatives ensure that forest recovery ensues (see Fig. 1). Eq. (12) indicates that better or worse governance also affects this land allocation decision. As weaker governance raises the effective discount rate r + θ, the opportunity cost of holding onto agricultural land – which is now the “reservoir” for forested land – has increased. The result is the following proposition. Proposition 2. . During the phase following the forest transition, worse (better) governance leads to a higher (lower) agricultural land to forest N0. restoration rate, i.e. ∂g ∂θ The proof is similar as for Proposition 1. From taking the time derivative of Eq. (9) and substituting it into Eq. (10), we have. : ðr þ θÞC 02 ðg Þ−R0 ð F−F 1 Þ þ q0 ðLÞ g¼ C ″2 ðg Þ

255

ð12”Þ

C 0 ðgÞ

∂g Differentiating Eq. (12″) with respect to θ yields ∂θ ¼ C ″2 ðgÞ N0, sug2

gesting that as θ rises so does the rate of restoration (i.e., there is an increase in the change over time in optimal forest restoration). The implication is that the perceived riskiness of holding onto this land for agricultural production has increased, and the result is higher current rate of forest restoration g. If θ declines, then the effective discount rate is lower and so is the rate of forest restoration. For an infinite horizon two-stage optimal control problem as Eq. (3), the solution also involves the following transversality conditions (Makris, 2001): lim H 1 ðsÞ ¼ 0; if t 1 →∞

s→∞

lim H 2 ðsÞ ¼ 0; if 0bt 1 b∞

ð13Þ

s→∞

The additional necessary conditions to define the optimal switching timet1⁎are (Tomiyama, 1985; Amit, 1986):             H 1 t 1 ¼ H 2 t 1 →−C    1 c t1  −μ  1 t 1 c t 1 ¼ −C 2 g t 1 þ μ 2 t 1 g t 1 ; 0bt 1 b∞

ð14Þ

H1 ð0Þ ≤H2 ð0Þ→−C 1 ðcð0ÞÞ−μ 1 ð0Þcð0Þ ≤−C 2 ðg ð0ÞÞ þ μ 2 ð0Þgð0Þ; 0 ¼ t 1 b∞

ð15Þ               H 1 t 1 ≥H 2 t 1 →−C 1 c t 1 −μ 1 t 1 c t 1 ≥−C 2 g t 1

ð16Þ

    þμ 2 t 1 g t 1 ; 0bt 1 ¼ ∞ Furthermore, if 0 b t1⁎ b ∞, then.           −μ 1 t 1 ¼ μ 2 t 1 →C 01 c t 1 ¼ C 02 g t 1

ð17Þ

Conditions (14)–(17) define the optimal switching time for the forest transition. Condition (14) says that if there is a time t1⁎ at which the returns H1 from Regime I just equal the returns H2 corresponding to Regime II, then it is optimal to start with forest conversion and switch to reforestation att1⁎ That is, t1⁎equalizes the marginal benefits and costs of postponing the switching time. However, if condition (15) or condition (16) holds, the problem is characterized by corner solutions - no interiort1⁎exists - and thus the problem effectively reduces to a standard optimal control problem.7 Condition (17) states that at the optimal finite switching timet1⁎ the shadow value of natural forest just equals that of restored forest. Note that, for conditions (11), (12), (14) and (17) to hold simultaneously att1⁎ requires thatC1′(c(t1⁎))= C2′(g(t1⁎))= c(t1⁎) = g(t1⁎) = 0.8 Consequently, the end of the pre-forest transition phase when primary forest is converted to agricultural c(t1⁎) = 0 corresponds with the postforest transition phase of forest recoveryg(t1⁎) = 0, which is the outcome depicted in Fig. 1.9 It also follows from Eqs. (11) and (12) that, at the 7 Specifically, condition (15) states that if the marginal benefit of delaying immediate regime switching to forest restoration is less than the corresponding marginal cost, it is optimal to immediately switch to restoration. Conversely, condition (16) states that if the marginal benefit of delaying forest restoration is always greater than or equal to its cost, it is optimal never to switch to forest restoration. 8 ⁎1))and marginal cost of restoraNote that when the marginal cost of conversion C′(c(t 1 ⁎1))are at their respective minimum, clearly average cost equals marginal cost. As tion C′(g(t 2 such, there could be a unique c(t⁎1) N 0 and g(t⁎1) N 0such that Condition (14) is satisfied. However, the fact that the two cost functions are fundamentally different means that this unique conversion and restoration solution does not necessarily satisfy condition (17). We thank a reviewer for pointing out this possibility. 9 As our model is in continuous time, by necessity the forest transition is depicted as an instant of timet⁎1. However, as noted by Mather (1992) and further elaborated by Barbier et al. (2010); Meyfroidt and Lambin (2011) and Rudel et al. (2005), in many countries the forest transition “phase” defining the transformation from net deforestation to a sustained period of forest recovery may last for considerable time. The main results of our model are consistent with and do not change if it is assumed that t⁎1 persists for a long interval of time.

