## 1. Saving Rate and Uncertainty

## Source details

**Canonical URL:** [1. Saving Rate and Uncertainty](https://www.imf.org/-/media/files/publications/wp/2023/english/wpiea2023150-print-pdf.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/wp/2023/english/wpiea2023150-print-pdf.pdf.md)
- [Structured JSON version](/-/media/files/publications/wp/2023/english/wpiea2023150-print-pdf.pdf.json)

---

### Introduction and objectives
- Aim: investigate causes of persistently low household savings in three Southern European countries: Cyprus, Greece, and Portugal (SE3).
- Key questions:
  - What are the main factors driving household savings in SE3 countries?
  - Why do SE3 countries have different household saving behavior from more advanced EA countries?
  - How will high inflation and monetary policy tightening affect saving behavior in SE3?
- Data and approach:
  - Household-level analysis using the Household Finance and Consumption Survey (HFCS) augmented via statistical matching with Household Budget Surveys (HBSs) and national accounts.
  - Multivariate regressions, quantile regressions, and policy simulation exercises based on estimated quantile models.

### Theoretical drivers (summary)
- Framework: life-cycle and permanent-income models, extended for bequest motives, precautionary savings, longevity risk, wealth, and borrowing constraints.
- Ambiguous channels highlighted:
  - Interest rates: substitution, income, and wealth effects can offset.
  - Uncertainty: precautionary motive predicts higher savings.
- Additional determinants emphasized: permanent and current income, education, age, family size, gender, borrowing constraints, housing wealth and house prices, fiscal policy and public pensions (Ricardian equivalence and public insurance substitution).

### Empirical findings — descriptive
- Historical patterns:
  - SE3 (Cyprus, Greece, Portugal) have among the lowest household saving rates in the EA; saving rates in SE3 declined before the GFC, spiked during the GFC, plunged during the SDC, and jumped during COVID-19.
  - During COVID-19, savings rose due to precautionary and involuntary savings and fiscal support.
- Distributional evidence (HFCS waves):
  - Saving rates declined across HFCS waves; SE3 show a longer tail of negative saving ratios.
  - Median saving rates display hump-shaped profiles by age; elderly saving rates declined across waves.
  - Savings are negative for first and second income quintiles and highest for the top quintile.
  - Self-declared motives: "protection against unexpected events" is the most cited motive; Greece shows the strongest precautionary motive among SE3.
- Household-characteristics correlations:
  - Positive associations: income and education correlate positively with savings.
  - Negative associations: household size, female-headed households, and higher debt service correlate with lower savings.
  - Wealth shows a positive association with savings in cross-sections where buffer-stock negative channel is dominated.

### Econometric results — key coefficients and patterns
- Estimation: OLS, median and quantile regressions; IV quantile methods, robust regressions, Tobit for robustness.
- Selected reported coefficients (EA sample and OLS/quantile examples preserved exactly where given):
  - Income: OLS 0.702*** (s.e. 0.00584); quantile coefficients fall from 0.694*** (0.1) to 0.274*** (0.9).
  - Age: OLS 0.00265*** (s.e. 0.000224).
  - Debt (debt service): OLS -0.00968*** (s.e. 0.000312).
  - Size (household members): OLS -0.0796*** (s.e. 0.00199).
  - Wealth: OLS -0.0896*** (s.e. 0.00374); declines toward higher quantiles (e.g., -0.00752*** at 0.9).
  - Deposit (real) rate: OLS 0.0225*** (s.e. 0.00247).
  - Inflation: negative and significant in specifications including macro variables (examples -0.03***, -0.02*** reported in narrative).
  - Private credit: OLS -0.000971*** (s.e. 0.000238).
  - Government budget balance: OLS -0.00467** (s.e. 0.00204).
  - Government pension expenditures: OLS -0.0192*** (s.e. 0.00251).
  - Income uncertainty (micro measure): OLS -0.106*** (s.e. 0.0139) but positive at high quantiles (e.g., 0.0443*** at 0.9, s.e. 0.00749).
- Sample and fit:
  - N: 56,103 observations in key tables.
  - OLS R-sq: 0.442; reported quantile R-sq vary (examples: 0.187 at 0.1, 0.363 at 0.9).

### Robustness checks (summary)
- Permanent income proxy, IV quantile, macro uncertainty/housing-price measures, robust regressions, and Tobit censoring generally yield qualitatively similar results.
- Notable robustness outcomes:
  - IV specifications: income positive but smaller (examples 0.0870*** and 0.0774*** with s.e. 0.0240 and 0.0214); wealth may flip sign (e.g., 0.129***) indicating endogeneity sensitivity.
  - Macro uncertainty: OLS 0.0108*** (s.e. 0.00184); QR 0.0131*** (s.e. 0.00120).
  - Robust regressions: deposit rate positive and larger in some specs (e.g., 0.0437***); government pension negative and sizable (e.g., -0.0415*** to -0.0827***).
  - Tobit: income ~0.698*** and 0.702*** in columns; wealth negative (e.g., -0.0881***).
- Wave-specific results:
  - Wave Ns: wave 1 N = 14,669; wave 2 N = 20,039; wave 3 N = 21,395.
  - Income positive across waves (examples reported: 0.453*** wave 1 to 0.411*** wave 3 in different columns).
  - Debt and size consistently negative; deposit rate large and positive in some wave-specific columns (e.g., 0.0979***).

