## _wp09118 - References

## Source details

**Canonical URL:** [_wp09118 - References](https://www.imf.org/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2009/_wp09118.pdf)

## Other formats

- [Markdown version](/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2009/_wp09118.pdf.md)
- [Structured JSON version](/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2009/_wp09118.pdf.json)

---

### I. Introduction
- Research question: Does aid promote economic growth?
- Context: Large infusions of aid have been recommended as a means of escaping poverty traps (Sachs et al., 2004; Sachs 2005a, 2005b); opposing views argue aid has been ineffective (Easterly, 2007a, 2007b).
- Empirical landscape: Evidence supports aid as growth-enhancing unconditionally or conditionally (Burnside and Dollar, 2000; Guillaumont and Chauvet, 2001; Hansen and Tarp, 2001; Collier and Dollar, 2002; Gomanee, Girma, and Morrissey, 2002; Dalgaard, Hansen and Tarp, 2004; Clemens, Radelet and Bhavnani, 2004), growth-neutral (Boone 1994, 1996; Easterly, Levine, and Roodman, 2004; Easterly, 2005), or growth-depressing (Bobba and Powell, 2007).
- Contribution of this paper:
  - Distinguishes between developmental and non-developmental aid.
  - Allows aid flows to translate into economic growth after long time periods (lags).
- Main findings:
  - Developmental aid has a positive, large, and robust effect on growth.
  - Non-developmental aid is mostly growth neutral and occasionally negatively associated with economic growth.
- Policy implications:
  - Shifting the composition of aid in favor of developmental aid or increasing its quantity can lead to sizable long-term benefits.
  - Results challenge arguments for reducing donor budgets based on claims that aid is inherently ineffective.
  - Provides insight on the “macro-micro paradox”: macro studies finding zero average effects of total aid may mask positive effects of project-level developmental aid.

### II. Literature review and identification challenges
- Common assumptions in prior literature:
  - (i) Aid has a solely contemporaneous effect on growth.
  - (ii) Different kinds of aid have the same effect on growth.
- Endogeneity concerns and instrumentation:
  - Lagged aid often used as exogenous variation (Dalgaard, Hansen and Tarp, 2004).
  - “Friends of the donors” / geostrategic instruments used (Alesina and Dollar, 2000; Easterly, 2003, 2005; Rajan and Subramanian, 2008): UN voting patterns, membership of strategic alliances, colonial ties, common language.
  - Pitfalls: geostrategic instruments may primarily capture geopolitically motivated aid that is not developmental, and may fail exogeneity/exclusion restrictions (Fleck and Kilby, 2006a; Headey, 2005, 2007; Bagchi, 1982; Bertocchi and Canova, 1996; Grier, 1999; Price, 2003).
- Heterogeneous effects by aid type:
  - Evidence that different aid types have different effects (Clemens, Radelet, and Bhavnani, 2004; Gomanee, Girma, and Morrissey, 2002; Miquel-Florensa, 2007; Mishra and Newhouse, 2007; Asiedu and Nandwa, 2006).
  - Clemens et al. find short-impact aid: a $1 increase in short-impact aid raises income, on average, by $1.64 (in present value). Long-term impact aid (democracy, environment, health, education) likely important but hard to identify statistically due to data limits.
  - Rajan and Subramanian (2008) find total aid ineffective when allowing only contemporaneous effects and using geopolitical instruments; interpretation complicated because geopolitical variation does not predict developmental aid well.
  - Headey (2007) and Bobba and Powell (2007) emphasize distinguishing geostrategic vs. developmental aid:
    - Bilateral aid (70 percent of total aid) showed differing effects across Cold War and post–Cold War periods in Headey’s work.
    - Bobba and Powell: aid to non-allies has positive contemporaneous effects; aid to allies has negative effects.
- Paper’s approach: simultaneously (i) distinguish developmental vs. non-developmental aid, and (ii) allow long-run effects of aid on growth.