256

E.B. Barbier, A. Tesfaw / Ecological Economics 119 (2015) 252–261

optimal switching timet1⁎, the capitalized values of forested and agricultural land are equal. Recall that condition (14) also states that it is always optimal to start with forest conversion and switch to reforestation att1⁎. However, Proposition 1 indicates that weaker governance during the pre-forest transition phase will lead to higher current conversion during Regime I. These two results imply the following proposition. Proposition 3. . Worse (better) governance delays (hastens) the optimal ∂t  time of a forest transition, i.e. ∂θ1 N0. Proof of this proposition follows the approach of Kamien and Schwartz (1991, pp. 169–171) for determining the comparative dynamics of a change in a parameter on the maximized objective function. The partial derivative of the maximized objective function (3) with respect to the governance risk parameter is. Z ∂J  ðθÞ ¼− ∂θ Z −

t1 0 ∞ t1

te−ðrþθÞt ðBð F  Þ þ qðL Þ−C 1 ðc ÞÞdt te−ðrþθÞt ðBð F 1  Þ þ Rð F  − F  1 Þ þ qðL Þ−C 2 ðg  ÞÞdt b 0 ð18Þ

where an asterisk denotes a variable at its optimal value. The partial derivative of the maximized objective function (3) with respect to the switching time t1 between Regimes I and II is   ∂J  ðθÞ ¼ e−ðrþθÞt 1 ½Bð F  ðt 1 ÞÞ þ qðL ðt 1 ÞÞ−C 1 ðc ðt 1 ÞÞ−ðB F 1 ∂t 1   þR F  ðt 1 Þ−F 1 þ qðL ðt 1 ÞÞ−C 2 ðg  ðt 1 ÞÞÞ

ð19Þ

  ¼ −e−ðrþθÞt 1 R F  ðt 1 Þ−F 1 b0 as by definitionB(F1⁎) = B(F⁎(t1)) and C1(c⁎(t1))= C2(g⁎(t1))=0 at t1 =  ∂t  t1⁎. It follows from Eqs. (18) and (19) that 1 ¼ ∂ J =∂θ N0. ∂θ

∂ J =∂t 1

Intuitively, because inadequate governance delays the time at which the forest transition occurs, the pre-transition phase when there is a wedge between the returns from agricultural and forested land continues on. This result for the pre-forest transition phase is illustrated graphically in Fig. 2. Worse governance therefore implies that at any time t ≤ t1⁎ the likelihood of attaining a forest transition is lower, whereas better governance improves this likelihood. The

Fig. 2. Governance impacts on Regime I (the pre-forest transition phase). The figure illustrates the effects of worse governance during the pre-forest transition phase of Regime I. As indicated in Proposition 3, the impact is to delay the time of the forest transition∂t⁎1/ ∂θN 0. The result is that the wedge between the capitalized values of agricultural and forested land continues on longer. Worse governance therefore implies that at any time t≤t⁎1 the likelihood of attaining a forest transition is lower.

next section empirically tests if this prediction holds for a crosssection of developing countries. 3. Estimation strategy and results. To summarize, the main hypothesis to emerge from our theoretical model is that worse (better) governance decreases (increases) the likelihood of a forest transition in developing countries. To test this hypothesis, we examine the probability of a forest transition for deforestation occurring for these countries over a specific time period, 1990 to 2010. This allows use a probit model, which takes the value of one when a country has experienced by 2010 a forest transition over the 1990– 2010 period, and zero otherwise.10 In identifying which countries have experienced a forest transition over 1990–2010, we follow Hosonuma et al. (2012), who categorize developing countries according to their stage along the forest transition path during this period. Their analysis suggests that 13 developing countries experienced a forest transition over this period. Applying the same approach to other studies, we identified seven more countries that showed evidence of a forest transition – a sustained shift from net deforestation to net reforestation – over 1990 to 2010 (Bray, 2010; Meyfroidt and Lambin, 2011; Rudel et al., 2005). We confirmed these results through our own appraisal based on the 2010 global Forest Resource Assessment (Food and Agriculture Organization of the United Nations (FAO), 2010). From our appraisal, we identified an additional seven countries that sustained net reforestation from 1990 to 2000, 2000 to 2005 and 2005 to 2010. Overall, as indicated in Appendix Table 1, 27 out of 132 developing countries over 1990–2010 attained a forest transition by 2010. In addition, we identify a range of governance variables that are consistent with our modeling approach of including a governance risk parameter in the effective discount rate. As discussed previously, these are governance indicators that alter the planner's perception of the risk of future versus current gains from any allocation. For example, this risk is greater when legal contracts are absent or weakly enforced (Deacon, 1994), ownership risk is high (Agrawal, 2007; Bohn and Deacon, 2000; Chhatre and Agrawal, 2008), political instability is pervasive (Ferreira and Vincent, 2010; Galinato and Galinato, 2012 and Galinato and Galinato, 2013), and comprehensive forest policy and regulations are absent (Agrawal et al., 2008; Food and Agriculture Organization of the United Nations (FAO), 2010). There are three categories of governance indicators that seem most consistent with our analysis: forest governance, economywide governance, and country risk classification and lending premium variables (see Appendix Table 2). Unfortunately, it is difficult to find global datasets corresponding to these categories that contain observations for all 132 developing countries of our sample, including the sub-set of 27 forest transition economies. For instance, reliable forest governance indicators across developing countries is generally lacking (Agrawal, 2007; Agrawal et al., 2008; Chhatre and Agrawal, 2008), although a few general indicators are reported in Food and Agriculture Organization of the United Nations (FAO) (2010). With the exception of country risk classification by the OECD, country risk classification and lending premium variables usually have poor coverage of developing countries. Whereas the World Bank's Worldwide Governance Indicators for political stability, regulatory quality and rule of law are present for all 132 developing countries of our sample, the Bank's more detailed Country Policy and Institutional Assessment indicators are available for only 72 of these countries. Appendix Table 2 lists the governance indicators that fit our three categories, and shows the number of observations available for each variable for all developing and forest transition countries of our sample. 10 Delacote and Garcia (2015) employ a similar probit model for the likelihood of a forest transition occurring over 1990–2010 in the first stage of their seemingly unrelated regression (SUR) of deforestation and land-use change in developing countries.

