## 1. Household Saving Panel Regression, 1990–2008 (OECD)

## Source details

**Canonical URL:** [1. Household Saving Panel Regression, 1990–2008 (OECD)](https://www.imf.org/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2010/_wp1069.pdf)

## Other formats

- [Markdown version](/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2010/_wp1069.pdf.md)
- [Structured JSON version](/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2010/_wp1069.pdf.json)

---

### I. Purpose and approach
- Examines likely impact of expanding social programs on household consumption in China.
- Three channels through which higher government social spending can affect household consumption:
  - (i) household age-specific propensities to consume out of (lifetime) disposable income;
  - (ii) distribution of household disposable income across income groups (with different propensities to consume);
  - (iii) overall level of household disposable income.
- Empirical strategy:
  - Use household income survey data and a generational accounting framework to estimate age-specific marginal propensities to consume for different income groups and lifetime resources by cohort.
  - Simulate aggregate consumption effects of alternative government social expenditure reforms.

### II. Key empirical findings on China’s consumption and savings
- Household consumption share of GDP:
  - 37 percent in 2008, down from about 55 percent in 1981.
- Household savings and income ratios:
  - Average household savings rate out of disposable income rose from 11 percent in 1990 to 25 percent in 2007.
  - The 2007 rate is 12 percentage points higher than the Asian average.
  - Comparative 2007 figures: EU average household savings rate ~11 percent; US below 2 percent.
  - Household disposable income as a share of GDP: about 54 percent; declined by about 8 percentage points between 1990 and 2007.
- Contribution to decline in household consumption ratio (1990–2007):
  - Increase in household savings rate accounts for about 9 percentage points of the ~13 percentage points decline in the household consumption ratio.
  - Remaining decline explained by fall in share of household disposable income in GDP.

### III. Determinants and mechanisms behind high savings / low consumption
- Structural and policy drivers:
  - China’s growth model and high level of precautionary savings.
  - Inadequate social protection for health and old age.
  - Elevated private cost of higher education.
  - Demographic trends and inadequate access to credit.
- Labor and income-side factors:
  - Weak wage growth due to high internal migration and a sizeable under- or unemployed share of labor.
  - Absence of effective unions and some monopsonistic employer power.
  - Limited redistribution of firms’ profits (investment income languished due to virtual absence of profit redistribution to the public, limited number of publicly listed firms, SOEs not paying dividends).
- SOE reforms and welfare provision:
  - 1990s SOE reforms reduced coverage of work-unit “welfare,” shifting health and education burdens to households, lowering lifetime incomes and increasing perceived income/expenditure risk.
- Life-cycle vs. precautionary saving motives:
  - Life-cycle: consumption/saving depend on lifetime resources and demographics; expected future income growth reduces current savings; higher working-to-nonworking ratio can increase savings.
  - Precautionary: higher income/expenditure risks raise savings; consistent with increasing savings rate with age.
- Financial development:
  - Financial underdevelopment, borrowing constraints, and low returns on financial assets amplified precautionary motives.

### IV. Empirical evidence on age and risk profiles
- Prior studies (selected findings):
  - Kraay (2000): expectations of future income growth and subsistence consumption affect rural saving; no significant relationship for urban saving.
  - Modigliani and Cao (2004): savings rate increases with share of working-age population (life-cycle).
  - Horioka and Wan (2006): demographic factors, lagged savings, income growth, interest rate, and sometimes inflation determine savings.
  - Chamon and Prasad (2008): U-shaped age profile of savings; households facing high health expenditure risk have savings rates 20 percentage points higher; households with small children save up to 5 percentage points more.
  - Wei and Zhang (2009): half of increase in household savings explained by increasing share of males post one-child policy (males save to accumulate assets for marriage market).
  - Barnett and Brooks (2010): a one yuan increase in government health spending is associated with a two yuan increase in urban household consumption.
- Age-risk patterns: younger and older households save relatively more; health and education risks are important drivers.

### V. Cross-country panel evidence (24 OECD countries, 1990–2008)
- Baseline regression: Saving_it = α + X_it’β + γZ_it + ν_i + ε_it
  - Saving expressed as percent of disposable income.
  - X_it includes: Real per capita GDP and its growth; old-age dependency ratio (65+/15–64); young-age dependency ratio (under 15/15–64); financial development measured by credit to private sector (percent of GDP).
  - Z_it is social spending (percent of GDP): public health, education (primary, secondary, higher), social security (pensions and social assistance).
  - Estimation methods: fixed-effects panel and dynamic panel GMM (Arellano-Bond); data are 5-year averages.

- Key empirical patterns:
  - Higher government social spending is generally associated with lower household saving.
  - Relationship is non-linear: marginal reduction in saving from increased social spending is largest when social spending (percent of GDP) is low; marginal decline shrinks as spending rises.
  - Public health spending has largest negative impact on household saving; social security has a smaller but significant effect; education shows sizable negative impact mainly when evaluated separately.

