## wpiea2019274-print-pdf

## Source details

**Canonical URL:** [wpiea2019274-print-pdf](https://www.imf.org/-/media/files/publications/wp/2019/wpiea2019274-print-pdf.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/wp/2019/wpiea2019274-print-pdf.pdf.md)
- [Structured JSON version](/-/media/files/publications/wp/2019/wpiea2019274-print-pdf.pdf.json)

---

### Data sources and coverage
- Primary data sources:
  - IMF’s CDIS: annual survey in 110-115 economies; most comprehensive source of information on FDI positions.
  - OECD FDI Statistics: decomposition into SPEs and non-SPEs for 30 economies; inward FDI by economy of the ultimate owner for 16 economies.
  - Orbis global firm database: unconsolidated accounting and ownership information for millions of corporate entities; used to infer ownership links and estimate FDI by ultimate investor economy where direct reporting is unavailable.
- Aggregate coverage statements:
  - Aggregate inward FDI across the 111 economies reporting to the CDIS was just above $30 trillion, of which around 95 percent was allocated to immediate investor economies.
  - The 16 economies reporting inward FDI in SPEs and non-SPEs separately account for almost 40 percent of the total inward FDI in CDIS.
  - The 12 economies further allocating inward Real FDI to ultimate investor economies account for around 20 percent.

### CDIS survey methodology and mirror data adjustments
- CDIS methodology:
  - Survey combines administrative data and survey information; local authorities aggregate to produce inward FDI by economy of the immediate investor and outward FDI by economy of the immediate investment.
  - Reported values are used without adjustment for the 110-115 economies reporting to CDIS (example: France reports inward FDI from Canada of around $3.8 billion in 2017).
- Mirror-data for non-reporting economies:
  - For the 130-135 economies not reporting to CDIS, mirror data are used (example: Cayman Islands inward FDI from Canada estimated from Canada’s reported outward FDI to Cayman Islands of around $34 billion).
  - Mirror-data undercoverage: simple summation of mirror estimates can substantially underestimate total inward FDI for non-reporting economies (example: sum of mirror estimates for Cayman Islands = $611 billion in 2017, likely an underestimate).
- Adjustment for mirror-data undercoverage:
  - Adjustment factor α_mirror_t defined as:
    - α_mirror_t = (∑_{h∈R} Inward FDI_reported_{hi,t}) / (∑_{i∈R} Outward FDI_reported_{ih,t})
  - The adjustment factor in 2017 is 1.17 (aggregate reported inward FDI is 17 percent higher than the sum of mirror positions).
  - Application example: Cayman Islands adjusted inward FDI = $712 billion ($611 billion * 1.17); adjustment amount $101 billion ($611 billion * 0.17) allocated to category “Not specified.”
  - Adjustment factors for the sample period are in the range 1.15-1.25 in every year.
  - Aggregate adjustments account for around 1.5 percent of global FDI because largest FDI recipients tend to report to CDIS and adjustments apply only to non-reporting economies.

### SPE definition, terminology, and conservative classification
- OECD SPE criteria (OECD, 2008): formally registered legal entity; few or no employees; little or no production in the host economy; little or no physical presence; ultimate owners are foreign residents; assets and liabilities mostly vis-a-vis non-residents; core business financing or holding activities.
- Terminology used in analysis:
  - Phantom FDI: FDI into SPEs.
  - Real FDI: foreign investment into non-SPEs.
- Conservative classification rule: foreign-owned corporations with mixed activities are classified as non-SPEs, implying all foreign investment they embody is classified as Real FDI.
- Reporting caveat: about half of the roughly 30 OECD economies that provide information about FDI into SPEs and non-SPEs report zero FDI into SPEs; survey evidence indicates such zeros often reflect absence of collection rather than literal zero and are recoded to missing in the main analysis.
- Retain reported decomposition only for 16 economies that report a value of FDI into SPEs different from zero.

### Orbis usage, construction of sample, and limitations
- Purpose and method:
  - Orbis provides balance sheets, employee counts, and ownership links at unconsolidated entity level to identify immediate and ultimate owners for observed entities.
  - Use Orbis to measure the ratio between FDI by immediate and ultimate investors (ownership-pattern based), not to capture absolute levels (coverage is partial).
  - Use Orbis only for pairs of economies where at least 50 percent of the bilateral FDI is represented in the database.
- Sample construction and exclusions:
  - Corporations extracted where at least one immediate investor with an ownership share of at least 10 percent is located in a foreign economy.
  - To focus on Real FDI, drop corporations resembling SPEs: at most five employees and assets per employee exceeding $10 million.
  - With this definition, around 8 percent of Total FDI covered by Orbis is Phantom FDI.
- Valuation and coverage caveats:
  - Aggregating Orbis-based FDI across all economies yields global Real FDI of around $8.4 trillion in 2016, compared to around $22 trillion from official FDI-statistics-based analysis.
  - Discrepancy mostly attributed to incomplete Orbis coverage and valuation differences (Orbis uses book values; official FDI statistics use market values whenever possible).
  - Only use Orbis conversion factors for pairs where coverage and validation criteria are met (see conversion-factor rules).

### Estimating Real vs Phantom FDI at economy and bilateral levels
- For OECD reporters of bilateral SPE/non-SPE splits: use reported bilateral values directly (example: Netherlands reports inward FDI from the United States of around $750 billion in SPEs and around $200 billion in non-SPEs in 2017).
- For OECD reporters with only aggregate SPE/non-SPE decomposition: apply aggregate SPE/non-SPE shares uniformly across counterpart economies (example: Luxembourg aggregate shares → Real FDI ≈ 4 percent, Phantom FDI ≈ 96 percent; apply to bilateral totals).
- For economies without direct decomposition, predictive model:
  - Univariate log model (2013-2017):
    - log(RealFDI_ht / TotalFDI_ht) = α + β log(TotalFDI_ht / GDP_ht) + ε_ht
  - Estimated slope β around -0.5: increasing Total FDI to GDP by 1 percent reduces predicted Real/Total FDI by around 0.5 percent.
  - Model fit: R^2 = 0.84 for baseline univariate relation.
  - Predicted pattern: Real/Total → 0 as Total FDI / GDP → ∞; lowest predicted ratios for tiny offshore centers (e.g., Cayman Islands, Bermuda, Gibraltar).
  - When model implies Real FDI > Total FDI, bound the share at unity.
  - Fallback for economies lacking GDP data: apply simple average across reporting OECD economies: 78 percent of Total FDI is Real FDI.

