## 1. The Impact of Reducing Consumption Volatility on Poverty

## Source details

**Canonical URL:** [1. The Impact of Reducing Consumption Volatility on Poverty](https://www.imf.org/-/media/files/publications/wp/2020/english/wpiea2020051-print-pdf.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/wp/2020/english/wpiea2020051-print-pdf.pdf.md)
- [Structured JSON version](/-/media/files/publications/wp/2020/english/wpiea2020051-print-pdf.pdf.json)

---

### Introduction, scope, and counterfactual metric
- Objective:
  - Study Sub-Saharan Africa households’ ability to maintain stable consumption via two case studies: South Africa and Tanzania using longitudinal micro-data (NIDS, LSMS).
- Counterfactual metric:
  - Define cαit = ci + α*(cit – ci) with 0≤α≤1; compute counterfactual poverty rates as share of households with cαit < cpoor for varying α.
- Key counterfactual findings (symmetric volatility reduction):
  - A 50-percent decline in consumption volatility (α that halves deviations) would:
    - Lower the baseline poverty rate by approximately one fifth in Tanzania.
    - Lower the poverty rate by one eighth in South Africa.
  - Example numeric movement (Tanzania): from just above 15 percent to just above 11 percent in the LSMS-adjusted series for a 50-percent volatility reduction.
- Interpretation:
  - Link between consumption volatility and poverty is stronger in Tanzania, likely due to weaker coping mechanisms.
  - Symmetric volatility reductions reduce poverty because fewer non-poor households fall below the line than poor households rise above it.
  - Mitigating only downside volatility (insurance-style) yields steeper aggregate poverty reductions than symmetric reductions.

### Poverty dynamics and transition probabilities (average across survey waves)
- South Africa transition matrix (Wave t-1 → Wave t):
  - Not poor → Not poor: 87%
  - Not poor → Poor: 13%
  - Poor → Not poor: 47%
  - Poor → Poor: 53%
- Tanzania transition matrix (average across waves 1–3):
  - Not poor → Not poor: 87%
  - Not poor → Poor: 13%
  - Poor → Not poor: 58%
  - Poor → Poor: 42%
- Stylized facts:
  - Non-poor households face on average a 13 percent probability of entering poverty across two survey periods.
  - About half of poor households exit poverty in each period.
  - Dynamic component of poverty is large and important for policy design.

### Coping mechanisms and cross-country comparisons
- Common mechanisms: borrowing, dissaving, transfers (public/private), adjusting labor supply.
- South Africa (Global Findex 2017 and survey):
  - Bank account ownership: 65% (total).
  - Borrowed in last 12 months: 33% (total).
  - By education: up to 9th grade — 46% account, 23% borrowed; secondary — 71% account, 33% borrowed; tertiary — 88% account, 53% borrowed.
  - By population group: Black African 63% account, 31% borrowed; Coloured 69% account, 40% borrowed; Whites 93% account, 58% borrowed.
  - By area: traditional 54% account, rural 62% account, urban 75% account.
  - Only 22% save at a financial institution (Global Findex Database 2017).
  - Public transfers targeted to bottom two-thirds; private transfers about 30% receive from residents.
- Tanzania:
  - Bank account ownership: 21% (total).
  - Borrowed in last 12 months: 13% (total).
  - Sources of borrowing (total sample figures): formal sector 14%, informal 51%, microcredit 13%, SACCO 30%.
  - By income source:
    - Agricultural households: 7% bank account, 9% borrowed, 5% formal sector borrowing, 65% informal borrowing, 9% microcredit, 26% SACCO.
    - Non-agricultural households: 24% bank account, 18% borrowed, 17% formal, 47% informal, 14% microcredit, 31% SACCO.
  - Mobile money:
    - IMF (2016): mobile money transactions amounted to 52 percent of GDP by 2015 (vs less than 1 percent in 2010).
    - Wave 4 LSMS: mobile money use higher than formal banking among low-income groups (example: lowest income percentile — 5 percent bank account vs 20 percent mobile money; 75th–90th percentile — 20–30 percent bank account vs 70 percent mobile money).
  - Tanzanian households more likely to adjust labor supply and rely less on formal savings.

### Policy-relevant interpretation
- South Africa: relatively high account ownership and social grants distributed via bank accounts facilitate mitigation; transfers and credit access materially reduce consumption losses.
- Tanzania: mobile money expands reach to unbanked and rural households, partially offsetting low banking access; agricultural volatility central to vulnerability.

---

### 2. Tanzania: Agricultural Shocks, Consumption, and Poverty

### Incidence of agricultural shocks (Waves 1–3 panel; by wave)
- All households:
  - Wave 2: 22% price shock, 30% quantity shock, 47% combined
  - Wave 3: 10% price shock, 31% quantity shock, 39% combined
- Non-Agricultural households:
  - Wave 2: 17% price shock, 26% quantity shock, 39% combined
  - Wave 3: 7% price shock, 28% quantity shock, 34% combined
- Agricultural households:
  - Wave 2: 29% price shock, 36% quantity shock, 59% combined
  - Wave 3: 14% price shock, 35% quantity shock, 46% combined

