Misallocation and Productivity in Agriculture

Transcription

1 Misallocation and Productivity in Agriculture Diego Restuccia University of Toronto and NBER University of Oslo August 14-18, 2017 Restuccia Macro Growth and Development University of Oslo 1 / 68

2 Overview Misallocation and Productivity in Agriculture Farm size facts across countries A model of farm size Aggregate factors vs. farm-specific distortions Evidence of misallocation Restuccia Macro Growth and Development University of Oslo 2 / 68

3 Misallocation in Agriculture Based on Adamopoulos and Restuccia (2014) The Size Distribution of Farms and International Productivity Differences, American Economic Review, Volume 104 (6), June 2014, pp Restuccia Macro Growth and Development University of Oslo 3 / 68

4 Motivation Poor countries are unproductive......but they are particularly unproductive in agriculture... In addition, poor countries devote most of their labor to agriculture As a result, agriculture important in accounting for income differences between rich and poor countries Restuccia Macro Growth and Development University of Oslo 4 / 68

5 Resource (Mis)allocation within Agriculture Farm structure and farm size differs substantially across countries Resource misallocation across production units generates aggregate productivity effects Restuccia and Rogerson (2008) Guner, Ventura, and Yi (2008) Hsieh and Klenow (2009) Focus on policies and institutions that create misallocation in agriculture above and beyond misallocation that affects economy-wide productivity (e.g. distortions from land market institutions) Restuccia Macro Growth and Development University of Oslo 5 / 68

6 Main Idea misallocation of resources across farms of different sizes = low agricultural productivity low agricultural productivity + subsistence constraint = high share of employment in agriculture, low average farm size Restuccia Macro Growth and Development University of Oslo 6 / 68

7 Why Study this Channel? Poor countries have mostly small farms compared to rich countries Larger farms have higher labor productivity Policies and institutions that encourage smallness are prevalent in poor countries Restuccia Macro Growth and Development University of Oslo 7 / 68

8 Research Questions What explains farm size and farm distribution differences across rich and poor countries? Does farm size matter for agricultural and aggregate productivity differences? Restuccia Macro Growth and Development University of Oslo 8 / 68

9 Approach Develop a two-sector model of agriculture and non-agriculture that features a non-degenerate distribution of farms, calibrated to U.S. farm-level data Use this framework to assess the quantitative importance of: (a) aggregate factors (b) farm-level policies that distort size in accounting for farm size and productivity Restuccia Macro Growth and Development University of Oslo 9 / 68

10 Main Findings Aggregate factors can account for 1/4 of the differences in farm size and agricultural and aggregate productivity Farm-level distortions can potentially account for most of the remaining differences Land reforms that cap size a prevalent policy in poor countries have non-trivial size and productivity effects Restuccia Macro Growth and Development University of Oslo 10 / 68

11 Data Data Restuccia Macro Growth and Development University of Oslo 11 / 68

12 Data Data Dataset of 63 countries Internationally comparable data on land holdings by size from the 1990 Report of the World Census of Agriculture, compiled by the Food and Agricultural Organization (FAO) Combine these data with aggregate output data from the Penn World Table and agricultural output data from the FAO Restuccia Macro Growth and Development University of Oslo 12 / 68

13 Data Average Farm Size Across Countries Corr = 0.61 AUS Log of Average Farm Size PRY BRA GER DNK FRA LUX 3 PER COL IRL AUT PRI ESP PAN ISR BEL NLD HND BHS CHE PRT NOR 2 IRN ITA WSM TUR FJI SVN BFA PAK GRC THA CYP 1 NAM UGA PHL DMA KNA LCA INDGIN LSO VCT BRB 0 GNB KOR JPN ETH NPL IDNEGY MWI GRD ZAR VNM Log of 1990 Real GDP per Capita ARG CAN USA GBR FIN Restuccia Macro Growth and Development University of Oslo 13 / 68

14 Data Average Farm Size By Quintile of Income Distribution GDP per Average Farm Size (Ha) capita Total Wheat Rice Maize Q1 1, Q2 3, Q3 6, Q4 16, Q5 23, Restuccia Macro Growth and Development University of Oslo 14 / 68

