## Measuring Multinational Production with Foreign Direct Investment Statistics (Executive Summary and Sections IV–V synthesis)

## Source details

**Canonical URL:** [Measuring Multinational Production with Foreign Direct Investment Statistics (Executive Summary and Sections IV–V synthesis)](https://www.imf.org/-/media/files/publications/wp/2023/english/wpiea2023113-print-pdf.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/wp/2023/english/wpiea2023113-print-pdf.pdf.md)
- [Structured JSON version](/-/media/files/publications/wp/2023/english/wpiea2023113-print-pdf.pdf.json)

---

### Importance of FDI as a data source
- FDI stocks accounted for some 30 percent of global cross-border liabilities in 2015 (Lane and Milesi-Ferretti (2021) dataset).
- FDI accounted for over half of foreign financing in more than a third of countries, most of them developing.
- FDI statistics capture financing aspects of MNEs and are widely available and comparable across countries, motivating their use as a proxy for multinational production (defined here as real activity of MNE affiliates such as output or sales).

### Main critiques and measurement challenges
- Conduit FDI and “phantom FDI”
  - Multinational enterprises can channel investments through an enterprise resident in one economy to an enterprise in another economy (conduit FDI or pass-through capital).
  - Conduit FDI is driven by motives including access to sophisticated tax and financial services, protection against claims on affiliates, and protection against political risk.
  - Conduit FDI inflates FDI statistics and obscures ultimate investors and destinations.
  - While conduit FDI can occur in any economy, it is most associated with offshore financial centers (OFCs).
  - It has been estimated that between 30 and 40 percent of total FDI stock is routed through OFCs.
  - Special Purpose Entities (SPEs) are often involved in channelling FDI through OFCs.
- Bilateral statistics and invisibility of ultimate investors
  - Channelling of FDI through intermediate jurisdictions weakens the link between immediate partner economy statistics and the true locations of investors and production.
  - The gap between immediate and ultimate investors in bilateral FDI statistics is a central measurement challenge addressed by recent methodological work (for example, Damgaard et al. (2019); Casella (2019)).
- Source of funds vs use of funds
  - FDI is a specific source of funds and does not capture local and non-affiliated sources of financing of multinational operations.
  - Empirical work has shown discrepancies between FDI reported positions and measures of foreign affiliate operations (employment, value added, gross fixed capital formation, assets), particularly where non-FDI financing and intangible-asset-intensive business models are important.

### Recent statistical progress, data collections, and proposed enhancements (Section IV)
- SPE and UIE data collection and recommendations
  - Several economies already disseminate FDI statistics that separately identify FDI to and from SPEs resident in their economies as a result of OECD and Eurostat data requests.
  - In 2021, the IMF launched a data collection on the cross-border transactions and positions of resident SPEs. The IMF collection includes:
    - FDI statistics for SPEs;
    - portfolio investment and selected trade in services transactions, including financial services, payments for the use of intellectual property, and merchanting.
  - Several economies already disseminate inward FDI positions by the ultimate investing economy (UIE). The BPM6 update proposes a recommendation that countries publish:
    - supplemental annual data of inward FDI statistics by UIE;
    - outward FDI statistics “looking through” any directly held SPEs to the first operating unit.
- Proposed BPM6 / BD4-related enhancements under consideration
  - Breakdown of FDI flows by purpose/type: (1) M&As (purchase/sale of existing equity); (2) greenfield investments (new resources and assets that lead to gross fixed capital formation); (3) extensions of capacity (additional new investments expanding existing affiliates); and (4) financial restructuring. Separate identification of corporate inversions is proposed where relevant.
  - Clarification of coverage of investment funds in FDI and treatment of distributions from retained earnings in affiliates.
  - Fewer recommended methods and more detailed implementation guidance for valuing unlisted equity to reduce bilateral asymmetries in FDI positions.
- Comparability and reconciliation efforts
  - IMF CDIS requests economies report FDI positions using one specific method (Own Funds at Book Value) to enhance comparability.
  - Bilateral reconciliation exercises and, within Europe, exchange of microdata through the FDI Network.
- Additional SPE-related data to assist analysis
  - The IMF data collection will help show the types of activities SPEs carry out in different economies, improving understanding of whether SPEs are conduits or have other functions.

