Environmental externalities in traditional agriculture and the impact of trade liberalization: the case of Ghana

ELSEVIER

Journal of Development Economics Vol. 53 (1997) 17-39

JOURNAL OF Development ECONOMICS

Environmental externalities in traditional agriculture and the impact of trade liberalization: the case of Ghana 1 Ram6n L6pez University of Maryland, College Park, MD 20742, USA

Abstract

This paper provides a framework for the evaluation of trade policies that explicitly considers the effects of such policies on the environmental resources (biomass). In doing this we estimate the effects of biomass as a factor of agricultural production and test the hypothesis that farmers efficiently exploit biomass, a resource held in common property. The main findings are: (i) biomass is an important factor of production in Ghanian agriculture; (ii) the biomass resource is exploited beyond the socially optimal level; (iii) a deepening of trade liberalization is likely to induce further losses of biomass and deforestation; and (iv) national income is likely to fall in response to further trade liberalization. © 1997 Elsevier Science B.V. JEL classification: 013 Keywords: Biomass; Externalities; Common property; Traditional agriculture; Trade policy; Ghana

1. I n t r o d u c t i o n The close dependence of rural incomes on environmental factors in poor areas of the world is by now widely accepted. Excessive deforestation, significant declines in natural biomass, reduction in soil quality, in many cases induced by

i I am grateful to two anonymous referees for very helpful comments. I am also grateful to Erik Lichtenberg for useful comments. Financial support for this research was provided by the World Bank. 0304-3878/97/$17.00 © 1997 Elsevier Science B.V. All rights reserved. PH S 0 3 0 4 - 3 8 7 8 ( 9 7 ) 0 0 0 1 5-1

18

R. Lbpez/ Journal of Development Economics 53 (1997) 17-39

rapid population growth, have been identified as key factors explaining an apparent fall or stagnation of rural income in Africa and in certain areas of Latin America and Asia (African Development Bank et al., 1984, FAO, 1986). The fact that a significant part of the land used in farming and most forested areas in the Third World are controlled through various forms of common property and, sometimes, public property has been emphasized as a source of resource overexploitation (Glantz, 1977, Allen, 1985, Sinn, 1988, Perrings, 1989, L6pez and Niklitschek, 1991). 2 Whether indigenous common property resource controls do cause overexploitation (i.e. whether the 'tragedy of the commons' applies) has been the subject of much controversy. Various authors have argued that traditional communities develop controls on the use of the resources inducing a socially efficient exploitation (Dasgupta and Maler, 1990, Larson and Bromley, 1990). That is, traditional systems would internalize the potential externalities arising because of lack of individual resource ownership. Whether or not this is true can only be elucidated through empirical work. To our knowledge, the only empirical test of the commons' efficient resource exploitation hypothesis has been developed by L6pez (1993) for western CSte d'Ivoire. L6pez rejects this hypothesis, documenting socially excessive deforestation and biomass depletion that causes large income losses for the rural communities. The large number of studies focusing on the implications of rural environmental degradation have permitted a qualitative understanding of how the income of rural communities is affected by policies that directly or indirectly impinge on the resource utilization patterns. However, these qualitative analyses have not yet been followed by empirical works that would allow us to obtain a more clear idea of the quantitative importance of the environment as a factor of production. 3 The absence of such estimates has, of course, made it impossible to measure the effects of government policies on agricultural income despite the fact that suitable theoretical models do exist (L6pez and Niklitschek, 1991, Deacon, 1992). The principal purpose of this paper is to estimate the value of environmental resources as factors of agricultural production and to measure the potential effects of various economy-wide policies on agricultural income by explicitly accounting for such environmental effects. In doing this we estimate a production function that, in addition to the conventional inputs, accounts for the effects of environmental factors. Based on these estimates we are able to obtain an idea of the extent of overexploitation of the environmental resource (i.e. the extent by which fallow periods are too short). A simple general equilibrium model is used to measure how

2 Feder et ad. (1988) have empirically documented the negative effects of insecure land tenure property rights on agricultural productivity. 3 An exception is the above-mentioned study by L6pez (1993) who obtained estimates for the contribution of biomass to agricultural production and for the effects of output prices or biomass and agricultural productivity.

R. L6pez/ Journal of Development Economics 53 (1997) 17-39

19

agricultural price policies and trade policies are likely to affect the contribution of agriculture to national income. In doing this analysis we use a unique data set for an area in western Ghana. These data match household survey information from 12 villages covering the period 1988 and 1989 with remote sensing data on the area under closed forest, area under natural bush (fallow area), and density of biomass in the fallow areas. The detailed survey includes data on production, land use, employment, use of conventional inputs, and demographic characteristics for a large number of farm households in the 12 villages. Field visits to the area provided important qualitative information to complement the survey data and the remote sensing data. Moreover, other parameters required for the general equilibrium model were obtained from published data sources. We try to shed some light on the implications of a deepening of the on-going trade liberalization reforms in Ghana taking explicit account of the existence of an environmental distortion in the agricultural sector. It is not clear that a complete elimination of tariffs and export taxes will necessarily cause an increase in national income in the presence of other distortions. Based on empirical estimates of the magnitude of the various distortions, environmental and conventional, we try to measure the impact of reducing import protection or export taxation on national income. The problem of evaluating economy-wide policy reforms in the context of environmental externalities is indeed widespread. First-best instruments to remove the externalities are not always available and the transaction costs associated with them can be extremely high. Setting first-best taxes is in many cases politically difficult and, moreover, many governments, particularly in poor developing countries, do not have the administrative and legal capabilities to enforce first-best regulations. This leaves second- or even third-best approaches as the only options to mitigating the negative effects of the externalities. The environmental impact of economy-wide policy reforms may, in this context, considerably affect the quantitative and even qualitative effect of these reforms on national income and welfare. The evaluation of these reforms, therefore, needs to explicitly incorporate the environmental as well as non-environmental channels by which such reforms affect national income. The problem is particularly important because, as illustrated by the ensuing analysis, the environmental and conventional effects of trade policies may have conflicting impacts on national income. To the best of our knowledge no previous empirical analyses evaluate the environmental trade-offs of economy-wide policies in a general equilibrium framework. The remainder of this paper is organized as follow: In Section 2 we provide a brief review of the evolution of the Ghanian economy as well as a discussion of the major current trade policy issues. Section 3 presents a discussion of the trade-offs between expanding area cultivated and agricultural productivity. A simple theoretical model of the private and collective land allocation decisions is presented in Section 4. Section 5 reports estimates of the production function

20

R. L6pez / Journal of Development Economics 53 (1997) 17-39

while Section 6 presents the testing of the hypothesis that individual farmers' decisions are socially efficient. Section 7 reports estimates of a land cultivation equation that is necessary for the general equilibrium simulations of the effect trade policy that are presented in Section 8. Section 9 concludes the paper.

