What Drives the Global “Land Rush”? Rabah Arezki, Klaus Deininger, and Harris Selod We review evidence regarding the size and evolution of the "land rush" in the wake of the 2007–8 boom in agricultural commodity prices, and we study the determinants of foreign land acquisition for large-scale agricultural investment. The use of data on bilat- eral investment relationships to estimate gravity models of transnational land-intensive investments confirms the central role of agro-ecological potential as a pull factor. However, this finding contrasts the standard literature insofar as the quality of the destination country’s business climate is insignificant, and weak tenure security is asso- ciated with increased interest for investors to acquire land in the country. Policy implica- tions are discussed. JEL codes: F21, O13, Q15, Q34 After decades of stagnant or declining commodity prices when agriculture was considered a “sunset industry”, recent increases in the level and volatility of com- modity prices and a concomitant rise in the global demand for land have taken many observers by surprise. Anticipation of future commodity price spikes and a lack of alternative assets for investments during the 2008 financial crisis led to marked increases in the demand for agricultural land. Reactions to this phenome- non have been mixed. Some, including many host-country governments, welcome it as an opportunity to overcome decades of under-investment in the Rabah Arezki is a Senior Economist with the Research Department of the International Monetary Fund, and a non-resident fellow with the Brookings Institution; his email address is rarezki@imf.org. Klaus Deininger is a Lead Economist with the Development Research Group of the World Bank; his email address is kdeininger@worldbank.org. Harris Selod is a Senior Economist with the Development Research Group of the World Bank, and a CEPR affiliate; his email address is hselod@worldbank.org. The authors thank Raphael Alomar and David Cousquer for making their historical data available to us for free, Charlotte Coutand and Caroline Silverman for help with data coding, Sudhir Singh, Daniel Monchuk, and Siobhan Murray for support with analysis and GIS, Gunther Fischer and Mahendra Shah for data, Jean-Franc ¸ ois Eudeline for building the synthetic indicator of tenure security, and Thierry Mayer and Jacques Ould Aoudia for their guidance and suggestions regarding the use of data on governance. The authors are also grateful to Derek Byerlee, Will Martin, and Jo Swinnen for insightful discussions and suggestions, the participants of the 2010 AAEA and NARSC conferences, and the Annual World Bank Conference on Land and Poverty 2012 for useful comments. Financial support from CEPREMAP, the TFESSD Trust Fund, PROFOR, the Bill and Melinda Gates Foundation and the Hewlett Foundation is acknowledged. The views expressed in this paper are those of the authors and do not necessarily reflect those of the International Monetary Fund or the World Bank, its Board of Directors, or the countries they represent. A supplemental appendix to this article is available at http://wber.oxfordjournals.org/. THE WORLD BANK ECONOMIC REVIEW, VOL. 29, NO. 2, pp. 207– 233 doi:10.1093/wber/lht034 Advance Access Publication October 21, 2013 # The Author 2013. Published by Oxford University Press on behalf of the International Bank for Reconstruction and Development / THE WORLD BANK. All rights reserved. For permissions, please e-mail: journals.permissions@oup.com 207 208 THE WORLD BANK ECONOMIC REVIEW sector, create employment, and provide access to financial services and technolo- gy. Others denounce this phenomenon as a “land grab” (Pearce 2012) and point to the irony of potentially large food exports from countries that may depend on food aid. A large body of case studies highlights that many projects seem speculative, lack a sound technical and economic basis, and fail to properly consult or compensate local people. Although the nature and desirability of the impacts are subject to debate, there is broad consensus that the wave of recent land acquisitions could have far-reaching implications for long-term food secur- ity, agricultural production patterns, and global stability. Thus, a better under- standing and analysis of the factors underlying the wave of large-scale land deals is desirable. Such an analysis is relevant for a number of development issues. One issue is the debate about the most appropriate structure of agricultural production. The exceptionally large poverty elasticity of growth in smallholder agriculture (Ligon and Sadoulet 2011, Loayza and Raddatz 2010), as demonstrated in recent rapid poverty reduction in Asian economies such as China, and the fact that the major- ity of the poor still live in rural areas have led many to highlight the importance of a farm structure based on smallholders for poverty reduction (Lipton 2009, World Bank 2007). However, disillusion with the limited success of smallholder- based efforts to improve productivity in Sub-Saharan Africa and the apparent export competitiveness of “mega-farms” in Latin America and Eastern Europe during the 2007–8 global food crisis have led many observers to suggest that despite a mixed record, the acquisition of land by large operators may provide a path out of poverty and to development (Collier and Venables 2011). However, high inequality associated with the concentration of land in large farms has nega- tively affected human and economic development because large farms have used their locally dominant position to monopolize markets (Binswanger et al. 1995), subvert the provision of public goods such as education (Nugent and Robinson 2010, Vollrath 2009), undermine financial sector development (Rajan and Ramcharan 2011), and restrict political participation (Baland and Robinson 2008). Although case studies are gradually being complemented by efforts to system- atically describe the scope of large land acquisitions (Anseeuw et al. 2012b), the reported data are often taken at face value and do not face proper scrutiny (Rulli et al. 2013).1 Our paper contributes to this debate in two ways. First, we compare estimates of large transnational land-based investment from three sources. With some important caveats, the available data point toward a boom in the wake of the combination of the 2008 food price spike and financial crises, a focus of institutional investors on more mature segments of the market, and a dominant role of the state rather than private parties as the supplier of land in 1. Problems with the existing data are documented at http://ruralmodernity.wordpress.com/2012/04/ 27/the-land-matrix-much-ado-about-nothing/, http://oilforfood.info/?p=423, and http://www.china africarealstory.com/2012/04/zombie-chinese-land-grabs-in-africa.html. Arezki, Deininger, and Selod 209 Africa. At the same time, we find that though they claim to focus on very differ- ent aspects of the phenomenon (expressions of demand at the height of the com- modity boom and signed or verified deals), the quantities reported by these data sources are strikingly similar. The scope of existing databases to trace and differ- entiate distinct stages of the process of acquiring land and implementing invest- ments seems very limited, and the ability to do so more clearly is likely to be a key criterion for reliable data sources. A second issue relates to the determinants of transnational land demand. Noting that, lacking better data, such an analysis will have to be limited to demand rather than actual land transfers, we estimate both unilateral regressions for land demand (expressed by the number of projects) in a destination country and bilateral gravity models as a function of traditional demand shifters, newly developed data on potential land supply, and political and institutional variables. The results support the importance of supply (agro-ecological potential) and demand-side variables ( population density and agricultural imports). Standard investment climate variables have less of a systematic effect than does land gover- nance, which is consistently highly significant. However, counter-intuitively, we find that countries with weak tenure security and governance have been most attractive for investors, a result that is robust across a range of estimators and controls. The paper is organized as follows. Section 2 contextualizes evidence on land demand and actual land transfers by drawing on the analysis of FDI in the macro-literature to suggest a methodological approach and to outline data needs. Section 3 presents cross-sectional data on land demand, discusses the economet- ric approach, and briefly presents relevant descriptive statistics. Key econometric results and robustness checks, as discussed in section 4, support the importance of food import demand as a motivation for countries to seek out land abroad (“push factors”) and of supply in the form of agro-ecological suitability as key determinants for the choice of destination (“pull factors”). These results and checks also highlight the extent to which weak land governance seems to encour- age rather than discourage transnational demand for land. Section 5 concludes by highlighting a number of implications for policy. CONCEPTUAL FRAMEWORK Although the recent interest in large land deals implies that the agricultural eco- nomics literature analyzing this issue remains limited, methodological and sub- stantive lessons can be drawn from studies on cross-country capital and investment flows. We briefly review conceptual considerations and econometric issues. This review is followed by a discussion of ways to measure the