## wp18145

## Source details

**Canonical URL:** [wp18145](https://www.imf.org/-/media/files/publications/wp/2018/wp18145.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/wp/2018/wp18145.pdf.md)
- [Structured JSON version](/-/media/files/publications/wp/2018/wp18145.pdf.json)

---

### General multi-country model with land investments and agricultural trade
- World: N countries; each produces agricultural good (A) and manufactured good (M) using labor (L) and land (T). Factors available in country i: L_i and T_i.
- Factor mobility:
  - Labor: domestically mobile between sectors.
  - Capital (knowledge capital/agricultural technology): internationally mobile.
  - Land: used only in agricultural production.
- Goods differentiation:
  - Varieties differentiated by country of production; agricultural goods additionally differentiated by investor country (multinational production allowed).
  - Agricultural technologies: Z_li for investor i in host l; Z_i ≡ {Z_li, l = 1...N}; Z_ll denotes own-country technology.
- Trade costs: iceberg costs t_nl ≥ 1.
- Investment motives modeled:
  - Platform FDI: produce abroad to re-export to world markets.
  - Food-independence (vertical FDI): produce abroad to reimport to investor’s domestic market only.

### Preferences, production, and equilibrium definitions (model equations and identities preserved)
- Preferences:
  - Upper-tier utility: u_n = (Q_A_n)^γ (Q_M_n)^{1−γ}, γ ∈ ]0,1[ (equation (1)).
  - Composite goods: CES across N country-of-production varieties (equations (2)–(4)), with elasticities ε_A > 1, ε_M > 1, σ > 1.
  - Double Armington specification: differentiation by production country and investor country; σ → ∞ removes investor-country differentiation.
- Technologies and costs:
  - Manufacturing: y_M_l = B_l L_M_l; p_M_l = (1/B_l) w_l (equations (5) and (8)).
  - Agriculture (investor i in host l): y_A_li = Z_li (L_A_li^α)^α (T_A_li^{1−α})^{1−α} (equation (6)).
  - Agricultural TFP decay: Z_li = Z_i F(t_il), F decreasing in t_il (equation (7)).
  - Agricultural unit cost: c_A_li = (w_l)^α (f_l)^{1−α} / Z_li (equation (9)).
- Demand and price indices:
  - Expenditures: D_A_n = γ I_n; D_M_n = (1−γ) I_n (equations (10)–(11)).
  - Marshallian demands: X_A_nl and X_M_nl (equations (12)–(13)).
  - Price indices: P_A_n and P_M_n (equations (14)–(15)); p_A_l factory-gate price (equation (18)).
- Revenues, market potential, and factor demands:
  - V_A_l and V_M_l definitions; revenue expressions (equations (19)–(20)).
  - Outward multilateral resistance indices Ω_A_l, Ω_M_l (equations (21)–(22)).
  - Aggregate land demand: T_A_li and T_A_l (equation (25)).
  - Aggregate labor demand: L_l expression.
- Market-clearing and equilibrium:
  - National income: I_l = w_l L_l + f_l T_A_l (equation (24)).
  - Excess demand: G_L_l and G_T_l (equations (26)–(27)).
  - Balanced trade: I_l = (p_A_l Ω_A_l)^{1−ε_A} V_A + (p_M_l Ω_M_l)^{1−ε_M} V_M (equation (28)).
  - Equilibrium: vector 〈w, T_A〉 ∈ R^{2N}_{++} such that G_L_l = 0 and G_T_l = 0 for all l.
  - Existence: Proposition 1 — at least one vector 〈w, T_A〉 ∈ R^{2N}_{++} with markets for agricultural land and labour clearing in every country (proof in Appendix C).
  - Interior equilibrium assumption: T_A_l < T_l for all l = 1,...,N.

### Determinants of large-scale land acquisitions: platform FDI versus food independence (theoretical contrasts)
- Platform FDI (Section 3.1; assumption ε_A = σ finite):
  - Equation (29) (source notation): T^A_li = (1−α) (c^A_li Ω^A_l)^{1−σ} f_l V_A.
  - Proposition 2 implications:
    - T^A_li decreases in host-country remoteness Ω^A_l.
    - T^A_li decreases in host return to labor w_l and land price f_l (via c^A_li).
    - T^A_li increases in investor-country agricultural productivity Z_li.
    - Intuition: remoteness deters platform FDI because re-exporting access matters.
- Food-independence (Section 3.2; σ → ∞ and absolute investor TFP advantage):
  - Equation (30) (source notation): T^A_li = (t_il Z_li ˜Ω^A_l ˜P^A_i)^{1−ε_A} γ^I_i T^A_l V_A.
  - Proposition 3 implications:
    - T^A_li increases in host remoteness ˜Ω^A_l.
    - T^A_li increases in investor remoteness ˜P^A_i.
    - T^A_li increases in Z_li, γ^I_i, T^A_l; decreases in bilateral trade costs t_il.
    - Intuition: remote hosts attract deals if investors seek domestic food independence (lower local production costs outweigh remoteness).
- Identification strategy:
  - Opposite signs on host remoteness provide testable hypotheses distinguishing motives using bilateral land-acquisition data.

### Empirical specification (theory-motivated) — equation (31)
- Empirical regression:
  - T^A_li = γ_0 + ∑_{m=1}^{m=4} γ_{1,m} ln(DIST_li,m) + γ_2 COL_ij + γ_3 ln(T_l) + γ_4 ln(w_l) + γ_5 ln(q_l)
    + γ_6 ln(Z_i) + γ_7 ln(L_i) + γ_8 ln(Ω^A_l) + γ_9 ln(P^A_i) + ε_li
  - DIST_li,m intervals:
    - m = 1: distances below 3,000 kilometers,
    - m = 2: 3,000 km ≤ distance < 7,000 km,
    - m = 3: 7,000 km ≤ distance < 10,000 km,
    - m = 4: distance ≥ 10,000 km.
  - Variable definitions preserved exactly:
    - T^A_li: investments of country i in country l (number of deals or total land covered).
    - COL_ij: historical colonial ties dummy.
    - T_l: host-country land suitable and “available”.
    - w_l: host-country GDP per capita (proxying labor costs).
    - q_l: host-country index of recognition of preexisting land rights / tenure security.
    - Z_i: investor-country agriculture value-added per worker.
    - L_i: investor-country population size.
    - Ω^A_l and P^A_i: host remoteness from agricultural markets and investor remoteness from agricultural producers.
- Expected signs and hypotheses preserved:
  - γ_7 > 0; γ_6 > 0; γ_3 > 0; γ_4 < 0; γ_5 < 0; γ_{1,m} < 0 ∀ m.
  - Host-country remoteness γ_8: γ_8 > 0 under food independence; γ_8 < 0 under platform FDI.
  - γ_9 > 0 under food independence.

### Estimation and remoteness measurement (four approaches)
- Estimation: Poisson Pseudo-Maximum Likelihood (PPML).
- Approaches for remoteness:
  1. Baier and Bergstrand (2009) linear approximations: Ω^A_l and P^A_i constructed as GDP- or agricultural-GDP-weighted average distances (equations (32)–(33)); removes internal distance component.
  2. Two-stage structural-gravity (Anderson and Yotov (2016); Anderson et al. (2015)):
     - Stage 1: PPML estimate of bilateral agricultural trade (equation (34)) to obtain pair fixed effects ˆμ_nl.
     - Stage 2: regress exp[ˆμ_nl] on geographic determinants to predict trade-cost matrix (equation (35)); pick ε_A to recover t_nl, authors use ε_A = 4.06.
     - Use predicted trade-costs to compute Ω^A_l and ˜P^A_i.
  3. Frankel and Romer (1999) style proxy: PPML regression of bilateral agricultural trade on exogenous geographic determinants; predicted agricultural trade used as inverse remoteness proxy.
  4. Fixed-effects absorption (Redding and Venables (2004); Feenstra (2004)): PPML with investor and host fixed effects to fully absorb remoteness (cannot estimate remoteness coefficients).