- Selected regression coefficients and diagnostics (Column references as in source):
  - Household saving (lagged): -0.41* (t-stat -1.73) in Column (1).
  - Growth, per capita GDP: -0.40** (t-stat -2.19) in Column (1).
  - Public social spending, total: -1.95*** (t-stat -4.17) in Column (1).
  - Public social spending total, squared: 0.03*** (t-stat 4.14) in Column (1).
  - Public health spending: -6.84*** (t-stat -3.45) in Column (3).
  - Public health spending, squared: 0.44*** (t-stat 4.09) in Column (3).
  - Social protection spending: -2.15* (t-stat -1.81) in Column (3).
  - Arellano-Bond AR(2) p-value: 0.47.
  - Hansen Test p-value: 0.95.
  - No. of Instruments: 14.
  - No. of Obs.: 78 (Column 1), 68 (Column 2), 78 (Column 3), 74 (Column 4).
  - No. of countries: 24 (Columns 1, 3, 4); 20 (Column 2).
  - R^2: 0.76 (Column 1), 0.8 (Columns 2 and 3).
  - Significance levels: *** 1 percent, ** 5 percent, * 10 percent.

- Non-linear (U-shaped) relationship for total social spending:
  - Minimum household saving reached when total social spending is around 31–36 percent of GDP.
  - Sample mean social spending: 28.1 percent of GDP.
  - Illustrative marginal effects at OECD average:
    - At total social spending 28.1 percent of GDP, a 1 percent of GDP increase in total spending leads to a fall in household saving of 0.26–0.45 percent of household disposable income (Columns 1, 2, and 4). Because average household disposable income is about 54 percent of GDP, this implies a reduction in household saving of 0.14–0.24 percent of GDP.
    - Simultaneous increase in each of the three components (1/3 each) yields similar impact ~0.13 percent of GDP on household savings (Column 3).

- Component-specific impacts (evaluated at sample means):
  - Public health spending (sample mean 6.3 percent of GDP): household saving falls by 0.70–0.78 percent of GDP in response to a 1 percent of GDP increase in health expenditure (Column 3 and Appendix). Health coefficient ~2.1 in regressions (consistent with Barnett and Brooks 2010 for urban China).
  - Public education spending (sample mean 5.8 percent of GDP): a 1 percent of GDP increase in public education spending leads to a decline in household saving by 0.79 percent of GDP (significant mainly when estimated separately).
  - Social protection / social security spending (sample mean 16.1 percent of GDP): a 1 percent of GDP increase reduces household saving by 0.22–0.29 percent of GDP (Column 3).

- Marginal reduction in household saving (As percent of GDP) of a 1 Percent of GDP Increase in Government Expenditure (Table 2 figures):
  - Total Social Spending (OECD average at 28.1 percent of GDP): 0.14 ~ 0.24
  - Health (OECD, at 6.3 percent of GDP): 0.70 ~ 0.78
  - Education (OECD, at 5.8 percent of GDP): 0.79
  - Social Security (OECD, at 16.1 percent of GDP): 0.22 ~ 0.29
  - Marginal reduction in China (measured at current Chinese levels):
    - Total social spending (at 6 percent of GDP): 0.56 ~ 1.03
    - Health (at 0.9 percent of GDP): 2.09 ~ 2.12
    - Education (at 2.9 percent of GDP): 1.26
    - Social security (at 2.2 percent of GDP): 0.68 ~ 0.72

### VI. Methodology for China-specific simulations (Boxes 2 and 3)
- Consumption identity for group i at time t: cit = αi rit, where αi is average propensity to consume out of lifetime resources and rit is net present value of group’s remaining lifetime resources.
- Decomposition of changes in total consumption into:
  - Changes in total lifetime household income,
  - Changes in distribution of lifetime income across groups,
  - Changes in group propensities to consume (insurance effect),
  - Changes in population distribution across groups (demographic effect; kept fixed in baseline).
- Implementation steps:
  1. Calculate lifetime resources for socio-economic groups using CHIP 2002 data.
  2. Estimate group average propensities to consume (αi) from current consumption / lifetime income.
  3. Translate increases in public expenditures into changes in lifetime resources and compute current consumption impact as change in group NPV × group propensity.
- Data details (CHIP 2002):
  - Urban: 12 provinces, 6,800 households, 20,600 individuals.
  - Rural: 22 provinces, 9,200 households, 38,000 individuals.
  - Migrant households in urban provinces: 2,000 households, 5,300 individuals.
- Lifetime income NPV assumptions:
  - Future wages follow age-specific wage profiles augmented by growth rate of 7.7 percent.
  - Discounted based on interest rate of 11.7 percent and survival rates from 2006 WHO Life Table for China, giving a discount rate of 0.96.
  - Lifetime resources include present value of net transfers, net pension benefits, in-kind education and health benefits, and financial assets.
- Propensity patterns:
  - Propensities increase with age and are higher for rural and lower-income groups.
- Simulated reforms:
  - Pension transfer: universal cash transfer to all individuals over 55 (interpreted as large expansion of coverage; McKinsey (2009): 2009 pension coverage ~90 percent urban, 20-25 percent rural).
  - Education transfer: proportional to current education expenditures by age profile.
  - Health transfer: proportional to current health expenditures by age profile.
  - Each reform is permanent and financed through a permanent decrease in fiscal surplus (deficit-financed in simulations); budget for each reform = 1 percent of GDP annually.