E.B. Barbier, A. Tesfaw / Ecological Economics 119 (2015) 252–261

As the main aim is to test whether governance influences the likelihood of a forest transition by 2010 over the 1990 to 2010 period for our sample of 132 developing countries, we estimate the following limited dependent variable regression.   Pr t 1 ≤2010 ¼ 1jG; X ¼ F ðGi βG þ X i βX Þ;

ð20Þ

Table 1 Probit estimates of the likelihood of a forest transition in developing countries by 2010. Basic regression

Including governance

−1.124 (−2.917)⁎⁎ −4.004 (−2.744)⁎⁎ 8.820 (1.263) 1.504 (2.082)⁎

−1.850 (−2.878)⁎⁎ −4.335 (−2.351)⁎ 5.537 (0.887) 2.036 (2.458)⁎⁎

Rule of law (2000–2010 average)

1.304 (1.971)⁎ –

Regulatory quality (2000–2010 average)

–

Political stability (2000–2010 average)

–

Dummy variable if forest policy and legal framework at national level exists Observations Pseudo-R2 Log-likelihood Hosmer–Lemeshow Observations predicted (%)

–

2.317 (2.917)⁎⁎ 1.631 (3.309)⁎⁎ −1.132 (−3.015)⁎⁎ −0.185 (−0.720) 0.890 (1.997)⁎ 108 0.74 30.18⁎⁎

Constant Agriculture, value added (% of GDP)/100

where i is each country observation, Gi is a vector of governance variables and Xi a vector of additional variables. Note that, if better (worse) governance increases (decreases) the probability of a forest transition occurring, thenβG N 0. Finally, assuming that F(·) is a standard normal cumulative distribution, Eq. (20) specifies a probit model, which takes the value one when the country has experienced a turning point over the 1990–2010 period (i.e., by 2010), and zero otherwise. We follow previous empirical analyses for individual and groups of countries in the forest transition literature and also include additional variables Xi.11 Appendix Table 2 lists the variables employed in our analysis. Our competing land use model suggests that two important factors that should influence the likelihood of a forest transition are the value, or returns, to agricultural production and the flow of production and environmental benefits of forests. We use agricultural value added as a percentage share of gross domestic product (GDP) in 2010 as our proxy for the value of agricultural production, and forest rents from roundwood harvest production as a percentage of GDP in 2010 as the proxy for forest benefits.12 However, reliable estimates of the various environmental benefits of forests are not possible for a wide range of developing countries (World Bank, 2011). In addition, a key outcome of our competing land use model is that, during the pre-transition phase, forest land is relatively abundant compared to agricultural land, but as the transition approaches, agricultural land is relatively more abundant. To test for this relationship, the percentage share of agricultural land to total land area is also included in our estimations. Finally, the occurrence of the forest transition may differ for developing countries located in tropical as opposed to temperate regions. To allow for this possible geographic effect, we use the latitude for the center point of a country expressed in degrees. Our strategy for estimating Eq. (20) involves two sets of regressions. First, we estimate a basic regression without any governance variables using the four variables suggested by our competing land use model: agricultural value added as a share of GDP, forest rents as a share of GDP, agricultural land share of total land, and latitude. This set of estimations is conducted with and without the various additional variables listed in Appendix Table 2. Our second set of regressions of Eq. (20) then adds to the basic regression different combinations of the available forest governance, economy-wide governance, and country risk classification and lending premium variables. Again, these regressions are estimated with and without additional covariates.13 Robust covariance matrix estimation was performed to adjust the estimated asymptotic covariance matrix for possible misspecification.14 To correct for unspecified latent heterogeneity in the covariance matrix, we use the standard sandwich estimator, which is the choice-based sampling estimator with weights equal to one. 11 See, for example, Bray (2010); Hosonuma et al., 2012; Köthke et al. (2013); Lambin and Mefroyidt (2010); Meyfroidt and Lambin (2011); Pfaff and Walker (2010); and Rudel et al., 2005; 12 As an alternative proxy for agricultural values we also use agricultural value added per worker (2005 $). However, the coefficient for this variable is statistically insignificant in all estimation specifications. Unfortunately, there is no alternative proxy for forest benefits. 13 Given that the independent variables in Appendix Table 2 are measured in various units, we tested whether each of these variables has sufficient range for estimating the probit model (20). The test reveal that the range of each independent variable is sufficient for the estimation. In addition, we tested and confirmed that each dummy variable has sufficient observations for estimating the dependent variable in Eq. (20). 14 We also employ two Lagrange multiplier tests for heteroskedasticity. The tests are not statistically significant for either estimation specification, and thus the null hypothesis of homoscedasticity cannot be rejected. However, given the low number of observations (108 after missing observations), the model used in the tests may be weakly identified.

