## 4.1 The Model

## Source details

**Canonical URL:** [4.1 The Model](https://www.imf.org/-/media/files/publications/wp/2023/english/wpiea2023133-print-pdf.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/wp/2023/english/wpiea2023133-print-pdf.pdf.md)
- [Structured JSON version](/-/media/files/publications/wp/2023/english/wpiea2023133-print-pdf.pdf.json)

---

### Purpose and empirical strategy
- Objective: Reassess whether stabilization SWFs help smooth government consumption and reduce fiscal policy volatility.
- Sample: 182 advanced and developing countries over the period 1980-2019.
- Main empirical challenge addressed: Non-random decision to establish a stabilization SWF (self-selection/endogeneity).

### Econometric approaches
- Two-step estimation method based on a treatment effect model (Maddala, 1983).
- Non-parametric estimation method based on Propensity Score Matching (PSM).
- Rationale: Use complementary methods to address endogeneity of SWF adoption; methodological novelty noted.

### Key empirical finding (policy-relevant result)
- Estimated average treatment effects show that a country with a stabilization SWF has about 14 percent less volatile government consumption relative to a country without an SWF.
- Result robustness: Described as robust under different specifications and sample periods.

### Theoretical and contextual considerations
- Commodity price volatility has increased, especially for oil, influenced by the pandemic and the war in Ukraine.
- Oil price example: increased from US$23.4 a barrel in April 2020 to US$114.7 a barrel in June 2022.
- Commodity-exporting countries defined as those with fuel/mining exports accounting for an average of 20 percent of their exports over a decade.
- Theoretical expectation: Optimal fiscal response for commodity exporters is countercyclical policy; stabilization SWFs accumulate resources when prices exceed a reference level and use resources when prices fall below another reference level.
- Caveat: In the absence of liquidity constraints, stabilization SWFs might not directly stabilize government expenditures because governments could borrow to meet financing needs.

---

### Model specification and measurement

### Baseline model
- Baseline linear relationship (Equation (1)):
  - σ_{i,t} = x_{i,t}′β + ρ SWF_{i,t} + η_i + δ_{i,t}
    - σ_{i,t}: measure of discretionary fiscal policy volatility.
    - x: vector of exogenous variables.
    - SWF: dummy variable equals one if country i has a stabilization SWF at time t and zero otherwise.
    - β and ρ: parameters to be estimated.
    - η: time invariant country-specific error term.
    - δ: time dependent random disturbance term.

### Measuring discretionary fiscal policy volatility
- Two approaches discussed:
  - Standard deviation of the annual growth rate of real government consumption (criticized for ignoring cyclical state).
  - Fatás and Mihov approach: volatility measured as the logarithm of the standard deviation of residuals from a fiscal reaction function:
    - Δln(G_{i,t}) = α + β Δln(Y_{i,t}) + δ W_{i,t} + ε_{i,t}
      - G_{i,t}: real government consumption.
      - Y_{i,t}: real GDP.
      - W_{i,t}: controls including inflation and its square, and real GDP per capita.
      - ε_{i,t}: discretionary fiscal policy (exogenous to output growth and automatic stabilizers).
- Implementation specifics in this paper:
  - Use panel data to estimate Equation (2) and control for unobservable country-specific effects.
  - Estimate Equation (2) by system GMM to address endogeneity via internal instruments (lagged values).
  - Control for the presence of fiscal rules.
  - Note: Fatás and Mihov originally used Instrumental Least Squares (ILS).

---

### Addressing endogeneity of SWF adoption

### Selection problems
- Two selection issues from including SWF dummy in Equation (1):
  1. Endogeneity of the dummy variable (selection bias): decision to establish a stabilization SWF is endogenous.
  2. Missing counterfactual: cannot observe σ_{i}|SWF_{i}=1 and σ_{i}|SWF_{i}=0 simultaneously for the same country.
- Conventional OLS would be biased and inconsistent.

### Methods to correct selection bias
- Regression-based treatment effect model (two-step selection and outcome equations).
- Non-parametric Propensity Score Matching (PSM).