15 Data Size Distribution of Farms - Small (< 5 Ha) Restuccia Macro Growth and Development University of Oslo 15 / 68

16 Data Size Distribution of Farms - Large (> 20 Ha) Restuccia Macro Growth and Development University of Oslo 16 / 68

17 Data Size Distribution of Farms - Selected Poor Ethiopia Malawi Congo Fraction of Farms Farm Size (Ha) Restuccia Macro Growth and Development University of Oslo 17 / 68

18 Data Size Distribution of Farms - Selected Rich United Kingdom Canada United States Fraction of Farms Farm Size (Ha) Restuccia Macro Growth and Development University of Oslo 18 / 68

19 Model Model Restuccia Macro Growth and Development University of Oslo 19 / 68

20 Model Economic Environment General equilibrium two-sector model of agriculture and non-agriculture Sectoral allocations driven by subsistence constraint for food Embed Lucas (1978) span-of-control model of firm size in agriculture, which allows for a non-degenerate distribution of farms Restuccia Macro Growth and Development University of Oslo 20 / 68

21 Model Production Non-Agriculture Representative firm in non-agriculture produces according to constant returns to scale technology, Y n = AK α n N 1 α n A = economy-wide productivity (TFP) K n = non-agricultural capital N n = non-agricultural labor Restuccia Macro Growth and Development University of Oslo 21 / 68

22 Model Production - Agriculture Production unit is a farm Farm: a technology that requires the input of an operator (farmer) with managerial skills s, land input (l) which defines its size, and capital (k) Farmer of ability s produces according to decreasing returns to scale technology, y a = Aκ [θk ρ + (1 θ) (sl) ρ ] γ ρ κ = sector-specific agricultural productivity γ = span-of-control parameter 1/(1 ρ) = capital-land elasticity of substitution Restuccia Macro Growth and Development University of Oslo 22 / 68

23 Model Stand-in Household Non-homothetic preferences over two goods (c a, c n ), u(c a, c n ) = {φ log(c a ā) + (1 φ) log(c n )} Endowments, supplied inelastically to the market: - capital stock K - arable land L - one unit of time per member Household members are heterogeneous in managerial ability in farming, s F (s) with support in S = [s, s] Workers in non-agriculture are homogeneous in ability Restuccia Macro Growth and Development University of Oslo 23 / 68

24 Model Market Clearing Conditions Land L = N a S l(s)df (s) Capital N a S k(s)df (s) + K n = K Goods c a = N a S y a (s)df (s) c n = Y n Restuccia Macro Growth and Development University of Oslo 24 / 68

25 Model Farmer Problem Farmer of ability s maximizes profits given prices (p a, q, r), { } p a Aκ [θk ρ + (1 θ) (sl) ρ ] γ ρ ql rk max l,k Optimal land and capital demands [l(s), k(s)] imply, [ ] 1 k(s) θ l(s) = q 1 ρ s ρ 1 ρ 1 θ r Resulting farm output (also labor productivity) and profits, y a (s) = Aκ [θk(s) ρ + (1 θ) (sl(s)) ρ ] γ ρ π(s) = (1 γ) p a y a (s) Restuccia Macro Growth and Development University of Oslo 25 / 68

26 Model Household Problem Household maximizes preferences given prices subject to, p a c a + c n = I (1 N a )w (1 ξ) + N a π (s) df (s) + ql + rk (1 ξ) = barrier to labor mobility out of agriculture Optimality with respect to N a implies, w(1 ξ) = π (s) df (s) Optimal consumption of the two goods given by, c n = (1 φ) (I ap a ) c a = a + φ (I ap a ) p a S S Restuccia Macro Growth and Development University of Oslo 26 / 68

27 Model Discussion Share of employment in agriculture mainly determined by subsistence relative to agricultural productivity In simple case with φ = 0, ρ 0, and θ = 0, where Then κal γ [E(s)N a ] 1 γ = ā E(s) = ( N a = S s γ 1 γ df (s) a κal γ E(s) 1 γ ) 1 1 γ Aggregate factors elasticity depends critically on land share γ Restuccia Macro Growth and Development University of Oslo 27 / 68

