## 1. Exogenously Calibrated Parameters

## Source details

**Canonical URL:** [1. Exogenously Calibrated Parameters](https://www.imf.org/-/media/files/publications/wp/2019/wpiea2019187-print-pdf.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/wp/2019/wpiea2019187-print-pdf.pdf.md)
- [Structured JSON version](/-/media/files/publications/wp/2019/wpiea2019187-print-pdf.pdf.json)

---

### Introduction: growth, inequality, and paper objective
- Real GDP per capita growth in Cambodia has averaged 6 percent over the last two decades; ASEAN average is 4 percent.
- GDP per capita in Cambodia rose to 11 percent of the ASEAN average in 2017 from 6 percent in 2001.
- Consumption Gini fell from 40.4 percent in 1997 to 29 percent in 2012 (SWIID 6.1).
- Government spending changes (1997 to 2012):
  - Education spending: 1.3 percent of GDP → 1.6 percent of GDP.
  - Subsidies and other transfers: 1.5 percent of GDP → 3 percent of GDP.
- Urban population: 17 percent in 1994 → 23 percent in 2017.
- Paper objective: analyze effects on GDP and inequality of increasing infrastructure spending financed by:
  - (i) higher property taxes,
  - (ii) higher Value-Added Taxation (VAT),
  - (iii) an increase in labor market taxation,
  using a general equilibrium model with heterogeneous agents calibrated to Cambodia.

### Key empirical patterns: distribution, regional gaps, and poverty
- Infrastructure gap: only 10.5 percent of all roads are paved.
- Regional population and income gaps:
  - Rural regions represent nearly 80 percent of the population.
  - Average rural household income is 60 percent of urban household income.
  - Average rural consumption is 70 percent of urban consumption.
- Consumption and income inequality:
  - Consumption Gini: 40.4 percent in 1997 → 29 percent in 2012.
  - Disposable income Gini: 36.7 percent in 1997 → 33.9 in 2012.
  - From 2012 to 2015: consumption inequality broadly unchanged; income inequality rose somewhat.
- Poverty reduction: absolute poverty declined from around 17 percent in 2007 to 2.2 percent in 2012.
- Average monthly household amounts:
  - Income: 1712 thousand riels.
  - Consumption: 1513 thousand riels.
- Rural–urban differences in welfare and services:
  - Sanitation access: 30.5 percent of rural population vs 88.1 percent in urban areas.
  - Rural households rely more on self-employment and net income transfers; urban households rely more on salary.
  - Average urban salary is almost double that in the rural sector.
  - Income from self-employment is on average 1.13 times larger in the urban area.

### Fiscal structure and redistribution tools
- Tax revenue developments:
  - Tax revenue: less than 10 percent of GDP in 2007 → around 17 percent of GDP in 2017.
  - Total tax revenue reported as 16.9 percent of GDP in source tables.
  - VAT generates 61 percent of total tax revenue.
- Progressive tax shares (small):
  - Personal income taxation and capital income and capital gains: 4.2 percent of GDP.
  - Property taxation: 0.1 percent of GDP.
- Social transfers and inequality:
  - Income transfers in 2015 reduced Gini by 0.7 percentage points in urban areas and 1.4 percentage points in rural areas.
  - Social assistance spending: 0.3 percent of GDP.
- Policy implication: income transfers generally more effective than taxes in reducing inequality (cited cross-country finding).

### Model framework and calibration
- Model type: small open economy with heterogeneous agents (Peralta-Alva et al., 2018); two regions (urban and rural); households work in agriculture, manufacturing, or services.
- Household features:
  - Infinitely lived; maximize expected utility over consumption of agricultural, manufacturing, and service goods.
  - Food subject to a subsistence constraint.
  - Total working hours per household fixed; no labor/leisure choice.
  - Urban: formal manufacturing and informal services choices; rural: formal wage work on large farms and informal own-plot work.
- Income shocks:
  - Wage income from formal markets in rural and urban areas subject to idiosyncratic shocks.
  - Efficiency of informal-sector working hours not subject to shocks.
- Government instruments in model:
  - VAT on food and manufacturing goods.
  - Personal income tax (PIT) on household income from formal markets.
  - Corporate income tax (CIT) on manufacturing firms and large farmers.
  - Government spending options: (i) income transfers, (ii) infrastructure investment boosting private-sector productivity.
  - Government runs a balanced budget every period.
  - Property taxes not included endogenously; property-tax results rely on microdata simulations.
  - Suggested administrative improvements: updated taxpayer register including property values; improved process for VAT refunds; broader coverage of personal income taxation.
- Infrastructure modelling:
  - Modeled as an increase in total factor productivity across all sectors.
  - Key parameter: elasticity of GDP with respect to infrastructure investment.
  - Simulations consider a permanent increase in infrastructure investments and compare steady-state productivity under alternative investment rates.