variables that may affect the country-level supply of and demand for land, institutional factors in terms of overall investor protection and the rule of law, and, specifi- cally, the security of property rights to land. 210 THE WORLD BANK ECONOMIC REVIEW Insights from the Literature on Foreign Investment Flows A large body of empirical literature demonstrates that, with the exception of a limited number of plantation crops, the production of agricultural crops is char- acterized by constant or decreasing (once a certain minimum farm size that fully utilizes certain lumpy inputs, such as machinery or managerial capacity, has been reached) returns to scale. A key issue is that if effort cannot be observed, sal- aried workers, in contrast to family members who are residual claimants to profits, will exert optimum levels of effort only if they are subject to costly super- vision. This phenomenon would reduce the competitiveness of large farms that rely on wage-labor compared to owner-operated farms. The former may have ad- vantages in acquiring the capital needed to expand into frontier areas or over- coming market imperfections because of the absence of public goods. The history of large land deals (Byerlee and Deininger forthcoming) suggests that, tra- ditionally, large operators’ advantages remained limited and transitory unless these advantages were upheld by distortions (Binswanger et al. 1995, Deininger 2003). In the recent past, developments in crop breeding, tillage, and information technology may have facilitated the supervision of wage labor (Deininger and Byerlee 2012), possibly making large farms more competitive economically. In countries where financial systems do not work well, large farms’ ability to reduce the cost of capital by accessing international equity markets can provide them with a distinct advantage (Deininger et al. 2013). Although no cross-country studies empirically analyze large-scale foreign land acquisitions, the literature on foreign investment has explored methodologically similar issues. This literature suggests that the magnitude and distribution of capital flows to recipient countries are determined by pull and push factors (Calvo et al. 1996)2 in addition to country-specific variables, such as cultural and geographical proximity or past bilateral ties (Benassy-Quere et al. 2007, Habib and Zurawicki 2002). Gravity models relating FDI between two countries to each partner’s size, distance, and proxies for transaction cost are widely used in the literature to explain bilateral FDI (Wei 2000).3 The results are largely con- sistent with the theoretical literature on trade and capital flows (Markusen and Venables 1998), suggesting that demand and supply factors complement the sector-specific drivers of FDI, such as a desire to be close to the market or to take advantage of lower production costs (Helpman 1984, Markusen and Venables 2000). 2. Push factors (e.g., the business cycle in industrialized countries) explain the magnitude of capital flows. Pull factors relate to the domestic country characteristics (e.g., economic performance) that help explain the distribution of capital flows across potential recipient countries. 3. The OECD defines FDI as "an activity in which an investor resident in one country obtains a lasting interest in, and a significant influence on the management of, an entity resident in another country. This may involve either creating an entirely new enterprise (“greenfield” investment) or, more typically, changing the ownership of existing enterprises via mergers and acquisitions”. A takeover by a foreign firm is considered FDI if the foreign firm holds at least 10 percent of the voting rights on the board. Arezki, Deininger, and Selod 211 A key stylized fact with regard to overall investment flows, commonly referred to as the Lucas paradox (Lucas 1990), is that the volume of such flows remains well below the levels that neoclassical theory predicts would be needed to equal- ize returns to capital. This remained the case even after capital market liberaliza- tion vastly increased capital flows to developing countries (Prasad et al. 2007). Explanations focus on fundamental differences in economic structure, such as technology, missing production factors, policy, or the institutional environment, particularly sovereign risk asymmetric information, and the previous track record (Fan et al. 2009). Countries with a weak rule of law, high political or default risk, incipient financial markets, high transaction costs, or deficient gov- ernance may attract limited investment, even if they offer exceptionally high rates of return (Shleifer and Wolfenzon 2002). Cross-sectional analysis supports the key role of institutional factors to explain the magnitude and nature of capital flows toward developing and emerging economies (Alfaro et al. 2008). Panel techniques have been used to show not only that time invariant factors such as social norms, culture, geography, and trust affect foreign capital flows but also that foreign investors tend to reward policy reforms by increasing bank lending once institutional reforms have been implemented (Papaioannou 2009). These techniques also suggest that institutional variables, rather than human capital or income, are key factors underlying this relationship. Investors have different vehicles at their disposal to realize any given level of investment. A key trade-off is between the length of commitment (and the ease of withdrawing funds) and the ability to exercise managerial control (Sawant 2010). The corporate finance literature suggests that a distinguishing feature of FDI vs. portfolio investment is the control investors enjoy over their assets. Asymmetric information, agency problems, and the use of proprietary technolo- gies are likely to give rise to a preference for direct over portfolio investment (Albuquerque 2003). Greater control can alleviate the adverse consequences of the limited ability to enforce investors’ rights (Schnitzer 2002), so direct invest- ment may be preferred over other forms of investment. Thus, although weak gov- ernance may deter investments in absolute terms, the share of FDI in total capital flows is likely to be higher in countries with weak governance because investors will demand ways of investing that provide them with greater control (Hausmann et al. 2007). Empirical Approach Because we are interested in explaining the number of planned or actual invest- ments in country j by investors from country i, we use a bilateral Poisson regres- sion to model the occurrence and count of projects in an origin-destination pair. Indexing host countries by j, we let Nij denote the number of investment projects received by host country j and originating in country i. Assuming that Nij follows a Poisson distribution lij, we can write 212 THE WORLD BANK ECONOMIC REVIEW N À Á eÀlij lij ij Prob Nij ¼ : Nij ! Specifying lij as a linear function of explanatory variables Xij allows us to express the expectation of Nij conditionally on a set of explanatory variables Xij. Denoting the conditional expectation by Lij, we obtain  à Lij ¼ E Nij jXij ¼ eXij ÀBij where Xij is a row vector of explanatory variables and Bji is a column vector of corresponding coefficients. Taking logs then allows us to formulate a model that can be estimated as lij ¼ Xij :Bij where lij is the logarithm of Lij and parameter Bij is estimated by maximum likeli- hood under the assumption that different realizations of the count variable Lij (i.e., the number of investment projects) are independent from each other. As we estimate in the logarithms, the coefficients can be interpreted as elasticities or semi-elasticities (depending on the unit of the explanatory variable), and each element of the coefficient vector Bij can then be interpreted as the change in the log of the conditional expectation of the planned or actual investments made in country j by investors from country i, resulting from a marginal increase in the value of the corresponding element of Xij. In principle, Xij can be partitioned into destination characteristics (VarDestj ), origin attributes (VarOrigi), and bilat- eral variables (VarBilati,j ) characterizing the specific origin-host pair. Formally, the bilateral count model (Poisson regression) is lij ¼ Var Origi :ai þ Var Destj :bj þ Var Bilatij :gij where the variables are defined as above. In our empirical application, VarOrigi includes food dependence and the population of the country of origin, VarDestj includes a country’s amount of “available” land or the maximum potential value of agricultural production on this land, the yield gap, institutional variables (see below), and the strength of investment protection, and VarBilat i,j includes the physical distance between the two countries and the existence of a historic colonizer/colonized relationship. Large numbers of zeros and the heteroskedasticity of errors may imply that the OLS results will be biased and inconsistent. The Poisson pseudo-maximum- likelihood estimator is suggested to address this issue (Silva and Tenreyro 2006). We follow this suggestion and use tobit and zero-inflated Poisson models to check the robustness of our estimates. Arezki, Deininger, and Selod 213 Specific Determinants of Cross-Border Farmland Investment Applying the above framework to explore the determinants of interest in cross-border farmland deals, although straightforward conceptually, requires in- formation on key supply- and demand-side variables as well as institutional factors. On the supply side, we focus on the availability of high potential “uncultivated” land which is not forested, not protected, and