257

Forest rents (% of GDP)/100 Agricultural land (% of land area)/100 Latitude/100

120 0.71 16.49⁎⁎ 3.17 93 (77.5%)

4.24 85 (78.7%)

Notes: the dependent variable is the probability that a forest transition occurs by 2010; tratios based on robust standard errors in parentheses; the Hosmer–Lemeshow diagnostic is a test of the null hypothesis that predicted values match actual values. ⁎⁎ Significant at the 1% level. ⁎ Significant at the 5% level.

Table 1 depicts the results of our two sets of estimations. The first estimation is for the basic regression without any governance variables. The version depicted is without the inclusion of additional covariates from the list in Appendix Table 2; the regression results do not change when these extra variables are added, and none of their coefficients are statistically significant. All the competing land use model variables have statistically significant coefficients, with the exception of forest rents as a percentage of GDP. Moreover, the coefficients have the expected sign. The likelihood of a forest transition decreases with agricultural value added as a percentage of GDP, increases with agricultural land's share of total area, and increases for countries at higher latitudes (i.e., in more temperate zones). The second estimation in Table 1 is the most robust regression with governance variables. Of all the governance variables utilized, either on their own or in different combinations, only the presence of forest policy, rule of law and regulatory quality have statistically significant coefficients. The third variable from the Worldwide Governance Indicators, political stability, is also included in the regression, although its estimated parameter is not significant. Whereas forest policy and the rule of law increase the likelihood of a forest transition, regulatory quality reduces this probability. Comparing the first and second regressions in Table 2, the inclusion of governance variables does not affect the significance or sign of the coefficients for the competing land use variables. In addition, the results for the second regression depicted in Table 2 do not change when additional control variables are added, and their coefficients are not significant. Table 2 depicts the marginal effects corresponding to the estimation that includes governance variables. The marginal effects are calculated for the mean of all observations, and for the mean of the observations of two subsets of developing countries: those with and without a forest policy and legal framework. The results confirm that the competing land use variables are an important influence on the likelihood of a forest transition (i.e., the dependent variable in Eq. (20) having a value of one). For all countries, a one-unit increase in agriculture's share of total land area raises this probability by 48%, a one-unit increase in

258

E.B. Barbier, A. Tesfaw / Ecological Economics 119 (2015) 252–261

Table 2 Marginal effects of the likelihood of a forest transition in developing countries by 2010. Observations used for means Forest Forest All policy policy dummy = 0 dummy = 1 Constant Agriculture, value added (% of GDP) / 100 Forest rents (% of GDP) / 100 Agricultural land (% of land area) / 100 Latitude / 100 Rule of law (2000–2010 average) Regulatory quality (2000–2010 average) Political stability (2000–2010 average) Dummy variable if forest policy and legal framework at national level exists

−0.280⁎⁎ −0.658⁎ 0.840 0.309⁎⁎ 0.351⁎⁎ 0.247⁎⁎ −0.172⁎⁎

−0.475⁎⁎ −1.113⁎ 1.422 0.523⁎⁎ 0.595⁎⁎ 0.419⁎⁎ −0.291⁎⁎

−0.434⁎⁎ −1.016⁎ 1.298 0.477⁎⁎ 0.543⁎⁎ 0.383⁎⁎ −0.265⁎⁎

−0.281 0.135⁎

−0.476 0.229⁎

−0.435 0.157⁎

Notes: The dependent variable is the probability that a forest transition occurs by 2010. ⁎⁎ Significant at the 1% level. ⁎ Significant at the 5% level.

latitude by 54%, and a one-unit decrease in agriculture's share of GDP doubles the likelihood of a forest transition. However, these effects are highly conditioned on whether or not countries have a forest policy. For example, across those countries without such a policy, a ones-unit decrease in agriculture's share of GDP increases the probability of a forest transition by 66%. The overall effect of a presence of a forest policy is indicated in the last row of Table 2. For all countries, if the forest policy dummy variable changes from 0 to 1, then the probability that a forest transition occurs rises by 16%. A one-unit rise in the rule of law increases the probability of a forest transition across all developing countries by 38%. For our sample of 132 countries, the average (and median) rule of law indicator is −0.55, and a one-standard deviation change is 0.60. Thus, if the average rule of law index across all countries improved by one standard deviation to 0.05, then the likelihood of a forest transition would increase by 23% (38.3 % × 0.60). In contrast, a one-unit increase in regulatory quality lowers the probability of a transition occurring by 27%. Average regulatory quality across our sample is −0.51 (median is −0.49), and a onestandard deviation is 0.68. Consequently, if average regulatory quality rises to 0.17, then the likelihood of a forest transition would decline by 18% (26.5 % × 0.68). In sum, our regression results provide strong support for the framework of the competing land use model that we develop to analyze the forest transition. Two variables that correspond closely to that model – agricultural share of GDP and agricultural land share of total land – have a significant influence on the occurrence of a forest transition in all estimated specification. Forest rents as share of GDP is not statistically significant, but this may be because it is a poor proxy for all the benefits of the forest, especially the myriad environmental benefits. We also find that the likelihood of a forest transition increases for developing countries at higher latitudes. In contrast, for nearly all our governance indicators, the hypothesis that governance influences the probability of a forest transition is rejected. Only the presence of forest policy, rule of law and regulatory quality have statistically significant coefficients in explaining the probability of a forest transition. Whereas forest policy and the rule of law increase the likelihood of a forest transition, regulatory quality reduces this probability. 4. Conclusion To our knowledge, this is the first paper that develops an economic model of the forest transition, including explicitly the influence of governance, which we then employ as the basis of an empirical analysis across developing countries. A key determinant of the occurrence of the transition is the difference between the returns from agricultural versus

