## _wp1373

## Source details

**Canonical URL:** [_wp1373](https://www.imf.org/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2013/_wp1373.pdf)

## Other formats

- [Markdown version](/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2013/_wp1373.pdf.md)
- [Structured JSON version](/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2013/_wp1373.pdf.json)

---

### Overview and research question
- Investigates effects of aid on exports, imports, and per capita GDP growth ("growth") from a time-series perspective to assess dynamic implications of the "Dutch Disease" (DD) hypothesis.
- Defines a "global" aid shock as the common component of each country’s aid-to-GDP ratio and studies joint responses of exports, imports, and growth to that shock.
- Uses a heterogeneous panel vector autoregression (VAR) model identified through factor analysis (country-specific VARs estimated and averaged).

### Model and identification
- Country system (for each i, t):
  - Equation (1): B_i^0 Z_i_t = B_i Z_i_{t-1} + C_i X_i_t + ϑ_i + ξ_i t + ε_i_t
  - Z_i_t (endogenous, 4 variables): global aid; per capita GDP growth; export-to-GDP ratio; import-to-GDP ratio.
  - X_i_t (controls): dummies (years of war, civil war, three years postconflict, earthquake, drought, flood/windstorm) and three continuous controls (real commodity export price index; trading partners' import volume growth index; index of trade liberalization).
  - ε_i_t: 4-dimensional vector of serially and cross-sectionally uncorrelated innovations with variance-covariance matrix Ω_i.
  - Identification: global aid defined as cross-section average (alternative factor constructions correlated above 0.9); global aid ordered first in Cholesky decomposition with identifying assumption that global aid does not respond contemporaneously (within a year) to country-specific shocks in growth, exports, imports.
- Estimation:
  - Mean group estimator (Pesaran and Smith (1995)); country-by-country OLS VARs; arithmetic averages for impulse responses; one lag of Z_i_t used.

### Data and sample
- Data sources:
  - Per capita GDP growth: Penn World Table.
  - Exports and imports (to-GDP ratios): IMF International Financial Statistics.
  - Aid: OECD "net official development assistance," entered as share of GDP.
- Sample selection:
  - Start: 55 countries (median aid-to-GDP > 2 percent and population > 1 million).
  - Selected: 37 countries with data > 20 years over 1960–2002.
  - Excluded 7 countries for explosive dynamic responses: Ethiopia, Ghana, Lesotho, Myanmar, Senegal, Chad, Zimbabwe.
  - Final estimation sample: 30 countries.
  - Global-aid average constructed from 42 countries with available aid data (five of these have incomplete series >20 years but retained for common component).
- Persistence statistics (sample autocorrelations, Table 1):
  - Per capita GDP growth: Mean 0.10; Median 0.15; Stdev. 0.19.
  - Aid-to-GDP: Mean 0.69; Median 0.75; Stdev. 0.22.
  - Export-to-GDP: Mean 0.74; Median 0.80; Stdev. 0.24.
  - Import-to-GDP: Mean 0.70; Median 0.79; Stdev. 0.28.

### Main empirical findings (impulse responses, heterogeneity)
- Global aid series behavior (1966–2002): upward until end of Cold War in 1991, then declines, with large increases in 1994, 2001, and 2002.
- Shock magnitude:
  - One-standard deviation shock to global aid ≈ one percent of GDP on impact on average; cumulative long-run disbursements reach 2 percent of GDP on average.
- Comovement and statistical association:
  - Scatter regression of cumulative per capita GDP growth response vs cumulative export-to-GDP response:
    - Coef. = 0.4698092
    - P-value = 0.001
  - First principal component of aid-to-GDP explains about 45 percent of variance of aid-to-GDP series in the sample.
  - Country-specific OLS loadings of individual aid-to-GDP on global aid: Mean 0.47; Median 0.69; Standard deviation 0.48; Adjusted Standard Deviation 2/0.08 (reported).
- Heterogeneity (by sign of cumulative growth response):
  - Positive-growth response group: N = 14 countries.
    - Long-run cumulative effects for a one-standard deviation (cumulative 2 percent of GDP) shock:
      - GDP increases cumulatively by about 1.5 percent.
      - Exports expand cumulatively by about 1 percent of GDP.
      - Imports expand cumulatively by about 4 percent of GDP.
  - Negative-growth response group: N = 16 countries.
    - Long-run cumulative effects:
      - GDP per capita falls cumulatively by more than 2 percent.
      - Exports decline cumulatively by about 1.5 percent of GDP.
      - Imports fall cumulatively by almost 2 percent of GDP.
  - Grouping by sign of export response yields similar groups (positive-export response group: N = 13; negative-export response group: N = 17).
  - Linearity: responses scale proportionally (example: a 4 percent of GDP cumulated aid shock — two-standard deviation shock — in negative group reduces cumulated imports and exports by about 3 and 4 percent of GDP, respectively, and GDP per capita by 4 percent).
- Interpretation of import behavior:
  - Positive-response group: imports increase more than exports by about cumulative aid response (~2 percent of GDP) → aid primarily finances a trade gap (imports in excess of exports) → less pressure on non-tradable prices/real exchange rate → better export performance.
  - Negative-response group: cumulative import response ≈ cumulative export response → aid spent on domestically produced goods (including non-tradables) → likely real exchange rate appreciation and worse export and growth outcomes.
- Association with covariates (second-stage, Table 3):
  - Measures of exchange rate overvaluation are statistically significant predictors of cumulative growth and export responses.
  - Financial development (M2/GDP) significant only for growth response.
  - Share of multilateral aid significant only for export response.
  - Overvaluation measures and multilateral aid share are highly negatively correlated (correlation reported as high).
- Magnitude relative to overall volatility:
  - Variance decompositions: share of variance of per capita growth and exports explained by global aid shock is around 10 percent of total variance (slightly above for exports; slightly below for growth), regardless of grouping.