### Data sources and construction (Section 5.1)
- Primary dataset: Land Matrix (as of June 2016).
  - 2,152 transnational deals negotiated between 2000 and 2016; split into 2,601 bilateral deals; after exclusions sample = 2,122 bilateral deals.
  - Deals meet four criteria: rights transfer; occurred after 2000; area ≥ 200 hectares; implies potential conversion from smallholder/community/ecosystem use to commercial use.
  - Data quality: 80% of deals based on two or more independent sources; 6% based on media reports only.
- Temporal coverage:
  - Unrestricted period: 2000–2016.
  - Main empirical period (restricted): 2006–2013 (almost 80% of deals occurred between 2006 and 2013).
- Auxiliary datasets:
  - GeoDist (Mayer and Zignago, 2011) for distance and colonial ties.
  - Land governance index: first principal component from Institutional Profiles Database (2009).
  - Regional Trade Agreements dummy: Mario Larch’s RTA Database.
  - World Development Indicators (2009): agricultural value added per worker, GDP per capita, population size.
  - CEPII TradeProd: gross food production and bilateral food trade; trade flows for food (ISIC 311) and beverages (ISIC 313) for 1995–2004.
  - Intranational trade imputation: Anderson and Yotov (2010) method.
- Construction of suitable available land (GAEZ):
  - Crop set: wheat, maize, oil palm, sugarcane, soybean.
  - Grid: 5-arc minute cells (~10km x 10km); maximum suitability index (0–100) under rainfed baseline climate (1961–1990).
  - Suitable available land: cells with max ≥ 70, excluding cultivated land, forests, protected land; expressed in 1,000 hectares.
  - World total suitable available land estimated at about 403 million hectares.
- Descriptive highlights:
  - Land Matrix (June 2016): 2,152 deals, cumulative size 58.4 million hectares, affecting 88 host countries.
  - Regional deal counts: Sub-Saharan Africa 884 deals; East Asia 611 deals; Latin-America 368 deals.
  - Distribution: 52 deals cover 40.9 percent of all land traded; ~66 percent of deals < 10,000 hectares; 51.9 percent of projects implemented; of 43.54 million hectares from implemented deals, 37.6 percent confirmed under production.
  - Fraction not implemented smaller in Sub-Saharan Africa: 37 percent.
  - Intended use: 76.5 percent linked to agricultural and biofuel projects; Table 5 intended use (All): Agriculture 62.3, Biofuels 14.2, Forestry/Industry/Other 14.3, “Green” 7.3, Unknown 1.9.
  - Suitable land (selected table entries preserved verbatim where given): total suitable available land: 403.2 1.3 208.2 130.9 24.8 22.1 0.4 7.2 8.1 (as in Table 4 entries).
- Descriptive statistics (selected):
  - projectb (2006–2013): N = 79,135; mean = 0.0208; sd = 0.5040; min = 0; max = 54.
  - distblog: N = 50,176; mean = 8.8090; sd = 0.838; min = -0.0048; max = 79.901.
  - colonyb: N = 50,176; mean = 0.0096; sd = 0.0975; min = 0; max = 1.
  - suitablenfnplog: N = 29,573; mean = 5.786; sd = 2.463; min = 1.085; max = 10.94.
  - landgovernance: N = 34,098; mean = -0.0196; sd = 2.279; min = -5.048; max = 4.194.
  - agri valueaddedolog: N = 37,645; mean = 8.412; sd = 1.556; min = 5.283; max = 11.53.
  - populationolog: N = 45,264; mean = 15.29; sd = 2.333; min = 9.19; max = 21.01.
  - log(PA i) (BB): N = 42,348; mean = 8.754; sd = 0.392; min = 6.735; max = 9.519.
  - log(ΩA l) (BB): N = 42,341; mean = 8.755; sd = 0.392; min = 6.735; max = 9.519.
  - Predicted agricultural exports (Frankel and Romer) logs: N = 63,066 / 63,067; mean = 23.59; sd = 0.430; min = 22.57; max = 24.44.

### Main empirical results (Section 6.1) — baseline PPML (number of deals 2006–2013)
- Baseline regression: dependent variable projectb (number of deals per investor-host pair, 2006–2013); PPML; N15960; r2p (column 1) = 0.404.
- Selected coefficient estimates (column (1), significance preserved):
  - dist1log -0.983***
  - dist2log -1.092***
  - dist3log -1.035***
  - dist4log -0.999***
  - colonyb 0.686**
  - agrivaolog 0.411***
  - populationolog 0.748***
  - gdpcaplog -0.559***
  - suitable nfnplog 0.346***
  - land governance 0.00483
- Key quantitative elasticities and interpretations:
  - Suitable available land: number of bilateral deals increases by 3.5 percent for a 10 percent increase in available land (non-cultivated, non-forested, non-protected).
  - Investor population: 10 percent increase in population raises expected deals by roughly 7.5 percent (populationolog 0.748***).
  - Investor agricultural productivity: 10 percent increase in agri value added per worker increases expected deals by 4.1–5.2 percent (agrivaolog coefficients across columns).
  - Distance intervals: coefficients similar across intervals (roughly −0.98 to −1.09); inclusion of FE increases absolute magnitudes substantially.
  - Colonial ties: positive and larger with FE (examples: 0.845***, 0.910***, 1.063*** in FE columns).
- Remoteness / market potential (columns (5)–(7) summarized):
  - Column (5) — Baier and Bergstrand (BB) remoteness proxies:
    - log(ΩA l) (BB) = 3.332*** (0.465).
    - log(P A i) (BB) = -0.158 (0.165) — not significant.
    - Interpretation: 10 percent increase in host ΩA l → up to 33 percent increase in expected bilateral deals.
  - Column (6) — Frankel and Romer predicted exports:
    - investor predicted exports coef: -0.505*** (0.182).
    - host predicted exports coef: -2.455*** (0.305).
    - Interpretation: predicted exports are inverse remoteness measures; negative coefficients imply bilateral investment positively associated with remoteness of both investor and host (quantified: 10 percent decrease in predicted exports of host → expected deals increase by 24.6 percent; investor analogous effect = 5.1 percent).
  - Column (7) — Anderson and Yotov (AY) inferred trade costs:
    - agritradecost log coefficient: -2.995*** (0.175).
    - log(P A i) (AY) = -0.0664 (0.155) — not significant.
    - log(ΩA l) (AY) = 3.035*** (0.295).
    - Interpretation: 10 percent increase in host ΩA l (AY) → bilateral investments increase by 30.4 percent.
- Overall empirical interpretation:
  - Consistent positive and statistically significant role for host-country remoteness in attracting large-scale land acquisitions across remoteness measures.
  - Investor-country remoteness evidence mixed; cannot unambiguously distinguish motives solely on investor remoteness.
  - Results (columns (5)–(7)) more consistent with food-independence motive than platform FDI.

### Quantitative examples and economic magnitude
- Angola example:
  - Angola deals: 19 deals (2000–2016).
  - Angola suitable land: 6.71 million hectares vs sample average 2.00 million hectares → Angola offers 236 percent more suitable land.
  - Using baseline suitable land coefficient: expected investor-host pair with Angola as host would have 2.36*0.34*100 = 80 percent more land acquisitions than average-host pair (authors’ arithmetic preserved).
  - Angola remoteness: average country GDP-weighted avg distance ≈ 6,340 km; Angola ≈ 8,554 km → 35 percent more remote.
  - Using Baier and Bergstrand remoteness coefficient: 0.35*3.32*100 = 116 percent more bilateral deals for Angola-host pair relative to average host.
- Standard-deviation thought experiment:
  - One standard-deviation increase in host suitable land → 0.346*(7580/3047)*100 = 86.1 percent increase in bilateral deals.
  - One standard-deviation increase in BB remoteness → 3.33*(2361/6790)*100 = 115.8 percent increase.
  - Using trade-cost remoteness: 3.04*(1.04/2.90)*100 = 109.0 percent increase.
  - Conclusion: host-country remoteness importance comparable to or exceeds that of land endowments.

### Overall empirical conclusions (Section 6.1)
- Large-scale land acquisitions target:
  - Countries with abundant available suitable land.
  - Countries not well integrated into global food trading systems (high host remoteness).
  - Productive, populous investor countries (investor agricultural productivity and population associated with more deals).
- Motive interpretation:
  - Evidence more consistent with food-independence (vertical FDI) than platform FDI, given positive host remoteness effects and investor characteristics.