forested land. In the phase preceding the forest transition, agricultural land is relatively scarce, and its capitalized value is significantly greater than that of primary forests. More forests are converted to agriculture. However, as conversion proceeds, agricultural land area expands while forests decline, and the difference between the capitalized value of agricultural and forest land narrows. The transition to the next phase of forest recovery begins once the two land values converge (see Fig. 2). We show that, if the most likely impact of good or bad governance is to alter the risk of future versus current gains from this land allocation decision, then there can be an impact on the timing of the forest transition. The key proposition to emerge from our analysis is that worse (better) governance delays (hastens) the time when a forest transition occurs. The result is that the wedge between the returns from agricultural and forested land widens, and the forest transition is postponed (see Fig. 2). Inadequate governance therefore implies that the likelihood of attaining a forest transition is lower. Our empirical test of this hypothesis across a sample of developing countries, 27 of which attained a forest transition by 2010, provides strong support for the competing land use approach of analyzing this transition. But, of all the governance indicators we analyzed (See Appendix Table 2), only the presence of forest policy, rule of law and regulatory quality are statistically significant in explaining the probability of a forest transition. Moreover, whereas forest policy and the rule of law increase the likelihood of a forest transition, regulatory quality reduces this probability. Our finding that better regulatory quality diminishes rather increases the likelihood of a forest transition may seem surprising. However, as pointed out by several studies, in recent decades, the main cause of deforestation in developing countries, has changed from state-funded enterprises and large-scale settlement investments to more decentralized decision-making by farmers, land speculators, Agri-business enterprises and ranchers (Chomitz et al., 2007; Gibbs et al., 2010; Lambin and Meyfroidt, 2011; Rudel, 2007). An improvement in economy-wide regulatory quality, which according to the Worldwide Governance indicators “captures perceptions of the ability of the government to formulate and implement sound policies and regulations that permit and promote private sector development”, could actually facilitate “enterprise-driven” deforestation by the private sector. Evidence for some developing countries has shown that a more favorable regulatory and private property rights structure may be enhancing private agents' legal claims to forested land and the profitability of their activities, but the result may be more rather than less deforestation (Angelsen and Rudel, 2013; Liscow, 2013). Further research analyzing the relationship between the returns to competing land use, governance and the forest transition would benefit from better cross-country indicators of forest governance. Reliable forest governance indicators across developing countries are generally lacking, yet very much needed not only for the type of analysis conducted here but also for guiding forest policy (Agrawal, 2007; Agrawal et al., 2008; Chhatre and Agrawal, 2008). Our finding that the presence of a national forest policy and legal framework, the rule of law and regulatory quality have a significant influence on the occurrence of a forest transition confirms the policy importance of better forest governance indicators. Finally, although our paper has emphasized the role of governance in influencing the forest transition, economy-wide and sector policies also have an important impact. The agricultural and forest sectors of many developing countries are rife with widespread policy and market failures, which have the consequence of distorting incentives and the relative values of agricultural versus forested lands (Angelsen and Rudel, 2013; Barbier et al., 2005 and Barbier et al., 2010; Bulte et al., 2007; Ferreira and Vincent, 2010;’ Grainger and Malayang, 2006; Lambin and Mefroyidt, 2010; López and Galinato, 2005; Rudel, 2007). A more complete model and empirical analysis of competing land uses and the forest transition in developing countries should take into account both the effects of governance and these wider market and policy failures.

