## 1. Full Sample, 1975–20001/

## Source details

**Canonical URL:** [1. Full Sample, 1975–20001/](https://www.imf.org/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2005/_wp05112.pdf)

## Other formats

- [Markdown version](/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2005/_wp05112.pdf.md)
- [Structured JSON version](/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2005/_wp05112.pdf.json)

---

### Introduction and research question
- Many developing and emerging market economies continue to rely heavily on trade taxes: in Sub-Saharan Africa, trade taxes account for an average of about one-quarter of all government revenues; in the developing countries of Asia and the Pacific they account for around 15 percent.
- Empirical question: for each $1 of trade tax revenue that countries lose, how many dollars have they, in fact, actually recovered from other sources?
- Focus: whether countries have been able, in practice, to make good from domestic sources any loss of revenue from trade taxes experienced during liberalization episodes.
- Note: trade liberalization does not always reduce trade tax revenues (early reforms can increase revenues); core concern is when further liberalization reduces trade tax revenues and whether such losses can be recouped elsewhere.

### Literature context and theoretical considerations
- Early trade reform steps (reducing prohibitively high tariffs, tariffication of quotas, elimination of exemptions, raising low tariffs toward uniformity) may plausibly increase trade tax revenues.
- Keen and Ligthart (2001): increasing domestic consumption taxes (slightly less than) one-for-one with tariff cuts can leave consumers better off, preserve production efficiency gains, and increase government revenue.
- Caveats and extensions: Keen and Ligthart (2001, 2003); Hatzipanayotou, Michael, and Miller (1994); Michael, Hatzipanayotou, and Miller (1993).
- Practical challenges: administrative ease of collecting customs duties relative to inland tax collection; much VAT revenue is collected at the border (often half or more, in many developing countries), while inland VAT implementation relies on self-assessment.

### Empirical gap and contribution
- Virtually no formal econometric assessment existed on the extent to which countries have offset reductions in trade tax revenues by increasing revenues from other sources.
- Prior work focused on effects of trade liberalization on trade tax revenues or simple tabulations of replacement, but not on conditioned econometric estimates of replacement rates.
- Simple correlations may give pessimistic conclusions because they fail to condition on variables associated with revenue performance.

### Data and empirical strategy
- Dataset: 125 countries, period 1975–2000 (unbalanced panel).
- Primary sources: Government Finance Statistics (GFS) and IMF Article IV country documents (Article IV used when conflicting with GFS).
- Revenue definitions:
  - Trade tax revenue = import and export duties (GFS or IMF documents).
  - Domestic tax revenue = total tax revenue excluding trade tax revenue (GFS or IMF documents).
- Regressors and definitions:
  - VAT dummy: 1 if a VAT in place in year t, 0 otherwise (from Ebrill and others (2001)).
  - Log per capita GDP: constant (1995) U.S. dollars (World Bank WDI).
  - Openness: imports plus exports in percent of GDP (IFS).
  - Log inflation: annual change in CPI (IFS).
  - Aid per capita: official development assistance (grants plus concessional loans) in U.S. dollars per capita (WDI).
  - Share of agriculture: share of agriculture in aggregate value added (WDI).
- Data limitations: GFS gaps, often central government only; reported “trade tax” can include border-collected VAT and sales taxes in some countries; efforts made to exclude such border-collected non-trade taxes and use general revenue data where possible.
- Empirical specification:
  - Dynamic panel baseline (all variables relative to GDP): includes lagged domestic revenue, trade tax revenue, interaction trade tax × VAT, VAT dummy, openness, VAT interactions with openness and log per capita GDP, log inflation, aid per capita, share of agriculture, country and time fixed effects.
  - Short-term recovery measure: 1 − β1 (β1 = coefficient on trade tax revenue).
  - Long-run recovery: θ ≡ (1 − β1)/(1 − β0) with lagged dependent variable coefficient β0; θV when VAT interacts with trade tax revenue.
  - Interpretation: θ = 1 (full recovery), θ = 0 (no recovery).
- Endogeneity/identification:
  - Potential endogeneity of trade tax revenue (policy/contemporaneous shocks; administration improvements).
  - IV: trade taxes instrumented by their own lag.
  - GMM (Arellano–Bond style): first-difference estimation with instruments (regressors lagged t-1 to t-3; trade tax lags t-2 to t-4).
  - Diagnostics reported: Hausman, m1 and m2 (serial correlation), Sargan test.