### Theoretical framing and interpretation
- Focus on "weak" Dutch Disease: whether export performance deteriorates in response to a temporary aid shock, curtailing import possibilities and temporarily reducing growth.
- Notes on persistence and "strong" DD:
  - Per capita growth is not persistent in the sample (Table 1), while aid, export, and import ratios are more persistent but stationary.
  - Aid shocks cannot affect trend growth in the aggregate data used; temporary growth losses are the relevant mechanism that can still entail permanent reductions in income per capita levels.
- Causal interpretation emphasized:
  - Evidence consistent with a causal link from aid to growth mediated by exchange rate overvaluation and resulting export decline; interpretation does not contradict DD hypothesis.
  - Composition of aid spending (import-intensive purchases or public capital/infrastructure supporting traded goods) may be as important as macroeconomic management of the real exchange rate in determining outcomes.

### Policy implications and limitations
- Policy-relevant implications:
  - Exchange rate policy should be an important area of focus to contain contagion from Dutch disease in response to aid inflows.
  - Allocation of aid between imported goods and domestically produced goods appears key: when aid finances imported goods (creating a trade gap), exchange rate appreciation and export contraction are less likely; when aid is spent on domestically produced non-tradables, appreciation and negative export/growth outcomes are more likely.
- Limitations and future research:
  - The paper documents association between exchange rate overvaluation and negative comovements but does not provide definitive policy prescriptions to avoid DD.
  - The empirical strategy is not designed to discriminate among alternative channels (e.g., productivity-enhancing infrastructure); investigation of specific channels and precise policy conditions enabling positive responses of exports, imports, and growth to aid is left to future research.

*Italic: Source — _wp1373 (Introduction; Sections 3–5, PDF chapter/section).*

### Introduction ...........................................................................................................

### _wp1373 - Introduction

### Overview and research question
- Investigates the effects of aid on exports, imports, and per capita GDP growth (referred to as "growth") from a time-series perspective to assess dynamic implications of the "Dutch Disease" (DD) hypothesis.
- Defines a "global" aid shock as the common component of each country’s aid-to-GDP ratio and studies the joint response of exports, imports, and growth to that shock.
- Uses a heterogeneous panel vector autoregression (VAR) model identified through factor analysis — described as a novel methodological approach in this literature.

### Main empirical findings
- Evidence is consistent with a DD mechanism in some countries, but not universally:
  - A global aid shock can affect exports, imports, and growth either positively or negatively.
  - The sign of the impact on exports and imports tends to be the same as that on growth within each country.
  - About half of the countries in the sample show a joint-negative response of exports, imports, and growth to a global aid shock — suggestive of DD.
  - The other half of the countries display a joint-positive response of exports, imports, and growth to a global aid shock.
- The first principal component of the aid-to-GDP series in the sample explains about 45 percent of their variance.
- Duration and intensity of exchange rate overvaluation periods in each aid-receiving country are negatively associated with the estimated responses of exports and growth to a global aid shock.
- Other standard growth covariates fail to account for the sign of the comovement between aid, exports, and growth.