### Treatment effect model details
- Latent selection equation and outcome equation formulation:
  - σ_{i,t} = x_{i,t}′β + ρ D^{*}_{i,t,SWF} + η_i + ε_{i,t}
  - D^{*}_{i,t,SWF} = z_{i,t}′γ + μ_{i,t}
  - Observed D_{i,t,SWF} = 1 if D^{*}_{i,t,SWF} > 0 and 0 otherwise.
- Outcome equation controls (motivated by literature and data availability):
  1. Log(GDP).
  2. Inflation.
  3. Government consumption as percent of GDP (Govsize).
  4. Degree of openness (exports + imports as percent of GDP).
  5. Commodity terms of trade index (CTOT), capital mobility degree, real effective exchange rate (REER).
  6. Institutional controls: sound institutions, corruption, democracy, fiscal rules.
- Selection equation determinants:
  1. Country-specific macro factors: level of development, real GDP growth rate, level of foreign reserves, inflation rates, commodity terms of trade.
  2. Institutional factors: democratic institutions and fiscal rules.
  3. Global factors: real oil prices.
- Estimation by full maximum likelihood; all variables except SWF measured as 5-year moving averages; SWF lagged (t-5) in treatment specifications.

---

### Propensity Score Matching (PSM) approach

### Notation and identifying assumptions
- Y_{1,i}: outcome if country i establishes a SWF.
- Y_{0,i}: outcome if country i does not establish a SWF.
- D_i: treatment indicator (1 if treated).
- δ_i = Y_{1,i} – Y_{0,i}; ATE = E(δ_i); ATT = E[Y_{1,i}|D_i=1] – E[Y_{0,i}|D_i=1].
- Propensity score: P(s) = P(D_i=1 | X=x).
- Key assumptions:
  1. Conditional independence (unconfoundedness): (Y_{1i}, Y_{0i}) ┴ D | X.
  2. Common support (overlap): 1 > P(D=1 | X) > 0.

### Implementation
- Propensity scores estimated via pooled panel Probit regression using determinants in Equation (4).
- Matching algorithm: Nearest Neighbor matching (1, 3, and 5 neighbors).
- Outcome: log of volatility of discretionary government expenditure over 1980-2019.
- All independent variables lagged one year; robust t-statistics reported.

---

### Data sources
- Unbalanced panel for 182 advanced and developing countries over 1980–2019.
- Real government expenditure and real GDP: World Penn table.
- Macroeconomic data: IMF World Economic Outlook database (2021) and World Bank, World Development Indicators (2021).
- Democracy index: sum of political and civil right indices from Freedom House (2021).
- SWF data: Sovereign Wealth Fund Institute and the International Forum of Sovereign Wealth Funds.
- SWF dummy: equals one if country has a stabilization SWF at time t and zero otherwise.
- Appendix contains full data descriptions, Appendix Table A1 descriptive statistics, Appendix Table A2 correlation matrices.

---

### Empirical results

### Fixed Effects confirmation
- Initial estimation of Equation (2) by Fixed Effects Model (FEM) confirms negative and robust statistical significance of the stabilization SWF coefficient across specifications (details in Table 1).

