Simulation Models in Economics: Issues, Design, and Implementation Sherman Robinson International Food Policy Research Institute (IFPRI)

Simulation Models in Economics: Issues, Design, and Implementation

Sherman RobinsonInternational Food Policy Research Institute (IFPRI)

Outline

• Simulation models: – Types– issues– design– Implementation

• Impact model• CGE models • Estimation and validation

2

Simulation Models

• Long history in economics– Econometric Models used in “simulation mode”– Models designed for simulation

• Level of aggregation– World models– Country models– Regional/sub-regional models– Enterprise/farm models

3

Types of Simulation Models

• Stylized: “putting numbers to theory”– Small, focused models—close to theory

• Applied– Larger, more detail (including institutions)– Broader range of issues

• Policy models– Explicit links between policy parameters and

economic outcomes

4

Types of Simulation Models

• “Reduced form” versus “structural”• Dynamic versus static• Partial versus general equilibrium• Coverage

– household/village/region/country/globe• Domain of application

– “Universe” of the model

5

“Reduced Form” Models

• Vague theoretical specification of relationships among variables – Econometric estimation: hypothesis testing– Unidentified/unidentifiable structural model

• Simulation mode: forecasting – E.g., macroeconometric models– Goal is to forecast endogenous variables, given

projections of exogenous variables– Less interested in “how” the economy works

6

Structural Models

• Goal is to simulate “how” the economy works– “Counterfactual” analysis: “What if” scenarios– Controlled experiments: parameters/policies

• causal chains/large numbers

• Model elements– Specify agents, technology, markets, institutions,

signals, motivation, and behavior– “Domain” of the model

7

Structural Models

• Model elements: structural models– Agents interacting, usually across markets– Specification of agent behavior– Specify institutional structure– Notions of equilibrium

• Partial versus general equilibrium• Static versus dynamic

8

Structural Models

• Partial equilibrium: commodity models– Single market models– Multimarket models

• Economywide models– “Economy” may vary in size and domain– Macro models– General equilibrium models– Microsimulation household models

9

Structural Models

• In a structural model, must specify: – Agents (producers, households)

• Economic actors in the model

– Motivation (profit maximizing producers, utility maximizing consumers)

– Signals (prices in markets)– Institutional structure (competitive markets)

• “Rules of the game”

10

Structural Models

• Describe agent behavior mathematically– Producers: supply behavior

• Production/cost functions, profit maximization– Input demand (K, L, Land, intermediate inputs)

• Supply curves (marginal cost function?)

– Consumers: demand behavior• Utility functions, utility maximization

– Income, expenditure equations

• Demand curves (Marshallian?)

11

Deep/Shallow Structural Models • “Deep” structural models

– explicit description of agent behavior– Utility functions, production/cost functions– Relevant factor and commodity markets

• “Shallow” structural models– Supply/demand functions which summarize agent

behavior (“reduced form” equations) – Only loosely based on theory

12

Structural Models

• Agent based models: – Opportunity, motive, ability– Not enough to describe operation of the economy

• Additional “constraints” on the economy– System constraints

• Supplies of primary factors (land, labor, capital)

– Equilibrium conditions• Supply-demand balance in all markets

13

Structural Models

• Market equilibrium: how markets work– Equilibrium conditions

• Supply = demand

– Equilibrating mechanisms • Price responsive supply and demand functions• International trade

– Equilibrating variables• Commodity and factor prices, domestic and global

14

Market Equilibrium in Models• A descriptive feature: If market clearing is a

reasonable assumption, then we can use the specification to describe a realistic result– Solve for market-clearing prices in the model,

which then correspond to actual prices• No need to specify the exact process by which

markets equilibrate, just the result– Powerful tool to simplify structural models

15

Partial Equilibrium Models

• Single commodity or multimarket– Do not cover the entire economy

• Supply and demand curves– Linear or nonlinear, loosely based on theory– Expenditure functions may or may not be based on

demand theory– “Shallow” structural models: reduced form equations

16

Simulation Models: Issues

• Growth and structural change– Investment/education– Role of trade– Productivity growth– Agriculture/water/land– Industrialization

• Long-run development strategies

18

Simulation Models : Issues

• Macro shocks and structural adjustment• Income distribution

– Long run: poverty and growth– Short run: impact of macro adjustment

• Fiscal policy– Tax system design and/or reform– Government expenditure policy

19

Simulation Models : Issues

• Globalization– Trade policy reform: GATT/WTO– Regional trade agreements

• Customs unions: EU, Mercosur• FTA’s: NAFTA, bilaterals, etc.• Preferential access: Cotonou, EBA, AGOA,etc

– Domestic policy reforms and trade system• Impact of OECD agricultural policies

20

Simulation Models: Issues

• Energy – Energy “system” and the economy– Oil price shocks– Biofuels

• Environment/climate change– Costs of environmental policy– Climate change: mitigation/adaptation

21

Model Design: Aggregation

• Macro (aggregates: C, I, G, E, M)– Macroeconometric models– Asset markets and financial variables

• Micro (household/firm/farm analysis)– Microsimulation models

• Mezzo (sectors: multi-market and CGE)– Structure of production, employment, trade, etc.