### Estimated impact on consumption (selected coefficients from Table 4)
- Main aggregated findings:
  - Agricultural shock associated with fall of 4½ percent in total consumption (Agricultural shock = -0.0449*; standard error 0.0248).
  - Impact almost three times larger for agricultural households (Agricultural shock*Agricultural Household = -0.116**; s.e. 0.0585 in Log Consumption).
  - Price shocks especially harmful for agricultural households (Price shock*Agricultural Household = -0.164**; s.e. 0.0706).
  - Quantity shocks often insignificant for aggregate but affect non-food consumption for agricultural households (Quantity shock*Agricultural Household = -0.187* in Non-Food Consumption).
- Selected coefficient table entries preserved:
  - Log (Consumption):
    - Agricultural shock: -0.0449* (0.0248)
    - Agricultural shock*Agricultural Household: -0.116** (0.0585)
    - Quantity shock: -0.0579** (0.0267)
    - Price shock*Agricultural Household: -0.164** (0.0706)
    - Log non-agricultural income: 0.0634*** (0.0120)
    - R-squared: 0.250; Number of households: 2,626
  - Log (Food Consumption):
    - Agricultural shock: -0.0934*** (0.0280)
    - Quantity shock: -0.0950*** (0.0309)
    - Price shock*Agricultural Household: -0.195** (0.0849)
    - R-squared: 0.146; Number of households: 2,605
  - Log (Non-Food Consumption):
    - Agricultural shock*Agricultural Household: -0.212** (0.100)
    - Quantity shock*Agricultural Household: -0.187* (0.108)
    - Log non-agricultural income: 0.105*** (0.0208)
    - R-squared: 0.227; Number of households: 2,626
- Notes:
  - Heteroskedasticity-robust standard errors in parentheses.
  - Controls: education and age of head, household size, number of workers, wave fixed effects, constant.
  - Significance codes: *** p<0.01, ** p<0.05, * p<0.1.

### Volatility and poverty (Table 5 summary)
- Agricultural shock increases poverty incidence by 4 to 5 percentage points:
  - OLS Agricultural shock = 0.0479*** (0.0139).
- Interaction Agricultural shock*Agricultural Household negative but not statistically significant (e.g., -0.0321 in OLS, not significant).
- Logit results corroborate OLS:
  - Agricultural shock Logit coefficients range 0.464*** to 0.606*** (s.e. 0.117 to 0.139).
- Interpretation:
  - Although consumption falls more for agricultural households, probability of crossing poverty line due to shocks similar across household types — agricultural volatility affects non-agricultural households as well.

### Agricultural income, volatility decomposition, and shock coding (Appendix B)
- Agricultural income constructed from sales, imputed non-sold production, wage labor and enterprise income.
- Primary crop accounts for on average 70 percent of household agricultural production value.
- Decomposition: quantities (within-basket changes and crop turnover) account for bulk of agricultural income volatility; prices account for small fraction.
- Price shock coded as price change below 25th percentile of pooled log changes; quantity shock coded as quantity change below 25th percentile or discontinuation.

---

### 3. South Africa: Labor Shocks, Transfers, Credit Access, and Poverty Mitigation

### Context and employment shock definition
- Labor income main pre-transfer source; agriculture <10 percent of income.
- Employment shock: transition out of formal employment or from informal employment to unemployment for the household member with highest labor income.

### Transition patterns and sample sizes
- Number of observations reported across groups (examples preserved from source): total observations: 5,590; rural: 3,489; urban: 4,229.
- Employment transition regression results (selected, Table A.6):
  - Transitions from formal to no employment: large negative effects (examples preserved): -0.696***, -0.671***, -7.688***, -7.585*** across specifications.
  - Transitions to formal employment: positive significant effects (e.g., 0.654***, 0.635***, 7.332***, 7.240***).

### Baseline difference-in-differences (Table 7) — average percent gains/losses from employment shock
- Pre-transfer income: average % gain/loss = -0.602*** (60 percent drop)
- Government transfers: 0.577*** (increase of about 60 percent)
- Private transfers: 0.570*** (increase about 60 percent)
- Total after-transfer income: -0.332*** (33 percent drop)
- Consumption: -0.172*** (17 percent drop)
- Effects of credit access:
  - Consumption loss with credit access: 7 percent on average
  - Consumption loss without credit access: 21 percent on average
  - Poverty probability increase after employment shock:
    - Without credit access: 9 percentage points
    - With credit access: 4 percentage points
- Observations for specifications: 12,066; 12,314 (varies by column)
- R-squared values reported: 0.321; 0.073; 0.062; 0.214; 0.180; 0.344

### Robustness and heterogeneity
- Non-food consumption drops more (24 percent) than total consumption (17 percent).
- Informal borrowing matters for mitigation; formal savings do not show significant mitigating effect.
- Triple difference-in-differences (Equation (4), Table A.8) indicates mitigating effect of financial access does not differ significantly across education, population, or living area groups (Wald tests fail to reject equality).

---

### 4. Data, poverty lines, and descriptive statistics (selected exact figures)

### South Africa (NIDS)
- Original sample: 7,305 households; longitudinal constructed sample: 5,400 households that responded in at least two consecutive waves.
- Waves: 2008; 2010/11; 2012; 2014/15; 2017.
- Household demographics (means by wave):
  - Mean household members: 4.02, 4.35, 4.34, 4.20, 4.15 (Waves 1–5).
  - Mean working-age adults: 1.56, 1.64, 1.56, 1.68, 1.70.
- Consumption percentiles (2010 prices; pooled):
  - Total consumption: 25th 20,615; 50th 33,185; 75th 62,479; Mean 64,838.
- National poverty lines (STATSSA; April 2018 prices):
  - Food poverty line: R547 per person per month.
  - Lower-bound poverty line: R785 per person per month.
  - Upper-bound poverty line: R1,183 per person per month.
- Percent below food poverty line by wave (total): 19% (2008), 26% (2010/11), 25% (2012), 19% (2014/15), 19% (2017).

### Tanzania (LSMS)
- Balanced panel Waves 1–3: 3,088 household-level observations; Wave 4 cross-section: 3,352 observations.
- Waves: 2008/09; 2010/11; 2012/13; 2014/15.
- Household demographics (means by wave):
  - Mean household members: 5.13, 5.30, 5.25, 4.74.
  - Mean working-age adults: 2.97, 3.07, 3.07, 2.70.
- Consumption percentiles (2010 prices; pooled):
  - Total consumption: 25th 1,260; 50th 2,034; 75th 3,328; Mean 2,842.
- Poverty line: 847 Tsh. per adult equivalent of weekly consumption in 2010 prices.
- Percent poor by wave (All): 12% (2008/09), 16% (2010/11), 18% (2012/13), 15% (2014/15).
- Main income source shares: Agricultural 40%, 40%, 40%, 30% (Waves 1–4); Non-agricultural 60%, 60%, 60%, 70%.

