Sharecropping, Insecure Land Rights and Land Titling Policies: A Case Study of Lac Alaotra, Madagascar

Transcription

1 Development Policy Review, 2009, 27 (1): Sharecropping, Insecure Land Rights and Land Titling Policies: A Case Study of Lac Alaotra, Madagascar Marc F. Bellemare Sharecropping between poor landlords and rich tenants has hitherto been the subject of very little academic scrutiny. Given that such reverse share tenancy contracts are mostly at odds with the canonical risk-sharing explanation for sharecropping, this article discusses a rationale for them that relies on weak property rights as well as the legal doctrine of adverse possession, and tests it using data from Lac Alaotra, Madagascar, where this type of tenancy accounts for one-third of land rentals. The empirical findings are discussed in relation to recent land-reform policies by the Government of Madagascar, the World Bank, IFAD, and the Millennium Challenge Corporation. Key words: Sharecropping, property rights, land titling, Madagascar 1 Introduction Ever since Adam Smith remarked in The Wealth of Nations (1776, 1976: 412) that sharecropping was an inefficient arrangement between slavery and the English fixedrent system, the existence of land-rental contracts in which the landlord and the tenant share the crop has been of interest to social scientists and the study of agrarian contracts has generated a wide body of literature. 1 Following the conventional wisdom, sharecropping is commonly thought of as a land-rental contract matching a rich landlord and a poor tenant in which the landlord leases out her 2 plot to the tenant in exchange for Assistant Professor of Public Policy and Economics, Duke University, 201 Science Drive, Box 90312, Durham, NC , USA (marc.bellemare@duke.edu). He thanks Chris Barrett, Phil Cook, Anirudh Krishna, Bart Minten, and an anonymous referee for useful comments and suggestions, and Anne Fletcher for research assistance. He is also grateful for financial support from the National Science Foundation through Doctoral Dissertation Improvement Grant SES ; from USAID through grant LAG-A to the BASIS CRSP; and from the Social Science Research Council s Program in Applied Economics with funds provided from the John D. and Catherine T. MacArthur Foundation. 1. Perhaps the best known of the debates has been between the Marshallian and the Cheungian schools of thought. Marshall (1920) saw sharecropping as inefficient because the landlord can only imperfectly supervise the tenant, who then provides less effort than under a fixed-rent contract, given that only a fraction of his labour effort is remunerated. Cheung (1968) saw sharecropping as efficient because the landlord could perfectly supervise the tenant. Whether supervision is perfect or not is ultimately an empirical question, and many have taken it upon themselves to test the two views. See, for example, Bell (1977); Shaban (1987); Laffont and Matoussi (1995); Pender and Fafchamps (2006); and Arcand et al. (2007). In most cases, the Cheungian view has been rejected in favour of the Marshallian. 2. In this article, the feminine is used to designate landlords, and the masculine to designate tenants. Published by Blackwell Publishing, Oxford OX4 2DQ, UK and 350 Main Street, Malden, MA 02148, USA.

2 88 Marc F. Bellemare a share (for example, one-third, one-half, two-thirds, and so forth) of the crop. 3 It is this precise wealth ordering rich landlord and poor tenant that has caused some to decry sharecropping as a mechanism by which the landed rich exploit the landless poor (Pearce, 1983), and others to describe it as a second-best arrangement trading off risksharing and incentives (Stiglitz, 1974). Indeed, for economists, a rich landlord-poor tenant match usually means that the former is risk-neutral and the latter risk-averse under the assumption of decreasing absolute risk aversion (Arrow, 1971). In this case, a wage contract would lead to the underprovision of labour by the tenant if the landlord cannot perfectly monitor labour, and a fixed-rent contract would lead to the tenant being overexposed to risk in the absence of insurance markets (Stiglitz, 1989). Sharecropping therefore serves as a useful, albeit inefficient, compromise. This explanation for the existence of sharecropping, given its widespread acceptance in economics, is hereafter referred to as the canonical model of share tenancy. Yet many instances of reverse share tenancy (that is, a sharecropping contract between a poor landlord and a rich tenant) have been reported in the development literature. For example, it has been mentioned in the context of Bangladesh (Pearce, 1983), Eritrea (Tikabo, 2003), Ethiopia (Bezabih, 2007), India (Singh, 1989), Lesotho (Lawry, 1993), Malaysia (Pearce, 1983), the Philippines (Roumasset, 2002), and South Africa (Lyne and Thomson, 1995). The problem, however, is that such contracts do not readily fit the canonical model of share tenancy described above. Indeed, an agreement between a poor, presumably risk-averse landlord and a rich, presumably risk-neutral tenant means that the former loses her comparative advantage in risk-bearing, which now resides with the tenant, who retains his comparative advantage in labour monitoring. Consequently, one would expect to observe a fixed-rent contract whenever a poor landlord leases out her plot to a rich tenant. 4 This article presents a detailed case study of sharecropping in Lac Alaotra, Madagascar s most important rice-growing region, where reverse share tenancy accounts for one-third of land rentals. A rationale for the existence of reverse share tenancy in Lac Alaotra based on weak property rights and the legal doctrine of adverse possession is also presented and tested empirically so as to discuss the policy implications of reverse share tenancy in the region. The article thus offers three contributions to the development literature. First, it gives the oft-observed yet understudied phenomenon of reverse share tenancy the attention it deserves by showing that it is not a mere statistical aberration and by broadening development practitioners understanding of sharecropping by suggesting an explanation for its emergence that lies at the intersection of law and economics and which is strongly supported by the data This rules out crop fixed-rent contracts, in which the landlord leases out the plot in exchange for a fixed amount of output per unit of land (for example, one metric ton per hectare). 4. The canonical model can accommodate the existence of reverse share tenancy contracts, but one must be willing to assume that both landlord and tenant are risk-averse. In addition, certain mathematical requirements must be met, which are discussed further in Bellemare (2008). These requirements, however, are unlikely to hold in practice when there exist significant wealth differences between landlords and tenants, as is the case in the empirical application in this article. 5. This article does not seek to answer the question of why poorer landlords contract with richer tenants, which is unfortunately beyond the scope of the data. It takes these matches as given and seeks to determine why poor landlords who contract with richer tenants choose sharecropping contracts.

