L’Étranger: Status, Property Rights, and Investment Incentives in Côte d’Ivoire

James Fenske

Abstract

This study uses data from Côte d’Ivoire to show that “outsider” households have relatively small land holdings and less complete rights over land than locals. It then demonstrates a link between property rights and investment incentives by looking at the decision to leave land fallow. “Outsiders” fallow a smaller proportion of their land, indicating that their farms are left uncultivated for shorter periods. This is only partly explained by their incomplete property rights and relative land-poverty. These differences in investment patterns lead to lower yields per hectare. (JEL O12, Q15)

I. Introduction

Recent work in both economics and history has shown institutions to be crucial determinants of long-run economic development (Acemoglu, Johnson, and Robinson 2001; Greif 2006). In particular, institutions that protect the property rights of citizens against expropriation by the state and other interests have been found to significantly impact growth, investment, and financial development (Acemoglu and Johnson 2005). Despite the contributions made by cross-country analyses to our understanding of the role institutions play, these studies suffer from several deficiencies; Pande and Udry (2006) note among these their coarse, urban-biased measures of institutions, the limited number of plausible instruments, and their inability to identify the impacts of changes in institutional quality over time. They argue that these studies must be supplemented with investigations of microdata on specific institutions of property rights, including those governing land tenure. The present paper seeks to make one such contribution by examining the nature and consequences of property rights over land in Cote d’Ivoire.

Land tenure in Africa has long been a subject of interest for both Africans and outside observers. The nineteenth-century explorer Richard Burton described inalienability of land as “incomprehensible to Europeans, but part and parcel of the African mind” (Burton 1863, 96). Atwood (1990) argues that the early debates on the advantages and disadvantages of African tenure systems focused on the supposed tension between efficiency and equity; while a lack of formal individual rights was thought to restrict productivity, policy makers feared that the formalization of tenure would lead to concentration and formation of a landless class.

Evidence on the connection between land rights and productivity has shown, however, that the actual operation of African tenure institutions is complex and that the case for their inefficiency is not clear. African cultivators may have only incomplete rights over their plots and yet have adequate security to permit fixed investments in improvements such as tree crops (Brasselle, Gaspart, and Platteau 2002; Ward-Price 1939). Investigations of the efficiency of property rights institutions must avoid generalization and should focus on the particulars of a specific context. The present paper tests whether the rights of a specific group (migrant farmers) in a narrow setting (the forest region of Cote d’Ivoire) are sufficient to motivate one form of investment (leaving land fallow).

This study uses data from the World Bank’s Cote d’Ivoire Living Standards Measurement Study (LSMS) to address the determinants and impacts of property rights in land. It finds evidence that the weaker land tenure rights of “outsiders”1 relative to “locals” discourages them from leaving their land fallow, rendering them less productive than they otherwise could be. This effect is further reinforced by the fact that these households have smaller parcels of land.

II. Literature and Background

Land Rights and Agricultural Investment

It is generally thought that more secure land rights produce better economic outcomes, although this conclusion comes with several caveats. The present section outlines some of the key issues involved; for a more thorough review, see Deininger and Feder (2002) or Place (2009). Besley (1995) identifies three channels through which land rights impact investment. First, tenure security promotes fixed investments by reducing the risk of expropriation. Second, the right to sell land encourages investment on a specific plot by raising the its value to potential buyers. Third, the ability to use land as collateral increases access to investment funds. This last effect will be important only in regions where the credit market will accept it. Feder and Onchan (1987), for example, attribute the differential impacts of land titling between the Thai provinces they study to differences in regional credit institutions.

Local context will as a result determine the outcomes of any policy intervention; for a summary of how the impacts of titling programs can differ across space and time, see Pande and Udry (2006). In addition, factors that may appear to be indicators of insecure tenure may not in fact limit the investment incentives of smallholders. Brasselle, Gaspart, and Platteau (2002) survey a number of studies from Africa, many of which find no significant impacts of titling or of the form of ownership on investments such as fencing and tree planting. They suggest that the present methods of measuring land rights fail to adequately capture the security actually experienced by African farmers. One measure of rights may be correlated with investment, while another produces no effect; Gavian and Fafchamps (1996) find that, while alienability, rights have no effect on manuring investment in Niger, farmers divert manure from borrowed to owned fields. Atwood (1990) has similarly charged that widespread belief in the benefits of land titling in Africa is premised on “a serious misreading of many African land tenure situations” (p. 660); in several African societies, cultivators have secure and heritable rights of use and transfer. Studies must be motivated by the local context. As discussed below, existing studies have shown that outsiders in Cote d’Ivoire do have insecure rights over land, and so there is reason to expect that this will affect their investment behavior.

A difficulty that must be addressed in studies of the impact of tenure security is the fact that land rights are not exogenous, but depend on the investments a cultivator has made in the land, his power within the allocation system, and the actions he has taken to secure his claims. Besley (1995) specifically has referred to Ghanaian tenure systems as “Lockean,” but this observation is not limited to Africa. Antle et al. (2003) find, for example, that the impact of land titling on terracing in Peru, while significant, is overstated if the endogeneity of titling is not taken into account. Similarly, Miceli, Sirmans, and Kieyah (2001) show that demand for title registration in Kenya is increasing in the value of the land to be registered and decreasing in the distance from the administrative center in Nairobi. Noting this aspect of the relationship between property rights and investment can reverse interpretations of their correlation; Brasselle, Gaspart, and Platteau (2002) find that while investment enhances tenure security, the evidence of a causal impact of land rights on investment is weaker. Baland et al. (1999) find analogous results for tree-planting in Uganda. Further, where property rights can be enhanced by improvements to the land, insecure tenure may in fact encourage investment. Deininger et al. (2003) find such an effect for tree planting in Ethiopia. Conversely, they find that terracing (which is less visible and thus less security-improving) is discouraged by incomplete rights. Methods for dealing with the issue of endogeneity are discussed below.