not populated, and the “yield gap”. On the demand side, we focus on population growth and food import dependence. Regarding the institutional environment, we consider vari- ables for land governance, investor protection, and law and order. The attractiveness of a country for farmland investment depends on the avail- ability of land with high agro-ecological potential that is not yet used for inten- sive crop production. We rely on the bio-physical modeling of potential crop yields to obtain an estimate of the value of the potential output from any given piece of land, even if it is not currently cultivated.4 To avoid problems (Young 2000), we use agro-ecological potential for rainfed cultivation, as defined by the global agro-ecological zoning project (Fischer et al. 2002). Because wheat, maize sorghum, soybean, sugarcane, oil palm, and cassava account for the majority of global agricultural output and span a wide range of agro-ecological conditions, we use them as indicator crops and simulate output for each of these crops using location-specific climatic conditions. The results of this approach, with output valued at 2005 prices (i.e., pre-crisis), are then compared for each five arc-minute grid-cell of the GAEZ v3.0 resource inventory to choose the crop with the highest output value, which then defines the output value for that grid-cell. Figure A.1 in the online appendix available at http://wber.orxfordjournals.org/ maps the resulting potential value of output per ha for all grid cells. To make these data useful for our regressions, we overlay the map of potential output with information on actual land use and population density from a variety of databases.5 This approach allows us to compute a measure of land supply as the potential output value from areas that are not forested, not protected, not already used for agricultural cultivation, and that have a population density below 25 inhabitants per km2.6 The rationale is that if potentially suitable land is forested or protected, it is likely to provide social or environmental benefits that 4. Note that our approach thus excludes the consideration of potential investment to establish irrigation, which would require more intensive modeling of hydrological flows and, furthermore, would encounter issues related to riparian rights and seasonal availability of water. 5. Our measure of agricultural land outside the forest and protected areas is constructed from various databases, including Global Land Cover 2000 (http://www-gem.jrc.it/glc2000), PAGE Global Agricultural Extent (http://www.ifpri.org/dataset/pilot-analysis-global-Ecosystems-page), Global Forest Resources Assessment 2000 (http://www.fao.org/forestry/32203/en), and the World Database on Protected Areas 2009 (http://www.wdpa.org/download.aspx). Population data are from LandScan 2003 Global Population (http://www.ornl.gov/Landscan/). 6. Based on this definition, the total land for potential expansion is 445 million ha, compared to approximately 1.5 billion ha already under cultivation. Most of this land (201, 123, and 52 million ha, respectively) is in Sub-Saharan Africa, Latin America, and Eastern Europe (Deininger et al. 2011a). 214 THE WORLD BANK ECONOMIC REVIEW would make its use by investors very costly and risky, and that population density needs to be low for land to be considered potentially “available” for agri- cultural use. We also compute the notional value of potential output on all areas that are currently covered with forest. If our hypothesis is correct, we would expect the first variable, but not the second variable, to be a significant driver of land demand. Furthermore, we aggregate the value of potential output on cur- rently cultivated areas at the country level and compare it with data on actual output to obtain a measure of the “yield gap”, the difference between observed and potential yields under existing technology that can be exploited by working with existing producers without bringing new areas under cultivation. We note that, with all other things being equal, a higher yield gap should increase interest by foreign investors who are interested in quickly establishing production. The literature suggests that much of the immediate demand for land in the wake of the 2008 crisis was driven by fears of political instability because of de- pendence on volatile food imports (Woertz 2013). To account for this demand, we complement standard bilateral information on physical, cultural, or geo- political proximity (a previous colonial relationship) with information on origin countries’ populations and past net food imports. We use three indicators to explore the links between foreign land acquisition and governance. First, data on regulatory quality, i.e., law and order, from the International Country Risk Guide (ICRG, 2009) serve as a proxy of general regulatory quality.7 Second, a measure of investor protection from the “Doing Business” database provides in- formation on the firm-specific regulatory environment.8 Finally, because agricul- tural investment is more land-intensive than other FDI, land governance and land rights security are likely to be of great relevance.9 We draw on a recent cross-country database on this issue (de Crombrugghe et al. 2009) to construct an indicator of tenure security for local users by using the first component from a principal component analysis on a set of key land governance variables.10 The effect of good land governance and strong protection of property rights on a country’s attractiveness for land-intensive investment is an empirical issue. 7. The variable comprises a law sub-component assessing the strength and impartiality of the legal system and an order sub-component assessing popular observance of the law. 8. The index consists of a weighted average of indices measuring the transparency of transactions, the liability of company directors and shareholders, and the power of administrators to hold directors accountable for misconduct. Our variable is defined as the country’s percentile in the ordered distribution of ranks regarding investor protection in the Doing Business database. 9. Key relevant aspects are the clarity of land rights and the way state land is managed, disposed of, and acquired because these elements have an important impact on land tenure security. For more details on land governance, see Deininger et al. (2011b). 10. The main contributing variables are (contributions in parentheses): “land tenure security” (16 percent), “public policies addressing land rights” (15 percent), “land ownership rights security” (14 percent), “diversity of tenure situations” (11 percent), “recognition by the State of the diversity of tenure situations” (10 percent), “scarcity of land-related conflicts” (10 percent), “traditional collective use and ownership” (9 percent), and “significance of land use policies” (6 percent). This first axis captures 40 percent of variance. Low values of the index imply low levels of tenure security. Arezki, Deininger, and Selod 215 On the one hand, the long-term horizon of some agricultural production cycles, particularly for perennials, is likely to make investors reluctant to tie up large re- sources in an environment where weak governance increases dangers of conflict with local users or of opportunistic government behavior and creeping expropri- ation (Schnitzer 1999). On the other hand, inexperienced investors may find it easier to establish property rights if (land) governance is weak, especially if they believe that it is easier and more “secure” to acquire land directly from govern- ments rather than to engage in dialogue with local populations.11 D ATA ON CRO S S -BO R D E R , L A R G E -S CA L E L A N D AC Q U I S I T I O N We document problems with the data and how these problems constrain the ability to analyze the “land rush”. Because the databases include few very large deals that did not materialize, the analysis is limited to proposed projects at the country level, and any further analysis is likely to require primary data from gov- ernment registries. With this caveat, we note that interest in large-scale agricul- tural investment of the type considered here hardly existed before a very rapid peak in 2008–9. The focus was on Africa, where most proposed deals involve firms interested in acquiring land from the government rather than private parties, in marked contrast to the predominant role of funds and the prevalence of market-based land transfers in more mature environments. Global Evidence In principle, information on cross-border, large-scale land acquisitions should be from national registries that are supported by periodically updated record track- ing, economic performance, and investors’ compliance with contractual obliga- tions.12 In practice, destination countries’ limited institutional capacities and weak regulatory framework often mean that such information is not systemati- cally gathered or analyzed (Deininger et al. 2011b).13 Consequently, much of the data underpinning the conclusions in the literature on large-scale land acquisi- tions originate from secondary sources, such as press reports. To explore the data quality and conduct a descriptive analysis, we draw on three distinct data sets that purport to refer to interest in land acquisition, signed deals, and transfers verified on the ground. 11. Weak protection of property rights by the state implies a greater need for private enforcement, an issue that has often proven problematic in the past. For an interesting perspective, see the story of Jarch capital in Southern Sudan (Funk 2010). 