### Selected exogenously calibrated parameter values (Table 1)
- Discount Rate: 0.96
- Depreciation Rate: 0.06
- Land Share in Food Production: 0.49
- Labor Share in Services Production: 0.37
- Capital Share in Manufacturing Production: 0.37
- Land Share in Cash Crop Production: 0.49
- Labor Share in Cash Crop Production: 0.32
- Relative Land Size of Food Farm: 0.27
- Persistence of Urban Income Shocks: 0.92
- Persistence of Rural Income Shocks: 0.92
- Urban Population Share: 0.28
- Rural Population Share: 0.69
- Large Farmer Share: 0.03

### Data targets and model matches (selected from Table 2)
- Service Share in Consumption: Data 0.43; Model 0.40
- Manufacturing Share in Consumption: Data 0.16; Model 0.17
- Service Share in GDP: Data 0.34; Model 0.35
- Agriculture Share in GDP: Data 0.32; Model 0.32
- Agriculture Exporting Share in GDP: Data 0.03; Model 0.02
- Manufacturing Share in GDP: Data 0.31; Model 0.31
- Tax Revenue as a fraction of GDP: Data 0.11; Model 0.15
- Share of Corporate Tax in Tax Revenue: Data 0.36; Model 0.30
- Share of Personal Tax in Tax Revenue: Data 0.13; Model 0.12
- Elasticity of Infrastructure Investments: Data 0.07; Model 0.07
- Urban Population Share: Data 0.21; Model 0.21
- Rural Population Share: Data 0.79; Model 0.79
- Urban Consumption Gini: Data 0.30; Model 0.294
- Rural Consumption Gini: Data 0.30; Model 0.285
- Empirical guidance: Calderon et al. (2015) long-run elasticities in range 0.07-0.10; simulations use elasticity = 0.07.

### Policy scenarios evaluated and calibration assumptions
- Baseline infrastructure increase:
  - Permanent rise in infrastructure investments by 0.5 percent of GDP from 1.9 percent to 2.4 percent.
  - Implies an increase of the steady-state stock of infrastructure capital by 29 percent (assumes elasticity 0.07).
- Three financing options to mobilize 0.5 percent of GDP:
  - Property taxation: increase from 0.1 percent of GDP to 0.6 percent of GDP.
  - VAT: necessary increase in effective VAT rate ≈ 1.2 percentage points to generate 0.5 percent of GDP.
  - Wage taxation: raise revenue from 1.0 percent of GDP to 1.5 percent of GDP.
- Implementation notes:
  - Revaluation of property prices, increases in stamp duty, and/or tax administration improvements could reduce required property tax rate increases.
  - Reducing VAT exemptions and/or tax administration improvements could reduce required VAT rate increases.
- Trade-off: higher taxation funds infrastructure that raises productivity but also creates distortions affecting GDP and inequality differently by instrument.

### Scenario A — Broad-based infrastructure increase (rural and urban productivity rise)
- GDP impacts (permanent 0.5 percent of GDP infrastructure increase):
  - Financing via higher property taxation: GDP increases by 2.7 percent.
  - Financing via higher VAT: GDP increases by 1.8 percent.
  - Financing via higher labor market taxation: GDP increases by 0.3 percent.
- Inequality (Gini) impacts:
  - Property taxation: Gini falls by up to 0.1 percentage points.
  - Labor market taxation: Gini falls marginally less than under property taxation.
  - Higher VAT: Gini increases by 0.2 percentage points.
- Mechanisms:
  - Property taxation modelled as least distorting; does not change workers' labor market choice.
  - Higher VAT distorts labor market choice by inducing more self-production exempt from VAT.
  - Higher labor taxation distorts formal employment decisions, pushing workers to informal sector.