### Policy simulations — scenarios and quantitative impacts
- Approach: quantile-regression-based comparative-static simulations using EA specifications (Table 1 columns 6 and 8) and SE3 specs (Table 2 columns 3 and 4); parameters estimated for each of the 99 percentiles; five stylized shocks.
- Scenario definitions (magnitudes as reported):
  - Scenario 1: inflation +5 ppt.
  - Scenario 2: real interest rate +200 bps.
  - Scenario 3: government budget balance worse by 2 ppt of GDP.
  - Scenario 4: untargeted lump-sum income transfer equal to 3 percent of average income to each household.
  - Scenario 5: same total transfers as Scenario 4 but distributed only to low-income households (targeted).
- Aggregate and distributional simulation findings (numeric impacts preserved):
  - A. Inflation Shock (inflation +5 ppt)
    - EA aggregate household saving rate: reduces by 3.6 ppts.
    - Top decile: household saving rate increases, on average, by about 2 pps.
    - SE3 specifics:
      - Greece: household saving rate impacted the least (by about 1.8 pps).
      - Cyprus: household saving rate estimated to drop by 4.3 pps.
    - Distribution: least-saving households hit disproportionately; positive effect on top savers absent in SE3.
  - B. Interest Rate Shock (real interest rate +200 bps)
    - EA average: raises household saving rate by 4.6 ppts.
    - Cyprus, Greece, Portugal: raise saving rates by around 1-2 ppts.
    - Distribution: high-saving households gain the most; lowest-saving households lose markedly.
  - C. Government Budget Balance Shock (fiscal balance - 2 ppt of GDP)
    - EA average: raises household savings by more than 1 ppts.
    - Cyprus, Greece, Portugal: raise household savings by almost 1 ppts.
    - Distribution: bottom 10 percent in EA see savings decrease by around 1 ppt on average; in SE3, lowest-saving decile increases by around 2.5 ppts while highest-saving decile increases by ~1 ppt.
  - D. Income Shock — Untargeted lump-sum transfer (3 percent of average income to all households)
    - EA median household saving rate: improves by about 1 percentage point.
    - Low-saving households: 4.2 percentage points increase.
    - High-saving households: 0.2 percentage point increase.
  - E. Income Shock — Targeted transfers to low-income households (same total transfer amount)
    - EA median saving rate: improves by 1.3 percentage points (0.3 ppt above untargeted).
    - Country median improvements reported:
      - Portugal: 1.3 percentage point improvement (additional 0.4 ppt above broad-based).
      - Cyprus: 2 percentage point improvement (additional 0.7 ppt above broad-based).
      - Greece: 1.8 percentage point improvement (additional 0.7 ppt above broad-based).
- Aggregate country-level simulation note:
  - "The change in country-level savings rate is proxied using the income-weighted average of the median response."
  - Figure panels plot simulated changes spanning -5 to 6 (Percentage point) on the vertical axis.

### Economic significance (standardized effects)
- Income is the most prominent determinant; reported standardized impacts:
  - A one standard deviation increase in household debt service ratio and private credit is estimated to decrease household saving ratio by around 0.1 standard deviation each.
  - A one standard deviation increase in deposit rate is associated with an increase in household saving ratio by around 0.1 standard deviation.
  - Family size and age have meaningful standardized impacts comparable to debt and deposit rate.

### Policy implications and recommendations
- Targeted income support:
  - Income-enhancing policies targeted to low-saving or low-income households are more effective at raising savings among the lowest-saving households than untargeted transfers.
  - Targeted transfers (same total resources) yield larger median and low-decile saving increases than untargeted broad-based transfers.
- Monetary policy considerations:
  - Monetary tightening (higher real interest rates) raises aggregate savings but unevenly benefits higher-saving households and can harm lowest-saving households; complementary income support or targeted measures recommended to protect vulnerable households.
- Inflation management:
  - Inflation episodes reduce aggregate savings and hit low-saving households disproportionately; policies to contain inflation and protect low-income households are important.
- Fiscal policy trade-offs:
  - Fiscal expansion can raise household savings via Ricardian-like effects on average, but distributional impacts vary; in SE3, well-targeted expansionary fiscal policy could support low-saving households while fiscal space and external imbalances constrain scope.
- Structural reforms:
  - Raise permanent income through education, labor market reforms, and growth-friendly fiscal composition.
  - Deepen financial markets to reduce borrowing constraints and improve deposit returns.
  - Reforms in social security and targeted antipoverty programs: recalibrate pension parameters (e.g., raise retirement age) and target transfers to raise savings among vulnerable households without undermining fiscal sustainability.
- Debt resolution and insolvency:
  - Stronger debt resolution frameworks to facilitate deleveraging and healthier household balance sheets, particularly for low-income highly leveraged households.

### Annex highlights: determinants and statistical matching
- Annex I — Determinants expected signs (as listed):
  - Income: +
  - Age: +/-
  - Uncertainty: +/-
  - Wealth: -
  - Housing: -
  - Debt: -
  - Size: -
  - Education: +
  - Household head gender (male): +
  - Interest rate: +/-
  - Inflation: +/-
  - Private credit: -
  - Government budget balance: -
  - Government pension: -
- Annex II — Statistical matching procedures:
  - HFCS to HBS: impute HBS total consumption to HFCS using covariates (inverse Engel curve approach) to correct HFCS underreporting of total consumption.
  - HFCS to OECD income tax: compute after-tax income by applying marginal tax rates to household taxable income plus 2/3 of self-employment income using OECD Tax Database rates.
  - HFCS to National Accounts: match aggregate income and consumption to national accounts to ensure sample savings rates align with national accounts while preserving relative household-level distributions.

