## _wp14204

## Source details

**Canonical URL:** [_wp14204](https://www.imf.org/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2014/_wp14204.pdf)

## Other formats

- [Markdown version](/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2014/_wp14204.pdf.md)
- [Structured JSON version](/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2014/_wp14204.pdf.json)

---

### I. Introduction — scope, extensions, and key empirical findings
- Research questions:
  - Determinants of private, household, corporate, and national saving rates and empirical measures to test them.
  - Whether the Global Financial Crisis changed behavioral relationships of private saving and its determinants.
  - Heterogeneity of saving determinants across country groups.
- Study extensions relative to prior literature:
  - Panel database covering 165 countries from 1981 to 2012, with a maximum of 4,137 observations.
  - Robustness across different estimation techniques.
  - Includes potential saving determinants identified by theory but not previously considered.
  - Explores differences across the 2008-10 crisis period and four country groups.
  - Examines private saving and searches for commonalities/differences across national, private, household, and corporate saving rates.
- Key summarized empirical findings:
  - Private saving rates are persistent and positively associated with income levels and income growth.
  - Terms-of-trade improvements raise saving through income effects; temporary terms-of-trade changes are saved more than permanent changes.
  - Inflation spurs saving, possibly via precautionary motives.
  - Increased credit availability depresses private saving.
  - Higher old-age dependency ratio reduces saving.
  - Urbanization lowers private saving rates.
  - Higher public saving reduces private saving (partial Ricardian offsetting).
  - Higher expected future growth and access to foreign borrowing raise private saving.
  - A higher share of young dependents reduces saving.

### II. Theoretical foundations and empirical expectations
- Theoretical models reviewed (categories preserved):
  - Permanent-Income Hypothesis (PIH) and variants including Ricardian-equivalence hypothesis (REH).
  - Life-Cycle Hypothesis (LCH).
  - Models of excess consumption sensitivity: durable goods, consumption habits, hyperbolic discounting.
  - Precautionary saving under classical and Knightian uncertainty.
  - Borrowing-constraint and buffer-stock saving models.
  - "Capitalist spirit" model.
  - Poverty/subsistence-consumption frameworks and Post-Keynesian views.
  - Household-corporate integration and “piercing the corporate veil” analogy to REH.
- Categories of determinants and expected sign notes:
  - Income: current income, temporary vs. permanent components, income growth, expected future income.
  - Wealth: total wealth, net assets, net foreign assets.
  - Rates of return: real deposit and lending rates (ambiguous overall).
  - Relative prices: CPI level, CPI inflation (current and expected), terms of trade (current and temporary/permanent), real exchange rate.
  - Risk and uncertainty: financial risk, macroeconomic instability, political risk, variance of innovations, Knightian uncertainty.
  - Borrowing constraints: domestic credit flows, foreign borrowing, current account deficits, sovereign debt premium, capital flow restrictions.
  - Financial depth: bank credit stock, financial assets, broad money.
  - Demographics: old-age dependency ratio, young-age dependency ratio, urbanization.
  - Poverty and distribution: absolute poverty measures, income concentration.
  - Fiscal policy: public saving, public budget balance, public consumption components.
  - Pension system: pay-as-you-go transfers, mandatory fully-funded contributions, fully-funded pension assets.
  - Households and firms: corporate saving effects on household saving.

### III. Data construction, descriptive statistics, and stylized facts
- Data sources and coverage:
  - IMF World Economic Outlook Database, World Bank Worldwide Development Indicators, UN National Accounts, OECD, Haver Analytics, central bank web pages.
  - Extensive data cleansing, splicing, and gap-filling; Appendix I documents details.
  - Panel is unbalanced; baseline private saving sample: 3,254 observations for 153 countries over 31 years.
  - Household and corporate saving subset: 48 countries and 674 observations.
- Definitions:
  - National saving rate = gross national saving (GNS) / gross national disposable income (GNDI).
  - Private saving rate = gross private saving (GPS) / gross private disposable income (GPDI).
- Descriptive statistics (panel sample; exact values reported in Table 2):
  - Private saving/GPDI: Obs. 153, 3,254; Mean 20.4; Std. Dev. 12.9; Min. -57.2; Max. 76.8.
  - Per capita GPDI (PPP): Obs. 153, 3,254; Mean 8,543; Std. Dev. 8,410; Min. 236; Max. 57,059.
  - Real growth rate of per capita GPDI (PPP): Obs. 153, 3,254; Mean 2.7; Std. Dev. 9.6; Min. -51.1; Max. 133.8.
  - Real deposit rate: Obs. 153, 3,254; Mean 1.0; Std. Dev. 5.3; Min. -27.5; Max. 42.7.
  - Terms of trade: Obs. 153, 3,254; Mean 106; Std. Dev. 34; Min. 42; Max. 6587.
  - Inflation: Obs. 153, 3,254; Mean 5.9; Std. Dev. 5.8; Min. -15.2; Max. 32.9.
  - Flow of private sector credit/GPDI: Obs. 153, 3,254; Mean 6.3; Std. Dev. 9.3; Min. -33.1; Max. 173.0.
  - Old age dependency ratio: Obs. 153, 3,254; Mean 11.7; Std. Dev. 7.1; Min. 0.6; Max. 39.0.
  - Share of urban population: Obs. 153, 3,254; Mean 55.0; Std. Dev. 23.9; Min. 4.7; Max. 100.0.
  - Public saving/GPDI: Obs. 153, 3,254; Mean 5.6; Std. Dev. 13.8; Min. -31.0; Max. 172.0.
  - Household saving/GPDI: Obs. 48, 674; Mean 8.9; Std. Dev. 7.3; Min. -12.3; Max. 35.1.
  - Corporate saving/GPDI: Obs. 48, 674; Mean 41.5; Std. Dev. 6.4; Min. -5.4; Max. 56.5.
- Correlation patterns (Table 3; panel correlations reported):
  - Private saving/GPDI with per capita GPDI (PPP): 0.44.
  - With real growth rate of per capita GPDI (PPP): 0.17.
  - With real deposit rate: -0.04.
  - With terms of trade: 0.06.
  - With inflation: -0.07.
  - With flow of private sector credit/GPDI: 0.12.
  - With old age dependency ratio: 0.22.
  - With share of urban population: 0.41.
  - With public saving/GPDI: 0.02.
  - With household saving/GPDI: 0.69.
  - With corporate saving/GPDI: 0.57.
- World trends and group patterns (selected exact figures):
  - Average national saving rate around 19 percent of GNDI until late 1990s; climbed to 22.6 percent of GNDI in 2006; fell to 19.5 percent in 2012.
  - Private saving component ~ four-fifths of national saving; private saving rate remained virtually constant at 16.3 percent of GNDI (sample median private saving rate stable at 20.5 percent of GPDI).
  - Advanced economies: average private saving rate of 27.2 percent of GPDI.
  - High-growth Asian economies: private saving reached 34.7 percent of GPDI by end-2012.
  - Low-income developing countries (LIDCs): average private saving rate 12.0 percent of GPDI.

### IV. Empirical strategy, identification, and estimation methods
- Dependent variable: private saving rate (Private saving/GNDI or Private saving/GPDI).
- Baseline static and dynamic modeling:
  - Baseline static model with endogenous/predetermined covariates (X_it) and strictly exogenous variables (Z_it).
  - Target adjustment equation with adjustment speed lambda and combined observable dynamic model.
  - Inclusion of country-specific heterogeneity term mu_i.
- Endogenous (or predetermined) covariates include:
  - log of real per capita GPDI (PPP);
  - real growth rate of per capita GPDI (PPP);
  - public saving in percent of GPDI;
  - inflation;
  - real deposit rate;
  - flow of private sector credit in percent of GPDI.
- Strictly exogenous variables include:
  - log of the terms of trade;
  - old-age dependency ratio;
  - share of urban population;
  - log of the real oil price.
- Estimation methods:
  - OLS on cross-section averages and pooled annual panel with heteroskedasticity and autocorrelation correction.
  - 2SLS using lags of endogenous variables as instruments.
  - Difference GMM and system GMM (S-GMM); two-step S-GMM with Windmeijer (2005) finite-sample correction is the preferred estimator.
  - Preferred final model: two-step S-GMM with time-fixed effects (equation (5)), using collapsed instrument matrix.