---

### 5. Selected conclusions and policy implications (preserved language and findings)
- Dynamic nature of poverty:
  - Sizeable flows into and out of poverty imply both policies targeting persistently poor and policies assisting households just above the poverty line are needed.
- Country-specific policy-relevant findings:
  - South Africa:
    - Employment loss reduces pre-transfer income by about 60 percent.
    - Government transfers halve the impact of the shock on total income.
    - Access to credit reduces the negative impact of employment loss on consumption by two thirds.
  - Tanzania:
    - Agricultural shocks are frequent across household types.
    - Consumption fall is significantly larger (approximately 10 percent) for agriculture-intensive households.
- Access to coping mechanisms:
  - Banking services usage strongly correlated with income in both countries.
  - Tanzania: mobile money partially offsets unequal banking access and facilitates remittances in shock years.
  - South Africa: government transfers combined with financial coverage are effective risk mitigators for those with access.

*Source: wpiea2020051-print-pdf — 1. The Impact of Reducing Consumption Volatility on Poverty (NIDS, LSMS, and authors’ calculations).*

### 1. The Impact of Reducing Consumption Volatility on Poverty _____________________________________________ 11

### 1. The Impact of Reducing Consumption Volatility on Poverty

### Main analytical themes
- Quantitative assessment of how reductions in consumption volatility affect poverty dynamics.
- Estimation methods likely include consumption and poverty regressions (see Tables 5 and 6).
- Cross-country focus on South Africa and Tanzania with country-specific appendices (A and B) and a joint appendix (C).

### Related tables and appendices referenced in this unit
- Table 2. Transition Probabilities into and out of Poverty in South Africa and Tanzania — page 9.
- Table 5. Tanzania: Estimated Coefficients of Consumption Regression — page 20.
- Table 6. Tanzania: Estimated Coefficients of Poverty Regression — page 21.
- Appendix A. South Africa: Data Details, Descriptive Statistics, Further Analysis — page 29.
- Appendix B. Tanzania: Data Details, Descriptive Statistics, Further Analysis — page 35.
- Appendix C. Both Countries: Data Details, Descriptive Statistics, Further Analysis — page 41.

### Expected empirical content (by section titles)
- Measurement of consumption volatility and its distribution across households.
- Regression-based identification of the effect of volatility reduction on consumption levels and poverty incidence.
- Transition analysis capturing movements into and out of poverty in both countries.

### 2. Incidence of Mechanisms for Consumption Insurance in South Africa and Tanzania from the Global Findex Database 2017

### Main analytical themes
- Incidence and distribution of consumption insurance mechanisms as captured in Global Findex 2017 for South Africa and Tanzania.
- Comparison across groups and likely linkage to poverty dynamics and consumption smoothing capacity.

### Related tables and appendices referenced in this unit
- Table 3. Use of Financial Services Across Groups — page 13.
- Table 2. Transition Probabilities into and out of Poverty in South Africa and Tanzania — page 9.
- Appendix B. Tanzania: Data Details, Descriptive Statistics, Further Analysis — page 35.
- Appendix A. South Africa: Data Details, Descriptive Statistics, Further Analysis — page 29.

### 3. Use of Traditional Financial Services by Income and Population Group

### Main analytical themes
- Patterns of traditional financial services usage disaggregated by income and population group.
- Likely identification of gaps in access and usage that affect the capacity to insure consumption shocks.

### Related table
- Table 3. Use of Financial Services Across Groups — page 13.

### 4. Tanzania: Use of Mobile Money Versus Traditional Banking

### Main analytical themes
- Comparative incidence and role of mobile money relative to traditional banking in Tanzania.
- Implications for consumption smoothing and poverty mitigation.

### Related tables
- Table 4. Tanzania: Incidence of Agricultural Shocks Among Households — page 18.
- Table 5. Tanzania: Estimated Coefficients of Consumption Regression — page 20.
- Table 6. Tanzania: Estimated Coefficients of Poverty Regression — page 21.
- Appendix B. Tanzania: Data Details, Descriptive Statistics, Further Analysis — page 35.

### 5. South Africa: Public and Private Transfer Beneficiaries by Income and Group

### Main analytical themes
- Distributional incidence of public and private transfers in South Africa across income and demographic groups.
- Analysis of transfers’ role in mitigating consumption volatility and affecting poverty transitions.

### Related tables and estimations
- Table 7. South Africa: Transitions Across Employment Statuses by Group — page 23.
- Table 8. South Africa: Baseline Difference-in-Difference Estimation — page 24.
- Table 9. Difference in the Magnitude of Mitigation across Groups: Triple Difference-in-Difference Estimation Results — page 25.
- Appendix A. South Africa: Data Details, Descriptive Statistics, Further Analysis — page 29.

*Source: wpiea2020051-print-pdf - 1. The Impact of Reducing Consumption Volatility on Poverty — canonical PDF.*

### REFERENCES ________________________________________________________________________________________________ 27

### wpiea2020051-print-pdf - REFERENCES ________________________________________________________________________________________________ 27

### Introduction: scope, approach, and key counterfactual metric
- Objective:
  - Study Sub-Saharan Africa (SSA) households’ ability to maintain stable consumption in adverse conditions via two case studies: South Africa and Tanzania.
  - Use longitudinal micro-data to examine consumption and poverty dynamics, quantify impact of consumption volatility on poverty, identify risk-sharing gaps and policy solutions.
- Data sources:
  - South Africa: National Income Dynamics Study (NIDS).
  - Tanzania: Living Standards Measurement Survey (LSMS).
- Core methodological innovation:
  - Propose a simple metric shrinking household-specific consumption deviations by a factor α (0≤α≤1). Define cαit = ci + α*(cit – ci). Counterfactual poverty rates computed as percent of households with cαit < cpoor for varying α.
- Key counterfactual findings (symmetric volatility reduction):
  - A 50-percent decline in consumption volatility (i.e., α reducing volatility by half) would lower the baseline poverty rate by approximately one fifth in Tanzania.
  - The same 50-percent decline would lower the poverty rate by one eighth in South Africa.
  - Interpretation: link between consumption volatility and poverty is stronger in Tanzania, likely due to weaker coping mechanisms.