2. Policy issues in Ghana Since 1983 the government of Ghana has introduced structural reforms to stabilize and liberalize the economy. The real exchange rate has been adjusted to levels that are consistent with external equilibrium, the public sector has been downsized, protection to the industrial sector has been substantially decreased, prices of agricultural goods, particularly exportables (cocoa), have been brought to levels much closer to international prices. The above reforms have had a positive impact on economic performance. Despite the impressive depth of the reforms, there are still several remaining policy issues: (1) Although quantitative import restrictions have been completely eliminated, the industrial sector is still protected via tariffs. The average import tariff rate including special taxes is about 20% ad-valorem with considerable variations across goods. (2) Although the industrial sector is the most important import substitution sector, a few agricultural goods, mostly cereals, are also protected via import duties that reach about 18%. (3) The implicit export tax on cocoa is still substantial, about 24% ad-valorem (this tax has been reduced from almost 50% in 1985). Each one of the above policy issues implies a potential need for further policy reforms. These additional reforms include further decreases in protection to the import substitution sector and further cuts on the implicit export tax to agriculture.

3. Biomass as a factor of agricultural production In the context of the traditional agricultural systems prevailing in western Ghana, the predominant cultivation practice is shifting cultivation. This involves long cycles over which land is cultivated for 1 or 2 years and left idle for 4 - 1 0 years to restore its productive capacity. The fallow or rest period allows the natural vegetation to reemerge generating biomass that is used as natural fertilizer in the next cultivation period. Also, the root system of the natural vegetation provides physical support to the inherently unstable tropical soils. At the same time the patches of bush spreading in the land help protect the cultivated areas from erosion, flooding and other detrimental factors. This provides a rationale to postulate that biomass constitutes an important factor of agricultural production and that its depletion, via a shortening of fallow periods and more extensive land cultivation, is likely to have a negative effect on productivity.

R. Ldpez / Journal of Development Economics 53 (1997) 17-39

21

Another important feature of traditional systems is that land property rights are not, in general, well defined. A large proportion of the land available in a village is reserved exclusively for use by the villagers. In general, the land is owned by the community (village or lineage that typically encompasses hundreds or even thousands of members) rather than by individuals but its use by outsiders is normally restricted. That is, this is a classic case of common property with closed or highly restricted access for external potential users. The community control of the resources is consistent with and facilitates the shifting cultivation practices. The individual has exclusive rights on the land that he/she actually cultivates but, once the land is left idle in fallow, the land can be reallocated. Common property in these conditions does not necessarily imply overexploitation of the resource. It is possible that the community exerts certain controls on the allocation of the land to allow for its optimal exploitation. The controls may be inadequate or insufficient, in which case the community as a whole may experience income losses associated with excessive loss of biomass, erosion, and flooding. Expanding the cultivated land occurs, of course, at the expense of forest lands a n d / o r of reducing the length of the fallow periods and, hence, of a reduction of the natural vegetation. Thus, the trade-off between cultivated land and forest/fallow is clear: An increase in land under cultivation has a direct output-increasing effect at the cost of reducing the natural capital and, hence, agricultural productivity. There is an optimal fraction of the land that should be cultivated in order to maximize the social income. If the level of land cultivated is above or below the optimum, income is reduced. In the absence of communal controls, individual cultivators are likely to overexploit the natural resource by cultivating too much. In deciding how much land to cultivate farmers in this case are likely to consider only the private costs, ignoring contemporary and intertemporal effects on other cultivators. If communal controls are adequate, individuals would behave as if they fully accounted for both the private and social costs of clearing land. One question the empirical model tries to elucidate is whether or not this happens.

4. The model

We assume the existence of a well-behaved production function relating agricultural output of an individual farmer to the village biomass and conventional inputs used by the farmer, Qi

= Fi( Li,xi,Ki, O)

where Qg is output of farmer i in a given village, L i is the labor input used by farmer i, x i is the level of cultivated land that farmer i uses, K i are fixed factors and 0 is the stock of biomass in the village where farmer i is located. The

R. L6pez/ Journal of Development Economics 53 (1997) 17-39

22

function F i ( . ) is assumed to be increasing and strictly concave in all factors of production. Note that output produced by farmer i is specified as a function of the conventional inputs used by farmer i and as a function of village level biomass. This reflects the fact that the effect of biomass on output is really external to the farm. The land cultivated-biomass trade-off is apparent by realizing that biomass and land cultivated are related. We now turn to this issue. Total biomass in a given village can be expressed as total fallow area times the average biomass density per acre of fallow land,

O=n.

yc-

xi

(1)

i=

where 7/is average biomass density per acre of fallow land, ~ is total land area in a village, and N is the total number of farmers in the village. Thus ~: - ~,x i is the village area under fallow or forest. The average density 7/ is not independent of the area cultivated either. Following L6pez (1993) we can postulate a simple specification for the dynamics of 7, +/t = & - t---~- ) ~/,

(2)

where ~ > 0 represents the natural growth of vegetation and Exi/Yc is really the rate of depreciation of biomass in the context of a rotational system. In the long run when the level of biomass becomes stable we have//, --- 0. In this case, n =

(3)

That is, the level of biomass density in the long run is inversely related to the proportion of the total land area that is cultivated. Using Eq. (3) in Eq. (1) we obtain,

0=

1

(13

That is, the total biomass volume is decreasing with the proportion of cultivated land (Y_,xi/Yc) and increasing with the rate of growth of natural vegetation, ~. From Eq. (1') is clear that the elasticity of total biomass with respect to area under cultivation can be defined by

,n0

din x

(

l

1-

~xi/2

)

(4)

That is, if for example, 25% of the village land is cultivated (~,xi/Yc = 0.25), increasing cultivated land by 1% will decrease total biomass by 1.33%, or, equivalently, the elasticity is - 1.33.