3 Madagascar: Sharecropping, Insecure Land Rights and Land Titling Policies 89 Second, and perhaps more importantly, it relates the empirical findings to recent landtitling efforts by the government, the World Bank, the International Fund for Agricultural Development (IFAD) and the Millennium Challenge Corporation (MCC) in Madagascar and explains how such a policy, although well-intentioned, may have only limited usefulness without a reform of the legal system that focuses on strengthening property rights from the bottom up, i.e., starting from the community level. Finally, it contributes to the literature on land tenancy in Madagascar by reporting the findings of a household survey, whereas studies of sharecropping there have so far been descriptive (Jarosz, 1990; 1991). The rest of the article is organised as follows. Section 2 presents a case study of sharecropping in Lac Alaotra based on descriptive statistics. In Section 3, a rationale for the existence of reverse share tenancy in Lac Alaotra relying on weak property rights and the legal doctrine of adverse possession is discussed that was motivated by field observations and preliminary conversations with landlords. 6 Section 4 uses regression analysis to show that the data strongly support the hypothesis put forward in the preceding section. In Section 5, the policy implications of the empirical findings are discussed and related to recent land-titling efforts by the government, the World Bank, IFAD, and the MCC in Madagascar. Section 6 concludes. 2 Data and descriptive statistics 2.1 Data The data used in this article come from a survey conducted in Lac Alaotra, Madagascar between March and August 2004 with the explicit aim of studying reverse share tenancy. Lac Alaotra is a lacustrine basin located 300 km to the northeast of Antananarivo, the country s capital, and constitutes the most important rice-growing region. Given that rice is the staple crop in Madagascar, where sharecropping mostly occurs on rice plots, and that the only previous studies (Jarosz, 1990; 1991) of sharecropping in Madagascar had been conducted in Lac Alaotra, the region was an obvious choice for this first formal study of reverse share tenancy. The sampling methodology was as follows. First, the six communes 7 with the highest density of sharecropping around Lac Alaotra were selected using the Cornell University/Ilo Project commune census of 2001 (Minten and Razafindraibe, 2003). Using communal records, the two villages with the highest density of sharecropping were selected within each selected commune. In each village, fifteen households known to enter sharecropping agreements and five known to enter fixed-rent agreements (either as landlords or tenants), plus five known not to lease plots in or out, were randomly selected from the sampling frame. This strategy of oversampling share tenancy was 6. Chris Barrett, who encouraged the author to pursue his dissertation research on reverse share tenancy, was the one who initially had these preliminary conversations. The author then had similar conversations with landlords, tenants, landowners, and local government officials during an exploratory, pre-fieldwork visit to Lac Alaotra. 7. Communes are administrative units inherited from the French colonial administration. They are roughly equivalent to counties in the United States.

4 90 Marc F. Bellemare adopted in an effort to increase the precision of statistical estimates pertaining to sharecropping. 8 The end result is a quasi-snowball sample (Goodman, 1961) of 300 selected households in Lac Alaotra. For each selected household, the survey collected household-, plot-, and contractlevel data along with information on agricultural production. Any randomly selected household, however, is a priori equally likely to lease plots in or out, except for the households which were chosen because they lease land neither in nor out. The survey thus also interviewed the selected households landlord and tenant households whenever feasible, 9 collecting household-level data for both types of households as well as contract-level data from landlord households and production data from tenant households. This sampling strategy was adopted so as to obtain a richer dataset of landtenancy contracts, and the end result is a survey of 1,029 plots in 12 villages in six communes around Lac Alaotra Land tenancy In order to offer as broad an overview as possible of land tenancy in Lac Alaotra, Table 1 presents descriptive statistics for selected land-tenancy indicators. More than 35% of plots are leased out in Lac Alaotra, two-thirds under a sharecropping agreement, and one-third under a fixed-rent contract. 11 More importantly, a fifth of the plots are under reverse tenancy, i.e., leased out by a landlord who is poorer than her tenant. 12 Finally, 12% of the plots in Lac Alaotra are under reverse share tenancy, meaning about one in every three plots under land tenancy. In either case, the data show that reverse share tenancy occurs in more than just a few degenerate cases. Table 1: Descriptive statistics for land tenancy Variable Mean (Std. error) Observations Plot under land tenancy indicator 0.37 (0.02) 1029 Plot under sharecropping indicator 0.24 (0.02) 1029 Plot under fixed rent indicator 0.12 (0.01) 1029 Plot under reverse tenancy indicator 0.19 (0.02) 1029 Plot under reverse share tenancy indicator 0.12 (0.01) The descriptive statistics and estimation results presented are weighted using probability weights so as to bring the sample as close as possible to a simple random sample. 9. Some non-selected households (that is, the landlords and tenants of the selected households) were impossible to survey. For example, one of the landlord households was that of the Prime Minister, and another landlord household resided in Réunion. This type of attrition explains most if not all missing observations in what follows. 10. Further information about the survey is omitted in the interest of brevity, but is available from the author upon request. 11. The sum of the proportions of plots leased out under sharecropping and of plots leased out under fixed rent does not exactly equal the proportion of plots leased out because these proportions are probabilityweighted means. 12. The precise definition of wealth used is explained in the next subsection.

