## _wp13162

## Source details

**Canonical URL:** [_wp13162](https://www.imf.org/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2013/_wp13162.pdf)

## Other formats

- [Markdown version](/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2013/_wp13162.pdf.md)
- [Structured JSON version](/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2013/_wp13162.pdf.json)

---

### Merging GFSM1986 with GFSM2001
- Major methodological changes addressed
  - Economic classification differences: GFSM2001 uses “net acquisition of non‑financial assets” (net concept); GFSM1986 uses gross capital expenditure.
  - Capital transfers: classified as capital expenditure under GFSM1986, but as current “expense” under GFSM2001.
  - Functional classification: GFSM2001 divides expenditures into 10 categories; GFSM1986 into 14 categories.
  - Reporting basis: GFSM1986 reported on a cash basis only; GFSM2001 mainly on an accrual basis; some GFSM2001 entries mix bases across subcategories within a year.
- Data processing and harmonization steps
  - Source: historical expenditure data reported to the IMF’s GFS yearbook from 1970 to 2010.
  - Conversion approach: followed Wickens(2002) to map GFSM1986 items into GFSM2001 concepts; unified series define capital expenditure as net.
  - Treatment of cash vs accrual reporting:
    - Analysis focuses on expenditure composition; ensured GFSM2001 data used only when all expenditure components of interest are reported on the same accounting basis within the same year.
    - When available on the accrual basis, accrual data were used.
- Dataset coverage and characteristics
  - Time span: 1970–2010.
  - Panel: unbalanced; covers 56 countries.
  - Country income groups: 14 low-, 16 medium-, and 26 high-income countries.
  - Institutional level: consolidated central government (CG) data used for main analysis; later checks with consolidated general government (GG) where available.
  - Caveat: mixing of cash and accrual concepts unavoidable in some cases; key assumptions made to preserve comparability.

### Subcomponents of Expenditure: economic and functional classifications
- Economic classification retained items and rationale
  - Three items retained: compensation of employees; the rest of expense; net acquisition of non‑financial assets.
  - Rest of expense includes: use of goods and services; consumption of fixed capital; interest; subsidies; grants; social benefits; and other expense.
- Functional classification: coverage and exclusions
  - GFSM2001 defines 10 functional categories; dataset covers 8 and excludes:
    - public order and safety
    - environmental protection
  - Exclusion reason: limited availability of these 2 subcomponents over 1970–2010.
- Construction of regressors and limitations
  - Key regressors: ratios of public spending components to total expenditure (spending shares).
  - No decomposition of changes in spending shares into relative prices versus relative quantities due to data limits.
  - Quality of spending not proxied; changes in quality abstracted from analysis.
- Government level and decentralization implications
  - Main analysis uses consolidated CG level because GFSM1986 generally reports only CG.
  - CG-level data can misrepresent national shares when fiscal decentralization differs across subcomponents (e.g., defense centralized; health and education more decentralized).
  - Robustness checks use GG-level data where both CG and GG coexist under GFSM2001.

### Additional macro variables and data construction
- Macroeconomic controls
  - GDP and exchange rates from WEO or IFS.
  - Key control: average years of schooling between ages 25 and 64 from Barro and Lee (2010).
  - Additional robustness controls: inflation rates; openness (imports+exports/GDP); population growth; terms of trade growth.
- Sample aggregation and graphical patterns
  - Countries ordered by GDP per capita (PPP); Figure 1 divides sample into deciles.
  - Coverage noted:
    - Economic classification: 86 countries reported all relevant components at least once (1970–2010).
    - Functional classification: 102 countries reported all relevant components at least once.
    - Transport and communication available for 101 countries.
  - Stylized cross‑country facts:
    - Total expenditure (share of GDP) increases with income until GDP per capita ≈ 20,000 US dollars (PPP), then flattens/slightly decreases (non‑monotonic; Wagner’s law).
    - Expense drives total expenditure behavior; compensation of employees relatively stable; social benefits rise with development; net acquisition of non‑financial assets declines with income.
    - Functional: health and social protection rise with development; education and transport and communication do not clearly increase with income.
    - Defense spending shows a clear downward pattern, large fall during the 1990s.
- Time aggregation and panel structure
  - Use 5‑year non‑overlapping averages: 1971–75, 1976–1980, ..., 2006–2010.
  - 5‑year average computed only if at least 3 observations exist within the period (compromise to smooth cycles and preserve sample).

### Empirical specification and estimation
- Theoretical basis: neoclassical growth models (Solow‑Swan); conditional convergence framework.
- Dynamic specification (summarized):
  - Growth equation: y_{i,t} - y_{i,t-1} = (a - 1) y_{i,t-1} + b x_{i,t-1} + \bar{f}'_{i,t} + n_i + e_{i,t}.
  - Fiscal components in \bar{f}'_{i,t}: d \bar{e}_{i,t} + Σ g_j \bar{s}_{i,j,t} + Σ h_j \bar{z}_{i,j,t}.
  - Reparameterization: coefficients on expenditure components interpreted as reallocation effects (rise in one component compensated by fall in omitted component).
- Estimation method: system GMM (Arellano and Bover 1995; Blundell and Bond 1997)
  - Reasons: handles unbalanced panels, fixed effects, endogeneity via internal instruments; better for persistent series than difference GMM.
  - Moment conditions specified for difference and level equations.
  - Instrument reduction: collapse option and limiting lags to avoid too many instruments (Roodman 2009b).
- Specification tests employed: Arellano‑Bond AR(1) and AR(2); Hansen test; Difference Hansen tests.
- Alternative estimators considered but not used (MG, PMG) due to large T requirements and business cycle concerns.

### Economic classification regression results (selected)
- Reallocation between expense and non‑financial assets (Table 1)
  - Main estimates (observations: 190; No. of countries: 52; No. of instruments: 51)
    - Nonfin. Assets/Total Exp coefficient (column (1)): 1.375*** (0.436).
    - Expense/Total Exp implicit: 1.209*** (0.391) in column (1).
    - Total exp/GDP: −0.416* (0.250) in column (1); −0.537** (0.248) in column (2).
    - Initial Human Capital: 4.600*** (0.996) column (1); 4.452*** (0.998) column (2).
  - Specification tests (both columns):
    - Arellano‑Bond AR(1) p‑value: 0.01
    - Arellano‑Bond AR(2) p‑value: 0.11
    - Hansen p‑value: 0.55 (col 1), 0.58 (col 2)
    - Diff Hansen 1 p‑value: 0.80 (col 1), 0.95 (col 2)
    - Diff Hansen 2 p‑value: 1.00 (both)
  - Interpretation and caveat:
    - Reallocating toward net acquisition of non‑financial assets (capital) at the expense of current spending associated with positive growth; effect not robust across specifications.

