Successfully reported this slideshow.
We use your LinkedIn profile and activity data to personalize ads and to show you more relevant ads. You can change your ad preferences anytime.

Day 3.1 robinson impact3-gfsf-rome-may-2015-sr2


Published on


Published in: Government & Nonprofit
  • Be the first to comment

  • Be the first to like this

Day 3.1 robinson impact3-gfsf-rome-may-2015-sr2

  1. 1. IFPRI IMPACT 3 Model System: Modularity Sherman Robinson IMPACT Model Team International Food Policy Research Institute (IFPRI) Rome, GFSF Team Meeting, May 2015
  2. 2. 5 The IMPACT 3 Model  International Model for Policy Analysis of Agricultural Commodities and Trade  Need for a multi-disciplinary approach: • CGIAR and other collaborators: – Economics, agronomy, hydrology, livestock, fish, crop models, nutrition/health • Civil engineering: infrastructure • Climate change (GCMs) • Energy (biofuels, inputs)
  3. 3. IMPACT version 3 • 58 Agricultural commodities 6
  4. 4. IMPACT 3 Model System 7
  5. 5. IMPACT 3.2: A Suite of Models  Multimarket model • Core global PE model  SPAM: • Spatial Production Allocation Model  Land-Use  DSSAT Crop Models  Welfare  Water models • Hydrology • Water Basin Management • Water Stress on yields  Sugar and oilseeds • Processing value chains  Livestock/meat/dairy • Current version running  Nutrition/health: • Current version running 8
  6. 6. 9 IMPACT 3.2 to 3.3: Improvements  Data on policies: tariffs and subsidies (GTAP data, OECD PSE/CSE data • Incorporated into IMPACT 3.2 data set  Review of productivity growth trends and model improvements by CGIAR centers • Rome GFSF meeting, May, 2015  Menu of possible model improvements • Priority setting
  7. 7. 10 IMPACT 3.3: Potential Improvements  Livestock module: under development with ILRI  Fish module: joint work with World Fish • Two stage work program underway  Linked global CGE model: joint work with IDS  Links to environmental models • Biodiversity: IFPRI and Bioversity • GHG emissions, nitrogen use efficiency: IFPRI  Water model improvements: • Ground water • Water quality
  8. 8. 11 IMPACT 3.3: Potential Improvements  Micronutrient module: IFPRI, PHND, A4NH and others • Under development  Health module: with Oxford (Martin Centre) • Under development  Land module: land supply and demand by type • Under development at IFPRI  Variability and extreme events • Work with UK/US collaborators (sponsored by Lloyds)
  9. 9. 12 Nutrition Module  IMPACT model solves for supply/demand for agricultural commodities in physical units • Nutrition module: post solution calculation of nutrition impacts  Nutrition content measured at the agricultural commodity level--extensive empirical work • FAO: Food balance sheets. Nutrition content of various agricultural commodities, focusing on “energy” (calories) • IFPRI (PHND, Haddad-Smith, Derek Headey), CSIRO (Mario Herrero), CIMSANS (Jerry Nelson), Oxford (Marco Springmann), Nestle Research – Extending food balance sheets to include more nutrients 12
  10. 10. 13 Nutrition Module: Regression Model  Reduced-form regression models • Statistical models linking nutrient supply at the commodity level to nutritional status at the household/ consumer level • Haddad/Smith cross-country regression model is currently used in IMPACT to calculate nutrition outcomes, focusing on energy – Nutrient supply is one variable among many in the model, but is the only variable linked to IMPACT – They are updating the regression model to include more nutrients and outcomes (obesity as well as under nourished) 13
  11. 11. 14 Nutrition Outcomes  Since actual household demand is for processed commodities (e.g., bread, not wheat), food balance sheets measure supply of nutrients, not what is actually consumed at the household level • Haddad-Smith regression model skips value chain to processed food commodities • Nestle: EcodEx (Product Ecodesign Tool) considers value chains to processed commodities and 32 nutrients • Tilman et al.0(2011), “Global food demand and the sustainable intensification of agriculture” – Demand functions for “nutrients”
