## _wp05102

## Source details

**Canonical URL:** [_wp05102](https://www.imf.org/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2005/_wp05102.pdf)

## Other formats

- [Markdown version](/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2005/_wp05102.pdf.md)
- [Structured JSON version](/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2005/_wp05102.pdf.json)

---

### Introduction and context
- Indirect trade through entrepôts is a common phenomenon in world commerce.
- Illustrative statistic: "for every $100 worth of goods that the United States exports to mainland China, approximately $23 goes through Hong Kong SAR."
- Indirect trade share increased from "5 percent in the mid-1980s to 17 percent today (Andriamananjara, Arce, and Ferrantino, 2004)."
- More than thirty countries participate in significant indirect trade; examples listed include Macao SAR, Cyprus, Fiji, Senegal, Jordan, Armenia, Seychelles, Honduras, Benin, Montserrat, St. Lucia, and Singapore.
- Hong Kong SAR identified as the world's largest entrepôt: trade was "259 percent of GDP in 1998."

### Hypothesis: entrepôts facilitate tariff evasion
- Proposed mechanism: entrepôt economies can be used to facilitate tariff evasion via specialized agents who transport goods without paying required tariffs.
- Rationale: the benefit of indirect trade for evading tariffs increases with tariff rates.
- Contextual data:
  - Chinese average statutory tariff rate declined from "23.6 percent in 1996 to 15.8 percent by 2001."
  - Average fraction of the rest of the world’s exports to China that goes through Hong Kong SAR declined from "26 percent in 1996 to 20 percent in 2001."

### Empirical framework and methodology
- Theoretical model:
  - Each trader exports a fixed amount, V.
  - Tariff rate for industry i is τi.
  - Benefit of evasion: Benefitik = τi V.
  - Cost of evasion: Costik = C + γ τi V + ηik, where C is fixed cost, γ < 1 is variable cost parameter, and ηik is a random component.
  - A representative trader evades iff ηik ≤ (1-γ) τi V - C.
  - Aggregate implication: Indirect Export Rate_i = F( (1-γ) τi V - C ) (Equation 1).
  - If F is uniform, linear approximation: Indirect Export Rate_i = α + β*τi + εi, with β > 0 (Equation 2).
- Identification strategy:
  - Unit: product-by-product indirect trade intensity = ratio of indirect exports to China via Hong Kong SAR to total exports to China.
  - Data disaggregated at "Harmonized Commodity Description and Coding System (HS) 6-digit level" for "1996–2001."
  - Controls: 6-digit HS fixed effects; differenced specifications to control for time-invariant product characteristics.
  - Placebo/contrast: compare Hong Kong SAR with Singapore.

### Data
- Three core datasets:
  - (a) Chinese tariffs (WITS → UNCTAD TRAINS), tariff rates at 8-digit HS aggregated to 6-digit HS where 8-digit variation is small; tariff_it is tariff rate on incoming goods in industry i in year t.
  - (b) Direct exports to China at 6-digit HS (WITS → United Nations' Comtrade); sample restricted to countries with full-period data; Africa and the Middle East omitted; final set of 29 countries (Appendix Table 1).
  - (c) Indirect exports to China via Hong Kong SAR at 6-digit HS (Smartal Solutions) for 1996–2001; Indirect_export_rate_it = Σ Indirect_exports_ict / Σ(Indirect_exports_ict + Direct_Exports_ict) (sums over c).
- Time coverage and handling:
  - Earliest detailed year: 1996; tariff data reflect year-end rates.
  - 1997 tariff measure uses weighted average of year-end 1996 and 1997 tariffs because changes occurred on October 1, 1997; 1998–2001 tariff changes implemented on January 1 and are uniform throughout those years.
- Summary statistics (Table 1):
  - Average indirect exports: 22 percent for full sample.
  - Average tariff rate: 18 percent for full sample; declined from approximately 23 percent in 1996 to 15 percent in 2001.
  - Panel years: 1996–2001; total observations reported as 27,577 for the Hong Kong SAR sample.