### Functional classification regression results (selected)
- Focus components: defense; transport and communication (Tracom); health; education; social protection.
- Pairwise compositional effects (Table 2)
  - Only education spending shows statistically significant growth‑enhancing effects.
  - Significant reallocations:
    - Increase in Education financed by fall in Health: Enhancing** (5 percent significance).
    - Increase in Education financed by fall in Social protection: Enhancing** (5 percent significance).
  - All other pairwise reallocations among the five components reported as Insignificant.
- Education detailed effects (Table 3)
  - Selected coefficients on Educ/Total exp:
    - Column (1) (compensating = All the rest): 1.098* (0.646).
    - Column (3) (compensating = Health): 1.553** (0.747).
    - Column (4) (compensating = Social Prot): 1.090** (0.411).
    - Column (5) (compensating = Tracom): −0.816 (2.251).
  - Observations: 175 in columns (1)–(4); 151 in column (5).
  - No. of countries: 56 in columns (1)–(4); 55 in column (5).
  - Interpretation:
    - A 1 percentage point increase in education spending offset by a 1 percentage point fall in social protection increases growth by 1.09 percentage points over 5 years (≈ 0.2 percentage points per annum).
    - Slightly larger effect when compensating component is health (1.553**).
- Transport and communication (Table 4)
  - Tracom/Total exp coefficients not statistically significant across specifications.
  - Observations: 151; No. of countries: 55.

### Robustness and additional explanatory variables
- Main robust findings summary
  1. Reallocation toward capital expenditure financed by reductions in current spending shows positive but not robust association with growth.
  2. Reallocation toward education spending appears statistically significantly associated with higher growth (most important result).
  3. Reallocations among other key functional components generally not statistically significant.
- Lagged fiscal variables (Table 5)
  - Educ/Total exp coefficients with one‑period lag:
    - Column (1) (All the rest): 1.278* (0.683).
    - Column (4) (Social Prot): 1.234** (0.552).
  - Observations: 175 in columns (1)–(4); 151 in column (5).
- Excluding G20‑Advanced (Table 6)
  - Educ/Total exp coefficients:
    - Column (4) (Social Prot): 0.804* (0.446).
  - Observations: 142 in columns (1)–(4); 125 in column (5).
  - Finding: education importance persists but some coefficients lose significance in reduced sample.
- Sensitivity to additional controls (Table 7)
  - Baseline replication (Column (1)): Educ/Total exp 1.098* (0.646); Observations 175; No. of countries 56.
  - Adding Inflation (Column (2)): Educ/Total exp 1.417** (0.580); Inflation −0.673* (0.379); Observations 150; No. of countries 50.
  - Adding Openness (Column (3)): Educ/Total exp 0.594 (0.601); Observations 158; No. of countries 53.
  - Adding Population growth (Column (4)): Educ/Total exp 1.380*** (0.515); Observations 165; No. of countries 52.
  - Adding Terms of trade growth (Column (5)): Educ/Total exp 1.555*** (0.557); Observations 153; No. of countries 50.
  - Conclusion: education spending shows robust positive association except when openness is added (Column (3) not significant).
- Central vs General Government (Table 8)
  - Fixed effects comparison (59 observations; Number of countries 32)
    - CG level (Column (1)): Educ/Total exp 0.286 (0.432); R‑squared 0.45.
    - GG level (Column (2)): Educ/Total exp 0.723* (0.384); R‑squared 0.54.
    - CG, compensating = Social Spend. (Column (3)): Educ/Total exp 0.602 (0.522).
    - GG, compensating = Social Spend. (Column (4)): Educ/Total exp 0.712* (0.352); R‑squared 0.56.
  - Caveat: short panel and lack of instrumentation make fixed effects results likely inconsistent; comparison intended as indicative.
  - Overall: education coefficients positive at both CG and GG levels; differences not statistically significant by paired t‑test noted in source.

### Main findings, magnitudes, and policy implications
- Main findings
  - Reallocations of government spending holding overall expenditure fixed do not generally show robust associations with growth.
  - Key exception: increases in education spending offset by reductions in social protection are robustly associated with higher growth.
  - Education–social protection reallocation robust to:
    - alternative timing (lagged fiscal effects),
    - excluding the richest countries,
    - addition of extra controls (inflation, population growth, terms of trade growth), though openness weakens significance.
  - CG versus GG: education result holds in both measures in smaller comparative sample.
  - Capital spending effects positive in some specifications but lack robustness.
- Magnitude highlighted in unit title
  - "0.286 percentage points increase in growth per annum" (unit title reference).
  - Example from Table 3: 1.090** (0.411) on Educ/Total exp implies 1.09 percentage points over 5 years (≈ 0.2 percentage points per annum) when compensated by social protection; GG fixed effect example shows 0.723* (0.384) in annual growth units.
- Policy implications
  - When overall expenditure cannot increase, reallocating toward education while reducing social protection is associated with higher growth in the sample/specifications examined.
  - Public capital spending can support long‑run growth if financed by reductions in current spending, but evidence is less robust and depends on spending efficiency.
  - Policymakers should weigh growth effects alongside other objectives (employment, income equality); social protection may serve distributional goals despite not being growth‑enhancing in these estimates.

### Robustness, limitations, and caveats
- Quality and efficiency of spending not observed; results reflect quantities.
- Fiscal decentralization and CG vs GG measurement issues can bias component shares when decentralization differs across functions and over time.
- Endogeneity and omitted variables (e.g., corruption, demographic change) addressed via system GMM instrumentation where feasible; sample constraints limit inclusion of some controls.
- Heterogeneity: subgroup analyses (HICs vs LICs+MICs) generally show no statistically significant differences, but some education effects significant only in HICs in particular specifications.

### Key statistics and sample descriptors (as presented)
- Sample growth (5‑year):
  - Mean growth of real GDP pc (over 5 years): 11.949.
  - Standard Deviation: 9.61.
  - Minimum: −24.21.
  - Maximum: 39.98.
- Economic classification (shares in percent)
  - Total Exp/GDP: Mean 33.03; Standard Deviation 8.90; Minimum 13.05; Maximum 53.87.
  - Compensation of Employees/Total Spend.: Mean 20.20; Standard Deviation 9.47; Minimum 5.85; Maximum 54.04.
  - Rest of Expense/Total Spend.: Mean 72.56; Standard Deviation 14.05; Minimum 33.83; Maximum 92.47.
  - Nonfin. Assets/Total Spend.: Mean 7.24; Standard Deviation 6.50; Minimum 1.24; Maximum 36.47.
- Functional classification (shares in percent)
  - Total exp/GDP: Mean 28.40; Standard Deviation 9.72; Minimum 10.43; Maximum 52.88.
  - Defense/Total Spend.: Mean 9.58; Standard Deviation 7.00; Minimum 1.98; Maximum 33.06.
  - Transport and communication/Total Spend.: Mean 5.69; Standard Deviation 2.80; Minimum 1.26; Maximum 13.72.
  - Health/Total Spend.: Mean 8.41; Standard Deviation 4.79; Minimum 0.83; Maximum 20.29.
  - Education/Total Spend.: Mean 11.39; Standard Deviation 5.53; Minimum 1.82; Maximum 23.34.
  - Social Prot./Total Spend.: Mean 27.11; Standard Deviation 15.43; Minimum 1.33; Maximum 54.89.
- CG versus GG summary (static panel regression sample, 32 countries: 20 HICs, 9 MICs, 3 LICs)
  - Total exp/GDP (CG): Mean 33.48; Std. dev. 7.04; Minimum 15.73; Maximum 44.29.
  - Total exp/GDP (GG): Mean 40.24; Std. dev. 7.45; Minimum 20.52; Maximum 57.66.
  - Educ/Total exp (CG): Mean 9.75; Std. dev. 3.96; Minimum 2.32; Maximum 20.44.
  - Educ/Total exp (GG): Mean 13.60; Std. dev. 2.77; Minimum 6.35; Maximum 19.69.
  - Social/Total exp (CG): Mean 49.93; Std. dev. 9.62; Minimum 8.22; Maximum 63.19.
  - Social/Total exp (GG): Mean 48.92; Std. dev. 8.11; Minimum 26.23; Maximum 59.74.

