Why Governments Should Stop Non-Social Subsidies:

                      Measuring Their Consequences for Rural Latin America

                                                        By

                                                 Ramón López

                                 University of Maryland at College Park

              (Tel. 301-405-1281; Fax: 301-314-9091; email: rlopez@arec.umd.edu)


Abstract. The provision of public goods and the amelioration of market failure are the classical
justifications for government intervention in the economy. In reality, (1) governments intervene
in markets that are not affected by failure, and (2) a large share of the government resources is
spent in private goods, not in public goods. In contrast to issue (1), issue (2) has received little
attention in the literature, in spite of the potentially large efficiency and equity losses arising
from misguided allocations of public expenditures. This paper empirically documents the size of
(2) in the rural sector and investigates its consequences for rural development for ten Latin
American countries over the 1985­2000 period. The econometric evidence suggests that the
structure of public expenditures is an important factor of economic development in the rural
sector, much greater than that of the level of public expenditures and of other factors on which
the development literature has traditionally focused. Expanding total public expenditure in rural
areas while maintaining the existing public expenditure composition prevailing in certain
countries does little to promote agricultural income and reduce rural poverty. Spending a
significant share of government resources in (non-social) subsidies causes less agriculture
income, induces an excessive reliance of agriculture on land expansion and reduces the income
of the rural poor.




World Bank Policy Research Working Paper 3609, May 2005

The Policy Research Working Paper Series disseminates the findings of work in progress to encourage the exchange
of ideas about development issues. An objective of the series is to get the findings out quickly, even if the
presentations are less than fully polished. The papers carry the names of the authors and should be cited
accordingly. The findings, interpretations, and conclusions expressed in this paper are entirely those of the authors.
They do not necessarily represent the view of the World Bank, its Executive Directors, or the countries they
represent. Policy Research Working Papers are available online at http://econ.worldbank.org.




Excellent research assistance was provided by Denise Falck, Gregmar I. Galinato, and Carolina Handall. I would
like to thank Daniel Lederman, Mark Wenner, and Bruce Gardner for comments and discussions on the issues of
this paper.

 Why Governments Should Stop Non-Social Subsidies: Measuring Their Consequences for

                                              Rural Latin America

I. Introduction

         The provision of public goods and the related issue of ameliorating the impact of market

failure are the classical justifications for government intervention in the economy. In reality,

governments often do not do a good job in fulfilling these roles: (1) they intervene in markets

that cannot be even remotely considered affected by failure or imperfections, and (2) a large

share of the financial, human, and institutional public resources is spent in private goods (non-

social subsidies), not on public goods.

          Misguided public policies that interfere with essentially efficient markets have been for a

long time a central concern to economists. The literature documenting and measuring the

efficiency costs of market interferences for many commodities and many countries around the
world is indeed massive.1 On the basis of these studies, international organizations as well as

several other economic institutions have forcefully argued for the removal of government

interference with markets for several decades.

         The high degree of attention that the economics literature has given to issue (1) contrasts

with the relatively little concern that issue (2) has generated. This despite the potentially large

efficiency and equity losses arising from misguided allocations of public expenditures. This

paper empirically documents the size of (2) in the rural sector and measures its consequences for

rural development (per capita income, agricultural land intensity, and rural poverty) for ten Latin

American countries over the period 1985­2000 using a newly assembled data set.

         Diverting part of the government resources to spending in private goods means that

government over-taxes the private sector, under-supplies public goods (including expenditures in
dealing with market failures), or both.2 Whatever the means used by the public sector to finance

spending in private goods, efficiency losses associated with the inevitable distortions caused by

raising more taxes than needed or by under-providing public goods are implied. For the

empirical part of this paper, however, we assume that the total fiscal resources (financial, human,

1The book by Krueger, Schiff, and Valdés (1991) constitutes a distinguished example of the great effort spent by the
profession in documenting and criticizing government interference with otherwise efficient markets.
2A third possibility is instead to increase public debt. Increasing government debt may over periods of time allow
governments to spend large volumes of revenue on private goods without decreasing expenditures in public goods,
while still keeping the tax burden low. But the debt mechanism is available mainly over the short run.


                                                            1

and institutional) available for the rural sector are exogenous and fixed. This allows us to

concentrate the empirical analysis on various forms of crowding out of expenditures on public

goods within a fixed rural budget. For the sake of keeping the issues in perspective, however, the

remainder of this introduction and the next section will retain a broader level of generality in this

respect.

         The under-provision of public goods may have serious negative consequences for the

productivity of private investment, as public goods are important complementaryassets with

private capital. As we shall see below, non-social subsidies, often available only to selected

individuals and firms, appear to contribute little to stimulate growth. Thus, the restricted supply

of public goods (or higher taxes) necessary to finance spending in subsidies may involve a poor

tradeoff as a policy to promote growth. A low supply of public goods means scarcity of human
capital,3 underinvestment in research and development (R&D), less infrastructure, insufficient

sanitary and environmental protection, etc. (World Bank, 2000). All these are important assets

that contribute to increase the productivity of private investments. Hence their scarcity is likely

to constitute an obstacle to economic expansion.

         The biased structure of public expenditure allocation may have important social equity

implications as well. For reasons to be discussed below, government expenditures in (non-social)

subsidies and other private goods tend to be directed toward the wealthier segments of society

and, at the same time, have little effect in promoting employment. The only significant assets of

the poor are their labor force and, for the rural poor, their natural resources. The income of the

poor is thus extremely dependent on their ability to enhance their human capital and to protect

the natural assets, both important public or semi-public goods.

         Over-taxationthat spending in private goods may cause also has negative equity

consequences. It is well known that in most developing countries the tax system is not

progressive, especially because of its great reliance on indirect taxationand because of the
prevalence of tax evasion.4 So the additional taxes needed to finance subsidies to selected


3Education and health care, though not pure public goods, have characteristics of public goods, especially in
developing countries. Their positive externalities are well documented in the literature. The poor, who often
constitute a large segment of the population in developing countries, are generally unable to finance even highly
profitable investments in human capital due to imperfections of capital markets; so if governments do not intervene
in some form, such investments are often not realized.
4"Given the low importance of personal income taxes and property taxes in Latin American countries, the direct
distributive leverage of the tax system in most countries should be expected to be very small or even negative"
(World Bank, 2004, p. 254). According to the World Bank, while in developed countries income taxes and property


                                                           2

individuals and firms are to a significant degree paid by the poor. In addition, the undersupply of

public goods affects the welfare of the poor much more than the welfare of the wealthy.

          Many empirical studies have measured the impact of specific public goods such as R&D,
rural roads, and other infrastructure on agriculture as well as on other sectors.5 What seems to be

missing, however, is a broader approach that may permit evaluation of the impact of public

goods vis-à-vis other types of government expenditures on efficiency and social equity. Without

such a comparison, the conceptual and policy implications of findings pointing to the payoffs of

certain specific public expenditures are in part overlooked.

          At the policy level, there seems to be a degree of consensus regarding the importance of

an adequate provision of public goods for economic expansion and social equity. However, the

same lack of a clear conceptual framework and the general weakness of empirical studies due in

part to such a conceptualdeficit also mean a certain weakness in the policy recommendations. It

is clearly a rather empty policy statement to advise governments to increase expenditures on

education, health, roads, etc., without pinpointing the source of the low expenditures on such

goods in the first place, and without indicating what expenditures to cut (or what taxes to raise)
in the second place.6

          Apart from affecting the rate of agricultural growth, the heavy emphasis of governments

on subsidizing the wealthy is likely to affect the patterns of growth of agriculture. In particular, a

hypothesis that we test here is that (non-social) subsidies tend to make agricultural growth much

more dependent on land expansion than on intensification. In many countries, especially tropical
ones, agricultural growth is linked to area expansion much more than to intensification.7 This

pattern of growth is often responsible for the large deforestation impacts attributed to agriculture

by many studies (Barbier, 2004). It has been shown that when governments are prone to



taxes constitute more than 10.5% of GDP, in Latin America they comprise only 3.7% of GDP. Another study, by
Chu, Davoodi, and Gupta (2000), corroborates this finding.
5 See, for example, the survey by Alston et al. (2000) of studies documenting returns to agricultural R&D and
extension; Psacharopoulos (1994) for returns to education; and the World Bank (2000) for an extensive review of
studies measuring returns to various other public goods.
6It can also be counterproductive because it may induce governments either to increase taxes, which, given the
weaknesses of the tax system discussed earlier, may have negative efficiency and equity effects, or worse to create
fiscal disequilibria that could over the intermediate run become very costly and cause an even greater undersupply of
public goods (Argentina is probably an important recent example of the latter).
7 Graham, Gauthier, and Mendonça de Barros (1987), for example, concluded that area expansion rather than yield
increases accounts for practically all of the growth of Brazilian agriculture between 1950 and 1980. Similar findings
are documented by López (1997) for Ghana.




                                                           3

provide non-social subsidies, expanding farm size beyond optimal levels is a rational response by

farmers signaling their disposition to offer bribes in exchange for subsidies (Bulte, Damania, and

López, 2004). Land expansion is thus a vehicle to promote subsidies, and knowing that

governments are prone to subsidize induces farmers to expand land area. That is, land expansion

and subsidies are the cause and effect of each other.