5 Madagascar: Sharecropping, Insecure Land Rights and Land Titling Policies Landowners Table 2 presents descriptive landowner household descriptive statistics for all plots in the sample. The average household in Lac Alaotra is made up of about six individuals, over half of whom are between the ages of 15 and 64. Thus, the average household has a dependency ratio of 0.42, 13 meaning that for every five individuals within the average household, two are either too young or too old to work, and are thus dependent on the remaining three individuals for their subsistence. Table 2: Descriptive statistics for the landowner households Variable Mean (Std. Observations error) Household Individuals 6.17 (0.11) 1005 Individuals under (0.07) 1005 Individuals between 15 and (0.10) 1005 Individuals over (0.02) 1005 Dependency ratio 0.42 (0.01) 1005 Liquidity constraint indicator 0.34 (0.02) 963 Working capital per are a (100,000 ariary) 9.19 (1.17) 998 Assets per capita (100,000 ariary) 2.85 (0.24) 989 Income cer capita (100,000 ariary) 2.01 (0.10) 998 Imputed land holdings (ares) (26.02) 1019 Household head Age (years) (0.57) 1005 Female indicator 0.09 (0.01) 1005 Single indicator 0.15 (0.01) 1005 Sihanaka indicator 0.84 (0.02) 1029 Agricultural experience (years) (0.73) 915 Household plots (months) 7.04 (0.13) 1005 Agricultural sector (months) 0.30 (0.03) 1005 Non-agricultural sector (months) 1.59 (0.12) 1005 No work (months) 3.08 (0.14) 1005 Wage (ariary) (7.94) 949 Fady days (1.61) 1005 Commune 1 indicator 0.14 (0.01) 1029 Commune 2 indicator 0.18 (0.02) 1029 Commune 3 indicator 0.16 (0.01) 1029 Commune 4 indicator 0.16 (0.01) 1029 Commune 5 indicator 0.20 (0.02) 1029 Commune 6 indicator 0.16 (0.02) 1029 Notes: a) One are is one-hundredth of a hectare, or 100 square metres. 13. A household s dependency ratio is obtained by dividing the number of the sum of individuals under age 15 and over age 64 by the total number of individuals in the household. It thus yields the percentage of dependants within the household, that is, a rough measure of how strained the household s resources are.

6 92 Marc F. Bellemare One-third of households report being liquidity-constrained, as proxied by whether the household head (or his wife, when the household head is married) has had a loan request, either formal or informal, denied during the previous 12 months. 14 On the one hand, the average household owned about $ in working capital (i.e., hoe, harrow, cart, plough, small tractor and tractor) and a little over $140 per capita in nonproductive assets (i.e., house, radio, television, car and bank account balance). 16 On the other hand, the average household had an annual income (i.e., animal sales receipts; agricultural and non-agricultural wages; proceeds from leases of cattle and equipment; proceeds from the sales of milk, eggs, meat, and manure; and the value of household rice consumption) per capita of about $100. The average landowner is a little over 50 years old. Almost 10% of all landowners are female, and 15% are single. Thus, given that female household heads are always single, only 6% of all landowners are single males. About 85% of landowners are Sihanaka, the dominant ethnic group in Lac Alaotra. The average landowner spent over seven months working on the household s plots, about a month and a half involved in the non-agricultural sector, and three months not working. In terms of opportunity cost of time, the average landowner estimated that his time was worth $0.15 hourly, or about $1.20 per day, assuming an eight-hour workday. Finally, the average landowner was forbidden by custom (that is, fady) from doing agricultural work for about two months per year, which may inflate the time landlords reported as not working Plots Table 3 presents descriptive statistics for all plots in the sample. 18 The average plot in the data covers about 1.3 hectares and is located over half an hour s walk from the landowner s house and about a two-minute walk from a road that is accessible by cart. Over half the plots are irrigated, but the majority of these plots are irrigated by rain or by a dike. 14. This is a lower bound, given that some households almost certainly have opted out of borrowing because they expected their request to be turned down, or may have obtained loans for lesser amounts than they had requested. 15. US$1 2,000 ariary over the March-August 2004 period, when the data were collected. 16. The value of a household s land holdings is omitted for two reasons. First, due to a mistake in survey design, information on land holdings was only collected from selected households. Second, due to the thinness of land sales markets in Madagascar (Randrianarisoa and Minten, 2001), given that plots are usually inherited and sold only in distress situations, landowners often do not know the value of their plots, and can thus offer only highly inaccurate estimates for the value of their holdings. 17. Fady roughly means taboo or prohibition in English. Only days that are fady for agricultural work are reported here, but thousands of fady are observed throughout the country, which vary by ethnic group, clan, and village, pertaining to things as different as work, food, buildings, etc. For a fascinating study of the multitude of fady observed in Madagascar, see Ruud (1960). The usual fady day for the Sihanaka is Thursday (Jarosz, 1994), and fady days adversely affect the productivity of agricultural households (Stifel et al., 2007). 18. Due to a mistake in the survey design, information on certain variables (for example, how long has the landowner owned the plot, how was it obtained, and so forth) was collected from the selected households, but not from their landlords. This explains why certain variables have over 1,000 observations, while others have only 736.