*Source: wpiea2023150-print-pdf — "1. Saving Rate and Uncertainty"; canonical URL: https://www.imf.org/-/media/files/publications/wp/2023/english/wpiea2023150-print-pdf.pdf*

### 1. Saving Rate and Uncertainty .........................................................................................

### 1. Saving Rate and Uncertainty

### Major sections (chapter/section headings)
- 1. Saving Rate and Uncertainty .......................................................................................................................  8
- 2. Household Disposable Income ................................................................................................................... 9
- 3. Household-Level Saving Distribution ........................................................................................................  10
- 4. Purpose of Saving ..................................................................................................................................... 11
- 5. Household Saving Rate by Age ................................................................................................................. 12
- 6. Household Saving Rate by Income and Education.................................................................................... 13
- 7. Household Saving Rate by Employment Status ........................................................................................ 14
- 8. Household Saving Rate by Net Wealth ..................................................................................................... 15
- 9. Household Saving Rate and Borrowing Constraints.................................................................................. 16
- 10. Household Saving Rate by Household Characteristics ............................................................................  17
- 11 Coefficients Across Distribution of Savings .............................................................................................. 21
- 12. Economic Significance ............................................................................................................................ 22
- 13. Simulation Results, Inflation Shock ......................................................................................................... 26
- 14. Simulation Results, Interest Rate Shock .................................................................................................. 27
- 15. Simulation Results, Fiscal Balance Shock ................................................................................................. 28
- 16. Simulation Results, income shock (broad-based).................................................................................... 29
- 17. Simulation Results, income shock (targeted) .......................................................................................... 30

### Tables
- 1. Baseline Regressions .............................................................................................................................. 20
- 2. Baseline Regressions for Southern-European Countries ........................................................................ 23

### Annexes and technical appendices
- Annex I. Determinants of Household Savings ........................................................................................ 33
- Annex II. Statistical Matching ................................................................................................................... 34
  - A.1 Matching HFCS to HBS ........................................................................................................... 34
  - A.2 Matching HFCS to OECD income tax ...................................................................................... 35
  - A.3 Matching HFCS to National Accounts ...................................................................................... 35
- Annex III. Empirical Investigation ............................................................................................................. 36
  - Econometric analysis ...................................................................................................................... 36
  - Robustness checks ........................................................................................................................ 38
  - Simulation Results, Country Aggregate .......................................................................................... 45

*Source: wpiea2023150-print-pdf - 1. Saving Rate and Uncertainty; canonical URL: https://www.imf.org/-/media/files/publications/wp/2023/english/wpiea2023150-print-pdf.pdf*

### References .............................................................................................................

### wpiea2023150-print-pdf - References .............................................................................................................

### Introduction and objectives
- Aim: investigate causes of persistently low household savings in three Southern European countries: Cyprus, Greece, and Portugal (SE3).
- Key questions:
  - What are the main factors driving household savings in SE3 countries?
  - Why do SE3 countries have different household saving behavior from more advanced EA countries?
  - How will high inflation and monetary policy tightening affect saving behavior in SE3?
- Data and approach:
  - Household-level analysis using the Household Finance and Consumption Survey (HFCS) augmented via statistical matching with Household Budget Surveys (HBSs) and national accounts.
  - Multivariate regressions, quantile regressions, and policy simulation exercises based on estimated quantile models.

### Theoretical drivers (summary)
- Life-cycle and permanent-income frameworks underpin the analysis; extensions include bequest motives, precautionary savings, longevity risk, wealth, and borrowing constraints.
- Ambiguous effects:
  - Interest rates: substitution, income, and wealth effects can offset.
  - Uncertainty: precautionary motive predicts higher savings.
- Additional determinants highlighted:
  - Income (permanent and current), education, age, family size, gender.
  - Borrowing constraints, housing wealth and house prices.
  - Fiscal policy and public pensions (Ricardian equivalence and public insurance substitution).

### Empirical findings — descriptive
- Historical patterns:
  - SE3 countries (Cyprus, Greece, Portugal) have among the lowest household saving rates in the EA; saving rates in SE3 declined before the GFC, spiked during the GFC, plunged during the SDC, and jumped during COVID-19.
  - During COVID-19, savings rose due to precautionary and involuntary savings and fiscal support.
- Distributional evidence (HFCS waves):
  - Saving rates declined across HFCS waves; SE3 show a longer tail of negative saving ratios.
  - Median saving rates display hump-shaped profiles by age; elderly saving rates declined across waves in general.
  - Savings are negative for first and second income quintiles and highest for the top quintile.
  - Self-declared motives: protection against unexpected events is the most cited motive; Greece shows strongest precautionary motive among SE3.
- Household characteristics correlations:
  - Positive relationships: income and education correlate positively with savings.
  - Negative relationships: household size, female-headed households, and higher debt service correlate with lower savings.
  - Wealth shows a positive association with savings in the cross-section (positive effect dominates buffer-stock negative channel).

