## Annex 3.2 Internet Penetration and Macroeconomic Impacts

## Source details

**Canonical URL:** [Annex 3.2 Internet Penetration and Macroeconomic Impacts](https://www.imf.org/-/media/files/publications/reo/afr/2020/april/english/onlineannexch3.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/reo/afr/2020/april/english/onlineannexch3.pdf.md)
- [Structured JSON version](/-/media/files/publications/reo/afr/2020/april/english/onlineannexch3.pdf.json)

---

### Overview
- Purpose: Provide an overview of the data and econometric approach underpinning analysis of the impact of internet penetration on macroeconomic outcomes in sub-Saharan Africa (SSA).
- Identification strategy: Exploit the large-scale arrival of submarine cables in SSA during 2009-12 as an exogenous source of variation for internet penetration at the country level.
- Instruments used:
  - Dummy: if a country is directly connected to a submarine cable after 2009 (IV1).
  - Continuous: data capacity of submarine cables (IV2).

### Data and sample
- Internet penetration measure: percent of population with access to the internet (International Telecommunication Union).
- Macroeconomic variables: IMF (real per capita GDP, PPP 2011 international dollar), United Nations (sector value added, constant 2010 prices), ILO (sector employment shares), Penn World Tables 9.1 (total employment: number of persons engaged).
- Sample: 33 SSA countries over 2005–17 (largest sample used in regression analysis).
- Internet penetration summary:
  - Average increases from 5.0 percent in 2008 to 24.9 percent in 2017.
  - 2017 penetration rates range from 1.3 percent to 62 percent.
  - Average in 2017: treatment countries 28.3 percent; control countries 18.2 percent.
  - Change between 2008 and 2017: treatment countries +22 percentage points; control countries +15.2 percentage points.

### Empirical strategy
- Method: Two-stage least squares (2SLS) with first-stage fuzzy difference-in-differences and alternative first-stage using cable capacity; country- and time-fixed effects included.
- First-stage (difference-in-differences):
  - Key regressor: Treatment*Post2009 interaction (Treatment = directly connected to a submarine cable; Post2009 includes 2009).
  - Controls X: lagged per capita income, public investment, trade openness, terms of trade, population density, institutional quality, human capital (years of schooling), private investment, fiscal position; robustness checks include access to electricity.
- Second-stage:
  - Dependent variables: real GDP per capita growth, sector shares in value added, sectoral shares in employment.
  - Fitted internet penetration from first stage (IV1 or IV2) used as regressor.
- Estimation details:
  - Robust standard errors clustered at the country level.
  - Alternative specifications include fixed-effects (non-IV) and IV estimates.

### Identification assumptions and instrument validity
- Identification assumptions:
  1. Absent the arrival of submarine cables, the difference in internet penetration rates between coastal and landlocked countries would have remained broadly unchanged.
  2. Timing of arrival and capacity of submarine cables depends little on macroeconomic conditions in the countries they connect.
- Relevance evidence:
  - Difference-in-differences estimate: internet penetration increased by 3.2 percentage points more in treatment countries than in control countries in 2009-17 (Table 3.2.1, column 1).
  - Installed cable capacity coefficient is positive and highly statistically significant (Table 3.2.1, column 2).
- Exogeneity evidence:
  - Scatterplot and chart do not show a systematic relationship between income and cable capacity (Figure 3.2.2).
  - Rationale: submarine cables are deployed regionally and funded by large international consortia; most deterministic factor of capacity is technology available at the time.
- Pre-treatment test:
  - Difference-in-differences impact of cable arrival on internet penetration is statistically insignificant in the pre-treatment period (2000-2008), supporting identification.