E.B. Barbier, A. Tesfaw / Ecological Economics 119 (2015) 252–261

259

Appendix A Appendix Table 1 Developing countries displaying forest transition, 1990–2010. Country

Source

Belarus Bhutan Bulgaria Cape Verde China Costa Rica Cote d'Ivoire Cuba Dominican Republic Egypt

Bray (2010) Meyfroidt and Lambin (2011) Bray (2010); Meyfroidt and Lambin (2011) Hosonuma et al. (2012) Hosonuma et al. (2012) Hosonuma et al. (2012) Hosonuma et al. (2012) Hosonuma et al. (2012) Rudel et al. (2005) Food and Agriculture Organization of the United Nations (FAO) (2010); positive net reforestation for 1990–2000, 2000–2005, and 2005–2010 Food and Agriculture Organization of the United Nations (FAO) (2010); positive net reforestation for 1990–2000, 2000–2005, and 2005–2010 Hosonuma et al. (2012) Hosonuma et al. (2012) Food and Agriculture Organization of the United Nations (FAO) (2010); positive net reforestation for 1990–2000, 2000–2005, and 2005–2010 Hosonuma et al. (2012) Rudel et al. (2005) Hosonuma et al. (2012) Bray (2010) Hosonuma et al. (2012) Food and Agriculture Organization of the United Nations (FAO) (2010); positive net reforestation for 1990–2000, 2000–2005, and 2005–2010 Hosonuma et al. (2012) Food and Agriculture Organization of the United Nations (FAO), 2010; positive net reforestation for 1990–2000, 2000–2005, and 2005–2010 Hosonuma et al. (2012) FAO (2010a); positive net reforestation for 1990–2000, 2000–2005, and 2005–2010 Food and Agriculture Organization of the United Nations (FAO) (2010); positive net reforestation for 1990–2000, 2000–2005, and 2005–2010 Bray (2010); Meyfroidt and Lambin (2011) Hosonuma et al. (2012)

Fiji Gambia India Lebanon Lesotho Morocco Philippines Romania Rwanda Serbia Swaziland Syria Thailand Tunisia Turkey Ukraine Vietnam

Developing countries are all low and middle-income economies with 2012 per capita income of $12,615 or less, as defined in the World Bank's World Development Indicators 2014.

Appendix Table 2 Variable definitions and data sources. Variable description

Forest governance variables Percentage (%) of public forest ownership and management rights, 2005 1 if forest policy and legal framework at national level exists in 2008, 0 otherwise Permanent forest estate (% of forest area), 2010 Forest within protected area (% of forest area), 2010 Forest with management plan (% of forest area), 2010

Economy-wide governance variables Estimate of political stability/no violence (ranges from approximately −2.5 (weak) to 2.5 (strong) governance performance), 2000–2010 average Estimate of regulatory quality (ranges from approximately −2.5 (weak) to 2.5 (strong) governance performance), 2000–2010 average Estimate of rule of law (ranges from approximately −2.5 (weak) to 2.5 (strong) governance performance), 2000–2010 average CPIA property rights and rule-based governance rating (1 = low to 6 = high), 2010 CPIA policy and institutions for environmental sustainability rating (1 = low to 6 = high), 2010 CPIA transparency, accountability, and corruption in the

Source

Number of observations for: All countries

Forest transition countries

Global Forest Resource Assessment (Food and Agriculture Organization of the United Nations (FAO), 2010) Global Forest Resource Assessment (Food and Agriculture Organization of the United Nations (FAO), 2010) Global Forest Resource Assessment (Food and Agriculture Organization of the United Nations (FAO), 2010) Global Forest Resource Assessment (Food and Agriculture Organization of the United Nations (FAO), 2010) Global Forest Resource Assessment (Food and Agriculture Organization of the United Nations (FAO), 2010)

113

26

117

26

73

17

84

19

72

19

Worldwide Governance Indicators 2014

132

27

Worldwide Governance Indicators 2014

132

27

Worldwide Governance Indicators 2014

132

27

World Development Indicators 2014

76

8

World Development Indicators 2014

76

8

World Development Indicators 2014

76

8 (continued on next page)

260

E.B. Barbier, A. Tesfaw / Ecological Economics 119 (2015) 252–261

Appendix Table 2 (continued) Variable description

public sector rating (1 = low to 6 = high), 2010 Country risk classification and lending premium variables Country risk classification (0–7), based on credit risk and qualitative assessment of other risk factors ((e.g. war, expropriation, revolution, civil disturbance, floods, earthquakes), 1999–2014 average Country risk premium (%) based on the latest bond ratings and appropriate default spreads for different countries, January 1, 2014 Risk premium on lending: the interest rate charged by banks on loans to private sector customers minus the “risk free” treasury bill interest rate at which short-term government securities are issued or traded in the market, 2010 Competing land use variables Agriculture, value added (% of GDP), 2010 Agriculture value added per worker (constant 2005 US$), 2010 Forest rents (% of GDP), 2010 Agricultural area (% of land area), 2010 Latitude for the center point of a country expressed in degrees Other variables Forest area (% of land area), 2010 Rural population (% of total population), 2010 Rural population growth (annual %), 2010 Road density (km of road per 100 sq. km of land area), 2010 Agricultural land (sq. km), 2010 Arable land (% of land area), 2010 Forest area (sq. km), 2010 Land area (sq. km) GDP per capita, PPP (constant 2005 international $), 2010 Trade measured by the sum of exports and imports as a percentage of GDP at 2005 constant prices, 2010 Rural Access Index: the percentage of rural people who live within 2 km (typically equivalent to a walk of 20 min) of an all-season road as a proportion of the total rural population, 2004