### Econometric results — key coefficients and patterns
- Estimation framework: OLS, median and quantile regressions; robustness checks include IV quantile methods, robust regressions, Tobit.
- Selected baseline coefficient highlights (EA sample, Table 1 and narrative):
  - Income: positive and significant (example coefficient 0.699*** in an OLS specification).
  - Age: positive and significant (examples 0.001*** to 0.003*** across specifications).
  - Debt: negative and significant (example -0.01***).
  - Size (household size): negative and significant (example -0.08***).
  - Gender (male head): often positive in quantile estimates (some QR coefficients 0.01***–0.02***).
  - Income uncertainty (unemployment proxy): mixed in OLS/median, but quantile regressions show positive correlation with savings for more than 50 percent of the distribution (precautionary motive relevant for a sizeable portion).
  - Wealth: negative coefficients reported in some specifications (e.g., -0.07***) reflecting countervailing effects; quantiles show higher sensitivity for lower-saving households.
  - Deposit (real) rate: positive and significant in richer specifications (examples 0.042*** to 0.057*** in QR results).
  - Inflation: negative and significant in specifications including macro variables (examples -0.03***, -0.02***).
  - Private credit: negative association in EA baseline (small magnitude but significant, e.g., -0.00***).
  - Government budget balance: negative coefficients (example -0.03***), consistent with Ricardian-like effects; government pension spending: negative association with household savings (example -0.03***).
- Economic significance (standardized coefficients):
  - Income is the most prominent determinant.
  - A one standard deviation increase in household debt service ratio and private credit is estimated to decrease household saving ratio by around 0.1 standard deviation each.
  - A one standard deviation increase in deposit rate is associated with an increase in household saving ratio by around 0.1 standard deviation.
  - Family size and age also have meaningful standardized impacts comparable to debt and deposit rate.
- SE3 vs EA differences:
  - Drivers broadly similar, but magnitudes differ—income, deposit rate, inflation, and government balance effects often smaller in SE3.
  - SE3 lower income levels imply higher shares of income spent on necessities, muting marginal saving responses.

### Robustness checks (summary)
- Accounting for permanent income proxy (fitted income conditioned on education, unemployment, age, gender, marital status) — results qualitatively similar.
- IV quantile methods for endogeneity — results qualitatively similar.
- Macro uncertainty and housing-price measures — results qualitatively similar; macro uncertainty shows stronger positive association with savings.
- Robust regression (down-weighting outliers) and Tobit censoring — broadly similar outcomes; with robust/Tobit approaches uncertainty becomes more strongly and positively associated with savings (precautionary motive reinforced).
- Wave-specific estimates — generally consistent, though Wave 1 shows some sign changes likely due to GFC volatility.

### Policy simulations — scenarios and quantitative impacts
- Approach: quantile regression-based simulations for EA sample (specifications in Table 1 columns 6 and 8) and SE3 subsample (Table 2 columns 3 and 4); parameters estimated for each of the 99 percentiles; five stylized shock scenarios (comparative-static exercise).
- Baseline: a sizeable portion of households projected to have negative savings; percentage of dissaving households higher in SE3.

Scenario definitions (magnitudes illustrative):
  - Scenario 1: inflation +5 ppt.
  - Scenario 2: real interest rate +200 bps.
  - Scenario 3: government budget balance worse by 2 ppt of GDP (fiscal expansion).
  - Scenario 4: untargeted lump-sum income transfer equal to 3 percent of average income to each household.
  - Scenario 5: same total transfers as Scenario 4 but distributed only to low-income households (targeted).

Key simulation findings (numeric impacts preserved as reported):
- A. Inflation Shock (inflation +5 ppt)
  - Aggregate household saving rate: reduces by 3.6 ppts for the EA.
  - Distributional pattern: least-saving households hit disproportionately; effect eases up the distribution and turns positive for top savers.
  - Top decile: household saving rate increases, on average, by about 2 pps (interpreted as precautionary response by highest savers).
  - SE3 specifics:
    - Greece: household saving rate impacted the least (by about 1.8 pps).
    - Cyprus: household saving rate estimated to drop by 4.3 pps.
    - Overall SE3 responses comparable in sign to EA but smaller on average; the positive effect on top savers is absent in SE3.

- B. Interest Rate Shock (real interest rate +200 bps)
  - EA average: raises household saving rate by 4.6 ppts.
  - Cyprus, Greece, Portugal: raise saving rates by around 1-2 ppts.
  - Distributional pattern: high-saving households gain the most; lowest-saving households lose markedly (income effect dominates for the poor), implying monetary tightening can worsen outcomes for lowest-saving households absent income support.
  - SE3 responses smaller in magnitude relative to EA and less linear across distribution.

- C. Government Budget Balance Shock (fiscal balance -2 ppt of GDP)
  - EA average: reduction in fiscal balance by 2 percent of GDP raises household savings by more than 1 ppts.
  - Cyprus, Greece, Portugal: raise household savings by almost 1 ppts.
  - Distributional effect: bottom 10 percent in EA see savings decrease by around 1 ppt on average; in SE3, lowest-saving households generally see higher increases in savings (example: average savings rate increases by around 2.5 ppts for lowest-saving decile in SE3, but only ~1 ppt for highest-saving decile).
  - Interpretation: expansionary fiscal policy could support low-saving households in SE3 if well targeted; caution due to external imbalances and high public debt.

- D. Income Shock — Untargeted lump-sum transfer (3 percent of average income to all households)
  - EA median household saving rate: improves by about 1 percentage point.
  - High-saving households: small change (0.2 percentage point increase).
  - Low-saving households: much stronger response — 4.2 percentage points increase for low-saving households.
  - SE3 responsiveness to income broadly aligned with EA despite income gaps.