“Outsider” Rights in Cote d’Ivoire

Cote d’Ivoire itself provides an interesting context in which to study land tenure issues involving migrants. Bonnecase (2001) argues that the colonial state attempted to supplant customary rights with private property, while at the same time encouraging migration of labor into the forest zone, which they believed was underpopulated and occupied by inferior tribes. Claims were established by giving the land value, a policy that necessarily favored those strangers who under custom would owe tribute to locals in order to access land. Blion and Bredeloup (1997) note that many Burki-nabe migrants at first went to the Gold Coast, to avoid taxation and forced labor in the French colonies, until Upper Volta was dismembered in 1932 and priority given to the development of Cote d’Ivoire.

Not all of the original inhabitants of the forest region met with the same experience. The Bete, Dozon (1997) argues, were considered by the French as one of the worst of the forest tribes. Particularly ill-equipped to control the influx of strangers, they fared worse than other groups such as the Agni. Large-scale migration both of Ivoirians and foreigners to the forested regions of Cote d’Ivoire has continued since independence. De Fina (1997) notes that the Baoule, as the pioneer migrant group, have had more negotiating power than later waves, which have included large numbers of Burkinabe. After 1946, the numbers of Burkinabe rose significantly and consisted largely of Mossi migration to plantations in the southeast. As land began to grow scarce in the southeast during the late 1970s, the Burkinabe began to shift into the southwest (Blion and Bredeloup 1997).

Jean-Pierre Chauveau has written extensively on relationships between locals and outsiders during the Houphouet-Boigny era, from 1960 to 1993. He argues that a “pact” between the state and the largely Baoule, Dioula, and Burkinabe migrants was formed in which the latter were granted access to land in return for electoral support (Chauveau 2000). Royalties on stranger farmers were banned, and migrants’ rights were supported by local courts and officials (Chauveau 2006b). Local communities and authorities were pressured to accept migrants. In order that village communities could retain some control over the process of settlement, this was accomplished largely through tutorat, an arrangement under which the newcomers retained a symbolic debt of gratitude to the patrons from whom they received land. Koné (2002, 18) describes the institution as follows:

The social obligations attached to gifted land are wide-ranging but vague. Oral contracts are concluded with a reminder by the owner to “make sure you look after me,” and the incomer is expected to do so for as long as he farms the land . . . visiting his patron on rest days and giving him food, drink and some of the harvest, . . . , giving the tuteur credit and helping his family through deaths, marriages or baptisms.

Chauveau (2006a) notes that a grantee’s concrete obligations include labor, credit, and expenses on funerals, disease, and scholarships. These institutional arrangements have evolved over time. The abolition of royalties has spurred tuteurs to make cash payments a more substantial component of these dues (Chauveau 2006b). Customary demands for free labor or other services rendered to traditional authorities are refused more now than in the past. Chauveau (1997) has pointed out that since at least the 1960s, family heads have circumvented lineage elders in order to grant land to strangers. Though the normative rules governing land tenure have changed very little, the pragmatic rules have evolved, while the strategies and institutions used for accessing land have “diversified” and “multiplied” in response to increased pressure on the land.

Tensions between locals and outsiders in Cote d’Ivoire grew steadily worse from the mid-1970s. Urban economic collapse, which began in the 1980s and intensified during the 1990s, pushed many townspeople back to the countryside in search of land (Chauveau 2000). At the same time, tens ofthousands of Burkinabe left Cote d’Ivoire’s urban centers for the rural sector (Blion and Bredeloup 1997). Local populations had tolerated the influx of migrants so long as the state was capable of compensating them with jobs, subsidies, infrastructure, and public services; during the 1980s, financial crises exhausted the state’s capacity to buy social peace (Chauveau 2006b). New political elites from the western region resented the relationships oflocal tuteurs with their clients and accused them of neglecting their duties toward youth (Chauveau 2006b). Data for the present study was collected during the 1980s; since then, the land question has become more intense and was an issue in the recent civil war (Chauveau 2006b).

This study, as a result, focuses on the land rights and investment behaviors of“outsiders” in the forest region of Cote d’Ivoire. Since many of these farmers migrated into the forests and entered into tenancy arrangements for the specific purpose of establishing plantations of coffee and cocoa, we should not expect their investment in tree planting to be diminished by their incomplete rights over land. Rather, following Goldstein and Udry (2008), this study focuses on their decisions to leave land fallow. Since leaving land uncultivated can send a strong signal that it is not needed, it is one of the investments that will be abandoned first when competition over land such as that described by Chauveau (2000) increases.

The focus here is on the risk of expropriation. There is no quantitative evidence on the frequency with which migrants lost land during this period, though indirect evidence is considered later in this paper. Jacoby and Minten (2007) have argued that the actual risk of expropriation in Madagascar is so small that it cannot significantly alter investment decisions. There are difficulties in applying this argument to the Ivoirian case. First, as shown in the model in Section III, farmers may alter their decisions in order to reduce the risk of expropriation. While insecurity reduces the level of investment, this reduced investment may in turn reduce insecurity, and so expropriation may be rare in practice. Second, while quantitative data is absent, ethnographic accounts describe social movements during the 1980s and 1990s that threatened migrants. During the 1980s and 1990s, Chauveau (2000, 113) argues:

The institution of guardianship by locals over “foreigners” increasingly concealed a situation of permanent dispute and was further called into question under the pressure of claims made by young people, townspeople returning to the countryside and other town-dwellers and officials with interests in their home areas.

Koné (2002), similarly, charges that returning women and youth, along with political officials, contested the arrangements by which their parents had granted land to migrants and joined a “drive to repossess land and give it back to indigenous people” (p. 24). Even during the 1980s, grantors “maintained a hold over the recipient that often enabled [them] to take back the land once it had been converted into a plantation, even though in theory it had been given on a permanent basis” (p. 18).

III. Model

While the literature in development economics has provided detailed evidence on the relationship between property rights and investment incentives, there have been fewer attempts to formally model why incomplete property rights exist over land. The notable exception is explanations of sharecropping. Early models of sharecropping stressed motives such as risk-sharing, shared input or cost provision, credit constraints, and screening of tenants; see Singh (2000) for a review. More recent contributions that differ from these themes have argued that landowners may risk losing their land by granting too large a share of output to the sharecropper (Bellemare and Barrett 2003), and that share tenancy can dampen the cultivator’s incentive to degrade the land through overexploitation (Ray 2005).