### Orbis-based conversion factors to allocate Real FDI to ultimate investor economies
- Definition:
  - Conversion factor δ_hi = estimated ratio between Real FDI in host h by ultimate investors in i and Real FDI in host h by immediate investors in i.
  - Real FDI_UIE_hi = δ_hi Real FDI_IIE_hi (Equation (3)).
- Pair-specific δ_hi (sufficient Orbis coverage and validation):
  - If |Real Orbis FDI_hi / Real FDI_hi − 1| < 0.5 and h ≠ i and numerator and denominator exceed $1 million:
    - δ_hi = Real Orbis FDI_UIE_hi / Real Orbis FDI_IIE_hi (Equation (4)).
  - δ_hi capped between 0.05 and 10.
  - Validation: rank-rank correlation between Orbis-derived δ_hi and actual ratios from reporting OECD countries is 0.72 when weighted with Real FDI; unweighted correlation is 0.43.
  - Example (Spain–United States): Official Real FDI_IIE ≈ $23 billion; Orbis observes $13 billion (coverage > 50 percent) and Orbis UIE ≈ $29 billion → δ_hi = $29 billion / $13 billion = 2.2 → estimated Real FDI_UIE ≈ $23 billion * 2.2 = just over $50 billion.
- Investor-specific δ_hi (low Orbis coverage for pair):
  - If |Real Orbis FDI_hi / Real FDI_hi − 1| > 0.5 and h ≠ i:
    - δ_hi = (Σ_{h∈S} Real Orbis FDI_UIE_hi) / (Σ_{h∈S} Real Orbis FDI_IIE_hi) if both sums exceed $50 million; otherwise set δ_hi = 1 (Equation (5)).
  - Example (Netherlands → Russia): Official Real FDI_IIE ≈ $41 billion; Orbis observes ≈ $19 billion → investor-specific δ_hi ≈ $294 billion / $597 billion ≈ 0.5 → estimated Real FDI_UIE ≈ $41 billion * 0.5 = $21 billion.

### Round-tripping, final adjustments, and internal consistency
- Round-tripping estimation:
  - Real FDI_UIE_hh cannot be derived from δ_hi directly; Equation (6) used:
    - Real FDI_UIE_hh = Real Orbis FDI_UIE_hh * (Σ_{h∈S} Real Orbis FDI_UIE_hi) / Real FDI_IIE_h
    - Right-hand ratio is round-tripping factor observed in Orbis.
  - For economies with Real FDI_IIE_h below $50 million, apply a global round-tripping factor (global round-tripping FDI in Orbis as a share of global FDI in Orbis).
  - Economy-specific round-tripping factors capped between 0 and 0.25.
  - Example (China): Orbis sample round-tripping = 22 percent; China total inward FDI position = $2.7 trillion → aggregate round-tripping FDI in China = $0.6 trillion (adjusted to $0.7 trillion below).
- Internal consistency adjustments:
  - Scale inward Real FDI by ultimate investor economy proportionally so that inward Real FDI aggregated over immediate investor economies matches inward Real FDI aggregated over ultimate investor economies (including the economy itself).
  - Example (Spain): inward Real FDI over immediate investor economies = $624 billion; raw aggregation over ultimate investors = $545 billion → adjustment factor = 1.14.
  - Global magnitude of these adjustments: around $2.8 trillion, or around 8 percent of global FDI.
  - Coverage note: Only around 5 percent of Real FDI is not adjusted due to low Orbis coverage.

### Key empirical findings, cross-sectional patterns and summary statistics
- Global totals and shares:
  - Global Phantom FDI position: $15 trillion (almost 40 percent of Total FDI globally).
  - Global Real FDI position: $25 trillion.
  - Phantom FDI has grown at a faster pace than Real FDI over the last decade.
- Largest Phantom FDI recipients (2017, self-reported or estimated):
  - Luxembourg: around $3.8 trillion (self-reported).
  - Netherlands: around $3.3 trillion (self-reported).
  - Hong Kong SAR: around $1.1 trillion (estimated).
  - British Virgin Islands: around $0.8 trillion (estimated).
  - Bermuda: around $0.8 trillion (estimated).
  - Singapore: around $0.8 trillion (estimated).
  - Cayman Islands: around $0.7 trillion (estimated).
- Major Real FDI patterns:
  - United States hosts Real FDI of around $4 trillion and ultimately owns Real FDI of around $3.6 trillion.
  - China’s inward position exceeds outward position by around $1.2 trillion; estimated ratio above two suggests Chinese investors are ultimate owners of more than twice as much FDI as indicated by outward FDI in standard statistics.
  - Germany, United Kingdom and France: joint outward position exceeds joint inward position by around $1.5 trillion.
  - Some large Real FDI economies (Netherlands, Singapore, Hong Kong SAR) have balanced Real FDI positions, indicating pass-through investment via non-SPEs.
- Ownership ratios and round-tripping:
  - Large developed economies (Italy, France, Spain, Germany, Japan): estimated ratios (ultimately owned Real FDI / immediately owned Total FDI) close to unity.
  - United States and United Kingdom: estimated ratios below unity.
  - Offshore financial centers: ratios much below unity.
  - China: estimated ratio above two.
  - Round-tripping shares:
    - China: around 25 percent of Real FDI ultimately owned by domestic investors.
    - Russia: around 25 percent similarly reported.
    - Most developed economies: round-tripping share far below 10 percent.
- Gravity-model evidence and implications:
  - Baseline OLS gravity for log(FDI): elasticities of Total FDI with respect to host and investor GDP far below unity; R-squared = 0.266.
  - Phantom FDI elasticity w.r.t. host GDP is negative; Real FDI elasticity w.r.t. host GDP is positive and larger than for Total FDI.
  - Replacing immediate investor characteristics with ultimate investor characteristics increases coefficients on market-size variables and raises R-squared to 0.334.
  - Poisson estimator results: coefficient on market size increases from around 0.55 for Total FDI by immediate investor economy to around 0.80 for Real FDI by ultimate investor economy.
  - Interpretation: standard FDI statistics systematically underestimate real links between large economies because investments are channeled through small offshore centers.
- Exposure to tax avoidance opportunities:
  - For outward FDI exposure (20 largest economies by GDP): cross-sectional estimates range from less than 10 percent (Spain) to more than 80 percent (Brazil); United States: more than half of its outward FDI is channeled through phantom corporations.
  - For inward FDI exposure (20 largest economies by GDP): estimates range from around 15 percent (Australia) to around 60 percent (Russia); United States: a bit more than one quarter of inward FDI derives from phantom corporations.
  - Statistically: raising the tax rate by 10 percentage points associated with an increase in exposure through outward FDI of more than 5 percentage points (Column 1, Table 5); tax-rate effect steeper in less developed economies (at lowest GDP per capita, raising tax rate by 10 percentage points associated with increase in exposure through outward FDI of more than 15 percentage points).
- Data summary statistics (selected):
  - Table 3 (2017, inward Total FDI by immediate investor economy): 23,062 pairs totaling 38,566.
    - Reported to CDIS: 16,230 pairs totaling 34,563.
    - Mirror data: 6,693 pairs totaling 3,435.
    - Own estimate: 139 pairs totaling 568.
  - Inward Phantom FDI by IIE: 23,062 pairs totaling 14,653.
    - Reported to OECD: 2,393 pairs totaling 7,436.
    - Estimated: 20,669 pairs totaling 7,217.
  - Inward Real FDI by UIE: 23,032 pairs totaling 23,913.
    - Reported to OECD: 1,598 pairs totaling 5,220.
    - Detailed Orbis: 625 pairs totaling 4,296.
    - Aggregate Orbis: 14,128 pairs totaling 12,965.
    - No adjustment: 6,681 pairs totaling 1,432.