*Source: _wp13162 — content excerpted from the supplied IMF PDF chapter.*

### 1.  Merging GFSM1986 with GFSM2001 . . . . . . . . . . . . . . . . . . .8

### 1. Merging GFSM1986 with GFSM2001

### Major methodological changes addressed
- Classification differences between GFSM1986 and GFSM2001:
  - Economic classification: both can be roughly divided into “current” and “capital” expenditures, but definitions differ.
  - Under GFSM2001, capital expenditure is denoted as “net acquisition of non-financial assets” and adopts a net concept (government revenue from the sales of fixed capital assets are taken into account).
  - Under GFSM1986, capital expenditure adopts a gross concept (revenue from capital sales is not deducted).
  - Capital transfers: part of capital expenditure under GFSM1986, but part of current expenditure (denoted as “expense”) under GFSM2001.
  - Functional classification: GFSM2001 divides expenditures into 10 functional categories; GFSM1986 divides them into 14 categories.
- Reporting basis differences:
  - GFSM1986 reporting is on a cash basis only.
  - GFSM2001 reporting is mainly on an accrual basis.
  - Accrual basis: flows are recorded when a transaction accrues, independently of the flow of cash.
  - Cash basis: transactions are recorded when cash effectively flows.
  - For some countries and years under GFSM2001, different subcategories are reported in different accounting bases even within a given year.

### Data processing and harmonization steps
- Source data retrieved: all historical expenditure data reported to the IMF’s GFS yearbook from 1970 to 2010.
- Conversion approach:
  - Followed Wickens(2002) to convert GFSM1986 expenditure items into the concepts defined by GFSM2001.
  - Unified series definition: capital expenditure defined as a net concept; functional components mapped into the 10 GFSM2001 functional categories.
- Treatment of cash vs accrual reporting:
  - Because the analysis focuses on expenditure composition (expenditure shares among subcomponents), timing differences between cash and accrual bases are less problematic provided all subcomponents within a year are reported on the same basis.
  - Ensured that GFSM2001 data were taken only when all expenditure components of interest are reported on the same accounting basis within the same year.
  - Whenever data for all components were available on the accrual basis (considered more economically relevant), those accrual-based data were used.

### Dataset coverage and relevant characteristics
- Time span: 1970–2010 (covering 40 years).
- Panel characteristics:
  - Unbalanced panel.
  - Covers a total of 56 countries.
  - Country income groups in the assembled dataset: 14 low-, 16 medium-, and 26 high-income countries.
  - Consolidated central government level data (with later checks for consolidated general government level).
- Purpose and caveats:
  - Objective: construct comparable expenditure series across the methodological transition from GFSM1986 to GFSM2001 to study compositional effects of public expenditure on growth.
  - Key caveat: mixing of cash and accrual concepts was unavoidable in some cases; important assumptions were made to bridge methodologies while aiming to preserve comparability of fiscal data across periods.

*Source: “1.  Merging GFSM1986 with GFSM2001” — dataset construction described in the paper based on IMF’s GFS yearbook data (1970–2010).*

### 2.  Subcomponents of Expenditure: Economic and Functional Classifications

### 2.  Subcomponents of Expenditure: Economic and Functional Classifications

### Economic classification: retained items and rationale
- Dataset follows categorization under GFSM2001 but does not cover all detailed subcomponents in the manual.
- Main distinction retained: expense versus net acquisition of non-financial assets.
- Compensation of employees is separated from the rest of expense due to frequent interest in the ‘wages’ subcomponent in the literature (e.g., Gupta and others(2005)).
- The dataset thus has the three items in the economic classification:
  - compensation of employees (as a proxy of wages)
  - the rest of expense
  - net acquisition of non-financial assets
- Rest of expense consists of: use of goods and services; consumption of fixed capital; interest; subsidies; grants; social benefits; and other expense.

### Functional classification: coverage and exclusions
- Total expenditure is grouped into 10 categories under GFSM2001; the dataset covers 8 of them and excludes:
  - public order and safety
  - environmental protection
- Reason for exclusion: limited availability of these 2 subcomponents throughout the sample period of 1970 to 2010.
- The 10 functional categories defined in GFSM2001 are: 1) general public services; 2) defense; 3) public order and safety; 4) economic affairs; 5) environmental protection; 6) housing and community amenities; 7) health; 8) recreation, culture, and religion; 9) education; and 10) social protection.

### Construction of regressors and limitations related to prices and quality
- Key regressors: ratios of the different public spending components to total expenditure (spending shares).
- Deflators among different spending components may differ, implying spending shares can be affected by both:
  - changes in relative prices
  - changes in relative quantities
- Due to data availability issues:
  - the dataset does not decompose observed changes in expenditure shares into changes in relative prices versus relative quantities.
  - issues related to the quality of spending are also left out.
- Note on quality: changes in the quality of public goods can affect quantities required for a given provision of a public service; lack of proxies for quality by function forces abstraction from this source of variation.

### Government level: institutional coverage and choice of series
- Under GFSM1986 countries report data at most at the consolidated central government (CG) level; under GFSM2001 they also provide consolidated general government (GG) level.
- Availability of lower-government (state/local) data is severely limited under GFSM1986.
- Main analysis uses consolidated CG level.
  - The consolidated CG level can be further divided based on whether the institutional unit is financed by the legislative budget or by extrabudgetary sources.
  - The CG unit based on the legislative budget is called budgetary CG.
  - Some works (e.g., Easterly and Rebelo(1993)) use budgetary CG along with consolidated CG to increase observations; however, non-trivial discrepancies are often found between consolidated CG and budgetary CG in some expenditure subcomponents such as social protection.
  - The dataset relies only on consolidated CG data.

### Implications of fiscal decentralization and potential biases
- When the degree of fiscal decentralization (measured by the share of spending at the CG level relative to that of the GG level) differs across subcomponents, CG-level data may not accurately capture national shares for those subcomponents.
- Evidence and examples:
  - Some subcomponents (e.g., defense) tend to be centralized in most countries.
  - Others (e.g., health and education) tend to be more decentralized.
  - Wyss and Lorenz(2000) discuss the vastly decentralized nature of the health sector in Switzerland.
  - Dziobek, Mangas, and Kufa(2011) report that in some countries (e.g., Spain and Switzerland), the level of fiscal decentralization in total expenditure has been unstable over time; their report does not cover different expenditure subcomponents except ‘compensation of employees’ in the economic classification, but unstable trends may be caused by decentralization of particular subcomponents such as health and education.
- Using CG-level data can be more problematic in panel analysis when trends of fiscal decentralization differ across components over time.

### Robustness check using GG-level data where available
- Recognizing limitations of CG data, the study will check robustness using GG-level data where both CG-level and GG-level data coexist under GFSM2001.
- Robustness approach:
  - Use a restricted sample with coexistence of CG-level and GG-level data.
  - Present evidence on whether results obtained with CG-level and GG-level data are statistically the same.
- Purpose: provide certain justification for relying on CG-level data in the main analysis.