### Main empirical results
- Basic specification: Indirect_export_rate_it = α + β*Tariff_it + δt + εit (Equation 3).
- Benchmark estimates (Table 2):
  - Column (1): year fixed effects only: Tariff coefficient ≈ 0.25 (standard error (0.027)).
  - Column (2): add 3-digit HS fixed effects: Tariff coefficient = 0.286*** (standard error (0.044)).
    - Interpretation: one percentage point increase in tariff rate → 0.29 percentage point increase in indirect export rate (benchmark reported as 0.29).
    - Economic significance: increase in tariff from 0 to 19 percent (mean tariff in sample) → 19*0.29 = 5.5 percent increase in indirect export rate; average indirect export rate = 0.23.
  - Column (3): include 6-digit industry fixed effects (≈ 3600 fixed effects): Tariff coefficient = 0.113*** (standard error (0.040)).
  - Column (4): differenced specification (2001–1996): ∆Tariff coefficient = 0.169*** (standard error (0.047)).
  - Column (5): include Tariff^2: Tariff = 0.705*** (std err (0.100)); Tariff^2 = -0.616*** (std err (0.134)) — implies diminishing marginal effect of tariff increases on indirect trade.
- Aggregation robustness (Table 3):
  - Aggregate by tax-bracket (53 distinct tax rates; 313 observations per specification).
  - Tariff coefficients (mean/median, weighted/unweighted) range from 0.262*** to 0.291***; adjusted R-squared increases to 0.26–0.29 for mean specifications.
- Singapore comparison and controls (Table 4):
  - Using Singapore indirect exports (1999–2001) as alternative entrepôt:
    - Singapore: Tariff coefficients near 0 (0.003 (0.011) and -0.008 (0.028)).
    - Hong Kong SAR (same sample years/industries): Tariff coefficients ~0.276*** and 0.268***.
  - Including Indirect_export_rate_SGP as control in Hong Kong regressions:
    - Indirect_export_rate_SGP coefficient positive and significant (0.259*** or 0.226***), indicating correlation across entrepôts.
    - Inclusion does not alter Tariff effect on Hong Kong.
- Tariff exemptions (Table 5):
  - Exemption intensity from Chinese Customs Statistics 1998 (6-digit): ratio of tariff-free imports to total imports.
  - Subsample estimates:
    - Exemption < 5th percentile (0.06): Tariff = 0.384* (std err 0.198), Observations 1,262, R-squared 0.48.
    - Exemption < 10th percentile (0.16): Tariff = 0.440*** (std err 0.147), Observations 2,526, R-squared 0.38.
    - Exemption > 90th percentile (0.996): Tariff = -0.174 (std err 0.220), Observations 2,526, R-squared 0.36.
    - Exemption > 95th percentile (0.999): Tariff = -0.279 (std err 0.296), Observations 1,262, R-squared 0.42.
  - Interpretation: for goods with low exemption intensity, Tariff effect is larger; for goods with very high exemption intensity, no positive tariff–indirect trade relation.
- Product differentiation (Table 6):
  - Classification by Rauch (1999) (matched via Feenstra (1996) concordance); clustering at 4-digit HS.
  - Tariff coefficients:
    - Homogeneous products: Tariff = 0.173** (std err 0.084), Observations 6,375, R-squared 0.21.
    - Differentiated products: Tariff = 0.280*** (std err 0.096), Observations 12,605, R-squared 0.19.
    - All products: Tariff = 0.182*** (std err 0.064); Differentiated*Tariff interaction = 0.087 (std err 0.073) not statistically significant.
  - Interpretation: positive tariff–indirect trade correlation holds for both differentiated and non-differentiated products; point estimate somewhat smaller for homogeneous products.
- Statistical practice:
  - Robust standard errors clustered at 6-digit HS level for most regressions; clustering at 4-digit HS for Rauch split.
  - Significance notation: * significant at 10%; ** significant at 5%; *** significant at 1%.

### Quantitative illustration of economic significance
- Reported specification: a "10 percent increase in tariff rate would lead to an increase in the indirect trade rate of 2.5 percentage points."
- Increase in tariff rate "from 0 to 16 percent (the average statutory tariff rate in China in 2001) would lead to an indirect export rate of about 4 percent."
- Interpretation offered by authors: about "1/5 of the indirect exports through Hong Kong SAR may be accounted for by the evasion motivation."

