## _wp10240 - References

## Source details

**Canonical URL:** [_wp10240 - References](https://www.imf.org/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2010/_wp10240.pdf)

## Other formats

- [Markdown version](/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2010/_wp10240.pdf.md)
- [Structured JSON version](/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2010/_wp10240.pdf.json)

---

### I. Introduction — scope and key research questions
- Data and period: bilateral dataset over the period 1970–2005.
- Sample: 22 donors to 113 recipients, giving 90,000 observations.
- Main questions:
  - To what extent does the business cycle in donor countries influence development aid outlays?
  - How do macroeconomic conditions in recipient countries influence aid receipts?
  - How do large adverse shocks in donors and recipients alter aid behavior?

### II. Core empirical findings
- Donor-cycle procyclicality
  - Aid flows are on average procyclical with respect to the donor output cycle.
  - A one percentage point increase in the donor output gap raises real aid outlays on average by between 8 and 11 percent (depending on the output gap estimate).
  - Expansions raise aid disbursements by one fifth in the full sample.
- Donor large-shock effects
  - In years with large negative output gap, aid outlays fall by between 32 and 89 percent in the full sample (depending on the output gap estimate).
  - Growth recessions reduce aid disbursements by 12 percent.
  - High donor public debt magnifies reductions in aid: each percentage point of GDP of public debt reduces aid flows by almost 2 percent; the negative effect is amplified during severe donor downturns (persistent up to two years).
- Recipient-cycle effects
  - Bilateral aid disbursements are procyclical with respect to the recipient output cycle on average: decreases of 15-21 percent in years of below-trend growth.
  - This procyclicality is driven mainly by middle-income countries; low-income countries’ output has a statistically zero effect on flows in baseline results.
- Recipient large-shock (countercyclical) effects
  - Large adverse recipient shocks attract more aid:
    - Large TOT shocks: bilateral aid increases on average by 16 percent in the full sample; TOT shocks result in a 28 percent increase in aid outlays for low-income countries.
    - Growth collapses (sustained decelerations to negative growth lasting at least three years): bilateral aid increases by 43 percent on average in the full sample; middle-income countries experience a 65 percent increase.
  - Dynamic persistence:
    - Large TOT shocks: aid flows almost double by the third year following a TOT collapse (especially to low-income countries).
    - Growth collapses: attract more aid later in the episode in middle-income countries.
- Donor–recipient simultaneous shocks
  - Interaction terms between donor and recipient cycle variables are statistically indistinguishable from zero: no additional impact when donor and recipient are simultaneously hit by negative shocks.
- Non-linearities
  - Evidence of a non-linear relationship between aid flows and donor/recipient cycles: severe recessions in donor countries lead to more than proportional reductions in bilateral aid flows.
  - Fitted relationships show aid increases with the donor output gap but at decreasing rates; the turning point (beyond which aid becomes countercyclical) occurs earlier for middle-income countries.
- Institutions and IMF programs
  - Better institutional quality in recipients (Polity IV score) is associated with higher aid inflows, particularly for low-income countries (level effect).
  - Interaction: aid has a stronger cushioning effect in countries with better institutions during TOT shocks and growth collapses.
  - IMF programs play a catalytic role: countries with IMF programs receive double the amount of bilateral aid than countries without an IMF program (when included in augmented specifications).

### III. Data, definitions, and methodology (concise)
- Dependent variable: real bilateral foreign aid defined as bilateral ODA net of principal repayments, subtracting humanitarian emergency aid, development food aid, and debt forgiveness grants.
- Treatment of zeros and negatives: semi-log transformation |),|1log()(*
  ijt
  aidaidsignaid used to retain zero and negative entries.
- Business cycle measures:
  - Donor output gap: permanent vs transitory component of GDP via log-linear regression against time; OECD dating methodology used as robustness (correlation 0.56 across donors).
  - Binary donor expansion indicator: above-trend real growth.
  - Recipient cycle proxies: HP-filtered output gap with λ=1 and binary below-trend GDP and consumption growth dummies.
- Large-shock definitions:
  - Donors: dummies for output gap or growth deviations falling in the bottom quartile of donor-specific distribution.
  - Recipients: (i) TOT shocks = year-on-year TOT growth rates falling in the bottom decile of recipient-specific distribution; (ii) growth collapses = sustained decelerations to negative growth lasting at least three years.
- Estimation:
  - Baseline: OLS with country-pair fixed effects and time dummies; standard errors clustered at country-pair level.
  - Robustness: gross aid flows as dependent variable; Tobit with random effects; models accounting for aid persistence with lagged aid.