### VII. Simulation results — income, insurance, and total effects
- Income effect (Simulation 1: 1 percent of GDP annual increase in each category) — impact on current household consumption (percent of GDP):
  - Pension: Total 1.42
    - Urban 0.92
    - Rural 0.50
  - Health: Total 0.77
    - Urban 0.46
    - Rural 0.32
  - Education: Total 0.51
    - Urban 0.24
    - Rural 0.27
- Budget shares by area:
  - Pension: Urban 0.75, Rural 0.25
  - Health: Urban 0.69, Rural 0.31
  - Education: Urban 0.58, Rural 0.42
- Rural impacts per unit of spending are substantially higher due to higher propensities and lower lifetime budgets (examples: pension impact rural/urban differential ~67 percent higher; health 56 percent; education 55 percent).
- Alternative scenario (smaller lifetime budget, Simulation 2):
  - Income effects lower: pensions 0.8 percent, health 0.5 percent, education 0.4 percent of GDP.
- Insurance effect:
  - Taiwan evidence (Chou, Liu, Hammitt (2006)): health insurance expansion (coverage 57 percent in 1994 to 96 percent in 2000) implies an increase in health expenditures of 1 percentage point of GDP increased current household consumption by 0.4–0.6 percent of GDP via the insurance effect.
  - Paper adopts ratio = 0.24 (0.5 / 2.1) of health insurance impact to total health impact and applies to education and pensions to obtain insurance impacts:
    - Education insurance impact = 0.31 percent of GDP.
    - Pension insurance impact = 0.17 percent of GDP.
- Total effects (income effect + insurance effect) — headline Table 4 impacts (In percent of GDP) for a 1 percent of GDP annual increase:
  - Pension: Total 1.6; Income Effect 1.4; Insurance Effect 0.2
  - Health: Total 1.3; Income Effect 0.8; Insurance Effect 0.5
  - Education: Total 0.8; Income Effect 0.5; Insurance Effect 0.3
- Composite and medium-term implications:
  - A 1 percentage point of GDP increase in social expenditures allocated evenly across pension, health, and education would increase household consumption permanently by 1.2 percent of GDP.
  - A sustained 1 percentage point of GDP increase in government spending is likely to lead to an increase in the household consumption ratio of up to 1¼ percentage points of GDP.
  - To raise household consumption by 3 percentage points of GDP:
    - Evenly distributed social expenditure increase scenario: requires 2.5 percent of GDP increase in total social expenditures maintained over medium term.
    - Targeted scenarios: requires 1.9 percent increase in pension expenditures or 2.2 percent increase in health expenditures.
  - Complementary structural reform:
    - A simulated 1 percent additional growth in labor income generates a 0.7 percent increase in current household consumption.

### VIII. Financing scenarios and distributional effects
- If the 1 percent of GDP increase in expenditures is financed by an increase in income taxes, financing would offset part of consumption gains: financing would lead to an offsetting decrease in consumption of nearly 0.6 percent of GDP.
  - Net increases in household consumption under income-tax financing:
    - Education: 0.2 percent of GDP
    - Health: 0.7 percent of GDP
    - Pensions: 1.0 percent of GDP
- Interpretation: tax-financed reforms yield smaller immediate net impacts because of redistribution from groups with high propensities to consume to groups with lower propensities.

### IX. Policy implications and recommendations
- Increasing government social expenditures on health, education, and pensions can raise household consumption and reduce precautionary saving.
- Cost-effectiveness and targeting:
  - Target a higher proportion of expenditure increases to health and pensions, or to rural and low-income households, to achieve larger consumption impacts for a given budget increase.
- Complementary reforms:
  - Combine expenditure reforms with structural measures to raise share of wages in national income and expand domestic consumption (credit availability, retail distribution reforms).
- Financing considerations:
  - Deficit-financed increases produce larger immediate consumption impacts than tax-financed equivalents, though tax-financed reforms still yield positive net impacts through redistribution.
- Contribution to external rebalancing:
  - Household consumption would have to increase by some 3 to 4 percentage points of GDP, assuming corporate and government savings remain unchanged, to help rebalance world demand and address China’s large external current account surplus.
  - Greater public provision of health, education, and pensions could make an important contribution alongside other structural reforms.

