Economy-Wide Models and Poverty Analysis

1
Economy-Wide Models and Poverty Analysis

• Sherman Robinson, IDS, Sussex
• Hans Lofgren, World Bank, Washington, D.C.
• Presentation at the WIDER Conference Frontiers
  of Poverty Analysis, held in Helsinki, Finland,
  September 26-27, 2008

2
Introduction

• Policy Issues Poverty and MDGs
• Economy-Wide framework
• Link development strategy choices with poverty
  and MDG outcomes
• Top-down versus bottom-up approaches
• Analytic and empirical models
• CGE models with representative households
• Households and microsimulation models
• MDG links

3
What do we want to capture?
Macroeconomic Environment
Structural features Binding macro
constraints General Equilibrium effects

• Factor markets

• Factor market functioning
• Segmentation Wage determination

Heterogeneity Human and physical
capital Demographic Composition Preferences
Access to Markets
Households
4
Typical model structure
Factor markets
Activities
Households
Commodity markets
Rest of world
5
A National SAM
Expenditures
Receipts
Activity
Factors
Institutions
World
Hshlds
Commodity
Activity
Total sales
Intermediate inputs
Final demand
Exports
Commodity
Consumption
Factors
Value added
Factor income
Indirect taxes and tariffs
Saving taxes
Saving taxes
Institutions
Inflows
Indirect taxes
Payments to hshlds
Factor income
Hshlds
remittances
transfers
World
Imports
Foreign Exchange inflow
Factor income
Institution income
Hshld income
Totals
Total costs
Total supply
6
CGE Models

• CGE models are widely applied to policy analysis
  both in developed and in developing countries.
• Many applications in trade policy
• Standard static and dynamic models
• Various approaches to incorporating
  distributional features poverty and MDGs

7
Modeling Impact of shocks on poverty and income
distribution

• Shocks include
• Macro shocks and structural adjustment
• Trade reform country, regional, global
• MDG development strategies
• Most existing studies of the distributive effects
  of shocks rely either on
• comparison of distributions before and after the
  shock,
• counterfactuals based on macro models with some
  disaggregation of the household sector.

8
Counterfactual Analysis

• Links between policy changes and impacts
• Single policies and strategies (e.g., MDGs)
• Decomposing the impact of shocks
• Exogenous shocks (e.g., Asian crisis)
• Policy responses
• Historical analysis analyzing causes
• Turkey 1972-77 Dervis, de Melo, and Robinson
• Model as a measurement device

9
Poverty Analysis

• Two basic approaches
• Representative households in country models
• Summary representation of income distribution
  within household groups
• Microsimulation models at household level
• Separate, linked, or integrated with economy-wide
  models

10
Representative Households

• Standard approach in early income-distribution
  focused CGE models. (Adelman-Robinson. Recent
  Decaluwe et al.)
• First, classify households into representative
  groups.
• Second, assume that the relative within-group
  income distribution for each representative group
  does not change, given the shocks being analyzed.
• CGE model generates changes in group mean
  incomes.
• Distributional statistics generated by
  aggregating within-group distributions
• Generate standard distribution/poverty measures

11
Representative Households

• Linking activities, factor incomes, households
• Functional distribution
• Extended functional distribution
• Livelihood strategies by households
• Household disaggregation
• Within-group distributions not affected, so model
  cannot explain or affect much poverty/inequality
• Representative to full household surveys

12
Microsimulation

• ... instead of aggregating observations within a
  household survey into a few household groups in
  conformity with the requirements of CGE-type
  models, our aim should be to work directly with
  all the individual observations of the survey. By
  doing so, we hope to achieve full consistency
  between macroeconomic reasoning and standard
  poverty evaluation. Bourguignon, 1999.

13
Microsimulation

• Integrated CGE - microsimulation model
• Top-down approach
• sequential approach with CGE model feeding
  microsimulation with price and income data
• No feedback from micro to macro levels
• Different degrees of complexity at the
  microsimulation level
• Models of household behavior

14
Simple Top-Down Approach

• Link model results to a household survey.
• Survey households are classified the same as
  representative households in CGE model.
• CGE model generates incomes and prices
• Individual survey observations scaled using
  simulated changes in representative household
  income and prices.
• Distributional measures computed from adjusted
  survey.