### V. Preferred-specification results (Two-step S-GMM with time-fixed effects — Table 4, Column 6)
- Estimation coverage and preferred model notes:
  - Number of observations across estimators ranges between 153 (average CS) and 3,254 (panel).
  - Preferred specification: Column 6 in Table 4 — two-step S-GMM with time-fixed effects on annual observations.
- Lagged dependent variable (persistence):
  - Point estimate: 0.593 (table shows 0.593*** in column 4; text reports estimated point of 0.59).
  - Interpretation: fairly high degree of persistence; long-run effect of permanent changes is 2.5 times the first-year effect.
- Core coefficient estimates (Table 4, column 6 exact reported values):
  - Ln real per capita GPDI (PPP): 0.171***.
    - Interpretation: "A one pp increase in the level of per capita GDPI raises the private saving rate by 0.17 pp."
  - Real growth rate of per capita GPDI (PPP): 0.255***.
    - Interpretation: "a one pp increase in its growth rate raises the private saving rate by 0.25 pp."
  - Ln terms of trade: 0.051***.
    - Interpretation: "A one pp improvement in the terms of trade raises the private saving rate by 0.05 pp."
  - Inflation: 0.387***.
    - Interpretation: "An increase of inflation by one pp is associated with a 0.39 pp rise in the private saving rate."
  - Flow of private sector credit/GPDI: -0.081**.
    - Interpretation: "An increase of one pp in the private credit ratio to GPDI reduces the private saving rate by 0.08 pp."
  - Old-age dependency ratio: -1.127***.
    - Interpretation: "a one pp increase in the old-age dependency ratio is associated with a reduction of the private saving rate by 1.13 pp."
  - Share of urban population: -0.391***.
    - Interpretation: "An increase of one pp in the urbanization rate reduces the private saving rate by 0.39 pp."
  - Public saving/GPDI: -0.252***.
    - Interpretation: "A rise of one pp in the public saving rate leads to a decline in the private saving rate by 0.25 pp in the same year."
    - Offset/long-run national saving effect: with lagged dependent variable, an increase of one pp in the public saving rate increases the national saving rate by 0.75 pp in the same year; Wald test rejects full offsetting with p-value 0.003.
  - Real deposit rate: coefficient not statistically different from zero in preferred specification.
- Significance summary in preferred model:
  - Out of 10 regressors, eight statistically significant at 1 percent, one at 5 percent, and one not different from zero (real deposit rate).
- S-GMM diagnostic test results (reported in Table 4):
  - AR(1) p-val (differenced residuals): 0.000.
  - AR(2) p-val: reported as 0.206, 0.318, 0.286, 0.388 across columns where applicable (no evidence of second-order autocorrelation).
  - Hansen J-test p-val: reported as 0.627, 0.753, 0.220 in S-GMM columns indicating instrument validity.

### VI. Robustness checks, additional explanatory variables, and main extensions (Table 5 and Table 6)
- Additional explanatory variables (representative coefficient patterns from Table 5):
  - Lag dependent variable examples: 0.639***, 0.599***, 0.600***, 0.591***, 0.400***, 0.621***, 0.633***, 0.662***, 0.749***, 0.592***, 0.653***.
  - Ln terms of trade examples: 0.046**, 0.051***, 0.051***, -0.0230, 0.052***.
  - Inflation examples: 0.344***, 0.390***, 0.398***, 0.383***, 0.202**.
  - Flow of private sector credit/GPDI examples: -0.058, -0.081**, -0.070**, -0.082**, -0.103***.
  - Old-age dependency ratio examples: -1.124***, -1.122***, -1.066***, -1.117***, -0.795***.
  - Share of urban population examples: -0.352***, -0.396***, -0.381***, -0.386***, -0.308***.
  - Public saving/GPDI examples: -0.345***, -0.251***, -0.227***, -0.250***, -0.344***.
- Permanent vs temporary components (HP filter, lambda = 6.25):
  - Higher permanent components of income and terms of trade are mostly saved (contrary to PIH/LCH predictions for income).
  - Higher temporary components of income have no significant effects on private saving.
  - Higher temporary components of the terms of trade are partly saved—and to a larger extent than permanent components—supporting PIH/LCH for terms-of-trade shocks.
- Other additional regressors (representative findings):
  - Expected GDP growth (5-year forecast): one pp increase raises private saving by 0.25 pp (Column 3) but causes loss of observations.
  - Conflict: coefficient not statistically different from zero in Column 4.
  - Current account balance / GPDI (endogenous): lower foreign saving by one pp associated with higher private saving by 0.54 pp (Column 5).
  - Capital account openness: fewer restrictions associated with lower private saving, coefficient not significant at conventional levels.
  - Financial system assets / GPDI: coefficient not significant.
  - Young-age dependency ratio: when income dropped, a one pp increase depresses private saving by 0.09 pp (Column 8).
  - Income distribution (Gini): coefficient not statistically different from zero.
  - Government social spending: public health/education expenditure coefficients not significant.
- Time-period and country-group differential effects (Table 6 highlights):
  - 2008-10 Global Financial Crisis:
    - Persistence fell: point estimate reduced from 0.62 to 0.46 (interaction X * Lag = -0.156***).
    - Income level effect fell: from 0.13 pp (non-crisis) to 0.07 pp (2008-10).
    - Urbanization effect declined: private saving response to one pp rise in urban share fell to -0.06 pp from -0.28 pp.
  - Advanced economies (33 countries):
    - Private saving more sensitive to growth: one pp increase in income growth raises private saving by 0.44 pp in advanced economies vs 0.26 pp in non-advanced.
    - Demographics: private saving almost non-sensitive to old-age dependency and urbanization in advanced economies (interaction X * Old-age dependency ratio = 0.941***).
  - LIDCs (47 countries):
    - Persistence lower: point estimate declines from 0.85 for non-LIDCs to 0.41 for LIDCs.
    - Growth sensitivity much lower: one pp income growth raises private saving by 0.10 pp in LIDCs vs 0.37 pp in non-LIDCs.
  - Oil exporters (20 countries):
    - Old-age dependency effect reversed: increases in old-age share raise private saving in oil exporters (positive interaction).
    - Sensitivity to public saving almost nil.
  - High-growth Asian economies (10 countries):
    - Real deposit rate sensitivity large and positive: one pp rise in real deposit rate increases private saving by 2.783** pp (interaction).

### VII. National saving and disaggregated household/corporate results
- National saving (two-step S-GMM preferred specification; sample 154 countries and 3,278 observations):
  - Preferred national saving specification similar to private saving in sign, size, and significance.
  - Difference: real deposit rate positive and significant for national saving — one pp increase raises national saving by 0.15 pp.
  - Interpretation: inclusion of public sector saving and government income effects.
- Household and corporate saving (unbalanced panel reduced to 48 countries, 674 observations; two-step S-GMM without time-fixed effects):
  - Many core private saving determinants confirmed at household level.
  - Household saving:
    - Responds significantly and positively to the real deposit rate and to inflation.
    - Not significantly affected by private credit flows.
  - Corporate saving:
    - Insensitive to the real deposit rate and to inflation.
    - Significantly lowered by private credit flows.
  - Substitution (piercing the veil):
    - Households pierce the corporate veil: coefficient offset ≈ 0.58.
    - Corporations pierce the household veil: coefficient offset ≈ 0.79.
    - Interpretation: increases in one private saving component are substantially offset by reductions in the other.
- Additional household saving findings (Table 9):
  - Household saving does not react to temporary/permanent terms-of-trade components, five-year growth forecast, or young-age dependency ratio.
  - Household saving reacts negatively and significantly to larger capital account openness.
  - Representative corporate saving/GPDI coefficients: -0.460***, -0.575***, -0.574***, -0.576***, -0.439***, -0.605***, -0.404***, -0.463***, -0.362**, -0.586***, -0.460**.