### X. Supporting quantitative details (selected)
- Cross-country panel coefficients (selected):
  - Public health spending coefficient: -6.74***; public health spending, squared: 0.42*** (FE specification A).
  - Public education spending coefficient: -5.76***; education spending, squared: 0.37*** (FE specification B).
  - Social security spending coefficient: -2.31**; squared: 0.06** (FE specification C).
  - Sample: No. of Obs. 78; No. of countries 24; R^2: 0.78 (A), 0.72 (B), 0.71 (C).
- Appendix household consumption pattern (average shares):
  - Urban: Food 0.364; Non-food 0.326; Housing 0.169; Education 0.074; Health 0.066.
  - Rural: Food 0.428; Non-food 0.282; Housing 0.160; Education 0.083; Health 0.047.

*Source: _wp1069*

### 1. Household Saving Panel Regression, 1990–2008 (OECD) .................................................10

### 1. Household Saving Panel Regression, 1990–2008 (OECD) .................................................10

### I. Introduction — purpose and approach
- Examines the likely impact of expanding social programs on household consumption in China.
- Identifies three channels through which higher government social spending can impact household consumption:
  - (i) household age-specific propensities to consume out of (lifetime) disposable income;
  - (ii) the distribution of household disposable income across different income groups (with different propensities to consume);
  - (iii) the overall level of household disposable income.
- Uses household income survey data and a generational accounting framework to estimate:
  - age-specific marginal propensities to consume for different income groups;
  - lifetime amount of resources available to each cohort.
- Simulates effects on aggregate consumption of alternative government social expenditure reforms.

### II. Key findings on consumption and savings in China
- Household consumption as a share of GDP:
  - 37 percent in 2008, down from about 55 percent in 1981.
  - China ranks at the bottom in the Asian region and among emerging markets in terms of household consumption share.
- Household savings rate:
  - Average household savings rate out of disposable income rose from 11 percent in 1990 to 25 percent in 2007.
  - The 2007 rate is 12 percentage points higher than the Asian average.
  - Comparative 2007 figures: average household savings rate of EU countries was about 11 percent; in the US it was below 2 percent.
- Household disposable income as a share of GDP:
  - About 54 percent.
  - Declined by about 8 percentage points between 1990 and 2007.
- Contribution to decline in household consumption ratio (1990–2007):
  - Increase in household savings rate accounts for about 9 percentage points of the approximately 13 percentage points of GDP decline in the household consumption ratio.
  - Remaining decline explained by fall in the share of household disposable income in GDP.

### III. Determinants and mechanisms behind high savings and low consumption
- Structural and policy-related drivers highlighted:
  - China’s specific growth model and a high level of precautionary savings.
  - Inadequate social protection programs relating to health and old age.
  - Elevated private cost of higher education.
  - Demographic trends and inadequate access to credit for a significant share of the population.
- Labor and income-side factors:
  - Weak wage growth due to high internal migration maintaining a high labor supply and a sizeable under- or unemployed share of labor.
  - Absence of effective union organizations and some degree of monopsonistic power of employers.
  - Limited redistribution of firms’ profits: investment income has languished because of the virtual absence of profit redistribution to the public, limited number of publicly listed firms, and tendency of State-Owned Enterprises (SOEs) not to pay dividends to the government.
- Role of SOE reforms:
  - Reform of SOEs at the beginning of the 1990s substantially reduced coverage of the effective “welfare” state previously provided by work units.
  - Shifted health and education expenditures burden to the private sector, effectively reducing households’ lifetime incomes and increasing perceived income and expenditure risk.
  - Increased risk of significant health or education expenditures likely contributed to the rise in the savings rate.
- Life-cycle versus precautionary savings:
  - Life-cycle hypothesis: consumption and saving depend on lifetime resources and demographic structure; expected future growth of income reduces current savings, while an increase in the ratio of working to nonworking population can increase savings.
  - Precautionary savings explanation: higher income and expenditure risks increase savings; consistent with an increasing savings rate with age.
- Financial development:
  - Financial underdevelopment, constraints on borrowing against future income, and low returns on financial assets may have amplified precautionary savings motives.

### IV. Empirical evidence on age and risk profiles
- Prior studies (Kraay 2000, Modigliani and Cao 2004, Horioka and Wan 2006) supported demographic factors in savings dynamics.
- Recent evidence (Chamon and Prasad 2008) findings:
  - U-shaped age profile of savings: younger and older households save relatively more.
  - Households facing high expenditure risk on health (typically older households) tend to have a savings rate 20 percentage points higher than households not facing these risks.
  - Households with small children tend to have a savings rate up to 5 percentage points higher than households without, to finance future education spending.

### V. Methodology and simulation overview (as described)
- Methodological components referenced:
  - Generational accounting framework to construct lifetime income and average propensities to consume.
  - Estimation of age-specific marginal propensities to consume for different income groups using household income survey data.
  - Simulation of alternative government social expenditure reforms to quantify aggregate consumption effects.

