## Monetary Policy, Inflation, and Distributional Impact: South Africa’s Case

## Source details

**Canonical URL:** [Monetary Policy, Inflation, and Distributional Impact: South Africa’s Case](https://www.imf.org/-/media/files/publications/wp/2021/english/wpiea2021078-print-pdf.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/wp/2021/english/wpiea2021078-print-pdf.pdf.md)
- [Structured JSON version](/-/media/files/publications/wp/2021/english/wpiea2021078-print-pdf.pdf.json)

---

### Introduction and research question
- South African Reserve Bank mandate: protect the value of the rand by keeping inflation low and steady.
- Research question: Could monetary policy tightening aimed at maintaining low and stable inflation also have distributional effects in South Africa?
- Empirical strategy:
  - Regress real per-capita household consumption on an estimated “exogenous” monetary policy shock (net of anticipated components), controlling for macro factors and individual characteristics.
  - Microdata: National Income Dynamics Study (NIDS) panel covering 2008–17 and five survey waves.
- Horizon of interest: 12–18 month horizon (transmission lag of monetary policy to the real economy and similar to distance between survey waves).

### Key descriptive findings on inflation and households
- Long-term trend:
  - Inflation fluctuated in a 10–20 percent range until the early-1990s, then declined with single-digit outturns by the end of 1992.
  - Average inflation during 1993–99 was around 8 percent.
  - Following formal inflation targeting (inception announced August 1999), inflation became more stable since 2010 and moderated toward the midpoint of the 3–6 percent target range in the late-2010s.
- Technical note:
  - Prime lending rate is generally 350 basis points above the policy rate.
- Distributional exposure (sample period 2008–17):
  - Average inflation: lowest two consumption deciles = 6.1 percent; highest two consumption deciles = 5.4 percent.
  - Volatility (standard deviation and coefficient of variation) for the lowest two deciles is about twice that faced by the highest two deciles.
  - Chance of facing the highest inflation: 56–57 percent for the lowest two deciles.
  - Food prices share of CPI basket: around 45 percent for the lowest two deciles; 15–25 percent for the highest two deciles.
- COVID-19 (2020) note:
  - Individuals in lower consumption deciles, most affected by the pandemic, generally faced relatively high inflation; those in the highest deciles also faced relatively high inflation in level and volatility.

### Channels of monetary policy transmission relevant for distribution
- Inflation channel:
  - Poorer individuals more negatively affected by higher inflation due to larger cash shares and less ability to hedge.
  - Lower inflation increases the real value of nominal transfers and pensions that are not fully indexed.
- Labor income channel:
  - Employment rates: lowest two consumption deciles = 28–34 percent; highest two consumption deciles = 54–65 percent. Labor income is a weaker channel for lower-consumption individuals.
- Transfers and fiscal support:
  - Social grants are important and relatively stable nominal income for the poor; they gain in real value as inflation declines and are relatively insulated from business cycles.
  - More than 80 percent of those in the lowest 4 consumption deciles receive social grants.
  - 2018 General Household Survey context: about 44 percent of households receive at least one kind of grant nationally; grants were the main source of income for almost 20 percent of households nationally.
- Asset and debt channels:
  - Assets disproportionately owned by higher-consumption individuals; low interest rates and higher asset prices could increase inequality.
  - Higher-consumption individuals borrow in far greater volumes, so borrowing-cost changes affect them more strongly.
- Structural limits:
  - High structural unemployment limits scope for monetary policy to generate broad-based employment gains.

### Identification and estimation of exogenous monetary policy shock
- Definition:
  - Forecast error of policy rate: FEF_t^i = AAt^i − FEEt^i.
  - Exogenous monetary policy shocks (MPS) are residuals ε_t from: FEF_t^i = α + β FEF_t^π + γ FEF_t^y + ε_t.
- Data sources:
  - Inflation and GDP growth from October WEO vintages.
  - Policy rate expectations from October vintages of Consensus Economics data.
- Characteristics of estimated exogenous MPS (South Africa):
  - Average around zero with standard deviation of 0.34 for 2000–19.
  - For NIDS wave years: average zero and standard deviation 0.25.
  - No autocorrelation detected.
- Comparable observed policy rate statistics:
  - Averages of –41 and –27 basis points for 2000–19 and 2008–17, respectively.
  - Standard deviation of 1.6 and 1.5 for 2000–19 and 2008–17, respectively.