R. L6pez / Journal of Development Economics 53 (1997) 17-39

23

Thus if farmer i increases the area that she/he cultivates, the total village biomass will decrease because the total area under fallow is reduced and because the growth of biomass will decline in the long run. This, in turn, would cause a reduction of agricultural productivity in both the short run and long run. Thus, increasing the area cultivated may cause two negative externalities: (a) the instantaneous area effect on biomass, and (b) the long-run effect through changes in 7/. Next, we consider an optimization model pertaining to a village where the full social cost of changing the area cultivated is imposed on individual cultivators by the village group, so that land allocation decisions maximize the village income. In reality, land allocation may not be optimal in this sense due to the lack of internalization of all the costs of decreasing biomass. The idea is to use this model as a benchmark to allow for generalizations that may permit us to empirically test the model. Consider a village that is a price taker in the output and input markets. Villagers also have an opportunity cost for their time (the wage rate) that is exogenous to the village. Maximization of the aggregate village wealth would then require, max xi'Li

pFi( Li,xi,Ki,O

) - wL i - cx i

i

s.t. 0 - - 7 /

~-j=

)

e-rtdt

xj

(5)

.~xj ~/= y - --w-~7 X

n(0)

=

where p is output price, w is the wage rate, c is the private cost of clearing land per acre, r is the discount rate, ~0 is the initial level of average biomass density per acre in the fallow land, and N is the total number of farmers that share the common lands of the village. As is well known, maximization of Eq. (5) is equivalent to maximizing the Hamiltonian or current income function at each point in time. Thus, the first order conditions are derived by max yA

pF i Li,xi,Ki,'r I x-

=

xj

_ wL i _ cx i

i

where IX is the current value co-state variable measuring the shadow value of "OThus, yA corresponds to the total true income of the village, which explicitly

R. L6pez / Journal of Development Economics 53 (1997) 17-39

24

accounts for the value of investment or disinvestment in the natural resource. This latter value is accounted for by / z [ y - ( X x y ) / ( ~ ) * / ] , while the first right-hand term is the current net income. The first-order conditions assuming an interior solution are the following: byA

--=pF~(')-w=0,

i=l ..... N

(7a)

aL i Oy A

0xi ~ =

=pF~(')-c-rl]~pFJ(.)-Ixrl/Yc=O, j

r+

_

x

~-

(x-

.~xj)}2pF4J(.)

:~xj

"i7= Y - - " : - 7 /

j

i=1 ..... N

(7b)

(7c) (7d)

x

where the subscripts of the F i ( . ) function denote partial derivatives with respect to the corresponding term. Condition Eq. (7a) is the usual profit maximization condition from which the demand for labor is derived. Condition Eq. (7b) indicates that farmer i should adjust her/his level of land under cultivation until the marginal effect of x i on the total current income yA is zero. This occurs at the point where the additional revenue of an acre of land cultivated is equal to the private cost of incorporating fallow land into production, c (the land clearing cost), plus the instantaneous loss of revenue for all producers in the village caused by the reduction of the fallow area (rlF,jpF~), and plus the future income losses for all farmers due to the reduction of r I that increasing x i causes as the effect of a shortened fallow cycle is manifested through time ( ~ r l / 2 ) . Condition Eq. (7c) is the well-known no-arbitrage condition indicating that the marginal social return of ~ should be equal to its marginal social cost. The total marginal social return is equal to the contribution of 77 to the revenue o f all farmers in the village plus the capital gains (or losses) associated with changes in /x. The marginal cost of 7/ is equal to the opportunity cost, rl.~, plus the value of the stock that is depleted, (~xj/2)l~. Finally condition Eq. (7d) is just a reinstatement of the control constraint. If a steady-state exists we have ~ = i7 = 0. Thus, the steady-state value of /.~ and rl are, N

/x* = ~(1 -- Z) Y'~ pF~(" ) / r + z

(8a)

j=l

71" = T/z

(8b)

R. L6pez / Journal of Development Economics 53 (1997) 17-39

25

where z =- F,j xj/Yc is the proportion of the total village land that is cultivated. Also, using Eq. (8a) in Eq. (7b) and rearranging we have, OY g

l+r N = pF~(" ) - c - 7 1 7 - ~ z j ~ p F 4 J ( • ) = O,

i = 1..... N

(9)

The first order conditions Eqs. (7a), (7b), (7c), (7d) and (7e) correspond to the case when the village is able to exert perfect controls on the use of the common property land resources by individual farmers. That is, it implies that land and labor are efficiently allocated. If social controls over individual farmers are non-existent or not perfect, the land allocation decisions will not, in general, follow the rule indicated by Eq. (7b) or, in the steady-state, by Eq. (9). Since the land clearing costs c are entirely private, a farmer is likely to fully consider this cost in her/his land cultivation decision but, unless h e / s h e is forced by regulations or taxation, h e / s h e will not take into account the full extent of the marginal costs to the rest of the village farmers. That is, in Eq. (9) for example farmer i may only consider a fraction of the cost r/(1 + r ) / ( r + Z)~tf=lpF4J(.). In the extreme case where there is no social regulation (or taxation) whatsoever, an individual farmer would only consider approximately 1 / N t h of this latter cost. Thus, in general we have oYA/Ox~ < O. If oYA/Ox~ < 0 for at least some i = 1, . . . . N, the social controls are imperfect and thus the community's income is less than its maximum due to excessive land cultivation. If o Y A / O x i = 0 for all i = 1, . . . . N, the community's income is maximized. As shown below, this observation forms the basis for a statistical test of the hypothesis of socially efficient land cultivation decisions and to estimate the approximate income loss of the potential inefficiency at least under the steady-state assumption. Expressing Eq. (9) in logarithmic form and using Eqs. (3) and (1') we obtain, 01nyA Oln x i

pFi(')

c}lnFi

~i

aln x i

yg

z 1

l+r

OlnFi 1

z r + z 01n0

'

i

1. . . . . N

J

(10) where E i = ( c x i ) / ( p F i ) , is the share of land clearing costs in the total value of output of farmer i. If all farmers have access to the same production function we have Fi(. ) = F and yA = N[ pF(" ) - wL - cx] in the steady-state (since T (Y.,xj)/(Yc) T1 = 0 in the steady-state). Thus, Eq. (10) can now be written as 01nY g 01nxi

1

[01nF

N(1--SL--e ) [~nx

i=l ..... N

Z

l+r

01nF]

e-- 1- z r+z-~nO]' (ll)

where s L is the share of labor in the total value of output. We are interested in the effect of an increase in the total land cultivated at the village level on total village

R. Ltpez / Journal of Development Economics 53 (1997) 17-39

26

income. This implies that we are interested in the effect on income of increasing land cultivated by all farmers by the same proportion. Thus, blnY A

blnY A

- - = N - - -

blnx

blnx i

1 l-sL-e

[blnF blnx

z e-

1

-

l+r01no]

z r+z

bln

(12)

Thus, the test of socially efficient land cultivation decisions consists in testing whether or not bln Y a / b l n x in Eq. (12) is equal to zero. Note that in order to test b In Y a / b In x = 0 all that we really need are the estimates of the production function, estimates of the share of land clearing costs in the total value of production and the observed z values. Also, note that if condition Eq. (7a) holds, i.e. if labor is optimally allocated for a given level of 0 and x, s L = b ln F / b i n L. Thus, from estimating a production function that explicitly considers 0 as a factor of production along with the other conventional factors we can obtain an estimate for Oln y A / o In x, and using the estimated variance-covariance matrix of the coefficients of the production function, we can also estimate the variance of expression Eq. (12). This allows us to statistically test whether b In Y a / b In x = 0 and, if b In y a / b In x < 0, to provide a point estimate with a confidence interval at the chosen level of significance.