Models of other aspects of land tenure are less common. Considering security, Banerjee and Ghatak (2004) argue that eviction threats may be used to encourage investment by the tenant, while Banerjee, Gertler, and Ghatak (2002) have made a similar argument about expulsion risk and labor inputs. Inalienability, indivisible inheritance, insecurity, and other deviations of observed institutions from complete rights of private property have received less treatment. Baker and Miceli (2005) have employed a two-period framework to show the relationship between scale economies, rent-seeking, and the optimality of primogeniture. Goldstein and Udry (2008) model offers of abusua land in the Akwapim region of Ghana for which fallowing is not allowed as a screening device that identifies the truly needy.2 Their model considers grants of land within a community as a poverty-reduction strategy. The present case considers the rights of households from outside the community, and so their explanation cannot apply here.

I argue instead that the problem that gives rise to inefficient investment incentives is one of limited commitment (Ligon, Thomas, and Worrall 2002); if land is left fallow, the authority who has granted it to a migrant household cannot guarantee that it will not repossess the land. In the Ivoirian context, the pressure to repossess comes from alternative claims on the land. A household that leaves a plot unused for a season is particularly vulnerable to the charge that the land should be tilled by the sons of the soil. This section introduces a model of farm investment by a migrant farmer who receives a grant of land from a local landlord who has no use for the land at present but who, in the future, may have a reason to wish to reclaim it.

Setup

There are two players: a migrant m and a landlord l. There are two periods, and the order of events is depicted in Figure 1. The variables that appear in the model are summarized in Table 1. The landlord has an amount ofland T that he is unable to use during the first period. At the beginning of Period 1, he chooses an amount ofland G to grant to the migrant. The migrant does not have cash with which to pay the landlord but instead renders him a tribute in labor during the first period of e units of effort for each unit of land granted. This is intended to capture the sort of obligations described by Koné (2002) and Chauveau (2006a). It is assumed that e is customary, and so it is set exogenously. For simplicity, the landlord values this tribute exactly as income. The migrant then takes the land, leaving a proportion f of the grant fallow and harvesting a fraction 1 − f. Land that has not been left fallow has a yield of 1 in the first period. In addition, in each period the migrant must expend c to remain above subsistence. During the first period, the landlord receives income Embedded Image, while the migrant receives income Embedded Image and pays tribute eG.

Table 1

Variables in the Model

During the second period, the landlord has outside income w. At the beginning of the period, his return on land r is revealed. r is 0 with probability 1 — p and θ > 0 with probability p. He then chooses whether or not to expropriate the land left fallow by the migrant; land under cultivation cannot be expropriated. Ifhe does not expropriate the land, he receives income Embedded Image. If he does expropriate the land, he must incur a cost of q < θ for each unit taken, namely, qfG. If he does so, his income in the second period is Embedded Image.

For the migrant, land that is left fallow yields nothing in Period 1 but provides a yield of ρ during the second period. The yield on land cultivated in the first period is 1 during the second period. If he is expropriated in the second period, the migrant receives income Embedded Image. If he is not expropriated, the migrant receives income Embedded Image.

The landlord’s utility over consumption is logarithmic, and he discounts Period 2 utility by the factor β. His problem in Period 2, once r has been revealed, can then be written as

Embedded Image [1]

The migrant’s utility over consumption is also logarithmic, and his disutility of effort is linear. He also discounts Period 2 utility by the factor β. For now, denote the probability that the migrant is expropriated, which may depend on f, as α(f). His problem in Period 1 is given by

Embedded Image [2]

In Period 1, the grantor knows that f will be a function of the G he chooses in the first period. He then solves

Embedded Image [3]

The remainder of this section solves for the subgame perfect Nash equilibrium of this game using backward induction.

The Landlord’s Second Period Decision

In the second period, if r = 0, the landlord has no incentive to expropriate the land. In this case, his income is simply Embedded Image. If r = θ, he must decide whether or not to take the land. If he does not, this Will give him Embedded Image. If he expropriates the land, he receives income Embedded Image. His decision then is trivial; he will expropriate the land if w + r(TG) + (rq) fG > w + r(TG), that is, whenever the return to expropriation is positive (r = θ or, equivalently, r > 0).

The Migrant’s Decision

The migrant takes into account the landlord’s response to his choice of f in the second period. He knows that he will be expropriated whenever r = θ, so that α(f) = π. This allows his problem in [2] to be rewritten as

Embedded Image [4]

Define

Embedded Image [5]

Assume that the return to leaving land fallow is great enough that, in the presence of expropriation risk, the migrant will still choose to leave some positive proportion of it fallow in Period 1. This is equivalent to (1 − π)ρ > (1/β) + 1. If this is the case, [4] can be solved from its first-order conditions. The migrant’s optimal fallow investment is then given by

Embedded Image [6]

Three results follow from the above:

Prediction 1. The proportion of land left fallow is increasing in the amount of land granted.

This follows immediately from [6], since f*(G) is increasing in G. This result is driven by the assumption that the migrant must exceed subsistence c, and will be observed even in the absence of expropriation risk.

Prediction 2. The risk of expropriation reduces the proportion of land left fallow.

This follows immediately from [6], since f*(G) is decreasing in π.

Prediction 3. The risk of expropriation depresses yields by reducing investment in fallow.

Output per unit land in the first period is 1. In the second period, if the migrant is expropriated, the yield observed on the land that remains in his hands is again 1. If he is not expropriated, the yield observed on his total land holding is

Embedded Image [7]

This is clearly increasing in f*(G), which is itself declining in π.

The Landlord’s First-Period Decision

The landlord knows that he will expropriate the land whenever r = θ, and that the migrant will follow the rule given in [6] when deciding how much land to leave fallow. The landlord takes this into account when solving his first-period problem. Combining [5] and [6], [3] can now be rewritten as

Embedded Image [8]

Since [8] can be solved from its first-order conditions, the optimal grant G* is given by

Embedded Image [9]

Two results follow from [9]:

Prediction 4. Migrants face expropriation risk in equilibrium.

This simply reiterates the conclusion that the landlord will expropriate the fallow portion in the state where his return is θ.