### Scenario B — Infrastructure increase focused in rural areas
- Aggregate impacts:
  - Maximum GDP increase up to 1.7 percent when investments only in rural area.
  - Smaller than broad-based 2.7 percent because productivity gains confined to rural area.
- Inequality impacts:
  - Gini falls by up to 0.2 percentage points.
  - Largest inequality reduction when financing via property taxation.
  - Inequality increases when financing via VAT.

### Scenario C — Tax revenue used for targeted transfers to poor rural households
- Distributional impacts:
  - Gini falls by up to 0.6 percentage points under targeted cash transfers to poorest rural households.
  - Reduction largest when financed via property taxation; smallest when financed via VAT.
  - Driver: higher taxation mostly borne by richer households (except VAT); transfers benefit poor rural households.
- GDP impacts:
  - GDP falls by 4 percent when financing via labor market taxation.
  - GDP falls by 1.9 percent when financing via VAT.
  - GDP impact zero or negative because transfers do not raise productivity while taxation creates distortions that reduce GDP; property taxation modelled to not negatively affect GDP.

### Main findings and policy implications (conclusion)
- Higher property taxation dominates among options studied:
  - Generates largest positive effect on GDP and largest reduction in inequality.
  - Rationale: property taxes are progressive and do not distort labor supply or investment incentives in the model.
  - Requires additional investments in tax administration to reap gains.
- VAT vs labor income taxation:
  - Financing through VAT delivers a larger increase in GDP than financing through labor income taxation in simulations.
  - Labor income taxes yield the largest reduction in inequality (relative to VAT) but are more distortive for GDP.
- Targeting and complementarities:
  - Targeting infrastructure to rural areas can increase Gini reduction up to 0.2 percentage points (with lower GDP impact).
  - Targeted social transfers can reduce Gini by up to 0.6 percentage points but can reduce GDP when financed by distortionary taxes.
- Overall trade-off: policymakers must balance GDP effects and distributional outcomes when choosing financing instruments; model results favor property taxation given its progressivity and limited distortionary effects.

*Source: wpiea2019187-print-pdf*

### 1. Exogenously Calibrated Parameters _________________________________________14

### 1. Exogenously Calibrated Parameters

### Introduction: growth and inequality trends
- Real GDP per capita growth in Cambodia has averaged 6 percent over the last two decades, compared with the ASEAN average of 4 percent.
- GDP per capita in Cambodia rose to 11 percent of the ASEAN average in 2017 from 6 percent in 2001.
- Gini coefficient for consumption fell from 40.4 percent in 1997 to 29 percent in 2012 (SWIID 6.1).
- Government spending changes (1997 to 2012):
  - Education spending increased from 1.3 percent of GDP to 1.6 percent of GDP.
  - Spending on subsidies and other transfers increased from 1.5 percent of GDP to 3 percent of GDP.
- Urban population increased from 17 percent in 1994 to 23 percent in 2017.
- Paper objective: analyze effects on GDP and inequality of increasing infrastructure spending financed by:
  - (i) higher property taxes,
  - (ii) higher Value-Added Taxation (VAT), or
  - (iii) an increase in labor market taxation,
  using a general equilibrium model with heterogeneous agents (Peralta-Alva et al., 2018) calibrated to Cambodia.

### Key findings on distribution and regional gaps
- Infrastructure gaps are substantial; only 10.5 percent of all roads are paved (Section III).
- Income disparities between regions:
  - Rural regions represent nearly 80 percent of the population.
  - Average household income in rural regions is only 60 percent of urban household income.
  - Average rural consumption is 70 percent of urban consumption.

### Research context and approach
- Model builds on Peralta-Alva et al. (2018) and literature on infrastructure and tax incidence (Calderon, Easterly, and Serven (2003); Briceno-Garmendia and Estache (2004); Calderon et al. (2015); Ghazanchyan et al. (2017)).
- Calibration uses aggregate and micro data for Cambodia; policy experiments conducted in calibrated general equilibrium heterogeneous-agent model.

### Structure of the paper
- Section II: recent developments in Cambodian inequality.
- Section III: policy tools for redistribution in Cambodia.
- Section IV: model description (Peralta-Alva et al., 2018).
- Section V: calibration to Cambodia.
- Section VI: policy experiments.
- Section VII: conclusion.