### Analytical methods and practical toolkits available to users
- Two broad estimation strategies to separate conduit/SPE investment from other FDI:
  - Fully estimated approach (e.g., Turban et al., 2020): extends estimation to all economies for comparability.
  - Hybrid approach (e.g., Bolwijn et al., 2018; Damgaard et al., 2019): uses reported SPE data when available and estimates only for non-reporters.
- Specific estimation methodologies described
  - UNCTAD’s implied investment method: relies on a direct linear relationship between inward (or outward) log FDI stock and log GDP; disproportionate FDI relative to GDP is attributed to conduit structures.
  - IMF/DEJ (Damgaard et al., 2019) method: estimates relationship between the share of non-SPE to total FDI and the ratio between total FDI and GDP (“FDI intensity”) using economies reporting non-zero SPEs (variables log-transformed), and extrapolates to others to assign shares of “real” and “phantom” FDI consistent with capacity to absorb real FDI.
  - UNCTAD probabilistic (absorbing Markov chain) approach (Casella, 2019): assigns transition rules to link recipient countries to ultimate investors using bilateral FDI stocks and estimates of conduit FDI.
  - OECD modification in Turban et al. (2020): probabilistic approach where conduit FDI is extrapolated from economies reporting ultimate investors.
  - DEJ firm-level approach: derives economy-pair conversion factors to convert distributions of immediate direct investors into distributions of ultimate investors.