Prediction 5. Migrant land holdings will be restricted in size.

There are two reasons for this. First, migrants observed in the data may have already suffered expropriation. Second, from [9], the grantor will withhold some land from the migrant. He does this in order to be able to use it without paying the cost of expropriation in the event that r = θ in the second period. Specifically, G* < T if either

Embedded Image

or

Embedded Image

Efficiency

Is this insecurity inefficient? Clearly, if θ > ρ, allowing the grantor to repossess the fallow land will allow it to be put to a better use in the second period. Make the following assumption:

Assumption 1. θ > 1.

This is clearly a necessary condition for expropriation to ever be efficient. However, consider an alternative case in which both parties can precommit to dividing the output ex post. In this scenario, let the landlord grant the entirety of his plot, T, to the migrant, who leaves a proportion f0 of it fallow, where f0 is defined as

Embedded Image [10]

This is the level of fallow investment that the migrant would choose if he received a grant of size T and faced no risk of expropriation. In the second period, let the two parties agree to put the land to its best use. If r = 0, the migrant continues using it, and total output is [(ρ − 1)f0 + 1]T ≥ [(ρ − 1)f*(G*) + 1f*(G*).3 Clearly, then, this alternative does better in the case where r = 0. The same logic holds in the event that r = θ, but expropriation does not occur. If r = θ and expropriation occurs, the total output generated will be max{θT, [(ρ − 1)f0 + 1]T}. In the above equilibrium without commitment, second-period output on the migrant’s plot is (1 − f*(G*))G*, while output on the landlord’s plot, including land repossessed from the migrant and subtracting the costs of expropriation, is θ(TG* + f*(G*)G*) — f*(G*)G*. Thus, total output in the second period, when r = θ, is given by

Embedded Image [11]

The inequality in [11] follows on Assumption 1. This alternative contract with commitment, then, does better in both states of the world. From [11], it is evident that the inability to commit to share output ex post leads to three sources of inefficiency. The first is that the plot of land may not be put to its best use in either the first or second periods. The second is that investment in fallow is too small and reduces output when r = 0. Third, expropriation costs the landlord q even when it is his preferred outcome.

IV. Econometric Specification and Data

Econometric Specification

Investigating the impact ofland rights on investment requires accounting for the “Lockean” impact of improvements on security of tenure. The standard method used in the literature is to assume that some determinant of land tenure impacts security without directly affecting investment, and use this as an instrument for the measure of land rights. Mode of plot acquisition, existence and type of title deed, size of farm, length of ownership, existence of litigation, improvements existing at the time of acquisition, and own or parents’ characteristics have been proposed as valid instruments (Antle et al. 2003; Besley 1995; Brasselle, Gaspart, and Platteau 2002; de Laiglesia 2004; Goldstein and Udry 2008; Place and Migot-Adholla 1998). Other strategies to deal with endogeneity include removing time-variant variables, estimating the anticipated change in value resulting from title acquisition, exploiting exogenous changes in government policy or titling programs, and simply ignoring it altogether (Alston, Libecap, and Schneider 1996; Deininger and Chamorro 2004; Deininger and Jin 2006; Feder and Onchan 1987; Field 2007; Lanjouw and Levy 2002; Place and Otsuka 2002; Tikabo 2003).

Brasselle, Gaspart, and Platteau (2002) point out that two-stage least squares is inappropriate with the discrete measures of rights and investment that are used in most studies. More critically, the exogeneity of many of these instruments is questionable. Unobservable determinants of secure tenure may (for instance) make titles unnecessary, cause plots to be owned for longer, and discourage litigation. Since the present paper focuses on the impact of timeinvariant household characteristics on land tenure, it is not possible here to control for unobserved heterogeneity with farmer or household fixed effects as does Shaban (1987) or Goldstein and Udry (2008). The manner in which land is acquired is the best of the instruments used in the literature.

This paper considers the following four-equation system:

Embedded Image [12]Embedded Image [13]Embedded Image [14]Embedded Image [15]

Here, Rit indicates the sum of the selfreported rights the household possesses to sell or transfer land.4 This is a proxy for security. The Sit are “status” variables that, in the analysis, will include dummies for outsider and foreign, as well as years of residence for outsiders. g(Lit) is a function of total land (in hectares) at the disposal of the household; in the analysis that follows, g will be quadratic. fit is the proportion of land that is left fallow.5 Yit is yield, measured as output (in CFA [Communauté Financiére Africaine Franc]) per hectare cultivated. The X matrices in each equation include shares of land planted to cocoa and coffee as primary crops by value,6 proportion of land in a co-op, proportion of land in a development scheme, several characteristics of the household head, other income and wealth of the household, characteristics of the head’s parents, and the number of men and women in the household. The matrix Ait includes dummies for four modes of land acquisition within the past 12 months: whether the household bought land, paid for the use of land, received land as a gift, or sharecropped land. All specifications also include year and village fixed effects.

Predictions 4 and 5 from the model, that the land holdings and tenure security of migrant households are deficient, are expected to be reflected in finding that the coefficients on outsider status in [12] and [13] are negative. Predictions 1 and 2, that insecurity and land poverty reduce fallow investment, are tested by investigating whether Prediction 3 and the coefficients on gF (Lit) are nonzero in [14]. Testing whether these results explain fully the differences in fallowing between outsiders and locals is equivalent to testing whether the coefficient on outsider status in [14] is zero. Similarly, testing whether the coefficient on outsider status is zero in [15] will show whether differences in yields between these two groups—if they exist—can be fully accounted for via fallow investment. Finally, testing whether δ3 = 0 in [15] will show whether land rights affect yields through mechanisms other than fallow investment. Together, these coefficients in [15] test Prediction 3 in the model.

There are several challenges in estimating this system of equations. In [12], potentially tenure-enhancing improvements such as permanent crops must be included, although these are potentially endogenous. Omitting these variables would be problematic, because they may be correlated with outsider status. In the Ivoirian context, the most important of such investments are coffee and cocoa trees, the planting of which may depend on tenure security. Since variables such as the endowment of permanent crops at the time of acquisition and soil characteristics are not available, there is no ideal instrument in the data available for remedying this. Rather, the strategy undertaken here is to use the proportions of total land planted with coffee and cocoa nearing the end of its productive life as instruments in [12].7 This will be valid if the mature permanent crops were planted long enough in the past to not have been determined by present levels of tenure security. A cocoa plantation can support increasing household consumption for 25 to 30 years before renewed investment in replanting is required (Ruf 1995).