### Alternative explanations and robustness
- High correlation between types of goods shipped via Hong Kong SAR and Singapore indicates common transshipment patterns; however, only Hong Kong shows a positive relation between tariffs and indirect trade.
- No tariff–indirect trade correlation for goods where the vast majority of incoming goods are tariff-exempt.
- Positive tariff–indirect trade correlation robust to:
  - year and industry fixed effects (3-digit and 6-digit),
  - long-difference estimation,
  - aggregation by tariff brackets,
  - inclusion of Singapore indirect export controls,
  - sample splits by exemption intensity and product differentiation.
- Authors rely on the accuracy of Hong Kong SAR customs statistics and do not allege lack of integrity of those statistics or customs.

### Conclusion and contributions
- Main finding: Tariff evasion is an important motivation for observed indirect trade via Hong Kong SAR; indirect trade intensity is positively and significantly related to product-level tariff rates in levels and differences.
- Contributions:
  - Conceptual: highlights a role for middlemen in facilitating tariff evasion via entrepôts.
  - Methodological: approach can be applied to other contexts and countries to compare susceptibility to tariff evasion through indirect trade.
- Caveat: result depends on Chinese border corruptibility being sufficient for Hong Kong SAR to serve as an intermediate step; the paper does not impugn Hong Kong SAR customs’ integrity.

### Appendix — Countries (annual observations)
- Argentina — Annual: 356
- Australia — Annual: 1,250
- Austria — Annual: 1,789
- Canada — Annual: 1,089
- Czech Republic — Annual: 645
- Denmark — Annual: 797
- Finland — Annual: 961
- France — Annual: 2,209
- Germany — Annual: 2,890
- Greece — Annual: 204
- Hungary — Annual: 290
- Indonesia — Annual: 1,292
- Ireland — Annual: 448
- Italy — Annual: 2,418
- Japan — Annual: 3,649
- Korea — Annual: 3,363
- Mexico — Annual: 257
- Netherlands — Annual: 1,453
- New Zealand — Annual: 426
- Norway — Annual: 564
- Poland — Annual: 107
- Portugal — Annual: 335
- Slovenia — Annual: 135
- Spain — Annual: 1,279
- Sweden — Annual: 1,390
- Switzerland — Annual: 1,791
- Turkey — Annual: 467
- United Kingdom — Annual: 2,246
- United States — Annual: 3,569

*Source: _wp05102 (PDF) — Sections II–V as provided.*

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

### _wp05102 - References

### Introduction and context
- Indirect trade through entrepôts is a common phenomenon in world commerce.
- Illustrative statistic: "for every $100 worth of goods that the United States exports to mainland China, approximately $23 goes through Hong Kong SAR."
- Indirect trade has grown faster than world trade: its share increased from "5 percent in the mid-1980s to 17 percent today (Andriamananjara, Arce, and Ferrantino, 2004)."
- There are "more than thirty countries" involved in significant indirect trade; examples include Macao SAR, Cyprus, Fiji, Senegal, Jordan, Armenia, Seychelles, Honduras, Benin, Montserrat, St. Lucia, and Singapore.
- Hong Kong SAR is the world's largest entrepôt: trade was "259 percent of GDP in 1998."

### Hypothesis: entrepôts facilitate tariff evasion
- New proposed explanation: entrepôt economies can be used to facilitate tariff evasion via specialized agents who transport goods without paying required tariffs.
- Rationale: the benefit of indirect trade for evading tariffs increases with tariff rates.
- Contextual data:
  - Chinese average statutory tariff rate declined from "23.6 percent in 1996 to 15.8 percent by 2001."
  - The average fraction of the rest of the world’s exports to China that goes through Hong Kong SAR declined from "26 percent in 1996 to 20 percent in 2001."

### Methodology
- Unit of analysis: product-by-product indirect trade intensity, defined as the ratio of indirect exports to China via Hong Kong SAR to total exports to China.
- Data: disaggregated at "Harmonized Commodity Description and Coding System (HS) 6-digit level" for the years "1996–2001."
- Identification strategy:
  - Examine whether indirect trade intensity systematically relates to product-level tariff rates.
  - Add "6-digit HS fixed effects" and also difference the data to control for time-invariant product characteristics.
  - Compare Hong Kong SAR with Singapore as a placebo/contrast.