### Main results — Macroeconomic impacts
- Real per capita GDP growth:
  - OLS-fixed effects: each percentage point increase in internet penetration is associated with 0.13 percentage point increase in the growth rate of real per capita GDP.
  - IV (using submarine cable capacity, IV2): each percentage point increase in internet penetration is associated with 0.37 percentage point increase in growth (column 6).
- Sectoral value added composition (Table 3.2.1):
  - Internet penetration boosts the share of services in total value added (column 7).
  - Internet penetration reduces the share of industry in total value added (column 9).
  - Impact on agriculture is insignificant (columns 11 and 12).
- Employment levels and composition (Table 3.2.2):
  - No significant impact on the level of total employment (columns 1 and 2).
  - No significant impact on industry employment share (columns 3 and 4).
  - Significant increase in services employment share: each percentage point increase in internet penetration leads to an increase of 0.3-0.4 percentage points in the share of services employment (columns 5 and 6).
  - Impact on services employment is larger for female than for male (columns 7, 9 and 10).

### Robustness checks and potential confounders
- External shocks addressed:
  - 2014-15 terms of trade shock: controlled for by including terms of trade as a control variable.
  - 2008-09 financial crisis: addressed by excluding countries heavily affected by the crisis in alternative regressions; results hold.
- Additional robustness:
  - Results robust to inclusion of access to electricity as a control.
  - Results robust to exclusion of outlier top performers on internet penetration.
  - Results hold when adjusting the sample end-year across different years in 2013-2017.

### Annex 3.3 — Digital Connectivity and Firm Performance in sub-Saharan Africa: Data and approach (summarized)
- Data source: World Bank Enterprise Surveys (WBES); formal manufacturing and services firms.
- Digital connectivity measure: binary indicator equal to one if firm reports using email to communicate with clients or suppliers at time of survey.
- Sample: more than 18,000 observations from 39 SSA countries.
- Key sample statistic: 57.2 percent of firms in SSA use email for business purposes (World average: 71 percent).
- Identification and instrument:
  - IV model with country-sector-year and location fixed effects.
  - Instrument: interaction between annual frequency of seaquakes and firm’s distance to closest connectivity infrastructure (submarine cable landing station or internet exchange point).
  - Instrument calibration: seaquakes with epicenter at least 100 km but less than 1,000 km from the submarine cable landing station; magnitude below 6.5.
- Main IV results (Annex Table 3.3.1 highlights):
  - First-stage: Seaquake frequency × log distance coefficients: -0.702***, -0.781***, -0.655***, -0.702*** (columns 1–4) and -0.384*** (manufacturing columns).
  - All Firms:
    - Log real sales: coefficient 0.965*; impact relative to non-email users: 2.6; impact in percent: 162%.
    - Log # workers: coefficient 2.100***; impact relative to non-email users: 8.2; impact in percent: 717%.
    - Log # temporary workers: coefficient -2.140***; impact relative to non-email users: 0.1; impact in percent: -88%.
    - Log # permanent workers: coefficient 2.229***; impact relative to non-email users: 9.3; impact in percent: 829%.
  - Manufacturing subsample:
    - Log # perm. non-production workers: coefficient 2.555***; impact relative to non-email users: 12.9; impact in percent: 1,187%.
    - Log # perm. production workers: coefficient 1.160***; impact relative to non-email users: 3.2; impact in percent: 219%.
    - Log # skilled production workers: coefficient 3.795***; impact relative to non-email users: 44.5; impact in percent: 4,348%.
    - Log # unskilled production workers: coefficient -8.877***; impact relative to non-email users: 0.0; impact in percent: -100%.
- Interpretation:
  - Email use is associated with higher sales and employment.
  - Digitally connected firms create more permanent, full-time jobs and reduce demand for temporary workers.
  - In manufacturing, digitalization shifts workforce composition toward non-production and skilled production workers, implying a skill-biased effect.
  - Policy implication: address skills shortages and mismatch in SSA labor markets to fully reap digital dividend.