Source

Number of observations for: All countries

Forest transition countries

OECD Trade and Agriculture Directorate http://www.oecd.org/tad/xcred/crc.htm

111

25

Aswath Damodaran, Stern Business School, New York University, http://pages.stern.nyu.edu/~%20adamodar/ New_Home_Page/datafile/ctryprem.html World Development Indicators 2014

74

21

61

14

World Development Indicators 2014 World Development Indicators 2014

129 109

27 24

World Development Indicators 2014 World Development Indicators 2014 CIA: The World Fact Book 20,104

121 132 131

26 27 27

World Development Indicators 2014 World Development Indicators 2014 World Development Indicators 2014 World Development Indicators 2014

132 132 132 50

27 27 27 13

World Development Indicators 2014 World Development Indicators 2014 World Development Indicators 2014 World Development Indicators 2014 World Development Indicators 2014 Penn World Table Version 7.1 2014

132 132 132 132 130 131

27 27 27 27 27 27

World Bank, Transport Group, Rural Access Index

125

26

References Agrawal, A., Chhatre, A., Hardin, R., 2008. Changing governance of the world's forests. Science 320, 1460–1462. Agrawal, A., 2007. Forests, governance, and sustainability: common property theory and its contributions. Int. J. Commons 1, 111–136. Alix-Garcia, J.M., de Janvry, A., Sadoulet, E., 2004. A tale of two communities: explaining deforestation in Mexico. World Dev. 33 (2), 219–235. Alston, L.J., Libecap, G.D., Schneider, R.R., 1996. The determinants and impact of property rights: land titles on the Brazilian frontier. J. Law Econ. Org. 12 (1), 25–61. Amacher, G.S., Koskela, E., Ollikainen, M., 2008. Deforestation and land use under insecure property rights. Environ. Dev. Econ. 14, 281–303. Amacher, G.S., Ollikainen, M., Koskela, E., 2009. Economics of forest resources. MIT Press, Cambridge, MA. Amit, R., 1986. Petroleum reservoir exploitation: switching from primary to secondary recovery. Oper. Res. 34, 534–549. Angelsen, A., 2007. Forest Cover Change in Space and Time: Combining the von Thünen and Forest Transition Theories. World Bank Policy Research working paper 4117. Washington, D.C, The World Bank,. Angelsen, A., Rudel, T.K., 2013. Designing and implementing effective REDD+ policies: a forest transition approach. Rev. Environ. Econ. Policy 7, 91–113. Araujo, C., Bonjean, C., Combes, J.-L.,., Motel, P.,., Reis, E.,., 2009. Property rights and deforestation in the Brazilian amazon,. Ecol. Econ. 68 (8–9), 2461–2468. Barbier, E.B., Burgess, J.C., 1997. The economics of tropical forest land use options. Land Econ. 73, 174–195. Barbier, E.B., Burgess, J.C., 2001. The economics of tropical deforestation. J. Econ. Surv. 15, 413–433. Barbier, E.B., Burgess, J.C., Grainger, A., 2010. The forest transition: towards a more comprehensive theoretical framework. Land Use Policy 27, 98–107. Barbier, E.B., Damania, R., Leonard, D., 2005. Corruption, trade, and resource conversion. J. Environ. Econ. Manag. 50, 276–299. Bhattarai, M., Hammig, M., 2004. Governance, economic policy, and the environmental Kuznets curve for natural tropical forests. Environ. Dev. Econ. 9, 367–382.

Bohn, H., Deacon, R.T., 2000. “Ownership risk, investment, and the use of natural resources. Am. Econ. Rev. 90 (3), 526–549. Bray, D.B., 2010. Forest Cover Dynamics and Forest Transitions in Mexico and Central America: Towards a ‘Great Restoration’? Chapter 5 In: Nagendra, H., Southworth, J. (Eds.), Reforesting Landscapes: Linking Pattern and Process. Springer, Netherlands, pp. 85–120 Bulte, E.H., Damania, R., López, R., 2007. On the gains of committing to inefficiency :corruption, deforestation and low land productivity in Latin America. J. Environ. Econ. Manag. 54, 277–295. Chhatre, A., Agrawal, A., 2008. Forest commons and local enforcement. Proc. Natl. Acad. Sci. 105, 13286–13291. Chomitz, K.M., Buys, P., De Luca, G., Thomas, T.S., Wertz-Kanounnikoff, S., 2007. At Loggerheads? Agricultural Expansion, Poverty Reduction, and Environment in the Tropical Forests. The World Bank, Washington DC. Damatte, O., Delacote, P., 2012. On the economic factors of deforestation: what can we learn from quantile analysis? Econ. Model. 29, 2427–2434. Deacon, R.T., 1994. Deforestation and the rule of law in a cross section of countries. Land Econ. 70 (4), 414–430. Wolfersberger, J., P. Delacote and S. Garcia. 2015. “An empirical analysis of forest transition and land-use change in developing countries.” mimeo. INRA-Laboratoire d'Economie Forestière, Nancy, France. Ferreira, S., Vincent, J., 2010. Governance and timber harvest. Environ. Resour. Econ. 47, 241–260. Food and Agriculture Organization of the United Nations (FAO), 2010,. Global Forest Resource Assessment. FAO, Rome. Galinato, G., Galinato, S., 2012. The effects of corruption control, political stability and economic growth on deforestation-induced carbon dioxide emissions. Environ. Dev. Econ. 17, 67–90. Galinato, G., Galinato, S., 2013. The short-run and long-run effects of corruption control and political stability on forest cover. Ecol. Econ. 89, 153–161. Gibbs, H.K., Ruesch, A.S., Achard, F., Clayton, M.K., Holmgren, P., Ramankutty, N., Foley, J.A., 2010. Tropical forests were the primary sources of new agricultural lands in the 1980s and 1990s. Proc. Natl. Acad. Sci. 107, 16732–16737.