### Inequality in Cambodia: empirical patterns
- Consumption and income inequality trends:
  - Consumption Gini: 40.4 percent in 1997 → 29 percent in 2012 (SWIID 6.1).
  - Disposable income Gini: 36.7 percent in 1997 → 33.9 in 2012 (Solt, 2016).
- From 2012 to 2015:
  - Inequality in consumption broadly unchanged (based on consumption survey microdata from 2015).
  - Inequality in income rose somewhat.
- Poverty reduction:
  - Absolute poverty declined from around 17 percent in 2007 to 2.2 percent in 2012.
- Income versus consumption distribution:
  - Average monthly household income is 1712 thousand riels.
  - Average monthly household consumption is 1513 thousand riels.
  - Income distribution is less equal than consumption distribution; the propensity to consume tends to fall with income, and higher-income households save more (hedging behavior).
- Regional consumption/income gaps:
  - Rural population share: around 80 percent.
  - Rural average income: 60 percent of urban average income.
  - Rural average consumption: 70 percent of urban average consumption.
- Consumption composition differences:
  - Urban–rural consumption gaps across housing-related spending, transportation, food and non-food consumption.
  - Health-related expenditures are higher among rural households.
  - Rural households have worse sanitation access: 30.5 percent of rural population have access to sanitation facilities versus 88.1 percent in urban areas.
- Income source differences by region:
  - Average rural household derives largest income share from self-employment.
  - Average urban household derives largest income share from salary.
  - Average urban salary is almost double that in the rural sector.
  - Income from self-employment is on average 1.13 times larger in the urban area.
  - Rural households rely relatively more on net income transfers than urban households.

### Policies with distributional impact: tax and transfer composition
- Tax revenue developments:
  - Tax revenue: less than 10 percent of GDP in 2007 → around 17 percent of GDP in 2017.
  - Total tax revenue reported as 16.9 percent of GDP in the source tables.
  - VAT generates 61 percent of total tax revenue.
- Progressive tax shares are small:
  - Revenue from personal income taxation and capital income and capital gains: 4.2 percent of GDP (lowest among peers).
  - Revenue from property taxation: 0.1 percent of GDP.
- VAT considerations:
  - VAT is usually considered among the most regressive sources of taxation because consumption comprises a larger share of income for poorer households.
- Social transfers:
  - Overall income transfers in 2015 reduced the Gini coefficient by 0.7 percentage points in urban areas and 1.4 percentage points in rural areas.
  - Cambodia’s social assistance spending is only 0.3 percent of GDP, below ASEAN and Low-Income Developing Countries (LIDCs) averages.
  - Income transfers include pensions, work assistance and training support, remittances and scholarships.
- Policy implication highlighted:
  - Income transfers are generally more effective than taxes in reducing inequality (cites Wang and Caminada (2011) finding across 36 countries).

### Infrastructure gaps and growth implications
- Infrastructure indicators and gaps:
  - Only 10.5 percent of all roads are paved.
  - Share of population with access to safe drinking water, basic sanitation services, and electricity remains relatively low.
- Potential impact of infrastructure investment:
  - Improvements in infrastructure stock could boost growth further, especially in the rural sector, by enhancing market access and factor mobility (Calderon et al., 2015).
  - Higher infrastructure quality in rural areas can increase human capital (education, health access, jobs).
- Efficiency caveat:
  - Effectiveness of additional investments would benefit from improvements in the public investment framework to achieve efficient spending and avoid leakages.

### The model: small open economy with heterogeneous agents
- Model framework:
  - Small open economy model with heterogeneous agents developed in Peralta-Alva et al. (2018).
  - Two regions: urban and rural, with fixed household shares.
  - Households work in agriculture, manufacturing, or services across regions.
  - Households face idiosyncratic income shocks that differ between regions, generating different income distributions.
- Household preferences and labor:
  - Households are infinitely lived and maximize expected utility over sequences of consumption of agricultural, manufacturing, and service goods.
  - Food is subject to a subsistence constraint.
  - Urban households allocate working hours between formal manufacturing and informal services.
  - Rural households allocate hours between formal work for a wage in large farms and informal work on own plots.
  - Total working hours per household are fixed; model contains no labor/leisure choice.
- Income shock specification:
  - Wage income from formal markets in rural and urban areas is subject to idiosyncratic shocks reflecting fluctuations in efficiency of supplied working hours.
  - Efficiency of working hours within the informal sectors is not subject to shocks.