### III. The pitfalls of misspecification (analytic summary)
- Key critique: assuming uniform effects of aid types leads to misspecification bias when aggregating into total aid (TA = DA + NDA).
- Intuition: aggregating components with positive, zero, or negative effects can produce a zero or biased estimate for TA.
- Formal setup:
  - True model: γ = β1 DA + β2 NDA + δC + ε (equation (1) in source, where γ denotes per capita GDP growth, DA developmental aid, NDA non-developmental aid, C controls).
  - Estimated (misspecified) model: γ = β CTA + δC + ε (equation (2), TA = DA + NDA).
- OLS and 2SLS bias:
  - OLS estimator of the coefficient on TA is a weighted function of the true coefficients on DA and NDA (equation (3)). Weights depend on variances and covariances of DA, NDA, and TA conditional on covariates.
  - If DA and NDA have opposite effects, the OLS-estimated coefficient on TA can be zero.
  - If DA is effective and NDA is ineffective, the OLS estimator will suffer from attenuation bias.
  - 2SLS estimator of TA is also a weighted function of the true coefficients (equation (4)); instrumenting TA without distinguishing DA and NDA can produce misleading inference.
- Implication: misspecifying the aid variable by aggregating heterogeneous components can lead to erroneous conclusions about aid effectiveness.

### IV. Empirical strategy and scope
- Definitions and construction:
  - DA defined as "aid expended in a manner that is anticipated to promote development, whether achieved through economic growth or other means."
  - Non-developmental aid (NDA) defined as "aid of all other kinds."
  - Multilateral aid (MA) treated separately and conjectured to be "more likely to be expended in a developmental manner." OECD‑DAC definition: “Multilateral transactions are those made to a recipient institution which conducts all or part of its activities in favor of development” (DAC, 2006).
  - Practical construction of DA due to data limits:
    - Sector-level disbursement data unavailable for 1960–2000.
    - DA proxied by pooled bilateral aid from development-friendly donor groups based on aid-quality rankings and literature on aid allocation.
    - Donor group proxies:
      - G1: Denmark, Finland, the Netherlands, Norway, Sweden.
      - G2: G1 plus Austria, Canada, Ireland, Luxembourg, Switzerland.
      - G3: G1 plus Belgium, France, Ireland, Switzerland, United Kingdom.
    - NDA computed as NDA = TA – MA – DA and allowed a distinct growth impact.
- Sample and variables:
  - Sample: developing countries over 1960–2000 (DAC, 2006).
  - Aid variable: grants plus net loans with a grant element higher than 25 percent.
- Econometric approaches:
  - Cross-sectional regressions: dependent variable average per capita GDP growth over 1990–2000; DA, MA, NDA averaged over 1960–90.
  - Panel regressions: eight five-year averages 1960–2000; system GMM estimator (Blundell and Bond, 1998).
- Controls and endogeneity treatment:
  - Baseline controls: initial per capita income; initial life expectancy; institutional quality (World Bank CPIA averaged 1960–1999); geography (average frost days and tropical land area); initial government consumption; revolutions indicator; terms-of-trade growth and standard deviation; initial policy (updated Sachs‑Warner openness dummy); continent dummies (SSA, East Asia).
  - Institutional quality and revolutions treated as contemporaneously uncorrelated with growth.
  - Geography and time dummies treated as strictly exogenous instruments.
  - Endogenous: beginning-of-period per capita income, inflation, policy (openness), government consumption, and one-period-lagged aid (DA, MA, NDA).
  - Validity checks: Hansen test of over-identifying restrictions; Arellano-Bond m1 and m2 tests for AR(1) and AR(2) in differenced residuals.

### Key empirical findings — cross-sectional evidence
- Replication of Rajan and Subramanian (2008) reproduces "aid’s persistent lack of power in explaining subsequent growth" in baseline short-lag specifications (Table 3).
- Including deeper lags of TA (lags 1960–80, 1960–90, 1970–90) makes the effect of TA positive:
  - Coefficients range between 6.8 and 8.5; interpreted as "an increase of total aid during these periods by 1 percentage point of GDP is associated with an average per capita GDP growth rate that is higher by approximately 0.068 to 0.085 percentage points in the 1990s."
- DA (pooled bilateral aid from G1–G3) shows large positive effects:
  - Bilateral aid from G1 and G2: "average growth in the 1990s was higher by as much as 1.2–1.3 percentage points for countries which had received 1 additional percentage point of GDP as aid transfers from these donor countries over the previous three decades."
  - Bilateral aid from G3: 1 percentage point increase associated with subsequent growth higher by 0.14 percentage points.
  - MA: positive but statistically insignificant effect on subsequent growth.
  - NDA: small, ambiguous effect on growth.
- Cross-sectional coefficient magnitudes (as reported):
  - Total lagged aid (1960–80, 1960–90, 1970–90): 8.45**, 8.22**, 6.77** (Table 4).
  - G1 aid/GDP (1960–90): 131.97*** (Table 5).
  - G2 aid/GDP (1960–90): 120.25*** (Table 5).
  - G3 aid/GDP (1960–90): 14.75** (Table 5).