         In summary, this paper conceptually justifies and empirically tests three hypotheses: (i)

that countries that, ceteris paribus, spend a greater share of rural public expenditures in private

goods have lower agricultural per capita income than countries that spend more on rural public

goods; (ii) that agriculture in these countries tends to be more dependent on the expansion of the

land area rather than on intensification; and (iii) that rural poverty in these countries is worse.

         To test these hypotheses we use a newly assembled data set for the period 1985­2000 for

the rural sector for several countries in Latin America. These data consist of annual rural public

expenditures collected using a consistent methodology across countries and over time. The high

degree of detail of these data is sufficient to allow us to aggregate rural expenditures in public

goods and in private goods.

         The remainder of this paper is organized as follows. First we discuss general conceptual

issues, in Section II. In Section III we present the empirical model. Section IV provides an

overview of the data used. Section V presents the econometric results. Section VI concludes.

II. Conceptual Issues

         The classification of public expenditures into public goods and private goods arises from

the conceptual view that governments' primary responsibility is to, using use taxpayers' money,

supply public goods, including goods and services that are undersupplied as a consequence of
missing markets or of market imperfections.8 The importance of the public versus private good

dichotomy is that while the public good tends to be complementary with private investment, the

private good, as we shall see, tends to be a substitute for private investment.

          Political motivation and the structure of public expenditures. The political motivation of

the government to invest in public goods is fundamentally different than its motivation to do so

in private goods, and such different motivations cause important socio-economic effects of the

public/private good choice. By definition the benefits of public or semi-public goods tend to



8 In practice, as we shall see, the distinction between public and private goods is not sharp. But by and large, more
often than not, a reasonable separation between these goods is possible.


                                                           4

disperse much more among the populationthan those of private goods. This implies that the

political lobby targets private goods that wholly benefit particular lobby groups much more than

they target public goods, of which the lobby group shares only a fraction of the benefits. The

struggle by lobby groups to elicit subsidies from the government is often their prime motivation.

It is well known that successful lobbies are typically constituted by small groups with a clear

common interest and objectives (Olson, 1965). In addition, and most important, successful lobby

groups often have enough financial means to bribe politicians and to contribute to their political

campaigns, or they have enough social status to be able to affect public opinion through a variety
of means.9

         These requirements for constituting a successful lobby are often satisfied by economic
elites and rarely by the poor.10 The implication is by now a stylized fact: Government
expenditures in private goods go almost always to the wealthy and only very rarely to the poor.11

This is in sharp contrast with expenditures in public goods, which generally produce benefits that

tend to disperse across all groups in society, poor and non-poor alike. The social equity effects

of the composition of public expenditures are obvious: The greater the share of government

expenditures in private goods, the worse, ceteris paribus, is income distributionand poverty.

Because the structure of public expenditures tends to be more biased towards private goods when

the economic power of the elites is in the first place already greater, such biased composition of

public expenditures may be both cause and effect of poverty and social inequity.

         The structure of public expenditures and economic efficiency. The public/private good

dichotomy is grounded on another factor which has to do more directly with economic efficiency

than with equity. Investment in public goods provides factors of production that are almost by

definition rarely supplied by the private sector. For this reason one can reasonably expect that

such investment will fill a vacuum that is not fulfilled by the private sector. As a consequence, a

complementary relationship between private investments and public goods is likely to arise.


9Deacon (2002), among others, has empirically shown that less democratic governments (and in general those that
give less opportunity to their citizenry to participate in decision making and that are therefore prone to lobby
pressures from the wealthy) tend to provide a lower share of public expenditures on public goods, instead placing
more emphasis on non-social subsidies.
10Sometimes organized workers, who are often not part of the wealthier segments of society, are able to also
become powerful lobbies. However, the successful workers' lobbies rarely encompass the poor. Recently certain
groups among the poor (some indigenous groups, grass-root organizations, etc.) have been able to become part of
the lobby forces, but their capacity to effectively lobby governments tends to be limited by disadvantages in terms of
group size, education, and other shortcomings.
11For an empirical illustration of this phenomenon see Calmon (2004) with dramatic evidence for Brazil.


                                                             5

Government expenditures in private goods (non-social subsidies), by contrast, often finance

assets that could equally well be financed directly by private entrepreneurs. Even more

important, as argued above these government subsidies go to the wealthy, who already face

generally no constraints to invest at optimal levels (who are likely to face no liquidity constraints

or other types of market imperfections). As shown below, the effect of these subsidies is almost

always detrimental for economic efficiency and, far from contributing to promote private

investment, the subsidies tend to crowd it out and to increase the consumption of the wealthy
instead.12

        Thus, we have another important contrast that justifies the emphasis on the public/private

goods dichotomy: While increasing public goods is likely to promote economic growth directly

(as factors of production) and indirectly (through its positive effect on private investment),

increasing subsidies is more often than not deleterious for growth and private investment.

        Crowding-out and agricultural performance. Government provision of private goods or

subsidies causes three forms of crowding-out with negative efficiency and equity effects:

         (i) Subsidies crowd out the supply of public goods through the government budget,

human, and institutional constraints. The budget crowding-out is obvious: government-provided

subsidies compete with the provision of public goods, not only for government's financial

resources but also for other public resources that also tend to be scarce. The administration of

subsidy programs often absorbs a large share of the limited availability of the government's

human capital. Scarce institutionalcapital is also usually preferentially allocated to subsidy

programs.

        Governments forego investments in public goods that, according to many studies, have

extremely large rates of return. The survey by Alston et al. (2000), for example, reviews

hundreds of studies around the world estimating rates of return to agriculture R&D revealing

extremely high and non-declining rates. Psacharopoulos's (1994) survey does the same for

studies measuring returns to education over many countries, also showing very high rates. Also,

a study by Fan, Hazell, and Thorat (2000) found unusually high returns to a variety of public

infrastructure investment in agriculture in India and China. The World Bank (2000) has gathered

evidence also pointing to very high rates of return to certain investments in protecting the


12Subsidies may, however, be directed in some cases to mitigate market imperfections and externalities that
otherwise may prevent even wealthy entrepreneurs from optimally investing. The evidence available, however,
seems to indicate that subsidies that can be justified in this way are rarely implemented.


                                                             6

environment. Thus the persistence of such unusually large rates of return suggests that under-

investment in public goods is likely to have large negative effects on per capita income. This

conclusion is reinforced by further empirical evidence showing that subsidies have low or even

negative economic rates of return and that they crowd out private investment. We turn to these

issues next.

         (ii) Subsidies may directly crowd out private investments. Consider what is often

regarded as a "desirable" subsidy: the government offers to pay a portion of the cost for a
particular investment.13 The subsidy is of course rationed as the funds are normally less than the

demand. Assume for simplicity that their allocation among producers is transparent, not subject

to corruption. Consider an investor who is able to extract a profitable return out of an investment

(even in the absence of the subsidy) who potentially may qualify for the subsidybut who does

not obtain the subsidy. The producer may go ahead with the investment (without getting the

subsidy) anyway, since it is a profitable one. Alternatively, she/he may opt to postpone the

investment and to try again the following year in the hope of then getting the subsidy. If the

expected value of the subsidy is sufficiently large to compensate the foregone profits of one year,
the producer may decide to delay the investment.14 Thus, investments that are privately (and

socially) profitable may be postponed as a consequence of the existence of the subsidy.

         Among the investors who actually get the subsidy there are two types: those who would

have implemented the investment anyway and those who otherwise would not have (because

they would not be able to get high enough returns). For the former the subsidy is ineffective. For

the latter the subsidy is effective insofar as it causes them to invest, but, in the absence of

positive externalities associated with the investment, they do so at a low social return. Thus, the

subsidy scheme does two things: (i) it increases consumption rather than investment by

producers who are able to obtain the subsidy but who would have invested anyway, and (ii) it

reallocates investment from those producers who potentially could have gotten high rates of

returnbut who did not obtain the subsidy (and who decide to postpone the investment, expecting


13This example is based on an actual irrigation-drainage public subsidy scheme for "small" projects (subsidy
amount up to US$275,000 per producer) in Chile that cost the government about US$30 million per year, using
parameters obtained from the program.
14Suppose the subsidy is 50% (the Chilean subsidy pays between 25% and 75% of the total investment) of the value
of the investment cost and that the rate of return per annum of the investment is quite high, 20%. Assume further
that the producer estimates that the probability of getting the subsidy the following year is 0.5 (in the Chile example,
about 45% of the proposals are funded). If the producer is risk-neutral, he/she will decide to delay the investment for
one year.


                                                            7

to get the subsidy in the future) to producers who obtain a low social returnbut invest only

because they are able to access the subsidy. The net effect on total investment is ambiguous, but

the efficiency impact is negative as a consequence of effect (ii).

         The above example is not just a curiosity. That subsidies, at least in the form in which

they are usually allocated, do not generally promote investment has been shown by several

empirical studies in various countries. Empirical studies using detailed firm-level data by

Bregman, Fuss, and Regev (1999) for Israel, Fakin (1995) for Poland, Lee (1996) for Korea,

Bergstrom (1998) for Sweden, Estache and Gaspar (1995) for Brazil, Harris (1991) for Ireland,

and several others have shown that subsidies and corporate tax concessions targeted to specific

firms are at best ineffective in promoting investment and technological adoption and, in some

instances, even counterproductive. Crowding-out of private investment as a consequence of the

subsidies occurs.