*Source: _wp13162 - 2.  Subcomponents of Expenditure: Economic and Functional Classifications*

### 4.  Additional Macro Variables

### 4.  Additional Macro Variables

### Macro variables and controls
- Dataset contains macroeconomic variables including GDP and exchange rates, obtained mainly from WEO or IFS databases of the IMF.
- Key control variable in reference regressions:
  - Average years of schooling between ages 25 and 64 (proxy for human capital accumulation) from the Barro and Lee (2010) dataset.
- Additional controls used in robustness checks:
  - Inflation rates
  - Openness (value of imports and exports relative to GDP)
  - Population growth
  - Terms of trade growth
- Note on use of GDP and exchange rates:
  - GDP information is used to create the ratio of total expenditure over GDP (to control for level effects).
  - Exchange rates are necessary to convert GFS data reported in national currency into US dollars, since nominal GDP in current prices is taken from WEO in US dollars.

### Graphical description — cross-country patterns and stylized facts
- Sample and aggregation:
  - Figures pool countries and order data by GDP per capita (PPP prices).
  - Figure 1 divides the sample into deciles by GDP per capita (PPP).
  - Figure 1 coverage:
    - Economic classification: covers 86 countries which have reported all relevant components at least once in 1970 to 2010.
    - Functional classification: covers 102 countries which have reported all relevant components at least once.
    - Transport and communication within economic affairs limited to 101 countries.
- Key empirical patterns from Figures 1–3 (median values in percent of GDP, by decile; and 10‑year unweighted averages):
  - Total expenditure (share of GDP) increases with income until GDP per capita reaches around 20,000 US dollars (PPP), then flattens and slightly decreases — described as a non-monotonic relation consistent with Wagner’s law.
  - The behavior of total expenditure is essentially driven by the expense subcomponent.
  - Compensation of employees (wages) shows a relatively more stable pattern across development.
  - Social benefits (within expense) exhibit an upward trend as countries become richer (not shown in figure notes).
  - Net acquisition of non-financial assets decreases noticeably with higher income.
  - Functional classification:
    - Health and social protection (public social spending) increase with development; spending on education and transport and communication do not appear to increase with income.
    - Over time, for HICs and for LICs and MICs combined, total expenditure and expense are higher in HICs; upward trend until the 1990s and slight fall during the 2000s for HICs.
    - Net acquisition of non-financial assets shows a downward trend over the 40-year period in both groups.
    - Transport and communication (proxy for economic infrastructure) declines notably in HICs.
    - Education spending relative to GDP has been relatively similar and stable across groups and time.
    - Health and social protection spending generally increased over time, with levels significantly higher in HICs.
    - Defense spending shows a clear downward pattern in both groups, with a large fall during the 1990s coinciding with the end of the Cold War.
  - Composition of expenditure (functional classification, 10‑year unweighted avg, percent of total expenditure):
    - Rapid increase in shares of social protection and health; shares are significantly smaller in LICs and MICs.
    - These upward trends have been accommodated by lowering shares in remaining categories including transport and communication and defense.
    - Share of education spending has not changed much over the last 40 years in HICs.
  - Notes on sample sizes for decade averages (Figure 2 and Figure 3):
    - For economic classification (Figure 2): unweighted averages for 19 HICs and 7 MICs and LICs.
    - For functional classification (Figure 2): 14 HICs and 6 MICs and LICs; transport and communication availability limited (7 HICs and 7 MICs and LICs).
    - For Figure 3 functional classification focused where transport and communication data are available: 7 HICs and 7 MICs and LICs.
    - Functional “total” excludes 2 of 10 functional subcomponents (‘public order and safety’, and ‘environmental protection’); remaining 8 components sum almost equal to actual total due to small shares of the excluded components.

### Descriptive statistics and data construction
- Time aggregation and sample structure:
  - Use 5-year non-overlapping averages to abstract from business cycle effects, yielding a maximum of 8 observations per country: 1971–75, 1976–1980, ..., 2006–2010.
  - Dataset is an unbalanced panel; 5-year average computed only if at least 3 observations are present within each 5-year period.
  - Rationale: allowing 1 observation per 5-year average maximizes observations but fails to smooth business cycles; requiring full 5 observations reduces sample severely; threshold of at least 3 observations chosen as compromise.

### Regression analysis — empirical specification and methodology
- Theoretical motivation:
  - Empirical specification motivated by neoclassical growth models (Solow-Swan); relates real GDP per capita growth to state variables (initial position) and control/environmental variables (steady-state determinants).
  - Implications referenced: conditional convergence and effects of steady-state changes on growth during adjustment.
- Dynamic empirical specification (equations reproduced in the source):
  - Growth equation (Eq.(1)):
    y_{i,t} - y_{i,t-1} = (a - 1) y_{i,t-1} + b x_{i,t-1} + \bar{f}'_{i,t} + n_i + e_{i,t}.
  - Fiscal components within \bar{f}'_{i,t} (Eq.(2)):
    \bar{f}'_{i,t} = d \bar{e}_{i,t} + \sum_{j=1}^{m} g_j \bar{s}_{i,j,t} + \sum_{j=1}^{k} h_j \bar{z}_{i,j,t}.
    - \bar{e}_{i,t}: share of total public expenditure to GDP (controls for level effect).
    - \bar{s}_{i,j,t}: share of expenditure components in total expenditure.
    - \bar{z}_{i,j,t}: other control/environmental variables (inflation, openness, population growth, terms of trade growth).
  - Reparameterization to avoid perfect multicollinearity by leaving out one component (component m), leading to Eq.(3) and rewriting as dynamic equation with lagged dependent variable on RHS (Eq.(4)):
    y_{i,t} = a y_{i,t-1} + b x_{i,t-1} + d \bar{e}_{i,t} + g_m + \sum_{j=1}^{m-1} (g_j - g_m) \bar{s}_{i,j,t} + \sum_{j=1}^{k} h_j \bar{z}_{i,j,t} + n_i + e_{i,t}.
  - Interpretation: coefficients on expenditure components are reallocation effects—effects on growth of a rise in a component compensated by a fall in the omitted component.
- Estimation approach:
  - Estimate the dynamic panel using system GMM (Arellano and Bover 1995; Blundell and Bond 1997) for reasons:
    - Flexibility with unbalanced panels.
    - Ability to handle country fixed effects.
    - Handling potential endogeneity via internal instruments (lagged values).
    - System GMM recommended over difference GMM when series are persistent to reduce finite-sample bias and weak instrument problems.
  - Moment conditions used (difference part and level part) are specified in the source:
    - Difference part: E[ y_{i,t-s} (e_{i,t} - e_{i,t-1}) ] = 0 and analogous for x and other explanatory variables, s ≥ 2, t = 3,...,T.
    - Level part: E[ (y_{i,t-1} - y_{i,t-2})(n_i + e_{i,t}) ] = 0 and analogous for \bar{x}, with t ≥ 3.
  - Instrument reduction strategies:
    - Combine instruments through additions to smaller sets (collapse option) to reduce instrument count.
    - Rationale: Too many instruments relative to number of countries undermines reliability (Roodman 2009b).
    - For some regressions, alternative reduction used: only one lag instead of all available lags.
- Specification tests implemented to validate the system approach:
  - Arellano-Bond test for absence of serial correlation in the error term.
  - Hansen test for overall validity of instruments.
  - Difference Hansen test for validity of instrument sets used in the level part of the system.
- Comparison with alternative estimators:
  - Mean-Group (MG) and Pooled Mean-Group (PMG) estimators allow heterogeneity in long-run relations and short-run adjustments but require large T (annual data), which exacerbates business cycle concerns; unbalanced nature of the highly disaggregated fiscal dataset prevents use of these estimators in this study.
- Endogeneity and omitted variables highlighted as important concerns:
  - Reverse causality, omitted variables such as corruption (Mauro 1995, 1998) and demographic change (aging) can affect both expenditure composition and growth; internal instrumentation in GMM used to address these concerns.