### Robustness checks (Section 6.2)
- Robustness exercises:
  - Controls for an “Africa effect” (Africa dummy): Africa dummy highly significant; inclusion leaves qualitative results unchanged though reduces magnitude of remoteness coefficients in some specifications.
  - Restrict to strictly positive bilateral flows (N = 409): host remoteness (BB and Frankel-Romer) remains significant at 1%; trade-cost remoteness positive but not significant; elasticities smaller — isolates intensive margin.
  - Use cumulative contract size as dependent variable (intensive margin): remoteness proxies (columns (5) and (7)) remain significant; magnitudes smaller than baseline (distance/remoteness matter more for extensive margin).
  - Expand sample to 2000–2016: remoteness proxies positive and significant; magnitudes little changed.
- Trade-cost estimation (Appendix A) results (selected):
  - Distance elasticities: around 0.77 (coefficients: −0.764 (0.050), −0.781 (0.041), −0.784 (0.040), −0.755 (0.038)).
  - Common language elasticity: 0.670 (0.068).
  - Colonial ties elasticity: 1.134 (0.133).
  - Contiguity effect: −0.024 (0.116), not significant.
- Robustness implication:
  - Multiple specifications and dependent-variable definitions reaffirm host-country remoteness as a robust positive determinant of bilateral land investments, supporting food-independence interpretation.

### Appendix C — Technical Appendix (existence proof and technical results)
- Proposition 1 — existence proof overview:
  - Define G(w, TA) excess demand vector; prove five properties for G(w, TA):
    - (i) Continuity: prices and indices continuous ⇒ G continuous.
    - (ii) Homogeneity degree zero: derived from homogeneity properties of cA_li, pA_l, pM_l and indices (equations (21), (22), (14), (15)).
    - (iii) Walras’ law: 〈w, TA〉 · G(w, TA) = 0 via balanced trade identities.
    - (iv) Lower bounds: GL_l(w, TA) > −k and GT_l(w, TA) > −k where k = max(maxh Lh, maxh fh).
    - (v) Boundary divergence: sequences approaching zero in some component force divergence in max excess demand terms; verifies condition for fixed-point theorem.
  - Conclusion: existence follows from Proposition 17.C.1 of Mas-Colell et al. (1995) (Kakutani’s Fixed-Point Theorem). Properties also hold in the case of fixed sold land TA_l = TA_l with solution vector 〈w, f〉 ∈ R2n++.
- Land acquisitions under food-independence motive (technical derivation):
  - Limiting case σ → ∞ yields identities (B.1) and definitions for ̃VA_l, ̃Ω^A_l, ̃P^A_i consistent with equation forms in main text.
- Equivalence and comparability of remoteness terms:
  - If Z_li = Zi F(t_il) and F(·) multiplicative, then food-independence remoteness terms are “magnified” versions of platform-FDI remoteness terms; under special form F(t_il) = 1/φ t_il, ̃Ω^A_l = φ (Ω^A_l)^2 and log terms differ by ln φ + 2·(sum θ_i ln t_il) (ln φ absorbed by constant, factor 2 a scaling).
  - Implication: in some special cases remoteness measures may be empirically indistinguishable up to scale.

*IMF Working Paper wp18145, sections 3.1–4.2, 5.1, 6.1–6.2, Appendix C.*

### Section 7 concludes.

### wp18145 - Section 7 concludes.

### The general setting: multi-country model with land investments and agricultural trade
- World consists of N countries; each country produces an agricultural good (A) and a manufactured good (M) using labor (L) and land (T).
- In country i = 1...N, factors available in quantities L_i and T_i.
- Labor is domestically mobile (between agricultural and manufacturing sectors within each country); capital (knowledge capital/agricultural technology) is internationally mobile.
- Land is used only for agricultural production.
- Goods are differentiated by country of production; agricultural goods additionally differentiated by investor country to allow for multinational production (multinational production = investor country i producing in host country l using host-specific variety of home technology).
- Agricultural technologies Z_li used by investor country i for host country l given by vector Z_i ≡ {Z_li, l = 1...N}; Z_ll denotes own-country technology.
- Trade is subject to iceberg costs: to sell one unit in importing country n, exporter in country l ships t_nl ≥ 1 units.
- Two investment motives modeled:
  - Platform FDI: investors produce abroad to re-export to world markets (compete for access to host resources and foreign consumer markets).
  - Food-independence (vertical FDI): investors produce abroad to reimport to their own domestic market only (obtain lower local production costs; host-market access channel shut down).

### Preferences and consumption
- Representative agent in country n has Cobb-Douglas upper-tier utility:
  - u_n = (Q_A_n)^γ (Q_M_n)^{1−γ}   (equation (1)), where γ ∈ ]0,1[ is expenditure share on agricultural goods.
- Composite agricultural and manufacturing goods are CES aggregates across N country-of-production varieties:
  - Q_A_n = (∑_{l=1}^N (q_A_nl)^{(ε_A−1)/ε_A})^{ε_A/(ε_A−1)}   (equation (2)), with ε_A > 1.
  - Q_M_n = (∑_{l=1}^N (q_M_nl)^{(ε_M−1)/ε_M})^{ε_M/(ε_M−1)}   (equation (3)), with ε_M > 1.
- Agricultural variety q_A_nl itself is a CES aggregate over investor-country differentiated varieties produced in country l:
  - q_A_nl = (∑_{i=1}^N (q_A_nli)^{(σ−1)/σ})^{σ/(σ−1)}   (equation (4)), with σ > 1.
  - If σ → ∞, agricultural varieties are not differentiated by investor country.
- This is a “double Armington” specification: differentiation by country of production (where) and by investor country (by whom).

### Production technologies
- Manufacturing production: constant returns to scale using labor only:
  - y_M_l = B_l L_M_l   (equation (5)); producing one unit requires 1/B_l units of labor.
  - Manufacturer price: p_M_l = (1/B_l) w_l   (equation (8)), where w_l is wage/return to labor.
- Agricultural production by investor i in host l: Cobb-Douglas constant returns using local land T_A_li and local labor L_A_li:
  - y_A_li = Z_li (L_A_li^α)^α (T_A_li^{1−α})^{1−α}   (equation (6)), with α and 1−α output elasticities to labor and land.
- Agricultural TFP may decay multiplicatively with distance/transport costs:
  - Z_li = Z_i F(t_il)   (equation (7)), with F decreasing in transport costs t_il.
- Agricultural unit cost (investor i in host l):
  - c_A_li = (w_l)^α (f_l)^{1−α} / Z_li   (equation (9)), where f_l is price of agricultural land in country l.

### Consumer demand, price indices, and market potentials
- National income and expenditure shares:
  - Expenditures on agricultural goods: D_A_n = γ I_n   (equation (10)).
  - Expenditures on manufactured goods: D_M_n = (1−γ) I_n   (equation (11)).
- Marshallian demands:
  - X_A_nl = (p_A_nl / P_A_n)^{1−ε_A} γ I_n   (equation (12)).
  - X_M_nl = (t_nl p_M_l / P_M_n)^{1−ε_M} (1−γ) I_n   (equation (13)).
- Consumer price indices (inward multilateral resistance):
  - P_A_n ≡ (∑_{h=1}^N (p_A_nh)^{1−ε_A})^{1/(1−ε_A)}   (equation (14)).
  - P_M_n ≡ (∑_{h=1}^N (t_nh p_M_h)^{1−ε_M})^{1/(1−ε_M)}   (equation (15)).
- The agricultural price p_A_nl is a CES aggregator over investor-specific producer prices (see equations (16)–(18)); factory-gate price for agricultural sector in l:
  - p_A_l ≡ (∑_{j=1}^N (c_A_lj)^{1−σ})^{1/(1−σ)} = p_A_nl t_nl   (equation (18)).

### Revenues, outward multilateral resistance (market potential), and demand for land and labor
- Total revenues (nominal outputs) by sector and country:
  - V_A_l ≡ ∑_{n=1}^N X_A_nl and V_M_l ≡ ∑_{n=1}^N X_M_nl.
  - Global nominal outputs V_A ≡ ∑_{l=1}^N V_A_l and V_M ≡ ∑_{l=1}^N V_M_l.
- Revenue expressions:
  - V_A_l = (p_A_l Ω_A_l)^{1−ε_A} V_A   (equation (19)).
  - V_M_l = (p_M_l Ω_M_l)^{1−ε_M} V_M   (equation (20)).
- Outward multilateral resistance indices (market potential):
  - Ω_A_l ≡ [∑_{n=1}^N (t_nl P_A_n)^{1−ε_A} γ I_n / V_A]^{1/(1−ε_A)}   (equation (21)).
  - Ω_M_l ≡ [∑_{n=1}^N (t_nl P_M_n)^{1−ε_M} (1−γ) I_n / V_M]^{1/(1−ε_M)}   (equation (22)).
- Aggregate land demand in host l (sum over investor demands):
  - T_A_li = (1−α) V_A_li / f_l = (1−α) (c_A_li / p_A_l)^{1−σ} (p_A_l Ω_A_l)^{1−ε_A} V_A / f_l   (equation (25)).
  - Aggregate T_A_l = (1−α)/f_l (p_A_l Ω_A_l)^{1−ε_A} V_A ≤ T_l.
- Aggregate labor demand in country l:
  - L_l = α/w_l (p_A_l Ω_A_l)^{1−ε_A} V_A + 1/w_l (p_M_l Ω_M_l)^{1−ε_M} V_M.