15
Microsimulation

• The essence of microsimulation is to model the
  behavior of individual agents (households or
  firms) that are included in a micro database.
• In order to extend the analysis from
  partial-equilibrium issues, microsimulation
  models can be linked to CGE models.
• Potential to link economy wide shocks to
  household outcomes

16
The Sequential Framework
Macro-level module (Extended CGE-type model) -
Occupational structure ?L - Price variables
?p - Wage and earnings ?w - All other
variables in macro module ?Y
Link variables ?L, ?w
Micro-simulation module (Household survey) -
Socio demographic characteristics Si -
Occupational/labor-supply choice li O(Si,?) -
Income yi E(Si,?).li Consistency with
macro. Find changes in parameters ? and ? such
that ??li ?L and ?Mean E(Si,?) ?w Outcome
change in distribution of income conditionally on
characteristics S.
17
Indonesia Model

• MACRO
• Static Computable General Equilibrium
• Software GAMS
• Data Social Accounting Matrix for 1995
• 38 sectors
• 5 agricultural
• 15 informal
• 18 formal
• 15 factors of production
• 8 types of labor
• 7 types of capital
• Segmented labor markets
• Working capital credit constraints

• MICRO
• Reduced Form occupational choice microsimulation
• Software STATA
• Data Savings-Investment module of SUSENAS 1996
• 9,800 households
• 33,400 individuals aged 10 years and older
• 8 labor segments
• urban/rural
• male/female
• skilled/unskilled
• 4 occupational choices
• inactive
• wage worker
• self employed
• wage worker self employed

18
UNDP Latin America Studies

• UNDP Projects on Latin America
• Trade 16 country studies
• Top-down CGE - microsimulation
• CGE models with focus on international trade
• Various degrees of household disaggregation
• Income, employment, and prices sent down
• Limited behavior at household level
• Latin American school of microsimulation

19
UNDP Latin America Studies

• UN Projects on Latin America
• Trade with IFPRI, 16 country studies
• MDG Strategies with World Bank 18 country
  studies
• Top-down CGE - microsimulation
• CGE models with focus on international trade
• Various degrees of household disaggregation
• Income, employment, and prices sent down
• Limited behavior at household level
• Latin American school of microsimulation

19
20
The Integrated Framework
Macro-level module (Extended CGE-type model) -
Occupational structure ?L - Price variables
?p - Wage and earnings ?w - All other
variables in macro module ?Y
Micro-simulation module (Household survey) -
Socio demographic characteristics Si -
Occupational/labor-supply choice li O(Si,E) -
Income yi E(Si,w).li E earning rate of
individual/household i in various occupations.
These personal rates are a function of a set of
standard market rates, w. Outcome change in
distribution of income conditionally on
characteristics S. Aggregating ??li ?L
21
Integrated micro-macro model Madagascar

• Cogneau and Robilliard. Model in Gauss.

• MACRO
• Stylized CGE Framework endogenous prices for
  labor and goods markets
• Data Social Accounting Matrix for 1995
• 3 goods
• 1 agricultural
• 1 informal
• 1 formal
• 5 factors of production
• 3 types of labor
• 2 types of capital

• MICRO
• Structural labor allocation model Endogenous
  occupational choice and time allocation
• Data Enquete Permanente aupres des Menages 1993
• 4,500 households
• 12,000 individuals aged 15 years and older
• 4 occupational choices at the household level
• farmer
• informal wage worker
• formal wage worker (rationed)
• farmer wage worker

22
Microsimulation vs RH models

• Comparing RH models and microsimulation
• When does disaggregation matter?
• Household impact and behavior
• If shocks affects variables such as prices or
  average wages, RH models do fine
• If shocks affect employment or discontinuous
  household behavior, microsimulation matters
• Labor participation
• Distinction is not sharp continuum of models