*Source: _wp13162 - 4.  Additional Macro Variables (authors’ calculations and text as presented in the source PDF)*

### 1.  Economic Classification

### 1.  Economic Classification

### Reallocation between expense and non-financial assets (Table 1)
- Key regression setup: dependent variable is GDP per capita growth over 5 years; compensating component in column (1) is “Expense”, in column (2) is “Nonfinancial assets”.
- Main estimated effects:
  - Nonfin. Assets/Total Exp coefficient (column (1)): 1.375*** (0.436). Interpretation: a rise in the share of non-financial assets by 1 percentage point, offset by an equal fall in expense, increases growth by about 1.4 percentage points over the 5 years period (thus about 0.27 percentage points per annum).
  - Expense/Total Exp coefficient (implicit in column (1) when non-financial assets are compensating): 1.209*** (0.391).
  - Total exp/GDP: −0.416* (0.250) in column (1) and −0.537** (0.248) in column (2).
  - Initial GDP p.c.: −0.805 (0.855) in column (1) and −0.763 (0.782) in column (2).
  - Initial Human Capital: 4.600*** (0.996) in column (1) and 4.452*** (0.998) in column (2).
  - Constant: −20.711 (16.789) in column (1) and 106.409*** (32.142) in column (2).
- Sample and instruments:
  - Observations: 190 in both columns.
  - No. of countries: 52 in both columns.
  - No. of instruments: 51 in both columns.
- Specification tests (both columns):
  - Arellano-Bond AR(1), p-value: 0.01
  - Arellano-Bond AR(2), p-value: 0.11
  - Hansen, p-value: 0.55 (column (1)), 0.58 (column (2))
  - Diff Hansen 1, p-value: 0.80 (column (1)), 0.95 (column (2))
  - Diff Hansen 2, p-value: 1.00 in both columns
- Interpretation and caveats:
  - Reallocating toward net acquisition of non-financial assets (capital spending) financed by offsetting reductions in expense (current spending) is associated with positive growth effects.
  - Statistical significance is sensitive to sample and specification changes; results under the economic classification tend to be non-robust.

### 2.  Functional Classification

### Overview and focus
- Five functional components analyzed: defense, transport and communication (Tracom), health, education, and social protection.
- Hypothesis: increases in education, infrastructure, defense, and health are potentially growth-enhancing while social protection is not directly productive; growth-enhancing reallocations expected when increases in those categories are financed by falls in social protection.

### Summary of compositional effects (Table 2)
- Only education spending shows statistically significant growth-enhancing effects.
- Statistically significant cases:
  - Increase in Education financed by a fall in Health: Enhancing** (significant at 5 percent)
  - Increase in Education financed by a fall in Social protection: Enhancing** (significant at 5 percent)
- All other pairwise reallocations among the five components are reported as Insignificant in the table.

### Detailed effects of education spending (Table 3)
- Dependent variable: GDP per capita growth over 5 years.
- Key coefficients on Educ/Total exp:
  - Column (1) (compensating factor = All the rest): 1.098* (0.646)
  - Column (2) (compensating factor = Defense): 0.539 (0.903)
  - Column (3) (compensating factor = Health): 1.553** (0.747)
  - Column (4) (compensating factor = Soc. Prot): 1.090** (0.411)
  - Column (5) (compensating factor = Tracom): −0.816 (2.251)
- Selected additional coefficients and statistics:
  - Total exp/GDP: 0.369 (0.382) in column (1); 0.532* (0.297) in column (3); 1.063* (0.557) in column (5)
  - Initial GDP p.c.: range from −0.066 (2.666) to −2.155 (2.690) across columns
  - Initial human capital: 5.644*** (1.862) in column (1); 5.216*** (1.888) column (2); 4.127** (1.954) column (3); 4.963*** (1.407) column (4); 2.349* (1.372) column (5)
  - Observations: 175 in columns (1)–(4); 151 in column (5)
  - No. of countries: 56 in columns (1)–(4); 55 in column (5)
  - No. of instruments: 33 in column (1); 39 in columns (2)–(4); 37 in column (5)
- Interpretation:
  - A 1 percentage point increase in education spending offset by a 1 percentage point fall in social protection spending causes a 1.09 percentage points increase in growth over the 5-year period (about 0.2 percentage points per annum).
  - Slightly larger effect when the compensating component is health (1.553** in column (3)).
  - Education is generally growth enhancing relative to all other functional components in these specifications.

### Transport and communication spending (Table 4)
- Dependent variable: GDP per capita growth over 5 years.
- Key coefficients on Tracom/Total exp across columns:
  - Column (1): 1.246 (1.937)
  - Column (2): −1.097 (1.353)
  - Column (3): 0.414 (1.494)
  - Column (4): 0.241 (2.240)
  - Column (5): 1.420 (1.881)
- Selected additional coefficients and statistics:
  - Total exp/GDP ranges from 0.076 (0.654) to 1.423*** (0.441) across columns.
  - Initial Human capital: 2.344* (1.327) column (1); 2.873* (1.535) column (2); 2.067 (1.473) column (3); 2.106 (1.305) column (4); 5.899* (3.447) column (5)
  - Observations: 151 in all columns
  - No. of countries: 55 in all columns
- Interpretation:
  - None of the coefficients on the share of transport and communication to total spending are statistically significant in these specifications. This holds even when its rise is offset by falls in health or social protection.

### 3.  Robustness

### Main robust findings (summary)
- Three main results:
  1. Reallocation toward capital expenditure (net acquisition of non-financial assets) financed by reductions in current spending shows a positive but statistically not robust association with growth.
  2. Reallocation toward education spending appears to have a statistically significant association with higher growth (most important and examined result).
  3. Reallocations among other key functional components (including increases in economic infrastructure financed by falls in social spending) do not show statistically significant relations with growth.

### Lagged fiscal variables (Table 5)
- Alternative timing assumption: steady state during period t−1 to t affected solely by fiscal changes in period t−1.
- Coefficients on Educ/Total exp:
  - Column (1) (All the rest): 1.278* (0.683)
  - Column (2) (Defense): 0.761 (0.826)
  - Column (3) (Health): 1.423 (0.959)
  - Column (4) (Social Prot): 1.234** (0.552)
  - Column (5) (Tracom): 0.785 (1.235)
- Observations: 175 in columns (1)–(4); 151 in column (5)
- No. of countries: 56 in columns (1)–(4); 55 in column (5)
- Interpretation:
  - The importance of education spending as growth-enhancing is reconfirmed under lagged fiscal specification; coefficients with social protection as compensating component are similar to Table 3, though the reallocation with health is not statistically significant in this specification.