It is also possible that the results found for the effect of outsider status in [12] or [13] are not statements about the position of migrants per se, but result instead from omitted heterogeneity; these households may differ from those of locals in ways we cannot observe, and that diminish their rights and land holdings. Two solutions are adopted here. The first is to include as many relevant observables as possible for both groups in X. The second is to adopt the indirect argument employed by Udry (1996) and Rangel and Thomas (2005). If outsider status is included without other household characteristics, and then X is added, the change in the estimated coefficient on outsider status will reflect the extent of bias due to unobservable characteristics. If observables and unobservables are related to the dependent variable in the same way, and if inclusion of observables causes the estimated coefficient to move further away from zero or to change only very slightly, then this suggests unobservable differences between outsiders and locals are not driving the results.

Only alienability rights are observed in the data. As was suggested in Section II, this is less than ideal; a lack of alienability rights may not be indicative of insecure ownership of fallow land. The identification here then rests on the assumption that sale and transfer rights are collinear with other elements of tenure security and can be taken as indicators of the completeness of a household’s rights.

Equation [14] presents its own difficulties. First, there is no direct information available in the data on how many years land is cultivated nor on how long it is kept fallow. The ratio of these should determine its fertility and, though this, yield. Instead, the proportion of total land kept fallow has been treated as a measure of this ratio (both as the dependent variable in [14] and a righthand-side variable in [15]). This should be a valid proxy. Assume, for example, that a cultivator has N plots of land, each of which has an area of Ai. Each plot of land is cultivated for Ci years and left fallow for Fi years. In a randomly selected year, that plot’s probability of being fallow will be fi = Fi/(Fi + Ci), while the ratio of years of fallow to years of cultivation for the same plot will be Embedded Image. Note that Embedded Image. Define Embedded Image as each plot’s share in the farmer’s total land area. The expected value of the proportion of total land kept fallow will be given by

Embedded Image [16]

The proportion of all land kept fallow, then, is a weighted average of an increasing function of the ratio of how long each parcel is left fallow to how long it is cultivated. This will smooth out heterogeneity across plots. If, for example, a farmer has one particularly poor plot that is small and is left fallow for a longer period than the rest, this will raise f (as it should), but only by a marginal amount, since the value of ai for that plot will be small.

The second difficulty with [14] is that the proportion of land kept fallow is censored from below at zero.8 Because the number of observations per village is generally more than 15, a tobit estimator with village fixed effects is used, rather than the trimmed least absolute deviations estimator proposed by Honoré (1992). Since it is possible that leaving land fallow can weaken rights or lead to a loss of land, both Rit and gF(Lit) may be endogenous. The Newey (1987) minimum distance estimator is used to control for this endogeneity, using the acquisition variables A as instruments. Lee (1992) has shown that the minimized distance for this estimator can also be used as atest of the over-identification restrictions.9

Ideally, [15] would be estimated using profit functions or instrumented production functions. Unfortunately, the data available do not permit this. Data on wages and land prices are too poor to accurately impute profits, while no instruments are available for variable inputs such as tools, equipment, manure, and fertilizer. Thus, yields per hectare are included without additional inputs. Manure and fertilizer, which could potentially substitute for fallow, are uncommon in the data. Of 1,345 households in the forest zone sample, three-quarters (998) have at least some land fallow. By contrast, fewer than 7% (91) had purchased fertilizer, and only a single household purchased manure. The proportion of land kept fallow is treated as potentially endogenous here and is instrumented using the modes of acquisition mentioned above.

Data

The data used are taken from the Cote d’Ivoire LSMS. This survey collected information on a rotating panel of households from 1985 to 1988. Each year, half the households were reinterviewed, resulting in three distinct two-year panels. Not all available observations are used for this analysis. Because this study is motivated by the literature on “outsider” rights in the forest region, only households from the “East Forest” and “West Forest” regions are used for the analysis. Outliers10 are also removed, as are households with zero or missing values for the sum of cultivated and fallow land. Missing years of residence for outsiders are recoded as zero.

Though the data set contains a rotating panel, this aspect of the data is not used until Section V. The reason for this is that the main question of interest is whether “outsider” households differ from others in their land rights, land holdings, fallow investment, and yields. Outsider status does not vary with time, and so including both observations from the households in the rotating panels would effectively double count them. Where a household appears twice, only the first observation is kept. The final sample thus contains 1,345 households. In Section V, the panel aspect of the data is used to investigate whether households that leave their land fallow in a given year are likely to lose it the next. This requires use of the panel; since lags are used, double counting is not an issue here.

The data suffer from several shortcomings that have been noted by other researchers (Jacoby 1995; Udry and Duflo 2004) and must be addressed here. The construction of the panels was faulty; if a household relocated, the new occupants of the old dwelling were surveyed without a change to the identification number. Though the data contains information on which households can actually be identified, this is accurate only for the first of the three rotating panels. Thus, for the panel analysis in the final table I keep only the households in which the household head aged between zero and two years between interviews and gender of the head is unchanged.

The other major shortcoming of the data is that sale and transfer rights are available only at the household level, precluding any plot-level analysis of land holdings or land rights, and thus replication of many of the methods (such as the use of household fixed effects) used in other papers. Since my focus is on household-level determinants of rights, this is justifiable, though Goldstein and Udry (2008) have shown that it is the interaction between individual status and the mode of plot acquisition that determines plot-specific rights. The analysis is thus unable to reject an alternate hypothesis in which unobserved household-specific variables both diminish tenure security and discourage fallowing, while there is no actual relationship between these two variables. Although aspects of tenure other than alienation rights, such as the ability to make permanent improvements, the need for approval of transfers or sales, tutorat status, or a direct measure of security of land left fallow are absent from the data, the significance of the results that follow implies that sale and transfer rights are a reasonable proxy for the overall security of a household’s land tenure.