- E. Income Shock — Targeted transfers to low-income households (same total transfer amount as untargeted case)
  - EA median saving rate: improves by 1.3 percentage points (0.3 ppt above the untargeted case).
  - Country-specific median improvements reported:
    - Portugal: 1.3 percentage point improvement (additional 0.4 ppt above broad-based scenario).
    - Cyprus: 2 percentage point improvement (additional 0.7 ppt above broad-based scenario).
    - Greece: 1.8 percentage point improvement (additional 0.7 ppt above broad-based scenario).

### Policy implications and recommendations (drawn from results)
- Income-enhancing policies targeted to low-saving or low-income households can be more effective at raising savings among the lowest-saving households than untargeted transfers.
- Monetary tightening (higher interest rates) raises aggregate savings but unevenly benefits higher-saving households and can harm lowest-saving households; consider complementary income support or targeted measures to protect vulnerable households.
- Inflation episodes reduce aggregate savings, hitting low-saving households disproportionately; policy efforts to contain inflation and protect low-income households are important.
- Fiscal expansion can raise household savings via Ricardian-like effects on average, but distributional impacts vary; in SE3, expansionary fiscal policy targeted to low-income households could bolster savings for low-saving households, but fiscal space and external imbalances constrain scope.
- Structural policies to raise permanent income (education, labor market reforms) and to deepen financial markets (reduce borrowing constraints, improve deposit returns) could sustainably increase household savings.

*Source: IMF staff analysis and empirical results presented in the chapter "Household Savings in Southern European Countries" (HFCS-based regressions, quantile simulations, and descriptive evidence) as provided in the PDF content.*

### 0.6 percentage point above the broad-based scenario). The impact on low-income households is estimated to

### V. Conclusions and Policy Implications

### Key findings on household saving rates and distributional patterns
- Study focus: household saving rates in three euro area (SE3) countries—Cyprus, Greece, and Portugal—and how they differ from average EA trends.
- Post-GFC and pre-pandemic: saving rates fell across EA, with SE3 countries recording among the lowest saving rates in the EA.
- Distributional pattern: EA saving rate distribution is negatively skewed; SE3 countries exhibit an even longer tail of negative saving ratios.
- Descriptive correlates:
  - Positive relationship between income and savings; saving gaps between SE3 and other EA countries tend to be wider for higher income households.
  - Education is positively correlated with savings, with highest saving rates for those with tertiary education.
  - Employment status positively influences savings; saving gaps between SE3 and other EA countries are larger for employed households.
  - Family size is negatively associated with savings.
  - Female-headed households tend to have lower savings.
- Deleveraging after the GFC in SE3 was accompanied by increasing household savings, reflecting a positive impact of tighter borrowing constraints on savings; this correlation appears stronger in SE3 than the EA average.

### Econometric and distributional insights
- Standard cross-sectional regressions:
  - Income, age, and education are positively associated with savings.
  - Household debt service and wealth are negatively correlated with savings.
  - Real interest rate on deposits tends to increase household savings but has important distributional differences.
  - Higher government deficits are positively associated with higher household savings (supporting Ricardian equivalence).
  - Higher government-provided pension benefits tend to be negatively correlated with household savings.
- Quintile regressions on household micro-level data:
  - Households at lower saving quantiles are more vulnerable to macroeconomic shocks.
  - Associations between household saving rates and income and debt are particularly strong for lower savers.
  - Higher uncertainty tends to increase savings for more than half of households.

### Simulation results and shock impacts
- Simulations indicate heterogeneous impacts by household type:
  - Higher inflation hits lower-saving households harder — controlling inflation is particularly critical for these households.
  - Under higher interest rates: low-saving households’ saving rates deteriorate; high-saving households’ saving rates improve.
  - Deterioration in fiscal balance tends to increase savings rates more for lower-saving households than higher-saving ones.
  - The impact of shocks is generally smaller for SE3 countries than the EA average.
- Income support simulations:
  - Income support has a noticeable effect on household saving rates, especially for low-saving households.
  - Targeted income support is stronger relative to broad-based income support in raising household savings among low-saving households.
  - Empirical example from simulations: household saving rates improve by 3, 4.4, and 4 percentage points in case of Portugal, Cyprus and Greece respectively (an additional 0.4, 1.2, and 1 percentage point increase in household savings on top of the effect of the broad-based income support).
  - Under the targeted income support scenario, the same amount of resources as under the broad-based income support scenario is spent, however is distributed only among households with income below the 25th income percentile.

### Policy implications and recommendations for SE3 countries
- Holistic approach required: addressing low household savings entails macroeconomic and distributional trade-offs.
- Fiscal sustainability constraint: policy measures should support household savings without undermining fiscal balances given high public debt levels in SE3.
- Policy mix to incentivize greater household savings in a fiscally sustainable way; specific policy takeaways:
  - Income and growth:
    - Income plays a central role for household savings; stronger economic growth is emphasized for higher household savings.
    - The income gap between SE3 and other EA countries largely explains differences in saving rates.
    - Fiscal policy that achieves a more growth-friendly composition of government spending can address growth bottlenecks and enhance growth potential.
    - Improving public investment management can enhance the efficiency and productivity of public investment.
  - Social security and targeted transfers:
    - Sustainable social security systems that guarantee adequate income for the most vulnerable can lower uncertainty and stimulate household savings.
    - Well-designed and targeted social transfers can increase savings of the most vulnerable households, which tend to respond more to changes in income and policy variables.
    - Targeting income transfers (e.g., by income level or employment status) or means-tested social safety nets would raise household savings while limiting associated costs.
    - Reforms adjusting pension parameters (such as raising retirement age) and recalibrating generosity of public pension schemes, accompanied by targeted antipoverty programs, could safeguard long-term sustainability of public pensions and provide incentives for higher private saving for retirement.
  - Structural reforms:
    - Education: support access to higher education, modernize curricula, increase financial literacy.
    - Labor market: improve quality of vocational training and active labor market policies; raise female labor participation and address gender inequality to narrow saving gaps.
    - Insolvency and debt resolution: stronger debt resolution frameworks can facilitate efficient deleveraging and help build healthier household balance sheets, particularly for low-income households with high leverage.

