## wp17156

## Source details

**Canonical URL:** [wp17156](https://www.imf.org/-/media/files/publications/wp/2017/wp17156.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/wp/2017/wp17156.pdf.md)
- [Structured JSON version](/-/media/files/publications/wp/2017/wp17156.pdf.json)

---

### 1. INTRODUCTION — definition, measurement, and study scope
- Definition and importance
  - Informal economy terminology includes "shadow economy", "black economy" and "unreported economy".
  - Feige (2005): informal economy comprises economic activities that circumvent costs and are excluded from the benefits and rights incorporated in laws and administrative rules covering property relationships, commercial licensing, labor contracts, torts, financial credit, and social systems.
  - Measuring informality matters because:
    - Workers in informal conditions have little or no social protection or employment benefits, undermining inclusiveness in the labor market.
    - Informal activity severely limits tax revenues for developing countries, reducing incentive to build a stable tax base.
- Methods to measure the informal economy
  - Direct approaches: rely on surveys, voluntary replies, tax auditing and compliance methods; sensitive to questionnaire formulation and respondents’ willingness to cooperate; unlikely to capture all informal activities.
  - Indirect (indicator) approaches: use discrepancies in national accounts, labor force participation, electricity consumption, monetary transactions, currency demand, or latent-variable models; many indirect methods consider just one indicator for all effects of the informal economy.
- MIMIC: features and criticisms
  - MIMIC considers multiple causes and multiple effects using a latent variable framework.
  - Criticisms include endogeneity when GDP is used as both cause and indicator, reliance on external studies to calibrate MIMIC to percent-of-GDP values, and sensitivity of estimated coefficients to specification, country sample and time span.
- This paper’s contributions and methodology
  - Key methodological advances:
    - (a) Modified MIMIC using night light intensity instead of GDP as an indicator to address endogeneity.
    - (b) Estimation via Predictive Mean Matching (PMM) by Rubin (1987) as an independent robustness check and to address calibration controversies.
    - (c) Comparison of these results with official estimations from countries’ national account statistics.
  - Novelty claim: first use of modified MIMIC with night light intensity combined with PMM to estimate informality and compare with national authority estimates.
  - Scope: MIMIC model covers a sample of 140 countries over the 1991-2014 interval.
- Variables and model design (summary)
  - Causes: fiscal freedom (measure of tax burden), institutions (lack of respect for the law), unemployment, trade openness.
  - Indicators: currency as a fraction of broad money, labor force participation, night lights intensity (Henderson, Storeygard, and Weil (2012)) used instead of GDP per capita and growth of GDP per capita.

### 5. RESULTS — MIMIC estimation findings
- A. MIMIC Estimation Results (benchmark and robustness)
  - Benchmark specification cause and indicator set: fiscal freedom index, institutions (rule of law), unemployment, trade openness; currency (M0/M1), labor force participation, night lights (size of the economy).
  - Cause variable standardized impacts (one standard deviation change on size of informal economy):
    - Fiscal freedom: -0.15 standard deviations.
    - Institutions (rule of law): -0.07 standard deviations.
    - Unemployment: 0.07 standard deviations.
    - Trade openness: -0.18 standard deviations.
  - Statistical significance: coefficients generally significant (mostly at the 1 or 5 percent level).
  - Robustness across specifications:
    - Trade openness: ~0.17 standard deviation decrease in informality per standard deviation increase, across specifications.
    - Unemployment: ~0.07 standard deviation increase in informality per standard deviation increase across specifications.
    - Size of government: first two specifications show a 0.1 standard deviation increase in informality per standard deviation increase; magnitude drops markedly when including government stability.
    - Fiscal freedom (alternative government distortion measure): across three specifications, a one standard deviation increase decreases informal activity by 0.16 standard deviations; remains similar when including government stability.
    - Rule of law: in specifications one and four, a one standard deviation increase decreases informal activity by 0.06 standard deviations.
    - Control on corruption: coefficient sign consistent with corruption encouraging informality but not statistically significant.
    - Government stability: statistically significant and decreases share of informal economic activity.
  - Indicators:
    - Coefficient on currency standardized to one.
    - Informal economic activity decreases both labor force participation and night light density across specifications (magnitudes vary by specification).
    - Informal activity decreases night light intensity across all specifications.
  - Sample split: developing vs advanced economies (differences summarized)
    - Trade openness: statistically significant for developing countries with magnitude similar to full sample; for advanced economies, effect larger when significant.
    - Unemployment:
      - Developing: no statistical significance for unemployment effect on informality.
      - Advanced: unemployment has a statistically significant and negative effect on informal economic activity.
    - Distortionary effect of government (size or taxation):
      - Similar significance and magnitude for developing economies.
      - For advanced economies, magnitude nearly double that for the whole sample and developing economies.
    - Institutions vs stability:
      - Rule of law: statistically significant for advanced economies, not for developing economies.
      - Government stability: significant for developing economies, not for advanced economies.
    - Indicators by sub-sample:
      - Informal economic activity decreases labor force participation and is statistically significant across specifications.
      - Night light density:
        - Developing economies: greater informal economic activity leads to a statistically significant decrease in night light density.
        - Advanced economies: more informal activity leads to an increase in night light density; hypothesized due to better ability to hide informal activity and informal activities occurring at night.