### Interpretation and implications
- The observed positive comovement of exports, imports, and growth in response to global aid shocks is a novel empirical result; prior literature often examined growth or exports separately and could not reveal this joint effect.
- The evidence suggests a causal link from aid to growth mediated by exchange rate overvaluation and resulting export decline; this interpretation does not contradict the DD hypothesis.
- Findings point to exchange rate policy as an area of focus to identify policy actions that could permit countries to avoid DD.
- Composition of aid spending (e.g., import-intensive purchases or public capital/infrastructure that supports the traded-good sector such as energy and transportation) may be as important as macroeconomic management of the real exchange rate in determining outcomes.

### Methodological contributions and strategy
- Identification of exogenous aid changes uses the common component of country-specific aid-to-GDP ratios ("global aid") rather than seeking external instrumental variables.
  - Global aid captures donors’ common aid policies (e.g., MDG-related surges, HIPC initiative, Cold War) and, by construction, does not respond to recipient-specific economic developments.
  - Global aid correlates well with individual country aid series (see above: first principal component explains about 45 percent of variance).
- Controls are included for potential confounders:
  - Global business cycle: index of commodity export prices and a measure of import demand by trading partners from the IMF World Economic Outlook.
  - Global waves of trade liberalization: index of trade openness of Wacziarg and Welch (2008).
- Rationale versus instrumental-variable approaches:
  - Global aid is intended to isolate donor-policy-driven variation less likely to constrain recipient responses (e.g., unlike tied aid instruments such as arms imports).
  - Unlike many instruments suited for cross-sectional analysis, global aid allows tracing effects over time and comovements with exports, imports, and per capita GDP growth.
- Panel VAR approach features:
  - Jointly models growth, exports, and imports to check for matching effects consistent with the weak DD hypothesis.
  - Allows a maximum degree of parameter heterogeneity by estimating a separate VAR for each country rather than imposing ex-ante heterogeneity through interacted covariates.

### Theoretical framing: "weak" versus "strong" Dutch Disease
- Focuses on a "weak" form of the DD hypothesis: whether export performance deteriorates in response to a temporary aid shock, curtailing import possibilities and temporarily reducing growth.
- Notes that standard DD models predict permanent losses to trend growth via lost learning-by-doing in tradable sectors, but this "strong" form conflicts with empirical evidence that per capita growth is not persistent.
  - Table 1 (referenced) indicates per capita growth is not persistent in the sample; aid, export and import ratios are more persistent but must also be stationary according to Table 1.
  - Hence, aid shocks cannot affect trend growth in the aggregate data used here; temporary growth losses are the relevant mechanism that can still entail permanent reductions in income per capita levels.
- Emphasizes that temporary exchange rate appreciation leading to reallocation away from tradables can produce temporary growth losses with potentially large welfare implications.

### Heterogeneity analysis and second-stage exploration
- Documents heterogeneous country responses to global aid shocks in the first stage.
- Second-stage analysis regresses estimated country-specific responses on a broad set of growth-related variables:
  - Institutional quality, geography, aid composition, macroeconomic policy stance, and measures of exchange rate overvaluation.
- This two-stage strategy allows investigation of alternative hypotheses for heterogeneity beyond those emphasized in prior work (e.g., Rajan and Subramanian).

### Scope, limitations, and suggested future work
- The paper identifies exchange rate overvaluation as closely associated with negative comovements but does not provide definitive policy prescriptions to avoid DD.
- Examining precise mechanisms and policy conditions that enable some countries to generate positive responses of exports, imports, and growth to aid (and thereby avoid DD) is left for future research.
- Notes relevant strands of literature (exchange rate policy; aid composition) but specifies that in-depth analysis of policy solutions is beyond the paper’s scope.