### Annex 3.4 — Impact of Digitalization on Fiscal Revenues: Data and approach (summarized)
- Dependent variable: government revenue as percentage of GDP; two fiscal variables used: total government revenue excluding grants, and tax revenue.
- Digitalization measures: proportion of households with internet access (ITU); share of mobile cellular subscriptions per 100 people (ITU).
- Controls: log GDP per capita, economic growth, trade openness, terms of trade, agriculture sector share, corruption index, inflation rate, and education.
- Estimation: panel fixed-effects and 2SLS (IV). Instrument for fixed broadband subscriptions: binary indicator equal to 1 if country is connected to a submarine cable.
- Sample: 37 sub-Saharan African countries over 2000-18.

### Annex 3.4 — Main results (Annex Table 3.4.1 highlights)
- Selected digitalization coefficients (preserve reported values):
  - OLS / Panel fixed-effects (Total revenue excluding grants, Access to internet): 0.0411*** (column 1).
  - OLS / Panel fixed-effects (Tax revenue, Mobile phone subscription): 0.0269** (column 2).
  - OLS / Panel fixed-effects (Total revenue excluding grants, Access to internet, column 3): 0.3781*.
  - OLS / Panel fixed-effects (Tax revenue, Mobile phone subscription, column 4): 0.2645*.
  - IV estimates (Total revenue excluding grants, Access to internet, column 5): 0.0814*.
  - IV estimates (Tax revenue, Mobile phone subscription, column 6): 0.1240***.
  - IV estimates (Total revenue excluding grants, column 7): 0.6743**.
  - IV estimates (Tax revenue, column 8): 0.5740***.
- Magnitude interpretation:
  - A move from the 25th to 75th percentile of digitalization is associated with an increase in government revenue excluding grants by up to 2.1 percent of GDP (column 3).
- First-stage:
  - Submarine cable indicator significant in first stage (reported coefficients 1.3871***; and 0.0484*** / 0.0511*** in other first-stage columns).
- Diagnostics and sample stats:
  - Observations reported across columns: 321, 321, 393, 390, 321, 321, 374, 371.
  - Number of countries: 37 in all columns.
  - Cragg-Donald Wald F statistic: 18.129; 132.251; 130.356 (reported across specifications).
  - Anderson-Rubin Wald test p-values: 0.0150; 0.0110; 0.0520; 0.026.
  - Hansen p-values: 0.7430; 0.9330; 0.5890; 0.319.

### Annex 3.5 — Mobile Phone Ownership and Welfare: Evidence from South Africa’s household surveys (summarized)
- Data: South Africa’s National Income Dynamic Survey, 5 waves (2008–17) with unique individual IDs.
- Trends:
  - Mobile phone ownership has grown faster among less affluent individuals between 2008 and 2017 (Annex Figure 3.5.1).
  - Individuals who owned a mobile phone in 2008 were more likely to move up the expenditure distribution by 2017 (Annex Figure 3.5.2).
- Channels (literature evidence):
  - Mobile phones help save on travel time and costs: 76 percent (survey evidence in Africa, World Bank 2016).
  - Mobile phones help in finding work: around 40 percent.
  - Mobile phones help access free education content: 8 percent.
  - Mobile money fosters risk sharing; direct evidence of promotion of welfare and savings is less robust (Aron 2018).
- Empirical findings (South Africa):
  - Estimation: dependent variable log of real expenditure; controls for macroeconomic conditions and individual characteristics; lagged real expenditure included.
  - Main quantitative result: Having a mobile phone would lead to a rise in real expenditure by about 10 percent over two years.
  - Distributional finding: Positive impact appears to accrue more for less affluent mobile phone owners.

*Source: Authors' estimates*

### Annex 3.2 Internet Penetration and Macroeconomic Impacts

### Annex 3.2 Internet Penetration and Macroeconomic Impacts

### Overview
- Purpose: Provide an overview of the data and econometric approach underpinning analysis of the impact of internet penetration on macroeconomic outcomes in sub-Saharan Africa (SSA).
- Identification strategy: Exploit the large-scale arrival of submarine cables in SSA during 2009-12 as an exogenous source of variation for internet penetration at the country level.
- Instruments used:
  - Dummy: if a country is directly connected to a submarine cable after 2009 (IV1).
  - Continuous: data capacity of submarine cables (IV2).