### VIII. Data construction, replacement, splicing, augmentation, and cleaning (Appendix I highlights)
- CPI and PPP handling:
  - CPI rebased to average CPI for 2000-04; Argentina uses private consumption deflator rebased to 2000-04.
  - PPP conversion factor = average PPP exchange rate observed in 2000-2004 (WEO).
- Saving series construction and sources:
  - GNS, GGS, GPS definitions and data sources: IMF WEO, OECD, UNNA, World Bank WSDB, central banks.
  - Household and corporate saving constructed via ratios enforcing adding-up constraints.
- Financial variables:
  - Real deposit rate computed via Fisher equation using CPI inflation; nominal deposit rate sources: IMF IFS, Haver Analytics, central bank pages.
  - Private sector credit flows constructed as first differences of averaged stock series and measured as share of nominal GPDI.
- Demographics and other variables:
  - Old-age dependency ratio: people older than 64 per 100 working-age (15-64) (World Bank WDI).
  - Share of urban population: World Bank WDI; Singapore and Hong Kong set to 100 percent with small stochastic noise added to avoid collinearity.
  - Real oil price: nominal oil price / US CPI × 100, then rebased to average 1981-2012 × 100 (IMF WEO and US CPI).
- Data replacement and augmentation examples:
  - Replaced WEO saving series for Spain (authorities) and Switzerland (OECD); missing saving series replaced with UNNA; augmented series via splicing with UNNA and World Bank WSDB.
  - Replaced WEO-constructed GPDI for Switzerland and Hungary with UNNA; supplemented deposit rates and private credit flows using Haver Analytics, central banks, BIS where needed.
- Cleaning steps and sample construction:
  - Initial raw dataset: 12,032 observations, 64 years, 188 countries.
  - Drop 2013 onward; restrict to years with adequate private saving coverage → sample spans 1981-2012.
  - Omit observations with public saving exactly zero or constant, or where GNS/GPS/GGS missing.
  - Exclude countries with long missing national accounts frameworks (Uzbekistan, Trinidad and Tobago).
  - Drop observations with annual inflation outside ± 50 percent and impose same restriction on real interest rate.
  - Remove specific faulty country-year observations (listed in Appendix).
  - Drop entire household/corporate series for countries with excessive variance/magnitude (list provided in Appendix).

### IX. Contributions to fitted values (Appendix III / Figure A1 key numbers)
- Private saving (percent of GPDI; full sample period contributions):
  - GPDI per capita (PPP): 24.3 percent of GPDI.
  - Terms of trade: about 24 percent of GPDI.
  - Lagged private saving rate: 12.1 percent of GPDI.
  - Largest negative contributions:
    - Share of urban population: -21.5 percent of GPDI.
    - Old-age dependency ratio: -13.2 percent of GPDI.
  - Other significant variables contribute by less than 1.1 percent of GPDI.
- National saving (percent of GNDI; key contributions):
  - Terms of trade: about 12 percent of GNDI.
  - GNDI per capita (PPP): more than 26 percent of GNDI.
  - Largest negative contributions:
    - Share of urban population: -13.6 percent of GNDI.
    - Old-age dependency ratio: -7.4 percent of GNDI.
- Household saving (percent of GPDI):
  - Terms of trade: about 19 percent of GPDI.
  - GPDI per capita (PPP): 28.3 percent of GPDI (1981-96) declining to 25.0 percent of GPDI (1997-2012).
  - Lagged household saving rate: declines from 8.1 percent of GPDI (1981-96) to 5.1 percent of GPDI (1997-2012).
  - Negative contributors: corporate saving and old-age dependency combined ≈ -17 percent of GPDI (full sample).
- Corporate saving (percent of GPDI):
  - Income per capita large positive contribution: 65.5 percent of GPDI in 1981-96, 60.4 percent of GPDI in 1997-2012.
  - Largest negative contribution: old-age dependency ratio: -13.9 percent of GPDI.
  - Other negative contributor: household saving: -7 percent of GPDI.
- Note on PPP conversion in contribution calculations:
  - Use of average PPP exchange rate to convert real GPDI/GNDI per capita inflates the income contribution and yields a large negative constant; contributions from GPDI per capita reported net of the difference between constants when using annual PPP exchange rates.

### X. Policy-relevant implications (as stated in text)
- Policies that boost income levels and income growth raise private saving rates:
  - Income level: +1 pp → private saving +0.17 pp.
  - Income growth: +1 pp → private saving +0.25 pp.
- Inflation moderation since the 1990s contributed to a decline in private saving:
  - Inflation: +1 pp → private saving +0.39 pp.
- Financial deepening via private sector credit reduces private saving:
  - Private credit ratio: +1 pp → private saving −0.08 pp.
- Population aging and urbanization lower private saving:
  - Old-age dependency: +1 pp → private saving −1.13 pp.
  - Urbanization: +1 pp → private saving −0.39 pp.
- Fiscal policy and Ricardian offsetting:
  - Public saving: +1 pp → private saving −0.25 pp in the same year; net national saving rises by 0.75 pp in the same year (partial offset; full offset rejected with p-value 0.003).
- Disaggregation recommendation:
  - Disaggregate private saving into household and corporate components to capture heterogeneous behavior and substitution dynamics; aggregate results can mask important opposing effects.

*Source: _wp14204 - Appendix I, Appendix II, Appendix III, and main empirical sections of the supplied PDF content.*

### Appendix I. Data .......................................................................................................

### Appendix I. Data

### A. Concept Construction and Variable Generation
- Section covers concept construction and variable generation (page 39).

### B. Data Replacement, Splicing, and Augmentation
- Section covers data replacement, splicing, and augmentation (page 43).

### C. Dataset Cleaning
- Section covers dataset cleaning (page 45).

### Appendix II. Country Groups
- Appendix II covers country groups (page 47).

### Appendix III. Contributions
- Appendix III covers contributions (page 48).

### Tables included in the appendix and surrounding material
- Table 1. Determinants of Private Saving in Previous Studies (page 8)
- Table 2. Descriptive Statistics (page 14)
- Table 3. Correlation Matrix of Core Private Saving Determinants (page 14)
- Table 4. Determinants of Private Saving, Different Estimators (page 23)
- Table 5. Determinants of Private Saving, Additional Explanatory Variables (page 26)
- Table 6. Determinants of Private Saving, Interactions (page 30)
- Table 7. Determinants of National Saving, Alternative Estimators (page 33)
- Table 8. Determinants of Household and Corporate Saving, Baseline Specification (page 35)
- Table 9. Determinants of Household and Corporate Saving, Additional Explanatory Variables (page 36)
- Table A1. List of Countries (page 47)

### Figures included in the appendix and surrounding material
- Figure 1. World Saving Rates, 1981-2012 (page 16)
- Figure 2. World Private Saving Rates, 1981-2012 (page 17)
- Figure 3. Pairwise Panel Correlations (page 18)
- Figure A1. Average Contributions of the Fitted Values (page 49)

*Source: _wp14204 - Appendix I. Data (PDF).*

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

### _wp14204 - References .............................................................................................................

### I. Introduction — scope and extensions
- Research questions:
  - Determinants of private, household, corporate, and national saving rates and empirical measures to test them.
  - Whether the Global Financial Crisis changed behavioral relationships of private saving and its determinants.
  - Heterogeneity of saving determinants across country groups.
- Study extensions relative to prior literature:
  - Uses a panel database covering 165 countries from 1981 to 2012, with a maximum of 4,137 observations.
  - Robustness across different estimation techniques.
  - Includes potential saving determinants identified by theory but not previously considered.
  - Explores differences across the 2008-10 crisis period and four country groups.
  - Examines private saving and searches for commonalities/differences across national, private, household, and corporate saving rates.
- Key summarized empirical findings:
  - Private saving rates are persistent and positively associated with income levels and income growth.
  - Terms-of-trade improvements raise saving through income effects; temporary terms-of-trade changes are saved more than permanent changes.
  - Inflation spurs saving, possibly via precautionary motives.
  - Increased credit availability depresses private saving.
  - Higher old-age dependency ratio reduces saving.
  - Urbanization lowers private saving rates.
  - Higher public saving reduces private saving (partial Ricardian offsetting).
  - Higher expected future growth and access to foreign borrowing raise private saving.
  - A higher share of young dependents reduces saving.
- Time- and group-specific differences:
  - During the Global Financial Crisis private saving inertia, income levels, and urbanization had diminished effects on saving.
  - Advanced economies: private saving more responsive to income growth and almost non-sensitive to demographic variables.
  - Low-income developing countries (LIDCs): lower response of private saving to income growth and less persistence of private saving rates.
  - Oil exporters: private saving positively associated with larger old-age population share.
  - High-growth Asian economies: private saving relatively more sensitive to real deposit rates.
- Robustness checks:
  - National saving regressions largely align with private saving results.
  - Separate regressions for household and corporate saving confirm many private saving findings.