12. Although inventory data suggest that a large share of large land acquisitions may be by domestic rather than foreign buyers, the existing databases fail to provide information on this. The implicit assumption seems to be that even a minority stake by a foreigner qualifies a deal as cross-border. 13. Reasons include the nature of a country’s land administration system (e.g., the role of chiefs in the case of Ghana), the balance between market and non-market transfers, gaps in capacity and resources with agencies often overwhelmed by unanticipated demand, and overlaps in responsibility whereby “approvals” are often given at different levels in the hierarchy or by institutions not authorized to do so. 216 THE WORLD BANK ECONOMIC REVIEW Our first dataset is based on media reports published at the height of the com- modity price boom, between October 2008 and August 2009 by the NGO GRAIN.14 In light of the limited time period covered by these data, the widely re- ported fact that only a small fraction of intended land acquisitions led to actual transfers, and the possibility that, without geo-referencing intended locations, the efforts to eliminate double counting by eliminating reports that refer to the same piece of physical land may not always have been fully successful, the use of this dataset is likely to provide an upper-bound estimate of the immediate re- sponse triggered by the 2007–8 boom. The second dataset, referred to below as A&C, is based on an algorithm of systematic automated web searches that has been used successfully as the basis for a commercial subscription service (initial- ly, to document closures of industrial plants in France and now applied to large- scale land investments).15 These data are limited to “signed deals” prior to 2012 (Alomar and Cousquer 2012). Both datasets may be biased if systematic cross- country differences in press freedom or internet access affect the reporting of deals. Our third source of data, the “land Matrix”, reports deals that have under- gone ground verification by NGOs affiliated with the International Land Coalition (Anseeuw et al. 2012a) and are thus not affected by such concerns. Differences in the variables covered across databases, together with data gaps and missing values for many of the variables, create challenges to efforts to distil simple stylized facts about the “land rush” (see table A.1 in the online appendix). For example, the size of the (proposed) land transfer is missing in approximately 57 percent of observations in the 2008–9 demand assessment. Additionally, in the land Matrix data, close to 80 percent of transfers lack information on transaction dates, making it impossible to assess whether such transfers accelerated recently. Information on the type of seller/investor and the projected amounts of investment or jobs to be created is absent virtually everywhere. More consistent data gathering with proper quality control procedures could have substantial benefits for analysts and policy makers who are currently unable to compare their country to others in terms of the “quality” and expected local benefits from such investments. Although our three sources refer to very different concepts, they provide very similar estimates of the phenomenon (table 1). With some 56 million ha in 390 projects and 54 million ha in 848 projects, respectively, A&C and the land Matrix arrive at higher estimates for the amount of land involved in signed or verified deals than the 45 million ha in 453 projects for which interest had been expressed, based on Grain data, at the height of the boom. This finding is unexpected because many intended deals are known to have never materialized or to have been imple- mented at a much smaller scale than originally envisaged (Schoneveld 2011). We also note that in all datasets, a few “mega”-projects above 1 million ha (nine with a total size of 23 million ha in Grain and seven with 24 and 12 million 14. All media reports can be accessed at www.farmlandgrab.org. 15. The ultimate goal is to continue collecting these data and to make them available to subscribers. We thank R. Alomar and D. Cousquer for kindly making historical data available to us for analysis. Arezki, Deininger, and Selod 217 T A B L E 1 . Comparing the Total and Regional Extent of Land Projects in Three Key Databases All AFR EAP ECA LAC MNA UEA 1. DEMAND IN 2008– 9 (Grain) Total area (mn ha) 45.179 27.232 8.562 4.556 3.181 1.420 0.228 Projects (#) 453 216 95 47 47 25 23 Countries affected (#) 82 35 7 12 11 6 11 Projects . 1 mn ha (#) 9 4 2 1 1 1 0 Projects . 250k ha (#) 30 15 8 6 1 0 0 Area , 1 mn ha 22.109 11.132 5.322 3.326 1.981 0.120 0.228 Area , 250k ha 7.134 3.548 1.270 0.886 1.081 0.120 0.228 2. SIGNED AFTER 2008 (A&C) Total area (mn ha) 56.223 34.202 2.528 6.482 4.121 1.527 7.363 Projects (#) 390 192 36 56 57 8 41 Countries affected (#) 67 28 5 9 11 5 9 Projects . 1 mn ha (#) 7 5 0 1 0 0 1 Projects . 250k ha (#) 34 19 2 4 4 2 3 Area , 1 mn ha 32.303 17.102 2.528 5.262 4.121 1.527 1.763 Area , 250k ha 17.344 8.393 1.704 3.390 2.818 0.327 0.711 3. GROUND VERIFIED (land Matrix) Total area (mn ha) 54.054 23.334 23.372 1.776 5.166 0.005 0.401 Projects (#) 848 439 270 18 117 1 3 Countries affected (#) 55 27 4 9 4 0 11 Projects . 1 mn ha (#) 7 2 5 0 0 0 0 Projects . 250k ha (#) 37 14 15 1 6 0 1 Area , 1 mn ha 42.715 20.354 15.013 1.776 5.166 0.005 0.401 Area , 250k ha 25.059 13.906 7.024 1.453 2.659 0.005 0.012 Note: AFR ¼ Sub-Saharan Africa, EAP ¼ East Asia and Pacific, ECA ¼ Eastern and Central Europe, LAC ¼ Latin America and Caribbean, MNA ¼ Middle East and North Africa, UEA ¼ United States, Europe, and Australia. Source: Authors’ computation from the relevant databases, as explained in the text. ha, respectively, in A&C and the land matrix) affect estimates of total area. Eliminating these reduces the estimated size of land deals, consistent with the notion that large parts of early demand may have been speculative and dominat- ed by a few very ambitious projects. Surprisingly, however, if transactions greater than 1 million ha are eliminated, the size of “ground-verified” deals, as reported by the land Matrix (42 million ha in total), amounts to almost double the demand articulated in 2008–9 (22 million ha) and significantly exceeds even the amount of supposedly signed deals during the 2008–12 period based on A&C (32 million ha). This finding suggests weaknesses in the field verification proce- dures applied by the land Matrix, although the paucity of variables reported makes it impossible to verify these systematically.16 The number of target coun- tries also varies across data sources, with 82 in the Grain data, followed by 67 in 16. Together with weak documentation and the fact that the database made available publicly is updated on a continuing basis without keeping track of previous versions, this reinforces the notion that the land Matrix seems be more of an advocacy tool than a rigorous scientific effort. Because the land Matrix does not include information on the timing of transactions, cross-checking is virtually impossible. 218 THE WORLD BANK ECONOMIC REVIEW A&C and 55 in the land Matrix. All data sources coincide in suggesting that there has been a disproportionate focus on Africa, which consistently accounts for some 50 percent of the area involved.17 Subject to caveats regarding the consistency of reporting and data quality, a comparison of A&C and Grain allows us to identify a few regularities in land in- vestment (table 2). Direct involvement by governments or SWFs appears to have been limited. In line with the literature on corporate finance, direct rather than portfolio investment predominates in Africa (where it makes up 75 percent of projects), whereas funds focus on more “mature” market segments, including North America, Europe, and Australia. Although joint ventures had limited rele- vance according to the data, the share of purchases in total acquisitions seems proportional to the level of institutional development; almost 90 percent of the reported transactions were purchases in the US, Europe, and Australia, compared to only about one third of the transactions in Africa. Africa also stands out in that, for almost 90 percent of the known cases, the “seller” is the government rather than a private party or user group, in line with the notion that in Africa, state usurpation of communal land rights is a key risk (Alden-Wily 2010). Although only one of our databases has information on the time of acquisition, it provides interesting insights (table 3). First, although there was little activity before 2008 (total transfers of only 2 million ha), the volume of reported signed deals increased to 6 million ha in 2008 and 30 million in 2009, followed by a drop to 9 and 10 million thereafter. This boom-bust cycle is more pronounced for biofuels (which account for 11 percent, 37 percent, and approximately 10 percent-15 percent of acquisitions before, during, and after 2008, respectively) and in Africa (53 percent in 2008, reduced to less than 20 percent in 2009 and less than 10 percent in 2011). Possibly as a result of limited alternative investment opportunities, the involvement of funds also peaked in 2008. Governments had not acted as buyers at all in the period before 2008, and although they were most active in 2008, their presence continued. Disaggregating country-level data for Africa indicates differences across the data- bases (table 4). The top destinations in terms of the number of investments are Ethiopia, Sudan, Mozambique, and Tanzania (22 percent, 15 percent, 13 percent, and 12 percent, respectively) for A&C; Mozambique, Ethiopia, Tanzania, and Madagascar (25 percent, 20 percent, 15 percent, and 11 