*Source: Excerpts from PDF chapter "2.1    Data sources"; Sections 2.4 and 3.2 as provided in the source content.*

### 2.1    Data sources

### 2.1    Data sources

### Overview of primary data sources
- IMF’s CDIS: most comprehensive source of information on FDI positions; survey conducted annually in 110-115 economies under the IMF’s Statistics Department.
- OECD FDI Statistics: provides additional information on FDI positions for a smaller group of OECD economies, including decomposition into SPEs and non-SPEs for 30 economies (aggregate SPE/non-SPE breakdown available for some additional economies) and inward FDI by economy of the ultimate owner for 16 economies.
- Orbis global firm database: unconsolidated accounting and ownership information for millions of corporate entities; used to infer ownership links and to estimate FDI by ultimate investor economy where direct reporting is unavailable.

### CDIS coverage and survey methodology
- Survey combines administrative data and survey information about foreign investors and foreign investments from local firms; follows detailed guidelines based on international macroeconomic statistical manuals.
- Local authorities aggregate to produce inward FDI by economy of the immediate investor and outward FDI by economy of the immediate investment.
- Aggregate inward FDI across the 111 economies reporting to the CDIS was just above $30 trillion, of which around 95 percent was allocated to immediate investor economies.
- The 16 economies reporting inward FDI in SPEs and non-SPEs separately account for almost 40 percent of the total inward FDI in CDIS.
- The 12 economies further allocating inward Real FDI to ultimate investor economies account for around 20 percent.

### OECD SPE definition and treatment
- SPE (Special Purpose Entity) criteria (OECD, 2008): formally registered legal entity subject to national law; few or no employees; little or no production in the host economy; little or no physical presence; ultimate owners are foreign residents; assets and liabilities are mostly vis-a-vis non-residents; core business consists of group financing or holding activities.
- FDI into SPEs falls within standard FDI definition and is generally included in CDIS reporting.
- Terminology used in analysis:
  - Phantom FDI: FDI into SPEs (label applied because such FDI often has no real economic significance for the host economy).
  - Real FDI: foreign investment into non-SPEs.
- Conservative classification: foreign-owned corporations with mixed activities (e.g., production to local market plus passive shareholdings) are classified as non-SPEs, implying all foreign investment they embody is classified as Real FDI.

### Orbis usage and limitations
- Orbis provides balance sheets, employee counts, and ownership links at unconsolidated entity level; allows identification of immediate and ultimate owners for observed entities.
- Method: use Orbis to measure the ratio between FDI by immediate and ultimate investors (ownership-pattern based), not to capture absolute levels of FDI (coverage is partial).
- Coverage constraint: Orbis represents a subset of corporations in any given economy; coverage varies across economies and over time.
- Use Orbis only for pairs of economies where at least 50 percent of the bilateral FDI is represented in the database.

### Estimating Total FDI by economy of immediate investor
- Estimated bilateral positions of Total FDI: direct investment in host economy h by immediate investors in economy i; estimated network includes all 246 economies and covers 2009-2017.
- For the 110-115 economies reporting inward FDI to CDIS, reported values are used without adjustment (example: France reports inward FDI from Canada of around $3.8 billion in 2017).
- For the 130-135 economies not reporting to CDIS, mirror data are used to estimate FDI positions (example: Cayman Islands inward FDI from Canada estimated from Canada’s reported outward FDI to Cayman Islands of around $34 billion).
- Mirror-data undercoverage issue: mirror positions available only for counterpart economies reporting to CDIS, so simple summation of mirror estimates can substantially underestimate total inward FDI for non-reporting economies (example: sum of mirror estimates for Cayman Islands = $611 billion in 2017, likely an underestimate).

### Adjustment for mirror-data undercoverage
- Adjustment factor α_mirror_t computed as:
  - α_mirror_t = (∑_{h∈R} Inward FDI_reported_{hi,t}) / (∑_{i∈R} Outward FDI_reported_{ih,t})
  - where InwardFDI_reported_hi,t is inward FDI in h from i as reported by h and OutwardFDI_reported_ih,t is outward FDI from i to h as reported by i.
- The adjustment factor in 2017 is 1.17, suggesting aggregate reported inward FDI is 17 percent higher than the sum of mirror positions for reporting economies.
- Application example: Cayman Islands estimated economy-level inward FDI = $712 billion ($611 billion * 1.17); the adjustment $101 billion ($611 billion * 0.17) is allocated to category “Not specified.”
- Adjustment factors for the sample period are in the range 1.15-1.25 in every year.
- Aggregate adjustments account for around 1.5 percent of global FDI because largest FDI recipients tend to report to CDIS and adjustments apply only to non-reporting economies.