### Data and sample
- Internet penetration measure: percent of population with access to the internet (International Telecommunication Union).
- Macroeconomic variables: IMF (real per capita GDP, PPP 2011 international dollar), United Nations (sector value added, constant 2010 prices), ILO (sector employment shares), Penn World Tables 9.1 (total employment: number of persons engaged).
- Sample: 33 SSA countries over 2005–17 (largest sample used in regression analysis).
- Internet penetration summary:
  - Average increases from 5.0 percent in 2008 to 24.9 percent in 2017.
  - 2017 penetration rates range from 1.3 percent to 62 percent.
  - Average in 2017: treatment countries 28.3 percent; control countries 18.2 percent.
  - Change between 2008 and 2017: treatment countries +22 percentage points; control countries +15.2 percentage points.

### Empirical strategy
- 2SLS framework with first-stage fuzzy difference-in-differences (Duflo 2001) and an alternative first-stage using cable capacity controlling for country- and time-fixed effects.
- First-stage specification (difference-in-differences):
  - Key regressor: Treatment*Post2009 interaction (Treatment = directly connected to a submarine cable; Post2009 includes 2009).
  - Controls vector X includes: lagged per capita income, public investment, trade openness, terms of trade, population density, institutional quality, human capital (years of schooling), private investment, fiscal position, and in robustness checks access to electricity.
- Second-stage specification:
  - Dependent variables: real GDP per capita growth, sector shares in value added, and sectoral shares in employment.
  - Fitted internet penetration from first stage (IV1 or IV2) used as regressor.
- Estimation details:
  - Robust standard errors clustered at the country level.
  - Alternative specifications include fixed-effects (non-IV) and IV estimates.

### Identification assumptions and instrument validity
- Identification assumptions:
  1. Absent the arrival of submarine cables, the difference in internet penetration rates between coastal and landlocked countries would have remained broadly unchanged.
  2. Timing of arrival and capacity of submarine cables depends little on macroeconomic conditions in the countries they connect.
- Relevance evidence:
  - Difference-in-differences estimate: internet penetration increased by 3.2 percentage points more in treatment countries than in control countries in 2009-17 (Table 3.2.1, column 1).
  - Installed cable capacity coefficient is positive and highly statistically significant (Table 3.2.1, column 2).
- Exogeneity evidence:
  - Scatterplot and chart do not show a systematic relationship between income and cable capacity (Figure 3.2.2).
  - Rationale: submarine cables are deployed regionally and funded by large international consortia; most deterministic factor of capacity is technology available at the time.

### Main results
- Real per capita GDP growth:
  - OLS-fixed effects: each percentage point increase in internet penetration is associated with 0.13 percentage point increase in the growth rate of real per capita GDP.
  - IV (using submarine cable capacity, IV2): each percentage point increase in internet penetration is associated with 0.37 percentage point increase in growth (column 6).
- Sectoral value added composition (Table 3.2.1):
  - Internet penetration boosts the share of services in total value added (column 7).
  - Internet penetration reduces the share of industry in total value added (column 9).
  - Impact on agriculture is insignificant (columns 11 and 12).
- Employment levels and composition (Table 3.2.2):
  - No significant impact on the level of total employment (columns 1 and 2).
  - No significant impact on industry employment share (columns 3 and 4).
  - Significant increase in services employment share: each percentage point increase in internet penetration leads to an increase of 0.3-0.4 percentage points in the share of services employment (columns 5 and 6).
  - Impact on services employment is larger for female than for male (columns 7, 9 and 10).

### Robustness checks and potential confounders
- Pre-treatment test: difference-in-differences impact of cable arrival on internet penetration is statistically insignificant in the pre-treatment period (2000-2008), supporting identification.
- External shocks addressed:
  - 2014-15 terms of trade shock: controlled for by including terms of trade as a control variable.
  - 2008-09 financial crisis: addressed by excluding countries heavily affected by the crisis in alternative regressions; results hold.
- Additional robustness:
  - Results robust to inclusion of access to electricity as a control.
  - Results robust to exclusion of outlier top performers on internet penetration.
  - Results hold when adjusting the sample end-year across different years in 2013-2017.