         (iii) Subsidies may indirectly crowd out private investment in the intermediate run. The

third type of crowding-out of private investment as a result of public expenditure biases toward

subsidies is long-run in nature. The low stocks of public goods caused by continuous under-

investment in public goods causes low productivity of private investments over the long run.

This, in turn, is translated into slower investment and slower productivity growthover the long

run (World Bank. 2000).

         The structure of public expenditure and rural poverty. Public goods affect rural poverty

through direct effects and through indirect effects:

(i) Direct effects. Important components of rural public goods are human capital and investment

in the protection of natural resources and the environment. Human capital is the main productive

asset for most poor. However, as a consequence of credit and other market failures they are not

in general able to fully finance investments in human capital, regardless of how high the rate of

return to these investments might be (World Bank, 2000). They are, therefore, largely dependent

on the public sector as a source of financing for these investments. On the other hand, the rural

poor are also highly dependent on natural resources as a source of subsistence (Barbier, 2004).

The price of resource degradation is often disproportionately paid by the rural poor. Increasing

investment to protect natural resources and to mitigate environmental externalities is, therefore,

vital to reducing rural poverty.




                                                  8

(ii) Indirect effects. As discussed earlier, an adequate stock of rural public goods contributes to

sustain growth for agriculture and related rural industries. In general much of the rural industry

linked to agriculture is highly intensive in unskilled labor (Lanjouw and lanjouw, 1995).

Moreover, it appears that the rural sector in many developing countries is important enough

within the broader unskilled labor market to significantly affect the real wages for unskilled

workers even at a national level. Thus a dynamic rural productive sector is a vital source of

employment for unskilled workers as it may play a role in supporting higher real wages. To the

extent that most poor are unskilled workers, these effects may be important in reducing poverty.

III. The Model

         The agriculture per capita value-added function. The approach is to specify a reduced-

form model that explains agricultural GDP per capita as a function of variables that are

exogenous to agriculture as well as of policy variables. We start with specifying a production

function,

(1)             Q = F(L, K, X; A) ,

where Q is agriculture output, L is labor used in the sector, K is a vector of primary factors of

production owned by farmers, X is a vector of purchased inputs, and A is a productivity index.

We assume that F( ) is concave in K, L, and X, and that it is subject to constant returns to scale;
that is, F( ) is homogenous of degree one in L, K, and X.15

         Agriculture value added or GDP is defined as the returns to primary factors, K and L.

Thus,

(2)              G( p,v; K,L; A)  max{ pF(L,K, X; A) - vX},
                                        X

where G( ) is agriculture GDP, p is output price, and v is a vector of purchased input prices. The

function G( ) is a (dual) revenue or value added function and must satisfy certain conditions.

The most important, from our point of view, is that, apart from being increasing and concave in

L and K, the agricultural value added function is also homogenous of degree one in K and L.

This follows directly from the constant returns to scale assumption. We can thus express per

capita agriculture GDP as




15The assumption of constant returns to scale in agriculture has not often been rejected by empirical studies using
farm household data. See, for example, López and Valdés (2000), who provide estimates for five countries in Latin
America.


                                                         9

                  _               G( p,v;K, L; A)
(3)               g( p,v;k; A)                   .
                                        L

Per capita GDP is a function of output and purchased input prices, and the per capita values of

the farmer's owned assets, k, as well as of the productivity factor, A. It is increasing in p, k, and

A, and decreasing in v. The remainder of this section is devoted to the estimation strategy of the

                                        _
per capita agricultural GDP function, g( p,v;k; A) .

       We postulate that output and purchased input prices are determined by world prices,

domestic government policies (including trade policy, subsidies, etc.), and the performance of

the non-agriculture economy, which may also play a role in affecting prices of non-traded

commodities. That is,

(4)                p = ( p*, H,Y) ; v = (v*, H,Y) ,

where a star indicates world prices, H stands for a vector of government policies affecting

domestic prices, and Y reflects conditions in sectors other than agriculture but ones that could

affect prices relevant to agriculture. The variables p*, H, and Yare all subject to change over

time as world market conditions, policies, and non-agricultural growth conditions vary.

       The variable A is also endogenous and affected by variables exogenous to the farmer,

including domestic market prices and government policies,

(5)                       A = A(H, p,v) .

The level of A is postulated to be affected primarily bygovernment policies, particularly public

expenditures in public goods such as R&D, extension, and education. It canalso be affected by

other market conditions that affect the profitability of adopting new technologies, represented by

p and v, as well as by k. Similarly, the conditions of the non-agriculture economy play a role on

agriculture through the market effects associated with demand conditions for the commodities

and inputs relevant to agriculture.

       Combining (4) and (5) we obtain a reduced-form function for A,

(6)                     A = ( p,v,H ,Y) .

We can disaggregate the vector H into T (trade policies), E (per capita total government

expenditures in the sector), and S (the share of government subsidies in the total expenditures in

the sector). We note that S includes both the direct non-social transfers as well as subsidies that




                                                 10

take place via markets that demand government outlays (credit subsidies, deficiency payments,

stabilization payments, etc.). Thus, we can write

(7)                       A = ( p*,v*,T,E,S,Y) .

We expect that A is non-decreasing in p*, E, and Y and decreasing inv* . The effect of S on A

is, according to our previous discussion, non-positive. The variable T is defined as an index of

trade openness. We expect that in countries that have used trade protection to discriminate

against agriculture, as have most countries in Latin America, the effect of T on A is non-

negative.

        Using (4), (5), and (7) in (3), we have now a specification for a reduced form for the per

capita agricultural GDP function,

(8)             g = g( p*,v*,T , E , S, Y ; k ) .
                       (+) (-) (+) ( +) (-) (+) (+)



The signs underneath the variables indicate the expected effects of the various exogenous

variables on the reduced-form per capita GDP. The sign pattern in (8) follows straightforwardly

                                                         _
from the sign pattern of the structural GDP function, g ( ), and from the signs of the effects of

the exogenous variables on k and A pointed out earlier. Equation (8) is the basis for the

specification of the estimating modelof per capita agricultural GDP.

        Econometric estimation of the per capita GDP function. Consider the following

benchmark econometric specification:

(9)              git = 1L[Eit]+2L[Sit] +3L[Tit] +4L[Yit]+5kit + µi +t +it ,

where git is the log of per capita agricultural GDP in country i at year t, L are lag operators, and

        Eit is total government-provided services to rural areas in country i and year t

        Sit is the share of subsidies or private goods in Eit

       Tit is an index of trade policy openness in country i at time t

        Yit is a measure of per capita non-agricultural GDP in country i at t

        µi is either a fixed country effect or, alternatively, a random country disturbance

        t is a common time effect

       it is a random, independently distributed disturbance with mean 0 and variance .       2




                                                     11

         We alternatively allow for a dynamic model where the lagged dependent variable is

included as an additional explanatory variable (git ) in (9). As we discuss below, this requires a
                                                         -1

more complex method of estimation to obtain consistent estimates with limited sample size
(Arellano and Bond, 1991).16 Moreover, this generalization of the model does involve tradeoffs

in terms of the efficiency of the estimates (more on this below). We estimate agricultural GDP

per hectare using the same right-hand-side variables as in (9). Alternatively, we use agricultural

GDP per hectare instead of agricultural per capita GDP as the dependent variable. In this case the

variable Eij is normalized by agricultural land area instead of rural population.

         Several important additional comments about (9) are in order:

(i) World prices. It is difficult to construct an appropriate price index for each country. As a

proxy we use annual dummies (t ) that are supposed to capture international shocks, including

world price changes that may affect the countries in the sample.

(ii) Unobserved country characteristics. There are many country-specific factors such as climate,

land quality, institutions, etc., that may affect GDP but that we do not explicitly control for.

These omitted variables may bias the coefficient estimated if they correlate with the explanatory

variables, or at least affect the efficiency of the estimation. The role of µi is to control for such

omitted variables as country-fixed effects or, alternatively, as random effects.

(iii) Government expenditures. While the vast majority of the government expenditures are

current expenditures, a portion of them are investments that accumulate over time as capital

goods. To control for this we experimented allowing both current and lagged values of E to

influence current GDP levels. The same approach was used for the variable S.

(iv) Simultaneous equation biases. Public expenditures in agriculture may be affected by the

performance of agriculture. This potentially causes simultaneity biases in the estimated

coefficient of Eand S. We thus use lagged values of E and S as instruments when we estimate

the regressions with the annual data. Similarly, the relationship betweenagriculture and non-

agriculture income is likely to be simultaneous rather than unidirectional. This may potentially

bias the estimates of the effect of Y in (9). To diminish the risk of such biases we use instruments

in such a way that predicted rather than actual values of Y are used in estimating (9). In the



16In this case a generalized method of moments (GMM) estimation is used. This means that the lag operators are no
longer needed as the GMM approach uses the lag values of the variables as instruments in a systematic way.


                                                      12

dynamic model, the GMM allows for a more comprehensive and systematic use of instruments

(Greene, 2003).