percent) for the land Matrix; and Sudan, Ethiopia, Nigeria, and Ghana (19 percent, 15 percent, 11 percent, and 11 percent) for Grain. Regarding the investors’ regions of origin, all databases attribute a significant role to investors from Western Europe, who account for between 40 percent and 43 percent of total investment. The databases diverge on the rest of the investors, however, indicating that the second most impor- tant origin region is the Middle East, according to Grain (29 percent); Africa 17. Beyond the focus on Africa, databases differ regarding the relative importance of other regions; whereas Grain and the Matrix coincide in pointing toward EAP, A&C has EAP as a distant fifth after UEA, ECA, and LAC, partly because of a stronger focus on “market” transactions. Arezki, Deininger, and Selod 219 T A B L E 2 . Key Characteristics of Land Projects according to the Three Databases, All Countries and by Region All AFR EAP ECA LAC MNA UEA 1. DEMAND IN 2008– 9 (Grain) Acquisition type (percents) Lease 44.6 59.1 55.4 43.3 15.6 33.3 11.1 Purchase 46.8 32.6 23.4 56.7 81.3 66.7 83.4 Concession 8.6 8.3 21.2 0.0 3.1 0.0 5.5 Intended use (percents) Biofuels 20.8 28.0 15.9 6.7 26.2 0.0 15.8 Food 37.8 34.4 52.3 48.9 11.9 70.0 0.0 Industrial/Plantation 22.0 21.7 20.4 20.0 35.7 15.0 15.8 Livestock 19.3 15.9 11.4 24.5 26.2 15.0 68.5 Type of buyer (percents) Public agency 25.8 25.7 33.3 27.8 26.3 21.1 0.0 Private firm 36.1 44.2 30.4 24.9 28.9 26.3 31.6 Private fund 38.1 30.1 36.2 47.3 44.7 52.6 68.4 2. SIGNED AFTER 2008 (A&C) Acquisition type (percents) Purchase 51.8 28.4 29.0 78.6 75.0 50.0 87.8 Lease 43.1 66.1 54.8 19.6 21.2 50.0 9.8 Joint venture 5.1 5.4 16.1 1.8 3.8 0.0 2.4 Type of seller (percents) Seller gov’t 54.8 88.1 70.9 11.9 18.0 57.1 2.8 Seller private 45.2 11.9 29.1 88.1 82.0 42.9 97.2 Type of buyer (percents) Buyer private firm 67.3 75.3 82.8 49.1 68.4 50.0 42.5 Private fund 25.4 15.3 5.8 45.4 31.6 37.5 52.5 Gov’t/SWF 7.3 9.5 11.4 5.5 0.0 12.5 5.0 Intended use (percents) Crop biofuel 22.8 39.1 16.7 0.0 12.3 12.5 0.0 Food 58.7 48.4 63.9 94.6 66.7 75.0 39.0 Other 18.5 12.5 19.4 5.4 21.1 12.5 61.0 Investment amount Info non-missing (percent) 23.3 14.6 25.0 14.3 29.8 25.0 65.9 Investment/ha (US$) 9,071 13,910 5,699 1,602 3,283 2,069 11,000 3. GROUND VERIFIED (land Matrix) Intended use (percents) Biofuels 20.0 26.9 17.4 0.0 5.6 0.0 Food 18.5 21.1 12.6 66.7 15.7 0.0 Industry/plantation 38.0 32.0 44.9 5.6 47.2 100.0 Other 23.5 20.0 25.1 27.8 31.5 0.0 Note: AFR ¼ Sub-Saharan Africa, EAP ¼ East Asia and Pacific, ECA ¼ Eastern and Central Europe, LAC ¼ Latin America and Caribbean, MNA ¼ Middle East and North Africa, UEA ¼ United States, Europe, and Australia. Source: Authors’ computation from relevant databases, as explained in the text. Only cases with information reported are considered, i.e., “not known” is coded as missing throughout. 220 THE WORLD BANK ECONOMIC REVIEW T A B L E 3 . Time Variation in the Nature of Signed Land Deals (A&C database) Year Type Total AFR UEA EAP ECA MNA LAC Total Area total (mn ha) 56.990 34.404 7.518 2.914 6.790 1.623 3.741 Biofuel (percent) 16.9 23.1 0.0 23.9 0.0 30.8 12.7 Buyer fund (percent) 24.2 4.4 82.7 11.2 45.2 49.7 49.6 Buyer gov’t (percent) 7.8 9.5 4.0 15.5 6.0 1.2 0.0 Seller gov’t (percent) 52.5 75.0 3.8 54.9 6.3 57.5 22.3 Before Area total (mn ha) 2.047 0.859 0.154 0.386 0.308 0.096 0.243 2007 Biofuel (percent) 11.3 23.2 0.0 8.5 0.0 0.0 0.0 Buyer fund (percent) 11.4 2.3 0.0 0.0 12.0 100.0 33.0 Buyer Gov’t (percent) 0.0 0.0 0.0 0.0 0.0 0.0 0.0 Seller gov’t (percent) 45.6 39.5 96.7 86.3 0.0 100.0 6.6 2008 Area total (mn ha) 5.932 3.806 0.202 0.593 0.844 0.002 0.485 Biofuel (percent) 36.7 53.0 0.0 16.9 0.0 0.0 12.4 Buyer fund (percent) 15.6 1.8 14.0 55.0 12.4 100.0 81.7 Buyer gov’t (percent) 11.9 18.6 0.0 0.0 0.0 0.0 0.0 Seller gov’t (percent) 39.6 54.6 69.4 20.2 0.0 100.0 1.7 2009 Area total (mn ha) 30.408 20.261 5.608 0.844 2.399 0.772 0.524 Biofuel (percent) 15.4 19.2 0.0 62.2 0.0 0.0 50.9 Buyer fund (percent) 31.5 4.3 99.9 0.0 87.3 92.0 57.6 Buyer gov’t (percent) 8.1 10.5 0.0 24.4 4.2 2.6 0.0 Seller gov’t (percent) 59.9 81.4 0.0 100.0 4.2 96.1 4.8 2010 Area total (mn ha) 8.514 3.251 0.201 0.790 2.141 0.754 1.378 Biofuel (percent) 22.6 39.6 0.0 3.5 0.0 66.4 7.6 Buyer fund (percent) 22.8 4.8 45.9 0.0 37.4 0.0 64.8 Buyer gov’t (percent) 1.5 3.9 1.8 0.0 0.0 0.0 0.0 Seller gov’t (percent) 32.4 70.3 0.0 4.3 4.2 12.5 18.3 2011 Area total (mn ha) 10.090 6.226 1.353 0.302 1.099 0.000 1.111 Biofuel (percent) 6.0 8.9 0.0 3.6 0.0 0.0 3.9 Buyer fund (percent) 11.0 6.4 37.0 0.0 2.9 0.0 16.6 Buyer gov’t (percent) 11.3 4.6 22.2 81.2 28.2 0.0 0.0 Seller gov’t (percent) 56.6 74.9 0.0 89.6 21.8 0.0 48.0 Note: AFR ¼ Sub-Saharan Africa, EAP ¼ East Asia and Pacific, ECA ¼ Eastern and Central Europe, LAC ¼ Latin America and Caribbean, MNA ¼ Middle East and North Africa, UEA ¼ United States, Europe, and Australia. Source: Authors’ computation based on all signed transfers from A&C database (see description in text). Percentages are weighted by area. (27 percent), according to the land Matrix; and East Asia or North America (17 percent each), according to A&C. Descriptive Statistics for Key Dependent Variables The means of the dependent variables for “origin” or “destination” countries based on the three databases, distinguishing between whether a project is report- ed as having started in the Grain or A&C databases, are displayed in table 5. Between 0.9 percent (in A&C) and 5.7 percent (based on the land Matrix) of the country pairs share a colonial heritage, with a mean distance of 5,500 to 7,500 km between them. The size of the cultivated area is approximately double Arezki, Deininger, and Selod 221 T A B L E 4 . Distribution of Projects among Main Targeted African Countries according to the Three Different Databases Area Projects Origin of investor (mn ha) percent among main East M. N. W. No targeted countries Africa Asia East America Europe 1. DEMAND IN 2008– 9 (Grain) Total Africa 27.23 216 14.9 13.9 28.7 3.1 39.5 Ethiopia 0.81 21 15.0 27.8 5.6 44.5 0.0 22.2 Ghana 0.53 15 10.7 14.2 7.2 7.2 0.0 71.4 Madagascar 1.94 14 10.0 9.0 18.2 18.2 0.0 54.6 Mali 0.60 9 6.4 11.1 0.0 33.3 11.1 44.4 Mozambique 0.18 14 10.0 33.4 8.3 8.3 8.3 41.7 Nigeria 0.03 16 11.4 43.8 12.5 18.8 0.0 25.0 Sudan 3.88 26 18.6 7.7 15.4 69.3 3.8 3.8 Sierra Leone 0.00 3 2.1 0.0 33.3 0.0 0.0 66.7 Tanzania 1.78 12 8.6 0.0 9.1 27.3 0.0 63.6 DRC 12.87 4 2.9 25.0 25.0 0.0 25.0 25.0 Zambia 2.00 6 4.3 0.0 50.0 33.3 0.0 16.7 Other 2.64 76 9.2 15.4 23.0 3.0 49.2 2. SIGNED AFTER 2008 (A&C) Total Africa 34.20 192 10.5 17.2 15.9 16.6 39.9 Ethiopia 2.19 32 22.2 17.7 5.8 29.4 17.7 29.4 Ghana 0.68 10 6.9 0.0 10.0 20.0 20.0 50.0 Madagascar 0.74 3 2.1 0.0 33.3 0.0 0.0 66.7 Mali 0.37 11 7.6 11.1 11.1 11.1 44.4 22.2 Mozambique 1.85 18 12.5 6.3 0.0 0.0 18.8 74.9 Nigeria 0.53 5 3.5 0.0 0.0 25.0 25.0 50.0 Sudan 6.12 22 15.3 15.0 15.0 50.1 20.0 0.0 Sierra Leone 0.49 11 7.6 0.0 30.0 10.0 0.0 60.0 Tanzania 1.32 17 11.8 0.0 25.0 6.3 18.7 50.1 DRC 15.04 10 6.9 30.0 30.0 0.0 20.0 20.0 Zambia 0.09 5 3.5 20.0 0.0 20.0 0.0 60.0 Other 4.78 48 11.6 25.6 9.3 11.6 41.9 3. GROUND VERIFIED (land Matrix) Total Africa 23.33 439 26.9 7.2 11.5 11.5 42.9 Ethiopia 4.77 71 20.1 22.5 4.1 24.5 18.4 30.6 Ghana 0.67 9 2.5 11.1 0.0 11.1 11.1 66.7 Madagascar 3.78 38 10.8 22.5 9.7 6.5 9.7 51.6 Mali 0.58 27 7.6 42.9 9.5 19.0 14.3 14.3 Mozambique 1.97 89 25.2 22.0 0.0 3.1 3.1 71.9 Nigeria 0.36 20 5.7 40.0 0.0 0.0 0.0 60.0 Sudan 3.92 18 5.1 5.9 0.0 52.9 23.5 17.7 Sierra Leone 0.72 13 3.7 14.3 0.0 0.0 28.6 57.1 Tanzania 1.38 54 15.3 24.2 9.2 0.0 12.1 54.5 DRC 0.24 6 1.7 0.0 0.0 0.0 0.0 0.0 Zambia 0.27 8 2.3 33.3 0.0 16.7 0.0 50.0 Other 4.66 86 37.7 14.5 2.9 7.2 37.7 Source: Authors’ computation from the relevant databases (see description in the text). 222 THE WORLD BANK ECONOMIC REVIEW in origin countries compared to destination countries throughout. The non- forested area that, according to our criteria, could be available for expansion is between two and three times larger in destination countries compared to origin countries, although the potential output per ha is slightly lower in destination countries. With between 65 percent and 72 percent in destination countries vs. 30 percent to 36 percent in target countries, the yield gap is approximately double in the former vs. the latter. On the demand side, origin countries are net importers of food and are signifi- cantly more populous than destinations, which are less populous and are charac- terized by small net exports. The values of our governance indices consistently point toward lower protection of investors’ interests, less robust law and order, and weaker land governance in destination countries compared to origin coun- tries.18 This finding suggests that the pull of supply-side factors (i.e., ample land availability) may outweigh concerns about limited institutional capacity. The bottom panel of table 5 also highlights the share of countries targeted in each of the regions together with the average number of projects per country in each of them, indicating that, according to the Grain data, 37 percent of the countries with an average of 2.11 investments per country (or 5.7 for each country with non-zero investment) were targeted and that 31 percent (with 1.4 projects) had some activity overall. The share was almost 70 percent among African countries (with an average of 4.4 projects per country) but only 8 percent (with 0.64 projects per country) in East Asia and the Pacific. These shares are slightly lower for the A&C and land Matrix data. ECO N O M E T R I C RE S U LTS Although our analysis is limited to demand rather than actual investment, in a scenario of high commodity prices, such demand may well be realized. Unilateral and bilateral models suggest that (i) the availability of suitable but uncultivated land for expansion is a key driver of land demand; (ii) the difference between po- tential and actual yield on land already cultivated (the “yield gap”), a key predic- tor of the ability to quickly increase production, has no consistent impact; and (iii) the quality of land governance and, in some cases, law and order is highly significant throughout, suggesting that land demand has been higher where pro- tection for such rights and the security of property remain weak.19 Results of Unilateral Regressions The results of a Poisson regression with the count of large-scale land acquisition projects in the country of destination as the dependent variable are reported in 18. Institutional variables are correlated with correlation coefficients of 2 0.44 and 0.63 between land governance and investor protection and law and order, respectively. 