- B. Estimation of the Size of the Informal Economy (levels and comparisons)
  - Heterogeneity in Sub-Saharan Africa (SSA):
    - Estimated range in SSA: low of 20 to 25 percent (Mauritius, South Africa, Namibia) to high of 50 to 65 percent (Benin, Tanzania, Nigeria).
  - Regional and global averages (2010-2014 unless otherwise indicated):
    - SSA unweighted average share of informality: almost 38 percent of GDP over 2010-14.
    - Comparators:
      - Latin America: 40 percent of GDP.
      - South Asia: 34 percent of GDP.
      - Europe: 23 percent of GDP.
      - OECD countries: 17 percent of GDP.
    - By income group (global averages over 2010-2014):
      - Low income countries: 40 percent of GDP.
      - Emerging economies: 32 percent of GDP.
      - Advanced economies: 18 percent of GDP.
    - Within Sub-Saharan Africa:
      - Low income countries: informal economy averages 40 percent.
      - Middleincome countries: informal economy averages 35 percent.
    - Oil exporters and fragile countries: more likely to harbor informality, with an informal economy well above 40 percent of GDP.
  - Comparison with national accounts / statistical agencies:
    - Rank correlation between MIMIC results and Statistical Agencies’ estimates for eight sub-Saharan African countries: about 85 percent.
    - Utility of MIMIC despite national estimates:
      - Not all countries publish national estimates.
      - Cross-country comparability of agencies’ methodologies is limited.
      - MIMIC produces panel data for most countries that is comparable and useful to test statistical relationships.

### 7. CONCLUSIONS — synthesis, validation, and policy implications
- Methodological approach summary
  - Modified MIMIC replacing GDP indicator with night lights to address endogeneity.
  - Independent PMM (Rubin (1987)) used to estimate size of the informal economy and calibrate MIMIC relative estimates.
  - Alternative treatment of informality as a missing-data problem using Multiple Imputations Predictive Mean Matching (MIPMM) matching survey-based countries to those without surveys.
- Key empirical findings (SSA emphasis)
  - Substantial heterogeneity in SSA:
    - Low estimates: 20 to 25 percent in Mauritius, South Africa and Namibia.
    - High estimates: 50 to 65 percent in Benin, Tanzania and Nigeria.
  - Informal economy in SSA remains among the largest in the world, though "very gradually declining".
  - Informality averages:
    - 40 percent in the region’s low income countries.
    - 35 percent for the region’s middle income countries.
  - Oil exporters and fragile countries: everything else equal, informal economy well above 40 percent of GDP.
- PMM country-grouping examples (informal economy as percent of GDP)
  - Low-Size Countries (0-20 percent): Mauritius, South Africa, Botswana, Lesotho, Swaziland, Namibia.
  - Middle-Size Countries (20-40 percent): Comoros, Cabo Verde, Zambia, Kenya, Ghana, Zimbabwe, Guinea, Eritrea, Tanzania, Gabon, Ethiopia, The Gambia, Mauritania, Uganda, Angola, Cameroon, Côte D'Ivoire, Malawi, Liberia, Madagascar, Equatorial Guinea, Mozambique, Burkina Faso, Burundi.
  - High-Size Countries (>40 percent): Congo, Republic of; Togo*; Guinea-Bissau; Nigeria*; Mali; Senegal*; Mali and others.
  - Note: entries marked with * are based on survey estimates.
- Robustness and validation statistics
  - MIMIC results robust when cross-checked with MIPMM; country groupings broadly aligned.
  - Cross-check against Statistical Agencies’ estimates for eight SSA countries:
    - Pearson correlation between National Accounts Statistics and MIMIC: 0.73.
    - Spearman's Rank Correlation: 0.857***.
  - Survey coverage: 49 countries identified with survey-based estimates, including nine in Sub-Saharan Africa.
  - MIMIC model fit statistics reported in Figure 1:
    - RMSEA: 0.052
    - Chi-Square: 158.781
    - Observations: 1211
    - Countries: 139
- Interpretation and comparative advantages of MIMIC-based estimates
  - MIMIC-generated panel data are useful because:
    - Not all countries publish national-account-based estimates.
    - National statistical methodologies and sampling methods may affect cross-country comparability.
    - National estimates may fail to account for recent changes in domestic economies.
  - The combination of modified MIMIC using night lights, PMM, and national accounts is presented as a novel application for generating robust estimates of informality in SSA.
- Policy implications
  - Existing policy literature is limited on determinants of variation in informal economy size and on policies to shift activity to the formal sector.
  - Policy advice emphasized in related studies includes the importance of allowing informal activity to act as a social safety net given that transition from informal to formal activity is likely to be long.