### Annex highlights: determinants and matching methodology
- Annex I — Determinants of Household Savings (variable expected signs as listed):
  - Income: +
  - Age: +/-
  - Uncertainty: +/-
  - Wealth: -
  - Housing: -
  - Debt: -
  - Size: -
  - Education: +
  - Household head gender: male +
  - Interest rate: +/-
  - Inflation: +/-
  - Private credit: -
  - Government budget balance: -
  - Government pension: -
- Annex II — Statistical matching procedures:
  - HFCS to HBS:
    - Household Budget Surveys (HBS) record household consumption on 12 subcategories; HFCS mixes detailed items on food, utilities, and rent with a one-shot question on total consumption.
    - One-shot total consumption questions in HFCS tend to understate total consumption relative to HBS and national accounts.
    - Imputation approach: follow Lamarche (2017) and earlier literature (Skinner (1987), Browning et al. (2003), Blundell (2004, 2008), Attanasio and Pistaferri (2014)) to impute total consumption from HFCS subsets using a linear inverse Engel curve specification incorporating interaction terms of food consumption with household income quintiles.
    - Practical form: impute HBS total consumption to HFCS using covariates that best match the distribution of total consumption.
  - HFCS to OECD income tax:
    - Follow Slacalek et al. (2020) to compute after-tax income by applying marginal tax rates to household taxable income plus 2/3 of self-employment income; tax rates obtained from OECD Tax Database.
  - HFCS to National Accounts:
    - To correct underreporting, follow Slacalek et al. (2020) to match aggregate income and consumption levels to national accounts, ensuring resulting savings rates in sample data match national accounts savings rates while not altering relative levels of income or consumption across households of the same country.

*IMF Working Papers — Household Savings in Southern European Countries (V. Conclusions and Policy Implications).*

### Annex III. Empirical Investigation

### Annex III. Empirical Investigation

### Variable definition and data sources
- Income: Logarithm of disposable income — HFCS
- Age: Household head age — HFCS
- Uncertainty: Unemployment rate — HFCS
- Wealth: Logarithm of wealth to disposable income — HFCS
- Housing: Share of housing wealth in disposable income — HFCS
- Debt: Debt service to disposable income — HFCS
- Size: Number of household members — HFCS
- Education: Education attainment — HFCS
- Household head gender: Gender of the reference person — HFCS
- Interest rate: Real deposit rate — IMF
- Inflation: Annual dynamics of Consumer Price Index — IMF
- Private credit: Private sector credit to GDP — IMF
- Government budget balance: General government budget balance to GDP — IMF
- Government pension: General government pension expenditures to GDP — IMF

### Main econometric findings (quantile regressions and OLS)
- Income
  - Positive and statistically significant across OLS and all reported quantiles.
  - OLS coefficient: 0.702*** (standard error 0.00584).
  - Quantile coefficients decline across quantiles from 0.694*** (0.1) to 0.274*** (0.9), with intermediate values precisely as reported.
- Income uncertainty (micro measure)
  - Mixed effects across quantiles:
    - Negative and significant at low quantiles (e.g., -0.106*** at OLS, s.e. 0.0139).
    - Weak/insignificant around median.
    - Positive and significant at higher quantiles (e.g., 0.0443*** at 0.9, s.e. 0.00749).
- Wealth
  - Negative and statistically significant across OLS and quantiles.
  - OLS: -0.0896*** (s.e. 0.00374); declines in magnitude toward higher quantiles (e.g., -0.00752*** at 0.9).
- Debt
  - Negative and statistically significant across most specifications.
  - OLS: -0.00968*** (s.e. 0.000312).
- Housing (share)
  - Generally negative and often significant.
  - OLS: -0.0538*** (s.e. 0.00486); quantile magnitudes vary, several significant at conventional levels.
- Age
  - Positive and statistically significant throughout.
  - OLS: 0.00265*** (s.e. 0.000224).
- Size (household members)
  - Negative and statistically significant throughout.
  - OLS: -0.0796*** (s.e. 0.00199).
- Gender (male)
  - Coefficients vary by specification; some positive and significant at upper quantiles or in robustness checks (e.g., 0.00553** at 0.9 in quantile table).
- Deposit (real) rate
  - Positive and significant in many specifications, but sign and magnitude vary by quantile.
  - OLS: 0.0225*** (s.e. 0.00247).
- Private credit
  - Small negative coefficients, often statistically significant.
  - OLS: -0.000971*** (s.e. 0.000238).
- Government budget balance
  - Mostly negative coefficients; significance varies across quantiles and specifications.
  - OLS: -0.00467** (s.e. 0.00204).
- Government pension expenditures
  - Negative and statistically significant across most specifications.
  - OLS: -0.0192*** (s.e. 0.00251).

- Sample sizes and fit
  - Key reported N: 56,103 observations for main quantile/OLS tables.
  - Reported R-squared values for quantiles: e.g., OLS R-sq 0.442; quantile R-sq reported per column (e.g., 0.187 at 0.1, 0.363 at 0.9).

