## 5. Average Efficiency Scores for Latin American Countries

## Source details

**Canonical URL:** [5. Average Efficiency Scores for Latin American Countries](https://www.imf.org/-/media/files/publications/wp/2023/english/wpiea2023101-print-pdf.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/wp/2023/english/wpiea2023101-print-pdf.pdf.md)
- [Structured JSON version](/-/media/files/publications/wp/2023/english/wpiea2023101-print-pdf.pdf.json)

---

### Introduction and context
- Public capital stock (as a share of GDP) trends:
  - Declined steadily in advanced economies (AEs).
  - Rose in low-income and emerging economies in the 1970s–1980s, then sharply dipped in the 1990s–2000s with modest recovery after 2008.
- Public investment ratios broadly follow capital stock trends but precede them by a few years.
- The Covid-19 pandemic likely exacerbated the decline in public investment in EMEs and LICs; some AEs are expected to raise public investment in recovery packages.
- Public investment can catalyze private investment and may have high multipliers, particularly for green investments.
- EMEs and LICs face tighter financing constraints with the end of the low interest rate era and higher debt levels.

### Benchmarking methodology: concepts and methods
- Benchmarking builds an efficiency frontier from best performers (decision-making units, DMUs); units below the frontier are inefficient.
- Two types of efficiency (Farrell, 1957):
  - Technical efficiency: ability to produce the same output with fewer inputs (input-oriented or output-oriented measures).
  - Allocative and cost efficiency discussed but not the paper’s focus due to lack of comparable multi-country input price data.
- Two families of estimation methods:
  - Data Envelopment Analysis (DEA) — non-parametric:
    - Output-oriented DEA yields factor φj ≥ 1; technical efficiency score = 1/φj (ranges from 1 to 0).
    - Constant Returns to Scale (CRS) vs Variable Returns to Scale (VRS); paper uses VRS.
    - Main shortcoming: deterministic and sensitive to outliers; captures uncontrollable factors.
  - Stochastic Frontier Analysis (SFA) — parametric:
    - Canonical form ln(yi) = f(xi; β) + εi, with εi = vi − ui, vi ~ iid N(0, σv^2), ui ≥ 0 (e.g., ui ~ iid N+(0, σu^2)).
    - Distinguishes statistical noise (vi) from inefficiency (ui); more robust to measurement error.
- Bootstrapping DEA:
  - Corrects DEA’s upward bias via resampling with replacement to estimate bias and confidence intervals.

### Inputs and outputs: choices and considerations
- Outputs classified as quantity, quality, and utilization (examples: road length, road condition indices, electricity consumption per capita).
- Multi-output vs composite single-output:
  - DEA with multiple outputs places greater weight on outputs where a DMU performs better and often yields higher mean efficiency scores than aggregated composite outputs.
  - Aggregation (standardize then average) can mask sectoral differences and may overstate efficiency.
- Inputs:
  - Monetary measures (investment spending, capital stock) are common for cross-country work.
  - Stock-producing analyses should use stocks (e.g., public capital stock); flow outputs should use flows (e.g., operating costs).
  - Inclusion of private investment, PPPs, SOE spending limited by data availability.
- Environmental (non-discretionary) factors:
  - Should be accounted for: in SFA as determinants of the inefficiency term mean; in DEA as non-discretionary variables or via second-stage regressions.

### Empirical models, outputs, and data
- Baseline focus: narrow physical infrastructure definition — roads, energy, telecommunications.
- Four models via principal components analysis:
  - Model 1 (baseline): roads, energy, telecommunications.
  - Model 2: roads, energy, telecommunications, railroads.
  - Model 3: roads, energy, railroads.
  - Model 4: roads, energy, water, health, education.
- Variables and units:
  - Length of the road network (kilometers).
  - Electric power consumption (kWh) per 100 people.
  - Number of fixed telephone lines per 100 people.
  - Kilometers of rail line route per 100 people.
  - Percent of the population using safely managed drinking water facilities.
  - Number of secondary school teachers per 100 people.
  - Number of hospital beds per 1,000 people.
- Data sources:
  - Roads: International Road Federation and CIA Factbook.
  - Electricity: International Energy Agency and U.S. Energy Information Administration.
  - Other variables: World Bank’s World Development Indicators (WDI).
- Standardization and aggregation:
  - Sectoral indicators standardized and combined via principal components analysis.
  - Income-group distributions scaled relative to the average of advanced economies.
- Coverage:
  - Baseline model maximum coverage: 168 countries.
  - Models 2 and 3 coverage: 112 countries (railroads variable limited).
  - Model 4 coverage: 153 countries.
- Main input for cross-country efficiency estimation: public capital stock per capita in monetary real dollars; per capita GDP included as proxy for economic development.