*Source: wp17156*

### 1. INTRODUCTION.........................................................................................................

### 1. INTRODUCTION

### Definition and importance
- The informal economy has no standard definition; terms used include "shadow economy", "black economy" and "unreported economy".
- Feige (2005): the informal economy comprises economic activities that circumvent costs and are excluded from the benefits and rights incorporated in laws and administrative rules covering property relationships, commercial licensing, labor contracts, torts, financial credit, and social systems.
- Measuring informality matters because:
  - Workers in informal conditions have little or no social protection or employment benefits, undermining inclusiveness in the labor market.
  - Informal activity severely limits tax revenues for developing countries, reducing incentive to build a stable tax base.

### Methods to measure the informal economy
- Direct approaches:
  - Rely on surveys, voluntary replies, tax auditing and compliance methods.
  - Sensitive to questionnaire formulation and respondents’ willingness to cooperate; unlikely to capture all informal activities.
- Indirect (indicator) approaches:
  - Use indirect information such as discrepancies in national accounts, labor force participation, electricity consumption, monetary transactions, currency demand, or latent-variable models.
  - Many indirect methods consider just one indicator for all effects of the informal economy.

### MIMIC: features and criticisms
- MIMIC (Multiple Indicator-Multiple Cause) main features:
  - Considers multiple causes of the existence and growth of the informal economy and multiple effects over time.
  - Based on an unobserved (latent) variable, using a set of causes and indicators to measure the phenomenon.
- Criticisms of MIMIC:
  - Use of GDP (GDP per capita and growth of GDP per capita) as both cause and indicator variables creates endogeneity concerns.
  - Reliance on another independent study to calibrate standardized MIMIC values to size of the informal economy in percent of GDP.
  - Estimated coefficients are sensitive to alternative specifications, country sample and time span chosen.

### This paper’s contributions and methodology
- Key methodological advances introduced:
  - (a) Use of a modified MIMIC model that addresses endogeneity concerns by using satellite data on night light intensity instead of GDP as an indicator variable proxying the size of the economy.
  - (b) Estimation of the size of the informal economy using the Predictive Mean Matching method (PMM) by Rubin (1987) as an independent robustness check and to address calibration controversies.
  - (c) Comparison of these results with official estimations from countries’ national account statistics.
- Novelty claim: First use of the combination of modified MIMIC with night light intensity and PMM to generate robust estimates of the size of the informal economy and compare them with national authority estimates.
- Scope: MIMIC model covers a sample of 140 countries over the 1991-2014 interval.

### Variables and model design (summary)
- Causes included in the MIMIC estimation:
  - fiscal freedom (measure of tax burden)
  - institutions (lack of respect for the law)
  - unemployment
  - trade openness
- Indicators included in the MIMIC estimation:
  - currency as a fraction of broad money
  - labor force participation
  - night lights intensity (Henderson, Storeygard, and Weil (2012)) used instead of GDP per capita and growth of GDP per capita to address endogeneity.

### Key results
- Heterogeneity in informality across Sub-Saharan Africa (SSA):
  - Low informality: "20 to 25 percent" of formal sector output in Mauritius, South Africa and Namibia.
  - High informality: "50 to 65 percent" in Benin, Tanzania and Nigeria.
- Regional and income patterns:
  - Share of informal economic activity in Sub-Saharan Africa remains among the largest in the world, though "very gradually declining".
  - Informality generally falls with the level of income:
    - averages "40 percent" in the region’s low income countries
    - averages "35 percent" for its middle income countries
  - Oil exporters and fragile countries tend to have an informal economy "well above 40 percent of GDP", regardless of income per capita.
- Robustness checks and validation:
  - Cross-checked MIMIC estimates with Multiple Imputations Predictive Mean Matching (MIPMM) treating informality as a missing data issue; country groupings broadly align with MIMIC estimates.
  - Rank correlation between MIMIC results and Statistical Agencies’ estimates for eight SSA countries is "about 80 percent".
  - Advantages of MIMIC despite national estimates existing:
    - Not all countries publish national estimates.
    - National methodologies and sampling affect cross-country comparability.
    - Some national approaches may fail to account for recent domestic economic changes.
    - MIMIC produces panel data for most countries that is comparable and useful for testing statistical relationships.