7 Madagascar: Sharecropping, Insecure Land Rights and Land Titling Policies 93 Table 3: Descriptive statistics for the sample plots Variable Mean (Std. error) Observations Plot size (ares) (11.15) 1005 Distance from the house (walking minutes) (2.06) 1005 Distance from a road (walking minutes) 2.23 (0.46) 1005 Irrigated plot indicator 0.57 (0.02) 1005 Irrigated by dam indicator 0.07 (0.01) 1004 Irrigated by dike indicator 0.40 (0.02) 1004 Irrigated by spring indicator 0.11 (0.01) 1004 Irrigated by rain indicator 0.42 (0.02) 1004 Titled plot indicator 0.32 (0.02) 736 Plot ownership duration (years) (0.71) 736 Plot purchased on the land market indicator 0.37 (0.02) 736 Plot inherited from a relative indicator 0.44 (0.02) 736 Plot given by a relative indicator 0.03 (0.01) 736 Plot given by a non-relative indicator 0.01 (0.01) 736 Plot obtained through exploitation indicator 0.14 (0.02) 736 Strong claim to the land indicator 0.77 (0.02) 736 Potential adverse possession indicator 0.19 (0.02) 736 Estimated cash rent per are (ariary) (534.33) 712 Off-season agriculture possible indicator 0.47 (0.02) 1005 Off-season crop grown indicator 0.07 (0.01) 1005 Reason for lack of off-season crop Lack of water indicator 0.62 (0.02) 918 Excess water indicator 0.07 (0.01) 918 Lack of financial capital indicator 0.25 (0.02) 918 Forested plot indicator 0.03 (0.01) 918 Other reason indicator 0.04 (0.01) 918 Soil type Tanimbary indicator 0.63 (0.02) 1005 Tanety indicator 0.19 (0.02) 1005 Bas-fonds indicator 0.15 (0.02) 1005 Forest indicator 0.03 (0.01) 1005 Plot slope Flat or almost flat indicator 0.76 (0.02) 1005 Weak slope indicator 0.18 (0.02) 1005 Moderate slope indicator 0.05 (0.01) 1005 Steep indicator 0.02 (0.01) 1005 Protection structure Hedge indicator 0.03 (0.01) 1005 Windbreaker indicator 0.01 (0.00) 1005 Anti-erosion pad indicator 0.01 (0.00) 1005 Canalisation with level curves indicator 0.01 (0.00) 1005 Canalisation without level curves indicator 0.16 (0.02) 1005 No protection structure indicator 0.78 (0.02) 1005 Time to drain excess rain from plot (days) 2.46 (0.19) 1005

8 94 Marc F. Bellemare Table 3: cont d Variable Mean (Std. error) Observations Dominant soil characteristic Sand indicator 0.05 (0.01) 1004 Clay indicator 0.40 (0.02) 1004 Loam indicator 0.47 (0.02) 1004 Other indicator 0.08 (0.01) 1004 Estimated price per are (ariary) 11, (1052.8) 731 Fady days for plot (1.74) 997 Investment over the last year indicator 0.27 (0.02) 990 Plot leased out indicator 0.37 (0.02) 1029 A third of all plots in Lac Alaotra are titled, 19 and almost half were inherited from a relative. For three-quarters of these plots, landowners felt that their claim to the land was strong, but in about 20% of cases, they felt that the community or a squatter could probably obtain the land through adverse possession. 20,21 Landlords estimate the annual cash rent per are to be about $1.20, or about $160 for the average plot. This is a rough estimate, however, given the large standard deviation around the estimated annual cash rent per are. Almost half of the plots in these data allow growing an off-season crop, but less than a tenth of the plots are actually used for that purpose. In about 60% of cases, this is due to insufficient water on the plot during the off-season, and in a quarter of cases to a lack of financial capital necessary to undertake agricultural production during the offseason. About 60% of the plots in these data are tanimbary (lowland rice plots), about 20% tanety (hillside or terraced rice plots), and 15% are bas-fonds (lowland non-rice plots), with a negligible fraction being forested plots. In 95% of cases, plots were either weakly sloped or flat, and most plots had no protection structure. In case of a cyclone, it took the average plot about two and a half days to get rid of the excess water, and the soil on most plots was either clay or loam. Landowners estimated that they could get about $5.75 per are on the land sales market, or about $770 for the average plot, that is, about 60% of the average landowner s combined productive and non-productive assets. A direct implication of the complexity of the many fady observed by the Malagasy is that not only do individuals and their families observe fady days, but there are also 19. Although high by African standards, this figure is consistent with the proportion of titled plots in Madagascar. Using data from a nationally representative survey of Malagasy households conducted in 1993, Randrianarisoa and Minten (2001) estimate that 31% of plots are titled in Madagascar, whereas Deininger (2004) reports that less than 10% of lands are titled in Africa. 20. Due to the same mistake, plot-level information was collected only for the plots owned by the selected households. As a result, for the plots owned by the landlords of the selected households (that is, the plots leased in by the selected households), the data do not include information on some variables, which explains why the number of observations falls drastically for the property rights variables. 21. The legal doctrine of adverse possession is summarised as follows by Posner (2007: 78): If for a specified number of years [ ] you hold property adversely to the real owner [ ], claiming it as your own, and he does not bring suit to assert his right, the property becomes yours. See also Shavell (2004).