23
MAMS Maquette for MDG Simulations

• Dynamic-recursive CGE Model for MDG analysis
  developed at World Bank
• Initial motivation need to address country-level
  MDG strategies How can government policies, with
  foreign aid providing part of the financing, be
  designed for achievement of the MDGs?
• Evolved into general framework for country-level,
  medium-to-long-run development policy analysis,
  with emphasis on fiscal issues and MDGs.
• Different versions (differing in data needs and
  issues they can address) ranging from aggregated
  macro to disaggregated MDG.
• Starting point standard dynamic-recursive CGE
  model
• Main innovation covers the generation of MDG and
  education outcomes.
• MAMS has been used with the standard approaches
  to poverty and inequality analysis.

23
24
MAMS

• Applications in many countries
• 18 in Latin America and the Caribbean
• 9 in Sub-Saharan Africa
• 5 in MENA region
• Used in the context World Bank country analysis
  (including Country Economic Memoranda, Public
  Expenditure Reviews, Poverty Assessments) as well
  as in joint work with the UN (UN-DESA and UNDP)
  on Latin America and the MENA region.

25
Issues in MDG strategy analysis

• A framework for analysis of MDG strategies should
  consider the following factors
• Synergies between different MDGs
• Role of non-government service providers
• Demand-side conditions (incentives,
  infrastructure, incomes)
• Role of economic growth
• Macro consequences of increased government
  spending under different financing scenarios
• Diminishing marginal returns (in terms of MDG
  indicators) to services and other determinants
• Role of efficiency and input prices (e.g. wages)
  in determining unit service costs

26
MAMS Model Structure

• An extended, dynamic-recursive computable general
  equilibrium (CGE) model designed for MDG
  analysis.
• Complementary to and draws extensively on sector
  and econometric research on MDGs.
• Motivation behind the design of MAMS
• An economywide, flexible-price model is required
  for development strategy analysis.
• Standard CGE models provide a good starting
  point.
• But standard CGE approach must be complemented by
  a satisfactory representation of social
  sectors.

27
MAMS Model Structure
2. Model Structure

• Many features are familiar from other CGE models
• Computable ? solvable numerically
• General ? economy-wide
• Equilibrium ?
• optimizing agents have found their best solutions
  subject to their budget constraints
• quantities demanded quantities supplied in
  factor and commodity markets
• macroeconomic balance
• Dynamic-recursive ? the solution in any time
  period depends on current and past periods, not
  the future.
• A real model only relative prices matter no
  modeling of inflation or the monetary sector.

28
MAMS Model Structure
2. Model Structure

• Extended to capture the generation of MDG
  outcomes.
• MAMS covers MDGs 1 (poverty), 2 (primary school
  completion), 4 (under-five mortality rate), 5
  (maternal mortality rate), 7a (water access), and
  7b (sanitation access).
• The main originality of MAMS compared to standard
  CGE models is the inclusion of (MDG-related)
  social services and their impact on the rest of
  the economy.
• Social services may be produced by the government
  and the private sector.

29
MAMS Role of Government
2. Model Structure

• Government services are produced using labor,
  capital, and intermediates (fixed coefficients
  for capital, intermediate inputs, and aggregate
  labor flexible coefficients for disaggregated
  labor).
• Government spending is split into
• Recurrent consumption, transfers, interest
• Capital (investment)
• Government demand (consumption and investment) is
  classified by function social services
  (education, health, water-sanitation),
  infrastructure and other government.
• Government spending is financed by taxes,
  domestic borrowing, foreign borrowing, and
  foreign grants.
• Model tracks government domestic and foreign debt
  stocks (including foreign debt relief) and
  related interest payments.
• Simplified versions of equations for government
  recurrent receipts, recurrent expenditure,
  savings, and investment expenditure..