28 Model Discussion Farmer of ability s chooses farm size, where l(s) = [ γaκ p ] 1 1 γ γ a s 1 γ q p a q = L γa Restuccia Macro Growth and Development University of Oslo 28 / 68

29 Calibration Calibration Restuccia Macro Growth and Development University of Oslo 29 / 68

30 Calibration Calibration Strategy: Calibrate a benchmark economy (BE) to U.S. farm-level and aggregate observations Distribution of farmer ability F (s) approximated by a log-normal distribution, with mean µ and variance σ, chosen to match the distribution of farm sizes from U.S. Census of Agriculture Restuccia Macro Growth and Development University of Oslo 30 / 68

31 Calibration Calibration Normalize A and κ to 1 α = 0.33 and γ = 0.54, to match sectoral capital income shares (Herrendorf and Valentinyi, 2008) φ = 0.01 to match a long-run share in agriculture of 1% Aggregate endowments (K, L) chosen to match: (a) a capital-output ratio of 2.5 (b) average farm size of 169 Ha ξ chosen to reproduce a ratio of average agricultural to non-agricultural labor productivity of 1/2 Restuccia Macro Growth and Development University of Oslo 31 / 68

32 Calibration Calibration Solve the model for (a, ρ, θ) to match three targets: (a) share of employment in agriculture of 2.5% (b) agricultural land income share of 18% (Herrendorf and Valentinyi, 2008) (c) disparity of capital-land ratio between minimum and maximum farm sizes of 84.8 Resulting ρ = 0.26 implies more substitution than Cobb-Douglas Restuccia Macro Growth and Development University of Oslo 32 / 68

33 Calibration Distribution of Farms by Size 0.35 Model U.S. Data Density Farms < Farm Size Class in Ha (USCA) Restuccia Macro Growth and Development University of Oslo 33 / 68

34 Calibration Land Share Distribution Model U.S. Data 0.5 Density Land < Farm Size Class in Ha (USCA) Restuccia Macro Growth and Development University of Oslo 34 / 68

35 Calibration Capital to Land Ratio by Farm Size Model U.S. Data Capital Land Ratio < Farm Size Class in Ha (USCA) Restuccia Macro Growth and Development University of Oslo 35 / 68

36 Calibration Output per Hectare by Farm Size Model U.S. Data Output Land Ratio < Farm Size Class in Ha (USCA) Restuccia Macro Growth and Development University of Oslo 36 / 68

37 Calibration Output per Worker by Farm Size Model VA Per Worker Adj. VA Per Operator VA Per Worker Raw 25 Output Per Worker < Farm Size Class in Ha (USCA) Restuccia Macro Growth and Development University of Oslo 37 / 68

38 Cross-Country Analysis Cross-Country Analysis Restuccia Macro Growth and Development University of Oslo 38 / 68

39 Cross-Country Analysis Aggregate Factors How important are aggregate factors (L, A, K) in accounting for: average farm size size and land distribution of farms share of employment in agriculture agricultural and aggregate productivity in poor countries? Restuccia Macro Growth and Development University of Oslo 39 / 68

40 Cross-Country Analysis Aggregate Factors In our data set we measure rich-poor country disparities in: L to match arable land per person differences 1.3-fold A to match non-agricultural real GDP per worker differences 6.8-fold K to match capital-output ratio differences 2.9-fold Restuccia Macro Growth and Development University of Oslo 40 / 68

41 Cross-Country Analysis Aggregate Factors BE Land (+ TFP, Capital) Data Relative Yn N n (i/b.e.) 1 1/1.0 1/6.8 1/6.8 N a (%) Relative Ya N a (i/b.e.) 1 1/1.0 1/11.2 1/46.7 AFS (Ha) 169 1/1.3 1/8.6 1/34 Size Distribution (%): Farms < 5 Ha Farms > 20 Ha Share of Land (%) Farms < 5 Ha Farms > 20 Ha Relative Y N (i/b.e.) 1 1/1.0 1/7.6 1/19.2 All aggregate factors together can account for roughly 1/4 of the disparities in the variables of interest Restuccia Macro Growth and Development University of Oslo 41 / 68