### Key empirical findings — panel evidence (system GMM)
- Lag structures tested: aid lags of 1, 3, and 5 periods (corresponding to 5, 15, 25 years).
- DA from G1 and G2:
  - A 1 percentage point increase in the DA/GDP ratio is associated with average growth higher by 0.2 percentage points 5 years later, and higher by 0.7 to 1.1 percentage points 25 years later (GMM results).
- Bilateral aid from G3: effect much smaller and not robust across specifications.
- MA and NDA: no statistically discernable impact; coefficients alternate sign and are imprecise.
- Selected reported GMM coefficient examples (Table 6 entries as reported):
  - G1 aid/GDP (5-year lag) 19.57***; (15-year lag) 14.22**; (25-year lag) 111.27**.
  - G2 aid/GDP (5-year lag) 18.41***; (15-year lag) 16.69***; (25-year lag) 74.62***.
  - G3 aid/GDP (5-year lag) 11.19**; (15-year lag) 4.77; (25-year lag) 0.52*.
- Instrument validity: Hansen test p-values indicate instruments are valid in reported specifications; m1 and m2 statistics typically show first-order but not second-order serial correlation, though validity becomes more questionable as sample size shrinks for deep lags.

### Further results and robustness checks
- Income threshold effects:
  - Interaction of DA with income-group dummies tested; data do not favor strong threshold effects.
  - Interaction term mostly insignificant, but DA coefficient remains positive, large, and statistically significant.
- Aid and policy environment:
  - Interaction of DA with policy measures (Sachs‑Warner variables; Burnside and Dollar policy index) yields inconclusive evidence.
  - Interaction term significant in 4 out of 9 cases; never reaches 1 percent significance.
- Robustness to alternative measures and specifications:
  - Results robust to alternative proxies for DA (e.g., share of total aid predicted by agricultural season variation; total bilateral aid from donors chosen by statistical criteria).
  - Replacing CPIA with ICRG index and adding initial literacy yields DA coefficients that remain significant and larger.
  - Weighing observations (Huber procedure) preserves main DA results.
  - Aggressively reducing instrument count (limit lag depth or collapse instruments) preserves panel DA results mainly for G1 and G2 proxies.

### Caveats and interpretation
- Potential confounders:
  - Donor "luck" or "cherry-picking" of recipients could bias results if development-friendly donors happened to choose high-growth recipients ex ante.
  - Evidence against extreme specialization: 90 percent of recipients in sample received aid from each development-friendly donor group over 1960–90; donor fragmentation documented in literature.
  - Growth in the 1960s and 1970s is a poor predictor of long-run growth; many early fast-growers later stagnated.
- Mechanisms remain unresolved due to incomplete donor-recipient sectoral allocation data. Possible channels include:
  - More efficient donor administrations (lower overhead; less bureaucracy).
  - Better selection of aid priorities and stronger partner relationships in recipient countries.
  - Aid from donors free of strategic preoccupations facilitating politically costly, growth-enhancing reforms.

### Policy implications
- Evidence counters the claim that aid is inherently ineffective; increasing aid and shifting composition toward developmental aid (DA) "are likely to create sizable returns in the long run."
- Donor characteristics matter for aid effectiveness; donor-recipient match quality may matter and donor motives may affect aid effectiveness independently of recipient characteristics.
- Calls for further research to:
  - Identify growth effects of distinct aid categories over relevant time periods.
  - Understand the strategies and mechanisms of the most effective donors to isolate channels through which DA works.