### Macroeconomic impacts: inflation and GDP
- Time-series/panel results:
  - Positive exogenous monetary policy shock (tightening) associated with a reduction in inflation and muted relationship with real GDP growth.
  - Quantitative example: a one standard deviation exogenous MPS tightening is associated with around 1 percentage point reduction (= –3.395 * 0.34) in the rate of inflation.
  - Observed increase in detrended policy rate associated with a rise in inflation (the “inflation puzzle”).
- Selected coefficients (examples reported):
  - Exogenous monetary policy shock: β = -3.780***, -3.395*** (std. error 0.993, 0.995).
  - Policy rate, detrended: β = 1.127*** (std. error 0.239).
  - Real GDP growth (L1): coefficients include 0.447** (std. error 0.190), 0.487** (std. error 0.185), 0.461** (std. error 0.194), 0.504** (std. error 0.195).
  - Sample size for these regressions: N = 20. Adjusted R^2 reported examples: 0.46, 0.50, 0.64, 0.20, 0.38, 0.40.

### Household-level consumption econometric specification and data
- Dependent variable:
  - Per-capita household real consumption in log (household consumption divided by number of adults, age-adjusted, divided by annual inflation index of survey year).
  - Age adjustment uses U.S. Consumer Expenditure Survey average consumption by age group (average for 2008-17, normalized such that 45-54 = 100).
- Specification:
  - Panel fixed effects without lagged dependent variable in baseline (lagged dependent variable found statistically insignificant; lagged results reported in Table A1).
  - Baseline regression (Equation (3)) includes:
    - Macro controls X_j: log real GDP level and mps.
    - Micro controls Y_k,i: grant receipt dummy, employment dummy, educational attainment dummies, food-to-total consumption ratio.
    - Interaction terms Z_l,h: mps interactions with food ratio, employment, grant dummy, consumption decile dummies.
    - Individual fixed effects α_i and constant c.
  - Extreme values winsorized at the 3rd and 97th percentiles.
- NIDS and sample construction:
  - NIDS started 2008 with over 28,000 individuals in 7,300 households.
  - Waves: 2008, 2010–11, 2012, 2014–15, 2017.
  - Sample restricted to adults surveyed across the 5 waves: about 6,700 adults.
- Education categories:
  - No education (including “other” and “don’t know”), lower primary (grades 1–7), upper primary (grades 8–9), secondary (grades 10–12, National Technical Certificate, National Vocational Certificate), tertiary (above secondary).
- Other variables:
  - Food consumption share, employment dummy (1 if employed), grant recipient dummy (1 if recipient), geography dummies (rural informal, rural formal, urban informal, urban formal); real GDP (log) interacted with geography dummies.

### Main empirical results (distributional effects and heterogeneity)
- Baseline model key patterns:
  - mps coefficient statistically insignificant when introduced alone.
  - Interaction with food-to-total consumption ratio:
    - Food share = 0.45 (lower-end deciles): total effect of mps = -0.001 (approximately zero).
    - Food share = 0.2 (higher-end deciles): total effect of mps = -0.124**.
    - Interpretation: one unit (one standard deviation) mps increase leads to a 12 percent (4 percent) reduction in per-capita household real consumption (Table 6, model 3 text reports both 12 percent and 4 percent linked to different standardizations).
  - Interaction with employment dummy:
    - Total effect for employed = -0.108* (model 4): one unit (one standard deviation) mps increase leads to an 11 percent (3.5 percent) reduction in per-capita household real consumption.
    - mps affects consumption of the unemployed very little.
  - Interaction with grant dummy:
    - Interaction term mps * grant dummy not statistically significant; total effect for grant recipient dummy = -0.041 (not significant).
- Distributional total effects by consumption decile (memo-item totals, baseline model 6):
  - Consumption decile 1: 0.409***
  - Consumption decile 2: 0.382***
  - Consumption decile 3: 0.284***
  - Consumption decile 4: 0.184*
  - Consumption decile 5: -0.007
  - Consumption decile 6: -0.081
  - Consumption decile 7: -0.205**
  - Consumption decile 8: -0.191*
  - Consumption decile 9: -0.121
  - Consumption decile 10: -0.445***
- Interpreted magnitudes (model discussion):
  - One unit (standard deviation) increase in mps leads to a 40 percent (13 percent) increase in consumption for the first two deciles (text reports both 40 percent and 13 percent figures).
  - Overall effects moderate to 0.3 for decile 3, 0.2 for decile 4, zero for deciles 5–6, negative –0.2 for deciles 7–8, and –0.45 for decile 10.
  - Effect for decile 9 not statistically different from zero.
- Table 6 selected coefficients (L1 examples):
  - Food share: 0.256*** across columns.
  - Employment dummy: 0.049*** across columns.
  - Grant recipient dummy: around -0.022 (not significant).
  - Interaction terms (mps * ...): mps * Food share = 0.491***; mps * Employment dummy = -0.122**; mps * Grant dummy = -0.004.
  - mps * consumption deciles 1–9: positive and statistically significant coefficients decreasing with higher deciles (examples: decile1 = 0.854***; decile9 = 0.324***).