### National income and market-clearing conditions
- National income:
  - I_l = w_l L_l + f_l T_A_l   (equation (24)).
- Excess demand functions (labor and land) for each country l:
  - G_L_l(w,T_A) = α/w_l (p_A_l Ω_A_l)^{1−ε_A} V_A + 1/w_l (p_M_l Ω_M_l)^{1−ε_M} V_M − L_l   (equation (26)).
  - G_T_l(w,T_A) = (1−α)/T_A_l (p_A_l Ω_A_l)^{1−ε_A} V_A − f_l   (equation (27)).
- Balanced trade (BT) condition (rearranged):
  - I_l = (p_A_l Ω_A_l)^{1−ε_A} V_A + (p_M_l Ω_M_l)^{1−ε_M} V_M   (equation (28)).
- Equilibrium defined as vector 〈w, T_A〉 ∈ R^{2N}_{++} such that G_L_l = 0 and G_T_l = 0 for all l = 1,...,N, with all price indices and multilateral resistance terms satisfying the model equations.
- Existence result: Proposition 1 asserts at least one vector 〈w, T_A〉 ∈ R^{2N}_{++} such that markets for agricultural land and labour clear in every country (G = 0). (Proof referenced to Appendix C.)
- Assumption for interior equilibrium: all countries own sufficient quantity of land so that T_A_l < T_l for all l = 1,...,N.

### Determinants of large-scale land acquisitions: platform FDI versus food independence
- Two alternative calibrated versions of the general framework produce distinct bilateral land investment (gravity-style) equations:
  - Platform FDI model (Section 3.1): investors produce abroad to export/re-export to other countries; host-country remoteness deters bilateral investment (because re-exporting and access to foreign markets is important).
  - Food-independence / vertical FDI model (Section 3.2): investors produce abroad to serve their own domestic market only (host-market access channel shut down); host-country remoteness attracts bilateral investment (because lower local production costs matter more than proximity to re-export markets).
- Comparative insight:
  - By comparing the gravity equations derived from the two models, testable hypotheses are formulated to distinguish evidence of food-independence-driven FDI versus platform-driven FDI using bilateral data on cross-border land acquisitions, land availability, land governance, and other determinants.
- Modeling implication:
  - Host-country remoteness plays opposite roles in the two models as a determinant of FDI patterns: deterring investment under platform FDI, attracting investment under food-independence FDI.

*Source: wp18145 - Section 7 concludes.*

### 3.1    Land acquisitions as platform-driven investments

### wp18145 - 3.1    Land acquisitions as platform-driven investments

### Platform FDI model (section 3.1)
- Assumptions:
  - Follow Ramondo and Rodŕıguez-Clare (2013): ε_A = σ and σ finite.
  - Under this simplifying assumption, equation (25) can be rewritten as equation (29):
    - T^A_li = (1−α) (c^A_li Ω^A_l)^{1−σ} f_l V_A  (equation (29) in source notation)
- Proposition 2 (implications):
  - Under the platform FDI motive, the quantity of land leased by investor country i in host country l is:
    - decreasing in host-country remoteness from agricultural consumer markets Ω^A_l,
    - decreasing in the return to labor w_l and land price f_l in host-country l (via c^A_li),
    - increasing in the agricultural productivity of the investor-country Z_li (via c^A_li).
  - The relationship between bilateral investment and host-country remoteness: investors tend to invest less in countries remote from large agricultural consumer markets because remoteness implies lower aggregate demand for their product.

### Food-independence motive (section 3.2)
- Alternative assumptions to shut down platform/horizontal mechanism:
  - Varieties are no longer differentiated by investor country (σ → ∞), so in every host country only one investor country produces (the most efficient).
  - Each investor country holds an absolute (agricultural TFP) advantage to produce abroad and reimport to its own country over all other investors (mathematically: t_nln / Z_nln < t_nli / Z_nli, for all n,l,i ≠ n, using notation Z_nli and t_nli).
- Resulting bilateral investment specification (equation (30)):
  - T^A_li = (t_il Z_li ˜Ω^A_l ˜P^A_i)^{1−ε_A} γ^I_i T^A_l V_A  (equation (30) in source notation)
  - Definitions redefined in text:
    - ˜Ω^A_l ≡ (∑_{i=1}^N (t_il Z_li ˜P^A_i)^{1−ε_A} γ^I_i V_A) ,
    - ˜P^A_i ≡ (∑_{l=1}^N (t_il Z_li ˜Ω^A_l)^{1−ε_A} ˜V^A_l V_A) ,
    - ˜V^A_l ≡ (w_l^α f_l^{1−α} ˜Ω^A_l)^{1−ε_A} V_A.
- Proposition 3 (implications):
  - Under the food independence motive, the quantity of land leased by investor country i in host country l is:
    - increasing in host-country remoteness from agricultural consumers ˜Ω^A_l,
    - increasing in investor-country remoteness from agricultural producers ˜P^A_i,
    - increasing in agricultural productivity of the investor-country Z_li,
    - increasing in expenditures on agricultural goods γ^I_i,
    - increasing in agricultural land T^A_l,
    - decreasing in bilateral trade costs t_il.
- Economic intuition:
  - Under food independence, market potential of destination is a negative determinant of land acquisitions (remote hosts → more deals), reversing the sign under platform FDI.
  - Conditional on available agricultural land, country pairs with remote hosts and remote investors feature more deals (analogy to importer/exporter remoteness in trade models).

### Differentiating motives empirically (section 3.3)
- Gravity-form expressions summarized (Table 1 in source):
  - Platform: T^A_li = (1−α) (c^A_li Ω^A_l)^{1−σ} f_l V_A
  - Food independence: T^A_li = (t^A_il Z_li ˜Ω^A_l ˜P^A_i)^{1−ε_A} γ^I_i T^A_l V_A
- Empirical implications for identification:
  - Findings suggestive of platform FDI if:
    - bilateral investment decreases in host-country remoteness,
    - bilateral investment does not depend on investor-country GDP nor investor-country remoteness from agricultural producers.
  - Findings suggestive of food-independence FDI if:
    - bilateral investment increases in investor-country GDP and investor-country remoteness from agricultural producers,
    - host-country remoteness from agricultural consumers stimulates investments.
- Note on remoteness terms:
  - Remoteness terms differ between models (t_nl vs t_il Z_li) because platform FDI investors do not compete directly, while food-independence investors do.
  - Z_li can be multiplicative in investor-country agricultural productivity and a bilateral distance term; thus ˜Ω^A_l and ˜P^A_i are generally “magnified” versions of Ω^A_l and P^A_i.
  - If F(t_il) = 1/φ t_il in distance decay function (7), empirical proxy for remoteness will be the same for both models — authors will use empirical proxies of Ω^A_l as defined in the platform-FDI model.