### Key empirical findings, validations, and quantitative results
- Global estimates of share of SPEs in total FDI stock (2016 replication)
  - Total FDI stock, Billion $: 34,183
  - UNCTAD estimated SPEs, Billion $: 12,332 (Share SPEs, Per Cent: 36%)
  - DEJ estimated SPEs, Billion $: 12,543 (Share SPEs, Per Cent: 37%)
- Countries reporting SPEs (2016)
  - Total FDI stock, Billion $: 20,453
  - UNCTAD estimated SPEs, Billion $: 9,382 (Share SPEs, Per Cent: 46%)
  - DEJ estimated SPEs, Billion $: 8,917 (Share SPEs, Per Cent: 44%)
  - Reported SPEs, Billion $: 8,819 (Share SPEs, Per Cent: 43%)
- Top recipient countries by size of inward FDI stock into SPEs (selected figures, 2016)
  - Luxembourg: Total FDI stock, Billion $: 3,775 — UNCTAD estimated SPEs, Billion $: 3,626 (Share SPEs, Per Cent: 96%) — DEJ estimated SPEs, Billion $: 3,480 (Share SPEs, Per Cent: 92%) — Reported SPEs, Billion $: 3,505 (Share SPEs, Per Cent: 93%)
  - Netherlands: Total FDI stock, Billion $: 4,185 — UNCTAD estimated SPEs, Billion $: 3,588 (Share SPEs, Per Cent: 86%) — DEJ estimated SPEs, Billion $: 3,044 (Share SPEs, Per Cent: 73%) — Reported SPEs, Billion $: 3,269 (Share SPEs, Per Cent: 78%)
  - Hong Kong, China: Total FDI stock, Billion $: 1,419 — UNCTAD estimated SPEs, Billion $: 1,048 (Share SPEs, Per Cent: 74%) — DEJ estimated SPEs, Billion $: 989 (Share SPEs, Per Cent: 70%) — Reported SPEs, Billion $: 1,097 (Share SPEs, Per Cent: 77%)
  - British Virgin Islands: Total FDI stock, Billion $: 796 — UNCTAD estimated SPEs, Billion $: 778 (Share SPEs, Per Cent: 98%) — DEJ estimated SPEs, Billion $: 779 (Share SPEs, Per Cent: 98%)
  - Switzerland: Total FDI stock, Billion $: 1,280 — UNCTAD estimated SPEs, Billion $: 727 (Share SPEs, Per Cent: 57%) — DEJ estimated SPEs, Billion $: 689 (Share SPEs, Per Cent: 54%) — Reported SPEs, Billion $: 249 (Share SPEs, Per Cent: 19%)
  - Singapore: Total FDI stock, Billion $: 1,001 — UNCTAD estimated SPEs, Billion $: 647 (Share SPEs, Per Cent: 65%) — DEJ estimated SPEs, Billion $: 654 (Share SPEs, Per Cent: 65%)
  - Ireland: Total FDI stock, Billion $: 841 — UNCTAD estimated SPEs, Billion $: 487 (Share SPEs, Per Cent: 58%) — DEJ estimated SPEs, Billion $: 523 (Share SPEs, Per Cent: 62%)
  - Cayman Islands: Total FDI stock, Billion $: 441 — UNCTAD estimated SPEs, Billion $: 404 (Share SPEs, Per Cent: 92%) — DEJ estimated SPEs, Billion $: 415 (Share SPEs, Per Cent: 94%)
  - United Kingdom: Total FDI stock, Billion $: 1,476 — UNCTAD estimated SPEs, Billion $: 297 (Share SPEs, Per Cent: 20%) — DEJ estimated SPEs, Billion $: 217 (Share SPEs, Per Cent: 15%) — Reported SPEs, Billion $: 415 (Share SPEs, Per Cent: 28%)
  - Mauritius: Total FDI stock, Billion $: 283 — UNCTAD estimated SPEs, Billion $: 218 (Share SPEs, Per Cent: 77%) — DEJ estimated SPEs, Billion $: 247 (Share SPEs, Per Cent: 87%)
  - Total top 10 (UNCTAD): Total FDI stock, Billion $: 11,722 — Estimated SPEs, Billion $: 8,194 (Share SPEs, Per Cent: 70%)
  - Total top 10 (DEJ): Total FDI stock, Billion $: 12,134 — Estimated SPEs, Billion $: 9,286 (Share SPEs, Per Cent: 77%)
  - Total top 10 (Reported): Total FDI stock, Billion $: 11,722 — Reported SPEs, Billion $: 8,536 (Share SPEs, Per Cent: 70%)
- Concordance between FDI stock and operational indicators after removing SPEs
  - Removal of FDI stock into SPEs from total inward FDI stock increases the correlation coefficient (R²) between inward FDI stock and foreign affiliates’ turnover from 0.44 to 0.87 (OECD country sample, 2016).
  - Near-unity elasticity between foreign affiliates’ turnover and FDI stock without SPEs: an x percent increase in the FDI stock is associated with an equivalent x percent increase in sales (as observed in OECD sample).
- BEA-based results for U.S. MNE affiliates (2018)
  - Estimated elasticity of affiliate sales with respect to US outward FDI position: 0.94, with a robust standard error of 0.04, R squared: 86 percent.
  - Estimated elasticity of affiliate total assets with respect to US outward FDI position: 0.95, with a robust standard error of 0.03, R squared: 91 percent.
  - Observations with residuals ≥2 when regressing ln(affiliate sales) on ln(FDI): 11 observations; six of those are on the IMF list of OFCs (Antigua and Barbuda, Singapore, Luxembourg, Malta, St. Lucia, and Anguilla); also UK Pacific Islands and others noted.
- On bilateral distributions and ultimate investor allocations
  - Around 40 percent of foreign affiliates have an economy of the immediate direct investor that differs from the economy of the ultimate investor (UNCTAD, 2016; Alabrese and Casella, 2019).
  - UNCTAD and DEJ methods provide substantial improvements in approximating distributions by ultimate investors compared with immediate-investor bilateral FDI statistics.
  - High correlation between immediate and ultimate ownership bilateral FDI observations in OECD data: correlation coefficient of 83 percent (Wacker, 2020); correlation remains high (81% to 80%) after trimming large observations and zeroes as described.