### Normality / skewness tests (Table 1)
- Skewness statistics (Model — Skewness, p-value, Test):
  - Model 1 (Roads, energy, telecommunications): Skewness = -1.670***, p-value = 0.000, Test = 84.285
  - Model 2 (Roads, energy, railroads, telecommunications): Skewness = -2.120***, p-value = 0.000, Test = 81.402
  - Model 3 (Roads, energy, railroads): Skewness = -1.753***, p-value = 0.000, Test = 66.975
  - Model 4 (Roads, energy, water, health, and education): Skewness = -4.601***, p-value = 0.000, Test = 187.044
- Note: *, **, and *** indicate statistical significance at the 10, 5, and 1 percent level, respectively.
- Skewness tests support using SFA.

### Key empirical findings relevant to Latin America and global comparisons
- Distribution of infrastructure stocks and outputs:
  - Stock of infrastructure is, for the most part, larger for EMs than for LIDCs, and larger still for AEs.
  - Income groups are quite far apart in infrastructure output; interquartile ranges of income groups rarely overlap across models.
  - LIDCs’ infrastructure concentrated in a narrow range; AEs display wider distribution (especially Model 3).
  - Capital stock measure includes general government capital stock and excludes SOEs’ and PPP capital stock (ICSD source).
- Efficiency scores:
  - Median efficiency score rises with income in nearly all models and estimation methods.
  - Exception: median SFA score of EMs nearly equals that of AEs in the baseline model and is higher in Model 4.
  - Country-level rankings are stable across methodologies: Spearman rank-correlations between methodologies range from 0.73 to 0.99 and are highly statistically significant.
- Correlations between methodologies (selected values from Table 3):
  - Model 1: DEA vs. Bootstrap DEA = 0.979*** (p-value = 0.000); DEA vs. SFA = 0.913*** (p-value = 0.000)
  - Model 2: DEA vs. Bootstrap DEA = 0.994*** (p-value = 0.000); DEA vs. SFA = 0.917*** (p-value = 0.000)
  - Model 3: DEA vs. Bootstrap DEA = 0.995*** (p-value = 0.000); DEA vs. SFA = 0.939*** (p-value = 0.000)
  - Model 4: DEA vs. Bootstrap DEA = 0.993*** (p-value = 0.000); DEA vs. SFA = 0.738*** (p-value = 0.000)
- Correlations across models for a given methodology range from 0.52 to 0.95 and are strongly significant.
- Estimated average efficiency gaps (Table 2):
  - Global public investment efficiency gap (1 minus average efficiency score) ranges between 23 and 79 percent depending on income group, model, and method.
  - Baseline model average efficiency gaps across methods (full efficiency gap):
    - AEs: 50 percent
    - EMs: 57 percent
    - LIDCs: 70 percent
  - Averaging all 12 indicators (4 models × 3 methodologies) — potential gains from removing inefficiencies:
    - Overall: 55 percent increase in infrastructure output
    - Advanced economies (AEs): 45 percent
    - Emerging market economies (EMs): 54 percent
    - Low-income developing countries (LIDCs): 65 percent
  - Closing the gap to each income group’s 90th percentile:
    - Overall: 30 percent increase
    - AEs: 29 percent; EMs: 27 percent; LIDCs: 35 percent
- Selected numeric cells from Table 2 (Overall average full efficiency gap):
  - DEA overall average full efficiency gap: AEs 0.450, EMs 0.538, LIDCs 0.647, Total 0.547.
  - Bootstrap DEA overall average full efficiency gap: AEs 0.493, EMs 0.607, LIDCs 0.734, Total 0.616.
  - SFA overall average full efficiency gap: AEs 0.433, EMs 0.449, LIDCs 0.525, Total 0.466.
  - Overall average across methods: AEs 0.450, EMs 0.538, LIDCs 0.647, Total 0.547.
  - Overall average gap to the 90th percentile across methods: AEs 0.287, EMs 0.275, LIDCs 0.350, Total 0.298.