5. P r o d u c t i o n f u n c t i o n e s t i m a t i o n

We specify a Cobb-Douglas functional form for the production function, In Qijt = fib + fix In x,7t + / 3 L In Liit + ~o In 0jz + [3k In Kij t + Intercept Dummies + e i j t, where Qijt is output of farmer i in village j at time t, xij t is land cultivated by farmer i, L i t is labor, 0jr is biomass in village j at time t, Kij t is farm capital (mostly toolsl and ~i~t is the disturbance term. We included several intercept dummies to control for productivity changes through time, village differences in productivity, farm size and whether or not the farmer produces cocoa or other tree crops. We also included in certain specifications the use of other purchased inputs, mostly fertilizers, but it tumed out that very few producers used any purchased input at all and those that did used very small amounts. Possibly for this reason, we could not detect any effect of purchased inputs on production. We also use certain other demographic variables obtaining very little significance. More on this below. Since individual farmers can move around within the village land areas in search of land to cultivate, the relevant biomass variable needs to be defined at the village level as well. Thus biomass is defined as the total village area under fallow times the average vegetation density per acre of fallow plus the forest areas times a density index. These data are obtained from remote sensing based on satellite observations for each village area in each year, 1988 and 1989 (see L6pez, 1993 for details on these data). We have data on 12 villages in western Ghana. These

R. L6pez / Journal of Development Economics 53 (1997) 17-39

27

Table 1 Production function estimation, Ghana 1988-1989 (1) OLS without village dummies

(2) OLS using fallow area as a proxy for biomass

(3) Controlling for village effects

Constant Land cultivated Total labor Village biomass Tools Dummy cocoa producers Dummy 89 Dummy size A1 Dummy size B 2 Village dummies (several) N R2

9.44 (13.30) 0.27 (3.16) 0.25 (2.47) 0.15 (1.86) 0.28 (3.13) 0.62 (4.37) -0.57 (-3.14) - 0.25 ( - 1.23) - 0 . 4 6 ( - 1.50) Not included 139 0.46

9.08 (13.6) 0.20 (2.49) 0.31 (3.01) 0.19 (2.48) 0.26 (3.15) 0.55 (4.02) - 0 . 3 0 ( - 1.71) - 0.40 ( - 2.02) - 0 . 8 6 ( - 2.79) Not included 139 0.42

9.51 (13.6) 0.26 (3.01) 0.24 (2.48) 0.18 (2.34) 0.27 (3.03) 0.60 (4.28) -0.61 (-3.30) - 0.51 ( - 1.66) - 0 . 2 5 ( - 1.32) All included 139 0.48

~2 F

0.43 9.60

0.38 10.92

0.42 10.32

Notes: a: t-statistics are in parentheses, b: the dependent variable is the log of real agricultural output per household, i Dummy size A is equal to one for extremely small farm size (less than 5 acres) and zero otherwise. 2 Dummy size B is equal to one for extremely large farm size (greater than 100 acres) and zero otherwise.

data were matched with Living Standard Survey (LSS) data that provide information on 16 farm-households sampled in each village and each year. The household data that we use are output value, family and hired labor used by the farm-household, land cultivated by each farmer, farm tools and labor used, use of fertilizers or other purchased inputs, and some demographic variables, such as education, age, family size, and ethnic background. In estimating the land allocation decisions we also use data on wages deflated by output prices per village. 4 Table 1 provides various estimates of a double-log production function for the 12 villages in Ghana. Not all household data turned out to be consistent and some household data was simply non-existent. For this reason we could only use data for 139 households over the 12 villages and 2 years. For the environmental variable we used two variables, one that accounts only for the proportion of fallow land and forest in the village area and another one that considers biomass volume properly defined as area under fallow times the average biomass density per acre of fallow plus forest biomass. The use of fallow land as a factor of production is appropriate under the assumption of steady state, i.e. when the stock of biomass is constant and, therefore, the fertility of the land actually cultivated is determined by the proportion of land under fallow. Outside the steady state the relationship between cultivated land fertility and the proportion of fallow land is less direct and

4 For a description of the data see Appendix A.

28

R. L6pez / Journal of Development Economics 53 (1997) 17-39

is also affected by the rate of biomass accumulation or disaccumulation. L6pez (1993), however, shows that even outside the steady state, cultivated land fertility is still affected by the proportion of fallow land. Besides the fertility effect, the level of biomass in the areas surrounding the cultivated land patches help protect against flooding and other physical damages. This positive effect on productivity is not dependent on whether or not a steady state prevails. The estimates of the production function appear highly plausible. Moreover, the goodness-of-fit is very good and the degree of stability of the coefficients to changes in the specification is quite remarkable. The specifications in Columns (2) and (3) passed the Breush-Pagan test for heteroskedasticity, which used family size, land cultivated and total labor as possible sources of heteroskedasticity. The estimates of the effect of the biomass variable are statistically significant and are comparable in magnitude with those associated with conventional factors. We also used other variables such as fertilizers used and family education but these variables turned out to be extremely insignificant. The fact that only a small proportion of all farmers use fertilizer and usually in small amounts, may explain the lack of significance of this variable. Also, education is quite homogeneously low among most farmers and thus the lack of explanatory power of this variable. The estimates suggest that farmers that produce cocoa tend to be more productive than those that do not and that productivity declined in 1989 vis-a-vis 1988.5 The coefficient of the biomass variable is significant at least at 10% in all regressions performed. Three of these regressions are shown in Table 1. The lowest value and level of significance of the biomass coefficient is obtained when the estimation does not control for village effects (Column (1)) where the coefficient is significant at 7% only. In all other specifications its level of significance is better than 5%. In general, the value of the coefficient tends to be sizable but somewhat lower than the values of the coefficients of the conventional factors of production. The coefficients of the environmental variable appear to indicate that the environmental factor is an important determinant of agricultural productivity. One possible interpretation, however, is that a village may have more biomass than another because it has better soil quality and climate that allows the natural vegetation to grow faster. This better climatic/soil characteristics would also lead to more productive agriculture in this village and thus the positive correlation found between output and biomass would be spurious. Two approaches were followed to check the relevance of this interpretation: (a) we used area under

5 Additionally, we used two dummy variables(Dummy size A and B) to control for the possibility of technological differences associated with either extremely small or extremelylarge farm sizes (i.e. for farm sizes outside the mean + 3 standard deviationsrange). Further, we also use a dummy variable (Dummy 89) to control for possible changes in productivitydue to weather or other reasons between 1988 and 1989.