### IV. Robustness checks and extensions — key results
- Gross vs net aid: qualitative findings unchanged; gross disbursements remain procyclical with respect to donor and recipient cycles and countercyclical for large recipient shocks.
- Tobit with random effects: results confirm baseline patterns (e.g., donor output gap increases real aid outlays by between 4 and 6 percent depending on cycle measure; large recession year reduces gross disbursements by 46 percent).
- Accounting for aid persistence:
  - Pair-wise bilateral aid autoregressive coefficient ≈ 0.5.
  - Core results (procyclicality; countercyclicality for large recipient shocks) remain material when including lagged aid.
- Post-Cold War regime:
  - Interactions between post-1989 indicator and donor cycle variables are zero → bilateral aid equally procyclical before and after 1990.
  - Recipient-side interactions: 3 out of 5 significant coefficients → average procyclicality vis-à-vis recipient output cycle and countercyclicality for large TOT shocks mainly reflect post-Cold War behavior.

### V. Policy-relevant implications and scenarios
- Historical downside risks to aid in the face of severe donor recessions:
  - Severe economic downturns in donor countries have historically triggered persistent declines in aid disbursements (effects can persist up to two years).
  - Non-linearities imply disproportionate aid cuts in severe recessions.
  - Higher donor public debt worsens aid reductions following large donor output contractions.
- Countercyclical buffering potential:
  - Bilateral aid can act as insurance for recipients exposed to large external shocks, notably:
    - Low-income countries: cushioning from large TOT shocks.
    - Middle-income countries: stronger bilateral responses to growth collapses.
  - Good recipient institutional quality enhances the cushioning effect.
- Catalytic role of IMF programs: presence of IMF programs is associated with substantially higher bilateral aid inflows (double the amount in augmented specifications).

### VI. Representative quantitative highlights (exact numbers retained)
- Sample and observations: 22 donors; 113 recipients; 90,000 observations; period 1970–2005.
- Share of zero entries: drops from 70 percent in 1970 to 24 percent in 2005; average over 1970–2005 is 43 percent.
- Correlation: two donor output gap estimates correlation = 0.56.
- Donor-cycle semi-elasticities: donor output gap → aid rises by between 8 and 11 percent per one percentage point increase.
- Large donor negative shock impacts: aid falls by between 32 and 89 percent (depending on output gap estimate) in large negative output gap years; growth recessions reduce aid by 12 percent.
- Recipient large-shock impacts:
  - TOT shocks: +16 percent aid (full sample); +28 percent (low-income countries).
  - Growth collapses: +43 percent aid (full sample); +65 percent (middle-income countries).
- Public debt effect: each percentage point of donor public debt-to-GDP reduces aid by almost 2 percent.
- Aid persistence: estimated autoregressive coefficient ≈ 0.5 in pair-wise bilateral aid.
- Frequency of growth collapses: 45 percent of recipient-year observations belong to a growth collapse episode.
- Figures/tables enumerated in source (listing preserved in source but not reproduced here): multiple tables (Table 1–17) and figures (Figure 1–6) document variable definitions, country lists, descriptive statistics, baseline and robustness regressions, and fitted relationships.

*Source: _wp10240 - References, IMF PDF content (excerpts provided).*

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

### _wp10240 - References .............................................................................................................

### I. Introduction — scope and key research questions
- Data and period: bilateral dataset over the period 1970–2005.
- Sample: 22 donors to 113 recipients, giving 90,000 observations.
- Main questions:
  - To what extent does the business cycle in donor countries influence development aid outlays?
  - How do macroeconomic conditions in recipient countries influence aid receipts?
  - How do large adverse shocks in donors and recipients alter aid behavior?