### Extensions and additional channels
- Extension 1 (grant-to-consumption ratio, bank accounts, credit cards):
  - Grant-to-consumption ratio included; bank account and credit card ownership dummies introduced.
  - Selected interactions and memo-item totals:
    - mps * grant to consumption ratio: 0.088*.
    - mps * credit card: -0.420*** (interaction); total effect for credit card owners: -0.443*** (memo).
    - Bank account interactions weak/insignificant; memo-item total for bank account: -0.087, -0.103* (selected columns).
    - Consumption decile totals remain positive for low deciles and strongly negative for top deciles (decile 10: -0.443***, -0.458***).
  - Interpretation:
    - Total effect of mps is weaker for individuals more reliant on social grants (e.g., total effect = –0.11 for grant-to-consumption ratio = 0.25; estimate = 0 for ratio = 0.9).
    - Credit card ownership amplifies negative mps effects (consistent with debt-service channel); bank account ownership not an important transmission channel in these estimates.
- Extension 2 (observed real policy rate rpol):
  - rpol defined as policy rate deflated by one year ahead inflation expectations and detrended by first differences.
  - Results somewhat weaker overall compared with mps-based models.
  - Memo-item total effects of rpol (selected):
    - Food share = 0.45: 0.028
    - Food share = 0.2: -0.017
    - Consumption decile 1: 0.34***
    - Consumption decile 5: -0.093*
    - Consumption decile 10: -0.195***
  - Interpreted magnitudes:
    - Average and standard deviation of rpol is –0.3 and 1.0 for 2008–17; a one standard deviation increase in rpol leads to a 34 percent increase in per-capita household real consumption for the first decile (text compares to 13 percent from mps one standard deviation).

### Robustness, limitations, and diagnostics
- Baseline diagnostics:
  - Sample: # of Obs. = 26,876; Cross section (N) = 6,719; Time series (T) = 4.
  - Adjusted R^2: around 0.115–0.126 across baseline specifications.
- Lagged dependent variable (Table A1 / Appendix):
  - Inclusion often yields lag coefficient L1 around –0.055 to –0.07 (models 16–21).
  - Some mps L1 estimates change sign/magnitude when lag included (example L1 mps: -0.236*** in one column; -0.457*** in another).
  - Memo-item Adjusted R^2 with lag: 0.115–0.127.
- Inference caveats:
  - Results suggestive; inclusion of lagged dependent variable and significance at mostly 10 percent level in some specifications indicate potential omitted variable or Nickell bias.
  - Additional waves of NIDS would help solidify findings.

### Policy-relevant implications and discussion
- Main policy implication:
  - Maintaining low and stable inflation may have favorable distributional effects in South Africa by protecting the real consumption of poorer households, primarily via the inflation channel and through insulation by social grants.
- Heterogeneous transmission:
  - Monetary policy transmission is heterogeneous across consumption groups; policymakers should consider distributional incidence alongside macro objectives.
- Complementary fiscal and structural measures:
  - Complementary fiscal measures (e.g., social transfers) enhance the protective effect of low inflation on lower-income households and interact with monetary policy to influence distributional outcomes.
  - Welfare-improving policies should address private investment, job creation, inclusive growth, and educational attainment to improve individual-level consumption outcomes.
- SARB context:
  - SARB reduced the policy rate by 300 basis points in 2020 as inflation fell toward the lower end of the official target band to cope with COVID-19 effects.

*Source: IMF Working Paper WP/21/78, “Monetary Policy, Inflation, and Distributional Impact: South Africa’s Case” (Ken Miyajima), March 2021, Sections 1–4 (wpiea2021078-print-pdf).*

### Section 1

### Monetary Policy, Inflation, and Distributional Impact: South Africa’s Case

### Introduction and research question
- The South African Reserve Bank has pursued its constitutional mandate to protect the value of the rand by keeping inflation low and steady.
- Research question: Could monetary policy tightening aimed at maintaining low and stable inflation also have distributional effects in South Africa?
- Empirical strategy: Regress real per-capita household consumption on an estimated “exogenous” monetary policy shock (net of anticipated components), controlling for macro factors and individual characteristics, using the National Income Dynamics Study (NIDS) panel covering 2008–17 and five survey waves.
- Horizon of interest: 12–18 month horizon, commonly understood as the transmission lag of monetary policy action to the real economy and similar to the distance between survey waves used in the analysis.

### Key descriptive findings on inflation and households
- Long-term trend:
  - Inflation fluctuated in a 10–20 percent range until the early-1990s, then declined with single-digit outturns by the end of 1992.
  - Average inflation during 1993–99 was around 8 percent.
  - Following formal inflation targeting (inception announced August 1999), inflation became more stable since 2010 and moderated toward the midpoint of the 3–6 percent target range in the late-2010s.
- Short technical note:
  - Prime lending rate is generally 350 basis points above the policy rate.
- Distributional exposure to inflation (sample period 2008–17):
  - Average inflation for the lowest two consumption deciles: 6.1 percent.
  - Average inflation for the highest two consumption deciles: 5.4 percent.
  - Volatility of inflation (standard deviation and coefficient of variation) for the lowest two deciles is about twice that faced by the highest two deciles.
  - The chance of facing the highest inflation is highest for the lowest two consumption deciles (56–57 percent).
  - Food prices account for around 45 percent of the CPI basket for the lowest two deciles, versus 15–25 percent for the highest two deciles.
- COVID-19 note (2020): Individuals in lower consumption deciles, most affected by the pandemic, generally faced relatively high inflation; those in the highest deciles also faced relatively high inflation in level and volatility.