### Selected quantitative summaries (preserve source figures)
- Bilateral share of total aid by decade: 1960s = .895466; 1970s = .633457; 1980s = .719724; 1990s = .650658 (Figure 1).
- Bilateral aid accounts on average for 6–7 pecent of recipients’ GDP; MA to GDP ranges between 1 and 3.9 percent.
- DA from G3 contributed 5–9 percent of recipient countries’ GDP over 1960–1990; DA from G1 accounted for at most 3 percent of recipients’ GDP over the same period.
- Cross-sectional coefficient magnitudes:
  - Total lagged aid (1960–80, 1960–90, 1970–90): 8.45**, 8.22**, 6.77** (Table 4).
  - G1 aid/GDP (1960–90): 131.97*** (Table 5).
  - G2 aid/GDP (1960–90): 120.25*** (Table 5).
  - G3 aid/GDP (1960–90): 14.75** (Table 5).
- Interpreted cross-sectional effect sizes:
  - 1 percentage point of GDP increase in bilateral aid from Scandinavian countries over 1960–90 associated with average per capita GDP growth in the 1990s higher by 1.2–1.3 percentage points.
- Panel (GMM) reported coefficient examples:
  - G1 aid/GDP (5-year lag) 19.57***; (15-year lag) 14.22**; (25-year lag) 111.27**.
  - G2 aid/GDP (5-year lag) 18.41***; (15-year lag) 16.69***; (25-year lag) 74.62***.
  - G3 aid/GDP (5-year lag) 11.19**; (15-year lag) 4.77; (25-year lag) 0.52*.
- Panel summary statistics (selected):
  - Per capita GDP growth mean = 1.4407; Std. Dev. = 3.7023; Min = -14.2798; Max = 22.9337 (Panel regressions, Table 2).
  - G1 aid/GDP mean = 0.0076; Std. Dev. = 0.0254; Min = -0.0005; Max = 0.3448 (Table 2).
  - G3 aid/GDP mean = 0.0353; Std. Dev. = 0.0644; Min = -0.0004; Max = 0.5828 (Table 2).

*Source: _wp09118 - References (excerpt of paper text and appendix listing).*

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

### _wp09118 - References

### I. Introduction
- Research question: Does aid promote economic growth?
- Context: Large infusions of aid have been recommended as a means of escaping poverty traps (Sachs et al., 2004; Sachs 2005a, 2005b); opposing views argue aid has been ineffective (Easterly, 2007a, 2007b).
- Empirical landscape: Evidence supports aid as growth-enhancing unconditionally or conditionally (Burnside and Dollar, 2000; Guillaumont and Chauvet, 2001; Hansen and Tarp, 2001; Collier and Dollar, 2002; Gomanee, Girma, and Morrissey, 2002; Dalgaard, Hansen and Tarp, 2004; Clemens, Radelet and Bhavnani, 2004), growth-neutral (Boone 1994, 1996; Easterly, Levine, and Roodman, 2004; Easterly, 2005), or growth-depressing (Bobba and Powell, 2007).
- Contribution of this paper:
  - Distinguishes between developmental and non-developmental aid.
  - Allows aid flows to translate into economic growth after long time periods (lags).
  - Main findings:
    - Developmental aid has a positive, large, and robust effect on growth.
    - Non-developmental aid is mostly growth neutral and occasionally negatively associated with economic growth.
- Policy implications:
  - Shifting the composition of aid in favor of developmental aid or increasing its quantity can lead to sizable long-term benefits.
  - Results challenge arguments for reducing donor budgets based on claims that aid is inherently ineffective.
  - Provides insight on the “macro-micro paradox”: macro studies finding zero average effects of total aid may mask positive effects of project-level developmental aid.