### Gravity estimation comparison (re-estimation of Wacker (2020), OECD data)
- Estimation setup and selected coefficients (PPML estimator, dropping sum of GDPs; robust standard errors in parentheses; significance stars preserved)
  - ln(GDP investor): OECD immediate: 0.384*** (0.100) — OECD ultimate: 0.668*** (0.109) — p-value for equality: 0.009***
  - ln(GDP recipient): OECD immediate: 1.047*** (0.159) — OECD ultimate: 0.900*** (0.178) — p-value for equality: 0.408
  - ln(distance): OECD immediate: -0.624*** (0.126) — OECD ultimate: -0.334*** (0.134) — p-value for equality: 0.030**
  - Relative skill endowment: OECD immediate: 2.709*** (0.613) — OECD ultimate: 2.462*** (0.560) — p-value for equality: 0.659
  - Constant: OECD immediate: -25.739*** (5.187) — OECD ultimate: -31.740*** (5.008) — p-value for equality: 0.231
  - Observations: 1,324 — R-squared: 0.169 (immediate), 0.192 (ultimate)
- Interpretation highlights
  - No qualitative differences in FDI determinants between immediate and ultimate ownership data.
  - Investor GDP coefficient doubles for ultimate vs. immediate investors; distance coefficient halves for ultimate vs. immediate investors — consistent with OFCs appearing as small immediate investors.
  - Recipient GDP parameter is near unity in both cases, consistent with horizontal gravity model predictions for foreign affiliates’ sales.

### Practical guidance and recommended steps for applied researchers (Section V summary and guidance)
- Metadata and coverage
  - Explore standardized reporting and metadata to understand whether FDI statistics:
    - include SPEs or not,
    - report SPEs separately, and
    - use a particular method for valuing unlisted equity in FDI positions.
- Institutional setting and potential distortions
  - Take into account whether an economy offers access to capital markets or sophisticated financial services, reduced regulatory and/or tax burdens, isolation of the ultimate owner from financial risk, or confidentiality features — these factors can signal SPE presence.
  - In economies with advanced financial markets, affiliates can access external financing, which could lead to understatement of multinational production by FDI positions.
- Adjustments and data choices
  - Where possible, use FDI statistics that separately identify SPEs and remove SPE investment when the objective is to measure multinational production.
  - Prefer inward-side FDI data where consistency is an issue; avoid mixing inward and outward sides in applications to prevent mismatches due to different valuation practices.
  - Use statistics by ultimate investing economy (UIE) where available; where not available, apply established estimation techniques (UNCTAD probabilistic method, DEJ extrapolation, OECD/EIA methods) to approximate UIE distributions or exclude/control for country pairs known to be heavily affected by conduit FDI and round-tripping.
  - Consider hybrid approaches combining reported SPE data with estimation for non-reporters to prioritize empirical accuracy while maintaining broad coverage.
  - Where valuation heterogeneity exists (market vs. book values), consider using book values (Own Funds at Book Value) for cross-country samples to reduce heterogeneity.
- Complementary data sources
  - Compare and complement FDI data with:
    - project-based data,
    - firm-level data,
    - survey-level data, and
    - trade and value-added trade data.
  - Examples cited include project-based data from fDi Markets and Thomson Reuter, firm-level data from ORBIS, survey-level data from Eurostat or US BEA Foreign Affiliates Statistics, and trade in value added data from UNCTAD-EORA GVC database or OECD TiVA database.
- Industry-level cautions
  - Conduit FDI and FDI as a source of funds are more likely in financial service activities; industry-level appropriateness of FDI as a proxy is cautioned and not addressed in depth in this paper.

### Overall assessment and outlook
- Statistical compilers and international organizations have developed data collections, reporting recommendations, and analytical methods that materially improve the ability to reconcile FDI statistics with measures of multinational production.
- The most severe distortions in using FDI as a proxy for multinational production are concentrated in specific OFC jurisdictions; available methods and increased reporting (SPE reporting, UIE statistics) make it feasible to mitigate these distortions substantially.
- While FDI statistics represent sources of funds and not direct measures of use of funds, empirical evidence (elasticities near unity, high R squared in BEA-based comparisons) supports using FDI stocks as appropriate measures of multinational production in most empirical settings, provided analysts account for SPEs, UIE issues, valuation differences, and known bilateral asymmetries.

*Source: IMF Working Paper — Executive Summary and Sections IV–V of "Measuring Multinational Production with Foreign Direct Investment Statistics" (contents provided).*

### Executive Summary ......................................................................................................

### Executive Summary — Measuring Multinational Production with Foreign Direct Investment Statistics

### Importance of FDI as a data source
- FDI stocks accounted for some 30 percent of global cross-border liabilities in 2015 (Lane and Milesi-Ferretti (2021) dataset).
- FDI accounted for over half of foreign financing in more than a third of countries, most of them developing.
- FDI statistics capture financing aspects of MNEs and are widely available and comparable across countries, motivating their use as a proxy for multinational production (defined here as real activity of MNE affiliates such as output or sales).