### Econometric specification (section 4.1)
- Theory-motivated empirical specification (equation (31)):
  - T^A_li = γ_0 + ∑_{m=1}^{m=4} γ_{1,m} ln(DIST_li,m) + γ_2 COL_ij + γ_3 ln(T_l) + γ_4 ln(w_l) + γ_5 ln(q_l)
    + γ_6 ln(Z_i) + γ_7 ln(L_i) + γ_8 ln(Ω^A_l) + γ_9 ln(P^A_i) + ε_li
  - Variable definitions:
    - T^A_li: investments of country i in country l (number of deals or total land covered).
    - DIST_li,m: physical distance between i and l if it falls in interval m, zero otherwise; intervals:
      - m = 1: distances below 3,000 kilometers,
      - m = 2: between 3,000 km (included) and 7,000 km,
      - m = 3: between 7,000 km (included) and 10,000 km,
      - m = 4: equal to or greater than 10,000 km.
    - COL_ij: dummy for historical colonial ties.
    - T_l: host-country land suitable and “available”.
    - w_l: host-country GDP per capita (proxying labor costs).
    - q_l: host-country index of recognition of preexisting land rights / tenure security (from Arezki et al. (2015)).
    - Z_i: investor-country agriculture value-added per worker (agricultural TFP proxy).
    - L_i: investor-country population size (proxying expenditures on agricultural consumption).
    - Ω^A_l and P^A_i: host-country remoteness from agricultural markets and investor-country remoteness from agricultural producers respectively.
- Expected signs and hypotheses (model-specific):
  - γ_7 > 0 (populous countries acquire more land).
  - γ_6 > 0 (more agriculturally productive investors acquire more land).
  - γ_3 > 0 (investors target countries with abundant suitable agricultural land).
  - γ_4 < 0 (cheaper labor costs attract investment).
  - γ_5 < 0 (protected preexisting land rights make investments more difficult).
  - γ_{1,m} < 0 for all m (greater distance reduces investment).
  - Host-country remoteness γ_8:
    - γ_8 > 0 under food independence motive,
    - γ_8 < 0 under platform FDI.
  - γ_9 > 0 under food independence (investor-country remoteness from producers increases outward investment).

### Estimation strategy and remoteness measurement (section 4.2)
- Estimation method:
  - Use Poisson Pseudo-Maximum Likelihood (PPML) to estimate equation (31), robust to measurement error, heteroskedasticity, and zeros.
- Four approaches to construct or account for remoteness:
  1. Baier and Bergstrand (2009) linear approximations:
     - Host-country remoteness Ω^A_l and investor remoteness P^A_i computed as GDP- or agricultural-GDP-weighted average distance to other countries:
       - Ω^A_l = ∏_{n=1,n≠l}^N t_{nl}^{θ_n}  (equation (32) in source notation)
       - P^A_i = ∏_{l=1,l≠i}^N t_{il}^{ξ_l}  (equation (33) in source notation)
     - t_nl (t_il) = physical distance; θ_n = importer’s share of global GDP; ξ_l = exporter’s share of global agricultural production.
     - Eliminates first-order endogeneity by removing internal distance component t_{ll}^{θ}.
     - Advantage: uses observable distance, GDP, agricultural GDP. Disadvantage: would prefer direct trade-cost measures instead of physical distance.
  2. Two-stage structural-gravity approach (Anderson and Yotov (2016); Anderson et al. (2015)):
     - First stage: estimate dynamic panel bilateral agricultural trade equation in multiplicative form using PPML to obtain estimates of bilateral fixed effects ˆμ_nl for observed imports:
       - X^A_nl,t = exp[π_l,t + χ_n,t + μ_nl + β_1 RTA_nl,t] × ε_nl,t  (equation (34) in source notation)
       - π_l,t = producer/exporter time-varying fixed effect; χ_n,t = consumer/importer time-varying fixed effect; μ_nl = pair fixed effect; RTA_nl,t = regional trade agreement dummy.
     - Second stage: regress exp[ˆμ_nl] on determinants of trade costs and geographic dummies to predict full trade-cost matrix:
       - exp[ˆμ_nl] = exp[π_l + χ_n + ∑_{m=1}^{4} η_{1,m} ln(DIST_nl,m) + η_2 CONTIG_nl + η_3 COMLANG_nl + η_4 COL_nl] × ε_nl  (equation (35) in source notation)
     - Predicted values yield ˆt_{nl}^{1−ε_A}; recover ˆt_nl by choosing a value for ε_A.
     - Trade elasticity choice for agriculture:
       - Authors follow Tombe (2015) and pick ε_A = 4.06.
       - Contextual numbers: Costinot et al. (2016) obtain 5.4 and 2.82 for crop-level trade elasticity and elasticity across different crops respectively.
     - Use resulting bilateral trade-cost matrix to calculate multilateral resistance indices Ω^A_l and ˜P^A_i (using Baier and Bergstrand approximations but substituting trade costs for physical distance).
  3. Frankel and Romer (1999) style proxy:
     - Run a PPML regression of bilateral agricultural trade on exogenous geographic determinants DIST_nl, CONTIG_nl, COMLANG_nl, COL_nl.
     - Use predicted agricultural trade from that regression as a proxy for the inverse of remoteness.
  4. Fixed-effects absorption (Redding and Venables (2004); Feenstra (2004)):
     - Run PPML regressions with investor-country and host-country fixed effects that fully absorb remoteness terms.
     - Limitation: cannot obtain coefficient estimates for remoteness effects when remoteness is absorbed by fixed effects.

*Italic source attribution: IMF Working Paper wp18145, sections 3.1–4.2*

### 5.1    Sources

### 5.1 Sources

### Primary dataset: Land Matrix (as of June 2016)
- Uses the Land Matrix online database, containing ground-verified information on land deals by NGOs affiliated with the International Land Coalition.
- The 2016 version used is an update compared to the 2011 version previously used in Arezki et al. (2015).
- The data include information on 2,152 transnational deals negotiated between 2000 and 2016, including the origin country of the investor and the total area covered by the investment.
- For deals involving multiple locations or multiple investor countries, deals are split into subprojects, yielding a sample of 2,601 bilateral land deals.
- After removing projects outside agriculture or biofuels (tourism, industry, other renewable energy), the sample is 2,122 bilateral deals.
- Deals in the dataset are defined to meet four criteria: (1) transfer of rights to use, control, or ownership of land through sale, lease, or concession; (2) occurred after the year 2000; (3) covers an area of 200 hectares or more; (4) implies potential conversion of land from smallholder production, local community use, or important ecosystem service provision to commercial use.
- Data quality notes from Nolte et al. (2016): 80% of deals are listed based on two or more independent sources; 6% of deals are based on media reports only.

### Temporal coverage and main empirical period
- All deals considered between 2000 and 2016 (unrestricted period).
- A restricted period is used for main empirical analysis: 2006-2013.
- Rationale for focusing on 2006-2013:
  - Monitoring of large-scale land deals was inconsistent before the 2008 global economic crisis; retroactive information for early years may be incomplete and increase mismeasurement.
  - Civil society scrutiny intensified after the 2008 crisis, potentially affecting reporting in later years.
- Results are reported to be similar across the restricted and unrestricted periods (see section 6.2).

### National-level determinants and auxiliary datasets
- Physical distance between countries and a dummy for former colonial relationship: GeoDist database (Mayer and Zignago, 2011).
- Land governance index at national level: first principal component of land governance variables in the 2009 Institutional Profiles Database (de Crombrugghe et al. 2009); measures recognition and tenure security of preexisting rights in rural areas (see Arezki et al., 2015).
- Dummy variable for regional trade agreements (1995-2004): Mario Larch’s Regional Trade Agreements Database from Egger and Larch (2008); includes all 468 multilateral and bilateral trade agreements as notified to the World Trade Organization for the last 66 years from 1950 to 2015.
- World Development Indicators (2009): agricultural value added per worker (in constant 2010 US$), GDP per capita (in current US$) and population size.
- Gross food production (current US$) and bilateral food trade (current US$): CEPII TradeProd database (de Sousa et al. (2012)), covering 26 sectors using ISIC revision 2 and 151 importing and exporting countries.
  - For trade-cost estimation, trade flow data are summed for food (ISIC 311) and beverages (ISIC 313) for each year in 1995-2004.
  - Intranational trade flows: CEPII TradeProd contains intranational flows for many countries; missing values imputed using the Anderson and Yotov (2010) method.
    - Imputation approach: determine for each country the ratio of aggregate intranational trade over total goods expenditures for sectors with available data; multiply total expenditures on food and beverages by this ratio to estimate intranational trade where missing.

### Construction of available suitable land (GAEZ)
- Source: Global Agro-Ecological Zones (GAEZ) data developed by FAO and IIASA (http://gaez.fao.org/).
- Crop set: five crops—wheat, maize, oil palm, sugarcane, soybean.
- For each 5-arc minute (˜10km by 10km) grid cell, calculate the maximum of the suitability indexes (rescaled between 0 and 100) under rainfed and baseline climate (1961-1990) conditions.
- Define suitable available land as total area of all cells for which this maximum for at least one crop is greater than 70 (corresponding to high or very high suitability under rainfed conditions), after excluding cultivated land, forests, and protected land.
- The resulting variable is expressed in 1,000 hectares.
- Using this approach, the authors estimate that the total quantity of suitable available land (uncultivated, non-forest, non-protected land suitable for agriculture outside urbanized areas) is about 403 million hectares for the whole world.