  12. 12. 15 Nutrition: • Oxford Martin Programme on the Future of Food, study linking IMPACT model results for fruit and vegetable and red-meat consumption on body weight and health outcomes using a Markov model and detailed information on nutritional content of foods – Springmann et al. paper – Health outcomes linked more to consumption of fruits and vegetables (micronutrients) than to energy (calories) 15
  13. 13. 16 Production to Nutrition  To support more structural models of food demand and nutritional status requires specifying the value chain in both PE and CGE models • Wheat to flour to bread/pasta/cake to retail sector to households • Production of “other” food commodities such as beverages, fish products, etc.  Extensive data on nutritional content of processed food commodities. • Feasible to use these data in models? 16
  14. 14. 17 Demand, Utility, and Nutrition  How to link nutritional status and commodity demand functions: • Current practice. Add equations determining nutritional status as an “add on” after commodity demands have been determined based – Data are available, both at the agricultural and food commodity levels. Current treatment is to work at the agricultural commodity level. – Thin links with utility/demand theory. Households demand food commodities, not agricultural products or nutrients. Only an indirect link with nutritional status indicators. 17
  15. 15. 18 Demand, Utility, and Nutrition  Add nutrition indicators to the utility function, with implications of commodity demand • Hedonic quantity/price indices: consumers do not demand nutrients, but commodities with nutritional attributes – K. Lancaster, “A New Approach to Consumer Theory”, 1966  Add nutritional status indicators as constraints in the utility optimization problem • Integrate the classic LP “diet problem” with an NLP utility maximization problem 18
  16. 16. 19 IMPACT 3.4: Potential New Modules  Production: better specification of technology and supply • Optimization given production/cost functions (CGE models) • Activity/process specification of production/costs – GLOBIOM (IIASA), MAgPIE (PIK, Potsdam) • Stylized “farm” simulation models  Value chains: more “processing” activities to move from crops/livestock to marketed “commodities” • From cows to hamburgers & milk (livestock module) • From wheat and corn to Wheaties and Cornflakes • Cassava: food vs industrial demand, tradability
  17. 17. 20 Current Value Chain Modules  Oilseeds and sugar (integrated in IMPACT) • Processing from crops (sugar cane/beet, various oil seeds) to “commodities”: processed sugar, oils, meal • Simple cost pricing: “markup” on cost of crop inputs • Implicit assumption of competitive markets  Livestock (standalone module and integrated) • Value chain from herds to dressed meat, eggs, milk • New livestock module: elaborate specification of feed inputs and livestock production “systems” • Simple model of commodity production: “markup” on inputs • Implicit assumption of competitive markets
  18. 18. 21 New Value Chain Modules  Interest in expanding range of value chain modules • Cassava, fish, wheat/maize/rice/soy beans • Welfare analysis using demand curves for intermediate inputs is problematic—consumer surplus calculation is suspect • Links to nutrition analysis: more detail on food commodities • Simple specification of competitive markets is suspect – E.g., sugar  Combine value chain with industry studies • “Structure, conduct, performance” analysis • Schmalensee and Willig: Handbook of Industiral Organization
  19. 19. 22 Linked Global CGE Model  New project: link IMPACT 3 with the GLOBE CGE model • GLOBE is based on GTAP data and written in GAMS • Includes activity/commodity distinction, as in IMPACT 3  One-way links: IMPACT to GLOBE • Crop/livestock production from IMPACT 3 passed to GLOBE, which then is run assuming those outputs are fixed • GLOBE solves for economywide impacts (direct and indirect links): production, employment, and prices • All welfare analysis is done in GLOBE (EV/CV, total absorption) • Links to labor markets, wages, and poverty done in GLOBE