9 Madagascar: Sharecropping, Insecure Land Rights and Land Titling Policies 95 fady days for the plots themselves. 22 Moreover, in about 30% of cases, the landowner had made an investment in infrastructure on his plot in the last twelve months. Finally, the fact that almost 40% of plots in these data are leased out is of direct relevance to this article: given that sampling weights were used to bring the sample at hand as close as possible to a simple random sample, this means that almost 40% of plots in Lac Alaotra are under some form of land tenancy, a proportion that is both statistically and economically significant. The next section takes a closer look at the households involved in reverse tenancy and examines whether there are systematic differences between landlord and tenant households. 2.5 Households involved in reverse tenancy Table 4: Descriptive statistics for the households involved in reverse tenancy (N = 190) Variable Landlords Tenants Mean (Std. error) Mean (Std. error) Difference a Household Individuals 5.60 (0.21) 5.99 (0.23) - Individuals under (0.15) 2.84 (0.16) - Individuals between 15 and (0.15) 3.13 (0.14) - Individuals over (0.06) 0.03 (0.01) + Dependency ratio 0.45 (0.02) 0.44 (0.02) Liquidity constraint indicator 0.19 (0.03) 0.33 (0.05) - Working capital (100,000 ariary) 2.01 (0.26) 9.50 (2.36) - Assets per capita (100,000 ariary) 3.71 (0.37) (1.33) - Income per capita (100,000 ariary) 0.70 (0.11) 1.27 (0.17) - Household head Age (years) (1.32) (0.84) + Female indicator 0.16 (0.03) 0.02 (0.01) + Single indicator 0.23 (0.03) 0.11 (0.03) + Sihanaka indicator 0.93 (0.02) 0.91 (0.02) Agricultural experience (years) (1.71) (0.87) + Household plots (months) 6.33 (0.28) 7.67 (0.22) - Non-farm agricultural sector (months) 0.30 (0.06) 0.53 (0.09) - Non-agricultural sector (months) 1.80 (0.30) 1.81 (0.25) No work (months) 3.60 (0.26) 1.98 (0.16) + Wage (ariary) (11.81) (16.31) - Fady days (3.38) (4.61) Note: a) This column consists in t-tests of the difference of the mean of the full and reverse tenancy samples for each variable. As such, a plus (minus) sign indicates that the mean of the variable is significantly higher (less) in the full than in the reverse tenancy sample. Only differences significant at least at the 90% level are reported as statistically significant. 22. Whether personal- and plot-level fady coincide with one another for most landowners in these data is anybody s guess, short of a direct study of fady and their impacts on agricultural productivity that tries to tease out the effect of personal- and plot-level fady.

10 96 Marc F. Bellemare Before describing the systematic differences between landlords and tenants involved in reverse tenancy (i.e., land-tenancy contracts in which the tenant is wealthier than the landlord), one must first define wealth. In this case, wealth is defined as the sum of the values of all non-productive assets per capita. Productive assets are left out of the definition because they could bias the sample in favour of the tenants, who are more involved in agriculture than their landlords. As for the per capita measure, it was initially adopted so as not to bias these estimates in case there were systematic differences in household sizes between landlords and tenants, but it turns out that this is not necessary. Indeed, whether one defines wealth as a household s level of assets or as its assets per capita makes no difference as to which variables are systematically different between landlord and tenant households, which is consistent with there being no systematic differences in household sizes. The remainder of this article thus retains the per capita definition of wealth. Turning to Table 4, in which the systematic differences between landlords and tenants in the reverse tenancy sample are presented, tenant households are on average significantly younger than landlord households. In other words, the composition of the latter is such that they include significantly fewer individuals under 64 and more individuals over 64. But although household composition differs between landlord and tenant households, intra-household labour quality does not, as dependency ratios do not differ between landlord and tenant households. Whereas in the larger sample of land-tenancy cases, landlord households are indistinguishable from tenant households in terms of working capital, assets per capita, and income per capita, 23 in the reverse tenancy sample, all three variables are systematically lower for the landlord household, although landlord households are less liquidity-constrained on average than tenant households. Obviously, it is true by construction that landlords have fewer assets per capita in this restricted sample, but the difference is statistically significant, and the definition of wealth adopted above in no way necessarily entails that landlord households should have less working capital and a lower income per capita. It thus seems that, in the restricted sample of reverse tenancy, landlord households are indeed systematically poorer than their tenants on all accounts. This finding, combined with the fact that reverse tenancy accounts for 51% of landtenancy contracts in these data, further strengthens the empirical relevance of reverse tenancy. Of particular interest here is the fact that landlords are on average older and more likely to be single females than their tenants. In Madagascar, the conventional wisdom holds that landlords in reverse share tenancy agreements are usually single, elderly women. The regression analysis below will be helpful in determining whether this stylised widows 24 hypothesis is supported by the data. Consistent with the age difference between landlord and tenant, landlords have more agricultural experience than their tenants, but they also spend less time working in the agricultural sector, and more time not working. Accordingly, the subjective hourly wage of landlords is significantly less than that of tenants, whose higher marginal 23.Statistics for the full sample of land tenancy are not shown, but they are available from the author on request. 24. The author thanks Clive Bell for sharing this expression in conversation.