30
MAMS MDG production
2. Model Structure

• Together with other determinants, government
  social services determine the "production" of
  MDGs.
• MDGs are modeled as being produced by a
  combination of factors or determinants (table
  following) using a (reduced) functional form that
  permits
• Imposition of limits (maximum or minimum) given
  by logic or country experiences
• Replication of base-year values and elasticities
• Calibration of a reference time path for
  achieving MDGs
• Diminishing marginal returns to the inputs
• Two-level function
• Constant-elasticity function at the bottom Z
  f(X)
• Logistic function at the top MDG g(Z)

31
MAMS Data Requirements

• Core needs are similar to other CGE models
• Social Accounting Matrix (SAM) stocks of
  factors, population, and debts (foreign and
  domestic) elasticities in trade, production, and
  consumption
• They depend on the (flexible) disaggregation of
  the model.
• The SAM is used to define most of these
  parameters.
• Requirements specific to MDG version
• In SAM government consumption and investment
  disaggregated by MDG-related functions labor
  disaggregated by educational achievement
• Education parameters stocks of students by
  educational cycle student behavioral patterns
  (ex rates of passing, repetition, dropout)
  population data with some disaggregation by age
• MDG data indicators for base-year and 1990
  elasticities calibration scenario for achieving
  each MDG.

32
MAMS Data

• Database draws on a wide range of sources.
• Likely key sources
• Standard national data publications (national
  accounts, government budget, balance of payments)
• World Development Indicators (WDI) (labor stocks
  value-added in agr/ind/ser population)
• Public Expenditure Reviews and Country Economic
  Memoranda
• Sectoral MDG studies (health, education,
  water-sanitation, public infrastructure)
• Existing SAMs and input-output tables
• Surveys (household, labor, DHS)

33
MAMS MDG Scenarios

• The BASE scenario is a business-as-usual
  continues that may have the following
  characteristics
• Growth in GDP reflects trend of last 5-15 years.
• Unchanged GDP shares for government demand,
  foreign aid, and debt stocks.
• Other policies are unchanged or adjusted
  according to trends.
• Other exogenous items grow at the same rate as
  GDP.
• The BASE scenario serves as a benchmark for
  comparisons.

34
Examples of MDG Scenarios

• Questions commonly addressed by non-BASE
  scenarios What happens if the government
• expands services sufficiently to reach the MDGs
  with additional financing provided by (a) foreign
  grants (b) domestic taxes (c) domestic
  borrowing?
• contracts in one area (e.g. human development or
  other government) and expands in another (e.g.
  infrastructure) with unchanged aid and domestic
  policies?
• expands in one area with additional financing
  from a, b, or c (as defined under 1)?
• becomes more/less productive, adjusting one or
  more types of spending or financing in response?

35
MAMS Ethiopia Study

• BASE (as described above)
• MDG-BASE (core MDG scenario)
• Government service growth is sufficient to
  achieve all HD MDGs (2, 4, 5, 7a, 7b) by 2015
• Foreign grants are unconstrained adjust to meet
  the government financing gap
• Various simulations exploring tradeoffs among
  MDGs and issues of timing and costs.

36
Ethiopia MDG Values
37
Key Ethiopia findings

• Foreign aid per capita increases five-fold to
  US79 in 2015 as compared to 2005.
• Heavy reliance on foreign aid appreciates the
  real exchange rate appreciation and skews
  production toward non-tradables.
• In the educated part of the labor market, wage
  increases are initially rapid but will later slow
  down when labor supplies increase and the
  scaling-up period is concluded.
• Relative to an emphasis on infrastructure, a
  human development focus puts the economy on a
  slower growth track.

38
Poverty Analysis The Road Ahead

• Improving microeconomic specifications
• Intertemporal household behavior
• savings and investment physical and human
  capital
• demographic changes and migration
• Intra-household allocation of resources
• Improving market specification in rural sector
• segmentation and market failures in factor
  markets (land, labor, credit),
• spatial and regional dimensions in markets for
  goods (access to markets, transaction costs)

39
Poverty Analysis The Road Ahead

• Microsimulation model of producers (farms, firms)
  as well as households.
• Issue use more representative actors without
  moving to specification of all observations in
  survey samples.
• Techniques of data reduction without loss of
  important information.
• In MDG/MAMS better models of links between govt
  policy and MDG outcomes