*Source: Authors' estimates*

### References

### onlineannexch3 - References

### References (selected)
- Autor, D. H., & Dorn, D. (2009). Inequality and Specialization: The Growth of Low-skill Service Jobs in the United States. NBER Working Paper Series, 15150.
- Cariolle, J. (2018). Telecommunication Submarine-Cable Deployment and the Digital Divide in Sub Saharan Africa. Ferdi Working Paper P241, May 2019.
- Czernich, N., Falck, O., Kretschmer, T., & Woessmann, L. (2011). Broadband infrastructure and economic growth. The Economic Journal, 121(552), 505-532.
- Duflo, E. (2001). Schooling and Labor Market Consequences of School Construction in Indonesia: Evidence from an Unusual Policy Experiment. American Economic Review. September.
- Hjort J. and Poulsen J. (2019). The Arrival of Fast Internet and Employment in Africa. American Economic Review, 109 (3): 1032-79.
- Houngbonon, Georges V.; Liang, Julienne (2018). The Impact of Broadband Internet on Employment in France. 29th European Regional Conference of the International Telecommunications Society (ITS).
- Nolen, P. (2008). Does ICT Benefit the Poor? Evidence from South Africa. Working paper.
- Qiang, C. Z. W., Rossotto, C. M., & Kimura, K. (2009). Economic Impacts of Broadband. Information and Communications for Development 2009: Extending reach and increasing impact, 3, 35-50.
- Sovbetov, Y. (2018). Impact of Digital Economy on Female Employment: Evidence from Turkey. International Economic Journal, 32(2), 256-270.
- World Bank. 2018. Innovative Business Models for Expanding Fiber-Optic Networks and Closing the Access Gaps.
- Zaballos, A. G., & López-Rivas, R. (2012). Socioeconomic Impact of Broadband in Latin American and Caribbean Countries. Inter-American Development Bank, 220.
- Additional references cited in annex text (Aceto et al., Carter et al., Clark, Clarke et al., Grubesic et al., Cariolle et al., etc.).

### Annex 3.3 — Digital Connectivity and Firm Performance in sub-Saharan Africa: Data and Approach
- Data source: World Bank Enterprise Surveys (WBES), covering formal manufacturing and services firms.
- Measure of digital connectivity: binary indicator equal to one if firm reports using email to communicate with clients or suppliers at time of survey.
- Key sample statistics:
  - 57.2 percent of firms in SSA use email for business purposes.
  - World average email use by firms: 71 percent.
- Sample coverage: more than 18,000 observations from 39 SSA countries.
- Standard errors clustered at the country-location-sector level.
- Geographical data sources: Telegeography, PCH, PeeringDB; seaquake data from Northern California Earthquake Data Center (University of Berkeley).

### Annex 3.3 — Identification and Instrument
- Estimation approach: instrumental variables (IV) model with country-sector-year and location fixed effects.
- Instrument: interaction between annual frequency of seaquakes and firm’s distance to closest connectivity infrastructure (submarine cable landing station or internet exchange point).
- Instrument calibration:
  - Only seaquakes whose epicenter is at least 100 km but less than 1,000 km away from the submarine cable landing station are considered.
  - Only seaquakes with magnitude below 6.5 on the Richter scale are considered.
- Rationale: maritime seismic events can cause submarine cable wear and breaks; locations distant from key nodes are more vulnerable to telecommunications disruptions.