### Treatment effect model results (Table 2)
- Dependent variable: log of volatility of discretionary government expenditure: 1980-2019.
- Estimated SWF (t-5) coefficients (Average Treatment Effects):
  - Full Sample: SWF (t-5) = -0.62*** (z = -6.23)
  - Developing Countries: SWF (t-5) = -0.60*** (z = -5.27)
  - Commodity-Exporters: SWF (t-5) = -0.77*** (z = -4.71)
  - Developing Commodity-Exporters: SWF (t-5) = -0.66*** (z = -3.80)
- Selected other coefficients (exact values and z-statistics):
  - Log (GDP):
    - Full Sample: -0.06*** (z = -14.57)
    - Developing Countries: -0.05*** (z = -8.60)
    - Commodity-Exporters: -0.03*** (z = -4.64)
    - Developing Commodity-Exporters: -0.02* (z = -2.53)
  - Log (Govsize):
    - Full Sample: -0.11*** (z = -4.46)
    - Developing Countries: -0.07* (z = -2.57)
    - Commodity-Exporters: -0.19*** (z = -4.75)
    - Developing Commodity-Exporters: -0.16*** (z = -3.64)
  - Degree of Openness:
    - Full Sample: -0.002 (z = -1.05)
    - Developing Countries: -0.001* (z = -2.27)
    - Commodity-Exporters: 0.001* (z = 2.29)
    - Developing Commodity-Exporters: 0.001* (z = 2.07)
  - Log (CTOT):
    - Full Sample: -0.7*** (z = -6.02)
    - Developing Countries: -0.8*** (z = -6.64)
    - Commodity-Exporters: -1.0*** (z = -7.19)
    - Developing Commodity-Exporters: -1.1*** (z = -7.27)
  - Log (REER):
    - Full Sample: -0.06 (z = -1.29)
    - Developing Countries: 0.00 (z = 0.05)
    - Commodity-Exporters: -0.21** (z = -2.94)
    - Developing Commodity-Exporters: -0.19* (z = -2.41)
  - Democracy index:
    - Full Sample: -0.72*** (z = -15.75)
    - Developing Countries: -0.42*** (z = -7.30)
    - Commodity-Exporters: -0.53*** (z = -6.57)
    - Developing Commodity-Exporters: -0.30** (z = -2.64)
- Sample sizes and diagnostics:
  - No. of Observations:
    - Full Sample: 4,026
    - Developing Countries: 3,140
    - Commodity-Exporters: 1,608
    - Developing Commodity-Exporters: 1,321
  - Wald Test statistics:
    - Full Sample: 5/30.05
    - Developing Countries: 20.33
    - Commodity-Exporters: 14.88
    - Developing Commodity-Exporters: 12.51
  - P-value for Wald Test: 0.000 for all reported groups.
- Interpretation: Estimated average treatment effects are negative and statistically significant at 1 percent in all reported sub-samples, indicating countries with stabilization SWFs have lower volatility of discretionary government expenditure than comparable countries without such funds.

### Propensity Score Matching results (Table 3)
- Nearest-Neighbor Matching estimates (t-statistics preserved; ***, **, * denote 1, 5, 10 percent significance):
  - Full sample:
    - SWF = -0.11* (t = -2.23) with 1 neighbor.
    - SWF = -0.13** (t = -2.89) with 3 neighbors.
    - SWF = -0.14** (t = -2.94) with 5 neighbors.
  - Developing Countries:
    - SWF = -0.12* (t = -2.27) with 1 neighbor.
    - SWF = -0.14** (t = -2.90) with 3 neighbors.
    - SWF = -0.14** (t = -2.87) with 5 neighbors.
  - Commodity-Exporters:
    - SWF = -0.18*** (t = -3.93) with 1 neighbor.
    - SWF = -0.19*** (t = -4.71) with 3 neighbors.
    - SWF = -0.19*** (t = -4.72) with 5 neighbors.
  - Developing Commodity-Exporters:
    - SWF = -0.20*** (t = -4.14) with 1 neighbor.
    - SWF = -0.21*** (t = -4.87) with 3 neighbors.
    - SWF = -0.22*** (t = -4.84) with 5 neighbors.
- Sample sizes and groups (examples reported in Table 3): No. of Observations: 5,309; 4,304; 3,042; 3,112; 2,007. No. of Groups: 166; 151; 80; 65 (varies by column).
- Interpretation: PSM confirms a negative effect of stabilization SWFs on fiscal policy volatility; PSM estimates are smaller in magnitude than the treatment-effect model but remain robust and statistically significant at 1 percent in commodity-exporting and developing commodity-exporting samples.

---

### Conclusion, caveats, and implications

### Main conclusion
- Presence of stabilization SWFs is associated with lower volatility of discretionary government expenditure: fiscal policy volatility in countries with stabilization SWFs is lower, relative to that in countries without such a fund, by about 14 percent.