Despite these difficulties, these data are of particular interest for three reasons. First, time period and context aside, they allow for further evidence to be brought to bear on the question of whether land tenure affects agricultural investment in Africa. Second, while the relationship between locals and outsiders in Cote d’Ivoire has received considerable attention as a matter of political economy, there has as of yet been no study of its consequences for economic welfare or efficiency. Third, the literature on Cote d’Ivoire has only stressed the deterioration of outsiders’ position since the economic collapse of the 1980s, especially during the 1990s. The data, coming from the mid to late 1980s, provide evidence that the insecure tenures of outsider households reduced investment even during the Houphouet-Boigny era, and that this may have been true even before the macroeconomic deterioration.

V. Results

Summary Statistics

Summary statistics for all observations are given in Table 2. Outsider-headed households account for nearly a third of the sample, while non-Ivoirian households are 15% of the whole. The head of a typical outsider household has lived in his current village of residence for 20 years.11 The average household in the sample has slightly more than 13 ha of land at its disposal, a little over 2 ha per person. A third of this is generally kept fallow. Using [16] this suggests a typical ratio of two years of cultivation for every year of fallow.12 More than half of households claim the right to sell their land, and two-thirds say that they can transfer at least some portion of it. The typical household produces 9,500 CFA/ha in output of annual crops.13 Between 10% and 15% of the typical land holding is planted with either coffee or cocoa as its primary crop.

Table 2

Summary Statistics

Additional summary statistics are presented in Tables 3 and 4 that compare “outsiders” and “locals.” These comparisons motivate the regressions that follow; the t-tests reported compare the means between migrants and locals. Outsider households are approximately 20% less likely to possess each of the alienability rights; this difference is statistically significant. They also have half a hectare less land per person on average; this is again significant. On average they keep 19% less of their land fallow, but the differences in output per hectare are not statistically significant. They also keep larger proportions of their lands planted to cocoa or coffee, though the former difference is insignificant. Table 4 compares the modes by which the two groups have acquired land in the past 12 months. Outsiders are more likely than locals to have bought, paid for, or sharecropped land. Land acquired by these means, particularly that paid for and sharecropped, forms a larger share of the total land holdings of outsiders than of locals.

Table 3

Differences in Means

Table 4

Modes of Land Acquisition

The remainder of the analysis in this section builds on these results, showing that the differences in land rights, land holdings, and fallow proportions for outsiders and locals are robust to the inclusion of other controls. The key empirical result of this paper is that outsiders have less land and fewer rights over it, which drives them to fallow less, potentially lowering their yields. This can be represented as a simple set of causal links:

Embedded Image

The rest of this section deals with each of these relationships in turn.

Alienation Rights

Table 5 tests Prediction 4 by estimating [12]. Column (1) regresses the sum of rights held on the status variables alone. Column (2) includes the function g(L), the acquisition variables A, and the X matrix of other controls. Column (3) repeats this, instrumenting for the shares planted to cocoa and coffee by the shares of land planted to coffee and cocoa nearing the end of its productive life. The Cragg-Donald F-statistic shows that these are strong predictors of the potentially endogenous regressors. Since this is just identified, no overidentification test is reported.

Table 5

Land Rights

The ordinary least squares (OLS) estimate in Column (1) is evidence that outsiders have fewer rights of alienability than others, and that there is an additional negative effect of being a foreigner on land rights held. Outsiders appear to gain rights over land as years of residence increase, but the point estimate implies that the rate of assimilation is slow; more than four decades of residence are needed to make up for the initial handicap. This may also reflect the different terms on which successive waves of migrants received land. Over time, as pressure on land grew, local communities may have become more concerned about their grants to migrants and tightened up the rules of acquisition.

Adding additional household characteristics causes the point estimate on outsider status to fall by about one-third. The point estimate on the added impact of being foreign falls by half and becomes insignificant. The coefficient on years of residence falls by a third, and is only marginally significant. This suggests that some of the difference in alienability rights held between outsiders and local inhabitants is not due to status alone, but also to differences between the unobservable characteristics of these groups. Still, since the coefficient on “outsider” remains large and significant in Columns (2) and (3), there does appear to be a direct impact of status on land tenure, though the coefficient in Column (2) is likely an upper bound on this.

Column (2) does not validate the “Lockean” interpretation of African tenure systems, since cocoa and coffee enter with insignificant effects on the sum of rights. Since the expected endogeneity bias is toward a positive relationship, it is not expected that the IV results will differ in interpretation. When cocoa and coffee are instrumented in Column (3) using tree crops nearing the end of their productive life, their impact remains insignificant. The acquisition variables matter, but only weakly. Of these, sharecropper status is the most significant. This will affect the validity of the IV results—the instruments may be weak, and there are strong a priori reasons to expect that sharecropping status will not be excludable.

Size of Land Holdings

Table 6 tests Prediction 5 from the model by looking at the amount of land available to the household. It gives estimates of [13]. Column (1) regresses the total land available to the household on the status variables alone. Column (2) includes the acquisition variables A and the X matrix of other controls. Outsiders clearly have less land, though there does not appear to be an additional effect of foreign status. The addition of other controls causes the estimated coefficient to fall by nearly half, suggesting unobservable characteristics of outsiders play a larger role in [13] than they do in [12].

Table 6

Land Holdings

There does not appear to be much of a reversal of this pattern with years of residence; though the effect is statistically significant, it would take a lifetime for an outsider household to make up for its initial land deficit. These results contrast with the depiction of Burkinabe migrants in Chauveau (1995) as aggressive land accumulators. The modes ofland acquisition are only very marginally significant in their effects. The positive coefficient may also reflect the possibility that earlier migrants, who acquired land when it was less scarce, received initially larger holdings. The data does not contain information on the history of land acquisitions, but only on land that has been acquired in the past 12 months.

From Table 4, between 7% and 16% of outsider households reported having acquired land by either purchase, payment, gift, or sharecropping in the past year. For these households, these acquisitions account on average for 30% to 60% of total holdings. If these rates were constant over time, they would suggest that, rather than adding small parcels piece by piece, migrants make acquisitions of 2.5 to 4.5 ha at a time, but do so only once every 5 to 10 years. These figures would, however, imply a much faster rate of accumulation than the coefficient estimate on years of residence. Pairwise correlations between years of residence and these modes of acquisition (presented in Table 7) are negative and imply that it is predominantly recent migrants who have acquired land by these means.