### Annex 3.3 — Main Results (summary of Annex Table 3.3.1)
- First-stage: Seaquake frequency × log distance has a negative and statistically significant relationship with email use (coefficients reported as -0.702***, -0.781***, -0.655***, -0.702*** for columns 1–4 and -0.384*** for manufacturing columns).
- IV estimates (email use effects, relative to non-email users):
  - Log real sales (All Firms): coefficient 0.965*; impact relative to non-email users: 2.6; impact in percent: 162%.
  - Log # workers (All Firms): coefficient 2.100***; impact relative to non-email users: 8.2; impact in percent: 717%.
  - Log # temporary workers (All Firms): coefficient -2.140***; impact relative to non-email users: 0.1; impact in percent: -88%.
  - Log # permanent workers (All Firms): coefficient 2.229***; impact relative to non-email users: 9.3; impact in percent: 829%.
- Manufacturing subsample (columns 5–8):
  - Log # perm. non-production workers: coefficient 2.555***; impact relative to non-email users: 12.9; impact in percent: 1,187%.
  - Log # perm. production workers: coefficient 1.160***; impact relative to non-email users: 3.2; impact in percent: 219%.
  - Log # skilled production workers: coefficient 3.795***; impact relative to non-email users: 44.5; impact in percent: 4,348%.
  - Log # unskilled production workers: coefficient -8.877***; impact relative to non-email users: 0.0; impact in percent: -100%.
- Controls with significant coefficients (examples):
  - Log age: positive and significant in multiple specifications (e.g., 0.345*** for log real sales).
  - Log initial # employees: positive and significant across many outcomes (e.g., 0.683*** for log real sales).
  - Foreign ownership, direct exporter, log manager experience, bank loan show varied significant associations as reported in the table.
- Diagnostics and sample stats:
  - Number of observations varies by specification (e.g., 16,501; 18,614; 17,794; 18,403 for all-firm columns; 7,681; 7,716; 7,090; 7,581 for manufacturing).
  - Number of countries: 38–39 for all-firm columns; 26–30 for manufacturing columns.
  - RMSE and Kleibergen-Paap F-statistics reported per column (e.g., Kleibergen-Paap F-stat: 39.48; 44.75; 31.31; 44.37; 17.21; 17.22; 14.53; 14.65).
- Interpretation highlighted in text:
  - Email use is associated with higher sales and employment.
  - Digitally connected firms create more permanent, full-time jobs and reduce demand for temporary workers.
  - In manufacturing, digitalization shifts workforce composition toward non-production and skilled production workers, implying a skill-biased effect of digitalization.
  - Policy implication: importance of addressing skills shortages and mismatch in SSA labor markets to fully reap digital dividend.

### Annex 3.4 — Impact of Digitalization on Fiscal Revenues: Data and Approach
- Dependent variable: government revenue as percentage of GDP; two fiscal variables used: total government revenue excluding grants, and tax revenue.
- Digitalization measures:
  - Proportion of households with internet access (from UN International Telecommunication Union).
  - Share of mobile cellular subscriptions per 100 people (from UN International Telecommunication Union).
- Controls include: log GDP per capita, economic growth, trade openness (sum of imports and exports as percent of GDP), terms of trade, agriculture sector share (percent of GDP), corruption index, inflation rate, and education (from IMF WEO and World Bank WDI).
- Estimation methods: panel fixed-effects and two-stage least squares (IV). Instrument for fixed broadband subscriptions: binary variable equal to 1 if country is connected to a submarine cable and 0 otherwise.
- Sample: 37 sub-Saharan African countries over the period 2000-18.