### Decomposition into Real vs Phantom FDI
- For OECD countries reporting bilateral SPE/non-SPE splits, use reported bilateral values directly (example: Netherlands reports inward FDI from the United States of around $750 billion in SPEs and around $200 billion in non-SPEs in 2017).
- For OECD countries reporting only an aggregate SPE/non-SPE decomposition, apply aggregate SPE/non-SPE shares uniformly across counterpart economies (example: Luxembourg reports aggregate inward FDI of $3,800 billion in SPEs and less than $200 billion in non-SPEs → Real FDI ≈ 4 percent, Phantom FDI ≈ 96 percent; thus Total FDI from Canada $194 billion → Real FDI $9 billion ($194 billion * 0.04), Phantom FDI $185 billion ($194 billion * 0.96)).
- Reporting caveat: about half of the roughly 30 OECD economies that provide information about FDI into SPEs and non-SPEs report zero FDI into SPEs; survey of statistical agencies indicates such zero often reflects absence of collection rather than literal zero; recoded to missing in main analysis.
- Retain reported decomposition only for 16 economies that report a value of FDI into SPEs different from zero; these 16 economies include financial centers (e.g., Netherlands, Luxembourg), advanced economies (e.g., Norway, Spain), and emerging markets (e.g., Poland, Chile).

### Estimating Real vs Phantom shares for non-reporting economies
- Key empirical relation exploited: strong correlation between ratio Real FDI / Total FDI and ratio Total FDI / GDP across 16 reporting economies.
- Estimated univariate model for 2013-2017:
  - log(RealFDI_ht / TotalFDI_ht) = α + β log(TotalFDI_ht / GDP_ht) + ε_ht
  - where RealFDI_ht is inward FDI in non-SPEs in economy h in year t; TotalFDI_ht is all inward FDI in economy h in year t; GDP is Gross Domestic Product in economy h in year t.
- Estimated slope around -0.5: increasing Total FDI to GDP by 1 percent reduces predicted Real/Total FDI by around 0.5 percent.
- Model fit: R^2 = 0.84 for baseline univariate relation.
- Predicted pattern: ratio Real FDI / Total FDI approaches zero as Total FDI / GDP approaches infinity; lowest predicted ratios for tiny offshore centers such as Cayman Islands, Bermuda, Gibraltar.
- When model implies Real FDI > Total FDI, bound the share at unity.
- For economies lacking GDP data, apply simple average across reporting OECD economies: 78 percent of Total FDI is Real FDI.

### Robustness checks and sensitivity
- Adding controls for market size, economic development, and openness barely changes coefficient on Total FDI / GDP (Table 2 results).
- Leave-one-out re-estimation across 16 reporting economies: 16 regression lines nearly indistinguishable; omission of Luxembourg flattens the slope noticeably, indicating potential sensitivity if Luxembourg is atypical.
- Year-by-year estimation for 2013-2017: estimates virtually identical for 2013-2016 and only slightly different for 2017, indicating temporal stability.
- Sensitivity summary (Figure 4, upper panel): baseline predictions compared with predictions after adding full set of controls (blue), excluding Luxembourg (red), and using only 2017 observations (green); controls occasionally change rankings but not systematic predicted ratios; excluding Luxembourg raises predicted Real/Total for economies with high Total FDI/GDP and lowers it for economies with low Total FDI/GDP; relying only on 2017 has opposite but smaller effects.

### Application examples and illustrative computations
- Example (Cayman Islands): predicted Total FDI is 95 percent Phantom FDI and 5 percent Real FDI; applying to bilateral inward FDI from Canada of $34 billion gives Phantom FDI $32.5 billion and Real FDI $1.5 billion.
- Example (Cayman Islands mirror sum vs adjusted): mirror sum $611 billion → adjusted by 1.17 → $712 billion; adjustment amount $101 billion allocated to “Not specified.”
- When decomposing Total FDI into Real and Phantom for economies without direct decomposition, apply the same predicted share uniformly across all counterpart economies for that host economy.

### Empirical findings and cross-sectional patterns
- Aggregate inward FDI across 111 CDIS reporters just above $30 trillion; around 95 percent allocated to immediate investor economies.
- 16 economies reporting SPE/non-SPE breakdown account for almost 40 percent of total inward FDI in CDIS.
- 12 economies allocating inward Real FDI to ultimate investor economies account for around 20 percent.
- Adjustment factors for mirror-based estimates in sample years range between 1.15 and 1.25.
- Baseline predicted average Real FDI share for reporting OECD economies used as fallback: 78 percent.
- Economies with the lowest predicted Real/Total ratios are predominantly offshore financial centers, overlapping with sink and conduit offshore financial centers identified in Garcia-Bernardo et al. (2017).

*Source: Excerpt from PDF chapter "2.1    Data sources".*

### 2.4    Real FDI by economy of ultimate investor

### 2.4    Real FDI by economy of ultimate investor

### 2.4.1    FDI statistics by economy of ultimate investor
- Purpose: break down inward Real FDI by the economy of the ultimate investors to look through chains of Phantom FDI and capture genuine cross-border investment links.
- Reporting coverage: the reporting of FDI by ultimate investor economy started in 2013, so the global FDI network covers this dimension for the period 2013-2017.
- Use of self-reported data:
  - For OECD reporters of Real FDI by economy of ultimate investor, the reported data are generally used directly.
  - Example: Germany reports inward FDI of $51 billion from ultimate owners in Luxembourg in 2016, compared to $143 billion of inward FDI from immediate owners in Luxembourg reported to the CDIS.
- Incomplete/inconsistent reports:
  - When reported FDI positions by ultimate investor economies are incomplete or inconsistent with FDI positions by immediate investor economies reported to the CDIS, they are ignored and treated as if no breakdown by ultimate investor economy were available.
  - Specific instances ignored ( OECD reports): (i) when less than 50 percent of Real FDI is allocated to specific ultimate investor economies; (ii) when the discrepancy between FDI positions by ultimate investor economy aggregated across all counterpart economies and FDI positions by immediate investor economy aggregated across all counterpart economies exceeds 20 percent.