Table 7

Mode of Acquisition and Years of Residence

Fallow Investment

The next stage in the causal link is the impact of tenure security and the availability of land on the proportion of land kept fallow. These are Predictions 1 and 2 in the model. Table 8 gives estimates of [14]. Column (1) includes only land rights and a quadratic function of land holdings as controls. Column (2) adds the status variables. Column (3) adds the matrix X of other controls. Column (4) includes the acquisition variables A that are used as instruments in Columns (5) and (6). In Column (5) land rights are treated as endogenous, while in Column (6) both rights and land holdings are treated as endogenous.

Table 8

Fallow Investment

The tobit results imply that land-poor households and those without rights of alienability fallow a smaller percentage of their land. This implies that alienability rights are a good proxy for the security with which a household may leave land unused without fear of losing it. Addition of other controls weakens the impact of land rights, suggesting again that unobservable variables are in part responsible for these results.

Since it is possible that leaving land fallow weakens rights, the expected bias on alienability rights due to endogeneity is to make the coefficient smaller in magnitude. The IV results in Column (5) suggest that this is the case, though the evidence here is poor. The value of the Cragg-Donald F-statistic is far below the critical values proposed by Stock and Yogo (2005), suggesting the instruments are weak. Further, the significance of the instruments in Column (4) and the failure of the overidentification test in Column (5) are evidence that they cannot be excluded. This should not be surprising, since the manner in which land is acquired is often linked to the conditions placed on its use. It is unlikely that a landowner would permit his sharecropper to leave his portion ofland unused, or that a cultivator would pay for use rights over a plot but fail to farm it within the same year. These concerns aside, the coefficient on alienation rights in Column (5) rises in magnitude, suggesting that the coefficient in Column (4) is a lower bound. Column (6) attempts to control, for the same reason, for the endogeneity ofland holdings. Here, however, the predictive power of the instruments is even weaker, and no inference is warranted.

Depending on the specification, the results here suggest that outsider households fallow between 16% and 25% less of their land than do locals, even when controlling for their observable characteristics, land rights, and land holdings. This point estimate is comparable to the mean difference of roughly 19% reported in Table 3. The additional impact of foreign status disappears as other controls are included. The proportion fallow rises with years of residence, but again the small point estimate suggests that assimilation of outsiders is slow. Together, these results imply that the inferior land rights and land holdings of outsider households partly explain the difference in fallow investment observed between them and locals, but cannot fully account for it.

Yields

Finally, the link between fallowing and productivity, Prediction 3 in the model, must be established. Table 9 presents estimates of [15]. Column (1) includes only proportion fallow as a control. Column (2) adds the status variables, land rights, and land holdings. Column (3) adds the matrix X of other controls. Column (4) includes the acquisition variables A that are used as instruments in Column (5). In Column (5) the proportion of land left fallow is treated as endogenous. To reduce the impact of outliers, yields are top-coded at 100,000 CFA/ha.

Table 9

Yields

The impact of proportion fallow is strong in the uninstrumented results, suggesting that a roughly one standard deviation increase in the proportion fallow will raise yields of annual crops by approximately 5,400 CFA/ha,14 more than half of the mean. The sign of the endogeneity bias here is not clear; it is possible that the returns to an additional year of fallow are greater on better land, but worse land may be cultivated for fewer years if it degrades faster. The IV results do not provide more robust estimates; the Cragg-Donald F-statistic suggests that the instruments are weak, while their near significance in Column (4) implies they are not excludable. Since the coefficient falls in magnitude, this suggests the endogeneity bias is positive—better land is left fallow longer.

The status and land rights variables here are almost always insignificant. If the sum of rights is included as the sole control, there is still no significant correlation with yields.15 The coefficient on outsider status suggests that, controlling for fallow, yields of annual crops are slightly better on outsider farms.

Fallow as a Source of Insecure Tenure

In the present paper, it is asserted that households who fallow their land risk signaling that they do not need it, and hence losing it. The panel aspect of the sample allows us to test if in fact households that leave their land fallow risk losing their holdings. Tables 10 and 11 present the results of an estimation of the following equation:

Embedded Image [17]
Table 10

Expropriation Risk

Table 11

Expropriation Risk

If fallowing creates risk, we should see total land holdings fall for those households that left land fallow in the previous year. Actual expropriation is not observed in the data, and so DLit+1, net of any land gained through purchase, payment, receipt as a gift, and sharecropping, and net of any land lost through sale, transfer, or gift is used as a proxy. This is the “unexplained” change in land holdings. It will include land lost through expropriation, but also any other unrecorded transfers and changes due to measurement error.

In this specification, it is possible that fis no longer a proxy measure for the variable of interest (length of fallow) but may be the variable that directly determines the risk of expropriation. To reduce the influence of outliers and measurement error, only households whose total holdings fluctuated for unexplained reasons by less than 20 ha are included. The OLS regressions are consistent with an interpretation that leaving land fallow creates a risk of loss; a one standard deviation increase in the proportion ofland held fallow, which for the mean household is approximately 3.6 ha,16 is associated with a loss of roughly 1 ha of land.17 The possible endogeneity here is that households more prone to losing land will leave less of it fallow, so this should be taken as an underestimate. The IV results in Columns (4) and (5), however, do not allow for this to be estimated more precisely, since the Cragg-Donald F-statistic suggests that the instruments are weak.

There is an additional concern in Table 10: measurement error may lead to these results holding mechanically. If a household’s fallow land is overstated in the first year and corrected in the second, this will drive up the lagged proportion fallow while appearing in the data as a loss of land. Thus, Table 11 repeats the results of Table 10 using indicator variables for whether any land was left fallow, and whether any land was lost. The results here are consistent with Table 10 only before additional controls are added. In Columns (2)-(5) this result is marginally insignificant.

The results also suggest in Tables 10 and 11 that, controlling for land left fallow, alienation rights reduce the risk that land will be lost.