_Italic: Source: _wp1069 - 1. Household Saving Panel Regression, 1990–2008 (OECD) .................................................10_

### Box 1. Determinants of Household Savings Rate: Survey of Evidence

### Box 1. Determinants of Household Savings Rate: Survey of Evidence

### Evidence from China (selected studies and findings)
- Kraay (2000); Rural and urban household survey of China’s National Bureau of Statistics (NBS); 1978–1995
  - Expectations of future income growth and the role of subsistence consumption play an important role in determining rural household saving levels. No significant relationship explaining urban saving levels.
- Modigliani and Cao (2004); Aggregate data from the China Statistical Yearbook; 1953–2000
  - Savings rate increases with the share of working age population, in line with the life-cycle hypothesis.
- Horioka and Wan (2006); Province-level data from China Statistics Yearbook; 1996–2005
  - Demographic factors affect savings in line with the life-cycle hypothesis; Savings rate across time and provinces determined mainly by the lagged savings rate, income growth, interest rate and, in some cases, inflation.
- Chamon and Prasad (2008); Urban household surveys (NBS); 1990–2005
  - Virtual absence of consumption smoothing over time. Savings rates of younger and older households have grown relatively more; the factor that best explains these patterns is the rising private expenditure on health, education, and housing.
- Wei and Zhang (2009); China population census, County social and economic statistical yearbook 2000, Chinese household income project; 2002
  - Half of the increase in the household savings rate can be explained by the increasing share of males in the population, experienced after the adoption of the one-child policy. Premise: males save to accumulate assets that would put them at a competitive advantage when searching for a spouse.
- Barnett and Brooks (2010); China provincial data from CEIC; 1994–2007
  - Spending on health, but not education, had an impact on household behavior. A one yuan increase in government health spending is associated with a two yuan increase in urban household consumption.

### Evidence from other countries on extending social safety nets
- Kotlikoff (1989); USA; 1950–1987
  - Savings rate is negatively correlated with the availability of public health insurance.
- Kantor and Fishback (1996); USA; 1917–1919
  - Introduction of workers’ compensation following injuries at work reduced private savings by approximately 25 percent of their baseline value.
- Gruber and Yelowitz (1999); USA; 1984–1993
  - Among the population eligible for Medicaid in 1993, each $1,000 of added coverage would increase household consumption by $538.
- Chou, Liu and Hammitt (2006); Taiwan; 1992–1997
  - Extension of health insurance coverage decreased the households’ savings rate by 3–10 percent; this means that an increase in health expenditures of 1 percentage point of GDP increased current household consumption by 0.4–0.6 percent of GDP.

### Panel analysis: Impact of public social expenditures on household saving (24 OECD countries, 1990–2008)
- Analytical framework
  - Baseline regression: Saving_it = α + X_it’β + γZ_it + ν_i + ε_it
    - Saving expressed as percent of disposable income.
    - X_it includes: Real per capita GDP and its growth; old-age dependency ratio (population aged 65+ to 15–64); young-age dependency ratio (population under 15 to 15–64); financial development measured by credit to private sector (percent of GDP).
    - Z_it is social spending variable (percent of GDP): public expenditures on health, education (primary, secondary, higher), and social security (pensions and social assistance).
  - Estimation methods: fixed-effects panel and dynamic panel GMM (Arellano-Bond); data are 5-year averages.

- Key empirical patterns
  - Higher government social spending is generally associated with lower household saving.
  - Estimated effect is non-linear: marginal reduction in saving from increased social spending is largest when social spending (percent of GDP) is low; marginal decline becomes smaller as spending levels rise.
  - Public spending on health care has the largest negative impact on household saving; social security has a smaller but significant effect; education shows a sizable negative impact but is significant mainly when evaluated separately.

### Empirical results (selected coefficients and interpretation)
- Non-linear (U-shaped) relationship for total social spending:
  - Minimum level of household saving reached when total social spending is around 31–36 percent of GDP.
  - Sample mean of social spending is 28.1 percent of GDP.
  - Illustrative marginal effects at the OECD average:
    - For total social spending at 28.1 percent of GDP, a 1 percent of GDP increase in total spending leads to a fall in household saving of 0.26–0.45 percent of household disposable income (Columns 1, 2, and 4). Because average household disposable income is about 54 percent of GDP, this implies a reduction in household saving of 0.14–0.24 percent of GDP.
  - Simultaneous increase in spending on each of the three components (assuming 1/3 outlay on each) yields a similar impact of around 0.13 percent of GDP on household savings (Column 3).

- Component-specific impacts (evaluated at sample means)
  - Public health spending (sample mean 6.3 percent of GDP)
    - Household saving falls by 0.70–0.78 percent of GDP in response to a 1 percent of GDP increase in health expenditure (Column 3 and Appendix).
    - The coefficient on health expenditures in regressions is about 2.1 (consistent with Barnett and Brooks 2010 for urban China).
  - Public education spending (sample mean 5.8 percent of GDP)
    - A 1 percent of GDP increase in public education spending leads to a decline in household saving by 0.79 percent of GDP (significant mainly when estimated separately).
  - Social protection / social security spending (sample mean 16.1 percent of GDP)
    - A 1 percent of GDP increase reduces household saving by 0.22–0.29 percent of GDP (Column 3).