### Channels of monetary policy transmission relevant for distribution
- Inflation channel:
  - Poorer individuals are more negatively affected by higher inflation due to larger cash shares and less ability to hedge.
  - Lower inflation increases the real value of nominal transfers and pensions that are not fully indexed.
- Labor income channel:
  - Employment rates suggest labor income is a weaker channel for lower-consumption individuals.
  - Employment rates: lowest two consumption deciles: 28–34 percent; highest two consumption deciles: 54–65 percent.
- Transfers and fiscal support:
  - Social grants are an important and relatively stable source of nominal income for the poor; they gain in real value as inflation declines and are relatively insulated from business cycles.
  - More than 80 percent of those in the lowest 4 consumption deciles receive social grants.
  - From the 2018 General Household Survey context (reported): about 44 percent of households receive at least one kind of grant nationally; grants were the main source of income for almost 20 percent of households nationally.
- Asset and debt channels:
  - Assets are disproportionately owned by higher-consumption individuals, so low interest rates and higher asset prices could increase inequality.
  - Higher-consumption individuals borrow in far greater volumes, so changes in borrowing costs affect them more strongly.
- Structural limits:
  - An important share of high unemployment is structural, limiting the scope for monetary policy to generate broad-based employment gains.

### Main empirical results (summary)
- In response to exogenous monetary policy tightening:
  - Real consumption of individuals at lower ends of the consumption distribution declines relatively modestly, or even increases.
    - Mechanisms: greater reliance on government transfers (less reliance on labor income) and relatively larger food consumption imply these individuals benefit mainly from lower inflation.
  - Real consumption of individuals at higher ends of the consumption distribution is more likely to decline.
    - Mechanisms: impact through lower labor income, weaker asset price performance, and higher debt service cost.
- Interpretation: Monetary policy tightening that maintains low and stable inflation can, through the inflation and transfer channels, reduce consumption inequality over the 12–18 month horizon.

### Methodological notes
- Identification:
  - The analysis isolates an “exogenous” monetary policy shock by netting out the anticipated component due to macroeconomic conditions that could influence both monetary policy action and consumption.
- Data:
  - Primary microdata source: National Income Dynamics Study (NIDS), the first national panel study in South Africa, tracking individuals across five waves during 2008–17.
  - Decile-level inflation index data start from January 2008 (thus inflation rates from January 2009 in the tabulations).

### Policy-relevant implications
- Maintaining low and stable inflation may have favorable distributional effects in South Africa by protecting the real consumption of poorer households, primarily via the inflation channel and through the relative insulation provided by social grants.
- Monetary policy transmission is heterogeneous across consumption groups; policymakers should consider distributional incidence alongside macro objectives.
- Complementary fiscal measures (e.g., social transfers) enhance the protective effect of low inflation on lower-income households and can interact with monetary policy to influence distributional outcomes.

*Source: IMF Working Paper WP/21/78, “Monetary Policy, Inflation, and Distributional Impact: South Africa’s Case” (Ken Miyajima), March 2021, Section 1.*

### Section 2

### Section 2

### Identifying the exogenous monetary policy shock: definition and estimation
- Unexpected changes in the policy rate i in year t are proxied by the forecast error FEF_t^i, defined as the difference between the policy rate observed at the end of year t (AAt^i) and the policy rate expected during the course of the year (FEEt^i):
  - FEF_t^i = AAt^i − FEEt^i. (Equation (1))
- Forecast errors of headline inflation (FEF_t^π) and GDP growth (FEF_t^y) are computed similarly.
- Exogenous monetary policy shocks (MPS) are the residuals ε_t from the regression:
  - FEF_t^i = α + β FEF_t^π + γ FEF_t^y + ε_t. (Equation (2))
- Inflation and GDP growth data are taken from different vintages of the October WEO. Policy rate expectations are taken from the October vintages of Consensus Economics data.

### Characteristics of the estimated exogenous monetary policy shock (South Africa)
- The estimated exogenous monetary policy shock:
  - averages around zero with standard deviation of 0.34 for 2000–19.
  - for the NIDS wave years (shaded areas in Figure 2) the statistics are similar: average zero and standard deviation 0.25.
- Comparable statistics for the observed policy rate:
  - averages of –41 and –27 basis points for 2000–19 and 2008–17, respectively.
  - standard deviation of 1.6 and 1.5 for the same periods.
- The estimates do not display autocorrelation.