### Key empirical findings
- A clear positive association exists between tariff rate and intensity of indirect trade via Hong Kong SAR.
- Results remain statistically significant at the "1 percent level" after adding HS fixed effects and differencing, though point estimates are somewhat reduced.
- Comparison with Singapore:
  - High correlation between types of goods shipped via Hong Kong SAR and Singapore, indicating commonalities in goods requiring transshipment.
  - No correlation found between tariff rates and indirect trade via Singapore.
- Sample-split robustness tests:
  - For industry classes where the vast majority of incoming goods are tariff-exempt, no tariff–indirect trade correlation is found.
  - For homogeneous and differentiated products (based on Rauch (1999) classification), a positive correlation between tariff rate and indirect trade intensity is found for both categories.

### Quantitative illustration of economic significance
- One reported specification: a "10 percent increase in tariff rate would lead to an increase in the indirect trade rate of 2.5 percentage points."
- An increase in tariff rate "from 0 to 16 percent (the average statutory tariff rate in China in 2001) would lead to an indirect export rate of about 4 percent."
- Interpretation: this suggests that about "1/5 of the indirect exports through Hong Kong SAR may be accounted for by the evasion motivation."

### Relation to literature and scope
- Distinction from Fisman and Wei (2004):
  - Fisman-Wei estimate the elasticity of tax evasion with respect to tax rates (a public finance question).
  - The current paper seeks to establish outsourcing of tariff evasion as an important explanation for entrepôt trade.
  - The Fisman-Wei result is necessary but not sufficient for the current paper’s result: Chinese border corruptibility is necessary for Hong Kong SAR to serve as an intermediate step to evade tariffs.
- The paper relies on the accuracy of Hong Kong SAR customs statistics and does not allege lack of integrity of those statistics or customs.

### Additional notes on data, figures, and tables (as presented in the source)
- Time span of analysis: "1996–2001."
- Tests and displays mentioned in the source:
  - Tables include: "Indirect Export Rates and Tariff Rates, 1996–2001;" "Effect of Tariff Rate on Hong Kong Indirect Export Rate;" "Aggregating the Indirect Export Rates by Tax Brackets;" "Effect of Tariff Rate on Singapore and Hong Kong’s Indirect Export Rates;" "Tariff-Exempted Versus Non-Exempted Industries;" "Homogeneous Versus Differentiated Products."
  - Figures include correlations between tariffs and Hong Kong’s indirect export rates and changes therein.
  - Appendix includes "List of Countries."

*Source: _wp05102 - References*

### Section III describes the datasets brought together for this research. Section IV presents our

### _wp05102 - Section III describes the datasets brought together for this research. Section IV presents our

### Empirical framework
- Model setup:
  - Each trader exports a fixed amount, V.
  - Tariff rate for industry i is τi.
  - Benefit of evasion for trader k in industry i: Benefitik = τi V.
  - Cost of evasion: Costik = C + γ τi V + ηik, where:
    - C is the fixed cost;
    - γ < 1 describes the variable cost;
    - ηik is a random component, realized before the trade decision.
  - A representative trader evades if and only if ηik ≤ (1-γ) τi V - C.
- Aggregate implication:
  - If ηik i.i.d. with cumulative distribution function F, the fraction of exports in industry i re-routed through Hong Kong SAR is:
    - Indirect Export Rate_i = F( (1-γ) τi V - C )  (Equation 1)
  - If F is uniform, this implies a linear relation:
    - Indirect Export Rate_i = α + β*τi + εi, where β > 0  (Equation 2)
- Discussion:
  - If ηik is non-uniform or cost is non-linear in τi, the indirect trade rate may be a non-linear function of the tariff rate.