### II. Private saving determinants — theoretical foundations and empirical expectations
- Theoretical models reviewed:
  - Permanent-Income Hypothesis (PIH) and variants (including Ricardian-equivalence hypothesis, REH).
  - Life-Cycle Hypothesis (LCH).
  - Models explaining excess consumption sensitivity: durable goods, consumption habits, hyperbolic discounting.
  - Precautionary saving under classical and Knightian uncertainty.
  - Borrowing-constraint and buffer-stock saving models.
  - "Capitalist spirit" model: consumption and wealth as potentially gross substitutes in utility.
  - Poverty and subsistence-consumption frameworks; Post-Keynesian views on functional income inequality.
  - Household-corporate integration and the “piercing the corporate veil” analogy to REH.
- Categories of determinants (as organized in Table 1) and theoretical expected signs (ambiguous where appropriate):
  - Income: current income, temporary vs. permanent components, income growth, expected future income.
  - Wealth: total wealth, net assets, net foreign assets.
  - Rates of return on financial assets: real deposit and lending rates; overall ambiguous effects.
  - Relative prices: CPI level, CPI inflation (current and expected), terms of trade (current and temporary/permanent), real exchange rate level and expected change.
  - Risk and uncertainty: financial risk, macroeconomic instability, political risk, variance of innovations, Knightian uncertainty.
  - Borrowing constraints: domestic credit flows/money flows, foreign borrowing, current account deficits, sovereign debt premium, capital flow restrictions.
  - Financial depth: bank credit stock, financial assets, broad money stock.
  - Demographics: old-age dependency ratio, young-age dependency ratio, urbanization.
  - Poverty and distribution: absolute poverty measures, income concentration.
  - Fiscal policy: public saving, public budget balance, public consumption and spending components.
  - Pension system: pay-as-you-go transfers, mandatory fully-funded contributions, fully-funded pension assets.
  - Households and firms: corporate saving effects on household saving.
- Empirical literature synthesis (15 panel studies):
  - Large heterogeneity across studies in sample size, coverage, data sources, saving definitions, model specifications, and estimation methods.
  - On average, individual studies include about six regressors out of 49 potential determinants listed in Table 1.
  - About half of the 49 potential regressors have not been used in any single previous private saving study.
  - Some determinants show consistent signs across studies (e.g., private saving offsets to public saving reported in 11 studies).
  - Other determinants show mixed or contradictory signs across studies (e.g., real interest rate, inflation, credit flows, old-age dependency).
  - Dispersion of point estimates and confidence intervals is large.
  - Core set of variables commonly included: income level and growth, real interest rate, inflation, terms of trade, demographics, public saving.
  - Non-standard variables (temporary/permanent components, income distribution, public consumption, pension variables) are infrequently used; financial assets, violent conflicts, and government spending components are largely absent due to data limitations.

### III. Data and stylized facts — dataset construction and descriptive statistics
- Data construction and coverage:
  - World dataset compiled from IMF World Economic Outlook Database, World Bank Worldwide Development Indicators, UN National Accounts, OECD, Haver Analytics, central bank web pages.
  - Extensive data cleansing, splicing, and gap-filling; details and series list documented in Appendix I.
  - Panel is unbalanced; sample restrictions applied for representative averages.
  - Baseline private saving sample: 3,254 observations for 153 countries over 31 years.
  - Household and corporate saving subset: 48 countries and 674 observations.
- Definitions:
  - National saving rate = gross national saving (GNS) / gross national disposable income (GNDI).
  - Private saving rate = gross private saving (GPS) / gross private disposable income (GPDI).
- Descriptive statistics (panel sample) — exact values reported in Table 2:
  - Private saving/GPDI: Obs. 153, 3,254; Mean 20.4; Std. Dev. 12.9; Min. -57.2; Max. 76.8.
  - Per capita GPDI (PPP): Obs. 153, 3,254; Mean 8,543; Std. Dev. 8,410; Min. 236; Max. 57,059.
  - Real growth rate of per capita GPDI (PPP): Obs. 153, 3,254; Mean 2.7; Std. Dev. 9.6; Min. -51.1; Max. 133.8.
  - Real deposit rate: Obs. 153, 3,254; Mean 1.0; Std. Dev. 5.3; Min. -27.5; Max. 42.7.
  - Terms of trade: Obs. 153, 3,254; Mean 106; Std. Dev. 34; Min. 42; Max. 6587.
  - Inflation: Obs. 153, 3,254; Mean 5.9; Std. Dev. 5.8; Min. -15.2; Max. 32.9.
  - Flow of private sector credit/GPDI: Obs. 153, 3,254; Mean 6.3; Std. Dev. 9.3; Min. -33.1; Max. 173.0.
  - Old age dependency ratio: Obs. 153, 3,254; Mean 11.7; Std. Dev. 7.1; Min. 0.6; Max. 39.0.
  - Share of urban population: Obs. 153, 3,254; Mean 55.0; Std. Dev. 23.9; Min. 4.7; Max. 100.0.
  - Public saving/GPDI: Obs. 153, 3,254; Mean 5.6; Std. Dev. 13.8; Min. -31.0; Max. 172.0.
  - Household saving/GPDI: Obs. 48, 674; Mean 8.9; Std. Dev. 7.3; Min. -12.3; Max. 35.1.
  - Corporate saving/GPDI: Obs. 48, 674; Mean 41.5; Std. Dev. 6.4; Min. -5.4; Max. 56.5.
- Correlation patterns (Table 3):
  - Private saving/GPDI correlations (panel sample upper triangle, cross section lower triangle) include:
    - With per capita GPDI (PPP): 0.44 (panel).
    - With real growth rate of per capita GPDI (PPP): 0.17.
    - With real deposit rate: -0.04.
    - With terms of trade: 0.06.
    - With inflation: -0.07.
    - With flow of private sector credit/GPDI: 0.12.
    - With old age dependency ratio: 0.22.
    - With share of urban population: 0.41.
    - With public saving/GPDI: 0.02.
    - With household saving/GPDI: 0.69.
    - With corporate saving/GPDI: 0.57.
- World trends and group patterns:
  - World national saving rate trends (1981-2012, Figure 1 panel (a)):
    - Average national saving rate around 19 percent of GNDI until late 1990s; climbed to 22.6 percent of GNDI in 2006; fell to 19.5 percent in 2012.
    - Private saving component ~ four-fifths of national saving.
    - Private saving rate remained virtually constant at 16.3 percent of GNDI.
    - Public saving fluctuated between 2 and 3 percent until end-1990s, peaked at 6 percent in 2006, then declined during the Great Recession.
  - Dispersion of private saving (Figure 1 panel (b)):
    - Sample median private saving rate stable at 20.5 percent of GPDI.
    - About one half of countries: private saving between 15.5 percent and 28.5 percent of GPDI.
    - About one-fifth of countries: private saving rates >35 percent or <5 percent of GPDI.
  - Private saving by country group (Figure 2 panel (a)):
    - Advanced economies: average private saving rate of 27.2 percent of GPDI.
    - Oil exporters: private saving fluctuated between 15.0 and 37.2 percent of GPDI.
    - High-growth Asian economies: private saving reached 34.7 percent of GPDI by end-2012.
    - Low-income developing countries (LIDCs): average private saving rate 12.0 percent of GPDI.
  - Private saving composition (Figure 2 panel (b)):
    - Subsample average private saving rate 16.3 percent of GPDI (subset of 48 countries), about 4 percentage points below whole-sample average.
    - Average household saving declined since mid-1990s; increasing corporate saving largely offset household decline.
- Simple cross-country associations (Figure 3):
  - Some correlations align with theory (income level, income growth, household-corporate saving).
  - Other correlations show no clear pattern or contradict theory (terms of trade, public saving rate, real deposit rate, flow of private credit, old-age dependency ratio, share of urban population).