### Main results — Full sample
- Estimation approaches: Fixed Effects (FE), Random Effects (RE), IV, one-step GMM.
- Key numerical findings:
  - FE implied θ = 0.243** (about 24 cents recovered per dollar lost; average delay ~2.5 years).
  - IV implied θ = 0.172 (insignificant).
  - GMM (preferred) implied θ = 0.505** (about 50 cents recovered per dollar lost; average delay under one year).
- Overall interpretation: degree of revenue recovery significant but far from complete; plausible range about 20–50 cents per dollar, with GMM pointing to upper part of this range.
- VAT interaction:
  - FE: Trade tax × VAT = 0.105* (positive and significant at 10 percent) but implied θV = −.119 (negative, insignificant).
  - IV and GMM: mixed/insignificant VAT interaction effects; GMM shows VAT-related terms largely insignificant.

### Results by income group
- Low-income countries (preferred GMM):
  - GMM implied θ = 0.299** (about 30 cents recovered per dollar lost).
  - FE and IV show small or insignificant recovery.
  - VAT presence: no systematic evidence that low-income countries with a VAT recover more than those without (θV insignificant).
  - Conclusion: low-income countries on average have largely failed to recover trade tax revenue losses from domestic sources.
- Middle-income countries:
  - GMM implied θ = 0.654** (about 65 cents recovered per dollar lost).
  - FE implied θ = 0.450**; IV weaker.
  - Middle-income countries show stronger recovery than low-income countries — estimates suggesting recovery in the order of 45–65 cents per dollar; near full recovery in some specifications when identifying episodes of falling trade taxes.
  - Some evidence VAT helps in middle-income group.
- High-income countries:
  - GMM implied θ = 1.529* (greater than one; excess recovery).
  - FE implied θ = 1.833*; IV and RE estimates large but sometimes imprecise.
  - Interpretation: high-income countries replace lost trade tax revenues more than fully; trade taxes play a secondary/protective role and revenue replacement is not a major constraint.

### Further analysis (asymmetric responses, non-linearities, heterogeneity)
- Asymmetries (decomposed responses to increases and decreases in trade tax revenues):
  - Low-income:
    - Implied θ(minus) = 0.277 (insignificant); implied θ(plus) = −0.234.
    - VAT may help somewhat (implied θ(minus)V = 0.425) but not statistically robust.
  - Middle-income:
    - Implied θ(minus) = 0.956** (essentially full recovery for episodes of falling trade tax revenues).
    - Implied θ(plus) = 0.378.
    - Middle-income countries with a VAT: Implied θ(minus)V = 1.143**.
  - Interpretation: recovery is asymmetric — responses differ for decreases versus increases in trade tax revenue; problem most relevant when trade tax revenues fall.
- Robustness and non-linearities:
  - Adding lags and a squared term for trade tax to GDP:
    - Low-income: trade-tax-related regressors insignificant.
    - Middle-income: evidence of non-linearity — recovery becomes harder the lower the initial level of trade taxes.
      - At average middle-income trade tax level (~5 percent of GDP): around 36 cents recovered per dollar lost.
      - At half that level: around 18 cents recovered.
  - Time and regional splits:
    - Degree of recovery lower in second half of sample (1987 onwards) than in first half.
    - Regional signals: recovery more complete in the Americas and, perhaps surprisingly, Africa than in Asia and the Middle East.
- Country heterogeneity (country-specific time-series):
  - Low-income sample: six of 38 (Benin, Côte d’Ivoire, The Gambia, Malawi, Pakistan, Zambia) show implied θ positive and significant at 10 percent or better (average for this group close to unity).
  - Niger: one case with significantly negative θ.
  - Implication: some low-income success stories exist and may hold policy lessons.