### II. Literature review and identification challenges
- Common assumptions in prior literature:
  - (i) Aid has a solely contemporaneous effect on growth.
  - (ii) Different kinds of aid have the same effect on growth.
- Endogeneity concerns and instrumentation:
  - Lagged aid often used as exogenous variation (Dalgaard, Hansen and Tarp, 2004).
  - “Friends of the donors” / geostrategic instruments used (Alesina and Dollar, 2000; Easterly, 2003, 2005; Rajan and Subramanian, 2008): UN voting patterns, membership of strategic alliances, colonial ties, common language.
  - Pitfalls: geostrategic instruments may primarily capture geopolitically motivated aid that is not developmental, and may fail exogeneity/exclusion restrictions (Fleck and Kilby, 2006a; Headey, 2005, 2007; Bagchi, 1982; Bertocchi and Canova, 1996; Grier, 1999; Price, 2003).
- Heterogeneous effects by aid type:
  - Evidence that different aid types have different effects (Clemens, Radelet, and Bhavnani, 2004; Gomanee, Girma, and Morrissey, 2002; Miquel-Florensa, 2007; Mishra and Newhouse, 2007; Asiedu and Nandwa, 2006).
  - Clemens et al. find short-impact aid: a $1 increase in short-impact aid raises income, on average, by $1.64 (in present value). Long-term impact aid (democracy, environment, health, education) likely important but hard to identify statistically due to data limits.
  - Rajan and Subramanian (2008) find total aid ineffective when allowing only contemporaneous effects and using geopolitical instruments; interpretation complicated because geopolitical variation does not predict developmental aid well.
  - Headey (2007) and Bobba and Powell (2007) emphasize distinguishing geostrategic vs. developmental aid:
    - Bilateral aid (70 percent of total aid) showed differing effects across Cold War and post–Cold War periods in Headey’s work.
    - Bobba and Powell: aid to non-allies has positive contemporaneous effects; aid to allies has negative effects.
- Paper’s approach: simultaneously (i) distinguish developmental vs. non-developmental aid, and (ii) allow long-run effects of aid on growth.

### III. The pitfalls of misspecification (analytic summary)
- Key critique: assuming uniform effects of aid types leads to misspecification bias when aggregating into total aid (TA = DA + NDA).
- Intuition: aggregating components with positive, zero, or negative effects can produce a zero or biased estimate for TA.
- Formal setup (as in paper):
  - True model: γ = β1 DA + β2 NDA + δC + ε  (equation (1) in source, where γ denotes per capita GDP growth, DA developmental aid, NDA non-developmental aid, C controls).
  - Estimated (misspecified) model: γ = β CTA + δC + ε  (equation (2), TA = DA + NDA).
- OLS bias on TA coefficient (as shown in source): the OLS estimator of the coefficient on TA is a weighted function of the true coefficients on DA and NDA (equation (3) in source). Weights depend on variances and covariances of DA, NDA, and TA conditional on covariates.
  - Consequences:
    - If DA and NDA have opposite effects, the OLS-estimated coefficient on TA can be zero.
    - If DA is effective and NDA is ineffective, the OLS estimator will suffer from attenuation bias.
- 2SLS bias under instrumentation: the 2SLS estimator of the effect of TA on growth is also a weighted function of the true coefficients (equation (4) in source). Instrumenting TA without distinguishing DA and NDA can therefore produce misleading inference.
- Implication: misspecifying the aid variable by aggregating heterogeneous components can lead to erroneous conclusions about aid effectiveness.

### IV. Empirical strategy and scope (as described)
- The paper:
  - Defines and operationalizes developmental aid.
  - Uses cross-sectional and panel regressions to evaluate the effect of developmental aid on growth.
  - Allows aid to affect growth over long time periods by including lag structures.
- Empirical findings preview:
  - Developmental aid: positive, large, robust effect on per capita GDP growth.
  - Non-developmental aid: mostly growth neutral; sometimes negatively associated with growth.
- Availability of supplementary results: Some results not included in the main text are available in the supplementary appendix of Minoiu and Reddy (2008).