### Macroeconomic impacts: inflation and GDP
- Time-series / panel econometric results (Table 3) indicate:
  - A positive exogenous monetary policy shock (tightening) is associated with a reduction in inflation and has a muted relationship with real GDP growth.
  - Quantitatively: a one standard deviation exogenous monetary policy shock (tightening) is associated with around 1 percentage point reduction (= –3.395 * 0.34) in the rate of inflation.
  - By contrast, an observed increase in the detrended policy rate is associated with a rise in inflation (the “inflation puzzle”).
- Selected estimated coefficients reported in Table 3 (as presented):
  - Exogenous monetary policy shock: β = -3.780***, -3.395*** (std. error 0.993, 0.995 shown).
  - Policy rate, detrended: β = 1.127*** (std. error 0.239).
  - Inflation (L1) γ coefficients reported (examples): 0.258 (std. error 0.174); -0.483** (std. error 0.203); -0.420** (std. error 0.181); -0.344 (std. error 0.202).
  - Real GDP growth (L1) δ coefficients reported (examples): 0.447** (std. error 0.190); 0.487** (std. error 0.185); 0.258 (std. error 0.162); 0.461** (std. error 0.194); 0.504** (std. error 0.195).
  - Constant α examples: 4.503*** (std. error 0.616); 2.920** (std. error 1.221); 5.001*** (std. error 1.192); 5.369*** (std. error 1.219); 3.747*** (std. error 1.277); 3.187** (std. error 1.437).
  - Sample size: N = 20 (for reported regressions).
  - Adjusted R^2 examples: 0.46, 0.50, 0.64, 0.20, 0.38, 0.40.

### Methodology for household-level consumption analysis
- Dependent variable:
  - Per-capita household real consumption in log (computed as household-level consumption divided by number of adults, adjusted for age, then divided by the annual inflation index of the survey year).
  - Age adjustment uses U.S. Consumer Expenditure Survey average consumption by age group (average for 2008-17, normalized such that 45-54 = 100).
- Econometric specification:
  - Panel fixed effects approach without the lagged dependent variable (lagged dependent variable found statistically insignificant; results with lag reported in Table A1).
  - Baseline regression (Equation (3)) regresses log per-capita real consumption pcap_i,t on:
    - Vector of macro controls X_j: log real GDP level and the exogenous monetary policy shock (mps).
    - Vector of micro controls Y_k,i: dummy variables for grant receipt status, employment status, educational attainment, food-to-total consumption ratio.
    - Interaction terms Z_l,h: interactions between mps and micro-level controls (food ratio, employment dummy, grant dummy, consumption decile dummies).
    - Individual fixed effects α_i, constant c, and error ε_i,t.
  - Extreme values are winsorized at the 3rd and 97th percentiles.

### Data: NIDS and sample construction
- National Income Dynamics Study (NIDS) details:
  - Started in 2008 with a nationally representative sample of over 28,000 individuals in 7,300 households.
  - Survey waves available: 2008, 2010–11, 2012, 2014–15, and 2017.
  - Sample restricted to adults successfully surveyed across the 5 waves: about 6,700 adults.
- Educational attainment categories used:
  - No education (including “other” and “don’t know”), lower primary (grades 1–7), upper primary (grades 8–9), secondary (grades 10–12, National Technical Certificate, National Vocational Certificate), tertiary (above secondary).
- Other individual-level variables:
  - Food consumption measured as share of total consumption.
  - Employment dummy = 1 if “employed”, 0 otherwise.
  - Grant recipient dummy = 1 if recipient, 0 otherwise.
  - Geography dummies for 4 types: rural informal, rural formal, urban informal, urban formal; real GDP (log) interacted with geography dummies.

### Distributional and micro-level findings: channels and heterogeneity
- Baseline model results (Table 6 and Figure 4) — key findings:
  - The coefficient on mps is statistically insignificant when introduced alone (Table 6, model 2).
  - Interaction with food-to-total consumption ratio:
    - For high food consumption ratio (e.g., 0.45, corresponding to lower-end consumption deciles in wave 5), the overall effect of mps is very low (impact ~ zero).
    - For low food consumption ratio (e.g., 0.2, corresponding to higher-end consumption deciles in wave 5), the overall impact of exogenous monetary policy tightening is –0.12.
    - Interpretation provided: a one unit (one standard deviation) mps increase leads to a 12 percent (4 percent) reduction in per-capita household real consumption (Table 6, model 3).
  - Interaction with employment dummy:
    - The overall effect of exogenous monetary policy tightening for the employed is –0.11.
    - Interpretation provided: a one unit (one standard deviation) mps increase leads to an 11 percent (3.5 percent) reduction in per-capita household real consumption (Figure 4; Table 6, model 4).
    - mps affects the consumption of the unemployed very little.
  - Interaction with grant dummy:
    - The interaction term between mps and the grant dummy does not generate statistically significant effects (Figure 4; model 5 in Table 6).
- Contextual notes on households’ financial positions:
  - Lower consumption deciles consume less and tend to own less assets, debt, and net worth.
  - NIDS data suggest those in the lowest consumption decile are as indebted relative to income as those in the highest consumption decile; borrowing from lenders charging very high interest rates implies small monetary policy moves (25 or 50 basis points) would have subdued effects on their debt service costs.