(v) Agricultural primary factors. The vector k should in principle include per capita agricultural

capital and agricultural land area. However, though for an individual farmer the land area may be

part of his/her optimizing behavior, for the economy as a whole it may in the short run be fixed

and in the long run greatly affected by government policy. Expanding the agricultural land area,

especially in countries with remaining frontier areas, can be considered to be at least in part a

government-controlled policy. If we assume that it is fully an endogenous variable there would

be no reason to explicitly include it in the reduced-form specification (9). If, however, land is in

part exogenous to the sector, then one needs to use it as an additional explanatory variable. We

thus estimate two versions of (9): One that includes land area (through instruments to allow for

its possibly endogenous component) as an additional explanatory variable, and another that does

not. With respect to the other primary factor, we do not have data for agricultural capital, so we

are forced to use the remaining explanatory variables in the reduced-form (9) as proxies. It may

not be unreasonable to assume that agricultural capital is correlated with these variables, which

may affect the profitability of capital in an important way. The interpretation of the coefficients

of the other explanatory variables is, therefore, that they encompass the full effect of such
variables, including the effects through k.17

        Estimating the determinants of agricultural land area. It follows quite naturally from

the previous discussion that we can postulate the following relationship,

landit =1L[Eit]+ 2L[Sit] +3L[Tit] + 4L[Yit]+i +it ,

where i is a random or, alternatively, fixed effect, and it is a random disturbance with the

usual desirable distribution properties.

        Estimating rural poverty. It is postulated that rural poverty is affected by government

policies and other exogenous factors through direct and indirect channels. Government

expenditures in rural areas are likely to affect rural poverty directly through social expenditures

that target the poor. In addition, government expenditures in rural infrastructure provide

employment to part of the rural poor, thus contributing to reduce poverty. Trade policy may also

have direct poverty effects by affecting domestic prices of food necessities and other



17That is, we assume thatk = ( p,v,T, E, S,Y ) , where  is a real value function.



                                                  13

commodities important to the poor. Also it may affect the incentives to production of

commodities mostly produced by the poor vis-à-vis those produced by the non-poor. These

policies also indirectly affect the rural poor if they have an impact on the performance of the

agricultural sector. Finally, the rural poor may also be affected by the non-rural sectors of the

economy through migration and remittances, and by affecting real wages in the rural areas.

        We estimate variants of the following poverty equation,

(10)          povit = 1L[git] + 2L[Yit]+ 3L[Eit]+ 4L[Sit] + 5L[Tit]+i +it,

where i is a random country effect and it is a random disturbance. We expect the effects of g,

E, and Y on poverty to be non-positive while the impact of T to be ambiguous and, according to

the discussion in the previous section, the effect of S to be positive. As in the case of the GDP

equation, we tried various lag structures to capture for cumulative effects of the explanatory

variable E over time as well as other instruments. The most popular measure of poverty is the

head count ratio. However, head count measures may be quite heterogeneous as the definition of

poverty lines tends to differ quite considerably across countries and over time. We use per capita

income of the poorest two quintiles as a proxy for the variable povit .

IV. The Data

        In this section we examine a new data set for rural public expenditures for ten countries

in Latin America recently assembled by the FAO regional office for Latin America and the

Caribbean with the support of the World Bank and others. The annual data for the ten countries
cover the period 1985­2000.18 An advantage of this data is that it is quite disaggregated while at

the same time it has been measured using a reasonably consistent methodology across countries

and over time. In addition the public expenditure data as provided is comprehensive, covering

the vast majority of the items in which the public sector spends money in the rural areas. The

level of detail of the data allows us to obtain a fairly accurate measure of the evolution of the

level and composition of government expenditures in rural areas over the period.

        The data for public expenditures in the rural sector cover all rural expenditures, including

expenditures for agriculture as well as rural infrastructure, animal and plant sanitary protection,

environmental expenditures, and social services to rural areas. The latter group covers


18The countries for which the data covers the full period are Costa Rica, Dominican Republic, Ecuador, Honduras,
Jamaica, Panama, Paraguay, Peru, Uruguay, and Venezuela. For a detailed description of the data see Kerrigan
(2001).


                                                         14

expenditures in rural education, health, and other social sectors. All in all, rural public

expenditures are disaggregated into 12 categories. The expenditure groups (including in brackets

their respective average country share in total public expenditures for the year 1995) are the

following: marketing assistance (28), irrigation(7.7), technological generation and transfers

(4.7), soil conservation(1.2), forest promotion (0.09), plant and animal sanitation(4.2),

communications and information (1.6), "focalized" expenditures directed to particular

commodities (4.9), integral rural development (IRD) (4.1), rural public infrastructure (16.3),

rural education, health, and other social services (25.5), and others (1.7).

        To allocate the above expenditures into private versus public goods we first account for

items that clearly fall into one or the other category. Next we deal with those categories that fall

into a gray area where it is difficult to determine a priori whether they correspond to public or

private goods. In these cases we have looked into more detailed components falling into these

categories in the various country statistics available. This allowed us to obtain an educated guess

as to what proportion of these expenditure items should be approximately allocated to public or

private goods. Finally there are some minor items for which we were not even able to have a

reasonable guess. In the latter cases we simply use a 50-50 rule.

         Among the items that clearly qualify as public goods are technology generation and

transfers, soil conservation, plant and animal sanitary protection, communications and

information services, rural roads, and social services (public education, health, and anti-poverty

programs). The above expenditures were classified as public goods, because they increase the

supply of services that generate important positive externalities (technology generation and

diffusion, soil conservation, plant and animal sanitation), were expenditures directed to at least in

part palliate the impact ofpoverty (public education, health, and social programs), which can be

regarded as a public "bad", and expenditures such as communication infrastructure and public

roads that may roughly satisfy both criteria for pure public goods (non-excludability and non-

rivalry).

         Items that were considered private goods are commodity-specific or focalized

expenditures, marketing promotion (excluding expenditures in communications and market

information that are among the public goods items), and irrigationexpenditures. An important

portion of "market promotion" includes direct subsidies and credit subsidies to mostly well-off

producers. Similarly, most of the irrigation services are completely or almost completely



                                                  15

subsidized and benefit non-poor farmers.19 The third group (the gray group) includes the

categories of forest promotion, IRD programs, and "others". Among these items, a high

proportion of the forest promotion expenditures were in the end allocated to public goods while a

smaller fraction of IRD were considered to qualify as public goods. An important portion of tree

planting expenditures is allocated to eroded areas, thus inducing positive externalities by

protecting soils, contributing to the regulation of water regimes, etc. IRD expenditures are mostly

subsidies to promote particular patterns of agricultural expansion that the planners desire. As
such, a high portion of them do not qualify as public goods.20

          Table 1 provides an overview of the total public expenditures in rural areas and its public

good/private good distribution for the ten countries studied using average annual values for the

1985­2000 period. On average, more than 54 % of the total government expenditures in rural

areas were spent in private goods, while only 45% was spent in public goods. The average

country in the sample spent about US$214 million (2000 dollars) per annum in the rural sector,

of which only $96 million corresponded to public goods and $116 million to private goods. This

large allocation of public monies to private goods may suggest a significant degree of waste. At

the same time it provides an indication of the importance of analyzing the impact of such an

unbalanced allocation of public resources especially on economic performance and social equity.

Even using a much weaker criterion for defining public good items (i.e., classifying all gray area

expenditures as public goods and also allocating 50% of irrigation expenditures to such

category), the average share of subsidies for the sample would still be a high value, about 48%.

          Total rural expenditures showed some real increases in several countries over time,

especially in the period 1995­2000, a period in which they increased an average of almost 20%

per country (from $177 to $215 million per country per year). But there is a dramatic variability

across countries in this respect. The structure of government expenditures changed quite



19 One may argue that classifying irrigation expenditures as private goods while other public infrastructure such as
roads, bridges, etc., as public goods may appear as arbitrary. However, unlike other forms of public infrastructure,
irrigation projects are targeted to a limited number of farmers (who usually pay little if at all for the water service),
excluding everyone else from directly benefiting from the irrigation investment. In addition, irrigation water
generally does not satisfy the indivisibility criterion of a public good; a farmer's consumption of irrigation water
directly diminishes the availability for other farmers of using it. The only remaining question is whether or not the
government irrigation investment helps to solve a market failure, which in general is not the case except for rare
situations where the irrigation project serves poor farmers who may be affected by credit market failure.
20 As reported in the next section, given the inevitable divergent views about the classification of some items into
one or the other category, we did perform some limited sensitivity analysis to check for the robustness of the results
to the classification criteria.


                                                             16

significantly over the period. The average per country share of subsidies in total expenditures

substantially declined, particularly over the last five years of the period. The average share of

subsidies steadily fell from 60% in 1985-90 to 58% in 1990-94, and to 44% in 1995­2000.

Moreover, in practically all countries the proportion of subsidies in total government

expenditures decreased or remained stable, the exceptions being Ecuador and Paraguay. The

structure of public expenditures, not surprisingly, varies a lot across countries. Moreover it also

exhibits a meaningful and large variability within countries over time. These large and seemingly

meaningful variances are extremely important features which allow us to appropriately use panel

data econometrics, including fixed effects methods (Nerlove, 2002).