*Source: _wp09118 - References (excerpt of paper text and appendix listing).*

### conclusions because of a “strategic bias” problem (Headey, 2005) arising from the failure to

### _wp09118 - conclusions because of a “strategic bias” problem (Headey, 2005) arising from the failure to

### Defining Developmental Aid (DA) and components of aid
- DA defined as "aid expended in a manner that is anticipated to promote development, whether achieved through economic growth or other means."
- Non-developmental aid (NDA) defined as "aid of all other kinds."
- Multilateral aid (MA) treated separately and conjectured to be "more likely to be expended in a developmental manner." OECD‑DAC definition: “Multilateral transactions are those made to a recipient institution which conducts all or part of its activities in favor of development” (DAC, 2006).
- Practical construction of DA due to data limits:
  - Sector-level disbursement data unavailable for 1960–2000.
  - DA proxied by pooled bilateral aid from development-friendly donor groups based on aid-quality rankings and literature on aid allocation.
  - Donor group proxies:
    - G1: Denmark, Finland, the Netherlands, Norway, Sweden.
    - G2: G1 plus Austria, Canada, Ireland, Luxembourg, Switzerland.
    - G3: G1 plus Belgium, France, Ireland, Switzerland, United Kingdom.
- Once DA and MA are extracted from total aid (TA), NDA computed as NDA = TA – MA – DA and allowed a distinct growth impact.

### Empirical strategy and variables
- Sample: developing countries over 1960–2000 (DAC, 2006).
- Aid variable: grants plus net loans with a grant element higher than 25 percent.
- Main approaches:
  - Cross-sectional regressions: dependent variable average per capita GDP growth over 1990–2000; DA, MA, NDA averaged over 1960–90.
  - Panel regressions: eight five-year averages 1960–2000; system GMM estimator (Blundell and Bond, 1998).
- Control variables (baseline): initial per capita income; initial life expectancy; institutional quality (World Bank CPIA averaged 1960–1999); geography (average frost days and tropical land area); initial government consumption; revolutions indicator; terms-of-trade growth and standard deviation; initial policy (updated Sachs‑Warner openness dummy); continent dummies (SSA, East Asia).
- Instrument and endogeneity treatment:
  - Institutional quality and revolutions taken as contemporaneously uncorrelated with growth.
  - Geography and time dummies treated as strictly exogenous instruments.
  - Endogenous: beginning-of-period per capita income, inflation, policy (openness), government consumption, and one-period-lagged aid (DA, MA, NDA).
  - Validity checks: Hansen test of over-identifying restrictions; Arellano-Bond m1 and m2 tests for AR(1) and AR(2) in differenced residuals.

### Key empirical findings — cross-sectional evidence
- Replication of Rajan and Subramanian (2008) reproduces "aid’s persistent lack of power in explaining subsequent growth" in baseline short-lag specifications (Table 3).
- Including deeper lags of TA (lags 1960–80, 1960–90, 1970–90) makes the effect of TA positive:
  - Coefficients range between 6.8 and 8.5; interpreted as "an increase of total aid during these periods by 1 percentage point of GDP is associated with an average per capita GDP growth rate that is higher by approximately 0.068 to 0.085 percentage points in the 1990s."
- DA (pooled bilateral aid from G1–G3) shows large positive effects:
  - Bilateral aid from G1 and G2: "average growth in the 1990s was higher by as much as 1.2–1.3 percentage points for countries which had received 1 additional percentage point of GDP as aid transfers from these donor countries over the previous three decades."
  - Bilateral aid from G3: 1 percentage point increase associated with subsequent growth higher by 0.14 percentage points.
  - MA: positive but statistically insignificant effect on subsequent growth.
  - NDA: small, ambiguous effect on growth.

### Key empirical findings — panel evidence (system GMM)
- Lag structures tested: aid lags of 1, 3, and 5 periods (corresponding to 5, 15, 25 years).
- DA from G1 and G2:
  - A 1 percentage point increase in the DA/GDP ratio is associated with average growth higher by 0.2 percentage points 5 years later, and higher by 0.7 to 1.1 percentage points 25 years later (GMM results).
- Bilateral aid from G3: effect much smaller and not robust across specifications.
- MA and NDA: no statistically discernable impact; coefficients alternate sign and are imprecise.
- Instrument validity: Hansen test p-values indicate instruments are valid in reported specifications; m1 and m2 statistics typically show first-order but not second-order serial correlation, though validity of instruments becomes more questionable as sample size shrinks for deep lags.