### Key descriptive statistics and dataset highlights
- Timing and coverage:
  - Almost 80% of the deals were negotiated between 2006 and 2013.
  - As of June 2016, the Land Matrix has information on 2,152 (trans)national deals, with a cumulative size of 58.4 million hectares, affecting 88 host countries worldwide.
  - Table 5 reports Total area (mn ha) for All = 58.9 (Table 5, column entry).
- Regional concentration:
  - Sub-Saharan Africa: 884 deals (Table 5 reports region-specific counts; Table 5 summary notes Sub-Saharan Africa and East-Asia as primary target regions).
  - East Asia: 611 deals.
  - Latin-America: 368 deals.
- Size distribution and implementation:
  - Only 52 deals cover 40.9 percent of all land traded (size distribution is very uneven).
  - Approximately 66 percent of all deals are smaller than 10,000 hectares (100 km2).
  - 51.9 percent of the projects have been implemented.
  - Of the 43.54 million hectares of land associated with implemented deals, 37.6 percent has been confirmed under production.
  - Fraction of deals not implemented is significantly smaller in Sub-Saharan Africa: 37 percent.
- Intended use:
  - 76.5 percent of all recorded deals have been linked to agricultural and biofuel related projects (text statement).
  - Table 5 intended use (percent) entries for All: Agriculture 62.3, Biofuels 14.2, Forestry/Industry/Other 14.3, “Green” 7.3, Unknown 1.9.
- Availability of suitable land by region (Table 4 entries, all in Mn ha):
  - Available uncultivated, non-forest, non-protected land: 403.2 1.3 208.2 130.9 24.8 22.1 0.4 7.2 8.1
  - Available uncultivated, forest, non-protected land: 669.8 0.1 152.4 242.2 113.2 49.5 7.6 103.2 1.4
  - Total available uncultivated, non-protected land: 10731.4 360.6 373.1 13871.6 8.0 110.4 9.5
  - (Notes: region abbreviations refer to Middle East and North Africa (MENA), Sub-Saharan Africa (SSA), Latin America and the Caribbean (LAC), East Asia and the Pacific (EAP), Europe and Central Asia (ECA), South Asia (SA) and Western Europe (WNE).)
- Descriptive table highlights (Table 2 / Table 3 variables and summary statistics):
  - projectb (number of deals per investor-host pair, 2006-2013): N = 79,135; mean = 0.0208; sd = 0.5040; min = 0; max = 54
  - distblog (log of physical distance in km): N = 50,176; mean = 8.8090; sd = 0.838; min = -0.0048; max = 79.901
  - colonyb (colonial relationship dummy): N = 50,176; mean = 0.0096; sd = 0.0975; min = 0; max = 1
  - suitablenfnplog (log of quantity of non-forest, non-protected land in ha suitable for agriculture): N = 29,573; mean = 5.786; sd = 2.463; min = 1.085; max = 10.94
  - suitabilityindex: N = 163,063; mean = 3,748; sd = 2,577; min = 0; max = 9,194
  - landgovernance: N = 34,098; mean = -0.0196; sd = 2.279; min = -5.048; max = 4.194
  - agri valueaddedolog: N = 37,645; mean = 8.412; sd = 1.556; min = 5.283; max = 11.53
  - populationolog: N = 45,264; mean = 15.29; sd = 2.333; min = 9.19; max = 21.01
  - gdpcaplog: N = 42,354; mean = 8.457; sd = 1.524; min = 5.249; max = 11.53
  - log(PA i) (BB) investor remoteness (Baier and Bergstrand 2009): N = 42,348; mean = 8.754; sd = 0.392; min = 6.735; max = 9.519
  - log(ΩA l) (BB) host remoteness (Baier and Bergstrand 2009): N = 42,341; mean = 8.755; sd = 0.392; min = 6.735; max = 9.519
  - Predicted agricultural exports (Frankel and Romer 1999) logs: N = 63,066 / 63,067; mean = 23.59; sd = 0.430; min = 22.57; max = 24.44
  - Alternative remoteness measures (Anderson and Yotov 2016): log(PA i) (AY): mean = 1.985 / 1.983; log(ΩA l) (AY): mean = 1.283 / 1.282 (table variations reported).
- Notes on alternative data sources and robustness:
  - Using UN Comtrade agricultural trade flow data produced comparable estimates of trade costs.

*Source: "5.1 Sources" (chapter section) from the provided IMF working paper content.*

### 6.1    Main results

### 6.1    Main results

### Baseline PPML regression: overview
- Dependent variable: projectb measures the number of deals concluded per investor-host country pair between 2006-2013.
- Estimation method: PPML estimator; constant and fixed effects included but not reported; robust standard errors clustered by country-pair.
- Sample size: N15960.
- Selected baseline coefficient estimates (column (1) of Table (6)):
  - dist1log -0.983***
  - dist2log -1.092***
  - dist3log -1.035***
  - dist4log -0.999***
  - colonyb 0.686**
  - agrivaolog 0.411***
  - populationolog 0.748***
  - gdpcaplog -0.559***
  - suitable nfnplog 0.346***
  - land governance 0.00483
- r2p for column (1): 0.404.
- Key interpretation: coefficients on regressors expressed in logs can be interpreted as elasticities; on levels as semi-elasticities.

### Role of land endowments and land governance
- Findings:
  - Available suitable land increases the frequency of land deals: "the number of bilateral deals increases by 3.5 percent for a 10 percent increase in the quantity of available land (non-cultivated, non-forested and non-protected) in the host country."
  - Elasticity stability: varies between 2.2-4.7 percent once controlling for multilateral resistance (columns (2)-(7)).
  - Land governance: coefficient generally negative in most columns (as in Arezki et al. (2015)) but not statistically significant in any regression in this analysis.
- Note on potential masking: land governance variable is correlated substantially with host-country GDP per capita; inclusion of host-country GDP per capita may mask effect of land governance.

### Investor-country characteristics
- Population:
  - A 10 percent increase in investor-country population raises expected number of deals roughly by 7.5 percent (baseline result; column evidence: populationolog 0.748*** in column (1)).
- Agricultural productivity:
  - Agricultural value added per worker in investor country (proxy for investor agricultural productivity) is positively associated with number of land deals: "If agricultural value added per worker in the origin country increases by 10 percent, the expected number of bilateral deals increases by 4.1-5.2 percent."
  - Baseline coefficient: agrivaolog 0.411*** (column (1)); column (2) agrivaolog 0.521***; column (3) 0.468***; column (4) 0.485***; column (5) 0.114**.

### Distance, trade costs, and fixed effects
- Distance coefficients across intervals:
  - Distance impacts (column (1)) are similar across four distance intervals (dist1log to dist4log roughly between -0.983 and -1.092) and not statistically different from one another.
  - Contrast with bilateral trade literature where long distances impede trade more strongly than short distances.
- Fixed effects amplify trade-cost roles:
  - Inclusion of country fixed effects (investor and/or host) increases magnitude of distance coefficients (columns (2)-(4)), with absolute magnitudes substantially larger than 1 in FE regressions.
  - Implication: (i) distance is a stronger deterrent of cross-border land acquisitions than of trade in goods; (ii) distance has a much stronger negative effect on number of land deals than initially found by Arezki et al. (2015).
- Former colony-colonizer dummy:
  - Coefficient increases in magnitude with inclusion of FE (e.g., colonyb 0.845*** in column (2), 0.910*** in column (3), 1.063*** in column (4)).