### Table 6 (selected reported estimates and notes)
- Dependent variable: per-capita household real consumption level in log.
- Macro control examples:
  - Real GDP L0 coefficients across models: 1.590***, 1.921***, 1.941***, 1.891***, 1.925***, 1.727***.
- Exogenous monetary policy shock (mps) reported in models:
  - L1 mps entries include: -0.039, -0.222***, 0.015, -0.036, -0.445*** (showing heterogeneity across specifications).
- Education dummy (examples, L1 coefficients):
  - No education: -0.134* (and similar values across models).
  - Lower primary: -0.275***.
  - Upper primary: -0.224***.
  - Secondary: -0.088***.
- Note: Table 6 memo item reports total effects of mps calculated using lincom. ***, **, and * denote statistical significance at 1, 5, and 10 percent levels.

*Italic: Source — wpiea2021078-print-pdf - Section 2*

### Section 3

### wpiea2021078-print-pdf - Section 3

### Major empirical estimates (baseline models and interactions)
- Key estimated coefficients (L1):
  - Food share of total cons.: 0.256***, 0.259***, 0.268***, 0.259***, 0.259***, 0.257***
  - Employment dummy: 0.049***, 0.048***, 0.049***, 0.043***, 0.048***, 0.047***
  - Grant recipient dummy: -0.022, -0.022, -0.022, -0.022, -0.023, -0.023
- Interaction terms with exogenous monetary policy shock (mps) (L1):
  - mps * Food share: 0.491***
  - mps * Employment dummy: -0.122**
  - mps * Grant recipient dummy: -0.004
  - mps * consumption decile 1: 0.854***
  - mps * consumption decile 2: 0.827***
  - mps * consumption decile 3: 0.729***
  - mps * consumption decile 4: 0.629***
  - mps * consumption decile 5: 0.438***
  - mps * consumption decile 6: 0.364***
  - mps * consumption decile 7: 0.239*
  - mps * consumption decile 8: 0.253**
  - mps * consumption decile 9: 0.324***
- Sample and model diagnostics (summary from baseline table):
  - # of Obs.: 26,876 (columns reported)
  - Cross section (N): 6,719
  - Time series (T): 4
  - Adjusted R^2: 0.115, 0.116, 0.126 (reported across columns)

### Distributional “total effects” of an exogenous monetary policy tightening (mps)
- Memo-item total effects (mps + interaction) reported:
  - Food share = 0.45: -0.001
  - Food share = 0.2: -0.124**
  - Employment dummy: -0.108*
  - Grant recipient dummy: -0.041
  - Consumption decile 1: 0.409***
  - Consumption decile 2: 0.382***
  - Consumption decile 3: 0.284***
  - Consumption decile 4: 0.184*
  - Consumption decile 5: -0.007
  - Consumption decile 6: -0.081
  - Consumption decile 7: -0.205**
  - Consumption decile 8: -0.191*
  - Consumption decile 9: -0.121
  - Consumption decile 10: -0.445***

- Interpreted magnitudes from model discussion (model 6 results):
  - One unit (standard deviation) increase in mps leads to a 40 percent (13 percent) increase in consumption for the first two deciles (text distinguishes 40 percent and 13 percent linked to different standardization; both figures reported in text).
  - Overall effect moderates to 0.3 for the third decile, 0.2 for the fourth decile, and zero for the 5–6 deciles.
  - Overall effect turns negative: –0.2 for the 7–8th deciles and –0.45 for the 10th decile.
  - The overall effect for the 9th decile is not statistically different from zero.

### Other important control-variable effects (baseline)
- Macro and micro controls (text summary and table entries):
  - A one percent increase in real GDP leads to a 1.5–2 percent increase in per-capita household real consumption.
  - Educational attainment: those with primary education tend to consume around 20 percent less than those with tertiary education; those with secondary education tend to consume around 10 percent less than those with tertiary education.
  - No education dummy: coefficient not statistically significant.
  - Employment dummy: employed consume around 5 percent more than the unemployed (coefficient positive but small).
  - Grant dummy: not statistically significant in baseline models.