        Turning to the remaining variables used, the indicator of trade openness is intended to

measure the degree of openness of trade policy rather than trade volumes. Large countries

(population and area wise) inherently tend to trade less than small countries, as they are naturally

more self-sufficient, regardless of trade policy. If we want to use trade (imports plus exports)

over GDP as a proxy for trade policy we must, therefore, normalize for country size and other

structural characteristics (access to sea ports, oil wealth, etc.) that may affect the natural

tendency of countries to trade. We normalize trade flows following a methodology used by

Pritchett (1996), which consists in controlling for countries' geographic size, population, per

capita income and other non-policy variables that are likely to affect the volume of trade. The

index can be interpreted as an indicator of a country's trade policy openness with respect to the
norm.21

        The data ofagricultural land area per country over time is obtained from FAO sources,

and the per capita non-rural income is estimated from national account data. Finally, since no

data for agricultural labor exist, we use the total rural labor force as a proxy for L. Summary

statistics for all the data used are provided in the appendix.

V. The Results

        We estimate the reduced-form for per capita agricultural GDP using various methods of

estimations and aggregating the data in several alternative ways. Our interest is primarily in

probing the robustness of the estimated coefficients. We present estimates using three

specifications: (1) regressions using country data aggregated for relatively long periods of time



21 The actual cross-country regressions used to estimate the norm included more than 120 countries. Details about
the methodology used are available fro m the author.


                                                       17

("long-run" estimates); (2) static estimates obtained using annual data with various specifications

for the lag operators; and (3) dynamic estimates, also using annual data.

          "Long-run" estimates. An attractive feature of (1) is that it is consistent with the fact that

much of the effects of the independent variables upon per capita agriculture GDP are likely to be

embodied through processes that may take considerably more than one year to have an effect.

Working with several year averages instead of annual data might help in elucidating what can be

considered long-run effects. A drawback of the approach is that one loses degrees of freedom by
reducing the number of observations. Alternatively, we used 3 or 5 year simple averages.22 In

general the absolute values of the coefficients tend to be higher when 5-year averages are used,

but their signs and degree of significance change very little.

          Table 2 provides the Random Effects (RE) and Fixed Effects (FE) estimators of the per

capita agricultural GDP regressions using 5-year simple averages as units of observation. The

most important message emerging from all these estimates (as well as from the several

alternative specifications used that we do not report here) is that while government expenditures

have a positive and highly significant effect on agriculture per capita income, the structure or

composition of such expenditures is quantitatively much more important and also of great

statistical significance. In fact, the results show that increasing the share of (non-social)

subsidies keeping the total public expenditures constant has a large negative impact on

agricultural per capita GDP. According to the estimates, a reallocation of just 10% of the

subsidy expenditures to supplying public goods instead may cause an increase in per capita

agriculture income of about 2.3%. And this is obtained without increasing the total government
expenditures.23 By contrast, increasing government expenditures without altering their

composition is much less effective in raising per capita agriculture incomes: A 10% expansion of

government outlays causes on average only a 0.6% increase of agriculture income.

          In 1996­2000, the 50% of the countries spending the least in rural areas spent about $35

per capita, while the 50% spending the most spent about $74 per capita. Thus, if the average

country in the bottom half increased its per capita outlays to levels comparable to the average

country in the top half, per capita agricultural GDP in that country could increase by more than

22We also use moving averages.
23It is important to emphasize that this result is purely a composition effect under the assumption that the efficiency
in use of resources in public goods (which several analysts have judged to be low) remains unchanged. Of course if
the efficiency of public good programs increases, the change in the composition of public expenditures would have
an even greater impact.


                                                            18

3%. With respect to the share of subsidies, in 1996­2000 it ranged from approximately 30% on

average for the bottom half of the countries to 65% for the top half. This means that if an average

country in the top half could readjust its public expenditure share to the level of the average

country in the bottom half, its per capita agricultural GDP might increase, ceteris paribus, by

12.5%.

          The results confirm the positive role of trade liberalization on agriculture income. This

reflects the fact that agriculture is still discriminated against through protective trade policies. An

opening up of trade policies consequently is not only good for overall economic efficiency but

also for agriculture income. The marginal effect of the trade openness index on agricultural per

capita income is about 0.002 and statistically significant. Moreover, this estimate is remarkably

stable and robust to changes of specification. To get an idea of the meaning of this number it is

useful to indicate that the trade openness index for the period for the most open country within

the sample (Costa Rica) is about 60 points higher than that of the country that is least open

(Uruguay). This means that if Uruguay had been as open as Costa Rica over the period, its per

capita agricultural income would have beenabout 0.12 % higher. Thus, though the effect of trade

openness is highly significant, its quantitative value is quite small, a result that is consistent with

other studies showing that the static gains of trade liberalization may not be too great.

          The estimates show an important positive effect of the rest of the economy upon

agriculture. This large effect may, however, be affected by biases due to omitted variables that

could exert similar effects upon agricultural and non-agricultural GDP. The FE estimators do

control for the impact of fixed omitted variables that are country-specific, but not for shocks that

happen over time and that could influence the dependent variable and be correlated with the

explanatory variable. We have also used instruments that could at least deal with reverse
causality biases, but certainly the issue of time-varying omitted variables remains.24 In any case,

it appears that the performance of the non-agricultural sector also positively affects agriculture,

as hypothesized by Gardner (2003), for example. A final important feature about the FE and RE

estimates is their mutual consistency. The method of estimation selected makes little difference

in the sign, statistical significance, and even the size of the coefficients.




24 In the annual estimates reported below we do control for time -varying shocks that have similar effects across
countries. As will be seen, this reduces the value of the coefficient, but the effect remains statistically significant.


                                                           19

         "Short-run" estimates: The static model. Table 3 presents estimates of agricultural per

capita GDP using annual data using RE and FE with robust standard errors. We tried various lag

structures for the explanatory variables. The results are robust to the method of estimation and to

other changes in specification. The sign pattern of the coefficients does not change, though their

sizes and standard errors are mildly affected by the actual lags used. We present in Table 3 the

estimates using one-year lags for the explanatory variables.

          The fact that now we have many more observations allows us to use annual time dummy

variables in the regressions. The time dummies control for unobserved common-to-the-countries
shocks that vary over time.25 The pattern of significance of the estimated coefficients and their

signs are highly consistent with the "long-run" estimates reported earlier. Consistent with the

short-run/long-run interpretation, in general the elasticity estimates using annual data have lower

absolute values than the coefficients estimated using the five-year averages.

          In particular, the effects of total government expenditures in rural areas are still positive

and statistically significant, but their values are about 50% lower than the values estimated using

five-year averages. Similarly, the share of government subsidies in total expenditures is now

negative and significant, as before, but the estimates are smaller in absolute value. Still, the

message is the same: total government expenditures are important in promoting agricultural

income, but the structure of public expenditures is even more important. Switching resources

from subsidies to public goods can be an extraordinarily effective instrument to achieve faster

agricultural growth.

         The reduction of the size of the effect of the non-agriculture income is even more

pronounced than that of the public expenditures variables. In fact, the effect of non-agriculture

GDP falls by about 66% when annual data is used compared to five-year averages, though the

coefficient is still significant. This drastic reduction in the size of this coefficient may reflect not

only the short-run/long-run dichotomy, but also a possible reduction of the size of the omitted

variable biases caused by the fact that we nowpartially control for effects of omitted variables

that do change over time. In any case, an elasticity of about 0.13 still reflects a sizable positive

impact of the rest of the economy upon agriculture.




25Both the two-way FE and RE estimates passed the White's test of heteroscedasticity at the 5% level of statistical
significance.


                                                    20

        The Dynamic Estimates. Table 4 presents the Arellano-Bond one-step procedure GMM

(heteroskedastic consistent) estimates of the per capita agricultural GDP function. Although the

GMM procedure has the advantage over the static estimates of accounting for the inertia that is

likely to exist in the determination of agricultural GDP and in mitigating potential endogeneity

biases of the explanatory variables, it also has some disadvantages. In particular, the Arellano-

Bond estimates tend to have low efficiency and, while first-order autocorrelation is expected, if

second-order autocorrelation exists it may lead to inconsistent estimates.

        The estimates in Table 4 suggest that per capita agricultural GDP adjusts relatively fast,

taking about two years to complete the adjustment to changing exogenous variables. The

estimates concerning the level and composition of government expenditures are highly consistent

with the static ones presented earlier. The sign pattern and significance and the relative

magnitudes of the total public expenditures vis-à-vis the share of expenditures in subsidies is the

same as in the static estimators. A problem with the GMM estimates in Table 4 is that they fail to

pass the second-order autocorrelation test. This of course may lead to inconsistent estimators.

        When we normalize the variables by land area, however, the resulting estimates are free

from second-order autocorrelation. These estimates are shown in Table 5, where for comparison

purposes we also include the land normalized RE and FE static estimators as well. We now

cannot reject the hypothesis of no second-order autocorrelation, even at low significance levels.

Moreover, the Sargan test for the validity of the instruments (a test of the hypothesis that the

instrumental variables are uncorrelated with the residuals) is passed, meaning that the

instruments are acceptable.