42 Cross-Country Analysis More on Aggregate Factors Interestingly, among developed economies, aggregate-factor differences account for most of the differences in average farm sizes (AFS) This is so for developed economies that have, relative to the B.E., either high AFS such as Canada (240) or low AFS such as Netherlands (17) and Germany (33) Restuccia Macro Growth and Development University of Oslo 42 / 68

43 Cross-Country Analysis Farm-Size Distortions Poor countries are exposed to several farm-level policies that distort size (farm-size distortions): Direct restrictions on size, e.g. land reforms Progressive land and output taxes Input subsidies for small farms Land rental restrictions Subsidized credit to small holders Differential application of regulations Social institutions, e.g. inheritance norms Transportation costs Restuccia Macro Growth and Development University of Oslo 43 / 68

44 Cross-Country Analysis Farm-Size Distortions Can farm-size distortions account for the unexplained productivity gap between rich and poor countries? We consider two types of farm-size distortions: Generic distortions that account for the pattern of farm size Specific relate to particular/observable policy choices, e.g. land reforms and progressive taxes Restuccia Macro Growth and Development University of Oslo 44 / 68

45 Cross-Country Analysis Farm-Size Distortions Follow approach in Restuccia and Rogerson (2008). Each farmer faces a farm-specific output tax τ s (idiosyncratic distortions). Parameterize taxes according to the following function: τ s = P (s) = 1 1 exp(ψs). Taxes as catch-all distortions, functional form motivated by features of farm-size distortions. Calibrate ψ to match average farm size in poor countries. Restuccia Macro Growth and Development University of Oslo 45 / 68

46 Cross-Country Analysis Farm-Size Distortions - Results Aggregate +Farm-Size Data Factors Distortions Size Distribution (%): Farms < 5 Ha Farms > 20 Ha Share of Land (%): Farms < 5 Ha Farms > 20 Ha N a (%) Ratio B.E./Poor: Average Farm Size Labor Prod in Ag Agg Labor Prod Farm-size distortions can potentially account for the disparities between rich and poor countries. Restuccia Macro Growth and Development University of Oslo 46 / 68

47 Cross-Country Analysis Empirical Farm-Size Distortions Measure of distortions from internationally comparable database. Nominal rates of assistance (NRAs) from World Bank. Available at the aggregate and crop categories across countries (Kruger, Schiff, and Valdes extended by Anderson and Valenzuela 2008). Measure of distortions to farmer prices induced by domestic policy: taxes/subsidies to farmers and price wedges (farm gate vs. world). Restuccia Macro Growth and Development University of Oslo 47 / 68

48 Cross-Country Analysis Empirical Farm-Size Distortions To uncover distortions by productivity level (size) we assume product level size in the United States Cotton 0.3 Relative Tax rate (Q1 Relative to U.S.) Coffee Tobacco Fruits and Vegetables Groundnuts Sorghum Maize Soybean Pulses Sunflower Wheat Sugar Rice Millet 0 Sweet Potatoes 0.05 Rapeseed U.S. AFS (ordered in ascending order) Restuccia Macro Growth and Development University of Oslo 48 / 68

49 Cross-Country Analysis Empirical Farm-Size Distortions Negative NRA is a tax on producers, we focus on mean, std and correlation with U.S. size as key statistics. We find high mean relative taxes (rich to poor) of 2.6 fold, normalizing rich to zero, then taxes in poor = Std of 0.33 (crop-specific NRAs) and correlation with U.S. size of 0.6. Parameterize taxes according to the following generalized function: τ s = P (s) = 1 ψ 0 exp(ψ 1 s). Calibrate (ψ 0, ψ 1 ) to match summary statistics on empirical measures of nominal rates of assistance: mean, std, and correlation with U.S. size. Restuccia Macro Growth and Development University of Oslo 49 / 68