### Literature review: existing evidence and gaps
- Tanzania literature:
  - Mkenda et al. (2010) on poverty 2000–2007 and demographic correlates.
  - Atkinson and Lugo (2010) on poverty and inequality (not longitudinal).
  - Sectoral consumption studies: Hosier and Kipondya (1993); Weinberger and Sai (2003); Abdulai and Auber (2004); Mason et al. (2015); Kaminski et al. (2016).
  - Sarris and Karfakis (2006) study agricultural shocks in two crop-producing regions.
  - Fiscal policy: Younger et al. (2016) find Tanzania’s taxation and expenditure programs highly redistributive but limited in poverty reduction relative to peers.
- South Africa literature:
  - Schotte et al. (2017) classify households by poverty transition likelihood using NIDS.
  - Flato et al. (2017) examine rainfall variation effects.
  - Anand et al. (2016) analyze unemployment flows and inequality implications.
  - Redistribution via social grants recognized as effective (Finn and Leibbrandt, 2017).
  - Tondini (2017); Tondini et al. (2017) show mixed effects of transfers on employment transitions.

### Micro-level data: samples, waves, and population groups
- South Africa (NIDS):
  - Original sample: 7,305 households interviewed bi-annually in five waves: 2008, 2010/11, 2012, 2014/15, 2017.
  - Constructed longitudinal sample: keep 5,400 households that responded in at least two consecutive waves.
  - Comparisons across ethnicity (Black African, Colored, White), education (up to 9th grade, secondary above 9th grade, tertiary), and living areas (urban, rural, traditional).
  - Nominal values reported in December 2016 prices (South African Rand, R.).
- Tanzania (LSMS):
  - Waves collected in 2008/2009, 2010/2011, 2012/2013, 2014/2015.
  - Balanced panel for Waves 1-3: 3,088 household-level observations (present in all three waves).
  - Wave 4 cross-section: 3,352 observations (fully new sample).
  - Nominal values reported in 2009 prices (Tanzanian Shillings, Tsh.).
  - Main heterogeneity: primary source of income categorized into “agricultural” (approximately 40 percent of the sample) and “non-agricultural”.
- Note on poverty lines:
  - South Africa: national food poverty line (corresponds roughly to World Bank USD 1.9 per day).
  - Tanzania: LSMS/World Bank consumption-based poverty definition.
  - Measures are country-specific and not directly comparable.

### Poverty dynamics and consumption smoothing: stylized facts and the volatility–poverty metric
- Aggregate and group-level poverty patterns:
  - South Africa: NIDS sample poverty levels around 20 percent (STATSSA, 2018); no clear upward trend from 2008 to 2017 in sample.
  - Tanzania: poverty rose from 12 percent to 18 percent between 2008/9 and 2012/3, then fell to 15 percent in the last wave.
  - In South Africa, poverty is more prevalent among less educated households, in rural/traditional areas, and among Black Africans.
  - In Tanzania, poverty is higher among agriculture-based households, which account for much of the increase over time.
- Transition probabilities (average across survey waves):
  - South Africa transition matrix (Wave t-1 → Wave t):
    - Not poor → Not poor: 87%
    - Not poor → Poor: 13%
    - Poor → Not poor: 47%
    - Poor → Poor: 53%
  - Tanzania transition matrix (average across waves 1–3):
    - Not poor → Not poor: 87%
    - Not poor → Poor: 13%
    - Poor → Not poor: 58%
    - Poor → Poor: 42%
  - Implication: sizeable flows into and out of poverty; dynamic component important for policy design.
- Counterfactual volatility metric and results:
  - Define household log consumption cit = ci + (cit – ci); shrink deviations by α to get cαit.
  - Compute counterfactual poverty rates across α∈[0,1].
  - Findings highlighted above: 50-percent volatility reduction → poverty lower by ~25 percent in Tanzania (from just above 15 percent to just above 11 percent in the LSMS-adjusted series) and lower by one eighth in South Africa.
  - Symmetric volatility reductions (muting both upside and downside) reduce poverty because more non-poor households are prevented from temporarily falling below the line than poor households are prevented from temporarily rising above it.
  - Mitigating only downside volatility (insurance-style policies) yields steeper reductions in aggregate poverty than symmetric reductions because downside mitigation does not reduce temporary exits from poverty.

### Coping mechanisms for consumption smoothing: cross-country comparisons and heterogeneity
- Common mechanisms: borrowing, dissaving (reducing savings/assets), transfers (public/private), adjusting labor supply.
- Country-level patterns (Global Findex Database 2017 and survey data):
  - South Africa:
    - Bank account ownership: 65% (total).
    - Evidence of borrowing in last 12 months: 33% (total).
    - By education: up to 9th grade — 46% have account, 23% borrowed; secondary — 71% account, 33% borrowed; tertiary — 88% account, 53% borrowed.
    - By population group: Black African 63% account, 31% borrowed; Coloured 69% account, 40% borrowed; Whites 93% account, 58% borrowed.
    - By living area: traditional 54% account, rural 62% account, urban 75% account.
  - Tanzania:
    - Bank account ownership: 21% (total).
    - Evidence of borrowing in last 12 months: 13% (total).
    - Sources of borrowing indicated: formal sector 14%, informal 51%, microcredit 13%, SACCO 30% (total sample figures from Table).
    - By main income source:
      - Agricultural households: 7% have bank account, 9% evidence of borrowing, 5% formal sector borrowing, 65% informal borrowing, 9% microcredit, 26% SACCO.
      - Non-agricultural households: 24% have bank account, 18% borrowed, 17% formal sector borrowing, 47% informal, 14% microcredit, 31% SACCO.
- Mechanism prevalence and variation:
  - South African households more likely to use savings and formal banking; 30 percent receive private transfers from residents (personal networks important); account ownership is high partly because social grants are distributed via bank accounts but only 22 percent save at a financial institution (Global Findex Database 2017).
  - Tanzanian households more likely to adjust labor supply in response to shocks and draw less from formal savings; mobile money coverage is large relative to banking.
- Mobile money (Tanzania):
  - IMF (2016): by 2015 mobile money transactions amounted to 52 percent of GDP versus less than 1 percent in 2010.
  - Wave 4 LSMS evidence: mobile money use is positively correlated with income but access is higher than formal banking especially among low-income groups.
    - Example comparisons: in the lowest income percentile, 5 percent have a bank account while 20 percent use mobile money.
    - In the 75th–90th percentile, 20–30 percent have a bank account while 70 percent use mobile money.
  - Mobile money can reach the unbanked and agricultural/rural households, facilitating receipt of remittances in shock years and buffering consumption.
- Public and private transfers (South Africa):
  - Public transfers targeted to bottom two-thirds of the income distribution (conditionalities such as disability, children, elderly).
  - Private transfers less common but more evenly distributed across income.
  - For all income categories, urban and educated households tend to benefit slightly less from public transfers.