### Policy implication
- Stabilization SWFs can help smooth government consumption during downturns, reducing fiscal policy volatility associated with commodity revenue fluctuations.
- However, establishment of a stabilization SWF does not by itself ensure fiscal insulation from commodity price fluctuations; SWFs are not a substitute for sound fiscal policy.

### Caveats and limitations
- Stabilization SWFs vary in size and deposit/withdrawal rules; the binary SWF dummy cannot capture heterogeneity in size or design.
- Future research could distinguish SWFs by size and by specific deposit/withdrawal rules to capture heterogeneous impacts.
- Data transformation note: with the exception of the SWF dummy variable, all variables are in 5-year moving average format.

*Italic: IMF Working Papers — Sovereign Wealth Funds and Fiscal Policy Procyclicality: Non-Parametric Approach (content unit: 4.1 The Model; empirical results; conclusion).*

### 4.1 The Model ..........................................................................................................

### 4.1 The Model

### Purpose and empirical strategy
- Objective: Reassess whether stabilization SWFs help smooth government consumption and reduce fiscal policy volatility.
- Sample: 182 advanced and developing countries over the period 1980-2019.
- Main empirical challenge addressed: Non-random decision to establish a stabilization SWF (self-selection/endogeneity).

### Econometric approaches employed
- Two-step estimation method based on a treatment effect model.
- Non-parametric estimation method based on Propensity Score Matching (PSM).
- Rationale: Use complementary methods to address endogeneity of SWF adoption; methodological novelty highlighted as the first-time fiscal policy is evaluated using these approaches in this context.

### Key empirical finding (policy-relevant result)
- The estimated average treatment effects show that a country with a stabilization SWF has about 14 percent less volatile government consumption relative to a country without an SWF.
- Result robustness: Described as robust under different specifications and sample periods.

### Contextual and theoretical considerations summarized in the model section
- Commodity price volatility has increased in recent years—especially for oil—following in large part the effects of the pandemic and the war in Ukraine.
- Oil price example cited: increased from US$23.4 a barrel in April 2020 to US$114.7 a barrel in June 2022.
- Definition used: Commodity-exporting countries are those with fuel/mining exports accounting for an average of 20 percent of their exports over a decade.
- Theoretical expectation: Optimal fiscal response for commodity exporters is countercyclical policy (save in booms, disburse in downturns); stabilization SWFs are an instrument to accumulate resources when prices exceed a reference level and use resources when prices fall below another reference level.
- Caveat noted: In the absence of liquidity constraints, a stabilization SWF might not directly stabilize government expenditures because governments could meet financing needs through borrowing.

### Literature synthesis informing model setup
- Mixed empirical evidence exists: some studies find stabilization SWFs reduce fiscal volatility; others find no significant effect or increased volatility.
- Prior empirical methods often failed to properly control for the endogeneity of SWF adoption; this study’s choice of treatment-effect and PSM methods aims to address that gap.

_Italic: IMF Working Papers — Sovereign Wealth Funds and Fiscal Policy Procyclicality: Non-Parametric Approach (content unit: 4.1 The Model)_.

### 4.    EMPIRICAL ANALYSIS

### 4.    EMPIRICAL ANALYSIS

### The Model
- Baseline linear relationship (Equation (1)):
  - σ_{i,t} = x_{i,t}′β + ρ SWF_{i,t} + η_i + δ_{i,t}
  - σ_{i,t}: measure of discretionary fiscal policy volatility.
  - x: vector of exogenous variables.
  - SWF: dummy variable equals one if country i has a stabilization SWF at time t and zero otherwise.
  - β and ρ: parameters to be estimated.
  - η: time invariant country-specific error term.
  - δ: time dependent random disturbance term.