R. L6pez/ Journal of Deoelopment Economics 53 (1997) 17-39

29

fallow per village as a proxy for biomass volume; (b) we control for village fixed effects such as climate and soil quality using village dummy variables. 6 Column (2) of Table 1 reports the estimates using fallow area as a proxy for biomass. The village fallow area is less likely to be correlated with soil quality and climatic conditions than total biomass. However, this procedure has the shortcoming of ignoring intervillage differences of biomass density in the fallow areas, which is likely to be an important factor affecting the fertility levels of the soils. Approach (b) is probably more satisfactory. Using village specific dummies should control for unobserved variables such as climate, soil quality and other village specific effects, thus eliminating the influence of such variables from the coefficients associated with the other explanatory variables. The use of village specific dummies (rather than household dummies) is an adequate procedure because the biomass variable used is also defined at the village level. Column (3) provides the estimates obtained using village effects. We will use these coefficients in the various experiments reported below. The estimators of the biomass coefficients remained large and significant. Thus, the relationship between biomass and agricultural output does not appear to be spurious. Apart from the statistical significance of the coefficients associated with biomass it is important to emphasize their quantitative levels. The biomass coefficient estimated using village effects was 0.18 (Table 1, Column (3)). This means that the contribution of biomass to agricultural production is about 18%. Or equivalently, that the share of biomass in total agricultural production once this factor is measured at its shadow value is 18%. This compares with a contribution of 26% for land cultivated, 24% for labor, and 27% for capital.

6. Land allocation and community income Using the estimates of the production function it is possible to measure Eq. (12) and thus test the hypothesis that land is socially optimally allocated (i.e. that a ln YA/aln x = 0). To do this we also must assume a discount rate, to use observed values of the rate of village land cultivation z, and to use information regarding the share of land clearing costs in total farm output value. It is important to note that with a Cobb-Douglas production function the coefficients aln F / a l n 0, a ln F / a l n x, and a ln F / a l n L become constants. Thus, the key variable that may allow for a In yA//o In x to vanish is the village level cultivation rate z. (z now becomes an endogenous variable.) This, of course, is consistent with the assumption that all farmers have identical production functions in which

6 Procedure (a) is consistent with a long-run interpretation because biomass density depends of the proportion of the land under fallow in steady state (see L6pez, 1993).

30

R. Lrpez / Journal of Development Economics 53 (1997) 17-39

case there is a representative farmer that has a cultivation rate equal to the aggregate village level cultivation rate. According to information obtained in the field, it was estimated that the share of land clearing costs in revenues fluctuate between 12% and 17%. We use the lower bound estimate and thus we assume that E = 0.12. We assume a 10% discount rate as a benchmark. Of course the lower the land clearing costs and the higher the discount rate assumed the less likely it is to reject the hypothesis that the observed land allocation ratio at the village level (z) is optimal, or equivalently, that ~ In yA/o In x = 0. Since we do not want to stack the odds against this hypothesis we assume a relatively large r and our most conservative estimate of From the production function estimates of Column (3) in Table 1 we have fix = 01n F / 0 1 n x = 0.26 and to = 3In F/31n 0 = 0.18. Additionally, using the assumption that labor is allocated to maximize profits (see Eq. (7a)) we have the share of labor equal to the estimated labor coefficient in the production function flL = sL = 0.24. Using these data we can obtain a point estimate for Eq. (12). In order to estimate the standard error of the estimated value of Eq. (12) we use the estimated variance-covariance matrix of the coefficients fix, ilL, and /30. Since Eq. (12) is a non-linear function of these coefficients we use the Delta method to estimate the variance of the expression for 0 In YA/oIn x. The Delta method provides the asymptotic distribution of a continuous function of random variables. 7 Using the above information we can thus obtain an estimate for expression Eq. (12) as well as for its standard error. Evaluating Eq. (12) at the mean level of z for the whole sample (which is 0.31) we obtain (~ In y g ) / ( O In x ) = --0.125 (0.06), where the number in brackets is the standard error of the estimate obtained using the Delta method. Thus, the t-statistic is - 2 . 0 8 indicating that 3 In Y/3 In x is negative and significant at 5% level of significance. That is, the hypothesis that land cultivation decisions are socially optimal is statistically rejected. Moreover, expanding land under cultivation is likely to cause a large reduction of the permanent or long-run income of the communities. That is, the short-run cash flow of increasing land cultivated falls short of compensating for the current losses and the present value of future losses associated with the loss in biomass that increasing land cultivation implies. Of course the above estimates are sensitive to the discount rate assumed. If the discount rate is lower than 10% the income elasticity with respect to land cultivation will be even more negative. If, however, a higher discount rate is allowed the elasticity becomes less negative. If the assumed annual discount rate is 15% instead of 10%, then ~ In YA/o In x = --0.094 (0.046), which is also significant with a t-statistic value of 2.04. The estimate becomes non-significantly

7 See Greene (1993, p. 297) for a description of the Delta method.

31

R. L6pez / Journal of Development Economics 53 (1997) 17-39

Table 2 Income elasticity of land cultivated Annual discount rate assumed 10% 15% 20%

Land cleating cost shares 12%

15%

17%

-0.125 -0.094 - 0.047

-0.171 -0.143 - 0.094

-0.203 -0.174 - 0.122

different from zero only at an annual discount rate of approximately 50%, which is clearly a very unrealistic rate. Using the estimates of the production function one can also calculate the optimal rate of land cultivation given a discount rate. That is, the rate of cultivation z that would be required so that aln YA/a In x = 0 in Eq. (12). Using a 10% discount rate we obtain an optimal rate of land cultivation of 22% compared with the average observed one of 31%. That is, the current area cultivated would need to be reduced by almost 30% in order to achieve a social optimum with full internalization of all externalities, or equivalently, the extent of overcultivation of the land is about 41%. If the assumed discount rate were 15% instead of 10%, the optimal rate of land cultivation should be 23.6%, still substantially below the observed rates of even the villages that cultivate the least amount of land (26.7%). Using our benchmark assumptions ( r = 0.10 and e = 0.12) we thus find that land is about 41% overcultivated and that the income losses associated with this are of the order of 4.0%. That is, village communities could increase their permanent income by about 4% if they could bring the area cultivated down to the socially optimal levels. Or, equivalently, that the costs of not fully internalizing the biomass externalities is of the order of 4% of the village income. Table 2 provides some simulations of the estimated income elasticity of cultivated land using various combinations of land clearing cost shares and interest rates.