- Table 2: Marginal reduction in household saving (As percent of GDP) of a 1 Percent of GDP Increase in Government Expenditure
  - Total Social Spending (OECD average at 28.1 percent of GDP): 0.14 ~ 0.24
  - Health (OECD, at 6.3 percent of GDP): 0.70 ~ 0.78
  - Education (OECD, at 5.8 percent of GDP): 0.79
  - Social Security (OECD, at 16.1 percent of GDP): 0.22 ~ 0.29
  - Marginal reduction in household saving in China (measured at current levels in China):
    - Total social spending (at 6 percent of GDP): 0.56 ~ 1.03
    - Health (at 0.9 percent of GDP): 2.09 ~ 2.12
    - Education (at 2.9 percent of GDP): 1.26
    - Social security (at 2.2 percent of GDP): 0.68 ~ 0.72
  - Source: Calculations based on Table 1 and Appendix Table 3.

- Additional numeric details from Table 1 (selected coefficients and tests)
  - Household saving (lagged): -0.41* (t-stat -1.73) in Column (1)
  - Growth, per capita GDP: -0.40** (t-stat -2.19) in Column (1)
  - Public social spending, total: -1.95*** (t-stat -4.17) in Column (1)
  - Public social spending total, squared: 0.03*** (t-stat 4.14) in Column (1)
  - Public health spending: -6.84*** (t-stat -3.45) in Column (3)
  - Public health spending, squared: 0.44*** (t-stat 4.09) in Column (3)
  - Social protection spending: -2.15* (t-stat -1.81) in Column (3)
  - Arellano-Bond test for AR(2), p-value: 0.47
  - Hansen Test of Joint Validity of instruments: 0.95
  - No. of Instruments: 14
  - No. of Obs.: 78 (Column 1), 68 (Column 2), 78 (Column 3), 74 (Column 4)
  - No. of countries: 24 (Columns 1, 3, 4), 20 (Column 2)
  - R^2: 0.76 (Column 1), 0.8 (Columns 2 and 3)
  - Note: Levels of significance indicated by asterisks: *** 1 percent, ** 5 percent, * 10 percent.

### Illustrative application to China and caveats
- Using OECD panel estimates to illustrate China:
  - Given China’s current total social spending (around 6 percent of GDP), the marginal reduction in household savings for a 1 percent of GDP increase in social spending could be in the range of 0.56–1.03 percent of GDP.
  - Using China’s spending levels and ratio of disposable income to GDP, the estimated impact of a 1 percent increase in government spending on health is about 2 percent of GDP on household savings.
- Caveats
  - Extrapolation from OECD sample to China requires caution: sample countries have higher levels of economic and institutional development than China.
  - Public education spending results should be interpreted with caution because education is only significant when considered separately from other social spending components.
  - Coefficients for health spending estimated when included alone (Appendix Column A) may pick up residual effects of other spending categories.

### Mechanisms and methodology for assessing China (generational accounting framework)
- Three channels through which increased government social expenditures can raise household consumption:
  1. Income effect via higher aggregate lifetime resources: expansion of government social expenditures that cover expenditures previously borne by households increases lifetime resources and current consumption.
  2. Age-concentration effect: expenditures concentrated on groups with higher average propensities to consume out of lifetime income (e.g., elderly) yield larger consumption impacts.
  3. Insurance (precautionary savings) effect: reduced need for precautionary savings when future health, education, and retirement risks are better covered.
- Quantification approach
  - Income and age-concentration effects estimated via generational accounting applied to household-level data (method based on Gokhale, Kotlikoff, and Sabelhaus (1997); see also Kirsanova and Sefton (2007)).
  - Insurance effect is more difficult to identify; inferred illustratively from Taiwan study and panel regression results.

*Source: Box 1, “Determinants of Household Savings Rate: Survey of Evidence,” _wp1069 - Box 1. Determinants of Household Savings Rate: Survey of Evidence*

### Box 2. Methodology for Estimating Consumption Impact of Social Expenditures

### Box 2. Methodology for Estimating Consumption Impact of Social Expenditures

### Methodology overview
- Consumption for group i at time t is defined as:
  - cit = αi rit
  - where αi is the average propensity to consume out of lifetime resources and rit is the net present value of the group’s remaining lifetime resources.
- Total consumption is decomposed across groups using population weights Pit and group lifetime resources rit.
- Changes in total consumption are decomposed into four components:
  - Changes in total lifetime household income, tt rP
  - Changes in the distribution of total lifetime household income across groups, tit r r
  - Changes in group propensities to consume out of group lifetime income, i α
  - Changes in the distribution of the population across groups, tit P P

### Interpretation of decomposition channels
- Income effect: combination of (i) changes in total lifetime household income and (ii) changes in its distribution across groups — reflects increase in household real incomes and is estimated via generational accounting applied to household survey data.
- Insurance effect: changes in average propensities to consume (i α) reflecting decreases in precautionary savings — requires microeconometric estimation.
- Demographic effect: changes in population distribution across groups (tit P P) — in the current version kept fixed, but can be imposed using projected demographic trends to assess impact on propensities and total consumption.