*Source: Authors' calculations and analysis from the study's dataset and Tables/Figures reported in the content unit.*

### Appendix II reports the classification of countries.

### _wp14204 - Appendix II reports the classification of countries.

### Figures and data coverage
- Year averages are calculated on the unbalanced panel of 165 countries. (Figure 1 notes)
- Distribution percentiles use the unbalanced panel of 165 countries. (Figure 1 notes)
- Average private, household, and corporate saving rates use the unbalanced panel of 48 countries. (Figure 2 notes)
- Pairwise panel correlations use the unbalanced panel of 153 countries. (Figure 3 notes)
- Light-grey dots represent LIDCs, dark-grey dots represent advanced economies, and black dots represent the rest of the countries in the sample. (Figure 3 notes)

### Empirical strategy (modeling approach and identification)
- Dependent variable: private saving rate, denoted Private saving/GNDI or Private saving/GPDI as used in tables and text.
- Baseline static model specified as equation (1), with endogenous/predetermined covariates (X_it) and strictly exogenous variables (Z_it).
- Endogenous (or predetermined) variables treated as correlated with present, past or future error terms include:
  - log of real per capita GPDI in PPP terms
  - real growth rate of per capita GPDI in PPP terms
  - public saving in percent of GPDI
  - inflation
  - real deposit rate
  - flow of private sector credit in percent of GPDI
- Exogenous variables include:
  - log of the terms of trade
  - old-age dependency ratio
  - share of urban population
  - log of the real oil price
- Dynamics addressed with target adjustment model and dynamic panel specification:
  - Target adjustment equation (2) with adjustment speed lambda.
  - Combined observable dynamic model (3).
  - Inclusion of time-invariant unobserved country-specific heterogeneity term mu_i leads to specification (4).
- Estimation methods and issues:
  - OLS on cross-section averages and pooled annual panel with heteroskedasticity and autocorrelation correction.
  - 2SLS using lags of endogenous variables as instruments.
  - Difference GMM (Arellano and Bond, 1991) and system GMM (S-GMM) to address endogeneity and persistence.
  - Two-step S-GMM with Windmeijer (2005) finite-sample correction for the covariance matrix is the preferred estimator.
  - Final preferred model includes time-fixed effects (but excludes real oil price from strictly exogenous set) and is estimated with two-step S-GMM (equation (5)).
- Robustness checks:
  - Five-year averages specification (advantages: abstracts from business cycle, reduces measurement error; cost: loses temporal dynamics).
  - Interaction terms with dummy D_g,t for time periods or country groups in nested model (6) to test differential behavior.
- S-GMM identification tests:
  - Second-order serial correlation test for residuals (AR(2) p-val reported).
  - Hansen J-test for overidentifying restrictions reported to assess instrument validity.

### Results — estimation strategy, sample sizes, and model selection
- Estimators applied: OLS on averages (average CS), pooled OLS, OLS with fixed effects (OLS FE), 2SLS FE, and two-step S-GMM (annual and five-year averages).
- Number of observations reported across estimators ranges between 153 (average cross-section estimation) and 3,254 (panel estimation fully exploiting the longitudinal and time dimension).
- Number of countries reported: 153, 152, 153 depending on estimator/column reported in Table 4.
- Preferred specification: Column 6 in Table 4 — two-step S-GMM with time-fixed effects on annual observations.
- Goodness of fit (reported for estimators other than two-step S-GMM) ranges between 39 and 52 percent.

### Key empirical findings (preferred specification — Column 6, two-step S-GMM with time-fixed effects)
- Lagged dependent variable:
  - Point estimate: 0.593 (table shows 0.593*** in column 4 and preferred reported 0.59 in text); persistence is high — estimated point of 0.59 in text.
  - Interpretation: fairly high degree of persistence; long-run effect of permanent changes is 2.5 times the first-year effect.
- Income level and growth:
  - Ln real per capita GPDI (PPP): coefficient in Table 4 column 6 = 0.171*** (Table shows 0.171*** under column 6).
  - Text: "A one pp (percentage point) increase in the level of per capita GDPI raises the private saving rate by 0.17 pp."
  - Real growth rate of per capita GPDI (PPP): Table 4 column 6 = 0.255*** (Table shows 0.255***).
  - Text: "a one pp increase in its growth rate raises the private saving rate by 0.25 pp."
- Real deposit rate:
  - Coefficient not statistically different from zero in preferred specification; retained in baseline due to theoretical centrality.
- Terms of trade:
  - Ln terms of trade coefficient in Table 4 column 6 = 0.051***.
  - Text: "A one pp improvement in the terms of trade raises the private saving rate by 0.05 pp."
- Inflation:
  - Coefficient in Table 4 column 6 = 0.387***.
  - Text: "An increase of inflation by one pp is associated with a 0.39 pp rise in the private saving rate."
  - Policy implication in text: moderation of inflation rates observed since the 1990s contributed to a decline in private saving.
- Flow of private sector credit/GPDI:
  - Table 4 column 6 coefficient = -0.081**.
  - Text: "An increase of one pp in the private credit ratio to GPDI reduces the private saving rate by 0.08 pp."
- Old-age dependency ratio:
  - Table 4 column 6 coefficient = -1.127***.
  - Text: "a one pp increase in the old-age dependency ratio is associated with a reduction of the private saving rate by 1.13 pp."
- Share of urban population:
  - Table 4 column 6 coefficient = -0.391***.
  - Text: "An increase of one pp in the urbanization rate reduces the private saving rate by 0.39 pp."
- Public saving/GPDI:
  - Table 4 column 6 coefficient = -0.252***.
  - Text: "A rise of one pp in the public saving rate leads to a decline in the private saving rate by 0.25 pp in the same year."
  - Offset: with lagged dependent variable, an increase of one pp in the public saving rate increases the national saving rate by 0.75 pp in the same year.
  - Wald test: long-run impact of the increase in public saving rate on private saving is statistically different from one; null of complete offsetting rejected with a p-value of 0.003.
- Significance summary in preferred model (Column 6):
  - Out of 10 regressors, eight statistically significant at 1 percent, one at 5 percent, and one not different from zero (real deposit rate).
- Robustness and comparisons:
  - Results are generally robust across different estimation methodologies.
  - Some coefficient magnitudes differ from Loayza et al. (2000): income level, inflation, and old-age dependency coefficients are two or three times larger than Loayza et al. (2000); income growth and private credit flows coefficients are much smaller than those in Loayza et al. (2000).
  - Correcting standard errors for cross-sectional dependence does not affect results (authors note results available upon request).

### Inference on dynamics, identification, and specification tests
- S-GMM diagnostic test results reported in Table 4 (columns for S-GMM estimations):
  - AR(1) p-val for differenced residuals: 0.000 in reported columns (indicates expected first-order autocorrelation in differences).
  - AR(2) p-val reported as 0.206, 0.318, 0.286, 0.388 across columns where applicable (no evidence of second-order autocorrelation in differences).
  - Hansen J-test p-val reported as 0.627, 0.753, 0.220 in S-GMM columns — used to assess overidentifying restrictions and instrument validity.
- Instrumenting and lags:
  - Table 4 reports lags/instruments such as "1/19", "1/48", "-3/37" across estimator columns as indicated in table footnotes and headers.
- Finite-sample correction:
  - Two-step S-GMM uses collapsed instrument matrix and Windmeijer (2005) correction of the covariance matrix.

### Policy-relevant interpretations and implications (as stated in text)
- Policies that boost income levels and income growth are effective in raising private saving rates (quantified: level +0.17 pp; growth +0.25 pp per one pp increase).
- Moderate inflation reduction has contributed to a decline in private saving (inflation +1 pp → private saving +0.39 pp).
- Financial deepening via private sector credit reduces private saving (credit ratio +1 pp → private saving −0.08 pp).
- Population aging and urbanization are associated with lower private saving (old-age dependency +1 pp → private saving −1.13 pp; urbanization +1 pp → private saving −0.39 pp).
- Fiscal contractions raise national saving but with partial private offset: public saving +1 pp → private saving −0.25 pp in same year, net national saving +0.75 pp.