7. Land cultivation decisions In view of the rejection of the hypothesis that producers choose land cultivation levels optimally, we cannot use Eq. (7b) (which assumes optimal choice) to derive a specification for a land cultivation equation. Instead we use a modification of Eq. (7b) or of its steady-state counterpart, Eq. (9), that allows for only a partial internalization of the external costs of cultivating land. Thus, we can generalize Eq. (9) in the following way, aY

l+r

N

=pF~ - c - Arl-~zEpF~(') ax i

= O,

i= 1..... N

(9')

32

R. Ltpez / Journal of Development Economics 53 (1997) 17-39

where A is a parameter that ranges between 1 and 1 / N . If A = 1 we have complete internalization of all the biomass costs of expanding the area cultivated, and if A = 1 / N farmers will only consider the effects of biomass reduction on their own outputs. The value of A can have intermediate values if there exists a degree of village control on individuals that force them to take into account some of the externalities. What the estimates of the production function show is that the efficiency parameter A < 1 but that its value is not necessarily equal to 1 / N . Thus, assuming that all farmers have identical production functions, N Ej= 1pFaJ( • ) = NpF~(L i, x i, K i, O) and ~,jxj = Nxi, we use Eq. (9'), the optimal labor allocation Eqs. (7a), (3) and (1') to solve for L i = L ( w / p , c / p , Ki; A) and x i = x ( w / p , c / p , Ki; A). If we assume that land clearing costs consist mostly of labor, the unit land clearing cost can be written as c = T w and thus in this case Li = L ( w / p , K~; A) and x~ = x ( w / p , Ki; A). The parameter A itself is likely to vary across villages according to the institutional setting of the village, population pressure, the degree of ethnical heterogeneity of the population, and the availability of fallow lands. Thus, the cultivated land equation considers real wages, farmers' capital, family size (to represent population pressure on the land), village fallow area and a dummy variable for the ethnic background to capture whether or not the household is immigrant or established (typically the Ashanti and Bron groups have been established in the area for a long time while the rest are mostly new immigrants). Table 3 provides the estimates of the land cultivation decisions by the individual farmers. Column (1) provides estimates conditional on a given level of family size while the estimators in Column (2) imply that family size is endogenous. In this sense, the estimates in Column (2) can be interpreted as more long run than those in Column (1). Both specifications passed the Breush-Pagan test for Table 3 Land equation, Western Ghana 1988-1989 (double-log specification)

(1)

(2)

Constant Family members Wage/agricultural output price Tools % of fallow land over total land Dummy for cocoa producer Dummy for Ashanti-Bron ethnic groups Dummy 89 N R2

7.10 2.66 0.37 2.33 - 1.01 ( - 2 . 1 9 ) 0.27 (1.94) - 0 . 1 9 ( - 0.91) 0.89 (4.37) - 0.57 ( - 2.44) 1.49 (5.77) 139 0.54

8.85 (4.40) - 1.26 ( - 3 . 1 3 ) 0.42 (3.32) 0.86 (4.34) - 0.69 ( - 2.97) 1.44 (6.57) 139 0.51

~2 F

0.51 21.67

0.49 27.94

Note: 1, t-statistics are in parentheses. 2, the dependent variable is the log of the area cultivated by each household.

R. Ltpez / Journal of Development Economics 53 (1997) 17-39

33

heteroskedasticity that used family size and total value of output as potential sources of heteroskedasticity. The key variable here is the agricultural wage deflated by the output price. This variable is defined per village for each year. Its coefficient suggests that the wage rate deflated by the output price exerts a negative effect on land cultivated. The elasticity for the wage rate is about minus one for the short-run estimates (Column (10) and about - 1 . 2 6 for the long-run estimates (Column (2)). Thus, decreasing w / p , which can occur if agricultural prices increase or the wage rate falls cause an expansion in the cultivated area and, therefore, a reduction in the fallow area and a fall in biomass, which, in turn, reduces agricultural productivity. According to the estimates in Table 3 the elasticity of land cultivated with respect to agricultural output price is in the 1-1.26 range. The other explanatory variables have all the expected signs and are significant. The family size has a large positive effect on the amount of land cultivated. This finding is very consistent with the idea that population pressure is an important factor behind environmental degradation. Capital equipment, represented by the number of tools and implements owned by a household, also exerts a positive effect on the area that families cultivate. Also it appears that, other conditions equal, established local residents (Ashanti and Bron tribes) tend to cultivate less land than relatively new immigrants (the rest of the population).

8. Trade liberalization effects In this section we consider the implications of trade reform in the presence of environmental distortions as well as trade distortions. The two-sector general equilibrium model used is extremely simple and thus we present the ensuing analysis mostly as an illustration of the potential consequences of economy-wide reforms that are implemented without consideration of environmental externalities. Despite deep trade reforms over the eighties Ghana still protects the industrial sector and taxes the agricultural sector, in particular the agricultural export sector. It is estimated that the implicit trade tax to the agricultural sector is of the order of 20%. This feature is interesting because in most sub-Saharan African countries agriculture is negatively protected although at even higher rates than in Ghana. Trade liberalization then implies a reduction of the implicit trade tax of agriculture. The question that we intend to elucidate here is whether or not such a policy increases national income when the environmental externalities are explicitly accounted for. Trade liberalization in this context has two effects on national income: (1) It diminishes the inefficiencies in resource allocation caused by the initial trade distortions. (2) It may magnify the environmental distortions by inducing an expansion of the land cultivated. Effect 1 is positive for national income while Effect 2 is detrimental. The net effect is ambiguous. We now turn to a conceptual and empirical evaluation of these effects.