### 2.4.2    Orbis database
- Objective: for economies without reliable self-reported ultimate-owner breakdowns, convert Real FDI by immediate investor economy into Real FDI by ultimate investor economy using conversion factors δ_hi derived from Orbis.
- Definition and formula:
  - Conversion factor δ_hi = estimated ratio between real investment in host economy h by ultimate investors in economy i and real investment in host economy h by immediate investors in economy i.
  - Real FDI_UIE_hi = δ_hi Real FDI_IIE_hi (Equation (3))
- Orbis sample construction and exclusions:
  - Corporations extracted where at least one immediate investor with an ownership share of at least 10 percent is located in a foreign economy (10 percent threshold used to delineate direct investment vs portfolio investment).
  - To focus on Real FDI, corporations resembling SPEs are dropped: at most five employees and assets per employee exceeding $10 million.
  - With this definition, around 8 percent of Total FDI covered by Orbis is Phantom FDI.
- Estimating shareholders’ investment from Orbis:
  - Equity measured as shareholders’ capital.
  - Internal debt not directly distinguished; assumption: ratio of internal debt to total debt is 0.2 in all non-financial corporations, and internal debt comes from owners in proportion to shareholdings.
  - Calibration: ratio 0.2 set so aggregates roughly match official FDI statistics where the ratio between FDI investment in the form of debt and equity is around 0.25.
  - For each non-financial corporation, estimated shareholders’ investment = shareholders’ capital + 20 percent of total debt.
  - For financial corporations, only shareholders’ capital is included.
- Ownership-share notation and Orbis-based FDI measures:
  - θ_i^c = share of corporation c controlled by ultimate investors in economy i.
  - λ_i^c = share of corporation c owned by immediate investors in economy i.
  - Real Orbis FDI_UIE_hi = Σ_{c∈H} θ_i^c investment_c
  - Real Orbis FDI_IIE_hi = Σ_{c∈H} λ_i^c investment_c
  - H is the set of corporations in host economy h in the Orbis sample; investment_c is estimated value of shareholders’ investment in corporation c.
- Coverage and valuation caveats:
  - Aggregating Orbis-based FDI across all economies yields global Real FDI of around $8.4 trillion in 2016, compared to around $22 trillion from the official FDI-statistics-based analysis.
  - Discrepancy mostly attributed to incomplete Orbis coverage and possible valuation differences (Orbis uses book values; FDI statistics use market values whenever possible).
- Computing pair-specific conversion factors (when Orbis coverage is sufficient):
  - For pairs where Orbis data deviates less than 50 percent from the Real FDI estimate and h ≠ i:
    - δ_hi = Real Orbis FDI_UIE_hi / Real Orbis FDI_IIE_hi if |Real Orbis FDI_hi / Real FDI_hi − 1| < 0.5 and h ≠ i (Equation (4))
    - Requirement: both numerator and denominator of δ_hi must exceed $1 million; otherwise proceed as if Orbis deviated more than 50 percent.
  - δ_hi is capped between 0.05 and 10 to avoid extreme conversions.
  - Validation: rank-rank correlation between Orbis-derived δ_hi and actual ratios from reporting OECD countries is 0.72 when observations are weighted with Real FDI; the unweighted correlation is 0.43, implying larger errors for pairs with lower bilateral FDI.
  - Example (Spain–United States):
    - Official estimate: Spain’s Real FDI from immediate owners in the United States ≈ $23 billion.
    - Orbis observes ≈ $13 billion of this investment (coverage > 50 percent), and Orbis observes investment from ultimate owners in the United States ≈ $29 billion.
    - δ_hi = $29 billion / $13 billion = 2.2.
    - Applying Equation (3): estimated Real FDI in Spain from ultimate owners in the United States ≈ $23 billion * 2.2 = just over $50 billion.
- Investor-specific conversion factors (when Orbis coverage is low):
  - For pairs where Orbis deviates more than 50 percent from Real FDI, compute investor-economy-specific δ_hi uniform across host economies:
    - δ_hi = (Σ_{h∈S} Real Orbis FDI_UIE_hi) / (Σ_{h∈S} Real Orbis FDI_IIE_hi) if |Real Orbis FDI_hi / Real FDI_hi − 1| > 0.5 and h ≠ i (Equation (5))
    - S is the set of host economies in Orbis.
    - Requirement: both numerator and denominator must exceed $50 million; if not met, set δ_hi = 1.
  - Example (Netherlands as investor into Russia):
    - Official estimate: Russia’s Real FDI from immediate owners in the Netherlands ≈ $41 billion.
    - Orbis observes ≈ $19 billion (coverage < 50 percent) so use investor-specific factor.
    - Orbis global: Real Orbis FDI_UIE (from ultimate investors in the Netherlands) ≈ $294 billion; Real Orbis FDI_IIE (from immediate investors in the Netherlands) ≈ $597 billion.
    - δ_hi ≈ $294 billion / $597 billion = around 0.5.
    - Applying Equation (3): estimated Real FDI in Russia from ultimate owners in the Netherlands ≈ $41 billion * 0.5 = $21 billion.