*Source: Authors’ calculations and text from Appendix II and main empirical sections of the supplied PDF content.*

### Section II, extending our preferred specification (Table 4, Column 6) by including additional

### _wp14204 - Section II, extending our preferred specification (Table 4, Column 6) by including additional

### Additional explanatory variables and key regression findings (Table 5)
- Estimation method: Two-step S-GMM; dependent variable: private saving/GPDI. Time-fixed effects included; System GMM with collapsed instrument matrix and Windmeijer (2005) correction.
- Representative coefficient results highlighted in the text:
  - Lag dependent variable (examples across columns): 0.639***, 0.599***, 0.600***, 0.591***, 0.400***, 0.621***, 0.633***, 0.662***, 0.749***, 0.592***, 0.653*** (standard errors reported in table).
  - Ln real per capita GPDI (PPP) (examples): 0.171***, 0.165***, 0.170***, 0.137***, 0.155***, 0.154***, 0.159***, 0.129***, 0.127***.
  - Real growth rate of per capita GPDI (PPP) (examples): 0.313***, 0.263***, 0.236***, 0.255***, 0.168***, 0.266***, 0.268***, 0.193***, 0.236***, 0.217***, 0.175***.
  - Ln terms of trade (examples): 0.046**, 0.051***, 0.051***, -0.0230, 0.052***, 0.043***, 0.044***, 0.037**, 0.051***, 0.049***.
  - Inflation (examples): 0.344***, 0.390***, 0.398***, 0.383***, 0.202**, 0.349**, 0.248*, 0.349*, 0.207*, 0.240*, 0.300*.
  - Flow of private sector credit/GPDI (examples): -0.058, -0.081**, -0.070**, -0.082**, 0.012, -0.103***, -0.088*, -0.067, -0.052, -0.072**, -0.069*.
  - Old-age dependency ratio (examples): -1.124***, -1.122***, -1.066***, -1.117***, -0.795***, -1.045***, -0.455***, -0.214**, -1.412***, -0.769***, -0.785***.
  - Share of urban population (examples): -0.352***, -0.396***, -0.381***, -0.386***, -0.308***, -0.340***, -0.237***, 0.079***, -0.392***, -0.263**, -0.285***.
  - Public saving/GPDI (examples): -0.345***, -0.251***, -0.227***, -0.250***, -0.344***, -0.243***, -0.205**, -0.245***, -0.162**, -0.204***, -0.151**.
- Model diagnostics (representative values):
  - AR(1) p-val.: 0.000, 0.013, 0.028 in some specifications; AR(2) p-val. generally not significant (e.g., 0.310, 0.268, 0.340).
  - Hansen J-test p-val.: e.g., 0.190, 0.690, 0.519, indicating instrument validity in reported columns.
  - Observations and country coverage examples: Observations 3,141; 3,245; 2,787; 3,073; 3,254; Number of countries examples: 144, 152, 153, 153, 110, 151.

### Permanent vs temporary components of income and terms of trade
- Method: Permanent and temporary components constructed by applying the Hodrick-Prescott (HP) filter to the (log) real GPDI per capita (PPP) and the (log) terms-of-trade index; these components were added to the baseline specification.
- Empirical findings:
  - Higher permanent components of income and terms of trade are mostly saved (contrary to PIH and LCH predictions).
  - Higher temporary components of income have no significant effects on private saving.
  - Higher temporary components of the terms of trade are partly saved—and to a larger extent than permanent components—supporting PIH and LCH for terms-of-trade shocks.
  - Comparison to prior work: Loayza et al. (2000) found no significant effect of income components on private saving and a larger positive effect of the temporary component of terms of trade compared to the permanent component.

### Other additional regressors and their effects
- Expected GDP growth (5-year forecast):
  - A one pp increase in expected GDP growth (five years ahead) raises the private saving rate by 0.25 pp (column 3).
  - Trade-off: expected-growth variable causes substantial loss of observations.
- Conflict:
  - Conflict coefficient not statistically different from zero in Column 4 (uncertain effect).
- Current account balance / GPDI (treated as endogenous):
  - Lower foreign saving by one pp (i.e., an equivalent increase in current account deficit) is associated with higher private saving by 0.54 pp (Column 5).
- Capital account openness:
  - Fewer restrictions on capital account transactions associated with lower private saving, but coefficient not significant at conventional levels.
- Financial system assets / GPDI:
  - Coefficient not significant (Column 6).
- Young-age dependency ratio:
  - Including it with income yields multicollinearity (correlation -0.83); VIF close to threshold value of 10.
  - When income is dropped (Column 8), young-age dependency ratio effect: a one pp increase depresses private saving rate by 0.09 pp.
- Income distribution (Gini):
  - Gini coefficient not statistically different from zero (Column 9).
- Government social spending:
  - Public health expenditure/GPDI and public education expenditure/GPDI coefficients are not significant (Columns 10 and 11).

### Time periods and country groups — differential effects (Table 6)
- Estimation context: Nested model with interaction terms between selected period/country-group dummies and every variable in baseline specification. Two-step S-GMM preferred model (Table 4, Column 6).
- 2008-10 Global Financial Crisis (Column 1):
  - Degree of persistence of private saving rate fell: point estimate reduced from 0.62 to 0.46 (text); table reports X * Lag private saving/GPDI = -0.156*** (interaction) with base lag ~0.621***.
  - Effect of income level on private saving fell: from 0.13 pp in non-crisis times to 0.07 pp during 2008-10.
  - Urbanization effect: private saving response to a one pp rise in share of urban population declined to -0.06 pp from -0.28 pp in non-crisis times (large reduction; rural populations relatively more affected).
- Advanced economies (Column 2, 33 countries):
  - Private saving more sensitive to growth: a one pp increase in income growth raises the private saving rate by 0.44 pp in advanced economies vs 0.26 pp in non-advanced economies (text).
  - Demographics: private saving almost non-sensitive to old-age dependency and urbanization in advanced economies (e.g., X * Old-age dependency ratio = 0.941*** interaction in Column 2 implies much smaller net negative effect).
- Low-income developing countries (LIDCs, Column 3, 47 countries):
  - Persistence lower in LIDCs: point estimate declines from 0.85 for non-LIDCs to 0.41 for LIDCs (text).
  - Growth sensitivity much lower: an increase of one pp in income growth raises private saving by 0.10 pp in LIDCs vs 0.37 pp in non-LIDCs.
- Oil exporters (Column 4, 20 countries):
  - Old-age dependency effect reversed: a one pp increase in old-age dependency lowers private saving by 0.99 pp in non-oil exporters but raises private saving by 0.58 pp in oil exporters (contradicting LCH).
  - Sensitivity of private saving to public saving is almost nil in oil exporters (contradicting REH).
- High-growth Asian economies (Column 5, 10 countries):
  - Real deposit rate sensitivity is large and positive: a one pp rise in real deposit rate increases private saving by 2.783** pp in the interaction (table) and text summarizes as 2.8 pp.
  - Inflation sensitivity appears larger but not significant at conventional levels.

### Synthesis and authors' decision on baseline specification
- Coverage of additional determinants: 13 additional potential determinants explored (four new to literature; nine considered only exceptionally before).
- Significance: Six of the additional regressors not significantly different from zero; seven significant.
- Reasons for excluding these variables from baseline specification:
  - Expected GDP growth: significant loss of observations.
  - Permanent and temporary components of income and terms of trade: approximations of unobservable variables; analytically questionable for baseline.
  - Current account balance: analytically questionable as it equals saving minus investment (endogeneity concerns).
  - Young-age dependency ratio: estimation imprecise due to very high collinearity with income variable.
- Conclusion on heterogeneity:
  - During 2008-10, global private saving exhibited less persistence, lower sensitivity to income levels, and lower sensitivity to urbanization.
  - Advanced economies: private saving more responsive to income growth and less sensitive to demographic composition.
  - LIDCs: lower persistence and lower income-growth response.
  - Oil exporters: unexpected positive sensitivity of saving to old-age share; muted response to public saving.
  - High-growth Asian economies: strong sensitivity of private saving to real deposit rates.

*Source: Authors' calculations.*

### 4. The two-step S-GMM results for five-year averages are weaker than those obtained for

### _wp14204 - 4. The two-step S-GMM results for five-year averages are weaker than those obtained for

### Two-step S-GMM results for national saving versus private saving
- Preferred specification (Column 6) for national saving is "remarkably similar" to private saving regarding sign, size, and significance of coefficient estimates.
- Moderate difference: the negative influence of demographic variables on private saving is smaller in magnitude for national saving.
- Major difference: the real deposit rate is:
  - not significant for private saving;
  - positive and significant for national saving: a one pp increase in the real deposit rate raises the national saving rate by 0.15 pp.
- Interpretation: national saving includes public sector saving; governments in most countries are net debtors with significant gross fixed-income debt, implying a positive income effect of higher interest rates on national saving.
- Sample/context numbers:
  - For the preferred specification, sample comprises 154 countries and 3,278 observations, including one additional country with 24 annual observations.
  - The authors note that about 80 percent of national saving is private.