50 Cross-Country Analysis Empirical Farm-Size Distortions Aggregate +Farm-Size Data Factors Distortions Size Distribution (%): Farms < 5 Ha Farms > 20 Ha Share of Land (%) Farms < 5 Ha Farms > 20 Ha N a (%) Ratio B.E./Poor: Average Farm Size Labor Prod in Ag Agg Labor Prod Aggregate factors and farm size distortions account equally for more than 50% in the disparities between rich and poor countries. Restuccia Macro Growth and Development University of Oslo 50 / 68

51 Cross-Country Analysis Summarizing Farm-size policies that distort size are harmful for productivity. Farm structure matters, to the extent generated by these policies. Not all problems lie within agriculture: aggregate factors still account for 1/4. What specific policies/institutions are most damaging for size and productivity in agriculture? Restuccia Macro Growth and Development University of Oslo 51 / 68

52 Cross-Country Analysis More Direct evidence of misallocation We have seen land market institutions in poor countries produce severe misallocation of land across farms: Restuccia and Santaeulalia (2017) for Malawi, Adamopoulos et al. (2017) for China, Chen (AEJM forthcoming) across countries, Adamopoulos and Restuccia (2015) for land reform in Philippines, etc. Present further evidence based on a land reform in Ethiopia, exploiting variation in rentals of land across households, across locations, and over time Based on Chen, Restuccia, and Santaeulalia-Llopis (2017) Land Markets, Resource Allocation, and Agricultural Productivity paper.pdf Restuccia Macro Growth and Development University of Oslo 52 / 68

53 Cross-Country Analysis Why Ethiopia? Growth miracle in Africa: GDP per capita (PPP) growth 6.9% ( ) Agriculture is an important sector: around 75% of employment Detailed micro data to study resource misallocation LSMS-ISA data from the World Bank Detailed information on farm-level input and output in physical units Interesting land institution; land certification reform Restuccia Macro Growth and Development University of Oslo 53 / 68

54 Cross-Country Analysis Institutional Background Current land institution shaped by historical events, but the prevailing characteristic has been state control over the allocation and use of land Property rules evolved from the imperial period (mid 19th century to 1974): Allocated land ownership to political supporters regardless of occupation or use in farming, which created a feudal regime Further emergence of private property resulted in powerful landlords Restuccia Macro Growth and Development University of Oslo 54 / 68

55 Cross-Country Analysis Institutional Background Social injustices lead to a communist regime from 1975 to 1991 Land to the Tiller : Expropriated all land in the country and redistributed, adjusting for soil quality and family size, among all rural households in the form of use rights Redistributions were frequent, every one to two years, to achieve an equitable allocation of use rights Prohibited land transactions Since 1991, a market oriented government has largely maintained the policies related to land: Ownership still resides with the state Households assigned use rights by village (kebele) and district (woreda) authorities Many of the restrictions to land transactions remain in place Restuccia Macro Growth and Development University of Oslo 55 / 68

56 Cross-Country Analysis The Restuccia percentage of rented Macro Growth land and differs Development greatly across University zones of Oslo 56 / 68 Institutional Background Land Certification Land certification reforms have been implemented since early 2000s to mainly promote tenure security by issuing land certificates of use rights The implementation of land reforms were decentralized to local governments: the timing, extent, and enforcement has differed across regions While severe rental restrictions remain in place (e.g. only a fraction of use-rights can be rented, lessee must dwell in the rural area and be engaged only in farming) rental markets have developed at differential pace across regions In 2013, 10.6 percent of agricultural land is rented; 24.7 percent of households formally or informally rent in some land for production

57 Cross-Country Analysis Data The Living Standards Measurement Study Integrated Surveys on Agriculture (LSMS-ISA) from the World Bank Main data: Ethiopia , 3,629 agricultural households with information on their location link households to geographical areas Detailed information on farm-household physical outputs and inputs in agricultural production Each household operates a farm consisting of on average 7 land plots, while typically only one crop is planted in each plot allows us to decompose productivity within and across households; and across crops Also explore Ethiopia for cross-time variation Restuccia Macro Growth and Development University of Oslo 57 / 68