### Extension 1 — Additional channels (grant-to-consumption ratio, bank accounts, credit cards)
- Key model changes:
  - Grant recipient dummy replaced by grant-to-consumption ratio.
  - Bank account and credit card ownership dummies introduced.
  - Consumption decile dummies re-specified (2nd through 10th deciles in alternative specification).
- Selected coefficients and interactions (from Table 7, L1):
  - Real GDP (log) coefficients: 2.668***, 2.823***, 1.841***, 1.923***, 1.932***, 2.018***, 1.750***, 1.832***
  - mps (L1) reported values across models: -0.146**, -0.177***, 0.000, -0.012, -0.02, -0.033, 0.389***, 0.371***
  - Food share of total consumption ratio: 0.166***, 0.149***, 0.253***, 0.245***, 0.251***, 0.243***, 0.250***, 0.243***
  - Employment dummy: 0.046**, 0.049**, 0.042***, 0.043***, 0.049***, 0.050***, 0.048***, 0.049***
  - Grant to consumption ratio: 0.032*, 0.027
  - Bank account (L1): 0.039**, 0.040**
  - Credit card (L1): -0.031, -0.030
  - Interactions:
    - mps * grant to consumption ratio: 0.088*, 0.088*
    - mps * bank account: -0.088, -0.092
    - mps * credit card: -0.420***, -0.432***
    - mps * consumption decile 5–10 (selected): decile 5: -0.399***, decile 6: -0.471***, decile 7: -0.591***, decile 8: -0.579***, decile 9: -0.515***, decile 10: -0.831*** (two columns reported with near-identical values)

- Memo-item total effects (selected) from Extension 1:
  - Grant to consumption = 0.9: -0.067, -0.098**
  - Grant to consumption = 0.25: -0.111**, -0.155***
  - Bank account: -0.087, -0.103*
  - Credit card: -0.443***, -0.469***
  - Consumption decile totals (selected): decile 1: 0.389***, 0.371***; decile 2: 0.381***, 0.368***; decile 3: 0.28***, 0.267***; decile 4: 0.185*, 0.175*; decile 10: -0.443***, -0.458***

- Interpretation (text summary):
  - Total effect of mps is weaker for individuals more reliant on social grants (less reliant on labor income): e.g., total effect of mps is –0.11 for grant-to-consumption ratio of 0.25 (roughly the 8th consumption decile, model 7); estimate is zero for ratio of 0.9 (roughly 3rd decile).
  - Credit card ownership emerges as an important transmission channel (likely via debt service cost): total effect of mps around –0.45 for credit card owners while zero for those without a credit card.
  - Bank account ownership not an important transmission channel in these estimates.
  - Lagged dependent variable often statistically significant (mostly at the 10 percent level), indicating potential omitted variable or Nickell bias.

### Extension 2 — Observed real policy rate (rpol)
- Estimation details (text):
  - Policy rate (period average) deflated by one year ahead inflation expectations and detrended by first differences.
  - Results for rpol are somewhat weaker overall compared with mps-based models.
- Selected coefficients and interactions (Table 8, L1):
  - Real GDP (L0): 1.588***, 1.588***, 1.587***, 1.577***, 1.624***
  - Real policy rate (rpol) (L1) reported values across models: 0.003, -0.052, 0.011, -0.028, -0.195***
  - Food share of total cons. L1: 0.385***, 0.436***, 0.385***, 0.386***, 0.251***
  - Employment dummy L1: 0.038**, 0.038**, 0.035**, 0.038**, 0.045***
  - rpol * Food share L1: 0.180**
  - rpol * Grant recipient dummy L1: 0.057**
  - rpol * consumption decile interactions (selected): decile 1: 0.536***, decile 2: 0.275***, decile 3: 0.232***, decile 4: 0.223***, decile 5: 0.103*, decile 6: 0.146**, decile 7: 0.167***, decile 8: 0.223***, decile 9: 0.182***

- Memo-item total effects of rpol (selected):
  - Food share = 0.45: 0.028
  - Food share = 0.2: -0.017
  - Employment dummy: -0.003
  - Grant recipient dummy: 0.029
  - Consumption decile 1: 0.34***
  - Consumption decile 2: 0.08
  - Consumption decile 3: 0.036
  - Consumption decile 4: 0.027
  - Consumption decile 5: -0.093*
  - Consumption decile 10: -0.195***

- Interpreted magnitudes (text summary):
  - Average and standard deviation of rpol is –0.3 and 1.0 for 2008–17.
  - Thus, a one standard deviation increase in rpol leads to a 34 percent increase in per-capita household real consumption for the first consumption decile (text compares this to 13 percent from a one standard deviation move in mps).
  - Impact for the 10th consumption decile: around 20 percent reduction in consumption (text), larger than the 15 percent reduction from mps-based standard deviation move.

### Synthesis of findings and interpretation
- Distributional consequences of monetary tightening:
  - Exogenous monetary policy tightening exerts a modest and less unfavorable effect, or even a relatively large favorable effect, on consumption of individuals at lower ends of the consumption distribution.
  - Lower-consumption individuals rely more on government transfers and benefit mainly from lower and more stable inflation; they are less negatively affected via labor income, asset prices, and higher debt service cost.
  - Higher-consumption individuals are more negatively affected (via lower labor income, asset price effects, and higher debt service costs).
- Channels highlighted:
  - Grants/social transfers dampen negative mps effects for more reliant individuals.
  - Credit card ownership amplifies negative mps effects (consistent with debt-service channel).
  - Bank account ownership not identified as a strong transmission channel in these estimates.