### Country-specific analysis of idiosyncratic shock mitigation
- South Africa: labor-market shocks (summary results)
  - Employment loss:
    - On average, employment loss reduces pre-transfer income by about 60 percent.
    - Government transfers halve the impact of the shock on total income.
    - Access to credit is associated with a reduction of two thirds in the negative impact of employment loss on consumption.
    - These effects hold across demographic groups (education, ethnicity, etc.).
- Tanzania: agricultural shocks (method and findings)
  - Agriculture context:
    - Agriculture comprised 32 percent of GDP in 2017 (NBS, 2018).
    - Almost three quarters of households report some personal agricultural production in each LSMS wave; approximately 40 percent rely on agriculture as their primary source of food and income.
  - Shock construction (Waves 1–3, panel):
    - Identify each household’s primary crop (main source of agricultural income).
    - Price shock: wave-to-wave log change in price in bottom quartile of pooled distribution for all crops.
    - Quantity shock: wave-to-wave log change in quantity in bottom quartile or full discontinuation (reported quantity 0 in latter wave).
    - Combined shock dummy: either a price or quantity fall.
  - Shock incidence and consumption impact:
    - About 50 percent of agricultural households in a given period face large drops in either price or quantity of their primary crop (Table 3).
    - Agricultural shocks are frequent for both agriculture-intensive and other household types but magnitude differs:
      - For agriculture-intensive households, agricultural shocks are associated with a significantly larger fall in consumption—approximately 10 percent.
    - Agricultural shocks predict a fall in food consumption and higher risk of poverty for all households regardless of primary income source.
  - Empirical approach:
    - Fixed-effects panel regression estimated: ln(consumption)it = β0 + β1·Shockit + γ′Xit + δi + ηt + εit, with household and wave fixed effects and household-level controls.
    - Specifications include separate price and quantity shocks and interactions with agricultural household dummy to capture differential effects.

*Source: LSMS, NIDS, and authors’ calculations, as presented in the supplied content.*

### Appendix B contains the details of the construction of these two variables, as well as an analysis of volatility in

### Appendix B contains the details of the construction of these two variables, as well as an analysis of volatility in agricultural value at the household level.

### Tanzania: Incidence of Agricultural Shocks Among Households
- Incidence by wave (fraction of households who experience either price or quantity shock):
  - All HH's:
    - Wave 2: 22% price shock, 30% quantity shock, 47% combined
    - Wave 3: 10% price shock, 31% quantity shock, 39% combined
  - non-Agri HH's:
    - Wave 2: 17% price shock, 26% quantity shock, 39% combined
    - Wave 3: 7% price shock, 28% quantity shock, 34% combined
  - Agri HH's:
    - Wave 2: 29% price shock, 36% quantity shock, 59% combined
    - Wave 3: 14% price shock, 35% quantity shock, 46% combined

### Tanzania: Estimated Impact on Consumption (summary of Table 4)
- Main aggregated findings:
  - Agricultural shock is associated on average with a fall of 4½ percent in total consumption (Column 1: Agricultural shock = -0.0449*).
  - Impact is almost three times larger for agricultural households (Agricultural shock*Agricultural Household = -0.116** in Log Consumption specification).
  - For agricultural households price shocks hurt mainly (Price shock*Agricultural Household = -0.164** in Log Consumption interactions).
  - Quantity shocks are often statistically insignificant for the aggregate but influence non-food consumption for agricultural households (Quantity shock*Agricultural Household = -0.187* in Non-Food Consumption).
- Selected coefficient estimates (as reported):
  - Log (Consumption):
    - Agricultural shock: -0.0449* (0.0248)
    - Agricultural shock*Agricultural Household: -0.116** (0.0585)
    - Quantity shock: -0.0579** (0.0267)
    - Price shock*Agricultural Household: -0.164** (0.0706)
    - Log non-agricultural income: 0.0634*** (0.0120)
    - R-squared: 0.250; Number of households: 2,626
  - Log (Food Consumption):
    - Agricultural shock: -0.0934*** (0.0280)
    - Agricultural shock*Agricultural Household: -0.0641 (0.0784)
    - Quantity shock: -0.0950*** (0.0309)
    - Price shock*Agricultural Household: -0.195** (0.0849)
    - Log non-agricultural income: 0.0276* (0.0142)
    - R-squared: 0.146; Number of households: 2,605
  - Log (Non-Food Consumption):
    - Agricultural shock: -0.0262 (0.0443)
    - Agricultural shock*Agricultural Household: -0.212** (0.100)
    - Quantity shock: -0.0624 (0.0457)
    - Quantity shock*Agricultural Household: -0.187* (0.108)
    - Log non-agricultural income: 0.105*** (0.0208)
    - R-squared: 0.227; Number of households: 2,626
- Notes:
  - Heteroskedasticity-robust standard errors in parentheses.
  - Other controls: education and age of head, household size, number of workers, wave fixed effects, constant term.
  - Significance: *** p<0.01, ** p<0.05, * p<0.1