### 2.4.3    Round-tripping and final adjustments
- Round-tripping: a considerable share of FDI in an economy is assigned to ultimate investors in the same economy (round-tripping), motivated by reasons such as tax planning.
  - Example: China’s historical use of Hong Kong SAR and other offshore centers induced round-tripping.
- Estimating round-tripping FDI (inward Real FDI in h with ultimate investor in h):
  - Cannot apply conversion factor to Real FDI_hh since Real FDI_hh is zero by construction.
  - Use Equation (6):
    - Real FDI_UIE_hh = Real Orbis FDI_UIE_hh * (Σ_{h∈S} Real Orbis FDI_UIE_hi) / Real FDI_IIE_h
    - The right-hand ratio is the round-tripping factor: share of Real FDI in h where the ultimate investor is also in h as observed in Orbis.
  - For economies where Real FDI_IIE_h is below $50 million, apply a global round-tripping factor (global round-tripping FDI in Orbis as a share of global FDI in Orbis).
  - Economy-specific round-tripping factors are capped at 0 and 0.25 to prevent extreme factors in economies with low Orbis coverage.
  - Example (China):
    - In the Orbis sample, 22 percent of inward FDI in China is by ultimate investors in China.
    - China’s total inward FDI position = $2.7 trillion.
    - Applying the round-tripping factor: aggregate round-tripping FDI in China = $0.6 trillion (adjusted to $0.7 trillion below).
- Internal consistency adjustments:
  - Adjust inward Real FDI by ultimate investor economy proportionally so that an economy’s inward Real FDI aggregated over immediate investor economies matches the inward Real FDI aggregated over ultimate investor economies (including the economy itself).
  - Adjustments address inconsistencies arising because Orbis sample may not be fully representative.
  - Example (Spain):
    - Aggregating Spain’s inward Real FDI over immediate investor economies = $624 billion.
    - Aggregating over ultimate investor economies (raw) = $545 billion.
    - Adjustment factor applied to Spain’s raw ultimate-owner breakdown = 1.14 (i.e. $624 billion / $545 billion).
  - Global magnitude of adjustments: around $2.8 trillion, or around 8 percent of global FDI.
- Coverage note: Only around 5% of Real FDI is not adjusted due to low coverage in Orbis.

*Source: 2.4    Real FDI by economy of ultimate investor (excerpt).*

### 3.2    Phantom FDI, Real FDI and round-tripping by economy

### 3.2 Phantom FDI, Real FDI and round-tripping by economy

### Distribution of Phantom FDI
- Global Phantom FDI position: $15 trillion.
- Largest economies by Phantom FDI (figures reported or estimated as stated):
  - Luxembourg: around $3.8 trillion (self-reported).
  - Netherlands: around $3.3 trillion (self-reported).
  - Hong Kong SAR: around $1.1 trillion (estimated).
  - British Virgin Islands: around $0.8 trillion (estimated).
  - Bermuda: around $0.8 trillion (estimated).
  - Singapore: around $0.8 trillion (estimated).
  - Cayman Islands: around $0.7 trillion (estimated).
- Self-reported positions (dark red bars in source) are not subject to estimation-method uncertainty; estimated positions (light red bars) are subject to some uncertainty.

### Distribution of Real FDI and ownership patterns
- Global Real FDI position: $25 trillion.
- United States:
  - Hosts Real FDI of around $4 trillion.
  - Ultimately owns Real FDI of around $3.6 trillion.
- China:
  - Inward position exceeds outward position by around $1.2 trillion.
  - Estimated ratio above two suggests Chinese investors are ultimate owners of more than twice as much FDI as indicated by their outward FDI position in standard statistics.
- Germany, United Kingdom and France:
  - Joint outward position exceeds joint inward position by around $1.5 trillion.
- Some economies with large Real FDI (e.g., Netherlands, Singapore, Hong Kong SAR) have strikingly balanced Real FDI positions, suggesting an important element of pass-through investment; some pass-through is handled by non-SPEs and thus classified as real.

### Ultimate vs immediate ownership (ratios)
- For large developed economies (Italy, France, Spain, Germany, Japan): estimated ratios (ultimately owned Real FDI / immediately owned Total FDI) are close to unity — standard FDI statistics accurately reflect aggregate outward FDI.
- United States and United Kingdom: estimated ratios below unity — standard statistics overstate aggregate outward FDI.
- Offshore financial centers (Luxembourg, Hong Kong SAR, Bermuda, British Virgin Islands): ratios much below unity — much outward FDI is ultimately owned by investors in other economies.
- China: estimated ratio above two (see above).

### Round-tripping (real inward FDI ultimately owned by domestic investors)
- Considerable cross-country variation in round-tripping shares.
- Transition economies (examples):
  - China: around 25 percent of Real FDI ultimately owned by domestic investors.
  - Russia: around 25 percent (similar magnitude reported).
- Most developed economies: round-tripping share far below 10 percent, sometimes completely negligible.
- Data source distinctions:
  - Some round-tripping shares are based on national statistical authorities (dark red bars).
  - Others based on the Orbis database (light red bars) and subject to more uncertainty.

### Real bilateral investment links and standard statistics
- Standard FDI statistics (immediate ownership) can understate real bilateral investment links (ultimate ownership).
- Example pair (Germany–United States):
  - CDIS (immediate ownership) reports FDI in Germany from the United States: around $90 billion.
  - Analysis suggests Real FDI link of almost $170 billion.
  - Likely reason: U.S. multinationals channel investments through intermediaries (e.g., Luxembourg, the Netherlands), concealing true bilateral ties.
- Gravity-model evidence (summary of findings):
  - Baseline OLS gravity for log(FDI) yields elasticities of Total FDI with respect to host and investor GDP far below unity and R-squared = 0.266.
  - Separate models for Phantom FDI and Real FDI:
    - Elasticity of Phantom FDI w.r.t. host GDP is negative — smaller economies attract more Phantom FDI in absolute terms.
    - Elasticity of Real FDI w.r.t. host GDP is positive and somewhat larger than for Total FDI.
  - Replacing immediate investor characteristics with ultimate investor characteristics increases coefficients on real variables (market size, distance) and R-squared to 0.334.
  - Poisson estimator (retaining zeros) results:
    - Coefficient on market size variables increases from around 0.55 for Total FDI by immediate investor economy (Column 5) to around 0.80 for Real FDI by ultimate investor economy (Column 8).
    - Interpretation: standard FDI statistics systematically underestimate real links between large economies because investments are channeled through small offshore centers.