*Italic source: _wp1373 - Introduction (PDF chapter/section)*

### Section 3 presents our data set. Section 4 reports the estimation results and discuss their

### _wp1373 - Section 3 presents our data set. Section 4 reports the estimation results and discuss their

### Methodology — empirical model
- Model: stationary, heterogenous panel vector-autoregression (PVAR) as in Holtz-Eakin, Newey and Rosen (1988); Attanasio, Picci, and Scorcu (2000); Canova and Ciccarelli (2004).
- Country system (for each i, t):
  - Equation (1): B_i^0 Z_i_t = B_i Z_i_{t-1} + C_i X_i_t + ϑ_i + ξ_i t + ε_i_t
  - X_i_t: vector of country-specific controls.
  - ε_i_t: 4-dimensional vector of serially and cross-sectionally uncorrelated innovations with variance-covariance matrix Ω_i.
  - ϑ_i: 4-dimensional vector of country-specific effects.
  - t: time trend.
  - B_i^0, B_i, C_i, ξ_i: coefficient matrices.
  - N: number of countries; T: number of time periods (years).
- Endogenous variables Z_i_t (four variables):
  - global aid (measure discussed below),
  - per capita GDP growth,
  - export-to-GDP ratio,
  - import-to-GDP ratio.
- Data choices and rationale:
  - Per capita GDP growth: Penn World Table.
  - Export-to-GDP and import-to-GDP: IMF International Financial Statistics.
  - Aid: OECD concept of “net official development assistance,” entered as share of GDP (as are import and export) to link via balance of payments identity under no access to private international capital markets assumption.
  - Imports included to distinguish channels (if aid is spent on imports vs domestically produced goods) and to interpret comovements among growth, exports, and imports.
- Time series properties:
  - All variables in Z_i_t assumed covariance-stationary.
  - Only one lag of Z_i_t included (annual data; degrees-of-freedom constraint).
  - VAR specification identical across countries to avoid specification-induced differences.

### Methodology — controls, estimation, and identification
- Controls X_i_t:
  - Dummies: (1) years of war; (2) years of civil war; (3) three years postconflict after end of war/civil war; (4) earthquake years; (5) drought years; (6) flood or windstorm years.
  - Three continuous controls: (i) real commodity export price index, (ii) trading partners' import volume growth index, (iii) index of trade liberalization (Wacziarg and Welch, 2008).
  - Sources: IMF World Economic Outlook database and Wacziarg and Welch (2008).
- Estimation:
  - Mean group estimator of Pesaran and Smith (1995); Pesaran, Smith, and Im (1995).
  - Country-by-country OLS estimation of VARs; arithmetic averages across countries used for impulse responses and variance decompositions. Weighted averages by inverse standard error computed as robustness check (similar results, not reported).
  - Matrix B_i in (1) left unrestricted (no imposed exclusions of lagged growth/export/import on global aid) to simplify estimation; absence of contemporaneous impact of country variables on global aid suffices for identification.
- Identification of global aid shock:
  - Global aid defined as cross-section average, for each year, of country-specific aid-to-GDP series (Forni & Reichlin, 1998 approach).
  - Alternative constructions considered: dynamic common factor, static common factor, first principal component — all highly correlated with the simple average (sample correlations above 0.9).
  - First principal component explains 45 percent of variability of aid-to-GDP series in the sample.
  - Country-specific OLS regressions of individual aid-to-GDP on global aid report average loading ~0.5 and median ~0.7 (Table 2: Mean 0.47; Median 0.69; Standard deviation 0.48; Adjusted Standard Deviation 2/0.08).
  - Identification assumption: global aid does not respond contemporaneously (within a year) to shocks to per capita GDP growth, exports, and imports in each country VAR.
  - Given that assumption, exogenous global aid shock and associated impulse responses derived from Cholesky decomposition with global aid ordered first.
  - Augmenting system with controls (real commodity price index, trading partners’ import volume index, trade liberalization index) to purge reduced-form residuals of common global shocks (global business cycle, trade liberalization waves) that might contemporaneously affect both global aid and country variables.

### Data
- Sample selection:
  - Start with 55 countries: median aid-to-GDP ratio greater than 2 percent and population above 1 million (out of 107 potential aid-receiving countries).
  - From these 55, select 37 countries with data availability for more than 20 years over 1960–2002.
  - Exclude 7 countries displaying explosive dynamic responses to global aid shock: Ethiopia, Ghana, Lesotho, Myanmar, Senegal, Chad, Zimbabwe.
  - Final estimation sample: 30 countries.
  - Average across countries of aid-to-GDP series for constructing global aid includes 42 countries for which aid data are available; five of these 42 lack complete series >20 years but are retained for computing the common component.
- Persistence statistics (Table 1: sample autocorrelations):
  - Per capita GDP growth: Mean 0.10; Median 0.15; Stdev. 0.19.
  - Aid-to-GDP: Mean 0.69; Median 0.75; Stdev. 0.22.
  - Export-to-GDP: Mean 0.74; Median 0.80; Stdev. 0.24.
  - Import-to-GDP: Mean 0.70; Median 0.79; Stdev. 0.28.