19. The main results using Law and Order from ICRG (2009) are robust to using Rule of Law from Kauffman et al. (2009). T A B L E 5 . Descriptive Statistics for Origin and Destination in the Three Different Databases Grain A&C Land Matrix All Started All Started All Origin Dest Origin Dest Origin Dest Origin Dest Origin Dest Descriptive statistics Distance (km) 5979.87 6024.46 7403.6 6828.7 5642.09 Former colonial relation (percent of all pairs) 0.023 0.017 0.009 0.014 0.057 Supply factors Cult. area (mn ha) 40.13 21.9 43.22 25.86 46.67 17.08 43.32 24.74 34.6 15.51 Non-forest land suit. (mn ha) 3.95 13.45 5.03 14.65 4.45 16 5.58 16.65 5.42 14 Max. poss. output non -forest ($US mn) 21,474 38,009 24,866 39,894 26,401 38,911 28,691 44,434 25,300 25,312 Forest land suit. (mn ha) 10.52 18.34 12.74 21.55 14.62 17.97 17.72 25.26 14.73 12.54 Max. poss. output forest ($US mn) 43,889 78,084 51,144 84,053 58,497 76,449 67,297 95,009 12,099 12,108 Yield gap (percent) 0.35 0.68 0.32 0.65 0.3 0.68 0.3 0.67 0.36 0.72 Demand factors Total population (mn) 26300 9500 28900 10600 29600 6660 20100 7430 15900 7120 Net food exports ($US bn) 2 3.23 1.33 2 3.51 1.73 2 1.61 1.33 2 0.60 2.52 0.11 1.11 Food exports ($US bn) 12.36 4.40 14.93 5.50 15.08 3.82 18.40 6.21 15.74 3.17 Arezki, Deininger, and Selod Institutional environment Land governance 1.02 2 1.22 1.10 2 1.15 0.97 2 1.18 1.81 2 0.90 1.02 2 1.42 Weak Investor protection 58.07 88.73 59.72 86.79 59.55 81.56 45.98 86.6 54 94.78 Law and order 4.55 3.24 4.6 3.36 4.53 3.17 4.71 3.35 4.56 3.1 Share of countries targeted (percent) Total 37.4 31.1 30.6 18.9 24.3 Africa 69.1 58.2 58.2 36.4 47.3 America 26.0 24.0 22.0 14.0 22.0 Asia 46.9 34.7 24.5 18.4 24.5 Europe 16.3 14.0 23.3 7.0 4.7 Pacific 8.0 8.0 12.0 12.0 12.0 223 (Continued ) 224 THE WORLD BANK ECONOMIC REVIEW TABLE 5. Continued Grain A&C Land Matrix All Started All Started All Origin Dest Origin Dest Origin Dest Origin Dest Origin Dest Average number of projects per country Total 2.11 1.35 1.99 0.63 2.68 Africa 4.36 2.69 3.98 1.38 5.53 America 1.08 0.82 1.40 0.36 1.78 Asia 2.84 1.84 1.53 0.71 3.84 Europe 0.47 0.28 0.88 0.09 0.19 Pacific 0.64 0.36 1.56 0.28 0.20 Notes: The table shows unweighted averages of country characteristics based on a total of 215 countries. The yield gap is the difference between the per- formance that is technically achievable and the effective yield observed (Source: FAO and IIASA). For the land governance indicator (see our footnote in the text regarding its construction from IPD 2009), low values imply low levels of tenure security. For weak investor protection (constructed from Doing Business), a high value corresponds to a situation in which investors’ rights are weakly protected. For rule of law (constructed from the Worldwide Governance Indicators database), a low value characterizes a country in which governance is poor. Source: Authors’ computation from the relevant databases (see description in the text). Arezki, Deininger, and Selod 225 T A B L E 6 . Results from Unilateral Regressions of the Number of Projects in the Three Databases Grain A&C Land Matrix Total Started Total Started Max potential outp. 1.1538*** 1.1255*** 1.7200*** 0.8008*** 0.4742*** non-forest [0.186] [0.150] [0.302] [0.144] [0.123] Max potential outp. 2 0.6333*** 2 0.6404*** 2 1.0907*** 2 0.0797 0.1301 forest [0.173] [0.139] [0.277] [0.133] [0.114] Landlocked 2 0.5110*** 2 0.4004*** 2 0.9878*** 2 0.1257 2 0.0769 [0.188] [0.146] [0.320] [0.142] [0.122] Yield gap 2 0.1367 0.1328 0.4108 1.0241*** 0.1505 [0.445] [0.361] [0.669] [0.365] [0.334] Land governance 2 0.4042*** 2 0.3947*** 2 0.3972*** 2 0.2081*** 2 0.6566*** (norm.) [0.094] [0.075] [0.137] [0.076] [0.069] Law and order (norm.) 2 0.0117 2 0.0099 2 0.0360 0.2959*** 0.1382** [0.083] [0.067] [0.122] [0.065] [0.058] Weak investor 2 0.0465 2 0.1099* 2 0.2145* 0.0508 2 0.0673 protection (norm.) [0.075] [0.060] [0.110] [0.066] [0.058] Observations 97 97 97 97 97 Pseudo R-squared 0.336 0.361 0.334 0.447 0.385 With region dummies Land governance 2 0.2592** 2 0.2378*** 2 0.1598 2 0.3437*** 2 0.2928*** [0.111] [0.088] [0.162] [0.094] [0.079] Notes: The dependent variable is the number of projects reported in a country. A constant is included throughout but not reported. Source: Authors’ analysis based on data as explained in the text. table 6 for information on all projects and for projects with some activity from Grain (columns 1 and 2) and A&C (columns 3 and 4) as well as total projects from the land Matrix (column 5). The point estimate of potential output on the non-forested area is positive and significant throughout, whereas the point esti- mate for potential on forested area is negative in all but two regressions (where it lacks significance). In terms of magnitude, the coefficients for potential output on non-forest and forested area suggest that, other things equal, a 10 percent in- crease in potential output value on non-forest or forest land would increase the number of projects by 5 –19 percent or reduce it by up to 10 percent, respectively. Surprisingly, the “yield gap” is not significant throughout for the total number of projects, which is consistent with the notion that a desire to better utilize poten- tial on land that is already cultivated was not a main driver of the “land rush”. To facilitate comparison, we normalize the land governance variables to have zero mean and unit variance. The coefficient on land governance is negative and significant throughout, whereas the coefficients on other governance variables 226 THE WORLD BANK ECONOMIC REVIEW are rarely significant. This finding supports the notion that, instead of land acqui- sition being contingent on the strong protection of rights, weak tenure security for existing occupants at the country level has been associated with higher inves- tor interest in land-related investment. Prima facie, this would imply that civil society concerns about extractive or speculative motives with little concern about benefits to local populations may not be entirely misplaced. The associa- tion with land governance is large enough to be economically meaningful; a re- duction in the land governance index by one standard deviation, equivalent to the difference between Brazil and Angola, is predicted to be associated with a total number of projects that is lower by between 36 percent (land Matrix) and 18 percent (A&C) and a number of started projects that is lower by 7 percent to 16 percent. Results of Bilateral Regressions Bilateral models provide a richer way of exploring determinants of the “land rush”. Poisson regressions of the number of projects for any bilateral investor/ host pair are thus estimated (see table 7, where the coefficient of the land gover- nance indicator from an equivalent regression including regional dummies is re- ported at the bottom). We note that distance (negative effect) and a former colonial relationship (positive effect) are strong predictors of an investment rela- tionship consistently across databases. In terms of supply-side characteristics, the regressions suggest that, as in the unilateral case, higher potential output from non-forested land is associated with the higher attractiveness of a country to in- vestors. According to these results, a 10 percent increase of potentially cultivable land would be associated with an increase in the number of projects in a host country of between 6 percent and 9 percent. The value of the potential output from forest land is significant in some cases. The coefficient on the yield gap, though positive, is insignificant or of marginal significance in all regressions for the total number of projects, except those for the land Matrix and started pro- jects in A&C, where it has a positive coefficient. Low yields and the associated opportunity to catch up or even leapfrog to the technology frontier seem to have been less important in terms of increasing a country’s attractiveness as a target for land acquisition than the availability of high-potential land that is not yet under cultivation. In terms of demand factors, higher population levels and per capita food imports in origin countries are strongly positively associated with higher demand for land investment.20 This finding may indicate that a desire to acquire land may increasingly complement traditional means of dealing with imbalances in food 20. Note that we do not include a measure of overall income in our regressions. One reason is that we want to focus on the effect of some specific characteristics of the agricultural sector rather than on the effect of overall economic performance on attracting investment. Another reason