### Exposure to tax avoidance opportunities
- Mechanisms by which phantom corporations enable tax avoidance (listed in source):
  - Notional ownership of financial assets, intangible assets and patents that would otherwise be taxed in affiliates with real activities.
  - Facilitation of hybrid financial instruments that create deductible interest expenses.
  - Accumulation and management of foreign profits to postpone repatriation taxes.
  - Exploitation of tax treaty networks to reduce or eliminate taxes on cross-border capital flows.
- Method for estimating exposure:
  - For outward FDI exposure: assume, for a given host economy, the share of inward Total FDI into phantom corporations is uniform across all counterpart economies when required.
    - Example: Luxembourg reports around 95 percent of inward Total FDI is Phantom FDI → assume this share applies to all economies’ outward FDI to Luxembourg.
    - Sweden reports around 5 percent → assume this share applies to all economies’ outward FDI to Sweden.
  - For inward FDI exposure: assume outward/inward Phantom FDI ratio observed in self-reporting economies (aggregate ratio between 1.15 and 1.18 over 2009-2017) applies to non-reporting economies to estimate outward Phantom FDI from inward estimates; also assume share of outward Total FDI coming from phantom corporations is uniform across counterpart economies.
- Estimated exposure through outward FDI (20 largest economies by GDP, Figure 10 upper panel):
  - Cross-sectional variation: exposure ranges from less than 10 percent in Spain to more than 80 percent in Brazil.
  - United States: relatively high estimated exposure (dark bar); more than half of its outward FDI is channeled through phantom corporations.
- Estimated exposure through inward FDI (20 largest economies by GDP, Figure 10 lower panel):
  - Cross-sectional variation: estimates range from around 15 percent in Australia to around 60 percent in Russia.
  - United States: only a bit more than one quarter of its inward FDI derives from phantom corporations (contrast with outward exposure above).
- Correlations with tax rates and development (Table 5):
  - Exposure through outward FDI is significantly higher in economies with higher corporate tax rates:
    - Raising the tax rate by 10 percentage points is associated with an increase in exposure through outward FDI of more than 5 percentage points (Column 1).
  - Tax gradient steeper in less developed economies:
    - At the lowest levels of GDP per capita, raising the tax rate by 10 percentage points is associated with an increase in exposure through outward FDI of more than 15 percentage points (Column 2).
  - Similar but somewhat weaker relationships for exposure through inward FDI (Columns 3-4).
  - Note: statistical power of the model is limited; other factors also play a role.

### Key conclusions and summary statistics
- Phantom FDI accounts for around $15 trillion, almost 40 percent of Total FDI globally.
- Real FDI accounts for around $25 trillion globally.
- Phantom FDI has grown at a faster pace than Real FDI over the last decade.
- Most Phantom FDI concentrated in a small group of offshore centers: Luxembourg ($3.8 trillion), Netherlands ($3.3 trillion), Hong Kong SAR ($1.1 trillion), British Virgin Islands ($0.8 trillion), Bermuda ($0.8 trillion), Singapore ($0.8 trillion), Cayman Islands ($0.7 trillion).
- Real FDI links between the largest economies are much higher than implied by standard FDI statistics due to channeling through offshore centers.

*Source: 3.2 Phantom FDI, Real FDI and round-tripping by economy (from provided IMF content).*

### References

### wpiea2019274-print-pdf - References

### Key empirical findings and highlights from figures and tables
- Largest inward FDI positions in 2017 (top economies listed in figure): Netherlands, United States, Luxembourg, China, United Kingdom, Hong Kong SAR, Switzerland, Singapore, Germany, Ireland, France, Bermuda, British Virgin Islands, Cayman Islands, Australia, Canada, Brazil, Spain, Belgium, Mexico.
- Global decomposition of Total FDI (2009–2017): Real FDI shown in blue bars and Phantom FDI shown in red bars; share of Phantom FDI in Total is plotted on right axis (figures show growth over 2009–2017).
- Validation of Orbis approach: regression slope reported as .72 when plotting Orbis conversion factor against FDI data conversion factor (weighted by value of Real FDI).

### Prediction model (Table 2 and related figures)
- Dependent variable: Real FDI / Total FDI (in logs).
- Baseline pooled OLS result for 2013-2017: coefficient on FDI / GDP (in logs) = -0.5044*** (robust s.e. 0.0357), observations = 75, R-squared = 0.8449.
- Alternative specifications (columns 2–5) show coefficient on FDI / GDP (in logs) = -0.5013***, -0.4884***, -0.5217***, -0.5059*** with R-squared up to 0.8581.
- Additional covariates and reported coefficients (examples):
  - Trade / GDP (in logs) coefficient reported as 0.1943** (s.e. 0.0820) in one specification.
  - GDP (in logs) and GDP / capita (in logs) included in some specifications with coefficients reported (e.g., GDP / capita (in logs) = -0.0934* in one column).

### Model sensitivity and alternative predictions (Figures 2–4, 33–35)
- Sensitivity checks:
  - Excluding single economies (Hungary, Netherlands, Luxembourg) changes regression lines; baseline slope reported as -.504 with robust s.e.: .036, r-squared: .845, period: 2013-2017, observations: 75.
  - Predictions compared across models: baseline univariate vs multivariate with full controls; univariate excluding Luxembourg; univariate using only 2017. Plotted predicted Real FDI/Total FDI values against alternatives (figure shows spread and clustering).
- Economies are distinguished by membership in a broad set of offshore financial centers (red) vs other economies (blue) in prediction comparison plots.

### Phantom FDI: levels, scenarios, and top economies (Figures 6–7, 37–38)
- Baseline global Phantom FDI series plotted 2009–2017; alternative scenarios shown:
  - Only reported Phantom FDI (assumes zero Phantom FDI in non-reporting economies).
  - Alternative prediction model (excluding Luxembourg).
  - Offshore-centers-only assumption (Phantom FDI in non-reporting economies constitutes 90 percent of Total FDI in offshore centers and zero elsewhere).
- Top 20 economies by inward Phantom FDI in 2017 (self-reported or estimated): Luxembourg, Netherlands, Hong Kong SAR, British Virgin Islands, Bermuda, Singapore, Cayman Islands, Switzerland, Ireland, Mauritius, United Kingdom, Cyprus, Malta, Hungary, Curacao, Barbados, Bahamas, Jersey, Gibraltar, Puerto Rico.