### Different development levels — excluding G20-Advanced (Table 6)
- Purpose: partial check by excluding G20-Advanced countries to examine robustness in a smaller, less-developed sample.
- Coefficients on Educ/Total exp:
  - Column (1) (All the rest): 0.741 (0.528)
  - Column (2) (Defense): 0.811 (1.003)
  - Column (3) (Health): 1.154 (0.756)
  - Column (4) (Social Prot): 0.804* (0.446)
  - Column (5) (Tracom): −0.601 (2.112)
- Observations: 142 in columns (1)–(4); 125 in column (5)
- Number of ifs: 49 in columns (1)–(4); 48 in column (5)
- No. of instruments: 33 in column (1); 39 in columns (2)–(4); 35 in column (5)
- Interpretation:
  - The importance of education spending persists in this reduced sample, though the coefficient with social protection as compensating factor is somewhat smaller than in the full sample.
  - Significance when compensating factor is health is reduced, indicating less robust effect across subsamples.

*Source: _wp13162 - 1.  Economic Classification*

### 3.  Additional Explanatory Variables

### 3.  Additional Explanatory Variables

### Sensitivity check: adding one extra control variable at a time
- Purpose: check sensitivity of results on education spending by adding one additional control/environmental variable at a time: the inflation rate, the degree of openness, population growth, and terms of trade growth.
- Treatment of variables: treated as endogenous; dynamic GMM framework used where feasible, but additional endogenous controls were added one-by-one to avoid an excessive number of instruments.
- Note on dataset coverage: dataset spans a long period and includes many countries including LICs; some desirable controls (e.g., institutional variables like corruption) cannot be included because they dramatically reduce sample size.

### Table 7 — Robustness Check with Additional Variables (Dependent variable: GDP per capita growth over 5 years)
- Reported coefficients and robust standard errors (in parentheses) for the share of education spending and each added control:

  - Column (1): baseline replication (no extra control)
    - Educ/Total exp: 1.098* (0.646)
    - Compensating factor: All the rest
    - Observations: 175
    - No. of countries: 56
    - No. of instruments: 33
    - Arellano-Bond AR(1), p-value: 0.01
    - Arellano-Bond AR(2), p-value: 0.19
    - Hansen, p-value: 0.26
    - Diff Hansen 1, p-value: 0.42
    - Diff Hansen 2, p-value: 0.68

  - Column (2): adding Inflation
    - Educ/Total exp: 1.417** (0.580)
    - Inflation: -0.673* (0.379)
    - Compensating factor: All the rest
    - Observations: 150
    - No. of countries: 50
    - No. of instruments: 39
    - Arellano-Bond AR(1), p-value: 0.06
    - Arellano-Bond AR(2), p-value: 0.15
    - Hansen, p-value: 0.64
    - Diff Hansen 1, p-value: 0.97
    - Diff Hansen 2, p-value: 0.86

  - Column (3): adding Openness
    - Educ/Total exp: 0.594 (0.601)
    - Openness: 0.100 (0.106)
    - Compensating factor: All the rest
    - Observations: 158
    - No. of countries: 53
    - No. of instruments: 39
    - Arellano-Bond AR(1), p-value: 0.01
    - Arellano-Bond AR(2), p-value: 0.15
    - Hansen, p-value: 0.29
    - Diff Hansen 1, p-value: 0.24
    - Diff Hansen 2, p-value: 0.84

  - Column (4): adding Population growth
    - Educ/Total exp: 1.380*** (0.515)
    - Population growth: -1.340 (3.716)
    - Compensating factor: All the rest
    - Observations: 165
    - No. of countries: 52
    - No. of instruments: 39
    - Arellano-Bond AR(1), p-value: 0.01
    - Arellano-Bond AR(2), p-value: 0.28
    - Hansen, p-value: 0.44
    - Diff Hansen 1, p-value: 0.55
    - Diff Hansen 2, p-value: 0.37

  - Column (5): adding Terms of trade growth
    - Educ/Total exp: 1.555*** (0.557)
    - Terms of trade growth: 0.516 (0.782)
    - Compensating factor: All the rest
    - Observations: 153
    - No. of countries: 50
    - No. of instruments: 39
    - Arellano-Bond AR(1), p-value: 0.07
    - Arellano-Bond AR(2), p-value: 0.18
    - Hansen, p-value: 0.62
    - Diff Hansen 1, p-value: 0.89
    - Diff Hansen 2, p-value: 0.47

- Key findings from Table 7:
  - Education spending shows a robust positive association with higher growth in all specifications except the case with openness as the additional control (Column (3) where the coefficient is not statistically significant).
  - Added controls have coefficients consistent with prior literature (e.g., inflation having a negative impact on growth in Column (2)).
  - Diagnostic tests (Arellano-Bond AR(1) and AR(2); Hansen and Diff Hansen tests) reported for each specification; no uniform rejection of instrument validity across specifications.

### D. Central vs General Government Level — motivation and approach
- Question: do main results on education spending hold when measured at the General Government (GG) level rather than the Central Government (CG) level?
- Data constraints: GFSM2001 dataset with both CG and GG consolidated levels available for a smaller sample; GFSM1986 yearbook discarded for this check. Resulting sample covers at most three 5-year periods (1996–2000, 2001–2005, 2006–2010) and often only two periods depending on migration timing to GFSM2001.
- Estimation approach: static fixed effects model (dynamic GMM with internal instruments not feasible due to short panel and smaller coverage). Dependent variable: average of the annual growth rate of real GDP per capita over 5 years. Independent variables: same as reference case (including initial real GDP per capita, initial human capital, relevant fiscal variables, fixed effects, and time dummies).
- Caveat: without instruments, results likely inconsistent; limited sample reduces estimation accuracy. Purpose is comparative (CG versus GG), not to obtain precise reallocation-effect estimates.

### Table 8 — Effects of Education Spending at the Central and General Government Levels (Dependent variable: Annual GDP per capita growth, averaged over 5 years)
- Fixed effects model. Robust standard errors in parentheses. Time dummies not shown.

- Coefficients and statistics by column:

  - Column (1): CG level, compensating factor = All the rest
    - Total exp/GDP: -0.240 (0.162)
    - Educ/Total exp: 0.286 (0.432)
    - Spend. ex. Social and Educ spend./Total exp: -0.327 (0.300)
    - Initial GDP p.c.: -8.294** (3.121)
    - Initial human capital: 0.290 (0.998)
    - Constant: 91.570** (34.109)
    - Level of government: CG
    - Observations: 59
    - R-squared: 0.45
    - Number of countries: 32

  - Column (2): GG level, compensating factor = All the rest
    - Total exp/GDP: -0.350*** (0.100)
    - Educ/Total exp: 0.723* (0.384)
    - Spend. ex. Social and Educ spend./Total exp: -0.181 (0.300)
    - Initial GDP p.c.: -10.981*** (3.210)
    - Initial human capital: 0.431 (1.133)
    - Constant: 117.404** (36.030)
    - Level of government: GG
    - Observations: 59
    - R-squared: 0.54
    - Number of countries: 32

  - Column (3): CG level, compensating factor = Social Spend.
    - Total exp/GDP: -0.359*** (0.129)
    - Educ/Total exp: 0.602 (0.522)
    - Initial GDP p.c.: -10.612** (3.976)
    - Initial human capital: -0.762 (1.556)
    - Constant: 141.183** (61.624)
    - Level of government: CG
    - Observations: 59
    - R-squared: 0.53
    - Number of countries: 32

  - Column (4): GG level, compensating factor = Social Spend.
    - Total exp/GDP: -0.364*** (0.097)
    - Educ/Total exp: 0.712* (0.352)
    - Initial GDP p.c.: -11.640*** (4.082)
    - Initial human capital: 0.370 (1.058)
    - Constant: 132.278** (54.041)
    - Level of government: GG
    - Observations: 59
    - R-squared: 0.56
    - Number of countries: 32

- Key findings from Table 8:
  - Coefficients on the share of education spending are positive at both CG and GG levels in the specifications reported.
  - Example interpretation (noting caveats on reliability): the value 0.286 (Column (1)) implies that one percentage point increase in the share of education spending is associated with 0.286 change in the dependent variable (annual GDP per capita growth, averaged over 5 years) in that specification.
  - Overall comparison: coefficients on education spending at CG and GG levels are broadly similar in sign and magnitude across specifications, suggesting that the reallocation effect involving education spending is not markedly different when measured at the GG versus CG level in this smaller sample (subject to the cautions about sample size and lack of instrumentation).