  20. 20. 23 Linked Global CGE Model  Two-way links: IMPACT to/from GLOBE • Agricultural output from IMPACT: GLOBE generates GDP originating in agriculture, and changes in total GDP • GDP from GLOBE sent back to IMPACT, so GDP in IMPACT reflects changes in agricultural productivity – Currently, GDP is exogenous in IMPACT  GLOBE and IMPACT need not run on the same time step • Both can be annual, but can run on different multiyear time steps (e.g., annual for IMPACT, every 5 years for GLOBE)  GLOBE linked via a standalone module that takes input from IMPACT and runs GLOBE
  21. 21. 24 Advantages of Modularity  “Standalone” modules can be run independently of IMPACT, but use inputs from IMPACT scenarios • Can be developed, calibrated, and tested by specialists (e.g, from various CGIAR centers). • Designed to be used in Center research programs  Design: separate modules can reflect their disciplines • No need to compromise to “fit” one model into another • E.g. water in economic models or economics in water models— always unsatisfactory  Model development, testing, and debugging is greatly facilitated if the modules can be run separately
  22. 22. 25 Desiderata for Modular Model Systems “Modules” should be designed to:  Operate in “standalone” mode;  Read its own parameters;  Initialize its own variables;  Accept variables/parameters passed to it from other modules and the environment;  Pass variables that are computed within the module to other modules or the main model;  Own its set of state variables;
  23. 23. 26 Modularity: Linking Modules  Modularity; “a la carte” model system • Use the models you need, turn off those you do not need • Separate models can be run independently • Modules can run with different time steps  Standardize data transfer • Information flows • Dynamic or iterative interaction  “Data driven” model specification • IMPACT 3 multimarket model can be run at any level of aggregation without changing the model code • Change input data and sets only: user need not even see the GAMS code
  24. 24. 27 Modularity: Linking Modules  Three ways to link modules: • Exogenous: Information flows in one direction – To IMPACT: hydrology, DSSAT, GCMs, SPAM – From IMPACT: welfare, nutrition/health, GLOBE/CGE • Linked dynamically: Two-way information flow between years – Water basin management, water stress on crops – Land use by type – GDP/economywide links: GLOBE • Endogenous: Module equations are solved simultaneously – Livestock, sugar processing, oilseeds/oils – Land allocation to crops
  25. 25. 28 IMPACT 3 Modules  Standalone modules, one-way links: • Welfare, nutrition, GLOBE (e.g., welfare, economywide impacts), hydrology, DSSAT, GCMs  Standalone modules, inter-period links: • Water models (IWSM, water stress), land use (by land type), livestock (herds), GLOBE (e.g., GDP, non-ag prices)  Standalone modules, intra-period links: • Land use (cropping, irrigated/rainfed), Livestock  Value chains, within IMPACT: sugar, oilseeds, livestock
  26. 26. 29 Water Models in IMPACT  Global hydrological module (GHM) assesses water availability  IMPACT Water Simulation Module (IWSM) optimizes water supply according to demands • Monthly time step • Domestic, industrial (linked to GDP/population) • Livestock, environmental, and irrigation demands • Optimizing model for irrigation demand/supply  Water stress module • Optimizing model: allocation of water to crops • Deliver crop yields to the IMPACT multimarket model
  27. 27. Water: Two-Way Model Integration Food Model • Crop areas • Population • GDP • Livestock numbers • Prices Water Models • Water supply • Water Stress: shock on crop yields Solve multimarket model given trends and variable crop areas Fix crop areas and livestock; call the water models: solve for water stress yields Re-solve the multimarket model with fixed crop areas and stress yields 32 In each year, solve in two steps:
  28. 28. 33 Standalone IMPACT Module: Template  GAMS IMPACT-compatible standalone module • Include file with definition of relevant IMPACT parameters • Include GDX file(s) of scenario output of IMPACT results • Load IMPACT data needed by the module  Data estimation and management • Module has its own data base, in addition to IMPACT data  Model specification and parameterization • If module is to be integrated with IMPACT, must avoid name collisions for parameters, variables, and equations  Linking to IMPACT 3 • Communication: exogenous, intra-period, within-period