### Real FDI and ownership links (Figures 8–9, 39–40)
- Top inward Real FDI positions in 2017 (blue bars) for 20 economies include: United States, China, Netherlands, United Kingdom, Germany, France, Canada, Brazil, Spain, Australia, Switzerland, Belgium, Mexico, Hong Kong SAR, India, Russia, Italy, Singapore, Ireland, Sweden.
- Ratio measures:
  - Ultimately owned Real FDI / Immediately owned Total FDI plotted for top outward FDI economies; chart displays ratios ranging above and below unity for listed economies (China, Italy, France, Spain, Germany, Japan, Sweden, Canada, United Kingdom, United States, Singapore, Switzerland, Belgium, Ireland, Cayman Islands, Netherlands, British Virgin Islands, Bermuda, Hong Kong SAR, Luxembourg).
- Round-tripping factors for 20 largest economies by GDP (share of inward Real FDI ultimately owned by investors in the same economy) plotted; country listing includes China, Russia, Canada, Indonesia, Germany, Italy, Spain, France, United Kingdom, Turkey, United States, Netherlands, India, Switzerland, South Korea, Brazil, Argentina, Mexico, Australia, Saudi Arabia.

### Exposure to tax avoidance opportunities (Figures 10, Table 5, 41, 46)
- Measures plotted:
  - For 20 largest economies by GDP: estimated share of outward Total FDI where the immediate investment is a foreign phantom corporation (top of panel includes Brazil, Indonesia, China, United States, Russia, India, Turkey, Canada, United Kingdom, Saudi Arabia, Taiwan Province of China, Germany, Italy, Australia, Mexico, Japan, France, South Korea, Spain, Argentina).
  - For 20 largest economies by GDP: estimated share of inward Total FDI where the immediate investor is a foreign phantom corporation (top of panel includes Russia, China, Indonesia, India, Germany, Italy, France, United Kingdom, Spain, Brazil, Japan, South Korea, United States, Turkey, Canada, Argentina, Australia, Taiwan Province of China, Mexico, Saudi Arabia).
- Table 5 (OLS regressions for 2016):
  - Dependent variables: share of outward Total FDI where immediate investment is foreign phantom corporation (Columns 1–2) and share of inward Total FDI where immediate investor is foreign phantom corporation (Columns 3–4).
  - Coefficients of interest:
    - Corp tax rate = 0.00598** (s.e. 0.00256) in Column (1).
    - Corp tax rate = 0.0381** (s.e. 0.0179) in Column (2).
    - Corp tax rate × GDP per cap (in logs) = -0.00352* (s.e. 0.00193) in Column (2); = -0.00226* (s.e. 0.00128) in Column (4).
    - Column (1) constant = 0.145** (s.e. 0.0633).
  - Observations: 106, 105, 124, 123 across columns; R-squared reported as 0.050, 0.093, 0.000, 0.027.

### Gravity models and FDI determinants (Table 4, 45)
- Dependent variables across specifications: All FDI, Phantom FDI, Real FDI (OLS pooled and Poisson pooled cross-section).
- Selected coefficient highlights (OLS columns 1–4):
  - Host economy GDP = 0.524*** (s.e. 0.00895) for All FDI; negative coefficient for Phantom FDI: -0.119*** (s.e. 0.0262); positive for Real FDI: 0.615*** (s.e. 0.00863).
  - Immediate investor economy GDP = 0.652*** (s.e. 0.00835) for All FDI; 0.718*** (s.e. 0.0218) for Phantom FDI; 0.667*** (s.e. 0.00806) for Real FDI.
  - Bilateral distance (in logs) negative and significant across specifications (e.g., -0.824***, -0.728***, -0.877***, -0.857*** with reported s.e.).
  - Common language (official) positive and strongly significant (e.g., 1.979*** in column 1).
  - Former colony and contiguity variables show positive associations in many specifications.
- Observations and fit:
  - OLS observations: 30,383 (All FDI), 13,536 (Phantom FDI), 30,369 (Real FDI); R-squared: 0.266, 0.140, 0.305 respectively.
  - Poisson pooled cross-section used for alternative estimations with observations up to 75,361 and reported coefficient estimates for Host, Immediate, and Ultimate investor GDPs (all positive and significant).

### Data sources and coverage (Table 1 and Table 3, 42–44)
- Table 1 (2016, FDI in $billions) aggregates coverage across three macro data sources:
  - Reporting economies: 111 (OECD FDI stats), 16 (IMF CDIS), 12 (OECD FDI stats — presumably different reporting subsets).
  - Total inward FDI totals by source segments:
    - Total inward FDI 30,960 with allocation to IIE: 29,372; unallocated 1,588.
    - Total inward FDI 11,850 with to SPEs ("Phantom FDI"): 6,761; to non-SPEs ("Real FDI"): 5,089.
    - Total inward FDI 6,465 with allocated to UIEs: 6,384; unallocated 81.
- Table 3 (Summary statistics, 2017, FDI in $billions):
  - Inward Total FDI by immediate investor economy (IIE): 23,062 pairs totaling 38,566.
    - Reported to CDIS: 16,230 pairs totaling 34,563.
    - Mirror data: 6,693 pairs totaling 3,435.
    - Own estimate: 139 pairs totaling 568.
  - Inward Phantom FDI by IIE: 23,062 pairs totaling 14,653.
    - Reported to OECD: 2,393 pairs totaling 7,436.
    - Estimated: 20,669 pairs totaling 7,217.
  - Inward Real FDI by IIE: 23,062 pairs totaling 23,912.
    - Reported to OECD: 2,393 pairs totaling 4,610.
    - Estimated: 20,669 pairs totaling 19,302.
  - Inward Real FDI by UIE: 23,032 pairs totaling 23,913.
    - Reported to OECD: 1,598 pairs totaling 5,220.
    - Detailed Orbis: 625 pairs totaling 4,296.
    - Aggregate Orbis: 14,128 pairs totaling 12,965.
    - No adjustment: 6,681 pairs totaling 1,432.

### Additional appendix diagnostics (Figures A1–A4, 48–51)
- Raw Orbis conversion factors and round-tripping factors plotted for up to 50 largest economies by outward and inward Real FDI; raw Orbis conversion factor panel notes "No correction: 1.00".
- Testing uniform share of SPEs in inward and outward FDI: scatter plots comparing imputed share of SPEs by counterpart-economy against reported aggregate share for reporting OECD countries; observations for large counterpart economies are noted to be close to the 45-degree line.

*Source: wpiea2019274-print-pdf - References*

---


_Source: https://www.imf.org/-/media/files/publications/wp/2019/wpiea2019274-print-pdf.pdf_