### Disaggregated input analysis (Latin America, INFRALATAM)
- INFRALATAM dataset:
  - 22 Latin-American countries, 2008–2021.
  - Includes general government and SOEs public investment by sector.
- Public investment trends in Latin America:
  - Declined steadily since 2016, with an accelerated decline in 2020.
  - Sectoral composition: roads constitute the largest portion of infrastructure investment, followed by electricity.
- Sectoral efficiency estimation:
  - Sectoral scores estimated for road, rail, telecommunications, and electricity.
  - Final country score = average of sectoral scores for Models 1–3.
  - Due to sample size, only non-parametric methods (DEA and bootstrap DEA) used.
- Aggregation-based vs sector-specific efficiency (Table 5) — selected results (average efficiency scores for Latin America):
  - DEA:
    - Model 1: Aggregation-based 0.626; Sector-specific 0.551; t-test p-value = 0.246 (not significant).
    - Model 2: Aggregation-based 0.564; Sector-specific 0.621; t-test p-value = 0.666 (not significant).
    - Model 3: Aggregation-based 0.631; Sector-specific 0.519; t-test p-value = 0.454 (not significant).
  - Bootstrap DEA:
    - Model 1: Aggregation-based 0.571; Sector-specific 0.436; t-test = 3.558*** (p-value = 0.002).
    - Model 2: Aggregation-based 0.644; Sector-specific 0.493; t-test = 2.183* (p-value = 0.072).
    - Model 3: Aggregation-based 0.611; Sector-specific 0.378; t-test = 3.281** (p-value = 0.017).
- Interpretation:
  - Aggregation-based scores are systematically higher than sector-specific scores when bootstrapping is applied.
  - Aggregated-input results likely represent a lower-bound for efficiency gaps; disaggregated data suggest larger gaps.

### Main conclusions and policy implications
- Public investment in infrastructure can support longer-term recovery from the pandemic, but many countries—especially low-income economies—face constraints to expanding infrastructure by substantially increasing investments when access to capital markets is limited.
- Increasing the efficiency with which capital expenditures generate infrastructure outputs is critical for EMEs and LICs facing financing constraints.
- Benchmarking can identify best performers and inform public investment management reforms.
- The IMF’s PIMA uses efficiency scores as a starting point for governance reform discussions, but scores should be complemented with granular information and caveats about methodology and data limitations.
- Empirical findings indicate sizeable efficiency gaps:
  - Removing all inefficiencies could increase infrastructure output by 55 percent overall, with larger potential gains in lower-income countries.
  - Closing the gap to the 90th percentile within income groups could increase infrastructure output by about 30 percent overall.
- Aggregation of inputs and outputs likely understates true efficiency gaps; disaggregated data suggest larger gaps and that main results should be viewed as a lower-bound.

### Suggestions for future research
- Expand empirical applications to additional conceptual options if data become available.
- Investigate drivers of investment inefficiencies (institutional, environmental, scale nonlinearities) and ways to address them.
- Use estimated efficiency measures to explore the relationship between public investment efficiency and economic growth.
- Improve data quality and sectoral disaggregation of public investment and infrastructure outputs.
- Explore nonlinear relationships between public capital stock and efficiency (e.g., low efficiency at very low capital levels due to lumpiness and network externalities; diminishing returns at very high capital stock).

*Source: Authors’ estimations, IMF Working Paper excerpt (text unit 100).*

### REFERENCES .............................................................................................................

### REFERENCES ............................................................................................................................................. 27

### Figures
- 1. Trends in Public Investment and Capital Stock .......................................................................................... 3
- 2. Technical and Allocative Efficiency ............................................................................................................ 6
- 3. Public Infrastructure Output Indexes, by Income Group .......................................................................... 16
- 4. Histograms of Residuals from OLS Regressions of Infrastructure on Inputs ........................................... 18
- 5. Public Investment Efficiency Scores ........................................................................................................ 19
- 6. Public Investment by Sector in Latin America .......................................................................................... 24

### Tables
- 1. Normality Tests ........................................................................................................................................ 18
- 2. Estimated Average Efficiency Gaps, by Income Group ........................................................................... 21
- 3. Correlations between Methodologies ....................................................................................................... 22
- 4. Correlations for Different Models and Estimation Methods ...................................................................... 23

*wpiea2023101-print-pdf - REFERENCES ............................................................................................................................................. 27*

### 5. Average Efficiency Scores for Latin American Countries ..............................................................

### 5. Average Efficiency Scores for Latin American Countries

### Introduction and context
- Public capital stock (as a share of GDP) has been on a steady long-term decline in advanced economies; in low-income and emerging economies it rose in the 1970s–1980s and then sharply dipped in the 1990s–2000s with modest recovery after 2008.
- Public investment ratios broadly follow the capital stock trend but precede it by a few years.
- The Covid-19 pandemic likely exacerbated the decline in public investment in EMEs and LICs; AEs (e.g., EU countries) are expected to raise public investment in recovery packages.
- Public investment can be a strong catalyst of private investment and may have high multipliers, particularly for green investments.
- EMEs and LICs face financing constraints with the end of the low interest rate era and higher debt levels.