### Remoteness / market potential: three proxy approaches and results
- Theoretical expectations:
  - Under food independence motive: host-country remoteness from agricultural consumers (ΩA l) and investor-country remoteness from agricultural producers (P A i) expected to increase land deals (γ8>0, γ9>0).
  - Under platform FDI motive: host-country remoteness expected to play a negative role (γ8<0); investor-country remoteness expected not to play a role (γ9=0).
- Column (5): Baier and Bergstrand (2009) remoteness proxies
  - Host-country remoteness (log(ΩA l) (BB)) coefficient: 3.332*** (0.465).
  - Investor-country remoteness (log(P A i) (BB)) coefficient: -0.158 (0.165) — not statistically significant.
  - Quantitative interpretation: "if the income-weighted average physical distance to agricultural markets increases by 10 percent, then the expected number of bilateral deals increase by as much as 33 percent."
- Column (6): Frankel and Romer (1999) predicted agricultural exports as inverse remoteness measures
  - log( ∑ l=N l=1,l≠i ˆX A il ) (investor predicted exports) coefficient: -0.505*** (0.182).
  - log( ∑ n=N n=1,n≠l ˆX A nl ) (host predicted exports) coefficient: -2.455*** (0.305).
  - Interpretation: predicted agricultural exports are inverse measures of remoteness; negative coefficients imply bilateral investment is positively associated with remoteness of both investor and host.
  - Quantitative magnitudes: for every 10 percent decrease in predicted exports (10 percent increase of predicted imports) of host country, expected bilateral deals increase by 24.6 percent; for investor country the comparable effect is 5.1 percent.
- Column (7): Anderson and Yotov (2016) structural gravity inferred remoteness and trade-costs
  - agritradecost log coefficient: -2.995*** (0.175).
  - log(P A i) (AY) coefficient: -0.0664 (0.155) — not statistically significant.
  - log(ΩA l) (AY) coefficient: 3.035*** (0.295).
  - Interpretation: using inferred remoteness from observed agricultural trade flows, a 10 percent increase in host-country remoteness increases bilateral investments by 30.4 percent; investor-country remoteness is smaller and not statistically significant.

### Robustness and interpretation across remoteness measures
- Consistent result across columns (5), (6), (7): positive and statistically significant role for host-country remoteness in attracting large-scale land acquisitions.
- Investor-country remoteness: mixed evidence
  - Coefficients on P A i (BB), log( ∑...ˆX A il ) (Frankel-Romer), and P A i (AY) suggest either no role or a negative role for investor-country remoteness.
  - Conclusion: cannot unambiguously distinguish food-independence motive from platform-FDI motive solely on investor-country remoteness evidence.
- Possible offsetting factors for investor-country remoteness:
  - Large land investments may require human capital and financial capital that are less abundant in remote investor countries; thus net effect of investor remoteness can be ambiguous.

### Quantitative examples: Angola and thought experiment on relative importance
- Angola example:
  - Angola recorded 19 deals between 2000 and 2016.
  - Angola offers 236 percent more available suitable agricultural land than sample average: 6.71 million hectares versus 2.00 million hectares.
  - Using baseline coefficient on suitable land: "we expect that an investor-host country pair (featuring Angola as the host) would have, by virtue of this difference alone, 2.36*0.34*100 = 80 percent more land acquisitions than an investor-host country pair with the average country in our sample as the host."
  - Remoteness example for Angola:
    - Average country GDP-weighted average distance ≈ 6,340 km.
    - Angola GDP-weighted average distance ≈ 8,554 km.
    - Angola is 35 percent more remote than average.
    - Applying Baier and Bergstrand remoteness coefficient (column (5)): 0.35*3.32*100 =116 percent more bilateral deals for a given investor-host pair with Angola as host versus average host.
- General thought experiment (standard-deviation impacts):
  - Increase in host suitable land endowment by one standard deviation increases bilateral deals by:
    - 0.346*(7580/3047)*100 = 86.1 percent.
  - Increase in Baier and Bergstrand distance-based remoteness of host-country by one standard deviation increases bilateral deals by:
    - 3.33*(2361/6790)*100 = 115.8 percent.
  - Using remoteness based on agricultural trade costs, the comparable increase is:
    - 3.04*(1.04/2.90)*100 = 109.0 percent.
  - Conclusion: host-country remoteness is a relevant determinant and its importance is comparable to or exceeds that of agricultural land endowments.

### Overall interpretation and implications
- Empirical picture sketched by Table (6):
  - Large-scale land acquisitions frequently target countries with abundant available suitable land (non-cultivated, non-forested, non-protected).
  - Large-scale land acquisitions frequently target countries that are not well integrated into global food trading system, as evidenced by significant host-country remoteness from existing food importers.
  - Productive, populous investor countries are the primary drivers behind the rush for land.
- Motive consistency:
  - Columns (5), (6), (7) are more consistent with the food independence motive for land acquisitions than with the platform FDI motive.
  - Positive relationship between agricultural productivity and population size in investor countries suggests productive investors from existing centers of agricultural production (Western-Europe, North-America and South-East Asia) invest away from centers of demand into potentially new emerging centers of food production.

*Source: wp18145 - 6.1    Main results*

### 6.2    Robustness Checks

### 6.2    Robustness Checks

### Overview of robustness exercises
- Multiple alternative regressions were run to verify robustness of main results.
- Tests address potential confounders (an “Africa effect”), sample composition (zero flows), alternative dependent variables (cumulative land quantity), and sample period (2000–2016 versus 2006–2013).

### Controlling for an “Africa effect”
- Motivation: more than 41 percent of all land deals target countries in Sub-Saharan Africa.
- Method: extended regressions in Table (6) with an African continent dummy variable.
- Findings:
  - The Africa dummy variable is highly significant.
  - Inclusion of the Africa dummy leaves all other results qualitatively unaffected.
  - Inclusion reduces the magnitude of remoteness regression coefficients in some cases, which is expected because the majority of African countries are substantially above-average in terms of remoteness and the Africa dummy picks up part of that effect.

### Restricting to strictly positive bilateral flows (extensive-margin focus)
- Sample restriction: constrained dependent variable (number of land deals by country pair) to strictly positive values only; sample size reduced to 409 bilateral flows.
- Estimation and notes:
  - Results reported in Table A.1; estimates obtained with the PPML estimator.
  - Reminder: PPML estimator is generally well behaved even if the proportion of zeros is very large (Silva and Tenreyro, 2011).
- Findings:
  - Host-country remoteness, measured by the Baier and Bergstrand proxy (column 5) and Frankel and Romer’s predicted food exports (column 6), remains statistically significant at the 1% level.
  - The trade-cost-induced remoteness proxy (column 7) enters with a positive sign but is no longer statistically significant.
  - Elasticities on host-country remoteness in columns (5), (6) and (7) are smaller in magnitude than in the unconstrained regression.
    - Interpretation: the unconstrained elasticity can be seen as a combination of an extensive margin effect (whether to acquire land) and an intensive margin effect (how many deals). Restricting to positive flows isolates the intensive margin and/or a different composition of observations.
  - Conclusion: results remain qualitatively unaffected; robustness exercise reaffirms the paper’s main message despite the smaller sample (and the typical caution that gravity regressions are not usually run on so few observations).

### Using cumulative contract size as dependent variable (intensive-margin focus)
- Dependent variable: quantity of land under each contract (cumulative contract size per investor-country host-country pair).
- Results reported in Table A.2 for 2000-2016.
- Findings:
  - Remoteness proxies in column (5) and (7) remain statistically significant; the proxy in column (6) does not.
  - Magnitudes of the coefficients on bilateral distance and host-country remoteness are smaller than in the baseline where the number of deals is the dependent variable.
  - Interpretation: distance and remoteness have larger effects on the decision whether to acquire land (extensive margin) than on the decision about the quantity of land to acquire once investments have started (intensive margin).

### Expanding the time horizon to 2000–2016
- Sample expansion: all land deals between 2000 and 2016 instead of 2006 and 2013.
- Results reported in Table A.3.
- Findings:
  - All three proxies for host-country remoteness are positive and statistically significant, as in Table (5).
  - The magnitude of the coefficients little changes relative to the baseline.
  - Conclusion: observed pattern of large-scale land acquisitions is consistent across a larger time span (starting well before the 2008 crisis) with the notion that investors aimed to secure land to reexport produce to their home countries.