        The implicit long-run elasticities (that is, the steady state effects) are obtained assuming

that per capita agricultural GDP does not change over time. Given that the coefficient of the

lagged dependent variable coefficient is about 0.52, the implicit long-run elasticity of total

government expenditures is about 0.097 (0.047/(1-0.52)), while the absolute value of the

elasticity of the share of expenditures in non-social subsidies is almost three times bigger, equal

to -0.26. These long-run estimates obtained using the Arellano-Bond dynamic approach are

similar to the corresponding long-run estimates obtained using 5-year averages static random

effects method, which were 0.055 and -0.228 (reported in Table 2). This high degree of




                                                   21

consistency among estimates obtained using very different approaches suggests that the
estimates are robust.26

         Rural poverty. The quantitative analysis of the impact of the level and structure of public

expenditures on rural poverty is hampered by insufficient availability of rural poverty estimates

for the countries considered and especially by issues of data quality and comparability. The

definition of the poverty line especially among countries and also over time varies a lot. This

may increase the "noise" affecting head count poverty measures. In addition, there are only two

or at most three observations by country, a fact that considerably limits the scope of the

econometric analysis. To increase the number of observations we were able to include three new

countries­Guatemala, Colombia, and Nicaragua­for which we were able to obtain two poverty

observations each, but restricted to the latter 6 or 7 years of the period, and for which measures

over the same part of the period were available for the other variables (although we missed

Uruguay from the old sample due to lack of rural poverty measures).

          The use of per capita income of the poor instead of poverty rate in part deals with the

problem associated with a lack of definition of a consistent poverty line. In fact the estimates

appear much more precise when we use per capita income of the poor instead of poverty rate as
the dependent variable.27         Table 6 shows the estimates of the determinants of per capita income

of the poorest 40% of the rural population for the period 1990­2002, which is the period for

which such income is measured. The data on income of the poor is not annual, but rather

consists of no more than three observations per country for different years during the period

1990­2002. We estimate the income of the poor function using average lagged values of the
explanatory variables over the previous 4 years of the observation.28 This is supposed to capture

the delayed effect of the explanatory variables on the per capita income of the poor.

         The main finding in Table 6 is that both the level of rural public expenditure and its

composition show statistically significant coefficients. Public expenditures contribute to increase

the income of the rural poor modestly, while the composition of public expenditures has a larger

effect. Altering the composition of public spending from subsidies to public goods can cause a


26 Both the GMM and RE coefficients associated with total government expenditures and share of non-social
subsidies were robust to changes in the sample as well. We experimented by removing from the sample one country
at a time and one year of data at a time. The sign and significance pattern was not affected and though the size of the
coefficients did change, their relative sizes changed little. These regressions are available from the author on request.
27 The estimates using poverty rate as a dependent variable are available from the author.
28 We also tried 3 and 5 year average, but the result changed little.


                                                            22

large increase in the income of the poor. This result is suggestive, but one should be cautious

because the data base for this analysis is weak and the number of observations is low.

        Agricultural land expansion. One of the most negative environmental impacts of

agriculture is its effect on deforestation. Agricultural expansion into forested areas has been

found to be an important direct cause of deforestation, particularly in tropical countries (Pfaff,

1999, Chomitz and Thomas, 2003). In land-rich countries agricultural growth has been based

more on extending the land area rather than on the intensification of production in already

cultivated lands (Abdelgalil and Cohen, 2001; Foster, Rosenzweig, and Behrman, 2002). The

impact of agricultural expansion on deforestation is particularly large because of a land

multiplier effect: According to empirical analyses, one additional permanent hectare of land

cultivated requires clearing of between 4 and 5 hectares of forests (Chomitz and Thomas, 2003;

López, 1997). This sizable land use multiplier is due to the fact that one additional permanent

hectare of land cultivated needs more infrastructure and more land for human settlements, and

the new land cultivated is often incorporated into a shifting cultivation cycle (to maintain one

hectare under cultivation in shifting cultivation typically needs intervening at 3 to 5 hectares).

        Table 7 shows the two-way FE and RE estimates of agriculture land. We explain

agricultural land area using agricultural GDP, non-agricultural GDP, total public expenditures in

the rural sector, the share of government subsidies in total rural public expenditures, and the

index of trade openness. We normalize all variables except the last two by the rural population.

To avoid problems of simultaneous equation biases affecting the per capita agricultural GDP

explanatory variable, we use lagged values of per capita agriculture GDP in Table 7

(alternatively various instrumental variables were used, but the results do not dramatically

change). In general the coefficient estimates are quite robust to changes in specification and the

FE and RE estimators are remarkably similar.

        Not surprising, agriculture GDP is an important determinant of the land area. The GDP

elasticity has a value of about 0.30, which is highly significant. This relatively low value of the

GDP elasticity suggests that agricultural growth in principle does not need to necessarily rely on

large area expansion. There are factors other than GDP growth that are at play. Another

important result in Table 7 is that more trade openness is a factor that contributes to intensify

agriculture. This effect is also highly significant. This suggests that in the countries considered




                                                  23

trade openness shifts the structure of incentives toward agricultural products that are less land-

demanding and perhaps more labor-intensive.

        The most important finding for us is, however, the strong effect that changing the

composition of public rural expenditures can cause. In fact, according to the estimates, reducing

the proportion of subsidies in total public expenditures from 50% to just 40% may cause, ceteris

paribus, a reduction of agricultural land area of more than 2%, with the consequent reduction of

the pressure imposed by agriculture on the remaining forests. As Bulte, Damania, and López

(2004) have shown, land expansion is an important instrument to elicit subsidies from

governments in exchange for bribes or other political contributions paid by wealthy farmers to

government politicians. For a variety of reasons discussed by Bulte, Damania, and López, the

ability of farmers to lobby governments and the effectiveness of such lobbying in eliciting

subsidies is related to their land size. Therefore, excessively expanding agricultural land when

governments are prone to subsidize the wealthy is privately rational but socially costly. The

empirical results found give support to this hypothesis.

VI. Conclusion

        This paper has provided evidence suggesting that the structure of public expenditures is

an important factor of economic development. The quantitative importance of this factor appears

to be greater than the traditional factors on which the development literature focuses. In

particular, expanding total public expenditures in rural areas while maintaining the existing

public expenditure composition prevailing in the countries does little to promote agricultural

income and reduce rural poverty. The key issue is not so much how much money is spent in the

sector but rather how public monies are being spent. Spending a significant share of government

resources in non-social subsidies causes not only less agriculture income but also induces an

excessive reliance of agricultural growth on land expansion, thus possibly exacerbating the

negative effects of agriculture on the remaining forests.

        The econometric results provide estimates that suggest that governments over-spend in

mostly politically motivated private goods and subsidies and under-spend in public goods and

anti-poverty programs. Though the very large estimated effects of expenditure composition may

be surprising to some, one should note that these estimates are highly consistent with two

previous pieces of evidence available in the literature. First, the literature reports extremely large

rates of return to many forms of public goods (often 25% to 60% per annum and above),



                                                  24

including R&D, education, healthcare, some infrastructure, and certain environmental protection

investments, and others. Also, those studies that have performed analyses over time generally

find that the rates of return show no tendency to decline over time. Recent surveys of this

literature, cited earlier in the text, find anamazing degree of consensus on the estimated rates of

returnobtained in many studies that evaluate rates of return to public goods around the world.

The fact that the studies reviewed have used different approaches and diverse data bases and

have been applied to a large number of countries around the world means that this consensus

should be taken seriously. These extremely high and non-declining rates of return suggest that

there are important investment opportunities in public goods that are not being fully exploited.

That is, it suggests under-spending in public goods.

        Second, many firm-level studies around the world (some cited earlier in the text) have

found that subsidies aimed at promoting investment, employment, and growth do not achieve

these objectives. In some cases they are counterproductive to these goals. Thus, while the rates of

return to public goods are very high, the rates of return to private goods or subsidies are very low

or even negative. Yet governments tend to spend a large share of their resources on the goods

that have low returns while at the same time foregoing investments with very large rates of

return. Switching spending priorities can obtain massive dividends.




                                                   25

                                           References


Abdelgalil, A.E., and S.I. Cohen (2001). `Policy Modelling of the Trade-off Between
       Agricultural Development and Land Degradation ­ the Case of Sudan'. Journal of Policy
       Modelling, 23, 847-874.

Alston, J., M. Marra, P. Pardey, and P. Wyatt (2000). "Research Return Redux: A Meta-Analysis
       and the Returns of R&D". Australian Journal of Agricultural Economics, 44, 1364-85.

Arellano, M., and S. Bond (1991). "Some Test of Specification for Panel Data: Monte Carlo
       Evidence and an Application to Employment Equations" The Review of Economic
       Studies 58, 277-97.

Barbier, E. (2004). "Natural Capital Resource Dependency, and Poverty in Developing
       Countries: The Problem of Dualism within Dualism", in R. López, J. Stiglitz and M.
       Toman (eds.) Sustainable Development: New Options and Policies, Oxford
       University Press, forthcoming

Bergstrom, F. (1998). "Capital Subsidies and the Performance of Firms," Stockholm School of
       Economics, Working Paper #285

Bregman, A., M. Fuss, and H. Regev (1999). "Effects of Capital subsidization on Productivity in
       Israeli Industry". Bank of Israel Economic Review, 77-101

Bulte, E., R. Damania, and R. López(2004). " On the Gains of Committing to Inefficient
       Production: Lobbying and Low Agricultural Productivity in Latin America", in
       http://www.arec.umd.edu/rlopez/govdev.