### II. Core empirical findings
- Donor-cycle procyclicality
  - Aid flows are on average procyclical with respect to the donor output cycle.
  - A one percentage point increase in the donor output gap raises real aid outlays on average by between 8 and 11 percent (depending on the output gap estimate).
  - Expansions raise aid disbursements by one fifth in the full sample.
- Donor large-shock effects
  - In years with large negative output gap, aid outlays fall by between 32 and 89 percent in the full sample (depending on the output gap estimate).
  - Growth recessions reduce aid disbursements by 12 percent.
  - High donor public debt magnifies reductions in aid: each percentage point of GDP of public debt reduces aid flows by almost 2 percent; the negative effect is amplified during severe donor downturns (persistent up to two years).
- Recipient-cycle effects
  - Bilateral aid disbursements are procyclical with respect to the recipient output cycle on average: decreases of 15-21 percent in years of below-trend growth.
  - This procyclicality is driven mainly by middle-income countries; low-income countries’ output has a statistically zero effect on flows in baseline results.
- Recipient large-shock (countercyclical) effects
  - Large adverse recipient shocks attract more aid:
    - Large TOT shocks: bilateral aid increases on average by 16 percent in the full sample; TOT shocks result in a 28 percent increase in aid outlays for low-income countries.
    - Growth collapses (sustained decelerations to negative growth lasting at least three years): bilateral aid increases by 43 percent on average in the full sample; middle-income countries experience a 65 percent increase.
  - Dynamic persistence:
    - Large TOT shocks: aid flows almost double by the third year following a TOT collapse (especially to low-income countries).
    - Growth collapses: attract more aid later in the episode in middle-income countries.
- Donor–recipient simultaneous shocks
  - Interaction terms between donor and recipient cycle variables are statistically indistinguishable from zero: no additional impact when donor and recipient are simultaneously hit by negative shocks.
- Non-linearities
  - Evidence of a non-linear relationship between aid flows and donor/recipient cycles: severe recessions in donor countries lead to more than proportional reductions in bilateral aid flows.
  - Fitted relationships show aid increases with the donor output gap but at decreasing rates; the turning point (beyond which aid becomes countercyclical) occurs earlier for middle-income countries.
- Institutions and IMF programs
  - Better institutional quality in recipients (Polity IV score) is associated with higher aid inflows, particularly for low-income countries (level effect).
  - Interaction: aid has a stronger cushioning effect in countries with better institutions during TOT shocks and growth collapses.
  - IMF programs play a catalytic role: countries with IMF programs receive double the amount of bilateral aid than countries without an IMF program (when included in augmented specifications).

### III. Data, definitions, and methodology (concise)
- Dependent variable: real bilateral foreign aid defined as bilateral ODA net of principal repayments, subtracting humanitarian emergency aid, development food aid, and debt forgiveness grants.
- Treatment of zeros and negatives: semi-log transformation |),|1log()(*
  ijt
  aidaidsignaid used to retain zero and negative entries.
- Business cycle measures:
  - Donor output gap: permanent vs transitory component of GDP via log-linear regression against time; OECD dating methodology used as robustness (correlation 0.56 across donors).
  - Binary donor expansion indicator: above-trend real growth.
  - Recipient cycle proxies: HP-filtered output gap with λ=1 and binary below-trend GDP and consumption growth dummies.
- Large-shock definitions:
  - Donors: dummies for output gap or growth deviations falling in the bottom quartile of donor-specific distribution.
  - Recipients: (i) TOT shocks = year-on-year TOT growth rates falling in the bottom decile of recipient-specific distribution; (ii) growth collapses = sustained decelerations to negative growth lasting at least three years.
- Estimation:
  - Baseline: OLS with country-pair fixed effects and time dummies; standard errors clustered at country-pair level.
  - Robustness: gross aid flows as dependent variable; Tobit with random effects; models accounting for aid persistence with lagged aid.

### IV. Robustness checks and extensions — key results
- Gross vs net aid: qualitative findings unchanged; gross disbursements remain procyclical with respect to donor and recipient cycles and countercyclical for large recipient shocks.
- Tobit with random effects: results confirm baseline patterns (e.g., donor output gap increases real aid outlays by between 4 and 6 percent depending on cycle measure; large recession year reduces gross disbursements by 46 percent).
- Accounting for aid persistence:
  - Pair-wise bilateral aid autoregressive coefficient ≈ 0.5.
  - Core results (procyclicality; countercyclicality for large recipient shocks) remain material when including lagged aid.
- Post-Cold War regime:
  - Interactions between post-1989 indicator and donor cycle variables are zero → bilateral aid equally procyclical before and after 1990.
  - Recipient-side interactions: 3 out of 5 significant coefficients → average procyclicality vis-à-vis recipient output cycle and countercyclicality for large TOT shocks mainly reflect post-Cold War behavior.