### Implementation steps
- Three implementation steps:
  1. Calculation of lifetime resources for socio-economic groups.
  2. Estimation of group average propensities to consume (αi).
  3. Translation of increases in public expenditures into changes in lifetime resources.
- Grouping dimensions: current income (income quintiles), urban/rural residence, and age.
- Lifetime resource calculation:
  - Use household budget survey data for assets, incomes, taxes, transfers.
  - Assign each individual the group mean for lifetime income components.
  - Assume future flows for a group member follow group averages for successive age groups (e.g., next year’s income for a 20-year-old equals current income of 21-year-old in same group).
  - Net present values calculated using a common discount rate and survival probabilities.
- Propensity estimation:
  - Group average propensities = current consumption / lifetime income.
  - Propensities increase with age and are higher for rural and lower-income groups.
- Translating expenditure reforms:
  - A reform is mapped into additional flows to each group over time; impact on current consumption = change in group net present value × group propensity to consume.
  - Example: increased public health spending reducing health care charges yields larger proportional NPV increases for the elderly.

### Data and calculation details (Box 3)
- Data source: Chinese Household Income Survey (CHIP) 2002.
  - Urban: 12 provinces, 6,800 households, 20,600 individuals.
  - Rural: 22 provinces, 9,200 households, 38,000 individuals.
  - Migrant households in urban provinces: 2,000 households, 5,300 individuals.
- Consumption at household level converted to per capita by dividing by household size.
- Individual income constructed from individual non-agricultural income plus per capita agricultural income where applicable.
- Per capita net assets = net household assets (including market value of privately-owned houses) / household size.
- Lifetime income NPV calculation assumptions:
  - Future wages follow age-specific wage profiles augmented by a growth rate of 7.7 percent.
  - Discounted based on an interest rate of 11.7 percent and on survival rates from the 2006 WHO Life Table for China, giving a discount rate of 0.96.
  - Lifetime resources include present value of net transfers, net pension benefits, in-kind benefits for education and health, and financial assets.
- Average propensity to consume = current consumption including education and health spending / lifetime resources.
- Propensities applied to 2007 national accounts and population data to compute group consumption levels.

### Simulation design and financing assumptions
- Simulated reforms: pension transfer (universal cash transfer to all individuals over 55), education transfer (proportional to current education expenditures by age profile), health transfer (proportional to current health expenditures by age profile).
- Pension reform interpreted as large expansion of coverage (McKinsey (2009): in 2009 pension coverage ~90 percent urban, 20-25 percent rural).
- Simulations assume reforms are permanent and financed through a permanent decrease in the fiscal surplus (i.e., financed from existing budgetary surpluses, not from increased taxes).
- Budget for each reform = 1 percent of GDP annually (absolute size rises with GDP).
- Note: If financed by taxes (budget-neutral) or implemented temporarily, net consumption impacts would be much smaller.

### Income effect — empirical findings (Simulation 1 and related notes)
- Under Simulation 1 (1 percent of GDP increase in each expenditure category annually), income effects on current household consumption (in percent of GDP):
  - Pension: Total 1.42
    - Urban 0.92
    - Rural 0.50
  - Health: Total 0.77
    - Urban 0.46
    - Rural 0.32
  - Education: Total 0.51
    - Urban 0.24
    - Rural 0.27
- Budget shares of expenditures by area:
  - Pension: Urban 0.75, Rural 0.25
  - Health: Urban 0.69, Rural 0.31
  - Education: Urban 0.58, Rural 0.42
- Additional observations:
  - Impact per unit of government social spending is substantially higher in rural areas due to higher propensities to consume and lower lifetime budgets.
  - Example: under Simulation 1, pension impact is 0.9 in urban areas versus 0.5 in rural areas; adjusting for budget shares implies the income effect of pension expenditures in rural areas is about 67 percent higher than in urban areas [(0.5/0.25) divided by (0.9/0.75)].
  - Equivalent rural/urban differentials: health 56 percent higher in rural areas; education 55 percent higher in rural areas.
  - Targeting expenditures to rural or low-income households increases cost-effectiveness in raising household consumption.
- Alternative scenario:
  - Under the smaller lifetime budget in Simulation 2, income effects are lower: pensions 0.8 percent, health 0.5 percent, education 0.4 percent of GDP.