### Measuring discretionary fiscal policy volatility
- Two main approaches discussed:
  - Standard deviation of the annual growth rate of real government consumption — criticized for ignoring the cyclical state of the economy.
  - Fatás and Mihov (2003, 2006) approach: volatility measured as the logarithm of the standard deviation of residuals from a fiscal reaction function (Equation (2)):
    - Δln(G_{i,t}) = α + β Δln(Y_{i,t}) + δ W_{i,t} + ε_{i,t}
    - G_{i,t}: real government consumption; Y_{i,t}: real GDP; W_{i,t}: controls including inflation and its square, and real GDP per capita.
    - ε_{i,t} represents discretionary fiscal policy (exogenous to output growth and automatic stabilizers).
- Implementation differences in this paper:
  - Use panel data to estimate Equation (2) to control for unobservable country-specific effects.
  - Estimate Equation (2) by system GMM to address endogeneity via internal instruments (lagged values).
  - Control for the presence of fiscal rules.
- Note: Fatás and Mihov (2003) originally used Instrumental Least Squares (ILS).

### The endogeneity of SWF
- Two selection problems from presence of SWF dummy in Equation (1):
  1. Endogeneity of the dummy variable (selection bias): decision to establish a stabilization SWF is endogenous (determined by other factors including episodes of poor macroeconomic performance).
  2. Missing counterfactual: cannot observe σ_{i}|SWF_{i}=1 and σ_{i}|SWF_{i}=0 simultaneously for same country.
- Conventional methods such as OLS would yield biased and inconsistent estimates.
- Approach: use two econometric methods to correct selection bias:
  - Regression-based treatment effect model (Maddala, 1983).
  - Non-parametric Propensity Score Matching (PSM).

### The Treatment Effect Model
- Two-step procedure: selection equation (probability of establishing a stabilization SWF) and outcome equation (impact on volatility).
- Latent endogenous selection equation (Equations (3) and (4)):
  - σ_{i,t} = x_{i,t}′β + ρ D^{*}_{i,t,SWF} + η_i + ε_{i,t}
  - D^{*}_{i,t,SWF} = z_{i,t}′γ + μ_{i,t}
  - Observed D_{i,t,SWF} = 1 if D^{*}_{i,t,SWF} > 0 and 0 otherwise.
- Consistent estimation requires correlation between ε and μ (if uncorrelated, OLS would be consistent).
- Outcome equation controls (variables motivated by literature and data availability):
  1. Log(GDP) — size of the economy.
  2. Inflation — proxy for macroeconomic stability.
  3. Government consumption as percent of GDP — government size.
  4. Degree of openness (exports + imports as percent of GDP).
  5. Commodity terms of trade index, capital mobility degree, real effective exchange rate.
  6. Institutional controls: sound institutions, corruption, democracy, fiscal rules.
- Selection equation determinants (factors leading to establishment of stabilization SWF):
  1. Country-specific macro factors: level of development, real GDP growth rate, level of foreign reserves, inflation rates, commodity terms of trade.
  2. Institutional factors: democratic institutions and fiscal rules.
  3. Global factors: real oil prices.

### Propensity Score Matching (PSM)
- Non-parametric alternative robust to selection bias under key assumptions.
- Notation:
  - Y_{1,i}: outcome if country i establishes a SWF.
  - Y_{0,i}: outcome if country i does not establish a SWF.
  - D_i: treatment indicator (1 if treated).
  - Treatment effect for country i: δ_i = Y_{1,i} – Y_{0,i}
  - ATE = E(δ_i) = E[Y_{1,i} – Y_{0,i}]
  - ATT = E[Y_{1,i}|D_i=1] – E[Y_{0,i}|D_i=1]
  - Propensity score: P(s) = P(D_i=1 | X=x)
- Key identifying assumptions:
  1. Conditional independence (unconfoundedness): (Y_{1i}, Y_{0i}) ┴ D | X
  2. Common support (overlap): 1 > P(D=1 | X) > 0
- Implementation in this paper:
  - Estimate propensity scores via pooled panel Probit regression (Equation (4) determinants).
  - Matching algorithm: Nearest Neighbor matching.
  - Compute average treatment effects using matched sample; standard errors calculated.