Discussion

The empirical results of this study are the following:

  1. Descriptive statistics suggest there are large differences in the alienation rights, land holdings, and fallow proportions of locals and outsiders.

  2. Outsiders are less likely to have sale or transfer rights than locals, and this improves only slightly with years of residence. This result cannot be attributed only to unobservable differences between outsiders and locals, though these do matter.

  3. Outsiders have less land than locals, and land available does not increase much with length of residence. Unobservables play a greater role here than in determining land rights.

  4. Sale rights and land holdings are significant determinants of fallow investment. Land holdings and alienation rights do not explain the entire difference in fallowing between outsiders and locals.

  5. While fallowing enhances yield per hectare, there is no direct effect of land rights on yields.

  6. There is suggestive evidence that leaving land fallow produces a risk of losing it, but this may be an artifact of measurement error.

The results of this paper are unsurprising when compared to other studies of African land tenure. Tenure security, in so far as it is imperfectly proxied by alienability, encourages fallowing, which may then in turn affect output. The results of Goldstein and Udry (2008) thus hold in an Ivoirian context. What is interesting here is the strength of the status variables. Brasselle, Gaspart, and Platteau (2002) find in their sample from Burkina Faso that immigrant farmers have the same rights as original residents (though controlling for migrant status does cause their mode of acquisition variables to become insignificant). Similarly, Berry (1975) finds no evidence of systematic discrimination in the terms on which outsiders were granted land, in her pioneering study of the cocoa-growing regions of southwestern Nigeria. The present results contrast sharply with these and are indicative of the difficulties of making generalizations, even across the West African forest region. The key determinants of property rights in African tenure systems vary across time and space.

VI Conclusion

This paper has used LSMS data from Cote d’Ivoire to provide further evidence that property rights matter for investment incentives. It finds clear effects of status on the terms by which households access land and the rights they possess over it. It also finds that the differences in property rights that arise as a result of status have significant impacts on investment behavior. In the present case, “outsiders” are found to invest less in fallow, due in part to their less secure land rights. There are, however, a number of qualifications that must be added to these conclusions. Most notably, the lack of strong, excludable instruments in the data make it impossible to estimate many of the coefficients of interest precisely, and many of the results presented can be understood only as bounds on the true parameters. Further, the necessity of aggregating the data to the household level prevents elimination of certain alternative hypotheses and limits the extent to which the determinants of insecure tenure can be investigated here. Despite these difficulties, the study makes a strong connection between status and rights, while drawing attention to a form of agricultural investment that has often been neglected in the literature on land rights and investment.

Footnotes

  • The author is a Ph.D. candidate, Department of Economics, Yale University. Many thanks to Michael Boozer, Timothy Guinnane, and Christopher Udry for their guidance and advice, and to Achyuta Adhvaryu, Rahul Deb, Tavneet Suri, and the participants of the Yale Development Lunch for their comments on this paper. I am also grateful for the feedback of two anonymous referees. All remaining errors are mine.

  • 1 I use the word “outsider” to refer to households whose head resides in a village that is not his village of birth. This will fail to capture second-generation outsider heads but is a less problematic measure than ethnicity, the only alternative available in the data.

  • 2 This model has been removed from this paper but is available from the author.

  • 3 This follows immediately from fκ(Gt)≤fo and G.≤T.

  • 4 This is the same method used by Besley (1995). Alternatively, Brasselle, Gaspart, and Platteau (2002) use an “internally consistent hierarchy of rights,” while Kabubo-Mariara et al. (2006) and Roth, Unruh, and Barrows (1994) have used factor analysis to aggregate together their measures of land tenure.

  • 5 An alternative measure would use the proportion of land not planted to permanent crops (which cannot be left fallow). Unfortunately, intercropping and measurement error in the acreage planted to these crops makes this problematic, and so fallow has been calculated as a proportion of all land.

  • 6 A typical entry in the data will be similar to the following example: For one household, five lines of data are recorded. In the first line, the “Crop” is cocoa, the “1st Other Crop” is Coffee, the “2nd Other Crop” is Taro, the area in hectares is 2, and the value of output is 1,000 CFA. In the second line, these are Coffee/Cocoa/ Taro/2/500. In the third line, these are Taro/Coffee/ Cocoa/2/100. In the fourth line, they are Yams/Cassava/ None/1/300. In the fifth line, they are Cassava/Yams/ None/1/200. In this example, the household has 2 ha planted to cocoa, coffee and taro, and 1 ha planted to yams and cassava. In the analysis, it would be treated as having 2 ha planted to cocoa as a primary crop and 1 ha planted to yams as a primary crop, since these are the most important by value on their respective parcels. It would also be treated as having an output of 600 CFA of annual crops—100 of taro, 300 of yams, and 200 of cassava.

  • 7 This is found by multiplying the portion planted to each tree crop as a primary crop by the proportion that is nearing the end of its productive life (since these were collected as separate variables in the data).

  • 8 Censoring at 1 occurs very rarely in the sample.

  • 9 This is implemented using the overid module written for Stata by Baum et al. (2007).

  • 10 Households with 212,222 ha of land, household sizes above 50, and yields (annual plus perennial) per cultivated hectare above 1,000,000 CFA/ha were treated as outliers.

  • 11 This variable is coded zero for nonoutsiders, since for them it would be perfectly collinear with age.

  • 12 Embedded Image

  • 13 There are 24 crops in the data. Nine (cocoa, coffee, rubber, coconut, oil palm, banana, fruit trees, wood, and cola nuts) are treated as perennial crops, while the others (cotton, peanut, tobacco, pineapple, sugarcane, cassava, yam, taro, sweet potato, maize, rice, millet/sorghum, vegetables, and “other”) are treated as annual crops. While leaving land fallow should improve the yield of annual crops, it should have no effect on perennials, and so only the yield of annual crops is included as a dependent variable. On average, annual crops account for 71% of total output, though the total yield per hectare averages 19,982 CFA, reflecting the greater importance of perennials for the most productive households.

  • 14 0.27 × 20 ≈ 5.4.

  • 15 Not reported.

  • 16 0.27 × 13.46 ≈ 3.6.

  • 17 0.27 × 4 ≈ 1.

References