### Conclusions and policy-relevant findings
- Core empirical finding (1975–2000): many developing countries did not fully recoup lost trade tax revenues from domestic sources.
- Key quantitative summary (preferred GMM unless otherwise stated):
  - Full sample: implied θ ≈ 0.505** (about 50 cents recovered per dollar lost).
  - Low-income: implied θ ≈ 0.299** (about 30 cents recovered per dollar lost).
  - Middle-income: implied θ ≈ 0.654** (about 65 cents recovered per dollar lost).
  - High-income: implied θ ≈ 1.529* (more than full recovery).
- Policy implications:
  - Further trade liberalization in many developing countries may be hindered unless alternative revenue sources can be found.
  - Standard prescription (replace tariff revenue with excises and a broad-based consumption tax like VAT) is theoretically sound but empirically mixed:
    - VAT adoption appears helpful in middle- and high-income countries in some specifications, but no systematic evidence that low-income countries with a VAT recover more than those without.
    - VAT design matters (examples cited: Uganda’s VAT consistent with orthodox design accompanied substantial offsetting revenue gains; Egypt’s VAT characterized by multiple rates and imperfect crediting coincided with revenue decline).
  - Successful recoveries often involved strengthening consumption taxes and income taxation and improvements in administration.
  - Indirect effects of trade liberalization (via growth and openness) are uncertain and not modeled here; countries should assess recovery prospects without relying on uncertain indirect gains.
- Research and policy priorities:
  - Case studies of low-income countries that succeeded in revenue recovery to identify actionable lessons.
  - Investigate VAT design features (single base, minimal exemptions, reasonable threshold) and complementary measures (income tax strengthening, administration) as determinants of successful revenue replacement.
- Cautionary note:
  - Results do not imply trade reforms lowering trade tax revenues were necessarily unwise; indirect benefits via growth or openness may offset direct revenue losses, but from a fiscal prudence perspective the revenue replacement challenge — particularly for low-income countries — is real.

*Source: _wp05112 - 1. Full Sample, 1975–20001/ (excerpt).*

### 1. Full Sample, 1975–20001/ ............................................................................................

### 1. Full Sample, 1975–20001/

### Introduction and research question
- Despite significant trade liberalization, many developing and emerging market economies continue to rely heavily on trade taxes as a source of government revenue: in Sub-Saharan Africa, trade taxes still account for an average of about one-quarter of all government revenues, and in the developing countries of Asia and the Pacific they account for around 15 percent.
- The paper asks the empirical question: for each $1 of trade tax revenue that countries lose, how many dollars have they, in fact, actually recovered from other sources?
- The focus is on whether countries have been able, in practice, to make good from domestic sources any loss of revenue from trade taxes experienced during liberalization episodes.
- The paper emphasizes that trade liberalization does not always reduce trade tax revenues (early reforms can increase revenues), but the core concern arises when further liberalization reduces trade tax revenues and whether such losses can be recouped elsewhere.

### Literature context and theoretical considerations
- Early steps of trade policy reform—reducing prohibitively high tariffs, tariffication of quotas, elimination of exemptions, and raising of low tariff rates toward uniformity—may plausibly lead to an increase in trade tax revenues (as discussed and documented by Ebrill and others (1999)).
- Keen and Ligthart (2001) show that for a small economy, increasing domestic consumption taxes (slightly less than) one-for-one with tariff cuts can leave consumers better off, preserve production efficiency gains, and increase government revenue.
- Caveats to the basic theoretical argument exist and are explored in Keen and Ligthart (2001, 2003); Hatzipanayotou, Michael, and Miller (1994) and Michael, Hatzipanayotou, and Miller (1993) consider infinitesimal shifts from trade to domestic commodity taxation.
- Practical challenges include the administrative ease of collecting customs duties relative to inland tax collection; much VAT revenue is collected at the border (often half or more, in many developing countries), while inland VAT implementation relies on self-assessment methods differing from customs procedures.

### Empirical gap and contribution
- There has been virtually no formal econometric assessment of the extent to which countries have successfully offset reductions in trade tax revenues by increasing revenues from other sources.
- Previous work (Ebrill and others (1999); Khattry and Rao (2002); Abgeyegbe, Stotsky, and WoldeMariam (2004)) focuses on how trade liberalization affects trade tax revenues or on tabulations of replacement, but not on conditioned econometric estimates of replacement rates.
- Simple correlations (as in Khattry and Rao (2002)) may give pessimistic conclusions because they fail to condition on variables associated with revenue performance; a country could maintain total revenues unchanged yet experience lasting effects relative to its income and other characteristics.