*Source: wp17156 - 1. INTRODUCTION*

### 5. RESULTS

### 5. RESULTS

### A. MIMIC Estimation Results
- Benchmark MIMIC specification (Figure 1): cause variables — fiscal freedom index, institutions (rule of law), unemployment, trade openness; indicator variables — currency (M0/M1), labor force participation, size of the economy (night lights).
- Coefficient signs and significance:
  - A one standard deviation increase in fiscal freedom index, institutions (rule of law), unemployment, and trade openness change the size of the informal economy by -0.15, -0.07, 0.07 and -0.18 standard deviations, respectively.
  - Coefficients generally statistically significant (mostly at the 1 or 5 percent level).
- Robustness across alternative MIMIC specifications (Tables 1, 2, and 3):
  - Trade openness: a standard deviation increase decreases the size of the informal economy by approximately 0.17 standard deviations across specifications.
  - Unemployment: a one standard deviation increase leads to an approximate 0.07 increase in the informal sector across specifications.
  - Size of government:
    - In first two specifications, a one standard deviation increase leads to a 0.1 increase in the informal economy.
    - Magnitude drops markedly when including government stability.
  - Fiscal freedom (alternative measure of government distortion):
    - Across three specifications, a one standard deviation increase in fiscal freedom decreases the share of informal economic activity by 0.16 standard deviations.
    - This coefficient remains similar even when including government stability.
  - Rule of law:
    - In specifications one and four, a one standard deviation increase in the rule of law decreases the share of informal economic activity by 0.06 standard deviations.
  - Control on corruption: no statistical significance, but coefficient sign consistent with corruption encouraging informality.
  - Government stability: statistically significant and decreases the share of informal economic activity.
- Indicator variables:
  - Coefficient on currency standardized to one (standard practice).
  - Informal economic activity is statistically significant and decreases both labor force participation and night light density across specifications (magnitudes vary by specification).
  - Informal activity decreases night light intensity across all specifications.
- Sample split (Tables 2 and 3: developing vs advanced economies) — notable differences:
  - Trade openness: statistically significant for developing countries with similar magnitude to full sample; effect for advanced economies has larger magnitude when significant.
  - Unemployment:
    - Table 2 (developing): no statistical significance for effect of unemployment on informality.
    - Table 3 (advanced): unemployment has a statistically significant and negative effect on informal economic activity.
  - Distortionary effect of government (size or taxation):
    - Similar significance and magnitude for developing economies.
    - For advanced economies, magnitude nearly double that for the whole sample and developing economies.
  - Institutions vs stability:
    - Rule of law: statistically significant for advanced economies, not for developing economies.
    - Government stability: significant for developing economies, not for advanced economies.
  - Indicators by sub-sample:
    - Greater informal economic activity decreases labor force participation and is statistically significant across all specifications in Tables 2 and 3.
    - Night light density:
      - Developing economies: greater informal economic activity leads to a statistically significant decrease in night light density.
      - Advanced economies: more informal activity leads to an increase in night light density.
    - Hypothesized reasons for advanced economies: better ability to hide informal activity; informal activities in advanced economies may occur at night, increasing night light intensity.

### B. Estimation of the Size of the Informal Economy
- Heterogeneity in Sub-Saharan Africa (Figure 2):
  - Estimated range in SSA: low of 20 to 25 percent (Mauritius, South Africa, Namibia) to high of 50 to 65 percent (Benin, Tanzania, Nigeria).
- Regional and global comparisons (Figure 3 and text):
  - SSA unweighted average share of informality: almost 38 percent of GDP over 2010-14.
  - Comparators:
    - Latin America: 40 percent of GDP.
    - South Asia: 34 percent of GDP.
    - Europe: 23 percent of GDP.
    - OECD countries: 17 percent of GDP.
  - By income group (global averages over 2010-2014):
    - Low income countries: 40 percent of GDP.
    - Emerging economies: 32 percent of GDP.
    - Advanced economies: 18 percent of GDP.
  - Within Sub-Saharan Africa:
    - Low income countries: informal economy averages 40 percent.
    - Middleincome countries: informal economy averages 35 percent.
  - Oil exporters and fragile countries: everything else equal, more likely to harbor informality with an informal economy well above 40 percent of GDP.
- Comparison with national accounts / statistical agencies:
  - Rank correlation between MIMIC results and Statistical Agencies’ estimates for eight sub-Saharan African countries: about 85 percent.
  - Utility of MIMIC despite national estimates:
    - Not all countries publish national estimates.
    - Cross-country comparability of agencies’ methodologies is limited.
    - MIMIC produces panel data for most countries that is comparable and useful to test statistical relationships.