### Benchmarking methodology: concepts and methods
- Benchmarking compares decision-making units (DMUs) under production-economics principles to build an efficiency frontier from best performers; units below the frontier are inefficient.
- Two types of efficiency (Farrell, 1957):
  - Technical efficiency: ability to produce the same output with fewer inputs (input-oriented measure: by how much inputs could be scaled back while leaving outputs unchanged; output-oriented: how much outputs could be boosted given inputs).
  - Allocative efficiency and cost efficiency are discussed but the paper focuses on technical efficiency due to lack of comparable multi-country input price data.
- Two families of estimation methods:
  - Non-parametric: Data Envelopment Analysis (DEA)
    - DEA constructs a piecewise-linear frontier and evaluates DMUs relative to it.
    - Output-oriented DEA model solution yields a factor φj ≥ 1; technical efficiency score = 1/φj which ranges from 1 (fully efficient) to 0 (fully inefficient).
    - Constant Returns to Scale (CRS) vs Variable Returns to Scale (VRS): the paper estimates DEA under VRS to allow comparisons among DMUs of similar size.
    - Main shortcoming: deterministic nature captures uncontrollable factors (weather, topography) and is sensitive to outliers, yielding a consistent but biased estimator of the true frontier.
  - Parametric: Stochastic Frontier Analysis (SFA)
    - Canonical SFA: ln(yi) = f(xi; β) + εi, with εi = vi − ui, where vi ~ iid N(0, σv^2), ui ≥ 0 (e.g., half-normal prior ui ~ iid N+(0, σu^2)).
    - Advantage: distinguishes statistical noise (vi) from inefficiency (ui) and is more robust to measurement error and uncontrollable factors.
- Bootstrapping DEA
  - Bootstrapping corrects the upward bias of DEA (DEA captures lower bound of true frontier) by sampling with replacement to estimate bias and confidence intervals for efficiency scores.

### Inputs and outputs: choices and considerations
- Outputs classified into three types: quantity, quality, and utilization.
  - Examples: road length (quantity), road condition indices (quality), electricity consumption per capita (utilization).
- Multi-output versus composite single-output:
  - DEA with multiple outputs places greater weight on outputs where a DMU performs better and often yields higher mean efficiency scores than aggregated composite outputs.
  - Aggregation approach (standardize then average) can mask sectoral differences and may overstate efficiency.
- Inputs: monetary measures (investment spending, capital stock) are most common for cross-country work; physical inputs are feasible mainly for sectoral/subsectoral analysis.
  - For stock-producing analyses, inputs should be stocks (e.g., public capital stock); for flow outputs, inputs may be flows (e.g., operating costs).
  - Decide whether to include private investment, PPPs, SOE spending — data limitations often preclude full inclusion.
- Environmental (non-discretionary) factors
  - Variables that affect infrastructure outputs but are not under managerial control (e.g., geography, climate, customer density) should be accounted for:
    - In SFA: as determinants of the mean of the inefficiency term (e.g., truncated-normal SFA).
    - In DEA: as non-discretionary variables in benchmarking or via second-stage regressions of efficiency scores on environmental variables.

### Empirical models, outputs, and data (models used in the paper)
- Baseline focus on a narrow definition of physical infrastructure: roads, energy, telecommunications.
- Four models constructed via principal components analysis to produce aggregate infrastructure output indexes:
  - Model 1 (baseline): roads, energy, telecommunications.
  - Model 2: roads, energy, telecommunications, railroads.
  - Model 3: roads, energy, railroads.
  - Model 4: roads, energy, water, health, education.
- Variables and units used:
  - Length of the road network (kilometers).
  - Electric power consumption (kWh) per 100 people.
  - Number of fixed telephone lines per 100 people.
  - Kilometers of rail line route per 100 people.
  - Percent of the population using safely managed drinking water facilities.
  - Number of secondary school teachers per 100 people (education proxy).
  - Number of hospital beds per 1,000 people (health proxy).
- Data sources:
  - Roads: International Road Federation and CIA Factbook.
  - Electricity: International Energy Agency and U.S. Energy Information Administration.
  - Other variables: World Bank’s World Development Indicators (WDI).
- Standardization and aggregation:
  - Sectoral indicators standardized (due to different scales) and combined via principal components analysis.
  - Income-group distributions scaled relative to the average of advanced economies.

### Key empirical findings relevant to Latin America
- Aggregated average efficiency scores vary by income group, estimation method, and measures of investment spending and outcomes.
- Efficiency gaps are large and range between about one-quarter to nearly 80 percent.
- Individual country efficiency rankings are broadly robust to estimation method.
- Sectoral analysis for Latin America indicates that using aggregated data may produce somewhat overestimated efficiency scores compared with more disaggregated, granular information.
- Improvements in data quality and more detailed sectoral spending data would provide a more complete picture and likely alter measured efficiencies.