11 Madagascar: Sharecropping, Insecure Land Rights and Land Titling Policies 97 productivity of labour is consistent with the well-known life-cycle hypothesis in economics. Naturally, one might be interested in whether these systematic differences persist in the even more restricted sample of reverse share tenancy, that is, the sample of sharecropping contracts in which the landlord household has a lower level of assets per capita than the tenant household. The results of this comparison, which are not shown for the sake of brevity and because the regression analysis below offers a more complete picture, are roughly similar, except for that of the differences between landlords and tenants as regards the number of household members under 15; the fact that the head of the household is single and the numbers of months spent working in the agricultural and non-agricultural sectors are no longer statistically significant. 2.6 Contract formation Having discussed the characteristics of the landlords and tenants as well as the systematic differences of the parties involved in reverse tenancy contracts in Lac Alaotra, it may be helpful to discuss how they come to contract with one another. Looking at Table 5, in Lac Alaotra, contracts occur overwhelmingly between members of the same extended family: 62% are between kin, about 34% occur between friends, and landlords only rarely contract with tenants to whom they have been introduced solely for land-tenancy reasons. The average landlord spends 1.29 days looking for a tenant, and about 85% of landlords indicate that they have had to negotiate one or two days with their tenant before agreeing on the terms of the contract. Assuming eight-hour workdays and considering that the average landlord estimates her hourly wage to be equal to about $0.13, this means that the transactions costs incurred by landlords when looking for and negotiating with a tenant run between $1.03 and $2.05 on average, which represents between 0.6% and 1.3% of the estimate cash rental rate per plot. This rules out an Allen and Lueck (1992, 1993) transactions cost-based explanation. In most cases, the landlord s and the tenant s first offers in the negotiating process were sharecropping contracts with either a different sharing rule for the output, for the costs, or for both with respect to the contract they eventually settled upon. This is surprising, considering the almost complete lack of variation in sharing rules and costsharing arrangements in Lac Alaotra. In every sharecropping contract observed in these data but one, the output was split fifty-fifty, and there was little to no cost-sharing, which again rules out an Allen and Lueck (1992, 1993) transactions costs-based explanation. The average landlord considered contracting with about two other tenants in addition to the tenants already contracted with, and in about half of the cases where the landlord could choose between tenants, the landlord indicated choosing this tenant because of kin relations. In about 30% of cases, the landlord chose this tenant because the tenant simply seemed more honest than the others. Finally, the average landlord has contracted with fewer than one tenant in the past, the average landlord and her tenant have been contracting for over four years, and contracts are signed, on average, for a little over a year.

12 98 Marc F. Bellemare Table 5: Descriptive statistics for the land-tenancy contracts Variable Reverse tenancy sample Mean (Std. error) Link between landlord and tenant Kin 0.62 (0.05) Met tenant through relative 0.03 (0.02) Met tenant through non-relative 0.01 (0.01) Friend 0.34 (0.05) Time spent looking for a tenant (days) 1.29 (0.21) Landlord had to negotiate with tenant 0.95 (0.02) Time spent negotiating with tenant Less than a day 0.03 (0.01) One or two days 0.86 (0.04) Three days to a week 0.08 (0.03) Over a week 0.04 (0.02) Landlord first offered Sharecropping with different crop sharing 0.16 (0.03) Sharecropping with different cost sharing 0.24 (0.04) Sharecropping with different sharing of both crop 0.26 (0.04) and costs Cash rent 0.26 (0.04) Crop rent 0.08 (0.02) Immediate agreement 0.00 (0.00) Tenant first offered Sharecropping with different crop sharing 0.14 (0. 03) Sharecropping with different cost sharing 0.26 (0.04) Sharecropping with different sharing of both crop 0.27 (0.04) and costs Cash rent 0.26 (0.04) Crop rent 0.07 (0.03) Immediate agreement 0.00 (0.00) Number of other potential tenants 1.96 (0.33) Reason why this tenant was chosen if there were other potential tenants Kin 0.48 (0.07) Has more capital than landlord 0.17 (0.05) Seemed more honest than others 0.32 (0.07) In a better position to shoulder risk 0.04 (0.02) Owed him a favour 0.00 (0.00) Number of other tenants in the past 0.49 (0.10) Relationship length (years) 4.39 (0.42) Contract duration (years) 1.37 (0.16)