### Data and empirical strategy
- To address data paucity, the authors assembled a dataset comprising 125 countries, covering the period 1975–2000.
- The empirical strategy is to examine the relationship between total tax revenues and revenue from trade taxes alone, conditioning on other variables likely to be associated with revenue performance.
- The two revenue series would be independent of one another if and only if countries are able, and choose, to replace lost trade tax revenues from other sources.

### Plan of the paper (as stated)
- Section II describes the new dataset in more detail and provides a first look at experience in dealing with fiscal consequences of reduced trade tax revenues.
- A conceptual framework for estimation is developed following the data description.

*Source: _wp05112 - 1. Full Sample, 1975–20001/ (excerpt).*

### Section III. The basic (and troubling) empirical results are in Section IV, with the

### _wp05112 - Section III. The basic (and troubling) empirical results are in Section IV, with the

### Data
- Unbalanced panel of 125 countries for the period 1975–2000.
- Primary data sources: Government Finance Statistics (GFS) and IMF Article IV country documents (Article IV used when conflicting with GFS).
- Trade tax revenue defined as import and export duties (GFS or IMF documents).
- Domestic tax revenue defined as total tax revenue excluding trade tax revenue (GFS or IMF documents).
- VAT dummy: 1 if a VAT in place in year t, 0 otherwise (from Ebrill and others (2001)).
- Other regressors and definitions:
  - Log per capita GDP: constant (1995) U.S. dollars (World Bank WDI).
  - Openness: imports plus exports in percent of GDP (IFS).
  - Log inflation: annual change in CPI (IFS).
  - Aid per capita: official development assistance (grants plus concessional loans) in U.S. dollars per capita (WDI).
  - Share of agriculture: share of agriculture in aggregate value added (WDI).
- Data limitations noted:
  - GFS has gaps and often records only central government revenues.
  - Reported “trade tax” can include border-collected VAT and sales taxes in some countries; effort made to exclude such border-collected non-trade taxes and to use general (not just central government) revenue data where possible.

### Empirical specification and identification
- Baseline estimating equation (dynamic panel, domestic tax revenue as dependent variable, all variables expressed relative to GDP):
  - Equation (6): includes lagged domestic revenue, trade tax revenue, interaction trade tax × VAT, VAT dummy, openness, VAT interactions with openness and log per capita GDP, log inflation, aid per capita, share of agriculture, country and time fixed effects.
- Key parameters of interest:
  - Short-term recovery measure: 1 − β1 (where β1 is coefficient on trade tax revenue).
  - Long-run recovery: θ ≡ (1 − β1)/(1 − β0) with lagged dependent variable β0; and θV when VAT interacts with trade tax revenue (defined in equation (7)).
  - Interpretation: θ = 1 (full recovery), θ = 0 (no recovery).
- Econometric concerns and approaches:
  - Potential endogeneity of trade tax revenue (policy/contemporaneous shocks and administration improvements).
  - IV: trade taxes instrumented by their own lag.
  - GMM (Arellano–Bond style) used to address endogeneity and dynamic-panel bias (in first differences, instruments: regressors lagged t-1 to t-3; trade tax lags t-2 to t-4).
  - Diagnostics reported: Hausman test, m1 and m2 (serial correlation), Sargan test.

### Main results — Full sample
- Summary of estimation approaches reported in Table 1: Fixed Effects (FE), Random Effects (RE), IV, one-step GMM.
- Key findings:
  - FE estimate: implied θ = 0.243** (i.e., about 24 cents recovered per dollar lost; average delay ~2.5 years).
  - IV estimates: implied θ = 0.172 (insignificant).
  - GMM (preferred): implied θ = 0.505** (about 50 cents recovered per dollar lost, average delay under one year).
  - Overall interpretation: across full sample, degree of revenue recovery is significant but far from complete; plausible range about 20–50 cents per dollar, with GMM pointing to upper part of this range.
- VAT interaction:
  - FE: Trade tax × VAT positive and significant at 10 percent (0.105*), but implied θV is negative (−.119) and insignificant.
  - IV and GMM produce mixed/insignificant VAT interaction effects; GMM shows VAT-related terms largely insignificant.