### Robustness checks and alternative specifications
- Permanent income and education (OLS and QR)
  - Permanent income: 0.689*** and 0.686*** (OLS), 0.424*** and 0.425*** (QR), with standard errors reported.
  - Income uncertainty: large negative OLS estimates in these specifications (e.g., -0.585***), negative QR estimates (e.g., -0.280***).
  - Education: positive and significant (e.g., 0.0972*** (s.e. 0.00205) in OLS).
  - Sample N: 55,835; R-sq about 0.442 (OLS) and 0.404–0.410 (QR).
- Endogeneity checks
  - IV/endo-aware specifications report:
    - Income: positive e.g., 0.0870*** and 0.0774*** (standard errors 0.0240 and 0.0214).
    - Income uncertainty: negative and significant (e.g., -0.0479***).
    - Wealth: positive in these specifications (e.g., 0.129***), indicating sensitivity to endogeneity treatment.
    - N: 55,835.
- Income uncertainty — macro measure
  - Macro uncertainty shows small positive coefficients:
    - OLS: 0.0108*** (s.e. 0.00184); QR: 0.0131*** (s.e. 0.00120).
  - Income remains strongly positive in these specs (e.g., 0.705*** OLS).
  - N: 56,103.
- Housing — macro measure
  - Housing macro coefficient small negative and significant in OLS and QR (e.g., -0.00339*** OLS, s.e. 0.000350).
  - Income remains positive (e.g., 0.753*** OLS).
  - N: 62,766.
- Sensitivity to outliers — robust regressions
  - Income positive across robust regressions (e.g., 0.371*** to 0.377***).
  - Income uncertainty (micro and macro) small positive or mixed in robust specs (micro 0.0297***; macro 0.0152*** in some columns).
  - Wealth shows negative or, in some columns, small positive coefficients depending on transformation (e.g., -0.0213*** or 0.00512***), reflecting specification sensitivity.
  - Deposit rate positive and significant (e.g., 0.0437***).
  - Government pension negative and sizable across robust regressions (e.g., -0.0415*** to -0.0827***).
  - N varies (e.g., 56,102; 62,765); adjusted R-sq range reported (e.g., 0.574 to 0.594).
- Tobit regressions
  - Results broadly similar to OLS/QR:
    - Income positive and significant (e.g., 0.698*** and 0.702*** across columns).
    - Income uncertainty negative in column (1) (e.g., -0.107***) and positive in column (2)/(4) for quantile-like specification (e.g., 0.0418***).
    - Wealth negative and significant (e.g., -0.0881***).
    - N: 56,103.

### Wave-specific results
- Separate regressions by survey wave (wave 1, wave 2, wave 3) show:
  - Income positive and significant across waves and columns (e.g., 0.453*** wave 1 to 0.411*** wave 3 in different columns), with standard errors reported.
  - Income uncertainty coefficients vary by wave and column and are often not significant.
  - Wealth negative and significant across waves (e.g., -0.0467*** in wave 1 to -0.0196*** in wave 3).
  - Debt and size consistently negative and significant across waves.
  - Deposit rate displays large positive coefficients in some wave-specific columns (e.g., 0.0979***).
  - N by wave: 14,669 (wave 1), 20,039 (wave 2), 21,395 (wave 3).

### Simulation results (aggregate and country-level)
- Figure A.III.1. Simulation Results, Country Aggregate (and country panels for Cyprus, Greece, Portugal)
  - Note: "The change in country-level savings rate is proxied using the income-weighted average of the median response."
  - Simulated scenarios (as presented on chart axes):
    - Inflation + 5%
    - Fiscal balance - 2% GDP
    - Real interest rate + 200 bps
  - The plotted range of simulated changes in aggregated saving rate spans from -5 to 6 (Percentage point) on the vertical axis in the figure panels.

*Source: IMF staff estimates.*

### References

### References

### Foundational theories and consumption/saving models
- Friedman, M. 1957. A Theory of the Consumption Function. Princeton, NJ: Princeton University Press.
- Keynes, John Maynard. 1936. The General Theory of Employment, Interest and Money. London: Macmillan.
- Modigliani, Franco and Richard Brumberg.1954. “Utility Analysis and the Consumption Function: An Interpretation of Cross-Section Data,” in Post-Keynesian Economics. Kenneth K. Kurihara, ed. New Brunswick, N.J.: Rutgers University Press, pp. 388–436.
- Hall, R. E. (1978). Stochastic implications of the life cycle-permanent income hypothesis: Theory and evidence. Journal of Political Economy, 86, 971–987.
- Deaton, Angus. (1992) Understanding Consumption (Oxford: Oxford University Press.)
- Deaton, Angus, 1991, “Saving and Liquidity Constraints,” Econometrica, Vol. 59, pp. 1121–1142.
- Friedman, M. 1957. A Theory of the Consumption Function. Princeton, NJ: Princeton University Press.
- Carroll Christopher D. (1997). “Buffer-Stock Saving and the Life Cycle/Permanent Income Hypothesis.” Quarterly Journal of Economics, 112, 1–56.
- Carroll, Christopher D. 1992. “The Buffer-Stock Theory of Saving: Some Macroeconomic Evidence.” Brookings Papers on Economic Activity.2, pp. 61–156.
- Carroll, Christopher D. (2001). “A Theory of the Consumption Function, with and without Liquidity Constraints.” Journal of Economic Perspectives, 15(3), 23–45.
- Hall, R. E. (1978). Stochastic implications of the life cycle-permanent income hypothesis: Theory and evidence. Journal of Political Economy, 86, 971–987.
- Zeldes, Stephen P, 1989, “Optimal Consumption with Stochastic Income: Deviations from Certainty Equivalence,” The Quarterly Journal of Economics, Vol. 104, No. 2, pp. 275-298.
- Barro, Robert J., 1974, “Are Government Bonds Net Wealth?,” Journal of Political Economy, Vol. 82, pp. 1095-1117.
- Barro, Robert J. (1989). The Ricardian Approach to Budget Deficits. Journal of Economic Perspectives, 3(2), p. 37–54.
- Seater, John J., 1993, “Ricardian Equivalence,” Journal of Economic Literature, Vol. 31, pp. 142-190.
- Browning, Martin, and Annamaria Lusardi (1996) Household Saving: Micro Theories and Micro Facts. Journal of Economic Literature 34 1797-1855
- Modigliani, Franco and Richard Brumberg.1954. “Utility Analysis and the Consumption Function: An Interpretation of Cross-Section Data,” in Post-Keynesian Economics. Kenneth K. Kurihara, ed. New Brunswick, N.J.: Rutgers University Press, pp. 388–436.
- Friedman, M. 1957. A Theory of the Consumption Function. Princeton, NJ: Princeton University Press.