### V. Policy-relevant implications and scenarios
- Historical downside risks to aid in the face of severe donor recessions:
  - Severe economic downturns in donor countries have historically triggered persistent declines in aid disbursements (effects can persist up to two years).
  - Non-linearities imply disproportionate aid cuts in severe recessions.
  - Higher donor public debt worsens aid reductions following large donor output contractions.
- Countercyclical buffering potential:
  - Bilateral aid can act as insurance for recipients exposed to large external shocks, notably:
    - Low-income countries: cushioning from large TOT shocks.
    - Middle-income countries: stronger bilateral responses to growth collapses.
  - Good recipient institutional quality enhances the cushioning effect.
- Catalytic role of IMF programs: presence of IMF programs is associated with substantially higher bilateral aid inflows (double the amount in augmented specifications).

### VI. Representative quantitative highlights (exact numbers retained)
- Sample and observations: 22 donors; 113 recipients; 90,000 observations; period 1970–2005.
- Share of zero entries: drops from 70 percent in 1970 to 24 percent in 2005; average over 1970–2005 is 43 percent.
- Correlation: two donor output gap estimates correlation = 0.56.
- Donor-cycle semi-elasticities: donor output gap → aid rises by between 8 and 11 percent per one percentage point increase.
- Large donor negative shock impacts: aid falls by between 32 and 89 percent (depending on output gap estimate) in large negative output gap years; growth recessions reduce aid by 12 percent.
- Recipient large-shock impacts:
  - TOT shocks: +16 percent aid (full sample); +28 percent (low-income countries).
  - Growth collapses: +43 percent aid (full sample); +65 percent (middle-income countries).
- Public debt effect: each percentage point of donor public debt-to-GDP reduces aid by almost 2 percent.
- Aid persistence: estimated autoregressive coefficient ≈ 0.5 in pair-wise bilateral aid.
- Frequency of growth collapses: 45 percent of recipient-year observations belong to a growth collapse episode.
- Figures/tables enumerated in source (listing preserved in source but not reproduced here): multiple tables (Table 1–17) and figures (Figure 1–6) document variable definitions, country lists, descriptive statistics, baseline and robustness regressions, and fitted relationships.

*Source: _wp10240 - References, IMF PDF content (excerpts provided).*

### REFERENCES

### _wp10240 - REFERENCES

### Literature and thematic coverage
- Extensive references on foreign aid determinants, effectiveness, and cyclicality, including studies on:
  - Who gives aid and why (Alesina and Dollar, 2000).
  - Corruption and aid reception (Alesina and Weder, 2002).
  - Aid, policies, and growth (Burnside and Dollar, 2000).
  - Aid as insurance and cyclical behavior (Pallage and Robe, 2001; Pallage, Robe, and Beroube, 2006).
  - Donor political motives and voting influence (Dreher and Sturm, 2006; Meernik, Krueger, and Poe, 1998).
  - Effects of financial crises on aid (Mold et al., 2008; Mendoza, Jones, and Vergara, 2009; Roodman, 2009).
  - Long-run relations between development aid and growth (Minoiu and Reddy, 2010).
- Methodology and data sources cited:
  - PWT Mark 6.3 (Heston, Summers and Aten, 2009).
  - OECD-DAC International Development Statistics (Organization for Economic Co-operation and Development, 2009).
  - World Development Indicators (WDI, 2009) and World Economic Outlook (WEO, 2009).
  - Polity IV project for institutional scores.
  - IMF Working Papers and World Bank Policy Research Working Papers for related empirical methods and findings.