### Additional methodological and empirical notes
- Trade-cost estimation (Appendix A) for agricultural bilateral trade costs using data on agricultural imports between 1995–2004 produced elasticities:
  - Distance elasticities around 0.77 (reported coefficients: −0.764 (0.050), −0.781 (0.041), −0.784 (0.040), −0.755 (0.038)).
  - Common language elasticity: 0.670 (0.068).
  - Colonial ties elasticity: 1.134 (0.133).
  - Contiguous border effect estimated at −0.024 (0.116), statistically insignificant.
- PPML estimation details in appendix tables:
  - Table A.1 (positive-only bilateral number of deals): N 409; r2 p values reported (e.g., 0.162, 0.483, 0.365, 0.404, 0.226, 0.256, 0.158 across columns).
  - Table A.2 (cumulative deal size, 2000–2016): N 15960 in many specifications; r2 p values reported (e.g., 0.400, 0.764, 0.583, 0.602, 0.411, 0.416, 0.445).
  - Table A.3 (number of deals, 2000–2016): N 15960; r2 p values reported (e.g., 0.432, 0.600, 0.643, 0.482, 0.503, 0.463).
  - Significance notation: *** p<0.01; ** p<0.05; * p<0.1.
- Data patterns highlighted: the majority of observations in the full dataset are zeros; restricting to positive observations yields smaller sample sizes and different magnitudes but preserves main qualitative results.

### Implications for the paper’s conclusions
- Robustness checks reinforce the primary finding that host-country remoteness plays a positive role in bilateral investment in large-scale land acquisitions across multiple specifications, dependent variables, and sample periods.
- This consistency supports the interpretation of land investment as food-independence driven vertical FDI rather than profit-seeking platform FDI, and suggests demand for suitable and available land may have been relatively large compared to supply (“investor-abundant, land-scarce”).

*Source: wp18145 - 6.2 Robustness Checks (IMF Working Paper content).*

### Appendix C. Technical Appendix.

### Appendix C. Technical Appendix.

### Proposition 1 — Existence proof overview
- Setup: fl is fixed in each country and the quantity of land TA l adjusts to clear the market for land.
- Goal: prove existence of at least one vector 〈w, TA〉 ∈ R2n++ such that G(w, TA) satisfies five properties.
- The five properties verified for G(w, TA):
  - (i) G(w, TA) is continuous.
    - Reason: all the prices cA li, pA l, PA n, pM l and PM n are continuous, so ΩA l and ΩM l are continuous too, and so is G.
  - (ii) G(w, TA) is homogenous of degree zero.
    - Reason: unit variety prices cA li and pA l are homogenous of degree α in 〈w, TA〉, while pM l is homogenous of degree one in 〈w, TA〉. From equations (21), (22), (14) and (15) the expressions
      - 1/TA l (pA l ΩA l)1−εA VA,
      - 1/w l (pA l ΩA l)1−εA VA, and
      - 1/w l (pM l ΩM l)1−εM VM
      are homogenous of degree zero, hence G is homogenous of degree zero.
  - (iii) Walras’ law: 〈w, TA〉 · G(w, TA) = 0 for all 〈w, TA〉 ∈ R2n++.
    - Interpretation: corresponds to balanced trade in each country. Substituting for ΩA l and ΩM l into the excess demand functions yields
      - ∑l=N l=1 wl GL l(w, TA) + ∑l=N l=1 TA l GT l(w, TA) =
        ∑l=N l=1 (∑n=N n=1 (pA l tnl PA n)1−εA γIn) +
        ∑l=N l=1 (∑n=N n=1 (pM l tnl PM n)1−εM (1−γ)In) − ∑l=N l=1 In,
      - and because consumers in each country spend all income on agricultural and manufacturing varieties:
        (∑l=N l=1 (pA l tnl PA n)γIn) + (∑l=N l=1 (pM l tnl PM n)1−εM (1−γ)In) − In = 0,
      - hence 〈w, TA〉 · G(w, TA) = 0.
  - (iv) Lower bounds: for k = max(maxh Lh, maxh fh), GL l(w, TA) > −k and GT l(w, TA) > −k for all l = 1,...,N and 〈w, TA〉 ∈ R2n++.
    - From inspection of equations (26) and (27), each excess demand function has a lower bound equal to inelastically supplied amount of labour or exogenous price of land, i.e., GL l(w, TA) > −Ll and GT l(w, TA) > −fl for all 〈w, TA〉 ∈ R2n++. Hence
      - GL l(w, TA) > −max(maxh Lh, maxh fh) and GT l(w, TA) > −max(maxh Lh, maxh fh) for all l = 1,...,N.
  - (v) Boundary divergence: if 〈w, TA〉 → 〈w0, TA,0〉 where 〈w0, TA,0〉 ≠ 0 and w0 l = 0 and/or TA,0 l = 0 for some l, then maxh Gh(w, TA) → ∞.
    - Strategy: show that along any factor price sequence 〈wm, TA,m〉 → 〈w0, TA,0〉 with some component tending to zero, at least one of the following diverges:
      - (i) maxk,n ((pM k tnk PM n)1−εM (1−γ)In wk) → ∞,
      - (ii) maxk,n ((pA k tnk PA n)1−εA γIn wk) → ∞, or
      - (iii) maxk,n ((pA k tnk PA n)1−εA γIn TA k) → ∞.
    - Verification uses that (i) and (ii) follow from (a) maxn wm n → maxn w0 n > 0 and minn wm k → minn w0 k = 0, while (iii) follows from (b) maxn TA,m n → maxn TA,0 n > 0 and min k TA,m k → min k TA,0 k = 0. Since either (a) or (b) was assumed, (v) is verified.
- Conclusion: Existence follows from Proposition 17.C.1. of Mas-Colell et al. (1995, p. 585), an application of Kakutani’s Fixed-Point Theorem. The five properties also hold for the case where countries sell a fixed quantity of land to agricultural producers, TA l = TA l, with the solution to the excess demand system given by vector 〈w, f〉 ∈ R2n++.

### Land Acquisitions under Food Independence Motive
- Limiting case: σ → ∞ and assume multinational production is the cheapest option for each consumer country to obtain every producer’s country unique agricultural product.
  - Then equation (17) becomes pA il = til cli, and subscript i refers to the consumer country (investor and consumer country are the same under the food independence motive).
- Definitions and identities:
  - Define ̃VA l ≡ ∑N i=1 XA il and write
    - ̃VA l ≡ (wα l f1−α l ̃ΩA l)1−εA VA
  - Where
    - ̃ΩA l ≡ ( ∑N i=1 (til Zli ̃PA i)1−εA γIi VA )
    - ̃PA i ≡ ( ∑N i=1 (til cA li)1−εA )1/(1−εA)
  - Using the power transform (wα l f1−α l)1−εA = VA (̃ΩA l)1−εA ̃VA l and substituting into ̃PA i and equation (12), and noting TA il = (1−α)1 f l VA il, we obtain:
    - (B.1) TA li = (til Zli ̃ΩA l ̃PA i)1−εA γIi TA l VA
    - ̃PA i ≡ ( ∑N l=1 (til Zli ̃ΩA l)1−εA ̃VA l VA )
    - Using ̃VA l = fl TA l1−α.

### The Equivalence of Remoteness under Platform FDI and Food Independence FDI
- Assumption: Zln can be written as multiplicative function of investor-country agricultural productivity and a bilateral distance term, i.e., Zli = Zi F(til).
- Simplification: substituting Zli = Zi F(til) into terms of (B.1) allows all til Zli terms to simplify to til F(til), eliminating the Zi term.
  - Since F′(til) < 0, the Baier and Bergstrand approximation to the Food Independence host-country remoteness term,
    - ∏N i=1,i≠l (til F(til))θi,
    - is a more magnified version of the remoteness term under the Platform FDI model (distance enters twice) and relies on the same observables (distance and population or agricultural production shares).
- Special functional form example: suppose F(til) = 1/φ til with φ > 0.
  - Then the host-country remoteness term in the Food Independence version can be written as:
    - ̃ΩA l ≡ ( ∑N i=1 (φ til2 ̃PA i)1−εA γIi VA )
  - Using Baier and Bergstrand’s linear approximation method gives the proxy:
    - ̃ΩA l = ∏N i=1,i≠l (φ til2)θn = φ (∏N i=1,i≠l (til)θi)2 = φ (ΩA l)2.
  - Taking natural logs yields ln φ + 2 ( ∑N i=1,i≠l θi ln(til) ); in regressions the ln φ term is absorbed by the regression constant and the factor 2 is an irrelevant scaling factor.
  - Conclusion: under this functional form the two host-country remoteness terms from the two models are indistinguishable. The same argument applies to investor-country remoteness terms.

*Source: wp18145 - Appendix C. Technical Appendix.*

---


_Source: https://www.imf.org/-/media/files/publications/wp/2018/wp18145.pdf_