*Source: wpiea2019187-print-pdf - 1. Exogenously Calibrated Parameters _________________________________________14*

### 17.      Sectors. The model has two sectors within both the rural and urban area, respectively. In

### 17. Sectors.

### Model structure
- Two sectors in the rural area:
  - One sector producing agricultural products for domestic consumption, done informally by workers laboring informally on their own land plots.
  - One sector where large farmers engage workers to produce agricultural products for both the domestic and international market.
- Two sectors in the urban area:
  - A manufacturing sector producing manufacturing goods by employing workers and capital rented by households.
  - An informal urban sector where households work informally to produce service goods.
- Migration and labor-choice margins, informality, and heterogeneous agents are part of the general equilibrium framework described.

### Government, taxation, and public spending
- Taxes available in the model:
  - Value added tax (VAT) on food and manufacturing goods.
  - Personal income tax (PIT) on household income from formal markets.
  - Corporate income tax (CIT) on manufacturing firms and large farmers.
- Government spending options for tax revenue:
  - (i) Income transfers.
  - (ii) Infrastructure investment boosting private sector productivity.
- Government is assumed to run a balanced budget every period.
- Property taxes are not included endogenously in the model; results involving property taxes rely on microdata simulations.
- Implementation/enabling administrative improvements suggested:
  - (i) an updated taxpayer register, including on property values,
  - (ii) an improved process for VAT refunds,
  - (iii) broader coverage of the existing personal income taxation.

### Modelling of infrastructure investment
- Effect modelled as an increase in total factor productivity lifting productivity across all sectors.
- Key parameter: elasticity of GDP with respect to infrastructure investment.
- Simulations consider a permanent increase in infrastructure investments and compare steady state productivity levels under alternative investment rates.

### Calibration (parameters and data targets)
- Two groups of parameters:
  - Parameters from literature (Peralta-Alva et al., 2018) used as exogenous calibration.
  - Parameters endogenously calibrated to match Cambodian data (consumption/output shares, tax structure, inequality).
- Selected exogenously calibrated parameter values (from Table 1):
  - Discount Rate: 0.96
  - Depreciation Rate: 0.06
  - Land Share in Food Production: 0.49
  - Labor Share in Services Production: 0.37
  - Capital Share in Manufacturing Production: 0.37
  - Land Share in Cash Crop Production: 0.49
  - Labor Share in Cash Crop Production: 0.32
  - Relative Land Size of Food Farm: 0.27
  - Persistence of Urban Income Shocks: 0.92
  - Persistence of Rural Income Shocks: 0.92
  - Urban Population Share: 0.28
  - Rural Population Share: 0.69
  - Large Farmer Share: 0.03
- Data targets and model matches (selected from Table 2):
  - Service Share in Consumption: Data 0.43; Model 0.40
  - Manufacturing Share in Consumption: Data 0.16; Model 0.17
  - Service Share in GDP: Data 0.34; Model 0.35
  - Agriculture Share in GDP: Data 0.32; Model 0.32
  - Agriculture Exporting Share in GDP: Data 0.03; Model 0.02
  - Manufacturing Share in GDP: Data 0.31; Model 0.31
  - Tax Revenue as a fraction of GDP: Data 0.11; Model 0.15
  - Share of Corporate Tax in Tax Revenue: Data 0.36; Model 0.30
  - Share of Personal Tax in Tax Revenue: Data 0.13; Model 0.12
  - Elasticity of Infrastructure Investments: Data 0.07; Model 0.07
  - Urban Population Share: Data 0.21; Model 0.21
  - Rural Population Share: Data 0.79; Model 0.79
  - Urban Consumption Gini: Data 0.30; Model 0.294
  - Rural Consumption Gini: Data 0.30; Model 0.285
- Empirical guidance for elasticity:
  - Calderon et al. (2015) find long-run elasticities for infrastructure investments across countries in the range 0.07-0.10.
  - Simulations use elasticity = 0.07.