### Policy implications and research priorities
- For EMEs and LICs facing financing constraints, increasing the efficiency with which capital expenditures generate infrastructure outputs is critical.
- Benchmarking can identify best performers shaping a frontier and inform public investment management reforms.
- The IMF’s PIMA uses efficiency scores as a starting point for governance reform discussions, but such scores should be complemented with granular information and caveats about methodology and data limitations.
- Recommended future work:
  - Focus on drivers of inefficiencies (institutional, environmental, scale nonlinearities).
  - Improve data quality and sectoral disaggregation of public investment and infrastructure outputs.
  - Explore the nonlinear relationship between public capital stock and efficiency (potentially low efficiency at very low capital levels due to lumpiness and network externalities; diminishing returns at very high capital stock).

*International Monetary Fund — Benchmarking Infrastructure Using Public Investment Efficiency Frontiers*

### 100. We see that irrespective of model applied, for the most part the stock of infrastructure is larger for EMs

### 100. We see that irrespective of model applied, for the most part the stock of infrastructure is larger for EMs than for LIDCs, and is larger still for AEs

### Distribution of infrastructure stocks and outputs
- The stock of infrastructure is, for the most part, larger for EMs than for LIDCs, and larger still for AEs.
- Income groups are quite far apart in infrastructure output: in all but one model, the interquartile range of the income groups never even overlap.
- LIDCs’ infrastructure is relatively concentrated within a narrow range, while AEs’ values display a wider distribution (especially in the third model that focuses on transport and energy).
- The measure of capital stock used consists of general government capital stock and does not include SOEs’ nor PPP capital stock (data limitations discussed in Section II.C). Capital stock data are obtained from IMF’s Investment and Capital Stock Dataset (ICSD).

### Methodologies and model coverage
- Efficiency estimation methods used: stochastic frontier analysis (SFA), data envelopment analysis (DEA), and bootstrap DEA.
- Main input: public capital stock per capita in monetary real dollars, and per capita GDP included as a proxy for economic development.
- Models and country coverage:
  - Baseline model maximum coverage: 168 countries.
  - Models 2 and 3 coverage: 112 countries (railroads variable has more limited coverage).
  - Model 4 coverage: 153 countries.
- Skewness tests on residuals support using SFA: the null hypothesis of no skewness is confidently rejected for all four models.

### Normality / skewness test statistics (Table 1)
- Model 1 (Roads, energy, telecommunications): Skewness = -1.670***, p-value = 0.000, Test = 84.285
- Model 2 (Roads, energy, railroads, telecommunications): Skewness = -2.120***, p-value = 0.000, Test = 81.402
- Model 3 (Roads, energy, railroads): Skewness = -1.753***, p-value = 0.000, Test = 66.975
- Model 4 (Roads, energy, water, health, and education): Skewness = -4.601***, p-value = 0.000, Test = 187.044
- Note: *, **, and *** indicate statistical significance at the 10, 5, and 1 percent level, respectively.

### Efficiency scores: central tendencies and variation
- Efficiency scores vary by model and estimation methodology.
- The median efficiency score rises with income in nearly all models and estimation methods.
- Exception: the median SFA score of EMs is nearly equal to that of AEs in the baseline model, and higher in Model 4 (akin to Baum, et al. 2020).
- Country-level rankings are stable across methodologies: Spearman rank-correlations between methodologies range from 0.73 to 0.99 and are highly statistically significant.

### Correlations between methodologies (Table 3) — selected values
- Model 1: DEA vs. Bootstrap DEA = 0.979*** (p-value = 0.000); DEA vs. SFA = 0.913*** (p-value = 0.000)
- Model 2: DEA vs. Bootstrap DEA = 0.994*** (p-value = 0.000); DEA vs. SFA = 0.917*** (p-value = 0.000)
- Model 3: DEA vs. Bootstrap DEA = 0.995*** (p-value = 0.000); DEA vs. SFA = 0.939*** (p-value = 0.000)
- Model 4: DEA vs. Bootstrap DEA = 0.993*** (p-value = 0.000); DEA vs. SFA = 0.738*** (p-value = 0.000)

### Correlations across models for given methodologies (Table 4) — selected ranges
- Correlation coefficients across models and within methodologies are strongly statistically significant and range from 0.52 to 0.95.

### Estimated average efficiency gaps (Table 2) — full efficiency gap and gap to the 90th percentile
- The average global public investment efficiency gap (1 minus the average efficiency score) is significant and ranges between 23 and 79 percent, depending on income group, infrastructure model, and estimation method.
- Baseline model average efficiency gaps across methods (AEs, EMs, LIDCs): 50 percent, 57 percent, and 70 percent, respectively.
- Averaging all 12 indicators (4 models and 3 methodologies):
  - Removing all inefficiencies could increase infrastructure output by more than half overall (55 percent).
  - By income group:
    - Advanced economies (AEs): 45 percent increase.
    - Emerging market economies (EMs): 54 percent increase.
    - Low-income developing countries (LIDCs): 65 percent increase.
- Closing the efficiency gap to each income group’s 90th percentile country yields substantial gains:
  - Countries could increase infrastructure by 30 percent overall.
  - AEs: 29 percent; EMs: 27 percent; LIDCs: 35 percent.