### Data Sources
- Unbalanced panel for 182 advanced and developing countries over 1980–2019.
- Real government expenditure and real GDP: World Penn table.
- Macroeconomic data: IMF World Economic Outlook database (2021) and World Bank, World Development Indicators (2021).
- Democracy index: sum of political and civil right indices from Freedom House (2021).
- SWF data: Sovereign Wealth Fund Institute and the International Forum of Sovereign Wealth Funds.
- SWF dummy: equals one if country has a stabilization SWF at time t and zero otherwise.
- All independent variables are lagged one year.
- Appendix contains full data descriptions, Appendix Table A1 descriptive statistics, Appendix Table A2 correlation matrices.

### Empirical results (overview)
- Initial estimation of Equation (2) by Fixed Effects Model (FEM) as a confirmatory step.
- Full sample includes advanced and developing economies; additional regressions for developing countries and commodity-exporting countries.
- Table 1 (Fixed Effects) summary statement: estimated coefficient of the stabilization SWF is negative and robustly statistically significant in all specifications (details in Table 1).

### Treatment effect model results (Table 2)
- Dependent variable: log of volatility of discretionary government expenditure: 1980-2019.
- Estimated SWF (t-5) coefficients and significance in Average Treatment Effects estimations:
  - Full Sample: SWF (t-5) = -0.62*** (z = -6.23)
  - Developing Countries: SWF (t-5) = -0.60*** (z = -5.27)
  - Commodity-Exporters: SWF (t-5) = -0.77*** (z = -4.71)
  - Developing Commodity-Exporters: SWF (t-5) = -0.66*** (z = -3.80)
- Other notable coefficients (exact values and z-statistics preserved):
  - Log (GDP):
    - Full Sample: -0.06*** (z = -14.57)
    - Developing Countries: -0.05*** (z = -8.60)
    - Commodity-Exporters: -0.03*** (z = -4.64)
    - Developing Commodity-Exporters: -0.02* (z = -2.53)
  - Log (Govsize):
    - Full Sample: -0.11*** (z = -4.46)
    - Developing Countries: -0.07* (z = -2.57)
    - Commodity-Exporters: -0.19*** (z = -4.75)
    - Developing Commodity-Exporters: -0.16*** (z = -3.64)
  - Degree of Openness:
    - Full Sample: -0.002 (z = -1.05)
    - Developing Countries: -0.001* (z = -2.27)
    - Commodity-Exporters: 0.001* (z = 2.29)
    - Developing Commodity-Exporters: 0.001* (z = 2.07)
  - Log (CTOT):
    - Full Sample: -0.7*** (z = -6.02)
    - Developing Countries: -0.8*** (z = -6.64)
    - Commodity-Exporters: -1.0*** (z = -7.19)
    - Developing Commodity-Exporters: -1.1*** (z = -7.27)
  - Log (REER):
    - Full Sample: -0.06 (z = -1.29)
    - Developing Countries: 0.00 (z = 0.05)
    - Commodity-Exporters: -0.21** (z = -2.94)
    - Developing Commodity-Exporters: -0.19* (z = -2.41)
  - Democracy index:
    - Full Sample: -0.72*** (z = -15.75)
    - Developing Countries: -0.42*** (z = -7.30)
    - Commodity-Exporters: -0.53*** (z = -6.57)
    - Developing Commodity-Exporters: -0.30** (z = -2.64)
- Sample sizes and diagnostics (preserved):
  - No. of Observations:
    - Full Sample: 4,026
    - Developing Countries: 3,140
    - Commodity-Exporters: 1,608
    - Developing Commodity-Exporters: 1,321
  - Wald Test statistics:
    - Full Sample: 5/30.05
    - Developing Countries: 20.33
    - Commodity-Exporters: 14.88
    - Developing Commodity-Exporters: 12.51
  - P-value for Wald Test: 0.000 for all reported groups.
- Interpretation: The estimated average treatment effects are negative and statistically significant at 1 percent in all reported sub-samples, indicating that countries with stabilization SWFs have lower volatility of discretionary government expenditure than comparable countries without such funds.
- Note: Models estimated by full maximum likelihood; all variables, except SWF, measured as 5-year moving averages.