### Results by income group (Tables 2–4)
- Low-income countries (Table 2):
  - Preferred GMM estimate: implied θ = 0.299** (about 30 cents recovered per dollar lost).
  - FE and IV estimates show small or insignificant recovery.
  - Presence of a VAT: no systematic evidence that low-income countries with a VAT recover more than those without (θV insignificant).
  - Conclusion: low-income countries on average have largely failed to recover trade tax revenue losses from domestic sources.
- Middle-income countries (Table 3):
  - GMM implied θ = 0.654** (about 65 cents recovered per dollar lost).
  - FE implies θ = 0.450**; IV weaker.
  - Middle-income countries show stronger recovery than low-income countries — estimates suggesting recovery in the order of 45–65 cents per dollar, and when separately identifying episodes of falling trade taxes, near full recovery in some specifications.
  - Some evidence VAT helps in middle-income group.
- High-income countries (Table 4):
  - GMM implied θ = 1.529* (greater than one; i.e., excess recovery).
  - FE implied θ = 1.833*; IV and RE estimates large but sometimes imprecise.
  - Interpretation: for high-income countries, lost trade tax revenues are replaced more than fully from domestic sources — trade taxes play a secondary/protective role and revenue replacement is not a major constraint.

### Further analysis (Section V)
- Asymmetries (Table 5):
  - Decomposed responses to increases (∆Trade tax plus) and decreases (∆Trade tax minus) in trade tax revenues.
  - Low-income countries:
    - Implied θ(minus) = 0.277 (insignificant); implied θ(plus) = −0.234.
    - Signs that VAT may help somewhat (implied θ(minus)V = 0.425), but not statistically robust.
  - Middle-income countries:
    - Implied θ(minus) = 0.956** (essentially full recovery for episodes of falling trade tax revenues).
    - Implied θ(plus) = 0.378.
    - For middle-income countries, those with a VAT recover noticeably more (Implied θ(minus)V = 1.143**).
  - Interpretation: revenue recovery is asymmetric — countries respond differently to decreases versus increases in trade tax revenue; the recovery problem is most relevant when trade tax revenues fall.
- Robustness and non-linearities (Table 6):
  - Including lags and a squared term for trade tax to GDP:
    - Low-income: trade-tax-related regressors now insignificant.
    - Middle-income: evidence of non-linearity — recovery becomes harder the lower the initial level of trade taxes.
      - At average middle-income trade tax level (~5 percent of GDP): around 36 cents recovered per dollar lost.
      - At half that level: around 18 cents recovered.
  - Time-splitting and regional checks (not fully tabulated in text):
    - Degree of recovery lower in second half of sample (1987 onwards) than in first half.
    - Regional signals: recovery more complete in the Americas and, perhaps surprisingly, Africa than in Asia and the Middle East.
- Country heterogeneity (country-specific time-series):
  - For low-income countries, only six out of 38 (Benin, Côte d’Ivoire, The Gambia, Malawi, Pakistan, Zambia) show implied θ positive and significant at 10 percent or better (average for this group close to unity).
  - Niger is one case with significantly negative θ.
  - Implication: some low-income success stories exist and may hold policy lessons.

### Conclusions and policy-relevant findings
- Core empirical finding: Over 1975–2000, many developing countries did not fully recoup lost trade tax revenues from domestic sources.
- Quantitative summary:
  - Full sample (preferred GMM): implied θ ≈ 0.505** (about 50 cents recovered per dollar lost).
  - Low-income (GMM): implied θ ≈ 0.299** (about 30 cents recovered per dollar lost).
  - Middle-income (GMM): implied θ ≈ 0.654** (about 65 cents recovered per dollar lost).
  - High-income (GMM): implied θ ≈ 1.529* (more than full recovery).
- Policy-relevant implications:
  - Further trade liberalization in many developing countries may be hindered unless alternative revenue sources can be found.
  - Standard prescription (replace tariff revenue with excises and a broad-based consumption tax like VAT) is theoretically sound but empirically mixed:
    - VAT adoption appears helpful in middle- and high-income countries in some specifications, but no systematic evidence that low-income countries with a VAT recover more than those without.
    - Quality and design of VAT likely matter (examples cited: Uganda’s VAT consistent with orthodox design accompanied substantial offsetting revenue gains; Egypt’s VAT characterized by multiple rates and imperfect crediting coincided with revenue decline).
  - Evidence that successful recoveries often involved strengthening not only consumption taxes but also income taxation.
  - Because indirect effects of trade liberalization (via growth and openness) are uncertain and not modeled here, countries should assess whether they can recover revenue without relying on these uncertain indirect gains.
- Heterogeneity and further research:
  - Diversity of country experiences suggests value in case studies of low-income countries that succeeded in revenue recovery.
  - Investigate VAT design features (single base, minimal exemptions, reasonable threshold) and complementary measures (income tax strengthening, administration) as potential determinants of successful revenue replacement.
- Caution:
  - Results do not imply trade reforms lowering trade tax revenues were necessarily unwise; indirect benefits via growth or openness may offset direct revenue losses, but from a fiscal prudence perspective the revenue replacement challenge — particularly for low-income countries — is real.