- Table 2 (selected average rows, Overall average):
  - DEA overall average full efficiency gap: AEs 0.450, EMs 0.538, LIDCs 0.647, Total 0.547.
  - Bootstrap DEA overall average full efficiency gap: AEs 0.493, EMs 0.607, LIDCs 0.734, Total 0.616.
  - SFA overall average full efficiency gap: AEs 0.433, EMs 0.449, LIDCs 0.525, Total 0.466.
  - Overall average across methods: AEs 0.450, EMs 0.538, LIDCs 0.647, Total 0.547.
  - Overall average gap to the 90th percentile across methods: AEs 0.287, EMs 0.275, LIDCs 0.350, Total 0.298.

### Disaggregated input analysis (Latin America, INFRALATAM)
- INFRALATAM dataset: 22 Latin-American countries, 2008–2021; includes general government and SOEs public investment by sector.
- Public investment in Latin America has declined steadily since 2016, with an acceleration of the decline in 2020.
- Roads constitute the largest portion of infrastructure investment, followed by electricity.
- Sectoral efficiency scores estimated for road, rail, telecommunications, and electricity; final country score = average of sectoral scores for Models 1–3.
- Due to sample size, only non-parametric methods (DEA and bootstrap DEA) used for this sectoral exercise.

### Latin America: aggregation-based vs sector-specific efficiency scores (Table 5)
- DEA results (average efficiency scores for Latin America):
  - Model 1: Aggregation-based 0.626; Sector-specific 0.551; t-test p-value = 0.246 (not significant).
  - Model 2: Aggregation-based 0.564; Sector-specific 0.621; t-test p-value = 0.666 (not significant).
  - Model 3: Aggregation-based 0.631; Sector-specific 0.519; t-test p-value = 0.454 (not significant).
- Bootstrap DEA results:
  - Model 1: Aggregation-based 0.571; Sector-specific 0.436; t-test = 3.558*** (p-value = 0.002).
  - Model 2: Aggregation-based 0.644; Sector-specific 0.493; t-test = 2.183* (p-value = 0.072).
  - Model 3: Aggregation-based 0.611; Sector-specific 0.378; t-test = 3.281** (p-value = 0.017).
- Interpretation: aggregation-based scores are systematically higher than sector-specific scores when bootstrapping is applied; aggregated-input results likely represent a lower-bound for efficiency gaps.

### Main conclusions and policy implications
- Public investment in infrastructure can support longer-term economic recovery from the pandemic, but many countries—especially low-income economies—face constraints to expanding infrastructure through substantially increasing investments when access to capital markets is limited.
- Increasing investment efficiency is a critical avenue for building infrastructure where financing is constrained.
- The paper’s empirical findings indicate sizeable efficiency gaps: removing all inefficiencies could increase infrastructure output by 55 percent overall, with larger potential gains in lower-income countries.
- Even more modest objectives—closing the gap to the 90th percentile within income groups—could increase infrastructure output by about 30 percent overall.
- Aggregation of inputs and outputs likely understates true efficiency gaps; disaggregated data (where available) suggest larger gaps and indicate the main results should be seen as a lower-bound.

### Suggestions for future research (as indicated in the text)
- Expand empirical applications to more of the conceptual options discussed in the paper if data become available.
- Contribute to the literature on the drivers of investment inefficiencies and possible ways to overcome these weaknesses.
- Use the estimated efficiency measures to explore the relationship between the efficiency of public investment and economic growth.