is that income per capita is often seen as an outcome of institutions and governance structure (Acemoglu et al. 2001), which are already included in our regressions. As indicated in table 3 in the online appendix, the main results presented in this paper are robust to the inclusion of both income indicators and regional dummies. T A B L E 7 . Results from Bilateral Regressions of the Number of Projects according to Three Different Databases Grain A&C Land Matrix Total Started Total Started Bilateral variables Distance (log) 2 0.6758*** 2 0.5954*** 2 0.2025* 2 0.4108*** 2 0.9163*** [0.057] [0.050] [0.122] [0.056] [0.032] Former colonial relation 0.5746* 0.9642*** 1.2569*** 1.4614*** 1.5348*** [0.312] [0.214] [0.356] [0.188] [0.170] Origin country variables Value net food imports 2.8002*** 3.3382*** 3.0151*** 1.1044*** 0.3883 [0.239] [0.186] [0.381] [0.259] [0.256] Population (log) 0.7041*** 0.7669*** 0.8364*** 0.6875*** 0.6863*** [0.038] [0.031] [0.056] [0.029] [0.027] Destination country variables Landlocked 2 0.4007** 2 0.3887** 2 1.0754*** 0.0281 0.2128* [0.195] [0.156] [0.329] [0.146] [0.127] Max potential outp. non-forest 0.6740*** 0.6588*** 0.8970*** 0.6355*** 0.5723*** [0.081] [0.066] [0.128] [0.069] [0.064] Net food import value 0.1761*** 0.1035*** 0.0012 0.1224*** 0.2084*** Arezki, Deininger, and Selod [0.048] [0.037] [0.063] [0.037] [0.034] Max potential outp. forest 2 0.1644*** 2 0.1534*** 2 0.2684*** 0.0589 0.0598 [0.055] [0.044] [0.084] [0.046] [0.042] Yield gap 0.7860 0.7568* 0.5882 1.6923*** 1.1934*** [0.531] [0.432] [0.796] [0.440] [0.420] Land governance (normalized) 2 0.5066*** 2 0.4574*** 2 0.4258*** 2 0.3200*** 2 0.7597*** [0.096] [0.079] [0.144] [0.078] [0.069] Law and order (normalized) 2 0.1089 2 0.0812 2 0.0306 0.2209*** 0.0039 [0.085] [0.070] [0.122] [0.065] [0.060] Weak investor protection (normalized) 2 0.0537 2 0.1140* 2 0.2455** 0.0487 2 0.0802 [0.077] [0.063] [0.113] [0.068] [0.060] 227 (Continued ) 228 THE WORLD BANK ECONOMIC REVIEW TABLE 7. Continued Grain A&C Land Matrix Total Started Total Started 18,333 18,333 18,333 18,333 18,333 Pseudo R-squared 0.243 0.276 0.275 0.276 0.350 With region dummies Land governance 2 0.2308* 2 0.2152** 2 0.1052 2 0.3498*** 0.0018 [0.118] [0.096] [0.172] [0.099] [0.088] Notes: The dependent variable is the number of projects in a country pair. A constant is included throughout but not reported. Source: Authors’ analysis based on data, as explained in the text. Arezki, Deininger, and Selod 229 supply through markets and storage. With the exception of land governance, co- efficients on institutional variables are weakly significant at most, suggesting that even when other factors are accounted for, high levels of institutional maturity are not a precondition for large amounts of land-related investment. On the con- trary, the coefficient on host countries’ quality of land governance, which ac- counts for the extent to which local rights are recognized, is highly significant and negative. Thus, consistent with the bilateral results, weak land governance seems associated with higher attractiveness to investors at the country level. From a substantive point of view, this finding resonates with evidence that, unless well-governed institutions exist to manage these resources, resource booms may fuel rent-seeking and corruption (Bhattacharyya and Hodler 2010) instead of development (Oechslin 2010). In the context of land-related invest- ment, transparency and disclosure, a proper regulatory framework, and the lack of market mechanisms to liquidate non-performing ventures have been of partic- ular concern. Robustness Checks Methodologically, our use of the Poisson pseudo-maximum-likelihood estimator follows the literature that suggests that this estimator is the most appropriate for the case at hand (Silva and Tenreyro 2006). Others have argued that in trade/in- vestment models, large numbers of zeros may pose greater challenges than the heteroskedasticity of errors so that, under certain conditions, it may be preferable to use tobit or even OLS (Martin and Pham 2011) or modified Poisson fixed-effects estimators, such as the zero-inflated Poisson (Burger et al. 2009).21 To check the robustness of our results, we complement Poisson regressions with tobit, zero-inflated Poisson, and OLS regressions. The results, reported in table A.2 in the online appendix, are in line with previous reports, allaying fears that our findings are driven by the choice of estimator. Coefficients for the main variables of interest are comparable to those obtained previously, supporting the impor- tance of bilateral factors, such as distance or colonial relationships, supply factors linked to agro-ecological potential, and, to some extent, food exports, demand shifters such as net food imports and population, and land governance rather than investor protection or a general rule of law index as a key institution- al factor. CONCLUSION AND PO LIC Y IMPLICATIONS Higher commodity prices and concerns about food security, a history of under- investment in agriculture, and wide variation in land scarcity and productivity 21. The zero-inflated models assume the existence of two latent groups within the population: one with zero counts and one with only positive counts. They are then estimated in two steps. A first step uses a logit regression to estimate the probability that there is no bilateral investment at all, and a second step is a Poisson regression of the probability of each count for the group with a non-zero probability. 230 THE WORLD BANK ECONOMIC REVIEW across countries have considerably increased interest by investors in agricultural land. Conceptually, it seems desirable for countries subject to such interest to adopt policies that encourage “pioneer” investors but to keep out speculators (Collier and Venables 2011). However, little systematic evidence or data exist to concretize such guidance. To advance this issue, we document available data, noting that limitations allow only crude inferences of interest in the number of projects involving land acquisitions at the country level rather than actual transfers. The use of different databases allows us to discern a boom-bust cycle associated with the 2007–8 commodity price spike, a strong focus of new interest on Africa, and distinct dif- ferences in the profile of transactions across regions with much greater state in- volvement in Africa than elsewhere. However, available databases suffer from common gaps and weaknesses that will have to be addressed on a priority basis to make reliable inferences on land sizes, proposed investment volumes and job creation, business models (outgrower or nucleus, greenfield or takeover of an ex- isting farm), and implementation progress in a consistent and meaningful way. Without these inferences, it will be difficult not only to dispel the air of secrecy currently surrounding this topic but also to allow countries and investors to draw lessons from successful (and unsuccessful) experiences to develop appropriate business models and approaches over time. Combining evidence on land demand by outsiders with country-level endow- ments allows econometric analysis to identify the drivers of this demand. Beyond bilateral links (distance, cultural proximity), the potential availability of hitherto uncultivated land and a history/infrastructure of food exports are relevant, as are food import dependence and population as demand factors. The insignificance of the “yield gap” and the consistent association of weak land governance with higher investor interest are surprising but are in line with the notion that, in our study period dominated by the immediate post-2008 peak, interest may have been focused more on the acquisition of “vacant” land rather than on helping improve agricultural productivity by integrating existing producers into value chains. This notion is in line with anecdotal evidence of countries that attracted large amounts of investor interest at the peak but found it difficult to translate in- vestors’ promises into production or benefits on the ground. This finding suggests that attracting high levels of diffuse interest by players who lack familiarity with the sector may not be conducive to quickly advancing agricultural productivity for the benefit of broader development and, if it leads to tracts of land being oc- cupied without utilization, may actually be detrimental to this goal. Although our data are too weak to make inferences on actual production, our evidence implies that better land governance,22 increased transparency, and a more consistent global and national effort at monitoring could be conducive to 22. In line with international agreements (Food and Agricultural Organization of the UN 2012), these could include recognition of local rights, education of rights holders, and allowing voluntary and transparent transfers of land. Arezki, Deininger, and Selod 231 attracting capable investors in a number of ways, particularly by (i) improving the ability to identify responsible and qualified investors ex ante and to effectively negotiate with them to maximize local benefits by integrating existing producers into value chains; (ii) ensuring that land occupied by non-viable ventures can be transferred to more efficient producers quickly; (iii) allowing responsible inves- tors to distinguish themselves to reduce risk and, ideally, their cost of capital; and (iv) providing a basis for learning from experience to develop successful busi- ness models. The recent slow-down of the “land rush” provides an opportunity for countries to act on this agenda now so they can be in a better position to dis- tinguish pioneers from speculators and can thus turn mere interest into real pro- gress for the agricultural sector if and when another commodity price boom hits. POSTSCRIPT Following acceptance of this article, the Land Matrix Global Observatory has in- formed the authors of a comprehensive re-launch of their database in June 2013 which now provides improved information on data sources, dates, implementa- tion status and sizes of intended, contracted, and failed land projects. Together with updates, e.g. on the type of investor and seller, envisaged in the future, this is expected to address many of the concerns raised in this article. Readers are en- couraged to access the most recent version of the data at http://landmatrix.org/ REFERENCES Acemoglu, D., S. Johnson, and J. Robinson. 