Calmon, P. (2004). "Evaluation of Subsidies in Brazil", Unpublished, The World Bank,
       Washington, DC.

Chomitz, K. and T. Thomas. (2003). "Determinants of Land Use in Amazonia: A Fine-Scale
       Spatial Analysis." American Journal of Agricultural Economics 85 (4): 1016-28.

Chu, K., H. Davoodi, and S. Gupta. (2000). "Income Distribution and Tax and Government
       Social Spending Policies in Developing Countries", United Nations University, Working
       Paper # 214, Helsinki.

Deacon, R. (2002). "Dictatorship, Democracy, and the Provision of Public Goods", Dept. of
       Economics, U. of California, San Diego.

Estache, A., and Gaspar, V. (1995). "Why tax Incentives do not Promote Investment in Brazil. In
       A. Shah (Ed.), Fiscal Incentives for Investment and Innovation. Baltimore: Oxford
       University Press.




                                               26

Fakin, B. (1995). "Investment Subsidies During Transition," Eastern European Economics,
        Sept./Oct.

Fan, S., P. Hazell and S. Thorat, (2000). "Government Spending, Growth and Poverty in Rural
        India" AJAE, 82: 1038-1051

Foster, A., M. Rosenzweig, and J. Behrman (2002). "Population Growth, Income Growth and
        Deforestation: Management of Village Common Land in India." Mimeo, Brown
        University.

Gardner, B. (2003). "Causes of Rural Economic Development", Elmhirst Lecture, Conference of
        the International Association of Agricultural Economists, Durban, S. Africa.

Graham, D., H. Gauthier and J.R. Mendonça de Barros, (1987). Thirty Years of Agricultural
        Growth in Brazil: Crop Performance, Regional Profile and Recent Policy Review,
        Economic Development and Cultural Change 36: 1-33

Greene, W. (2003). Econometric Analysis. Prentice Hall, New Jersey.

Harris, R. (1991). "The Employment Creation Effects of Factor Subsidies: Some Estimates for
        Northern Ireland". Journal of Regional Science 31:49-64.

Kerrigan, G. (2001). "Gasto Público hacia el Sector Agricola y Desarrollo de Areas
        Rurales",Seminario Gestión del Gasto Público para el desarrollo Agricola y rural en
        America Latina y el Caribe, Santiago, Chile, August.

Krueger, A., Schiff, M. and A. Valdés (1991) eds. Political Economy of agricultural Pricing
        Policy. Baltimore, Johns Hopkins University Press.

Lee, J. (1996). "Government Interventions and Productivity Growth" 1:392-415

López, R. (1997). "Environmental Externalities in Traditional Agriculture and the Impact of
        Trade Liberalization: The Case of Ghana." Journal of Development Economics 53 (1):
        17-39.

López, R. and A. Valdés (2000). Rural poverty in Latin America, MacMillan Publisher, UK and
        Saint Martin's Press, USA.

Nerlove, M. (2002). Essays in Panel Data Econometrics. Cambridge University Press, New
        York, NY

Olson, M. (1965). The logic of collective action: Public goods and the theory of groups,
        Cambridge, MA: Harvard University Press.

Pfaff, A. (1999). "What Drives Deforestation in the Brazilian Amazon?" Journal of
        Environmental Economics and Management 37: 26-43.



                                                27

Pritchett, L. (1996). "Measuring Outward Orientation in LDCs: Can It Be Done?" Journal of
       Development Economics v.49: 307-335.

Psacharopoulos, G. (1994). "Returns to Investment in Education: A Global Update". World
       Development, 22(9), 1325-1343

The World Bank (2000). The Quality of Growth. Oxford University Press, Washington, DC.

The World Bank (2004). Inequality in Latin America: Breaking With History?. The World Bank,
       Washington DC




                                             28

                                         TABLE 1

GOVERNMENT EXPENDITURES IN THE RURAL SECTOR: ANNUAL COUNTRY AVERAGES, 1985­2000

                                  (MILLIONS US$ OF 2000)




                                 Subsidies as a               Public Goods

                                 Percentage of      Public   as a Percentage   Total
   Countries      Subsidies    Total Expenditure    Goods       of Total     Expenditure
                                       (%)                     Expenditure
                                                                   (%)

   Costa Rica       41.6              47.4           46.1         52.6         87.7


   Dominican       174.6              65.4           92.2         34.6         266.8
    Republic

   Honduras          3.3              10.8           27.6         89.2         31.0


    Panama          82.9              80.8           19.6         19.2         102.5


   Paraguay        106.5              86.5           16.6         13.5         123.1


      Peru         197.3              55.0          161.4         45.0         358.7


   Venezuela       283.8              54.2          239.9         45.8         523.8


    Ecuador         89.8              67.3          43.61         32.7         133.4


    Uruguay          7.7              19.1          32.42         80.9         40.1




Source: FAO, Regional Office for Latin America and the Caribbean




                                            29

                                            Table 2

  DETERMINANTS OF LONG-RUN PER CAPITA AGRICULTURAL GDP. DEPENDENT
VARIABLE:LOG OF AVERAGE ANNUAL PER CAPITA AGRICULTURAL GDP (FIVEYEAR
                                         AVERAGES)



                                                         Panel Regression
Variables
                                           Fixed Effects                Random Effects

Log of average annual public                   0.057***                     0.055***
expenditures in the rural sector                 (0.010)                      (0.010)

Share of non-social subsidies in              -0.231***                    -0.228***
expenditures                                     (0.053)                      (0.053)

Index of trade openness                        0.002***                     0.002***
                                                 (0.001)                      (0.001)

Log of (predicted) non-                        0.396***                     0.405***
agricultural per capita GDP                      (0.033)                      (0.032)

                                               2.606***                     2.563***
Constant                                         (0.253)                      (0.287)


Notes: *** - 5% level of significance. Number of observations: 30; Number of countries: 10




                                              30

                                       TABLE 3
 ESTIMATING PER CAPITA AGRICULTURAL GDP USING ANNUAL DATA WITH ANNUAL DUMMIES.
            DEPENDENT VARIABLE: LOG OF PER CAPITA AGRICULTURAL GDP

               Variables                                Panel Regression

                                          Fixed Effects                 Random Effects

Log of total per capita public                  0.031***                     0.028***
  expenditures in the rural sector                (0.009)                      (0.009)
Share of expenditures on non-social            -0.117***                     -0.10***
  subsidies in expenditures                       (0.052)                      (0.050)

Index of trade openness                         0.002***                     0.002***
                                                  (0.001)                      (0.001)
Log of (predicted) per capita non-              0.133***                     0.140***
  agriculture GDP                                 (0.042)                      (0.042)

Land Area in agriculture                        0.851***                     0.712***
                                                  (0.129)                      (0.105)

Constant                                        4.804***                      3.98***
                                                  (0.788)                      (0.687)

D85                                            -0.245***                    -0.245***
                                                  (0.039)                      (0.039)

D86                                            -0.213***                    -0.214***
                                                  (0.038)                      (0.037)

D87                                            -0.210***                    -0.212***
                                                  (0.037)                      (0.036)

D88                                            -0.169***                    -0.171***
                                                  (0.036)                      (0.035)

D89                                            -0.160***                    -0.162***
                                                  (0.036)                      (0.035)

D90                                            -0.167***                    -0.167***
                                                  (0.037)                      (0.037)

D91                                            -0.145***                    -0.143***
                                                  (0.036)                      (0.034)

D92                                            -0.133***                    -0.132***
                                                  (0.035)                      (0.035)

D93                                            -0.149***                    -0.149***
                                                  (0.031)                      (0.031)

D94                                            -0.109***                    -0.109***
                                                  (0.031)                      (0.031)

D95                                            -0.070***                    -0.070***
                                                  (0.030)                      (0.030)

D96                                                -0.030                       -0.029
                                                  (0.030)                      (0.029)

D97                                                -0.025                       -0.025
                                                  (0.030)                      (0.030)

D98                                                -0.022                       -0.022
                                                  (0.029)                      (0.029)

D99                                                -0.010                       -0.010
                                                  (0.029)                      (0.029)
Number of observations: 133; Number of countries (groups): 10



                                          31

                                                 TABLE 4
DYNAMIC ARELLANO-BOND GMM ESTIMATES OF PER CAPITA AGRICULTURAL GDP. DEPENDENT
                           VARIABLE: LOG OF PER CAPITA AGRICULTURAL GDP