### Data definitions and variable construction (Table 1)
- Dyadic variables:
  - Real bilateral aid (dyadic): Bilateral aid flows, 2007 US$, source: OECD-DAC.
  - Real bilateral trade (dyadic): Real trade between donor and recipient, source: Qureshi and Tsangarides (2010), originally from IMF Direction of Trade statistics.
  - Common language: dummy for shared language (Qureshi and Tsangarides, 2010).
  - Former colony: dummy for donor-recipient former colonial relationship (Qureshi and Tsangarides, 2010).
  - UNGA Voting variable: fraction of times a recipient votes with the donor (Dreher and Sturm, 2006).
- Donor- and recipient-level variables:
  - Real GDP (per capita), Population, and Recipient final consumption: PWT Mark 6.3 (Heston et al., 2009).
  - Donor output gap & trend: obtained through log-linear regression on trend (authors' calculations); OECD methodology reference: Beffy et al. (2006).
  - Recipient output gap & trend: obtained through HP filter with λ=1 (Rand and Tarp, 2002).
  - Recipient terms-of-trade shocks: TOT growth rates in the bottom decile of the recipient-specific distribution; TOT variable smoothed using two-year moving average (authors' calculations).
  - Recipient growth collapse: episodes as defined in Hausmann, Rodriguez and Wagner (2008) that lasted at least three years.
  - Recipient Polity IV score: scale from -10 to +10 (Polity IV project).
  - IMF program dummy: at least five months of active IMF program in any given year (Dreher, 2006, updated February, 2010).

### Descriptive statistics and sample composition (Tables and Figures)
- Sample coverage:
  - Donors: N=22.
  - Recipients: N=111 (list of countries provided across low income and middle income categories).
- Selected summary statistics (Table 3; exact values preserved):
  - Donor Output gap (% PGDP): # obs. 89,496; Mean 0.1; St. Dev. 5.3; Min -19.1; Max 21.8.
  - Donor Output gap (OECD) (% PGDP): # obs. 78,987; Mean -0.4; St. Dev. 2.3; Min -9.2; Max 6.7.
  - 1=Above-trend growth (donor): # obs. 87,010; Mean 0.5; St. Dev. 0.5; Min 0.0; Max 1.0.
  - Public debt (% GDP) (donor): # obs. 74,354; Mean 54.7; St. Dev. 29.5; Min 4.1; Max 191.6.
  - Recipient Output gap* (% PGDP): # obs. 79,662; Mean 0.7; St. Dev. 12.0; Min -40.9; Max 19.2. (Note: minimum output gap for recipients is -40.9, Rwanda, 1994.)
  - Log-real aid (net flows): # obs. 89,496; Mean 7.6; St. Dev. 8.3; Min -20.8; Max 21.8.
  - 1=IMF program (recipient): # obs. 87,010; Mean 0.3; St. Dev. 0.5; Min 0.0; Max 1.0.

- Appendix Figures (exact captions preserved):
  - Figure 1. Bilateral ODA, 1970–2008
    - Note: Bilateral Net ODA* excludes humanitarian aid, development food aid, and debt relief grants. All figures in 2008 US$ billion for 22 OECD donors.
  - Figure 2. Geographical composition of bilateral net ODA, 1970–2007
    - Note: Bilateral Net ODA excludes humanitarian aid, development food aid, and debt relief grants. All figures in 2007 US$ billion.
  - Figure 3. Distribution of output gap estimates (Panel 1: Donors (N=22); Panel 2: Recipients (N=111)). Extreme output gap observations—the top and bottom ten percent—have been dropped.
  - Figure 4. Correlation coefficients between the business cycle and aid (Donors N=22; Recipients N=111). Notes: donor-specific contemporaneous correlation coefficients between the aid-to-GDP ratio and the donor cycle variable; histograms for recipient-level correlations; extreme output gap observations (below the 10th percentile and above the 90th percentile) have been dropped.
  - Figure 5. Yearly probability of a large shock in aid-receiving countries, 1970–2004 (Panel 1. Growth collapses; Panel 2. Terms-of-trade shocks). See Table 1 and Section III.A for definitions.
  - Figure 6. Conditional empirical relationship between the donor output gap and aid flows (Panel 1. LICs; Panel 2. MICs). Notes: coefficient curves with 95 percent confidence intervals based on estimates available upon request.