13 Madagascar: Sharecropping, Insecure Land Rights and Land Titling Policies 99 3 Rationale for reverse share tenancy Having shown empirically that reverse share tenancy is a common phenomenon, this leaves open the question of why we observe this form of sharecropping in Lac Alaotra. Consequently, this section discusses a possible theoretical explanation (the mathematical underpinnings of which can be found in Bellemare, 2008) for the emergence of reverse share tenancy in Lac Alaotra that is rooted in the interplay between weak property rights and the legal doctrine of adverse possession. Conditional upon having chosen to lease out her plot, the landlord must choose between a sharecropping or fixed-rent agreement. In the latter type of contract, rent can be paid in cash before the beginning of the agricultural season or after the harvest either in cash or crop. One should not mistake crop fixed rent for sharecropping: in the former case, a pre-determined amount of crop is delivered to the landlord ex post as payment (for example, one ton of rice per hectare), whereas in the latter case, a fraction of the crop is delivered to the landlord ex post as payment (for example, 50% of the crop). This seemingly esoteric difference has important implications for risk sharing: in a cash (or crop) fixed-rent contract, the tenant bears all of the production risk inherent to agriculture, whereas in a sharecropping contract, the risk is shared by the landlord and the tenant according to the percentage of the crop each party receives. 25 In Madagascar, the question of who bears this risk is all-important. Whereas under traditional property rights, direct cultivation ensures continued access to the land, local Malagasy customs are such that the party who bears the risk in agricultural production and takes possession of the output could claim the land as their own, as peasants talk about landlords who choose not to bear risk with scorn and contempt. This surprising form of the adverse possession doctrine was first mentioned by landowners in preliminary conversations, and one could liken it to use it or lose it water rights in the Western United States (Milgrom and Roberts, 1992). Formal and informal institutions co-exist in Madagascar, and this co-existence can give rise to inefficiencies. According to Le Roy (1996), because land in Madagascar is perceived as the land of the ancestors (tanindrazana), and because the living must preserve tanindrazana, plots of land cannot ever truly become private property unless they are titled. The allocation of lands is often left to the rayamandreny (village elders), who allocate plots to clan members. Tenancy contracts are for the most part signed for only one crop season, and landowners contract with new tenants every two to three years so as to reduce the effect of weak property rights: after only a few years, a tenant can declare the plot as his own, especially if he has some knowledge of the law. Finally, Le Roy confirms that sharecroppers are for the most part wealthier landowners rather than poor landless peasants, especially in Lac Alaotra. The next section empirically tests this explanation for the emergence of reverse share tenancy contracts in Lac Alaotra using 25. This article abstracts from price-risk considerations. When the price of the output and thus the value of the harvest is uncertain, a fixed-rent contract leaves the tenant with all the production and price risk, a sharecropping contract allows the landlord and the tenant to share both types of risk, and a crop fixed-rent contract leaves the tenant with all the production risk while the price risk is shared between them. As such, the absence of crop fixed-rent contracts from these data points to the relative unimportance of price risk in Lac Alaotra.

14 100 Marc F. Bellemare data on landlords subjective expectations regarding the strength of her property rights over the plot under consideration. 4 Empirical test In order to test the hypothesis that weak property rights and adverse possession cause the emergence of reverse share tenancy in these data, one must use contract-level data on the type of contract signed between the landlord and the tenant and on the landlord s perceived probability of losing her claim to the land under the observed and the alternative, hypothetical contract. A valid concern when dealing with hypotheticals is that respondents may anchor their answer to a hypothetical question with their answer a corresponding question about actual behaviour, thus causing the two sets of answers actual and hypothetical to be spuriously correlated (Kahneman et al., 1982). In order to prevent such anchoring, data on the likelihood of losing claim to the land under the observed and alternative contracts were collected three months apart. In what follows, let y i denote the contract signed between the landlord and the tenant on plot i, where y i = 1 if a sharecropping contract is observed and y i = 0 if a fixed-rent contract is observed; 26 let r i1 and r i0 denote the landlord s perceived probability of losing her claim to the land under sharecropping and fixed rent, respectively; and let x i denote a vector of all other explanatory variables (for example, landlord and tenant household characteristics, plot-level controls, and commune dummies). Furthermore, let ry = ( ri0 ri 1) denote the change in the landlord s subjective likelihood of losing her claim to the land as the contract changes. If the explanation of this section holds, r y should be positive for the average landlord. That is, the landlord s perceived likelihood of losing her claim to the land should be higher under a fixed-rent contract than under share tenancy. Thus, if the equation of interest is y i = α + βx + γ ) + i ( ri 0 ri 1 ε i (1) where α, β, and γ are parameters to be estimated, and ε is a randomly distributed error term, a test of whether weak property rights cause the emergence of share tenancy in these data is then such that H 0 : γ = 0 versus H A : γ > 0. In other words, the sharper the difference in r between a fixed-rent and a sharecropping contract, the more likely the landlord will be to choose to enter a share tenancy with her tenant. One may naturally be inclined to think that r y is endogenous to contract choice because r i1 and r i0 both depend directly on y i, but the exogeneity of r y can be understood in two ways. First, because contract choice is determined by the perceived difference in the likelihood of losing claim to the plot of land rather than by the likelihood associated with each contract, and because this perception is determined by social norms rather than by contract choice itself, then r y is not endogenous to y. Second, one can also appeal to mathematical reasoning to show that r y is exogenous in 26. The change is discrete rather than continuous because although y theoretically lies anywhere in the (0, 1] set, that is, any sharing rule is feasible in principle, there are only two observed contracts in the data, namely, sharecropping with sharing rule, or fixed rent.