### Household and corporate saving — sample and estimation
- Sample for disaggregated analysis:
  - Unbalanced panel reduced to 48 countries over the period 1981-2012, corresponding to 674 observations.
- Estimation method: two-step System GMM (S-GMM) with collapsed instrument matrix and Windmeijer (2005) correction; time-fixed effects not included in these S-GMM specifications to avoid instrument proliferation (log real oil price used as substitute).
- Benchmark: Column 1 reports private saving results for the restricted sample used for household and corporate regressions.

### Key empirical findings for household and corporate saving
- Core commonalities:
  - Many core private saving determinants confirmed at household level.
  - Strong evidence of significant substitution between corporate and household saving.
- Differences across components:
  - Household saving:
    - Responds significantly and positively to the real deposit rate and to inflation.
    - Not significantly affected by private credit flows.
  - Corporate saving:
    - Insensitive to the real deposit rate and to inflation.
    - Significantly lowered by private credit flows.
  - Urban population share and public saving reduce national saving but do not significantly affect household or corporate saving separately.
- Substitution (piercing the veil):
  - Households pierce the corporate veil: coefficient offset ≈ 0.58.
  - Corporations pierce the household veil: coefficient offset ≈ 0.79.
  - Interpretation: Much of an increase in one private saving component is offset by a reduction in the other; corporations pierce the household veil to an even larger degree than households pierce the corporate veil.
- Additional sample/estimation diagnostics reported for S-GMM estimations (selected):
  - Observations: 674
  - Number of countries: 48

### Additional explanatory variables for household saving (Table 9 findings)
- The baseline regressors for household saving are generally robust to inclusion of 13 additional saving regressors.
- Specific distinctions relative to private saving:
  - Household saving does not react to:
    - temporary and permanent components of the terms of trade (nor to total terms of trade),
    - the five-year growth forecast,
    - the young-age dependency ratio.
  - Household saving reacts negatively and significantly to larger capital account openness (in contrast to private saving).
- Example coefficient mentions (from Table 9):
  - Corporate saving/GPDI coefficients in various specifications reported as -0.460***, -0.575***, -0.574***, -0.576***, -0.439***, -0.605***, -0.404***, -0.463***, -0.362**, -0.586***, -0.460** (standard errors shown in the source).
- Sample and coverage details in Table 9:
  - Observations: ranges reported (e.g., 645, 674, 630, 674, 644, 674, 674, 674, 512, 651, 616)
  - Number of countries: ranges reported (e.g., 45, 48, 48, 48, 45, 48, 48, 41, 48, 46)

### Overall conclusions and methodological notes
- Database and scope:
  - The paper exploits a large panel database covering 163 countries from 1981 to 2012 in the main analysis.
  - Appendix I describes the final dataset: 4,137 observations, spanning 31 years for 165 countries.
- Main empirical summary (baseline private saving specification):
  - Of 10 private saving regressors consistent with theory, nine are statistically significant in the authors' preferred estimation: the lagged dependent variable, the real income level, real income growth, the terms of trade, inflation, private credit flows, old-age dependency, urban population, and public saving (partial Ricardian offsetting).
  - The real deposit rate is the only core regressor not significantly different from zero in the baseline private saving regression.
- Extensions and additional determinants:
  - When including 13 additional potential determinants, seven are significant: the permanent component of income, the permanent and temporary components of the terms of trade, future growth forecasts, the current account balance, capital account openness, and young-age dependency.
- Heterogeneity across groups/periods:
  - Private saving rates evolved heterogeneously across regions (e.g., booming in high-growth Asia; volatile in oil exporters).
  - During the 2008-10 crisis period private saving rates were less persistent and less sensitive to income levels and urbanization.
  - Cross-group differences include:
    - Advanced economies: private saving more responsive to income growth and almost non-sensitive to demographic variables.
    - LIDCs: lower response of private saving to income growth and less persistence.
    - Oil exporters: sensitivity of private saving to old-age population share is positive (contradicting LCH).
    - High-growth Asian economies: private saving rates highly sensitive to real deposit rates.
- Recommendation/implication:
  - It is important to disaggregate private saving into household and corporate components to gather evidence on heterogeneous saving behavior; aggregate results can mask important differences and substitution dynamics.

*Source: Authors' calculations, as reported in the supplied content unit.*

### 1. Price and Inflation Variables

### 1. Price and Inflation Variables

### Consumer Price Index (CPI) and Rebasing
- CPI data source: IMF World Economic Outlook (WEO) database.
- To maintain cross-country comparability when calculating real variables, each country’s CPI index is rebased with the average CPI for 2000-04.
- Note on Argentina: use the private consumption deflator from the national accounts and rebase it using the average for 2000-04.

### CPI-based inflation
- Inflation calculated as the growth rate of the CPI, measured in percent.

### Bounded CPI-based inflation indicator
- Indicator derived from the intertemporal consumption optimization problem in discrete time.
- Indicator = (inflation rate) / (1 + inflation rate) with inflation rates expressed in decimals (not percent).
- Indicator is bounded between zero and one.

### Purchasing Power Parity (PPP) Index
- PPP conversion factor varies across countries but not across time.
- For each country, factor = average PPP exchange rate observed in 2000-2004 (from WEO database).

---

### 2. Saving Variables

### Definitions and sources
- GNS: gross national disposable income less final consumption expenditure, plus adjustment for change in equity positions in pension funds. Data collected from IMF WEO, OECD, UNNA, World Bank WSDB, and several central banks.
- Gross government saving (GGS): gross disposable income of the general government less general government final consumption expenditure. Same data sources as GNS.
- GPS: GNS less GGS.
- Household saving: portion of GPS corresponding to households. National, government, private, and household saving data from UNNA and OECD.
  - Derived ratio: share of household saving to GNS, all over one minus the share of GGS to GNS. This transformation enforces the adding-up constraint that household + corporate saving = private saving.
  - Constructed ratio applied to GPS to obtain household saving in local currency.
- Corporate saving: portion of GPS from financial corporations, non-financial corporations, and non-profit institutions serving households.
  - Derived ratio: (one minus share of GGS to GNS minus share of household saving to GNS) all over one minus share of GGS to GNS. Corporate saving ratio is the residual of GPS minus household saving ratio.
  - Ratio applied to GPS to obtain corporate saving in local currency.

---

### 3. Income Variables

### GPDI and real GPDI per capita (PPP terms)
- GPDI: GPS plus private consumption (WEO), measured in local currency.
- Real GPDI per capita in PPP terms:
  - Deflate nominal GPDI by rebased CPI deflator.
  - Divide by rebased PPP exchange rate.
  - Divide by total population (World Bank WDI).

### Terms-of-trade index
- Defined as (deflator of exports of goods and services) / (deflator of imports of goods and services) × 100.
- All indexes re-scaled such that 2000 = 100.
- Sources: IMF WEO and World Bank WDI.

---

### 4. Financial Variables

### Real deposit rate
- Calculated using the Fisher equation and CPI inflation.
- Nominal deposit rate data sources: IMF IFS, Haver Analytics, central bank web pages.

### Flow of private sector credit to GPDI
- Private sector credit as share of nominal GDP from World Bank WDI.
- Converted to stocks in local currency.
- Private sector credit flows = first difference of average of current year and previous year stocks (current year prices), measured as share of nominal GPDI.
- Additional country data from BIS, IMF IFS, and central bank web pages when not found in WDI.

---

### 5. Demographic Variables

### Old-age dependency ratio
- Ratio of people older than 64 to working age population (ages 15-64), expressed as elderly dependents per 100 people of working age.
- Source: World Bank WDI.

### Share of urban population
- Number of people in urban areas divided by mid-year total population × 100.
- Source: World Bank WDI.
- Note: Singapore and Hong Kong share of urban population constant at 100 percent in-sample; small stochastic white noise is added to avoid collinearity with country fixed effects.