                   Variables                              Model                              Model
                                                             1                                  2
 Lag of log of per capita agricultural GDP                        0.520***                           0.551***
                                                                 (0.194)                            (0.199)
 Log of total per capita public expenditures                      0.022**                            0.027***
 in the rural sector                                             (0.012)                            (0.011)
 Share of expenditures on non-social                              -0.075**                          -0.090***
 subsidies in total expenditures                                 (0.041)                            (0.042)
 Index of trade openness                                           0.001                            0.0003
                                                                 (0.001)                            (0.001)
 Log of non-agriculture GDP                                        0.156                             0.136
                                                                 (0.116)                            (0.087)
 Log of land area in agriculture                                   0.202                             0.212
                                                                 (0.222)                            (0.223)
 Constant                                                          0.003                             0.002
                                                                 (0.005)                            (0.005)
 D87                                                                                                 -0.034
                                                                                                    (0.053)
 D88                                                                                                 0.042***
                                                                                                    (0.019)
 D89                                                                                                 0.002
                                                                                                    (0.028)
 D90                                                                                                 -0.014
                                                                                                    (0.031)
 D91                                                                                                 0.012
                                                                                                    (0.026)
 D92                                                                                                 -0.004
                                                                                                    (0.038)
 D93                                                                                                 -0.016
                                                                                                    (0.037)
 D94                                                                                                 0.005
                                                                                                    (0.034)
 D95                                                                                                 0.022
                                                                                                    (0.026)
 D96                                                                                                 0.026
                                                                                                    (0.025)
 D97                                                                                                 0.004
                                                                                                    (0.017)
 D98                                                                                                 0.004
                                                                                                    (0.023)
 D99                                                                                                 0.007
                                                                                                    (0.020)
 Arellano-Bond test autocovariance in                   1.82**                                     1.96***
 residuals of order 2
Notes: *** 5%, ** 10%, level of significance. Number of observations: 129; Number of countries (groups): 10. All
standard errors are robust.




                                                      32

                                                  Table 5

             ESTIMATINGAGRICULTURAL GDP NORMALIZED BY LAND AREA, 1985­2000.

                 Variables                 Generalized Method of            Panel Regression
                                                Moments              Fixed Effects      Random Effects
Lag of agricultural GDP per hectare                    0.522***
                                                          (0.099)
Log of total public expenditures in rural              0.047***              0.031***           0.032***
areas (normalized by land area)                           (0.014)              (0.009)            (0.009)
Share of non-social subsidies in total                 -0.120***            -0.119***          -0.123***
expenditures                                              (0.059)              (0.048)            (0.049)
Index of trade openness                                     0.001            0.002***           0.002***
                                                          (0.001)              (0.001)            (0.001)
Log of per capita non-agriculture GDP                       0.085            0.140***           0.162***
                                                          (0.056)              (0.044)            (0.043)
Constant                                                    0.004            4.662***           4.515***
                                                          (0.004)              (0.346)            (0.383)
D85                                                                         -0.245***          -0.234***
                                                                               (0.039)            (0.039)
D86                                                                         -0.211***          -0.201***
                                                                               (0.038)            (0.038)
D87                                                        -0.028           -0.208***          -0.197***
                                                          (0.030)              (0.037)            (0.037)
D88                                                      0.047**            -0.167***          -0.157***
                                                          (0.028)              (0.036)            (0.036)
D89                                                         0.013           -0.158***          -0.149***
                                                          (0.028)              (0.036)            (0.036)
D90                                                        -0.016           -0.166***          -0.156***
                                                          (0.028)              (0.037)            (0.037)
D91                                                         0.007           -0.143***          -0.134***
                                                          (0.027)              (0.036)            (0.036)
D92                                                        -0.011           -0.132***          -0.124***
                                                          (0.028)              (0.035)            (0.035)
D93                                                        -0.028           -0.147***          -0.142***
                                                          (0.025)              (0.032)            (0.032)
D94                                                         0.005           -0.107***          -0.102***
                                                          (0.025)              (0.032)            (0.032)
D95                                                         0.020           -0.069***          -0.066***
                                                          (0.025)              (0.030)            (0.031)
D96                                                         0.027                -0.028            -0.025
                                                          (0.025)              (0.030)            (0.030)
D97                                                         0.004                -0.025            -0.023
                                                          (0.025)              (0.030)            (0.030)
D98                                                       -0.0003                -0.022            -0.020
                                                          (0.026)              (0.029)            (0.030)
D99                                                         0.003                -0.010            -0.009
                                                          (0.026)              (0.029)            (0.030)
Arellano-Bond test autocovariance in
residuals of order 2                                         0.99
Note: ***-5% level of significance; ** - 10% level of significance. Number of observations: 133; Number
of countries: 10.




                                                     33

                                            TABLE 6

     ESTIMATING THE DETERMINANTS OF PERCAPITA INCOME OF RURAL POOR, 1990­2002*.

 DEPENDENT VARIABLE: LOG OF INCOME PER CAPITA OF THE 40% POOREST IN THE RURAL SECTOR



                                                           Panel Regression
             Variable
                                              Fixed Effects                 Random Effects

Log of average annual public
expenditures per capita in the rural
sector                                                     0.182***                     0.172***
                                                             (0.095)                       (0.087)

Share of non-social subsidies in
public expenditures in the rural
sector                                                   -2.043***                     -1.967***
                                                             (0.854)                       (0.850)

Log of lagged agricultural GDP                               -1.504                         0.148
per capita                                                   (0.813)                       (0.367)

Log of lagged non-agricultural                                 2.028                        0.266
GDP per capita                                               (0.881)                       (0.279)

                                                           1.649***                     5.296***
Constant                                                     (1.768)                       (0.575)
Number of countries                                             12                            12

Adj R-Sq                                                     0.742                        0.6208

# Observations                                                   34                            34

Hausman test Chi square                                                                     0.056


Note: * Using 5 year lagged averages for all variables but for the dependent variable. All
standard errors are heteroskedastic consistent. ***-5% level of significance;




                                                34

                                                 Table 7.
    ESTIMATING THE DETERMINANTS OF AGRICULTURAL LAND AREA FOR TEN
                                           COUNTRIES, 1985­2000.


                 Variable                                       Panel Regression
                                                  Fixed Effects                 Random Effects

Log of lagged agricultural GDP per capita              0.307***                     0.339***
                                                         (0.061)                      (0.046)
Log of average annual public                          -0.020***                    -0.019***
expenditures per capita in the rural sector              (0.006)                      (0.006)
Share of non-social subsidies in public                0.119***                     0.113***
expenditures in the rural sector                         (0.033)                      (0.030)
Log of (predicted) non-agricultural GDP                    0.028                        0.035
per capita                                               (0.026)                      (0.026)

Index of trade openness                               -0.002***                    -0.002***
                                                        (0.0004)                     (0.0004)
                                                       0.082***                     0.097***
D85                                                      (0.036)                      (0.026)
                                                       0.080***                     0.093***
D86                                                      (0.034)                      (0.024)
                                                       0.071***                     0.084***
D87                                                      (0.032)                      (0.024)
                                                        0.059**                     0.071***
D88                                                      (0.032)                      (0.023)
                                                        0.058**                     0.069***
D89                                                      (0.030)                      (0.023)
                                                       0.068***                     0.081***
D90                                                      (0.031)                      (0.023)
                                                       0.061***                     0.071***
D91                                                      (0.030)                      (0.023)
                                                         0.049*                     0.059***
D92                                                      (0.031)                      (0.022)
                                                        0.056**                     0.064***
D93                                                      (0.028)                      (0.020)
                                                         0.042*                     0.048***
D94                                                      (0.027)                      (0.019)
                                                           0.031                     0.035**
D95                                                      (0.027)                      (0.018)
                                                           0.012                        0.015
D96                                                      (0.028)                      (0.018)
                                                           0.010                        0.012
D97                                                      (0.028)                      (0.018)
                                                           0.011                        0.012
D98                                                      (0.028)                      (0.018)
                                                           0.008                        0.008
D99                                                      (0.032)                      (0.018)
Adj R-Sq                                    0.999                         0.696

Note: All standard errors are heteroskedastic consistent. ***5% level of significance; **10%
level of significance; * 15% level of significance. There are 10 countries and 153 observations.




                                                   35

                                        Appendix


           Summary Statistics of Selected Variables Used in the Regressions

               Variables                 Mean     Std. Dev.   Minimum    Maximum
Share of non-social subsidies in public
expenditures in the rural sector           0.477       0.264      0.030      0.920

Index of trade openness                  -16.698     21.343     -63.210     55.452
Agricultural GDP per capita in 000s
2000 dollars                               0.844       1.487      0.220      7.144

Log of agricultural GDP per capita        -0.170       0.789     -1.514      1.966
Total per capita public expenditures in
the rural sector in 000s dollars           0.060       0.065      0.000      0.412
Log of total per capita public
expenditures in the rural sector          -3.868       2.264    -10.820     -0.886
Non-socialsubsidies per capita in
0000s dollars                              0.033       0.038      0.001      0.228

Log of non-social subsidies per capita    -4.859       2.541    -11.513     -1.479
Public goods expenditure per capita in
000s dollars                               0.027       0.034      0.001      0.184
Log of public goods expenditure per
capita                                    -4.668       2.212    -11.513     -1.691
Land area in agriculture per capita in
000s of hectares                           0.008       0.014      0.001      0.055
Log of land area in agriculture per
capita                                    -5.757       1.260     -7.808     -2.898
Non-agricultural GDP per capita in
000s of dollars                            1.270       1.821      0.112      8.172
Log of non-agricultural GDP per
capita                                     0.207       0.109     -2.189      2.100




                                          36