### Policy scenarios evaluated
- Baseline infrastructure increase considered:
  - Permanent rise in infrastructure investments by 0.5 percent of GDP from 1.9 percent to 2.4 percent.
  - This entails an increase of the steady state stock of infrastructure capital by 29 percent (assumes elasticity 0.07).
- Three financing options considered to mobilize 0.5 percent of GDP:
  - Increase in property taxation from current level of 0.1 percent of GDP to 0.6 percent of GDP.
  - Increase in effective VAT rate; necessary increase to generate 0.5 percent of GDP in revenue would be about 1.2 percentage points.
  - Raise revenue from wage taxation from current level of 1.0 of GDP to 1.5 percent of GDP.
- Key implementation notes:
  - Revaluation of property prices, increases in stamp duty, and/or improvements in tax administration could reduce necessary property tax rate increases.
  - Reducing VAT exemptions and/or improvements in tax administration could reduce necessary VAT rate increases.
- Trade-offs emphasized: higher taxation raises productivity through infrastructure but also creates distortions that can dent GDP and affect inequality differently by tax instrument.

### Scenario A — Higher infrastructure in both rural and urban areas (broad-based)
- GDP impacts of the permanent 0.5 percent of GDP infrastructure increase (productivity across all sectors):
  - Financing via higher property taxation: GDP increases by 2.7 percent.
  - Financing via higher VAT: GDP increases by 1.8 percent.
  - Financing via higher labor market taxation: GDP increases by 0.3 percent.
- Inequality (Gini coefficient) impacts:
  - Property taxation: Gini falls by up to 0.1 percentage points.
  - Labor market taxation: Gini falls marginally less than under property taxation.
  - Higher VAT: Gini increases by 0.2 percentage points.
- Mechanisms:
  - Property taxation is modelled as least distorting because it does not change workers' labor market choice.
  - Higher VAT is assumed to distort the labor market choice by inducing more self-production exempt from VAT.
  - Higher labor taxation distorts formal employment decisions, pushing workers to the informal sector.

### Scenario B — Higher infrastructure focused in rural areas
- Policy: additional infrastructure projects undertaken only in rural areas (rural productivity rise only).
- Aggregate impacts:
  - Maximum GDP increase up to 1.7 percent when investments are only in the rural area.
  - This is smaller than the 2.7 percent from the broad-based case because productivity gains are confined to the rural area.
- Inequality impacts:
  - Gini coefficient falls by up to 0.2 percentage points.
  - Largest inequality reduction achieved when financing is secured through property taxation.
  - Inequality increases when financing is secured through VAT.

### Scenario C — Targeted transfers to poor rural households (tax revenue used for transfers)
- Distributional impacts:
  - Gini coefficient falls by up to 0.6 percentage points under the package of targeted cash transfers to the poorest rural households.
  - The reduction is largest when financing is via property taxation and smallest when financed via VAT.
  - Drivers: higher taxation is mostly borne by richer households (except VAT), and transfers benefit poor rural households specifically.
- GDP impacts:
  - GDP falls by 4 percent when financing is secured through labor market taxation.
  - GDP falls by 1.9 percent when financing is secured through VAT.
  - GDP impact is zero or negative because transfers do not raise productivity whereas taxation creates distortions that reduce GDP; property taxation is modelled to not negatively affect GDP.

### Main findings and policy implications (Conclusion)
- The reform option based on higher property taxation dominates among the options studied:
  - Generates the largest positive effect on GDP and the largest reduction in inequality.
  - Rationale: property taxes are progressive (wealthier households pay more) and do not distort labor supply or investment incentives in the model.
  - Reaping gains from property taxation requires additional investments in tax administration.
- Ranking between VAT and labor income taxation is ambiguous:
  - Financing through VAT delivers a larger increase in GDP than financing through labor income taxation in the simulations.
  - Labor income taxes yield the largest reduction in inequality (relative to VAT) but can be more distortive for GDP.
- Targeting infrastructure spending to rural areas or complementing infrastructure with more efficient social spending can further reduce inequality:
  - Targeting infrastructure to rural areas can increase the Gini reduction up to 0.2 percentage points (at the cost of lower GDP impact).
  - Targeted social transfers can reduce the Gini by up to 0.6 percentage points but can reduce GDP when financed by distortionary taxes.
- Trade-off emphasized: choosing financing instruments involves balancing GDP effects and distributional outcomes; property taxation emerges as the preferred financing instrument in the model given its progressivity and limited distortionary effects.

*Source: https://www.imf.org/-/media/files/publications/wp/2019/wpiea2019187-print-pdf.pdf*

---


_Source: https://www.imf.org/-/media/files/publications/wp/2019/wpiea2019187-print-pdf.pdf_