40
Poverty Analysis The Road Ahead

• No single approach is likely to dominate
• Informational demands and operational constraints
  vary across applications
• Data reconciling household/firm data with
  national accounts and SAM data
• Important for any poverty analysis.
• Separation of economywide and household analysis
  represents a methodological failure
• Need for reconciliation and integration

41
References

• Bourguignon, François, Maurizo Bussolo, and Luiz
  A. Pereira da Silva, eds. 2008. The Impact of
  Macroeconomic Policies on Poverty and Income
  Distribution Macro-Micro Evaluation Techniques
  and Tools. Washington, DC and New York World
  Bank and Palgrave Macmillan.
• Bourguignon, François, Carolina Diaz-Bonilla and
  Hans Lofgren. 2008. Aid, service delivery and
  the MDGs in an economy-wide framework. In
  Bourguignon, Bussolo, and Pereira da Silva, eds.
  Also as World Bank Policy Research Working Paper
  4683.
• Bourguignon, François , Sherman Robinson, and
  Anne-Sophie Robilliard. 2005. Representative
  versus Real Households in the Macroeconomic
  Modeling of Inequality In Frontiers in Applied
  General Equilibrium Modeling Essays in Honor of
  Herbert Scarf edited by Timothy Kehoe, T. N.
  Srinivasan and John Whalley. Cambridge Cambridge
  University Press.
• Cogneau, Denis, and Anne-Sophie Robilliard. 2007.
  Growth, Distribution, and Poverty in Madagascar
  Learning From a Microsimulation Model in a
  General Equilibrium Framework. In
  Microsimulations as a Tool for the Evaluation of
  Public Policies Methods and Applications, ed.
  Amedeo Spadero. Fundacion BBVA.

42
References

• Lofgren, Hans and Carolina Diaz-Bonilla. 2008.
  Foreign Aid, Taxes, and Government Productivity
  Alternative Scenarios for Ethiopias Millennium
  Development Goal Strategy. In Delfin S. Go and
  John Page, eds. Africa at a Turning Point?
  Growth, Aid, and External Shocks. Washington
  World Bank.
• Lofgren, Hans and Carolina Diaz-Bonilla. 2008.
  "MAMS An Economywide Model for Development
  Strategy Analysis." Draft. September 24. World
  Bank. Downloadable from www.worldbank.org\mams.
• Robilliard, Anne-Sophie and Sherman Robinson.
  2003. Reconciling Household Surveys and National
  Accounts Data Using a Cross Entropy Estimation
  Method. The Review of Income and Wealth, Series
  49, No. 3, September.
• Robilliard, Anne-Sophie and Sherman Robinson.
  2006. The Social Impact of a WTO Agreement in
  Indonesia. In Thomas W. Hertel and L. Alan
  Winters, eds. Poverty and the WTO Impacts of the
  Doha Development Agenda. Washington and New York
  World Bank and Palgrave Macmillan.

43
References

• Robilliard, Anne-Sophie, Francois Bourguignon,
  and Sherman Robinson. 2008. Examining the Social
  Impact of the Indonesian Financial Crisis Using a
  Macro-Micro Model. In Bourguignon, Bussolo,
  Pereira da Silva, eds.
• Robinson, Sherman, Andrea Cattaneo and Moataz
  El-Said. 2001. Updating and estimating a SAM
  using cross entropy methods, Economic Systems
  Research, 13 (1) 47-64
• Vos, Rob, Enrique Ganuza, Sam Morley and Sherman
  Robinson, eds. 2006. Who Gains from Free Trade?
  Export-Led Growth, Inequality and Poverty in
  Latin America. London Routledge.
• Vos, Rob, Enrique Ganuza, Hans Lofgren, Marco V.
  Sánchez y Carolina Díaz-Bonilla, eds. 2008.
  Políticas Públicas para el Desarrollo Humano
  Cómo lograr los Objetivos de Desarrollo del
  Milenio en América Latina y el Caribe? Editado
  por PNUD ONU/DAES Banco Mundial ONU/CEPAL.