### Insurance effect — empirical estimation and application
- Empirical evidence is limited; one relevant study (Chou, Liu, and Hammitt (2006)) evaluated Taiwan’s health insurance expansion (coverage from 57 percent in 1994 to 96 percent in 2000).
  - Their estimates: an increase in health expenditures of 1 percentage point of GDP increased current household consumption by 0.4 to 0.6 percent of GDP via the insurance effect.
- For this paper:
  - Compute ratio of health insurance impact to total health impact estimated in Section III: ratio = 0.24 (i.e., 0.5 divided by 2.1).
  - Apply ratio 0.24 to total consumption impacts estimated for education (1.3) and pensions (0.7) to obtain estimated insurance impacts:
    - Education insurance impact = 0.31 percent of GDP
    - Pension insurance impact = 0.17 percent of GDP

### Total effect
- Total household consumption impact = income effect (Table 3) + insurance impacts (above).
- The resulting total consumption impacts range from 1.6 percent of GDP for pensions,

*Source: Box 2 and Box 3, _wp1069 - Methodology for Estimating Consumption Impact of Social Expenditures (staff estimates based on 2002 CHIP household survey data).*

### 1.3 percent for health,

### 1.3 percent for health,

### Household consumption impact of expenditure reforms
- Table 4 headline impacts (In percent of GDP) for a 1 percent of GDP annual increase in each expenditure category:
  - Total: Pension 1.6; Health 1.3; Education 0.8
  - Income Effect: Pension 1.4; Health 0.8; Education 0.5
  - Insurance Effect: Pension 0.2; Health 0.5; Education 0.3
- A 1 percentage point of GDP increase in social expenditures allocated evenly across pension, health, and education would result in a permanent increase in household consumption of 1.2 percent of GDP.
- A sustained one percentage point of GDP increase in government spending is likely to lead to an increase in the household consumption ratio of up to 1¼ percentage points of GDP.

### Simulation scenarios and required fiscal increases
- Evenly distributed social expenditure increase scenario:
  - To raise household consumption by 3 percentage points of GDP, a 2.5 percent of GDP increase in total social expenditures maintained over the medium term is required.
- Targeted expenditure scenarios:
  - The same 3 percentage point of GDP increase in household consumption would require a 1.9 percent increase in pension expenditures.
  - Alternatively, it would require a 2.2 percent increase in health expenditures.
- Complementary structural reform scenario:
  - Simulating a 1 percent additional growth in labor income in China within the framework generates a 0.7 percent increase in current household consumption.

### Financing considerations and distributional effects
- If the 1 percent of GDP increase in expenditures in each category is financed by an increase in income taxes, financing would lead to an offsetting decrease in consumption of nearly 0.6 percent of GDP.
  - Net increases in household consumption under income-tax financing:
    - Education: 0.2 percent of GDP
    - Health: 0.7 percent of GDP
    - Pensions: 1.0 percent of GDP
- The smaller net impact under tax financing reflects the redistributive effect of combined tax and expenditure reforms, redirecting resources from those with low to those with high propensities to consume.

### Drivers, channels, and broader macro implications
- Identified channels through which increased government social expenditures raise consumption:
  - (i) Direct income channel
  - (ii) Distributional channel
  - (iii) “Insurance” channel
- Underlying causes of high household savings and low consumption ratio include:
  - Greater uncertainty facing households following structural changes
  - Reduced provision of support for education, health, and old age pensions by the government
- Contribution to external rebalancing:
  - Estimates suggest household consumption would have to increase by some 3 to 4 percentage points of GDP, assuming corporate sector and government savings rates remain unchanged, to help rebalance world demand and address China’s large external current account surplus.
  - Greater provision of public expenditure on health, education, and pensions could make an important contribution alongside other structural reforms (availability of credit, retail distribution, measures to increase the share of wages in national income).

### Supporting empirical evidence (selected quantitative results)
- Cross-country household saving panel (1990–2008, OECD sample) regression highlights:
  - Public health spending coefficient: -6.74*** with Public health spending, squared coefficient: 0.42*** (FE specification A)
  - Public education spending coefficient: -5.76*** with Public education spending, squared coefficient: 0.37*** (FE specification B)
  - Social security spending coefficient: -2.31** with Social security spending, squared coefficient: 0.06** (FE specification C)
  - Sample: No. of Obs. 78; No. of countries 24; R^2: 0.78 (A), 0.72 (B), 0.71 (C)
- Appendix household consumption pattern (average shares):
  - Urban average consumption shares: Food 0.364; Non-food 0.326; Housing 0.169; Education 0.074; Health 0.066
  - Rural average consumption shares: Food 0.428; Non-food 0.282; Housing 0.160; Education 0.083; Health 0.047

### Policy implications and recommendations
- Increasing government social expenditures on health, education, and pensions can be an effective tool to raise household consumption and reduce household precautionary saving.
- Policy options and priorities:
  - Target a higher proportion of expenditure increases to health and pensions, or to rural and low-income households, to achieve larger consumption impacts for a given budget increase.
  - Combine expenditure reforms with structural measures to raise the share of wages in national income and expand domestic consumption (credit availability, retail distribution reforms).
  - Consider financing methods carefully: deficit-financed increases in social spending produce larger immediate consumption impacts than equivalent increases financed by income taxes, though tax-financed reforms still yield positive net impacts through redistribution.

*Source: Staff estimates and analyses from the provided document.*

---


_Source: https://www.imf.org/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2010/_wp1069.pdf_