E.B. Barbier, A. Tesfaw / Ecological Economics 119 (2015) 252–261 Grainger, A., 2004. Societal change and the control of deforestation in Thailand. Int. J. Sustain. Dev. 11, 364–379. Grainger, A., Malayang, B., 2006. A model of policy changes to secure sustainable forest management and control of deforestation in the Philippines. For. Policy Econ. 8, 67–80. Hansen, M.C., Potapov, P.V., Moore, R., Hancher, M., Turubanova, S.A., Tyukavina, A., Thau, D., Stehman, S.V., Goetz, S.J., Loveland, T.R., Kommareddy, A., Egorov, A., Chini, L., Justice, C.O., Townshend, J.R.G., 2013. High-resolution global maps of 21st-century forest cover change. Science 342, 850–853. Hartwick, J., van Long, N., Tian, H., 2001. Deforestation and development in a small open economy. J. Environ. Econ. Manag. 41, 235–251. Hosonuma, N., Herold, M., De Sy, V., De Fries, R., Brockhaus, M., Verchot, L., Angelsen, A., Romijn, E., 2012. An assessment of deforestation and forest degradation drivers in developing countries. Environ. Res. Lett. 7, 1–12. Kamien, M.I., Schwartz, N.L., 1991. Dynamic Optimization: The Calculus of Variations and Optimal Control in Economics and Management. 2nd ed. Elsevier, Amsterdam. Kaufmann, D., Kraay, A., Mastruzzi, M., 2009. Governance Matters VIII: Aggregate and Individual Governance Indicators 1996-2008. Policy Research Working Paper 4978. The World Bank, Washington, D.C. Köthke, M., Leischner, B., Elasser, P., 2013. Uniform global deforestation patterns – an empirical analysis. For. Policy Econ. 28, 23–37. Lambin, E., Mefroyidt, P., 2010. Land use transitions: socio-ecological feedback versus socio-economic change. Land Use Policy 27, 108–118. Lambin, E.F., Meyfroidt, P., 2011. Global land use change, economic globalization, and the looming land scarcity. Proc. Natl. Acad. Sci. 108, 3465–3472. Liscow, Z., 2013. Do property rights promote investment but cause deforestation? Quasiexperimental evidence from Nicaragua. J. Environ. Econ. Manag. 65, 241–261. López, R., Galinato, G., 2005. Deforestation and forest-induced carbon dioxide emissions in tropical countries: how do governance and trade openness affect the forest-income relationship? J. Environ. Dev. 2005 (14), 73–98.

261

Makris, M., 2001. Necessary conditions for infinite horizon discounted two-stage optimal control problems. J. Econ. Dyn. Control. 25, 1935–1950. Mather, A.S., 1990. Global forest resources. Bellhaven Press, London. Mather, A.S., 1992. The forest transition. Area 24, 367–379. Mather, A.S., 2007. Recent Asian forest transitions in relation to forest-transition theory. Int. For. Rev. 9, 491–501. Mather, A.S., Needle, C.L., 1999. Development, democracy and forest trends. Glob. Environ. Chang. 9, 105–118. Meyfroidt, P., Lambin, E., 2011. Global forest transition: prospect for an end to deforestation. Annu. Rev. Environ. Resour. 36, 343–371. Meyfroidt, P., Rudel, T.K., Lambin, E., 2010. Forest transitions, trade, and the global displacement of land use. Proc. Natl. Acad. Sci. 107, 20917–20922. Nguyen-Van, P., Azomahou, T., 2007. Nonlinearities and heterogeneity in environmental quality: an empirical analysis of deforestation. J. Dev. Econ. 84, 291–309. Ollivier, H., 2012. Growth, deforestation and the efficiency of the REDD mechanism. J. Environ. Econ. Manag. 64, 312–327. Pfaff, A., Walker, R., 2010. Regional interdependence and forest ‘transitions’: substitute deforestation limits the relevance of local reversals. Land Use Policy 27, 119–129. Rudel, T.K., 2007. Changing agents of deforestation: from state initiated to enterprise driven processes, 1970–2000. Land Use Policy 24, 35–41. Rudel, T.K., Coomes, O.T., Moran, E., Achard, F., Angelsen, A., Xu, J., Lambin, E., 2005. Forest transitions: towards a global understanding of land use change. Glob. Environ. Chang. 15, 23–31. Tomiyama, K., 1985. Two-stage optimal control problems and optimality conditions. J. Econ. Dyn. Control. 9 (3), 317–337. World Bank, 2011. The Changing Wealth of Nations: Measuring Sustainable Development in the New Millennium. The World Bank, Washington, D.C.