### Data (Section III)
- Three crucial datasets:
  - (a) Chinese tariffs,
  - (b) direct exports to China at a product level,
  - (c) indirect exports to China via Hong Kong SAR at a product level.
- Tariff data:
  - Source: World Bank's WITS database, derived from UNCTAD TRAINS; tariff rates at the 8-digit HS level.
  - Aggregation: Rates aggregated to 6-digit HS level where 8-digit variation within a 6-digit category is relatively small; sample restricted to uniform 8-digit rates within 6-digit categories.
  - Time coverage: earliest detailed year is 1996; data reflect year-end tariff rates.
  - Handling mid-year changes: 1997 tariff measure uses a weighted average of year-end 1996 and 1997 tariffs because changes occurred on October 1, 1997; 1998–2001 tariff changes implemented on January 1 and are uniform throughout those years.
  - Definition: Tariff_it is the tariff rate on incoming goods in industry i in year t.
- Direct export data:
  - Source: WITS → United Nations' Comtrade; exports recorded at 6-digit HS level.
  - Sample: countries with export data available for the entire period; Africa and the Middle East omitted due to very low export rates; final set of 29 countries (listed in Appendix Table 1).
  - Definition: Direct_exports_ict is value in US dollars of direct exports in industry i from country c to China in year t.
- Indirect export data:
  - Source: Smartal Solutions (official provider of Hong Kong SAR export statistics); indirect exports to China by country of origin, at the 6-digit HS level for 1996–2001.
  - Aggregate indirect export rate by industry-year:
    - Indirect_export_rate_it = Σ Indirect_exports_ict / Σ(Indirect_exports_ict + Direct_Exports_ict)  (sums over c)
  - Rationale: aggregate to avoid clustering complications across two group types and robustness to aggregation-level regressions.
- Summary statistics (Table 1):
  - Average indirect exports on average: 22 percent for full sample (noted in text).
  - Average tariff rate: 18 percent for full sample; declined from approximately 23 percent in 1996 to 15 percent in 2001.
  - Panel: years 1996–2001; observations reported as counts in Table 1 (e.g., total observations 27,577 for Hong Kong SAR sample).

### Results (Section IV)
- Basic specification:
  - Model: Indirect_export_rate_it = α + β*Tariff_it + δt + εit  (Equation 3), with year fixed effect δt.
- Benchmark estimates (Table 2):
  - Column (1): basic specification (year fixed effects only): Tariff coefficient ≈ 0.25 (standard error (0.027)).
  - Column (2): add industry-year fixed effects (3-digit HS): Tariff coefficient = 0.286*** (standard error (0.044)).
    - Interpretation: a one percentage point increase in the tariff rate leads to a 0.29 percentage point increase in the indirect export rate (benchmark estimate reported as 0.29).
    - Economic significance: increase in tariff from 0 to 19 percent (the mean tariff in the entire sample) leads to 19*0.29 = 5.5 percent increase in indirect export rate; average indirect export rate in sample is 0.23.
  - Column (3): include 6-digit industry fixed effects (≈ 3600 fixed effects): Tariff coefficient = 0.113*** (standard error (0.040)).
  - Column (4): differenced specification, (Indirect_export_rate_i2001 – Indirect_export_rate_i1996) = α + β*(Tariff_i2001 – Tariff_i1996) + δt + εit: ∆Tariff coefficient = 0.169*** (standard error (0.047)).
  - Column (5): include quadratic term Tariff^2: Tariff coefficient reported as 0.705*** (std err (0.100)) and Tariff^2 = -0.616*** (std err (0.134)), implying diminishing marginal effect of tariff increases on indirect trade.
- Aggregation robustness (Table 3):
  - Aggregate outcome by tax-bracket mean/median over 53 distinct tax rates (313 observations per specification across years).
  - Tariff coefficients (mean/median, weighted/unweighted) range from 0.262*** to 0.291***; adjusted R-squared increases to 0.26–0.29 for mean specifications.
- Alternative explanations and robustness checks:
  - Singapore comparison (Table 4):
    - Use Singapore indirect exports (1999–2001) as alternative entrepôt; specification: Indirect_export_rate_Country_it = α + β*Tariff_it + δHS3,t + εit (Equation 5), Country ∈ {SGP, HK}.
    - Results: for Singapore, no relation between tariff rates and fraction routed through Singapore (Tariff coefficients near 0: 0.003 (0.011) and -0.008 (0.028) in columns (1) and (2)).
    - For Hong Kong SAR using same sample years/industries, Tariff coefficients ~0.276*** and 0.268*** in columns (3) and (4).
    - Including Indirect_export_rate_SGP as control (Equation 6): Indirect_export_rate_HK_it = α + β1*Tariff_it + β2*Indirect_export_rate_SGP_it + δHS3,t + εit. Indirect_export_rate_SGP coefficient positive and significant (0.259*** or 0.226***), indicating correlation across entrepôts, but inclusion does not alter Tariff effect on Hong Kong.
  - Tariff exemptions (Table 5):
    - Use Exemption intensity from Chinese Customs Statistics 1998 (aggregated from 8-digit to 6-digit): Exemption = ratio of value of imports tariff-free to total imports.
    - Subsamples:
      - Exemption < 5th percentile (0.06): Tariff = 0.384* (std err 0.198), Observations 1,262, R-squared 0.48.
      - Exemption < 10th percentile (0.16): Tariff = 0.440*** (std err 0.147), Observations 2,526, R-squared 0.38.
      - Exemption > 90th percentile (0.996): Tariff = -0.174 (std err 0.220), Observations 2,526, R-squared 0.36.
      - Exemption > 95th percentile (0.999): Tariff = -0.279 (std err 0.296), Observations 1,262, R-squared 0.42.
    - Interpretation: for goods with low exemption intensity, Tariff effect is larger; for goods with very high exemption intensity, no positive relation between tariff and indirect export rate.
  - Product differentiation (Table 6):
    - Sample split by Rauch (1999) classification (matched via Feenstra (1996) concordance); cluster at 4-digit HS level.
    - Tariff coefficients:
      - Homogeneous products: Tariff = 0.173** (std err 0.084), Observations 6,375, R-squared 0.21.
      - Differentiated products: Tariff = 0.280*** (std err 0.096), Observations 12,605, R-squared 0.19.
      - All products: Tariff = 0.182*** (std err 0.064); interaction Differentiated*Tariff = 0.087 (std err 0.073) not statistically significant.
    - Interpretation: positive correlation holds for both differentiated and non-differentiated products; point estimate somewhat smaller for homogeneous products.
- Statistical notes:
  - Robust standard errors clustered at 6-digit HS level for most regressions; clustering at 4-digit HS for Rauch split.
  - Significance notation: * significant at 10%; ** significant at 5%; *** significant at 1%.