*Source: wp17156 - 5. RESULTS*

### 7. CONCLUSIONS

### 7. CONCLUSIONS

### Methodological approach
- Applied a modified Multiple Indicator-Multiple Cause (MIMIC) model that replaces GDP as an indicator with a "light intensity" approach (night lights) to address endogeneity concerns associated with GDP as an indicator variable.
- Employed a completely independent method, the Predictive Mean Matching method (PMM) by Rubin (1987), to estimate the size of the informal economy and to calibrate relative MIMIC estimates.
- Compared model-based estimates with official estimations from countries’ national account statistics.
- Treated informality as a missing-data problem under an alternative approach: Multiple Imputations Predictive Mean Matching (MIPMM) developed by Rubin D.B., matching countries with survey-based information to those without.

### Key empirical findings
- Substantial heterogeneity in the size of the informal economy across Sub-Saharan Africa (SSA):
  - Low estimates: 20 to 25 percent in Mauritius, South Africa and Namibia.
  - High estimates: 50 to 65 percent in Benin, Tanzania and Nigeria.
- Regional characterization:
  - The informal economy in Sub-Saharan Africa remains among the largest in the world, although this share has been very gradually declining.
- Income-level patterns:
  - Informality averages 40 percent in the region’s low income countries.
  - Informality averages 35 percent for the region’s middle income countries.
- Other correlates:
  - Oil exporters and fragile countries are, everything else equal, more likely to harbor informality, with an informal economy well above 40 percent of GDP.
- Examples from PMM country groupings (size of informal economy as percent of GDP):
  - Low-Size Countries (0-20 percent): Mauritius, South Africa, Botswana, Lesotho, Swaziland, Namibia.
  - Middle-Size Countries (20-40 percent): Comoros, Cabo Verde, Zambia, Kenya, Ghana, Zimbabwe, Guinea, Eritrea, Tanzania, Gabon, Ethiopia, The Gambia, Mauritania, Uganda, Angola, Cameroon, Côte D'Ivoire, Malawi, Liberia, Madagascar, Equatorial Guinea, Mozambique, Burkina Faso, Burundi.
  - High-Size Countries (>40 percent): Congo, Republic of; Togo*; Guinea-Bissau; Nigeria*; Mali; Senegal*; Mali and others.
  - Note: Entries marked with * are based on survey estimates.

### Robustness and validation
- MIMIC results are found to be robust when cross-checked with MIPMM results; countries ordered in groups by MIPMM are broadly aligned with MIMIC findings.
- Cross-check against Statistical Agencies’ estimates for eight SSA countries shows:
  - Pearson correlation between National Accounts Statistics and MIMIC: 0.73.
  - Spearman's Rank Correlation: 0.857***.
- Survey coverage: 49 countries were identified to have survey-based estimates of the size of their informal economies, including nine in Sub-Saharan Africa.
- MIMIC model fit statistics reported in Figure 1:
  - RMSEA: 0.052
  - Chi-Square: 158.781
  - Observations: 1211
  - Countries: 139

### Interpretation and comparative advantages of MIMIC-based estimates
- MIMIC-generated panel data are useful because:
  - Not all countries publish national-account-based estimates of informality.
  - National statistical methodologies and sampling methods may affect cross-country comparability.
  - National estimates may fail to account for recent changes in domestic economies.
- The combination of modified MIMIC using night lights, PMM, and national accounts is presented as a novel application for generating robust estimates of informality in SSA.

### Policy implications
- Existing policy literature is limited on the determinants of variation in the size of the informal economy and on policy prescriptions to shift activity to the formal sector.
- Policy advice emphasized in related studies (for example International Monetary Fund 2017) includes:
  - Importance of allowing informal activity to act as a social safety net given that transition from informal to formal activity is likely to be long.

*Source: 7. CONCLUSIONS, wp17156 - 7. CONCLUSIONS*

---


_Source: https://www.imf.org/-/media/files/publications/wp/2017/wp17156.pdf_