15 Madagascar: Sharecropping, Insecure Land Rights and Land Titling Policies 101 Equation 1. Indeed, because there are only two types of contract, it follows that r y, the slope of asset risk as a function of contract choice, is such that r y r r(1) r(0) = = = r(1) r(0) (2) y 1 0 Since r y is the slope of r (y), that is, whether asset risk increases or decreases as a result of moving from a fixed-rent to a sharecropping contract, which in this case is the discrete change in r due to a change in y, the second derivative of r (y) with respect to y, that is, the curvature of r (y), or r yy, is undefined, which allows one to assume that r y is fully exogenous to y. Intuitively, because there are only two possible values of contract choice, which contract is chosen has no effect on the rate at which the likelihood of losing claim to the land changes as contract choice changes, simply because such a rate consequently does not exist. In other words, there are no secondorder effects of y on r, so that the direction of the change in the likelihood due to a change of contract choice is exogenous to contract choice. Given that the canonical explanation for the existence of sharecropping relies on a risk-sharing argument, proxies for the landlord s and the tenant s levels of absolute risk aversion (that is, their assets per capita, as in Laffont and Matoussi, 1995) are included in Equation 1 so as to test the risk-sharing hypothesis. In this case, in order to find support for the risk-sharing hypothesis, one would need to reject the null hypothesis H 0 : β W L = 0 and β W T = 0 versus H A : β W L > 0 and β W T < 0, where β W L ( β W T ) denotes the marginal effect of the landlord s (tenant s) wealth on contract choice. In other words, the less risk-averse the landlord (tenant), the more likely a sharecropping (fixed-rent) contract. The estimation results are presented in Table 6. Once again, sampling weights were used in the estimation so as to bring the sample as close as possible to a representative random sample, which is a necessary condition when applying the maximum likelihood estimator used here. The further the plot is from the landlord s house, the less likely she is to choose a sharecropping contract, presumably because it gets more costly to supervise sharecroppers so as to prevent them from shirking as this distance increases. In direct contradiction to the stylised widow hypothesis, female and elderly landlords are more likely to choose a fixed-rent contract rather than a sharecropping agreement, which constitutes a rejection of the conventional wisdom among the donor community in Madagascar. The longer the landlord and the tenant have been contracting together, the more likely the tenant is to be offered a sharecropping contract. This is consistent with a similar finding by Laffont and Matoussi (1995), which they explain as the landlord being better able to detect shirking by the tenant the longer they have been contracting together. As for the tenant household covariates, the bigger the tenant household, the more likely the landlord is to choose a fixed-rent contract. Thus, it seems that landlords believe that tenant households with more available labour are in a better position to shoulder risk. Finally, the weak property rights hypothesis is strongly supported by the data in the full sample, but the null hypothesis of no risk-sharing cannot be rejected. Given the low p-value of the test and the identification problem associated with such

16 102 Marc F. Bellemare tests of risk-sharing (Bellemare and Brown, 2008), however, it is preferable to err on the side of caution and remain agnostic about whether risk preferences drive contract choice in the restricted sample. 5 Policy implications The empirical results imply that a mixture of weak property rights and customarily recognised adverse possession pushes landlords into choosing sharecropping over fixed rent. By choosing a sharecropping contract, these landlords expose themselves to more production risk than would be otherwise optimal for them. The welfare loss incurred by this sub-optimal amount of risk borne is thus the premium paid by landlords in reverse share tenancy agreements in order to maintain their claim to their plot of land. In addition, it is interesting to note that the empirical findings of the previous section go against conventional wisdom, which states that landlords in reverse share tenancy agreements are often single, elderly females. In fact, the data indicate that these stylised widows, who are a priori more vulnerable, are more likely to choose a fixedrent contract which, while exposing them to no production risk, makes them more likely to lose claim to the plots that are often their only productive assets. Given the results of the previous section, the policy implications may seem obvious. That is, one may expect a simple policy aiming to assign secure, tradable property rights in other words, a land-titling policy to prevent these landlords from exposing themselves to too much risk by allowing them to choose a fixed-rent contract without having to worry about losing plots of land, especially if this policy is targeted towards single, elderly women, who are often HIV/AIDS widows, or toward individuals whose lack of collateral makes them unable to access credit. Yet, without an overhaul of the legal and judicial system in Madagascar, such a policy would have little to no effect. Indeed, what good is an official title to a plot of land when legal recourse is too costly to resort to, both in terms of time and money (Fafchamps and Minten, 2001) and when public order institutions (Platteau, 1994a and 1994b) are weak? On this reasoning, recent titling-policy interventions by the government, the World Bank, IFAD, and the MCC in Madagascar, 27 while wellintentioned, may have missed the mark altogether. This is especially true in light of Platteau (1996) noting how grossly overestimated the effects of such reforms have been relative to their high costs, and given that Jacoby and Minten (2007) find that land titles have no effect on agricultural productivity and investment in Lac Alaotra. It is thus difficult to be enthusiastic about any land-titling programme that would not be accompanied by a (presumably extremely costly) reform of the legal and judicial system. Rather, it seems as if the solution lies in a better understanding of how land is allocated and has been allocated in the past at the community level. Armed with such an 27. The government undertook an ambitious National Land Policy Programme in The MCC rapidly followed suit by signing a four-year, $110 million compact with the government that emphasises land titling as one of its objectives (MCC, 2008). Likewise, IFAD (2008) has provided the Menabe and Melaky regions of Madagascar with a $13 million loan and a $365,000 grant to be used for land reform.