### Conclusion (Section V)
- Main finding:
  - Tariff evasion is an important motivation for observed indirect trade via Hong Kong SAR; indirect trade intensity (ratio of indirect trade to total trade) is positively and significantly related to product-level tariff rates in levels and differences.
- Supporting evidence:
  - Positive tariff–indirect trade correlation robust to:
    - year and industry fixed effects (3-digit and 6-digit),
    - long-difference estimation,
    - aggregation by tariff brackets,
    - inclusion of Singapore indirect export controls,
    - sample splits by exemption intensity and product differentiation.
  - For goods with widespread tariff exemptions, no correlation between tariff rate and indirect trade intensity.
- Contributions and extensions:
  - Conceptual: highlights potential darker role of middlemen in facilitating tariff evasion.
  - Methodological: approach applicable to other contexts and countries to compare susceptibility to tariff evasion via indirect trade.
- Caveat:
  - Paper relies on accuracy of Hong Kong SAR customs statistics and does not impugn Hong Kong SAR customs’ integrity.

*Source: _wp05102 (PDF) — Sections II–V as provided.*

### Appendix Table 1. List of Countries

### Appendix Table 1. List of Countries

### Country data (Annual; Observations)
- Argentina — Annual: 356
- Australia — Annual: 1,250
- Austria — Annual: 1,789
- Canada — Annual: 1,089
- Czech Republic — Annual: 645
- Denmark — Annual: 797
- Finland — Annual: 961
- France — Annual: 2,209
- Germany — Annual: 2,890
- Greece — Annual: 204
- Hungary — Annual: 290
- Indonesia — Annual: 1,292
- Ireland — Annual: 448
- Italy — Annual: 2,418
- Japan — Annual: 3,649
- Korea — Annual: 3,363
- Mexico — Annual: 257
- Netherlands — Annual: 1,453
- New Zealand — Annual: 426
- Norway — Annual: 564
- Poland — Annual: 107
- Portugal — Annual: 335
- Slovenia — Annual: 135
- Spain — Annual: 1,279
- Sweden — Annual: 1,390
- Switzerland — Annual: 1,791
- Turkey — Annual: 467
- United Kingdom — Annual: 2,246
- United States — Annual: 3,569

*Source: Authors’ calculations.*

---


_Source: https://www.imf.org/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2005/_wp05102.pdf_