*Source: _wp13162 - 3.  Additional Explanatory Variables*

### 0.286 percentage points increase in growth per annum.

### _wp13162 - 0.286 percentage points increase in growth per annum.

### Main findings
- Reallocations of government spending across different outlays (holding overall expenditure envelope fixed) do not appear to be robustly associated with growth in general.
- Key exception: increases in education spending, when offset by a fall in social protection spending, are robustly associated with higher growth.
- The education–social protection reallocation result is robust to:
  - alternative assumptions of delayed fiscal policy effects;
  - use of a smaller sample excluding the richest countries;
  - addition of various extra control variables.
- The main analysis uses consolidated central government (CG) level data; the education-related finding also holds at the consolidated general government (GG) level.
- In the economic classification, increases in capital spending financed through reductions in current spending (including public wages) seem to be positively associated with growth, but this result lacks robustness across specifications.

### Magnitude and statistical notes
- The unit title references a "0.286 percentage points increase in growth per annum."
- Coefficient on education spending is larger at the GG level and is statistically significant at the 10 percent level in that specification, but a paired t-test indicates the difference between CG and GG coefficients is not statistically significant.
- The dependent variable in main regressions is an (averaged) annual growth rate.
- Results for economic classification coefficients under CG and GG levels do not appear to be statistically different.

### Robustness, limitations, and caveats
- Results may be sensitive to quality/efficiency of spending:
  - The paper analyzes quantities of spending without explicit controls for efficiency/quality.
  - Public infrastructure spending effects may depend on cross-country differences in efficiency; without proxies for spending quality, quantity-based results can be misleading.
- Other policy objectives beyond growth are not addressed in the analysis:
  - Employment and income equality may be influenced differently by spending composition.
  - Social protection, while not growth-enhancing in these results, may promote income equality.
- Heterogeneity and subgroup analyses:
  - Regressions by country groups (HICs vs LICs+MICs) generally find coefficients do not differ significantly across groups, though some education spending effects are significant only in HICs when offset by falls in health or social protection.
  - Preliminary country-group estimations sometimes required treating fiscal variables as exogenous due to instrument count concerns; some specifications produced counterintuitive signs (e.g., positive coefficient on initial GDP per capita), indicating caution in interpretation.

### Policy implications
- When overall expenditure cannot be increased (e.g., during prolonged fiscal austerity), reallocating spending matters:
  - Prioritizing education spending while reducing social protection spending is associated with higher growth in the sample and specifications examined.
  - Public capital spending can support long-run growth when financed by reductions in current spending, but evidence is less robust and depends on spending efficiency.
- Policymakers should weigh growth effects alongside other objectives (employment, income equality) when reallocating public spending.

### Key statistics and sample descriptors (as presented)
- Sample coverage and growth:
  - The sample countries grew, on average, around 12 percent in per capita terms over the 5-year period.
  - Growth rate of real GDP pc (over 5 years): Mean 11.949, Standard Deviation 9.61, Minimum -24.21, Maximum 39.98.
- Economic classification (shares in percent):
  - Total Exp/GDP: Mean 33.03, Standard Deviation 8.90, Minimum 13.05, Maximum 53.87.
  - Comp. of Employees/Total Spend.: Mean 20.20, Standard Deviation 9.47, Minimum 5.85, Maximum 54.04.
  - Rest of Expense/Total Spend.: Mean 72.56, Standard Deviation 14.05, Minimum 33.83, Maximum 92.47.
  - Nonfin. Assets/Total Spend.: Mean 7.24, Standard Deviation 6.50, Minimum 1.24, Maximum 36.47.
- Functional classification (shares in percent):
  - Total exp/GDP: Mean 28.40, Standard Deviation 9.72, Minimum 10.43, Maximum 52.88.
  - Defense/Total Spend.: Mean 9.58, Standard Deviation 7.00, Minimum 1.98, Maximum 33.06.
  - Transport and communication/Total Spend.: Mean 5.69, Standard Deviation 2.80, Minimum 1.26, Maximum 13.72.
  - Health/Total Spend.: Mean 8.41, Standard Deviation 4.79, Minimum 0.83, Maximum 20.29.
  - Education/Total Spend.: Mean 11.39, Standard Deviation 5.53, Minimum 1.82, Maximum 23.34.
  - Social Prot./Total Spend.: Mean 27.11, Standard Deviation 15.43, Minimum 1.33, Maximum 54.89.
  - Rest/Total Spend.: Mean 37.83, Standard Deviation 11.19, Minimum 16.64, Maximum 76.76.
- CG versus GG summary (in percent) for the static panel regression sample (32 countries: 20 HICs, 9 MICs, 3 LICs):
  - Total exp/GDP (CG): Mean 33.48, Std. dev. 7.04, Minimum 15.73, Maximum 44.29.
  - Total exp/GDP (GG): Mean 40.24, Std. dev. 7.45, Minimum 20.52, Maximum 57.66.
  - Educ/Total exp (CG): Mean 9.75, Std. dev. 3.96, Minimum 2.32, Maximum 20.44.
  - Educ/Total exp (GG): Mean 13.60, Std. dev. 2.77, Minimum 6.35, Maximum 19.69.
  - Social/Total exp (CG): Mean 49.93, Std. dev. 9.62, Minimum 8.22, Maximum 63.19.
  - Social/Total exp (GG): Mean 48.92, Std. dev. 8.11, Minimum 26.23, Maximum 59.74.
  - Rest/Total exp (CG): Mean 40.33, Std. dev. 10.00, Minimum 26.36, Maximum 86.24.
  - Rest/Total exp (GG): Mean 37.48, Std. dev. 7.37, Minimum 26.39, Maximum 55.32.

*Source: _wp13162 - 0.286 percentage points increase in growth per annum.*

### 11.39 percent at the CG level and 13.50 percent at the GG level).

### _wp13162 - 11.39 percent at the CG level and 13.50 percent at the GG level).

### Methodological notes
- To be consistent with our main regression analysis conducted above, we calculate the 5-year average if we have 3 observations or more (of all the expenditure subcomponents of our interest) out of the maximum of 5.
- As before, total expenditure is calculated as the addition of the 8 functional subcomponents out of 10. That is, public order and safety and environmental protection are excluded.