---

### 6. World Oil Price

### Real oil price
- Nominal oil price divided by US CPI × 100, then rebased as (real price at time t) / (average real price between 1981 and 2012) × 100.
- Data sources: IMF WEO for oil price and US CPI.

---

### 7. Additional Explanatory Variables

### Permanent and temporary components of terms of trade
- Construct permanent component by applying HP filter to terms-of-trade series and WEO projections through 2019.
- HP filter smoothing parameter lambda = 6.25 (recommended for annual data).
- Permanent component = trend from HP filter; temporary component = cycle (actual minus trend).

### Permanent and temporary components of income
- Apply same HP filter procedure (lambda = 6.25) to GPDI and its components (with WEO projections through 2019).
- Temporary component = actual GPDI minus trend component.

### Five-year forecast of real GDP growth
- Use archived WEO vintages (Fall versions 1990–2013) to construct five-year-ahead real GDP growth forecasts.
- Denote vintage publication year as year t; the five-year-ahead growth forecast is taken for each country and year.

### Other explanatory variables (definitions and sources)
- Conflict: dummy = 1 when battle-related deaths > 1,000, else 0 (World Bank WDI).
- Current account balance to GPDI: WEO current account balance in U.S. dollars transformed to local currency using period-average exchange rates, then measured as share of nominal GPDI.
- Capital account openness: Quinn Index (0–100).
- Financial system assets to GPDI: deposit money bank assets as share of nominal GPDI using World Bank GFDD and central banks.
- Young-age dependency ratio: share of people < 15 to working age population (15–64), percent (World Bank WDI).
- Gini index: from Standardized World Income Inequality Database, ranges 0–100.
- Public health and education expenditure to GPDI: sum of public health and education expenditure; series retrieved as shares of nominal GDP from World Bank WDI and IMF Fiscal Monitor Database, then converted to ratios to nominal GPDI.
- Real world GDP growth: rate of growth from IMF WEO aggregated real GDP index.

---

### Data Replacement, Splicing, Augmentation, and Cleaning

### Data replacement and augmentation highlights
- Replaced WEO saving series for Spain (country authorities) and Switzerland (OECD).
- Missing saving series replaced with UNNA; extended GNS via splicing with UNNA and World Bank WSDB.
- Applied UN and OECD saving ratios to augment private/public, household/corporate saving series; prefer OECD-derived household and corporate saving when both available.
- Replaced WEO-constructed GPDI with UNNA for Switzerland and Hungary; replaced GDP and included missing GNDI for Switzerland using OECD.
- Missing nominal deposit rates replaced with nominal outstanding deposit rates; U.S. deposit series spliced using Haver Analytics and Federal Funds rate.
- Augmented deposit rate data for Norway, Sweden, Poland using Haver; splicing for UK, EU members, Iceland, Colombia, El Salvador, Iran, Iraq, Kuwait, Saudi Arabia, UAE, India, Burundi.
- Private sector credit flow gaps supplemented using central bank/regulator web pages for Austria, Ireland, Australia, New Zealand, South Africa, Brazil, Colombia, Mexico, Hong Kong, China.
- Deposit money bank assets series extended by proxing with liabilities and using percent changes from World Bank GFDD; replaced entire series for Chile and supplemented Hong Kong and Venezuela with central bank data.
- For Singapore and Hong Kong urban share, added small stochastic noise to avoid collinearity.

### Interpolation and initial dataset
- Gaps of two missing observations or less filled by interpolating averages.
- Resulting initial raw dataset: 12,032 observations, spanning 64 years for 188 countries.

### Dataset cleaning steps
- Drop data from 2013 onward.
- Drop years in which more than 50 percent of the raw dataset contains missing private saving rates; resulting sample spans 1981 to 2012.
- Omit observations where public saving is exactly zero or constant as share of GDP.
- Exclude observations where GNS, GPS, or GGS are missing.
- Drop countries lacking national accounts framework for long periods (Uzbekistan and Trinidad and Tobago).
- Drop country-year observations where observed annual inflation rates fall outside the ± 50 percent range; impose same restriction on the real interest rate.
- Graphical inspection of faulty GNS/GPS/GGS as shares of GNDI resulted in removal of 11 country-year observations: Cabo Verde 1981; Kuwait 1991; Chad 2002; Equatorial Guinea 1981, 1989, 1990, 1997, 1999; Republic of Congo 1987-1989.
- In several cases, household and corporate saving series with excessive variance or magnitudes are dropped entirely. Series dropped completely include: Luxembourg, Venezuela, Bahrain, Kuwait, Botswana, Guinea-Bissau, Côte d'Ivoire, Burkina Faso, Armenia, Azerbaijan, Kazakhstan, Kyrgyz Republic, Saudi Arabia, and Moldova. Observations for Bulgaria 2007 and Romania 1981-1997 are also omitted.

*Source: _wp14204 - 1. Price and Inflation Variables*

### Appendix III. Contributions

### Appendix III. Contributions

### Overview
- Figure A1 shows contributions to the fitted values of the significant variables in the preferred specifications for:
  - private saving (Table 4, Column 6),
  - national saving (Table 7, Column 6),
  - household saving (Table 8, Column 3),
  - corporate saving (Table 8, Column 5).
- Contributions are averaged for sample sub-periods 1981-96, 1997-2012, and 1981-2012.
- Charts highlight relative importance of explanatory variables and changes over time.
- Only statistically significant contributions are reported.

### Private and National Saving: Key Contributions and Drags
- Private saving (percent of GPDI):
  - GPDI per capita (PPP): 24.3 percent of GPDI (full sample period).
  - Terms of trade: about 24 percent of GPDI.
  - Lagged private saving rate: 12.1 percent of GPDI.
  - Largest negative contributions:
    - Share of urban population: -21.5 percent of GPDI.
    - Old-age dependency ratio: -13.2 percent of GPDI.
  - Other significant variables contribute by less than 1.1 percent of GPDI.
- National saving (percent of GNDI):
  - Terms of trade: about 12 percent of GNDI.
  - National saving scaled by GNDI: about 12 percent of GNDI.
  - GNDI per capita (PPP): more than 26 percent of GNDI.
  - Largest negative contributions:
    - Share of urban population: -13.6 percent of GNDI.
    - Old-age dependency ratio: -7.4 percent of GNDI.

### Household and Corporate Saving: Key Contributions and Drags
- Household saving (percent of GPDI):
  - Positive contributors:
    - Terms of trade: about 19 percent of GPDI (virtually constant across the two time periods).
    - GPDI per capita (PPP): 28.3 percent of GPDI (1981-96) declining to 25.0 percent of GPDI (1997-2012) for the contribution of income level.
    - Lagged household saving rate: declines from 8.1 percent of GPDI (1981-96) to 5.1 percent of GPDI (1997-2012).
  - Negative contributors:
    - Corporate saving and old-age dependency ratio combined: about -17 percent of GPDI (full sample period).
- Corporate saving (percent of GPDI):
  - Income per capita has a very large positive contribution reflecting a large coefficient, but declines over time:
    - 65.5 percent of GPDI in 1981-96.
    - 60.4 percent of GPDI in 1997-2012.
  - Largest negative contribution:
    - Old-age dependency ratio: -13.9 percent of GPDI.
  - Other negative contributor:
    - Household saving: -7 percent of GPDI.

### Figure A1 and Estimation Notes
- Panels in Figure A1:
  - (a) Private saving — coefficients from Table 4, Column 6.
  - (b) National saving — coefficients from Table 7, Column 6.
  - (c) Household saving — coefficients from Table 8, Column 3.
  - (d) Corporate saving — coefficients from Table 8, Column 5.
- Note on PPP conversion and contribution calculation:
  - Use of the average PPP exchange rate to convert real GPDI/GNDI per capita in national currency into PPP yields an inflated contribution from GPDI per capita and a large and negative constant.
  - Given that contributions from GPDI per capita and the constant term have a considerably smaller size in regressions where real GPDI/GNDI per capita is converted in PPP terms by using the PPP exchange rate of every year (instead of the average PPP exchange rate), the contribution of log of real GPDI/GNDI per capita (PPP) is calculated net of the difference between the constants in these two regressions.

*Source: Authors' calculations.*

---


_Source: https://www.imf.org/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2014/_wp14204.pdf_