### Further results and robustness checks
- Income threshold effects:
  - Interaction of DA with income-group dummies tested; data do not favor strong threshold effects.
  - Interaction term mostly insignificant, but DA coefficient remains positive, large, and statistically significant.
- Aid and policy environment:
  - Interaction of DA with policy measures (Sachs‑Warner variables; Burnside and Dollar policy index) yields inconclusive evidence.
  - Interaction term significant in 4 out of 9 cases; never reaches 1 percent significance.
- Robustness:
  - Results robust to alternative proxies for DA (e.g., share of total aid predicted by agricultural season variation; total bilateral aid from donors chosen by statistical criteria).
  - Replacing CPIA with ICRG index and adding initial literacy yields DA coefficients that remain significant and larger.
  - Weighing observations (Huber procedure) preserves main DA results.
  - Aggressively reducing instrument count (limit lag depth or collapse instruments) preserves panel DA results mainly for G1 and G2 proxies.

### Caveats and interpretation
- Potential confounders:
  - Donor "luck" or "cherry-picking" of recipients could bias results if development-friendly donors happened to choose high-growth recipients ex ante.
  - Evidence against extreme specialization: 90 percent of recipients in sample received aid from each development-friendly donor group over 1960–90; donor fragmentation documented in literature.
  - Growth in the 1960s and 1970s is a poor predictor of long-run growth; many early fast-growers later stagnated.
- Mechanisms remain unresolved due to incomplete donor-recipient sectoral allocation data. Possible channels include:
  - More efficient donor administrations (lower overhead; less bureaucracy).
  - Better selection of aid priorities and stronger partner relationships in recipient countries.
  - Aid from donors free of strategic preoccupations facilitating politically costly, growth-enhancing reforms.

### Policy implications (drawn from results)
- Evidence counters the claim that aid is inherently ineffective; increasing aid and shifting composition toward developmental aid (DA) "are likely to create sizable returns in the long run."
- Donor characteristics matter for aid effectiveness; donor-recipient match quality may matter and donor motives may affect aid effectiveness independently of recipient characteristics.
- Calls for further research to:
  - Identify growth effects of distinct aid categories over relevant time periods.
  - Understand the strategies and mechanisms of the most effective donors to isolate channels through which DA works.

### Selected quantitative summaries (preserve source figures)
- Bilateral share of total aid by decade: 1960s = .895466; 1970s = .633457; 1980s = .719724; 1990s = .650658 (Figure 1).
- Bilateral aid accounts on average for 6–7 pecent of recipients’ GDP; MA to GDP ranges between 1 and 3.9 percent.
- DA from G3 contributed 5–9 percent of recipient countries’ GDP over 1960–1990; DA from G1 accounted for at most 3 percent of recipients’ GDP over the same period.
- Cross-sectional coefficient magnitudes:
  - Total lagged aid (1960–80, 1960–90, 1970–90) coefficients: 8.45**, 8.22**, 6.77** (Table 4).
  - G1 aid/GDP (1960–90) coefficient: 131.97*** (Table 5).
  - G2 aid/GDP (1960–90) coefficient: 120.25*** (Table 5).
  - G3 aid/GDP (1960–90) coefficient: 14.75** (Table 5).
- Interpreted cross-sectional effect sizes:
  - 1 percentage point of GDP increase in bilateral aid from Scandinavian countries over 1960–90 associated with average per capita GDP growth in the 1990s higher by 1.2–1.3 percentage points.
  - Panel (GMM) results for DA (lagged):
    - G1 aid/GDP (5-year lag) coefficient examples: 19.57***; (15-year lag) 14.22**; (25-year lag) 111.27** (Table 6 entries as reported).
    - G2 aid/GDP (5-year lag) 18.41***; (15-year lag) 16.69***; (25-year lag) 74.62*** (Table 6 entries as reported).
    - G3 aid/GDP (5-year lag) 11.19**; (15-year lag) 4.77; (25-year lag) 0.52* (Table 6 entries as reported).
- Panel summary statistics (selected):
  - Per capita GDP growth mean = 1.4407; Std. Dev. = 3.7023; Min = -14.2798; Max = 22.9337 (Panel regressions, Table 2).
  - G1 aid/GDP mean = 0.0076; Std. Dev. = 0.0254; Min = -0.0005; Max = 0.3448 (Table 2).
  - G3 aid/GDP mean = 0.0353; Std. Dev. = 0.0644; Min = -0.0004; Max = 0.5828 (Table 2).

*Source: Authors’ calculations and text in _wp09118 (IMF Working Paper PDF content provided).*

---


_Source: https://www.imf.org/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2009/_wp09118.pdf_