### Policy implications and discussion (text summary)
- Monetary policy and structural policies:
  - The South African Reserve Bank (SARB) uses monetary policy to safeguard low and stable inflation. As inflation fell to around the lower end of the official target band, SARB reduced the policy rate by 300 basis points in 2020 to help cope with COVID-19 effects.
  - Welfare-improving policies need to address other key determinants of individual-level consumption: attract private investment, create jobs, boost economic growth, and make growth more inclusive (agenda of key structural reforms identified by the authorities needs fuller implementation).
  - Educational attainment needs enhancement to gain skills and employment—important given COVID-19 scarring risks and the future of work.
- Data and robustness:
  - Results are suggestive: inclusion of lagged dependent variable and significance at mostly 10 percent level in some models indicate potential omitted variable or Nickell bias.
  - Additional waves of NIDS are needed to help solidify the paper’s findings.

*Source: Section 3, wpiea2021078-print-pdf*

### Section 4

### wpiea2021078-print-pdf - Section 4

### Table A1 — Determinants of Per-Capita Household Real Consumption in South Africa: Exogenous Monetary Policy Shock, with Lagged Dependent Variable (Estimated coefficients)

- Sources: Haver, NIDS, and author’s calculations.
- Note: This table reports estimated results from equation (3). Dependent variable is per-capita household real consumption level in log. L0 and L1 signify contemporaneous value and one period lag. ***,**, and * when statistically significant at the 1, 5, and 10 percent levels. Memo item reports the total effects of rpol is calculated using Stat command lincom.

- Model columns: #16 17 18 19 20 21

- Macro variables:
  - Real GDP (log) L0.: 1.562*** 2.009*** 2.032*** 1.983*** 2.009*** 1.823***
  - Exogenous monetary policy shock (mps) L1.: .... -0.053 -0.236*** 0.000 -0.053 -0.457***

- Micro variables:
  - Per-capita household real cons. (log) L1.: -0.055 -0.068 -0.07 -0.07 -0.068 -0.069
  - Education dummy:
    - No education L1.: -0.130* -0.132* -0.134* -0.125 -0.132* -0.105
    - Lower primary L1.: -0.273*** -0.275*** -0.276*** -0.266*** -0.275*** -0.244***
    - Upper primary L1.: -0.223*** -0.226*** -0.229*** -0.220*** -0.226*** -0.212***
    - Secondary L1.: -0.091*** -0.091*** -0.096*** -0.093*** -0.091*** -0.105***
    - Tertiary L1.: ...................
  - Food share of total consumption L1.: 0.242*** 0.244*** 0.253*** 0.244*** 0.244*** 0.243***
  - Employment dummy L1.: 0.051*** 0.050*** 0.051*** 0.045*** 0.050*** 0.049***
  - Grant recipient dummy L1.: -0.02 -0.022 -0.022 -0.021 -0.022 -0.022

- Interaction with mps:
  - mps * Food share L1.: ....... 0.488*** .........
  - mps * Employment dummy L1.: .......... -0.123** ......
  - mps * Grant recipient dummy L1.: ............. 0.001...
  - mps * consumption decile 1 L1.: ................ 0.828***
  - mps * consumption decile 2 L1.: ................ 0.825***
  - mps * consumption decile 3 L1.: ................ 0.725***
  - mps * consumption decile 4 L1.: ................ 0.633***
  - mps * consumption decile 5 L1.: ................ 0.436***
  - mps * consumption decile 6 L1.: ................ 0.364***
  - mps * consumption decile 7 L1.: ................ 0.244**
  - mps * consumption decile 8 L1.: ................ 0.255**
  - mps * consumption decile 9 L1.: ................ 0.317***
  - mps * consumption decile 10 L1.: .................

- Other controls:
  - Location L1.: yes yes yes yes yes yes
  - Location*real GDP L1.: yes yes yes yes yes yes
  - Consumption decile L1.: yes yes yes yes yes yes
  - Individual effects: ... yes yes yes yes yes yes

- Sample and fit statistics:
  - # of Obs.: 26,876 26,876 26,876 26,876 26,876 26,876
  - Cross section (N): 6,719 6,719 6,719 6,719 6,719 6,719
  - Time series (T): 4 4 4 4 4 4
  - Adjusted R^2: 0.115 0.116 0.117 0.116 0.116 0.127

*Italic source attribution: Extracted from wpiea2021078-print-pdf - Section 4*

---


_Source: https://www.imf.org/-/media/files/publications/wp/2021/english/wpiea2021078-print-pdf.pdf_