### Empirical determinants and regional studies of saving
- Loayza, Norman, Klaus Schmidt-Hebbel, and Luis Servén, 2000, “What Drives Private Saving Across the World?,” The Review of Economics and Statistics, Vol. 82, No. 2, pp. 165-181.
- Grigoli, F., Herman, A., & Schmidt-Hebbel, K. (2014). World Saving. IMF Working Paper no. 14/204. Washington, D.C.: International Monetary Fund
- De Serres, Alain, and Florian Pelgrin, 2003, “The Decline in Private Saving Rates in the 1990s in OECD Countries: How Much Can Be Explained by Non-Wealth Determinants?,” OECD Economic Studies, Vol. 36, No. 4, pp. 117-153.
- Bebczuk, R., Gasparini, L., Garbero, N. and Amendolaggine, J. (2015), “Understanding the determinants of household saving: micro evidence for Latin America”, Technical Report, CEDLAS, Universidad Nacional de La Plata/IDB, La Plata.
- Callen, Tim and Christian Thimann (1997). Empirical determinants of household saving: Evidence from OECD countries. IMF Working Paper, 97/181.
- Dossche Maarten, Georgi Krustev and Stylianos Zlatanos, 2020, “COVID-19 and the increase in household savings: precautionary or forced?”, Economic Bulletin, Issue 6, ECB, 2020.
- Dynan KE, J Skinner and SP Zeldes (2004), ‘Do the Rich Save More?’, Journal of Political Economy, 112(2), pp 397–444.
- Bebczuk, R., Gasparini, L., Garbero, N. and Amendolaggine, J. (2015), “Understanding the determinants of household saving: micro evidence for Latin America”, Technical Report, CEDLAS, Universidad Nacional de La Plata/IDB, La Plata.

### Policy, pension design, and macroeconomic implications
- Amaglobeli, D, H. Chai, E. Dabla-Norris, K. Dybczak, M. Soto, and A. Tieman, The Future of Saving: The Role of Pension System Design in an Aging World, IMF Staff Discussion Note, SDN/19/01
- Feldstein, M. (1985), “The optimal level of social security benefits”, The Quarterly Journal of Economics, Vol. 100, No 2, pp. 303-320.
- Mody, A., F. Ohnsorge, and D. Sandri. 2012. “Precautionary Savings in the Great Recession.” IMF Economic Review, Vol. 60, No. 1, 114-138.
- Blanchard, O., and D. Leigh. 2013. ““Growth Forecast Errors and Fiscal Multipliers.” American Economic Review: Papers and Proceedings 103 (3): 117–120.
- Elmendorf, Douglas. 1996. “The Effect of Interest-Rate Changes on Household Saving and Consumption: A Survey.” Finance and Econ. Discussion Ser.,Working Paper no. 1996-27. Washington, D.C.: Fed. Reserve Bd.
- Muellbauer, J. 2007. “Housing, Credit and Consumer Expenditure.” Housing, Housing Finance, and Monetary Policy, a Symposium Sponsored by the Federal Reserve Bank of Kansas City, Jackson Hole, Wyoming, August 30–September 1, 267–334.
- Fisher, P.J. (2010), “Gender differences in personal saving behaviors”, Journal of Financial Counseling and Planning, Vol. 21 No. 1, pp. 14-24.

### Econometric methods and quantile/IV approaches
- Koenker, R., & Bassett, G. (1978). Regression quantiles. Econometrica, 46(1), 33-50.
- Chernozhukov, V., and C. Hansen. 2005. An IV Model of Quantile Treatment Effects. Econometrica 73(1): 245{261.
- Chernozhukov, V., and C. Hansen. 2008. Instrumental Variable Quantile Regression: A Robust Inference Approach. Journal of Econometrics 142(1): 379{398.
- Kaplan, D. M., and Y. Sun. 2017. Smoothed estimating equations for instrumental variables quantile regression. Econometric Theory 33(1): 105{157.
- Machado, J. A. F., and J. M. C. Santos Silva. 2019. Quantiles via moments. Journal of Econometrics 213(1): 145{173.

*References list from wpiea2023150-print-pdf - References*

---


_Source: https://www.imf.org/-/media/files/publications/wp/2023/english/wpiea2023150-print-pdf.pdf_