### Core empirical findings (selected coefficient highlights from tables)
- Donor cycle effects on aid (Table 4 / Table 5 / Table 8 — baseline and dynamic):
  - Donor output gap: positive and statistically significant. Example coefficients reported:
    - Donor output gap 0.11*** (standard error 0.01) in multiple specifications.
    - Donor output gap (OECD) 0.08*** (standard error 0.02).
    - 1=Donor above-trend growth 0.21*** (standard error 0.05).
  - Dynamic large-shock effects (Table 8):
    - Large shock 1/ -0.464*** (0.08); one-year lag -0.307*** (0.06); two-year lag -0.462*** (0.07).
    - Large shock 2/ -0.305*** (0.07); one-year lag -0.113* (0.06); two-year lag -0.310*** (0.07).
    - Large shock 3/ coefficients reported as generally smaller and less often significant.
- Recipient cycle effects on aid (Table 6 / Table 7 / Table 9):
  - Recipient output gap: mixed; selected coefficients:
    - Recipient output gap: 0.02** (0.01) in some specifications.
    - 1=Rec. below-trend GDP growth: -0.15*** (0.05) in a specification; other specifications report negative and significant coefficients for below-trend consumption and GDP growth indicators.
  - Recipient large shocks:
    - 1=Recipient Large terms of trade shock 0.16* (0.09) in one column; 0.28*** (0.10) in LICs; -0.05 in MICs.
    - 1=Recipient Growth collapse 0.43*** (0.12) in Full sample; 0.15 (0.15) in LICs; 0.65*** (0.20) in MICs.
  - Dynamic recipient shocks (Table 9):
    - 1=Large shock 1/ contemporaneous 0.13 (0.08) to 0.25** (0.10) depending on subsample; two-year lag 0.38*** (0.08) in full sample.
- Simultaneous shocks and interactions (Table 10):
  - Donor output gap (log-linear): 0.113*** (0.01).
  - Recipient output gap: 0.022** (0.01) in some columns.
  - 1=Large donor shock: -0.895*** (0.11) in full sample specification.
  - 1=Large recipient shock: 0.430*** (0.13) in full sample specification.
  - Interaction 1=Large donor x recipient shock coefficients reported small and not consistently significant (e.g., 0.010 (0.14)).
- Donor public debt and aid (Table 11):
  - Public debt (% GDP): negative coefficient -0.012** (0.01) to -0.017*** (0.00) across specifications.
  - Public debt x Large shock interaction: negative and sometimes significant (e.g., -0.005*** (0.00); -0.006*** (0.00); -0.010*** (0.00)).
- Institutions and shock responses (Table 12):
  - Recipient Polity IV score: positive coefficients in several specifications (e.g., 0.42*** (0.13); 0.31*** (0.12)).
  - 1=Large shock x Polity IV score: positive and often significant (e.g., 0.04*** (0.01); 0.03** (0.01); 0.07*** (0.02)), indicating that higher Polity IV scores amplify aid responsiveness to large recipient shocks in some specifications.
- Augmented specifications and robustness (Table 13 and Tables 14–17):
  - Core donor cycle coefficient remains positive and significant in augmented models:
    - Output gap (donor) 0.12*** (0.01) across baseline augmented columns.
  - Additional determinants with significant coefficients in augmented models (Table 13):
    - Donor government revenue 0.14*** (0.02).
    - 1=IMF program 1.07*** (0.10) in some specifications.
    - Log-Real bilateral trade 0.36*** (0.05).
    - 1=Common language 2.11*** (0.36); 1=Former colony 1.44** (0.61).
  - Robustness checks (Tables 14–17) report consistency of donor cycle positive effects across:
    - Gross aid flows (Table 14).
    - Tobit with Random Effects (Table 15).
    - Accounting for aid persistence (Table 16): Log-Real aid, one year lag 0.52*** (0.01).
    - Post-Cold War regime interactions (Table 17): donor cycle x (1=Post-Cold War) coefficients vary, with some negative and significant interactions (e.g., -0.37** (0.16)).

### Key empirical magnitudes and statistical notes
- Regression framework:
  - Dependent variable: semi-log transformed real aid flows (see Section III.A).
  - Estimation: country-pair and time fixed effects in most specifications; standard errors clustered at the country-pair level.
- Statistical significance notation used across tables:
  - *** p<0.01, ** p<0.05, * p<0.1.
- Within R-squared ranges reported across tables typically between 0.04 and 0.33 depending on specification and sample.

*Italic: Source — _wp10240 - REFERENCES (appendix, tables, and references as provided).*

---


_Source: https://www.imf.org/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2010/_wp10240.pdf_