2001. “The Colonial Origins of Comparative Development: An Empirical Investigation.” American Economic Review 91 (5): 1369–401. Albuquerque, R. 2003. “The Composition of International Capital Flows: Risk Sharing Through Foreign Direct Investment.” Journal of International Economics 61 (12): 353–83. Alden-Wily, L. 2010. “Whose Land Are You Giving Away, Mr. President?” Paper Presented at the Annual Bank Conference on Land Policy and Administration, Washington, DC, April 26-27. Alfaro, L., S. Kalemh-Ozcan, and V. Volosovych. 2008. “Why Doesn’t Capital Flow from Rich to Poor Countries? An Empirical Investigation.” Review of Economics and Statistics 90 (2): 347– 68. Alomar, R., and D. Cousquer. 2012. “A Global Land Purchase Monitor.” Paper presented at the Annual Bank Conference on Land and Poverty, Washington, DC. Anseeuw, W., L. Alden Wily, L. Cotula, and M. Taylor. 2012a. “Land Rights and the Rush for Land. Findings of the Global Commercial Pressures on Land Research Project.” Rome, Italy: International Land Coalition. Anseeuw, W., W. Boche, T. Breu, M. Giger, J. Lay, P. Messerli, and K. Nolte. 2012b. “Transnational Land Deals for Agriculture in the Global South: Report Based on the Land Matrix Database.” Bern, Montpellier, Hamburg, Germany: Bern University, CDE, CIRAD, CIGA, ILC, GIZ. Baland, J.M., and J.A. Robinson. 2008. “Land and Power: Theory and Evidence from Chile.” American Economic Review 98 (5): 1737– 65. Benassy-Quere, A., M. Coupet, and T. Mayer. 2007. “Institutional Determinants of Foreign Direct Investment.” World Economy 30 (5): 764–82. Bhattacharyya, S., and R. Hodler. 2010. “Natural Resources, Democracy and Corruption.” European Economic Review 54 (4): 608–21. 232 THE WORLD BANK ECONOMIC REVIEW Binswanger, H.P., K. Deininger, and G. Feder. 1995. “Power, Distortions, Revolt and Reform in Agricultural Land Relations.” Handbook of Development Economics 3B 2659– 772. Burger, M., F. Van Oort, and G.-J. Linders. 2009. “On the Specification of the Gravity Model of Trade: Zeros, Excess Zeros and Zero-Inflated Estimation.” Spatial Economic Analysis 4 (2): 167–90. Byerlee, D., and K. Deininger. Forthcoming. “Growing Resource Scarcity and Global Farmland Investment.” Annual Review of Resource Economics. Calvo, G.A., L. Leiderman, and C.M. Reinhart. 1996. “Inflows of Capital to Developing Countries in the 1990s.” Journal of Economic Perspectives 10 (2): 123– 39. Collier, P., and A.J. Venables. 2011. “Land Deals in Africa: Pioneers and Speculators.” Discussion Paper 8644. Centre for Economic Policy Research, London. de Crombrugghe, D, K. Farla, N. Meisel, C. de Neubourg, J. Ould Aoudia, and A. Szirmai. 2009. “Institutional Profiles Database III. Presentation of the Institutional Profiles Database 2009 (IPD 2009).” Documents de Travail de la DGTPE, No.2009/14. Treasury Directorate General of the French Ministry of the Economy, Industry and Employment, Paris. Deininger, K. 2003. Land Policies for Growth and Poverty Reduction. A World Bank Policy Research Report. Oxford and New York: World Bank and Oxford University Press. Deininger, K., and D. Byerlee. 2012. “The Rise of Large Farms in Land Abundant Countries: Do They have a Future?” World Development 40 (4): 701–14. Deininger, K, D. Byerlee, J. Lindsay, A. Norton, H Selod, and M. Stickler. 2011a. Rising Global Interest in Farmland: Can it Yield Sustainable and Equitable Benefits? Washington, DC: World Bank. Deininger, K., H. Selod, and A. Burns. 2011b. Improving Governance of Land and Associated Natural Resources: The Land Governance Assessment Framework Washington, DC: World Bank. Deininger, K., D. Nizalov, and S.K. Singh. 2013. “Are Mega Farms the Future of Global Agriculture? Exploring the Farm Size-Productivity Relationship for Large Commercial Farms in Ukraine.” World Bank Policy Research Working Paper. World Bank, Washington, DC. Fan, J.P.H., R. Morck, J.C. Xu, and B. Yeung. 2009. “Institutions and Foreign Direct Investment: China versus the Rest of the World.” World Development 37 (6): 852 –65. Fischer, G., H. v. Velthuizen, M. Shah, and F. Nachtergaele. 2002. “Global Agro-ecological Assessment for Agriculture in the 21st Century: Methodology and Results.” IIASA and FAO, Laxenburg, Austria, and Rome, Italy. Food and Agricultural Organization of the UN. 2012. Voluntary Guidelines on the Responsible Governance of Tenure of Land, Fisheries and Forests in the Context of National Food Security. Rome, Italy: FAO. Funk, M. 2010. “Will Global Warming, Overpopulation, Floods, Droughts and Food Riots Make this Man Rich? Meet the New Capitalists of Chaos.” Rolling Stone 21: 62– 66. Habib, M., and L. Zurawicki. 2002. “Corruption and Foreign Direct Investment.” Journal of International Business Studies 33 (2): 291 –307. Hausmann, R., F. Sturzenegger, A. Sibert, and C. Tille. 2007. “The Missing Dark Matter in the Wealth of Nations and its Implications for Global Imbalances.” Economic Policy (51): 69–518. Helpman, E. 1984. “A Simple Theory of International-Trade with Multinational-Corporations.” Journal of Political Economy 92 (3): 451–71. Kaufmann, D., A. Kraay, and M. Mastruzzi. 2004. “Governance Matters III: Governance Indicators for 1996, 1998, 2000, and 2002.” World Bank Economic Review 18 (2): 253–87. Ligon, E., and E. Sadoulet. 2011. “Estimating the Effects of Aggregate Agricultural Growth on the Distribution of Expenditures.” CUDARE Working Paper 1115. University of California, Berkeley. Lipton, M. 2009. Land Reform in Developing Countries: Property Rights and Property Wrongs. New York: Routledge (or Taylor and Francis). Loayza, N.V., and C. Raddatz. 2010. “The Composition of Growth Matters for Poverty Alleviation.” Journal of Development Economics 93 (1): 137 –51. Arezki, Deininger, and Selod 233 Lucas, R.E. 1990. “Why Doesnt Capital Flow from Rich to Poor Countries.” American Economic Review 80 (2): 92 –96. Markusen, J.R., and A.J. Venables. 1998. “Multinational Firms and the New Trade Theory.” Journal of International Economics 46 (2): 183 –203. ———. 2000. “The Theory of Endowment, Intra-Industry and Multi-National Trade.” Journal of International Economics 52 (2): 209 –34. Martin, W., and C.S. Pham. 2011. “Estimating the Gravity Model When Zero Trade Flows are Frequent.” Policy Research Working Paper. World Bank, Washington, DC. Nugent, J.B., and J.A. Robinson. 2010. “Are Factor Endowments Fate?” Revista de Historia Economica 28 (1): 45 –82. Oechslin, M. 2010. “Government Revenues and Economic Growth in Weakly Institutionalised States.” Economic Journal 120 (545): 631–50. Papaioannou, E. 2009. “What Drives International Financial Flows? Politics, Institutions, and Other Determinants.” Journal of Development Economics 88 (269): 281. Pearce, F. 2012. The Landgrabbers: The New Fight Over Who Owns The Earth. London: Transworld Publishers. Prasad, E., K. Rogoff, S.J. Wei, and A. Kose. 2007. “Financial Globalization, Growth, and Volatility in Developing Countries." In A. Harrison, ed., Globalization and Poverty. Cambridge, MA: National Bureau of Economic Research. Rajan, R., and R. Ramcharan. 2011. “Land and Credit: A Study of the Political Economy of Banking in the United States in the Early 20th Century.” The Journal of Finance 66 (6): 1895– 931. Rulli, M.C., A. Saviori, and P. D’Odorico. 2013. “Global Land and Water Grabbing.” Proceedings of the National Academy of Sciences of the United States of America 110 (3): 892– 97. Sawant, R.J. 2010. “The Economics of Large-Scale Infrastructure FDI: The Case of Project Finance.” Journal of International Business Studies 41 (6): 1036– 55. Schnitzer, M. 1999. “Expropriation and Control Rights: A Dynamic Model of Foreign Direct Investment.” International Journal of Industrial Organization 17 (8): 1113– 37. ——— 2002. “Debt v. Foreign Direct Investment: The Impact of Sovereign Risk on the Structure of International Capital Flows.” Economica 69 (273): 41 –67. Schoneveld, G.C. 2011. “The Anatomy of Large-Scale Farmland Acquisitions in Sub-Saharan Africa." Working Paper 85. Center for International Forestry Research, Bogor, Indonesia. Shleifer, A., and D. Wolfenzon. 2002. “Investor Protection and Equity Markets.” Journal of Financial Economics 66 (1): 3 – 27. Silva, J. M.C.S., and S. Tenreyro. 2006. “The Log of Gravity.” Review of Economics and Statistics 88 (4): 641–58. Vollrath, D. 2009. “How Important are Dual Economy Effects for Aggregate Productivity?” Journal of Development Economics 88 (2): 325– 34. Wei, S.J. 2000. “Local Corruption and Global Capital Flows.” Brookings Papers on Economic Activity (2): 303– 54. Woertz, E. 2013. Oil for Food: The Global Food Crisis and the Middle East. Oxford: Oxford University Press. World Bank. 2007. Agriculture for Development: World Development Report 2008. Washington, DC: World Bank and Oxford University Press. Young, A. 2000. “How Much Spare Land Exisits?” Bulletin of the International Union of Soil Sciences 97 (1): 51 –55.