### Propensity Score Matching results (Table 3)
- PSM estimation approach: Nearest-Neighbor Matching, using pooled panel Probit to estimate propensity scores.
- Estimated SWF impacts (Nearest-Neighbor Matching):
  - Full sample:
    - SWF = -0.11* (t = -2.23) with 1 neighbor.
    - SWF = -0.13** (t = -2.89) with 3 neighbors.
    - SWF = -0.14** (t = -2.94) with 5 neighbors.
  - Developing Countries:
    - SWF = -0.12* (t = -2.27) with 1 neighbor.
    - SWF = -0.14** (t = -2.90) with 3 neighbors.
    - SWF = -0.14** (t = -2.87) with 5 neighbors.
  - Commodity-Exporters:
    - SWF = -0.18*** (t = -3.93) with 1 neighbor.
    - SWF = -0.19*** (t = -4.71) with 3 neighbors.
    - SWF = -0.19*** (t = -4.72) with 5 neighbors.
  - Developing Commodity-Exporters:
    - SWF = -0.20*** (t = -4.14) with 1 neighbor.
    - SWF = -0.21*** (t = -4.87) with 3 neighbors.
    - SWF = -0.22*** (t = -4.84) with 5 neighbors.
- Sample sizes and groups (preserved from Table 3):
  - No. of Observations examples: 5,309; 4,304; 3,042; 3,112; 2,007 (varies by column).
  - No. of Groups examples: 166; 151; 80; 65 (varies by column).
- Interpretation: PSM estimates confirm the negative effect of stabilization SWFs on fiscal policy volatility; estimated effects are smaller in magnitude than treatment-effect model but remain robust and statistically significant at 1 percent in commodity-exporting and developing commodity-exporting samples.
- Notes:
  - Treatment variable is whether a country has a stabilization SWF at time (t-5).
  - Outcome is log of volatility of discretionary government expenditure over 1980-2019.
  - (1), (3), (5) refer to the number of nearest neighbors used.
  - Robust t-statistics in parentheses; ***, **, * indicate statistical significance at 1, 5, and 10 percent levels respectively.
  - All variables, except SWF, are measured as 5-year moving averages.

*Source: IMF Working Paper (extracted section 4 and related tables).*

### 6.    CONCLUSION

### 6.    CONCLUSION

### Research question and theoretical rationale
- Assessed the relationship between the presence of stabilization SWFs and the degree of volatility of fiscal policy.
- The issue is motivated by recent volatility in commodity prices following the pandemic and the war in Ukraine.
- Theoretical perspective: a stabilization SWF helps smooth fluctuations in budget resources by reducing or eliminating the uncertainty and volatility of resource-related revenue flowing into the budget.

### Data and empirical approaches
- Dataset: an unbalanced panel for 182 advanced and developing countries during 1980-2019.
- Two econometric approaches used to address the self-selection problem:
  - A two-step regression-based method that estimates the outcome and selection equations simultaneously.
  - A non-parametric approach in which the average treatment effect is estimated by propensity score matching.

### Main empirical findings
- The empirical findings support the argument that stabilization SWFs help smooth government consumption during bad times.
- Fiscal policy volatility in countries with stabilization SWFs is lower, relative to that in countries without such a fund, by about 14 percent.
- The establishment of a stabilization SWF does not in itself ensure that fiscal policy will insulate the domestic economy from commodity price fluctuations since these funds are not a substitute for fiscal policy.

### Caveats, limitations, and scope for future research
- Stabilization SWFs vary in size and deposit/withdrawal rules; the use of a simple binary variable in this study cannot capture those differences.
- Future work could distinguish SWFs by size and by specific deposit/withdrawal rules to capture heterogeneous impacts on fiscal policy behavior.
- Note on data transformation: with the exception of the SWF dummy variable, all variables are in 5-year moving average format.

*Source: wpiea2023133-print-pdf - 6.    CONCLUSION*

---


_Source: https://www.imf.org/-/media/files/publications/wp/2023/english/wpiea2023133-print-pdf.pdf_