*Source: Authors’ calculations and empirical results as presented in the supplied IMF working paper content.*

### References

### _wp05112 - References

### References
- Agbeyegbe, T., J. Stotsky, and A. Woldemariam, 2004, “ Trade Liberalization, Exchange Rate Changes, and Tax Revenue in Sub-Saharan Africa” (unpublished, Washington: International Monetary Fund).
- Arrelano, M. and S. Bond, 1991, “Some Tests of Specification for Panel Data: Monte Carlo Evidence and an Application to Employment Equations,” Review of Economic Studies, Vol. 58, pp. 277–97.
- Ebrill, L., J. Stotsky, and R. Gropp, 1999, Revenue Implications of Trade Liberalization, IMF Occasional Paper No.180 (Washington: International Monetary Fund).
- Ebrill, L., M. Keen, J-P. Bodin, and V. Summers, 2001, The Modern VAT (Washington: International Monetary Fund).
- Gupta, S., B. Clements, A. Pivovarsky, and E. R. Tiongsin, 2004, “Foreign Aid and Revenue Response: Does the Composition of Aid Matter?” in Sanjeev Gupta, Benedict Clements and Gabriela Inchauste (eds.), Helping Countries Develop: The Role of Fiscal Policy (Washington: International Monetary Fund).
- Hatzipanayotou, P., M.S. Michael, and S. M. Miller, 1994, “Win-Win Indirect Tax Reform: A Modest Proposal,” Economics Letters, Vol. 44, pp. 147–51.
- Keen, M., and B. Lockwood, 2005, “The Value Added Tax—Consequences and Causes,” in preparation.
- Keen, M., and J. Ligthart, 2001, “Coordinating Tariff Reductions and Domestic Tax Reform,” Journal of International Economics, Vol. 56 (2001), pp. 407–425.
- _____, 2005, “Coordinating Tariff Reduction and Domestic Tax Reform Under Imperfect Competition,” Review of International Economics, Vol. 13, pp. 385–90.
- Keen, M., and A. Simone, 2004, “Tax policy in Developing Countries: Some Lessons from the 1990s, and Some Challenges Ahead,” in Sanjeev Gupta, Benedict Clements, and Gabriela Inchauste (eds.): Helping Countries Develop: The Role of the Fiscal Policy (Washington: International Monetary Fund).
- Khattry, B., and M. Rao, 2002, “Fiscal Faux Pas? An Analysis of the Revenue Implications of Trade Liberalization,” World Development, Vol. 30, pp.1431–44.
- Michael, M. S., P. Hatzipanayotou, and S. M. Miller, 1993, “Integrated Reforms of Tariffs and Consumption Taxes,” Journal of Public Economics, Vol. 52, 417–28.
- Rodrik, D., 1998, “Why Do More Open Economies have Bigger Governments?” Journal of Political Economy, Vol. 106, pp. 997–1032.
- Tanzi, V., 1987, “Quantitative Characteristics of the Tax Systems of Developing Countries,” in The Theory of Taxation for Developing Countries, ed. by D. Newbery and N. Stern (Oxford: Oxford University Press), pp. 205–41.
- Rajaraman, I., 2004, “ Fiscal Restructuring in the Context of Trade Reform,” National Institute of Public Finance and Policy, Delhi, Working Paper No. 7.
- Wacziarg, R., and K. H. Welch, 2003, “Trade Liberalization and Growth: New Evidence,” NBER Working Paper 10152.

*Source: _wp05112 - References*

---


_Source: https://www.imf.org/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2005/_wp05112.pdf_