*Source: Authors’ estimations, IMF Working Paper excerpt (text unit 100).*

### References

### References

### Cited works

- Agrell, Per J., Peter Bogetoft, and Jørgen Tind (2005) “DEA and dynamic yardstick competition in Scandinavian electricity distribution,” Journal of Productivity Analysis, 23: 173–201.  
- Albino-War, Maria, Svetlana Cerovic, Francesco Grigoli, Juan Carlos Flores, Javier Kapsoli, Haonan Qu, Yahia Said, Bahrom Shukurov, Martin Sommer, and SeokHyun Yoon, 2014, “Making the most of public investment in MENA and CCA oil-exporting countries,” IMF Staff Discussion Note 14/10.  
- Ali, Md Kamar, and K. K. Klein (2014) “Water use efficiency and productivity of the irrigation districts in southern Alberta” Water Resource Management 28: 2751–2766.  
- Bartels, Robert and Towhidul Islam (2002) “Supply restricted telecommunications markets: The effect of technical efficiency on waiting times” Journal of Productivity Analysis, 18: 161-169.  
- Batini, Nicoletta, Mario di Serio, Matteo Fragetta, Giovanni Melina, and Anthony Waldron. 2021. “Building back better: How big are green spending multipliers?” IMF Working Paper No. WP/21/87.  
- Baum, Anja, Tewodaj Mogues, and Geneviève Verdier (2020) “Getting the Most from Public Investment,” Ch. 3 (pp. 30 – 49), In: Schwartz, Gerd, Manal Fouad, Torben Hansen, and Geneviève Verdier, Well Spent: How Strong Infrastructure Governance Can End Waste in Public Investment. Washington DC: International Monetary Fund.  
- Bogetoft, Peter, and Lars Otto (2011) “Benchmarking with DEA, SFA, and R,” Springer.  
- Brasili, Andrea, Atanas Kolev, Debora Revoltella, and Jochen Schanz (2022) “Public Investment in the Pandemic—Europe at a Glance,” In: Floriana Cerniglia, Francesco Saraceno, and Andrew Watt (eds.), The Great Reset: 2021 European Public Investment Outlook, OpenBook Publishers.  
- Calderón, César, Enrique Moral-Benito, and Luis Servén (2015) “Is infrastructure capital productive? A dynamic heterogeneous approach,” Journal of Applied Econometrics, 30(2): 177–198.  
- Canning, David (1998) “A Database of World Stocks of Infrastructure, 1950-95,” World Bank Economic Review, 12(3): 529-547.  
- Cavalieri, Marina, Calogero Guccio, and Ilde Rizzo (2017) “On the role of environmental corruption in healthcare infrastructures: An empirical assessment for Italy using DEA with truncated regression approach,” Health Policy, 121: 515-524.  
- Coelli, Timothy, and Sergio Perelman (1999) “A comparison of parametric and non-parametric distance functions: With application to European railways,” European Journal of Operational Research 117: 326-339.  
- Coelli, Timothy, and Sergio Perelman (2000) “Technical efficiency of European railways: A distance function approach,” Applied Economics, 32(15): 1967-1976.  
- Coelli, Timothy, Dodla Sai Prasada Rao, Christopher J. O'Donnell, and George Edward Battese (2005) An Introduction to Efficiency and Productivity Analysis. New York: Springer.  
- D’Agostino, Ralph, Albert Belanger, and Ralph D’Agostino Jr. (1990) “A suggestion for using powerful and informative tests of normality,” American Statistician, 44(4): 316–321.  
- Estache, Antonio and Marianela González (2002) “Efficiency gains from port reform and the potential for yardstick competition: Lessons from Mexico,” World Development, 30(4): 545-560.  
- Farrell, Michael J. (1957) “The measurement of productive efficiency,” Journal of the Royal Statistical Society, 120(3): 253–290.  
- Fritzsche, Carolin (2019) “Analyzing the efficiency of county road production: evidence from Eastern German counties,” German Economic Review 20(4): e415–e435.  
- Gramlich, Edward M. (1994) “Infrastructure investment: A review essay,” Journal of Economic Literature, 32(3): 1176-1196.  
- Grigoli, Francesco (2014) “A hybrid approach to estimating the efficiency of public spending on education in emerging and developing economies,” IMF Working Paper N 14/19.  
- Grigoli, Francesco and Javier Kapsoli (2018) “Waste not, want not: The efficiency of health expenditure in emerging and developing economies,” Review of Development Economics 22(1): 384-403.  
- Growitsch, Christian, Tooraj Jamasb and Michael Pollitt (2009) “Quality of service, efficiency and scale in network industries: an analysis of European electricity distribution,” Applied Economics, 41(20): 2555-2570.  
- Gupta, Sanjeev and Marijn Verhoeven (2001) “The efficiency of government expenditure: experiences from Africa” Journal of Policy Modeling, 23(4): 433-467.  
- Hepburn, Cameron, Brian O Callaghan, Nicholas Stern, and Dimitri Zenghelis (2020) “Will COVID-19 fiscal recovery packages accelerate or retard progress on climate change?” Oxford Review of Economic Policy, 36 (20): 1–48.  
