Two Blades of Grass: The Impact of the Green Revolution

Two Blades of Grass: The Impact of the Green Revolution Douglas Gollin 1 Casper Worm Hansen 2 Asger M. Wingender 3 Nairobi, July 7, 2017 1 Oxford University 2 University of Copenhagen 3 University of Copenhagen 1 / 1

Introduction Question: How does agricultural productivity growth affect aggregate economic growth? Prior empirical research has focused on local effects of agr. productivity growth (e.g., Foster and Rosenzweig, 2004; Bustos et al., 2016). We estimate the general equlibrium effect of adopting HYVs at the national level We analyze the diffusion of high-yielding crop varieties (HYVs) in 84 developing countries 1960-2000 Causal identification main challenge we use an IV strategy 2 / 1

Estimation equation Our main diff-in-diff estimation equation is: y it = β 0 + β 1 HYVs it + τ t + δ i + ε it, y it is the outcome variable (e.g., GDP/capita) in country i in year t. HYVs it is the HYV adoption rate for the 10 most important food crops (data from Evenson and Gollin 2003) τ t and δ i are year and country fixed effects Since HYVs it is endogenous, we construct an IV based on: 1 Agro-climatically attainable yields for HYVs (spatial variation in the intensity of the treatment). 2 The different timing of HYV research successes for different crops 3 / 1

Spatial variation: Agro-climatically attainable yields Idea: countries with high benefit from HYVs are more likely to adopt We measure the potential benefit by agro-climatically attainable yields (hypothetical maximum yields given climate when a high input level is used, i.e. HYVs, fertilizer, mechanization, and irrigation) 4 / 1

0 10 20 30 40 Adoption rate (percent of cropland) Time variation in development of HYVs The Green Revolution initiated by philanthropic efforts in the 1950s. First HYVs of wheat and rice available in the 1960s, HYVs of other crops at later dates Timing of research successes exogenous to individual countries. 1960 1970 1980 1990 2000 Year Rice Barley Wheat Cassava Notes: Adoption rate of HYVs in the average country in our sample 5 / 1

Constructing the instrument 1 Estimate crop-specific adoption rates by a DD strategy: HYVs j it = 2000 k=1965 α j k acy j i τ k t + τ t + δ i + ε j it acy j i is attainable yield of crop j spatial component of IV Interaction with year fixed effects, τ k t timing component of IV 2 Predict adoption rates for each of the 10 crops, ĤYVs j it. 3 Weight them by the crops share of harvested land in 1960: ĤYVs it = J ĤYVs j it harvested areaj i1960 j=1 harvested area i1960 6 / 1

Constructing the instrument 1 Estimate crop-specific adoption rates by a DD strategy: HYVs j it = 2000 k=1965 α j k acy j i τ k t + τ t + δ i + ε j it acy j i is attainable yield of crop j spatial component of IV Interaction with year fixed effects, τ k t timing component of IV 2 Predict adoption rates for each of the 10 crops, ĤYVs j it. 3 Weight them by the crops share of harvested land in 1960: ĤYVs it = 4 This is our instrument! J ĤYVs j it harvested areaj i1960 j=1 harvested area i1960 6 / 1

Main results: GDP/pc and population Table 3: The effect of HYV on population and GDP/capita (1) (2) (3) (4) (5) (6) Dependent Variable (in logs): GDP/capita Population Actual HYV adoption 0.987*** 1.482*** -0.198*** -0.543*** (0.178) (0.402) (0.0701) (0.178) Predicted HYV adoption 1.801*** -0.659*** (0.539) (0.198) Observations 420 420 420 420 420 420 Countries 84 84 84 84 84 84 Estimator OLS 2SLS OLS OLS 2SLS OLS Kleibergen-Paap. 25.98.. 25.98. 2SLS results: HYV adoption up by 10pp GDP/cap. up by 15%, and population down by 5%. 7 / 1

Robustness checks Controls (Interacted with time dummies): Geographic confounders (e.g., climate and agricultural suitability) Distance (or presence) of research centers Asian regional dummies. Initial values: outcomes, constraint on the executive, population density, openness to trade. Cash crops (cotton, soybeans and sugar) Time periods: Other: Unbalanced panel from 1940. Subperiods (e.g., stop in 1980). Country specific linear time trends. Alternative specifications of instrument. Outliers Falsification checks of IV strategy 8 / 1

Channels Table 9: The agricultural sector (1) (2) (3) (4) (5) Dependent variable (in logs): Yield/ Harvested Fertilizer/ Agri. Agri. employworker area hectare population ment share Actual HYV adoption 1.919*** -0.538* 2.162** -1.339*** -0.767*** (0.468) (0.326) (0.905) (0.338) (0.252) Observations 405 405 405 405 405 Countries 81 81 81 81 81 Estimator 2SLS 2SLS 2SLS 2SLS 2SLS Kleibergen-Paap 24.75 24.75 24.75 24.75 24.75 HYVs increased agricultural productivity Direct effect on yields Increased use of other inputs (e.g., fertilizer, capital) 9 / 1

Demographics Table 10: Demographic effects (1) (2) (3) (4) (5) (6) (7) Dependent variable: (in logs) (in rates) Life Infant Adult Mortality Fertility Rate of Pop- Expec- mortality female male rate natural ulation tancy increase growth Actual HYV adoption 0.134-1.958*** -1.689*** -0.996*** -1.524*** -0.270*** -0.274*** (0.0870) (0.382) (0.314) (0.254) (0.293) (0.0636) (0.0774) Observations 420 381 420 420 420 420 420 Countries 84 84 84 84 84 84 84 Kleibergen-Paap 25.98 23.70 25.98 25.98 25.98 25.98 25.98 HYV adoption decreases mortality (col. 1-4), especially for females. Fertility decreases more (col. 5) population growth No Malthusian effect. No significant effect on migration (col. 6 minus col. 7). 10 / 1

Implications: The effect of the Green Revolution Table 11: The impact of HYV adoption 1960-2000 Average developing country Developing world total HYV adoption rate year 2000 27 % 59 % HYVs contribution to GDP/cap. growth 50 pp 139 pp Actual GDP/capita growth 57 % 171 % HYVs contribution to population growth -14 pp -27 pp Actual population growth 162 % 129 % The Green Revolution had a huge positive impact on economic growth from 1960-2000. It also reduced population growth Catch-all GE effect: All amplifying effects part of the estimated effect of adopting HYVs 11 / 1

Thank you 12 / 1