### Annex 3.4 — Main Results (summary of Annex Table 3.4.1)
- Digitalization coefficient (selected reported values):
  - OLS / Panel fixed-effects (Total revenue excluding grants, Access to internet): 0.0411*** (column 1).
  - OLS / Panel fixed-effects (Tax revenue, Mobile phone subscription): 0.0269** (column 2).
  - OLS / Panel fixed-effects (Total revenue excluding grants, Access to internet, column 3): 0.3781*.
  - OLS / Panel fixed-effects (Tax revenue, Mobile phone subscription, column 4): 0.2645*.
  - IV estimates (Total revenue excluding grants, Access to internet, column 5): 0.0814*.
  - IV estimates (Tax revenue, Mobile phone subscription, column 6): 0.1240***.
  - IV estimates (Total revenue excluding grants, column 7): 0.6743**.
  - IV estimates (Tax revenue, column 8): 0.5740***.
- Magnitude interpretation noted in text:
  - A move from the 25th to 75th percentile of digitalization is associated with an increase in government revenue excluding grants by up to 2.1 percent of GDP (column 3).
- First-stage results for IV:
  - Submarine cable indicator significant in first stage (reported coefficients 1.3871*** in first-stage columns; and 0.0484*** / 0.0511*** in other first-stage columns).
- Diagnostics and sample stats:
  - Observations: reported values include 321, 321, 393, 390, 321, 321, 374, 371 across columns.
  - Number of countries: 37 in all columns.
  - R-squared ranges reported (e.g., 0.374, 0.401, 0.378, 0.418, 0.519, 0.589, 0.549, 0.616).
  - Robust standard errors reported; significance levels: ***, **, * indicate p<0.01, p<0.05, p<0.1.
  - Instrument diagnostics: Cragg-Donald Wald F statistic reported (e.g., 18.129; 132.251; 130.356), Anderson-Rubin Wald test p-values reported (e.g., 0.0150; 0.0110; 0.0520; 0.026), Hansen p-values reported (e.g., 0.7430; 0.9330; 0.5890; 0.319).

*Source: onlineannexch3 - References*

### Annex 3.5. Mobile Phone Ownership and Welfare: Evidence from South Africa’s

### Annex 3.5. Mobile Phone Ownership and Welfare: Evidence from South Africa’s Household Surveys

### Context and data
- Analysis based on South Africa’s household surveys because of relatively large survey data and a level of mobile phone ownership suited for the analysis.
- South Africa’s National Income Dynamic Survey data include 5 waves in about 2-year increments (2008–17) with unique individual IDs.

### Trends in mobile phone ownership
- Mobile phone ownership over the past decade has grown faster among less affluent individuals (illustrated by Annex Figure 3.5.1 comparing 2008 and 2017 by expenditure decile).
- Individuals who owned a mobile phone in 2008 were more likely to move up the expenditure distribution by 2017 (illustrated by Annex Figure 3.5.2 showing excess probability of moving to higher expenditure deciles from 2008 to 2017).

### Channels through which mobile phones may affect welfare (literature evidence)
- Mobile phones help save on travel time and costs: 76 percent (survey evidence in Africa, World Bank 2016).
- Mobile phones help in finding work: around 40 percent (same survey evidence).
- Mobile phones help access free education content: 8 percent (same survey evidence).
- Additional channels noted in the literature:
  - Greater and more affordable access to information, increasing market efficiency (e.g., sharing information on agricultural prices and labor markets).
  - Facilitation of networking and starting a business.
  - Increased disposable income, particularly for the most vulnerable.
- Mobile money fosters risk sharing; however, direct evidence of promotion of welfare and savings is less robust (Aron 2018).

### Empirical findings from South Africa’s surveys
- Estimation approach:
  - Dependent variable: log of real expenditure level.
  - Controls included: macroeconomic conditions and individuals’ characteristics.
  - Additional controls tested in different specifications: gender, ethnicity, employment status, access to finance (store card, credit card, access to different types of loans), per capita GDP (log level), location, location-by-time, and time.
  - Lagged real expenditure (log level) was included to help control for potential reverse causality (mobile phone ownership may be motivated by resource availability).
- Main quantitative result:
  - Having a mobile phone would lead to a rise in real expenditure by about 10 percent over two years.
- Distributional aspect:
  - The positive impact appears to accrue more for less affluent mobile phone owners.

### Reference cited in annex
- Aron, J. 2018. “Mobile Money and the Economy: A Review of the Evidence.” The World Bank Research Observer. Volume 33, Issue 2, August, 135–88.

*Prepared by Ken Miyajima.*

---


_Source: https://www.imf.org/-/media/files/publications/reo/afr/2020/april/english/onlineannexch3.pdf_