22

Implementation: Construction

• Explicit mathematical statement of theoretical model– Specify functional forms, endogenous variables,

parameters, and exogenous variables– Transforms inputs to outputs

• Computer code: modeling languages– GAMS, Matlab, Mathematica, Stella, Vensim,

system dynamics

23

Implementation: Validation

• Validation is linked to issues to be analyzed– Focus of the model application– Intended “domain of applicability” of the model

• Need to “test” the model with historical data relevant to its domain of applicability– How well does the model “explain” past events?– How well does it capture the important causal

chains? Validity of the underlying deep/shallow structural model

24

Multi-Market: IMPACT Model

• Impact is a suite of models: – Core Impact multi-market global trade model – “Water" model of FPU river basins,– “Water stress" model that converts hydrological

output into yield shocks – Crop models – Biofuels, livestock, and fish models– Links to GCM climate change models

25

Economywide CGE Models

• “General equilibrium”: many markets, factors and commodities– Simultaneous equilibrium across inter-dependent

markets• “Behavior” consistent with general

equilibrium theory– Deep structural relations

26

CGE Model Design: Theory

• Walras-neoclassical-structuralist-Keynes: theoretical roots– Role of product and factor markets– Role of assets and financial markets

• Dynamic versus static– Time horizon: short, medium, long– Notion of equilibrium: flows and stocks

• Rational expectations, forward looking, etc.

27

CGE Models

• Numerical application of the Walrasian general equilibrium model – Market economy where a many agents maximize their

objective functions (utility or profit) subject to their constraints (budget or technology)

– Single-period, static model • Equilibrium model

– No global objective function– Optimizing, price-responsive behavior of individual actors– Complete specification of both supply and demand sides of all

markets (goods and factors)

28

Background

• Johansen 1960: MSG Model of Norway– Still used for planning and forecasting

• 1970s: Confined mostly to universities and research institutes

• 1980s and beyond: wider use (including government agencies in many countries)

29

30

What do we want to capture?

Factor markets Factor market functioning SegmentationWage determination

Economywide

Environment

Households

Structural featuresBinding macro constraintsGeneral Equilibrium effects

HeterogeneityHuman and physical

capitalDemographic

CompositionPreferencesAccess to Markets

Typical CGE Model Features

• Simulation model– No forecasting or macro cyclical analysis

• “Micro-macro” model in structure– Explicit specification of micro/agent behavior– Simultaneous economywide and micro outcomes

• Set up in “real” terms: – No asset markets, – Money is neutral, – Decisions are a function of relative prices

• Representative household assumption

31

CGE Models

• Actors: producers, consumers, government, rest of the world

• Motivation: profit maximization, utility maximization

• Institutions and signals: competitive markets and prices

• Agent constraints: technology, factor endowments (budget constraints)

32

CGE Models

• System constraints: – Resources (land, labor, capital), – International: foreign trade balance

• Equilibrium conditions: – Supply-demand balance in all markets– Macro balances: government, savings-investment,

foreign trade balance

33

Stylized Model Structure

34

Activities

Commodity Markets

Factor Markets

Rest of the World

Households Government Sav./Inv.

FactorCosts

Wages& Rents

IntermediateInput Cost

Sales

PrivateConsumption

Taxes

Domestic Private Savings

GovernmentConsumption

Gov. Savings

Investment Demand

ImportsExports

Foreign Savings

Transfers

Foreign Transfers

35

SAM Structure

Expenditures

Receipts Activities Commodities Factors

Domestic Institutions

Rest of World

Totals

Activities Market sales

Home con- sumption

Activity income

Commodities

Intermediate

inputs

Trans- actions costs

Final

market demands

Exports Commodity

demand

Factors Value added

Transfers Factor income

Domestic Institutions

Taxes Tariffs, Taxes

Income, Taxes

Transfers, Taxes,

Savings

Transfers, Savings

Institution income

Rest of World

Imports Foreign

exchange outflow

Totals Activity

spending Commodity

supply Factor

spending Institution spending

Foreign exchange

inflow

Solving CGE Models

• Direct approaches– Scarf algorithm– Log linearization (Johansen, Orani, GTAP)– Simultaneous nonlinear equations

• Scarf algorithm.• Tâtonnement algorithms• Newton techniques (GAMS)

• Optimization methods– Negishi Theorem (Ginsburgh-Waelbroeck-Keyzer)– Nonlinear programming problem (NLP)– Shadow prices = market prices

36

Calibration of CGE Models

• Equivalent to a “backward” solution of the model in order to determine the set of parameter values consistent with the initial structure of the economy.

• Assume that the initial data (e.g., SAM) represent an equilibrium model solution.– Share parameters from SAM data.– Elasticity parameters from other sources.

37

Estimation and Validation

• Define “domain of applicability” of model• Econometric models: simultaneous estimation

and validation– Sample data used for both parameter estimation and

within-sample “prediction” of endogenous variables (validation). With lots of data, one can save some data for separate validation exercise.

• Notion of “information” for estimation and validation

38

Estimation and Validation

• Structural versus reduced-form models– “Deep” behavioral parameters for structural

simulation models• Tastes, technology, and institutions

– Issue of use of prior information about parameters in estimation

• Separation of estimation and validation• Not enough data to do both simultaneously• Need to use variety of information

39

Estimation and Validation

• Estimation using MaxEnt econometrics– Zellner: “Efficient” information processing rule.

Use all, but only, the information available. Do not assume information you do not have.

– Use of prior information on parameters• Bayesian in spirit, but not formal Bayesian estimation• Distinction between “precision” and “prediction”

– Tradeoffs, different from classical regression analysis

40

Conclusion

• Gap between theory and empirical implementation has narrowed

• Simulation models are widely used, and will become even more common

• Advances in econometrics applicable to structural parameter estimation:– Information theoretic estimation methods

41