### Tanzania: Volatility and Poverty (summary of Table 5 and associated analysis)
- Linear probability (OLS) and Logit estimates:
  - Agricultural shock increases incidence of poverty by 4 to 5 percentage points (Agricultural shock OLS = 0.0479***; Columns 1–4).
  - Interaction Agricultural shock*Agricultural Household: negative but not statistically significant (e.g., -0.0321 in OLS, not significant).
  - Logit coefficients corroborate OLS findings (Agricultural shock Logit = 0.464*** to 0.606*** across columns).
- Selected coefficient estimates (as reported):
  - OLS:
    - Agricultural Household: -0.0180 (0.0185)
    - Agricultural shock: 0.0479*** (0.0139)
    - Agricultural shock*Agricultural Household: -0.0321 (0.0341)
    - Log non-agricultural income: -0.0518*** (0.00589)
    - Poor in t-1 (when included): 0.227*** (0.0264)
    - Number of household-wave observations: 4,469; R-squared: 0.100 to 0.144
  - Logit:
    - Agricultural Household: coefficients range (e.g., -0.207 to 0.0655) with large standard errors
    - Agricultural shock: 0.464*** to 0.606*** (0.117 to 0.139)
    - Agricultural shock*Agricultural Household: -0.472* to -0.279 (large s.e.)
    - Log non-agricultural income: -0.504*** to -0.462*** (0.0529 to 0.0547)
    - Pseudo R-squared: 0.138 to 0.177
- Interpretation:
  - While shocks reduce consumption more for agricultural households (Table 4), the likelihood of crossing the poverty line due to shocks is similar across agricultural and non-agricultural households (interaction not significant). This implies some non-agricultural households are also vulnerable to agricultural volatility.
- Additional note:
  - Appendix B shows turnover of crops accounts for a large fraction of total income volatility.

### South Africa: Coping with Labor Shocks
- Context:
  - Labor income is the main source of pre-transfer earnings in South Africa (less than 10 percent of income comes from agriculture).
  - Employment statuses: formal employment, informal employment (including self-employment), unemployment.
  - Employment shock defined as transition out of formal employment or from informal employment to unemployment for the household member with the highest labor income.
- Transition patterns (Table 6 summary):
  - Table reports population shares and transition probabilities across employment statuses for geographic, education, and population groups (specific percentages provided in the source table).
  - Number of observations by groups reported (e.g., total observations: 5,590; rural: 3,489; urban: 4,229; other group breakdowns provided).
- Baseline difference-in-difference estimation (Table 7) — average percent gains/losses from employment shock:
  - Pre-transfer income: average % gain or loss from employment shock = -0.602*** (i.e., a 60 percent drop)
  - Government transfers: 0.577*** (increase of about 60 percent)
  - Private transfers: 0.570*** (increase about 60 percent)
  - Total after-transfer income: -0.332*** (33 percent drop)
  - Consumption: -0.172*** (17 percent drop)
  - Average % change without credit access (selected):
    - Pre-transfer income: -0.336***
    - Government transfers: -0.206***
    - Total/consumption: 0.091**
  - Average % change with credit access (selected):
    - Pre-transfer income: -0.320***
    - Government transfers: -0.074***
    - Total/consumption: 0.041**
  - Difference between changes with/without credit access:
    - For some specifications differences reported (e.g., 0.016; 0.132***; 0.050**)
  - Observations for specifications: 12,066; 12,314; 12,314; 12,314; 12,314; 12,314
  - R-square values reported: 0.321; 0.073; 0.062; 0.214; 0.180; 0.344
- Key findings and interpretation:
  - Employment losses are associated with a 60 percent drop in income before transfers.
  - Both private and public transfers increase by about 60 percent when households are hit by the shock, mitigating the pre-transfer income loss so after-transfer income falls by 33 percent on average.
  - Government transfers response explained by means-tested programs (Old-Age pension, Child Support grant) becoming accessible when households lose income.
  - Consumption falls by 17 percent on average after employment loss.
  - Household access to borrowing significantly mitigates consumption losses:
    - Consumption loss with credit access: 7 percent on average
    - Consumption loss without credit access: 21 percent on average
  - Poverty probability change:
    - Increase in probability of being poor following an employment shock without credit access: 9 percentage points
    - Increase with credit access: 4 percentage points
- Robustness and additional analyses:
  - Non-food consumption drops more (24 percent) than total consumption (17 percent).
  - Informal borrowing matters for mitigation; formal savings do not show a significant mitigating effect.
  - Triple difference-in-differences (Equation (4), Table 8) shows the mitigating effect of financial access does not differ significantly across education, population, or living area groups (Wald tests fail to reject equality), implying coping mechanisms have similar quantitative effects across groups.

### Conclusions (selected)
- Dynamic nature of poverty:
  - Non-poor households face on average a 13 percent probability of entering poverty across two survey periods due to a fall in consumption.
  - About half of poor households exit poverty in each period as consumption rises.
- Policy implications:
  - Some fiscal policies should target persistently poor households; other policies should focus on helping households just above the poverty line cope with adverse shocks.
- Country-specific summary:
  - South Africa: employment loss reduces pre-transfer income by about 60 percent; government transfers halve the impact on total income; access to credit reduces the negative impact of employment loss on consumption by two thirds.
  - Tanzania: agricultural shocks are frequent across household types; consumption fall is significantly larger (approximately 10 percent) for agriculture-intensive households.
- Access to coping mechanisms:
  - Use of banking services is highly correlated with income in both countries.
  - Tanzania: unequal access partially offset by spread of mobile money services among those without a traditional bank account.
  - South Africa: government transfers combined with relatively large financial coverage are effective risk mitigators for those with access.