### References
- Agénor, Pierre-Richard, 2010, “A theory of infrastructure-led development,”Journal of Economic Dynamics & Control, Vol. 34, pp. 932–950.
- Aghion, Philippe, Leah Boustan, Caroline Hoxby, and Jerome Vandenbussche, 2009, “The Causal Impact of Education on Economic Growth: Evidence from U.S.”Brookings Papers on Economic Activity, Vol. Spring, pp. 1–73.
- Arellano, Manuel, and Stephen Bond, 1991, “Some tests of specification for panel data: Monte Carlo evidence and an application to employment equations,”Review of Economic Studies, Vol. 58, pp. 277–297.
- Arellano, Manuel, and Olympia Bover, 1995, “Another look at the instrumental-variable estimation of error-components models,”Journal of Econometrics, Vol. 68, pp. 29–52.
- Arnold, Jens Matthias, Bert Brys, Christopher Heady, Åsa Johansson, Cyrille Schwellnus, and Laura Vartia, 2011, “Tax policy for economic recovery and growth,”Economic Journal, Vol. 121, pp. F59–F80.
- Aschauer, David Alan, 1989, “Is public expenditure productive?”Journal of Monetary Economics, Vol. 23, pp. 177–200.
- Barro, Robert, 1990, “Government spending in a simple model of endogenous growth,”Journal of Political Economy, Vol. 98, No. 5, pp. S103–S125.
- ———, 2004,Economic Growth(Cambridge, Massachusetts: MIT Press), second ed.
- Barro, Robert, and Jong-Wha Lee, 2010, “A New Data Set of Educational Attainment in the World, 1950–2010,” NBER Working Paper 15902.
- Blankenau, William F., and Nicole B. Simpson, 2004, “Public education expenditures and growth,”Journal of Development Economics, Vol. 73, pp. 583–605.
- Blundell, Richard, and Stephen Bond, 1997, “Initial conditions and moment restrictions in dynamic panel data models,” University College London, Discussion Paper in Economics 97-07.
- Bond, Stephen, Anke Hoeffler, and Jonathan Temple, 2001, “GMM Estimation of Empirical Growth Models,” Discussion Paper No. 2048, Centre for Economic Policy Research.
- Bose, Niloy, M. Emranul Haque, and Denise R. Osborn, 2007, “Public expenditure and economic growth: A disaggregated analysis for developing countries,”The Manchester School, Vol. 75, No. 5, pp. 533–556.
- Caselli, Francesco, Gerardo Esquivel, and Fernando Lefort, 1996, “Reopening the convergence debate: A new look at cross-country growth empirics,”Journal of Economic Growth, Vol. 1, pp. 363–389.
- Clements, Ben, David Coady, Frank Eich, Sanjeev Gupta, Alvar Kangur, Baoping Shang, and Mauricio Soto, 2012,The Challenge of Health Care Reform in Advanced and Emerging Economies, Occasional Paper No 275 (Washington DC: International Monetary Fund).
- Clements, Ben, David Coady, and Sanjeev Gupta, 2012, “The Challenge of Health Care Reform in Advanced and Emerging Economies,” in Ben Clements, David Coady, and Sanjeev Gupta (eds.),The Economics of Public Health Care Reform in Advanced and Emerging Economies, pp. 3–21 (Washington DC: International Monetary Fund).
- Dabla-Norris, Era, Jim Brumby, Annette Kyobe, Zac Mills, and Chris Papageorgiou, 2012, “Investing in public investment: An index of public investment efficiency,”Journal of Economic Growth, Vol. 17, No. 3, pp. 235–266.
- Dziobek, Claudia, Carlos Gutierrez Mangas, and Phebby Kufa, 2011, “Measuring fiscal decentralization – Exploring the IMF’s database,” IMF, Working Paper 11/126.
- Easterly, William, and Sergio Rebelo, 1993, “Fiscal policy and economic growth,”Journal of Monetary Economics, Vol. 32, pp. 417–458.
- Gemmell, Norman, Richard Kneller, and Ismael Sanz, 2011, “The timing and persistence of fiscal policy impacts on growth: Evidence from OECD countries,”Economic Journal, Vol. 121, pp. F33–F58.
- Glomm, Gerhard, and B. Ravikumar, 1997, “Productive government expenditures and long-run growth,”Journal of Economic Dynamics and Control, Vol. 21, pp. 183–204.
- Gupta, Sanjeev, Benedict Clements, Emanuele Baldacci, and Carlos Mulas-Granados, 2005, “Fiscal policy, expenditure composition, and growth in low-income countries,”Journal of International Money and Finance, Vol. 24, pp. 441–463.
- Holtz-Eakin, Douglas, 1994, “Public-sector capital and the productivity puzzle,”Review of Economics and Statistics, Vol. 76, No. 1, pp. 12–21.
- Holtz-Eakin, Douglas, Whitney Newey, and Harvey Rosen, 1990, “Estimating vector autoregressions with panel data,”Econometrica, Vol. 56, pp. 1371–1395.
- Kaganovich, Michael, and Itzhak Zilcha, 1999, “Education, social security and growth,”Journal of Public Economics, Vol. 71, pp. 289–309.
- Kneller, Richard, Michael F. Bleaney, and Norman Gemmell, 1999, “Fiscal policy and growth: evidence from OECD countries,”Journal of Public Economics, Vol. 74, pp. 171–190.
- Levine, Ross, Norman Loayza, and Thorsten Beck, 2000, “Financial intermediation and growth: Causality and causes,”Journal of Monetary Economics, Vol. 46, pp. 31–77.
- Lucas, Robert, 1988, “On the mechanics of economic development,”Journal of Monetary Economics, Vol. 22, pp. 3–42.
- Mauro, Paolo, 1995, “Corruption and growth,”Quarterly Journal of Economics, Vol. 110, pp. 681–712.
- ———, 1998, “Corruption and the composition of government expenditure,”Journal of Public Economics, Vol. 69, pp. 263–279.
- Pesaran, M. Hashem, Y Shin, and Ron P. Smith, 1999, “Pooled mean group estimation of dynamic heterogeneous panels,”Journal of the American Statistical Association, Vol. 94, pp. 621–34.
- Pesaran, M. Hashem, and Ron P. Smith, 1995, “Estimating long-run relationships from dynamic heterogeneous panels,”Journal of Econometrics, Vol. 68, pp. 79–113.
- Pritchett, Lant, 2000, “The tyranny of concepts: CUDIE (cumulated, depreciated, investment effort) is not capital,”Journal of Economic Growth, Vol. 5, pp. 361–384.
- Ram, Rati, 1990, “Wagner’s Hypothesis in Time-Series and Cross-Section Perspectives: Evidence from “ Real” Data for 115 Countries,”Review of Economics and Statistics, Vol. 69, No. 2, pp. 194–204.
- Rodrik, Dani, 2008, “The real exchange rate and economic growth,”Brookings Papers on Economic Activity, Vol. Fall, pp. 365–412.
- Roodman, David, 2009a, “How to do xtabond2: An instroduction to difference and system GMM in Stata,”Stata Journal, Vol. 9, No. 1, pp. 86–136.
- ———, 2009b, “A Note on the Theme of Too Many Instruments,”Oxford Bulletin of Economics and Statistics, Vol. 71, No. 1, pp. 135–58.
- Wickens, Tobias, 2002, “Government Finance Stataistics Manual 2001 Companion Material: Classification of GFSM 1986 data to the GFSM 2001 framework,” International Monetary Fund.
- Wyss, Kaspar, and Nicolaus Lorenz, 2000, “Decentralization and central and regional coordination of health services: The case of Switzerland,”International Journal of Health Planning and Management, Vol. 15, pp. 103–114.

*Content excerpted from the supplied IMF PDF chapter.*

---


_Source: https://www.imf.org/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2013/_wp13162.pdf_