R. Lgpez / Journal of Development Economics 53 (1997) 17-39

34

To specify a general equilibrium model we need to first specify the land clearing cost (c) in terms of real factor use. We assume that land clearing requires labor in fixed proportion. We specify that y days of labor are necessary to clear 1 acre of land. In this case c = y w and the agricultural production function needs to be modified slightly to account for this, F ( L - y x , x, K, 0). This specification accounts directly for the cost of clearing land. An increase in x has a direct positive effect on output and two indirect negative effects, decreasing 0 and reducing the amount of labor that is effectively devoted to production. National income for a small economy evaluated at the opportunity cost of resources needs to be specified in terms of world prices rather than at domestic prices. Thus national income can be defined as, Y* = p* F ( L - y x , x , K , O ) + txil + q* G( L N ; K s ) (13) where p * and q * are the international prices of agriculture and the urban goods, respectively, G(.) is the production function of the urban goods, Ls is labor, and K s are other factors of production used in production of the urban good. The other components of the model are the market clearing conditions for labor ( 1 - ~')p*Fl(.) = w = ( 1 + t)q* G I ( . ) (14a) L + L s = L, (14b) where ~- is the export tax to agriculture, t is the tariff to industrial imports, and is the total labor force. Thus Eq. (14a) reflects the competition for labor between agriculture and industry and Eq. (14b) the full employment condition. Consider now trade liberalization implemented by a reduction of the export tax affecting agriculture that causes an increase in the domestic price of agriculture, p - ( 1 - ~')p*. (Given the Lerner's symmetry condition this is equivalent to a reduction in the import tariff that reduces the domestic price of the industrial good by the same proportion.) Totally differentiating Eq. (13) using Eqs. (14a) and (14b) we obtain dY * [ 1+ r ] dx dL d-p- = P * [ F 2 - - TF1 - ~ ' rl - + - UzF 4 1 ]--~p + ~p* Vl(") -d-p (15)

We can relate the effect of trade liberalization to the elasticities obtained in the previous section by expressing Eq. (15) in logarithmic form, 01n F din L dlnY* S A /[OlnF_ e-z l + r O l n F ] d l n x dlnp

~[0~nx

1

f

z r + z -d ~-O J -d-~nnp + 7 01n - L dln p (16)

where: din x

~ln x

dlnp din L -

-

dlnp

+

- -

+

~lnp ~ln L

=

~ln x ~lnw

- -

01np

- -

(16a)

- -

(16b)

Olnw ~ l n p ~ln L ~lnw ~lnw b l n p

R. Lrpez/ Journalof DevelopmentEconomics53 (1997) 17-39

35

and s A is the share of agriculture income in total income. F r o m the labor market cleating condition L(w, p ) + LN(w, q ) = L , we can derive an expression for In w/O In p by totally differentiating it with respect to w and p, In w ln-----p-

~ In L/~ In p Lu ~ In L u - - + L Olnw

a In L

(17)

Olnw

which is positive if labor is a normal input in agriculture (~ In L/O In p > 0). W e note that as long as both x and L are normal inputs, the total or net effects o f p on x and L ( d l n x / d l n p and d l n L / d l n p ) are both positive. This can be seen by using Eq. (17) in Eqs. (16a) and (16b) and noting that Oln LN/OW < O, and ~In L/~ In w < 0. Thus, if the environmental externality is fully internalized by individual producers the expression in square brackets on the right-hand side of Eq. (16) vanishes (see Eq. (12)) and thus trade liberalization will unambiguously increase income as long as r > 0 initially. If, however, land cultivation decisions are socially inefficient the term in square brackets is negative and the net effect of trade liberalization is, in general, ambiguous. Using the estimates of the agricultural production function, the land cultivated elasticity with respect to output price as well as measures of the share of agriculture in national income, and total employment and the labor demand elasticity in the non-agricultural sector, we can evaluate expression Eq. (16). Table 4 provides the values of the parameters used to estimate expression Eq. (16). To estimate the standard error of this estimate we used the Delta method. Using the parameters in Table 4 we obtain a point estimate for the effect of trade liberalization on national income. The estimated value for Eq. (16) is - 0 . 0 0 4 5 with a standard error of 0.0016. That is, a further improvement in real agricultural prices that trade liberalization induces is slightly detrimental for national income. The income losses associated with the magnification of the environmental distortion are greater than the gains related to the decrease of the conventional price distortions.

Table 4 Basic parameters and elasticies used in evaluating the effect of trade liberalization S A = 0.49 OlnF/01nx = 0.26 T = 0.12 z = 0.31 L N/L = 0.33

r = 0.10 ~lnF/01n0 = 0.18 Olnx/Olnp = 1.01 Olnx/Olnw = 1.01 OlnLh/alnw = -0.40

"r = 0.17 alnF/OlnL = 0.24 alnL/31np = 1.26 OlnL/Olnw = - 1.26

Note: the parameters L N/L, S n, T are obtained from World Bank sources. T is the ad-valorem implicit net tax rate for agriculture. The estimate for ~lnLn/~lnw is obtained from estimates obtained by Riveros for other sub-Saharan African countries.

36

R. L6pez/ Journal of Development Economics 53 (1997) 17-39

We have also estimated the effects of a complete trade liberalization on national income. The net effect of such a policy is to decrease national income by about 0.10%. As indicated above, the main source of the net losses is the fact that trade liberalization causes a reduction of the biomass with the consequent fall of productivity of the conventional factors in agriculture. It is estimated that a complete trade liberalization is likely to induce a fall in biomass of the order of 2.9% over the long run. Several other simulations were performed by considering alternative values for the parameter a ln Llv/aln w, which is the only elasticity that is not directly estimated in the regression analysis. It turned out that both the effect of trade liberalization on national income or the biomass level are only mildly sensitive to changes in the labor demand elasticity of the non-agricultural sector. As labor demand in the non-agricultural sector becomes less elastic both the income-decreasing environmental effect (the first right-hand-side term in Eq. (16)) and the income-increasing price efficiency effect (the second right-hand-side term in Eq. (16)) of trade liberalization become smaller. The net effect of this is to cause only a moderate reduction on the effect of trade liberalization on income. Under the various simulations executed for a wide range of non-farm labor demand elasticities the net income elasticity fluctuates between - 0 . 0 0 2 and -0.0078. Also the loss of biomass is likely to range between 2.48% and 4.4% for the non-farm labor demand elasticity ranging between - 0.1 and - 0.6.