*Source: LSMS, NIDS, and authors’ calculations as presented in the appendix material.*

### References

### References and Appendices (wpiea2020051-print-pdf)

### References
- Bibliographic list of cited works includes empirical studies and working papers on household demand, labor market dynamics, mobile money and risk sharing, poverty dynamics, energy use, food markets seasonality, and fiscal incidence. Authors and sources cited include Abdulai & Auber (2004); Anand, Kothari & Kumar (2016); Aron (2017); Atkinson & Lugo (2010); Carlson et al. (2015); Carroll (1997); Deaton & Muellbauer (1980); Flatø, Muttarak & Pelser (2017); Finn & Leibbrandt (2017); Hill & Mejía-Mantilla (2014); Hosier & Kipondya (1993); IMF (2016); Jack & Suri (2014); Kaminski, Christiaensen & Gilbert (2016); Lucas & Stokey (1987); Mason et al. (2015); Mkenda, Luvanda & Ruhinduka (2010); National Bureau of Statistics (2013, 2018); Riley (2018); Sarris & Karfakis (2006); Schotte, Zizzamia & Leibbrandt (2017); Schulhofer‑Wohl (2008); Statistics South Africa (2018); Tondini (2017); Tondini, Ardington & Woolard (2017); Weinberger & Swai (2006); World Bank (2014, 2015); Younger, Myamba & Mdadila (2016).

### South Africa: Data Details and Key Descriptive Statistics
- Survey waves and years: Wave 1: 2008; Wave 2: 2010/11; Wave 3: 2012; Wave 4: 2014/2015; Wave 5: 2017. Source: NIDS and authors’ calculations.
- Household demographics (means by wave):
  - Mean no. of household members: 4.02, 4.35, 4.34, 4.20, 4.15 (Waves 1–5).
  - Mean no. of working age adults: 1.56, 1.64, 1.56, 1.68, 1.70.
  - Mean no. of children: 1.44, 1.50, 1.47, 1.45, 1.42.
- Adult age composition (percent by age group across waves):
  - 15–29: 29%, 28%, 29%, 29%, 26%.
  - 30–39: 41%, 44%, 44%, 43%, 45%.
  - 40–64: 30%, 28%, 27%, 29%, 29%.
- Education of head (percent):
  - Up to grade 9: 46%, 43%, 37%, 30%, 26%.
  - Secondary: 35%, 37%, 41%, 43%, 44%.
  - Tertiary: 19%, 19%, 22%, 27%, 30%.
- Population group shares (percent): Black African ~82–83% across waves; Coloured ~13–14%; Asian/Indian ~1%; White ~3–4%.
- Area type shares (percent): Traditional 39%–52%; Urban 15%–21%; Rural 32%–39%.

### South Africa: Consumption and Income Summary (Table A.2)
- Consumption percentiles and mean (deflated to 2010 prices; all waves pooled):
  - Total consumption: 25th percentile 20,615; 50th percentile 33,185; 75th percentile 62,479; Mean 64,838; % of HHs with > 0 = 100%.
  - Food: 8,139; 12,144; 18,578; Mean 15,672; % with > 0 = 100%.
  - Health & Education: 2,650; 5,911; 16,243; Mean 16,160; % with > 0 = 32%.
  - Other non-food: 4,782; 10,510; 25,901; Mean 30,246; % with > 0 = 100%.
- Income percentiles and mean:
  - Total income: 27,963; 49,824; 96,849; Mean 92,665; % with > 0 = 100%.
  - Wages: 20,643; 45,541; 100,058; Mean 87,919; % with > 0 = 65%.
  - Government Grants: 8,744; 17,878; 27,292; Mean 20,684; % with > 0 = 68%.
  - Remittances: 6,072; 11,045; 18,766; Mean 16,935; % with > 0 = 17%.
  - Other: 3,401; 6,227; 13,733; Mean 20,923; % with > 0 = 90%.

### South Africa: Poverty Lines and Poverty Dynamics
- National poverty lines (STATSSA; April 2018 prices):
  - Food poverty line (extreme): R547 per person per month.
  - Lower-bound poverty line: R785 per person per month.
  - Upper-bound poverty line: R1,183 per person per month.
- Percent of population below the food poverty line by wave (Table A.3):
  - Total: 19% (2008), 26% (2010/11), 25% (2012), 19% (2014/15), 19% (2017).
  - By education (up to 9th grade): 28%, 37%, 35%, 29%, 31%.
  - By population group (Black African): 22%, 30%, 28%, 21%, 21%.
  - By living area (traditional): 29%, 38%, 39%, 31%, 30%.
- Transition frequencies (Table A.4 — average population shares in wave-pairs):
  - Poor in t and t-1: 53% overall; 58% for up to 9th grade; 50% for secondary; 33% for tertiary.
  - Poor in t, non-poor in t-1: 47% overall.
  - Non-poor in t, poor in t-1: 13% overall.
  - Non-poor in t and t-1: 87% overall.

### South Africa: Transfers and Employment Transitions (Selected Results)
- Transfers (share of beneficiaries; median amount as share of total income):
  - Government transfers: beneficiaries 78% / median 59% (example group); other reported combinations include 70%/49%, 83%/59% across subgroups (Table A.5).
  - Private transfers: beneficiaries 15% / median 26% (example).
- Employment transition impacts on income (Table A.6; coefficients reported for log changes in income per capita):
  - Transition from formal employment to self/informal: negative and significant effects (e.g., -0.190***, -0.226***, -0.389***, -0.466*** across specifications).
  - Transition from formal employment to no employment: large negative effects (e.g., -0.696***, -0.671***, -7.688***, -7.585*** — exact values preserved from table).
  - Transition from no employment to formal worker: positive and significant (e.g., 0.654***, 0.635***, 7.332***, 7.240***).
- Estimation notes: specifications include region-time fixed effects; standard errors clustered at region and household level; controls (when included) comprise changes in household composition and average education/age of working-age adults. Significance: *** p<0.01, ** p<0.05, * p<0.1.