- Herrera, Santiago, and Gaobo Pang (2005) “Efficiency of public spending in developing countries: An efficiency frontier approach,” World Bank Policy Research Working Paper N 3645.  
- International Monetary Fund (2015) “Making public investment more efficient,” IMF Board Paper.  
- _____ (2020) “Fiscal Monitor: Policies for the Recovery,” IMF, Washington DC, October 2020.  
- _____ (2022a) “Fiscal Monitor: Helping People Bounce Back,” IMF, Washington DC, October 2022.  
- _____ (2022b) “Saudi Arabia—Selected Issues,” IMF, Washington DC, August 2022.  
- _____ (2022c) “World Economic Outlook: Countering the Cost-of-Living Crisis,” IMF, Washington DC, October 2022.  
- Kapsoli, Javier, and Iulia Ruxandra Teodoru (2017), “Benchmarking social spending using efficiency frontiers,” Working Paper WP/17/197. International Monetary Fund.  
- Kalb, Alexander (2014) “What determines local governments’ cost-efficiency? The case of road maintenance” Regional Studies, 48(9): 1483–1498.  
- Kneip, Alois, Byeong Park, and Léopold Simar (1998) “A note on the convergence of nonparametric DEA estimators for production efficiency scores.” Econometric Theory, Vol. 14, N 6, pp. 783-93.  
- Kyriacou, Andreas P., Leonel Muinelo-Gallo, and Oriol Roca-Sagalés (2019) “The efficiency of transport infrastructure investment and the role of government quality: An empirical analysis,” Transport Policy, 74: 93–102.  
- Lozano, Sebastián and Ester Gutiérrez (2011) “Efficiency analysis and target setting of Spanish airports,” Networks and Spatial Economics, 11:139-157.  
- Lv, Kangjuan, Anyu Yu, and Yiwen Bian (2017) “Regional energy efficiency and its determinants in China during 2001–2010: a slacks-based measure and spatial econometric analysis,” Journal of Productivity Analysis, 47:65–81.  
- Lorenzo, José Manuel Prado and Isabel María García Sánchez (2007) “Efficiency evaluation in municipal services: an application to the street lighting service in Spain,” Journal of Productivity Analysis 27:149-162.  
- Marcelo, Darwin, Cledan Mandri-Perott, Schuyler House, Jordan Schwartz (2016) “Prioritizing infrastructure investment: A framework for government decision-making,” Policy Research Working Paper No. 7674, World Bank, Washington DC.  
- Martín, Juan Carlos, Concepción Román, and Augusto Voltes-Dorta (2009) “A stochastic frontier analysis to estimate the relative efficiency of Spanish airports” Journal of Productivity Analysis 31: 163–176.  
- Moszoro, Mariano and Mauricio Soto (2022) “Road quality and mean speed score,” IMF Working Paper WP/22/95, Washington DC.  
- Murillo-Zamorano, Luis (2004) “Economic efficiency and frontier techniques,” Journal of Economic Surveys, 18(1): 33-77.  
- Obstfeld, Maurice and Kenneth Rogoff (1996) Foundations of International Macroeconomics, Cambridge, MA: MIT Press.  
- Pina, Vincente, and Lourdes Torres (2006) “Public-private efficiency in the delivery of services of general economic interest: The case of urban transport” Local Government Studies, 32(2): 177 – 198.  
- Rayp, Glenn and Nicolas Van De Sijpe (2007) “Measuring and explaining government efficiency in developing countries,” Journal of Development Studies, 43(2): 360-381.  
- Simar, Léopold, and Paul W. Wilson (1998) “Sensitivity analysis of the efficiency scores: How to bootstrap in nonparametric frontier models,” Management Science, 44(1): 49–61.  
- _______ (2000) “A general methodology for bootstrapping in non-parametric frontier models,” Journal of Applied Statistics, 27(6): 779–802.  
- Tanzi, V., and H. Davoodi (2002) “Corruption, public investment, and growth.” In Governance, Corruption, and Economic Performance, edited by G. Abed and S. Gupta, 280–299. Washington: International Monetary Fund.  
- Wanke, Peter F. (2013) “Physical infrastructure and shipment consolidation efficiency drivers in Brazilian ports: A two-stage network-DEA approach,” Transport Policy 29: 145-153.  
- Wheat, Phill, Alexander D. Stead, William H. Greene (2019) “Robust stochastic frontier analysis: a Student’s t-half normal model with application to highway maintenance costs in England,” Journal of Productivity Analysis, 51: 21–38.  
- Xiao, Yuan, David Amaglobeli, and Riki Matsumoto (2021) “IMF Investment and Capital Stock Dataset (ICSD) 2021: Manual & FAQ - Estimating Public, Private, And PPP Capital Stocks,” (mimeo).

*Benchmarking Infrastructure Using Public Investment Efficiency Frontiers — Working Paper No.WP/2023/101*

---


_Source: https://www.imf.org/-/media/files/publications/wp/2023/english/wpiea2023101-print-pdf.pdf_