9. Conclusion The major findings emerging from this paper are: 1. Biomass is an important factor of production in western Ghana. Its contribution to agricultural output ranges from 15% to 20% of the value of output. Given the importance of agriculture in the economy, this means that biomass is also an important determinant of national income. 2. The empirical evidence suggests an overexploitation of biomass through a more than optimal level of land cultivated. Or, equivalently, fallow periods appear to be too short, and the level of deforestation too high. Thus, the stock of the environmental resource is below the socially optimum levels. In this paper we have defined the social value of biomass quite narrowly, considering only its contribution to agricultural output. We have excluded further roles of biomass that transcend the Ghanian economy, such as its contribution to biodiversity and carbon sequestration. 3. The quantitative importance of agricultural prices, rural wages and population pressure as a source of biomass degradation has been clearly demonstrated. 4. The impact of a deepening of trade liberalization on biomass depletion may be quite serious. Further losses of biomass of the order of 2.5-4.4% by completing the trade liberalization process are likely.

R. Lrpez / Journal of Development Economics 53 (1997) 17-39

37

5. The evaluation of a deepening of trade liberalization in Ghana suggests that, in general, their impact on national income is rather small once the existence of the rural environmental distortion is considered. The effects of furthering trade liberalization (decreasing the implicit tax to agriculture or reducing tariff protection) is, in general, ambiguous. The simulations performed indicate that a deepening of trade liberalization is likely to cause a fall in national income. The negative income effect of greater biomass losses is likely to more than off-set the positive income effect of reducing trade distortions. The above implications are obtained using parameters estimated by exploiting cross-sectional and time variability during the 1988-89 years in western Ghana. A key empirical result underlying the policy simulations is that the main source of supply response in agriculture is the expansion of the cultivated area rather than agricultural intensification. If agricultural price responsiveness relied less on land expansion and more on intensification the policy evaluation of trade liberalization would probably be more favorable. Currently, Ghana is gradually improving its agricultural research and extension services, which could, in the future, allow for greater reliance on yield increases. On the other hand, it is plausible to assume that as long as (forest) land is available for cultivation that a significant component of the supply response is likely to continue to be based on agricultural area expansion. To the extent that land is a normal factor of production (and there is no reason to question this), any increase in real output prices is likely to induce a greater demand for land. This, in turn, will be manifested by an expansion of the agricultural frontier as long as more land is available.

Appendix A Part of the data used come from the Living Standards Survey (LSS) conducted in Ghana in 1988 and 1989. This panel data includes a wide range of information such as: labor force activity, individual characteristics, consumption and production at the household level, as well as wages at the village level. The sample used consists of all observations repeated in 1988-1989 for which information on agricultural revenue, land cultivated, hours worked by family members, and hired labor exist. Data on fallow, forest and agricultural land and biomass density for each village were provided by a special study done for this project by the Earth Satellite Corp, Bethesda, MA, based on satellite images that cover the west region of Ghana for the years 1988 and 1989. The total number of villages considered was 12, all in the West Region of Ghana. The villages are New Bansakrom, Yankye, Akantamwa, Tanoso, Sireso, Bibiani, Asunsu, Doduoso, Susuanso, Dormaa, Apronsie, and Boasi. The LSS and the EARTHSAT data were matched to obtain a data set that included the usual individual characteristic variables with information on natural resources at the village level. The total number of households considered is 139.

38

R. Ltpez / Journal of Development Economics 53 (1997) 17-39

A.1. Variables

1. Real Agricultural Output per Household: Total output in Cedis 1988. It is the sum of total sales to the market, payments in kind to hired labor, seeds kept, home consumption and home production. 2. Wage/Agricultural Output Price: Average wage per village earned per day by a male working in the fields, divided by the average price of agricultural output at the village level. 3. Family Members: Number of family members older than 10 years and younger than 60 years old. 4. Total Labor: Total number of days worked in the field by family members or hired labor. 5. Land Cultivated per Household: Land cultivated by each farmer (in acres). 6. Tools: Total number of tools owned by each farmer. 7. % of Fallow over Total Land: LFALLOW/(LFALLOW + LAGRIC), where LFALLOW is total land under fallow in each village and LAGRIC is total land cultivated in each village. Source: EARTHSAT. 8. Village Biomass: LFALLOW × IFALLOW/(LFALLOW + LAGRIC), where IFALLOW is the average biomass density in each village. Source: EARTHSAT.

References African Development Bank (ADB), The Economic Commission for Africa (ECA), Organization of African Unity (OAU), 1984. The African food crisis and the role of African Development Bank in tackling the problem. A paper presented jointly by the staff of the ADB, ECA and OAU to the Symposium on the future of ADB and the Food Crisis of Africa, Tunis, Tunisia, May. Allen, J., 1985. Wood energy and preservation of woodlands in semi- and developing countries, the case of Dodoma Region, Tanzania. Journal of Development Economics 19, 59-84. Dasgupta, P., Maler, K., 1990. The environment and emerging developing issues. In: World Bank Annual Conference on Development, World Bank Occasional Paper, Washington, DC. Deacon, R., 1992. Controlling tropical deforestation: an analysis of alternative policies. Working Paper #1029, Country Economic Department, The World Bank. FAO, 1986. Natural resources and human environment for food and agriculture in Africa. Environment and Energy Paper No. 6, Rome. Feder, G., Onchan, T., Chalamwong, Y., Hongladarom, C., 1988. Land Policies and Farm Productivity in Thailand. The World Bank, Johns Hopkins University Press, Baltimore, MD. Glantz, M., 1977. Dese~fication: Environmental Degradation in and around Add Lands. Westview Press, Boulder, CO. Greene, W.H., 1993. Econometric Analysis. MacMillan Publishing Co., New York. Larson, B., Bromley, D., 1990. Property rights, externalities, and resource degradation. Journal of Development Economics 33, 235-262. L6pez, R., 1993. Resource degradation, community controls and agricultural productivity in tropical areas. Working Paper No. 93-08, Dept. Agricultural and Resource Economics, Working Paper No. 93-08, Dept. Agricultural and Resource Economics, University of Maryland, College Park.

R. L6pez/ Journal of Development Economics 53 (1997) 17-39

39

L6pez, R., Niklitschek, M., 1991. Dual economic growth in poor tropical areas. Journal of Development Economics 36, 189-211. Perrings, C., 1989. An optimal path to extinction? Poverty and resource degradation in the open agrarian economy. Journal of Development Economics, 30, 1-24. Sinn, H., 1988. The Sahel problem. KIKLOS, May, 41.