### South Africa: Difference-in-Difference and Triple-Difference Results (Tables A.7–A.9)
- Baseline DiD (Table A.7) reports effects across dependent variables: Pre-transfer income, Government transfers, Remittances, Total income, Total consumption. Key reported bottom-line averages (exact values preserved):
  - average % gain or loss: -0.602***, 0.577***, 0.570***, -0.332***, -0.172***, 0.074*** (across columns).
  - average % change without credit access: -0.336***, -0.206***, 0.091**.
  - average % change with credit access: -0.320***, -0.073***, 0.041**.
  - difference between changes w./w.o credit access: 0.016, 0.132***, 0.050**.
- Triple-difference (Table A.8) and alternative DiD (Table A.9) provide heterogeneous impacts across education groups (1: up to 9th grade; 2: secondary; 3: tertiary), population groups (1: White; 2: African; 3: Coloured), and geographic groups (traditional, rural, urban). Reported group-level net gains/losses preserve values:
  - Example entries: A: group 1 % net gains (without credit) -0.256***; B: group 1 % net gains with credit access -0.106; C: group 2 % net gains (without credit) -0.290***; etc. (see table for full set of exact values).
- Financial access measures (Table A.9):
  - financial access (dummy) coefficients: 0.336*** (0.0321), 0.183*** (0.0391), 0.338*** (0.0460) across specifications.
  - lag financial access: -0.429*** (0.0348), -0.303*** (0.0615), -0.296*** (0.0387).
  - new formal borrower (dummy): 0.329*** (0.0419); interactions with employment loss and borrower status are reported.

### Tanzania: Data Details and Key Descriptive Statistics
- Survey waves and years: Wave 1: 2008/09; Wave 2: 2010/11; Wave 3: 2012/2013; Wave 4: 2014/2015. Source: LSMS and authors’ calculations.
- Household demographics (means by wave):
  - Mean no. of household members: 5.13, 5.30, 5.25, 4.74.
  - Mean no. of working age adults: 2.97, 3.07, 3.07, 2.70.
  - Mean no. of children: 2.16, 2.22, 2.18, 2.04.
- Adult age composition (percent by age group):
  - 15–29: 36%, 33%, 28%, 37%.
  - 30–39: 44%, 47%, 49%, 42%.
  - 40–64: 19%, 20%, 23%, 21%.
- Education of head (percent):
  - Below primary: 42%, 42%, 42%, 40%.
  - Primary: 48%, 48%, 48%, 45%.
  - Secondary or above: 10%, 10%, 11%, 15%.
- Main income source (percent): Agricultural 40%, 40%, 40%, 30%; Non-agricultural 60%, 60%, 60%, 70%.

### Tanzania: Consumption and Income Summary (Table B.2)
- Consumption percentiles and mean (deflated to 2010 prices; all waves pooled):
  - Total consumption: 25th percentile 1,260; 50th percentile 2,034; 75th percentile 3,328; Mean 2,842; % of HHs with > 0 = 100%.
  - Food: 813; 1,263; 1,868; Mean 1,564; % with > 0 = 98%.
  - Health & Education: 33; 87; 242; Mean 273; % with > 0 = 87%.
  - Other non-food: 193; 477; 1,258; Mean 1,053; % with > 0 = 100%.
- Income percentiles and mean:
  - Total income: 498; 1,304; 2,932; Mean 3,211; % with > 0 = 100%.
  - Wages: 702; 1,353; 2,787; Mean 3,617; % with > 0 = 51%.
  - Agriculture: 107; 253; 590; Mean 723; % with > 0 = 70%.
  - Self-Employment: 321; 798; 1,927; Mean 1,934; % with > 0 = 47%.

### Tanzania: Poverty Definition and Dynamics
- Poverty line used: 847 Tsh. per adult equivalent of weekly consumption in 2010 prices (Tanzania LSMS Wave 2 Final Report).
- Percent of poor households by wave (Table B.3):
  - All: 12% (2008/09), 16% (2010/11), 18% (2012/2013), 15% (2014/15).
  - Non-Agricultural: 11%, 12%, 14%, 13%.
  - Agricultural: 14%, 22%, 24%, 19%.
- Transition probabilities (Table B.4): reported as fractions by t-1 state; example averages across wave-pairs:
  - poor in t and t-1: 42% overall; agricultural 43%; non-agricultural 41%.
  - poor in t, non-poor in t-1: 58% overall.
  - non-poor in t, poor in t-1: 13% overall.
  - non-poor in t and t-1: 87% overall.

### Tanzania: Agricultural Income Computation and Volatility
- Agricultural income: constructed by summing reported income from self-employment/household enterprises, wage labor, agricultural sales, and imputed value of non-sold agricultural production (including byproducts as detailed).
- Definition of "agricultural" household: households where agricultural production constitutes the largest average share of income across three waves. Approximately 40% of households in Waves 1–3 classify as agricultural; Wave 4 share is 10 percentage points lower when coded using Wave 4 only.
- Decomposition of changes in agricultural harvest value across waves:
  - Total change ΔY is decomposed into contributions from:
    - Change in quantities within constant basket.
    - Change in prices within constant basket.
    - Interaction of price and quantity changes.
    - Contributions from newly grown crops and discontinued crops.
  - Empirical finding: price changes account on average for a small fraction of agricultural income volatility; quantities account for the bulk of volatility (both within-basket quantity changes and crop turnover) (Figure B.2).
- Price and quantity shocks coding:
  - Price shock: household-level change in price below the 25th percentile of the distribution of all log changes in price (pooling waves).
  - Quantity shock: household-level change in quantity below the 25th percentile of log changes in quantity or discontinuation of primary good in the second wave.
- Main crop contribution: the primary crop accounts on average for 70 percent of total agricultural production value at the household level (Figure B.4).
- Nonparametric evidence (Figure B.3): for most crops (except banana, mango, orange) volatility of quantities is much larger than volatility of prices.

### Cross-Country Summary
- Survey-based headline poverty rates (Table C.1):
  - South Africa (percent population below country-specific consumption poverty line): 19% (2008), 26% (2010/11), 25% (2012), 19% (2014/15), 19% (2017).
  - Tanzania: 12% (2008/09), 16% (2010/11), 18% (2012/13), 15% (2014/15).

*Source: wpiea2020051-print-pdf — References and Appendices (NIDS, LSMS and authors’ calculations).*

---


_Source: https://www.imf.org/-/media/files/publications/wp/2020/english/wpiea2020051-print-pdf.pdf_