### Results — impulse responses and heterogeneity
- Global aid series behavior (1966–2002):
  - Trends upward until end of Cold War in 1991, then declines steadily with three large increases in 1994, 2001, and 2002 (debt relief and HIPC-related deliveries).
- Main impulse-response evidence:
  - Shock magnitude: one-standard deviation shock to global aid ~ one percent of GDP on impact on average across aid-receiving countries; cumulative long-run disbursements reach 2 percent of GDP on average.
  - Figure 3: scatter of cumulative per capita GDP growth response vs cumulative export-to-GDP response to a one-standard deviation global aid shock.
    - Regression coefficient: Coef. = 0.4698092
    - P-value = 0.001
  - First main result: long-run comovement between growth and export responses to global aid shock is positive, economically sizable, and statistically significant — consistent with the weak form of the Dutch disease (DD) hypothesis considered.
- Heterogeneity across countries:
  - Approximately half the sample: global aid shock produces positive cumulative effects on per capita GDP growth and exports; the other half: negative cumulative effects.
  - Grouping by sign of cumulative growth response:
    - Positive-growth response group: N = 14 countries.
      - Long-run cumulative effects for a one-standard deviation (cumulative 2 percent of GDP) shock:
        - GDP increases cumulatively by about 1.5 percent.
        - Exports expand cumulatively by about 1 percent of GDP.
        - Imports expand cumulatively by about 4 percent of GDP.
    - Negative-growth response group: N = 16 countries.
      - Long-run cumulative effects:
        - GDP per capita falls cumulatively by more than 2 percent.
        - Exports decline cumulatively by about 1.5 percent of GDP.
        - Imports fall cumulatively by almost 2 percent of GDP.
    - Note: linearity of VAR implies scaling responses proportionally (example given: 4 percent of GDP cumulated aid — two-standard deviation shock — in negative group reduces cumulated imports and exports by about 3 and 4 percent of GDP, respectively, and GDP per capita by 4 percent).
  - Grouping by sign of export response yields very similar dynamic responses (positive-export response group: N = 13; negative-export response group: N = 17).
- Interpretation of import behavior and transmission channels:
  - Positive-response group: imports increase more than exports by about the cumulative aid response (~2 percent of GDP), implying aid primarily finances a trade gap (imports in excess of exports) rather than purchases of domestically produced goods — less upward pressure on non-tradable prices and real exchange rate, better export performance.
  - Negative-response group: cumulative import response approximately equals cumulative export response (no trade gap opened by aid shock) — suggests aid spent on domestically produced goods (including non-tradables), possibly inducing real exchange rate appreciation and worse export and growth outcomes.
- Association with other covariates (Table 3):
  - Univariate OLS regressions of cumulative responses (growth and exports) on selected covariates (subset from Rajan and Subramanian (2007) with sample correlation > 0.1).
  - Three measures of exchange rate overvaluation included.
  - Key finding: cumulative growth and export responses are associated with measures of overvaluation; measures of overvaluation are statistically significant at conventional levels for both growth and export responses.
  - Other covariates:
    - Financial development (M2/GDP) significant only for growth response.
    - Share of multilateral aid significant only for export response.
    - Overvaluation measures and multilateral aid share are highly negatively correlated (correlation not reported).
- Magnitude relative to overall volatility:
  - Variance decompositions (Figure 6): share of variance of per capita growth and exports explained by global aid shock is around 10 percent of total variance (slightly above for exports; slightly below for growth), regardless of grouping criterion and response group.
  - Conclusion: although dynamic impulse responses are sizable, global aid shocks explain only a modest fraction of growth and export volatility.

### Conclusions and policy implications
- Summary of main empirical findings:
  - (i) A global aid shock causes cumulative imports, exports, and per capita GDP growth to comove positively and strongly in the long run (weak DD form).
  - (ii) Export and growth responses correlate closely with measures of exchange rate overvaluation, more so than with many typical growth covariates used in the literature.
  - Heterogeneity: about half the sample exhibits DD-consistent negative cumulative effects on exports, imports, and growth; the other half exhibits positive cumulative effects — both economically sizable and statistically significant.
- Interpretation and policy-relevant implication:
  - Differences in allocation of aid between imported goods and domestically produced goods appear to explain why some countries avoid DD effects: when aid finances imported goods (creating a trade gap), exchange rate appreciation and export contraction are less likely; when aid is spent on domestically produced non-tradables, appreciation and negative export/growth outcomes are more likely.
  - Exchange rate policy should be an important area of focus to contain contagion from Dutch disease in response to aid inflows.
- Limitations and avenues for future research:
  - The empirical strategy is designed to identify and illustrate the DD hypothesis and is not designed to discriminate among alternative channels through which aid can spur exports and growth (e.g., productivity-enhancing infrastructure). Investigation of specific channels is left to future research.

*Italic: Source — _wp1373 (Section 3–5 content, PDF chapter/section).*

---


_Source: https://www.imf.org/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2013/_wp1373.pdf_