58 Cross-Country Analysis Data Output: quantities of each crop produced and price of crops Intermediate inputs: fertilizer and seeds, quantity and price Land inputs: GPS measured field size very detailed information on land quality and rainfall Capital inputs: physical measure of agricultural items and market prices livestocks for agricultural production purposes Labor inputs: measured in hours including family labour, hired labour, and free labour Restuccia Macro Growth and Development University of Oslo 58 / 68

59 Cross-Country Analysis Data Focus on the cropping sector We have detailed measure of land quality: Nine elements of land quality: elevation, slope, terrain roughness, nutrient availability, nutrient retention, rooting conditions, excess salts, toxicity, and workability Rain shock: annual precipitation in millimetres We control for these items on output dispersion Restuccia Macro Growth and Development University of Oslo 59 / 68

60 Cross-Country Analysis Misallocation in Agriculture Land Input (log) Farm Productivity (log) Restuccia Macro Growth and Development University of Oslo 60 / 68

61 Cross-Country Analysis The Role of land Markets Farmers Operating with/without Rented Land Whole Sample No Rental Some Rental - (0%) (>0%) S.D. (log TFPR) S.D. (log MPLa) Efficiency gain Observations 2,672 2, Sample (%) Restuccia Macro Growth and Development University of Oslo 61 / 68

62 Cross-Country Analysis A Detailed Investigation of Land Rentals Standard Deviation (log) (1) (2) (3) Obs. MPLa No rental ,011 Some rental Non-market rental Some market rental TFPR No rental ,011 Some rental Non-market rental Some market rental Control variables: Land quality Yes Yes Yes Zone fixed effect Yes Yes Farm productivity Yes Restuccia Macro Growth and Development University of Oslo 62 / 68

63 Cross-Country Analysis Land Rental Markets across Locations Four levels of administrative divisions in Ethiopia: regions, zones (counties), woreda (districts), and kebele (wards). We mainly focus on the zone level analysis Reasonable number (56) of zones and enough observations (around 50) within each zone Exploit the variation of implementation across locations to study the impact of land rentals Approach similar to Banerjee-Iyer (2005) The percentage of land rentals differs from 0 to more than 60% at the zone level, smaller range for market rentals Re-compute the percentage of rentals as well as measures of misallocation and efficiency gains in each zone Restuccia Macro Growth and Development University of Oslo 63 / 68

64 Cross-Country Analysis Land Rental Markets across Locations Dispersion of Log MPL Rental Percentage Restuccia Macro Growth and Development University of Oslo 64 / 68

65 Cross-Country Analysis Variations of Rentals across Time Use two waves of data: Ethiopia and Aggregate rentals do not change over this time period, hence, explore household-level variation. Consider the following regression ξ zit = γ z + λ t + βd zt + ε zit, ( ) ξ zit : household level misallocation log TFPRzit TFPR or ( ) log MPLazit MPLa d z,2015 = 1 if zone z increases market rental in 2015; d zt = 0 otherwise. Restuccia Macro Growth and Development University of Oslo 65 / 68

66 Cross-Country Analysis Variations of Rentals across Time Dependent Variable: TFPR i MPLa i β (0.032) (0.030) Observations 4,422 4,422 R Restuccia Macro Growth and Development University of Oslo 66 / 68

67 Cross-Country Analysis Limited Role of Land Rentals Land sales prohibited by law in Ethiopia Given egalitarian distribution of land-use rights, rentals only mean to achieve a better land-use allocation Two features of rentals required to achieve an efficient allocation: (1) Substantial land rentals: 76% of land would be rented, whereas only 10.6 percent in data (2) Substantial concentration: Top 10% most productive farms would operate 95% of rented land, whereas only 19% in data Restuccia Macro Growth and Development University of Oslo 67 / 68

68 Cross-Country Analysis Take Away Market allocation leads to more efficient use of land resources across households, across locations, and across time Land certification reform intended to increase tenure security has not lead to substantially better resource allocation (most rentals within family and friends, many not market oriented) As a result, not expected to lead to within farm productivity growth Importance of assessing market oriented reforms Restuccia Macro Growth and Development University of Oslo 68 / 68