### Main critique 1 — “Phantom FDI” and conduit FDI through OFCs
- Multinational enterprises can channel investments through an enterprise resident in one economy to an enterprise in another economy (conduit FDI or pass-through capital).
- Conduit FDI is driven by motives including access to sophisticated tax and financial services, protection against claims on affiliates, and protection against political risk.
- Conduit FDI inflates FDI statistics and obscures ultimate investors and destinations.
- While conduit FDI can occur in any economy, it is most associated with offshore financial centers (OFCs).
- It has been estimated that between 30 and 40 percent of total FDI stock is routed through OFCs.
- Special Purpose Entities (SPEs) are often involved in channelling FDI through OFCs.

### Main critique 2 — Bilateral statistics and invisibility of ultimate investors
- Channelling of FDI through intermediate jurisdictions weakens the link between immediate partner economy statistics and the true locations of investors and production.
- The gap between immediate and ultimate investors in bilateral FDI statistics is a central measurement challenge addressed by recent methodological work (e.g., Damgaard et al. (2019); Casella (2019)).

### Main critique 3 — Source of funds vs use of funds
- FDI is a specific source of funds and does not capture local and non-affiliated sources of financing of multinational operations.
- Empirical work has shown discrepancies between FDI reported positions and measures of foreign affiliate operations (employment, value added, gross fixed capital formation, assets), particularly where non-FDI financing and intangible-asset-intensive business models are important.

### Recent statistical progress and analytical methods
- The international statistical community has developed guidelines and initiated collection of data on Special Purpose Entities (SPEs), enabling separate identification of SPE transactions and positions to reveal pure conduit activities.
- Where FDI data to and from SPEs are not available, analytical estimation methods exist to estimate the role of SPEs in aggregate FDI statistics; these methods build on statistical correlations (for example between gross domestic product and FDI stock) and other approaches proposed in recent literature.
- Guidance has been developed on recording FDI positions by the Ultimate Investing Economy (UIE) to look through conduit economies and reveal ultimate origins of FDI.
- Damgaard et al. (2019) proposed an estimation of the SPE component for countries not reporting such information and an analytical procedure to estimate the distribution of ultimate investors.
- UNCTAD (Casella (2019)) proposed a Markov chain–based method to derive bilateral FDI stocks by ultimate investors; different analytical approaches provide complementary solutions to the immediate-versus-ultimate investor gap.

### Counterpoints and mitigating factors
- When the bilateral link involves non-OFC economies, standard bilateral FDI statistics already identify ultimate investors relatively well.
- With careful use and adjustments, IIP-based FDI statistics remain an important data source to study and understand multinational production.

### Practical guidance for researchers using FDI as a proxy for multinational production
- Explore metadata to understand coverage of FDI statistics (for example, whether it includes SPEs and whether SPEs are reported separately).
- Take into account the institutional setting of the economies studied to assess presence of factors that can distort the relationship between FDI positions and multinational production.
- Adjust for FDI to and from SPEs by using published data combined with available analytical methods.
- In bilateral FDI analysis, identify ultimate investors using published data combined with available analytical methods.
- Compare and complement FDI data with other relevant data sources, including project-based data, firm-level data, survey-level data, and trade and value added trade data when available.

### Empirical and literature context (selected findings cited)
- Lipsey (2007): Differences between U.S. outward FDI and U.S. MNE activity metrics (employment, value added, property, plant and equipment), with significant differences in industry distributions attributed to tax avoidance strategies and intangible assets.
- Beugelsdijk et al. (2010): Systematic over-estimation of foreign affiliate operations in jurisdictions classified as OFCs and underestimation in countries with advanced financial systems due to non-FDI financing.
- Leino and Ali-Yrkkö (2014): Recorded annual inflows of FDI do not accurately measure annual real investments in foreign-owned companies; factors include non-FDI financing, cross-border M&As, and conduit investment.
- Blanchard and Acalin (2016): Evidence suggesting a prominent role for conduit investment based on a high correlation between quarterly inward and outward FDI flows.
- Sauvant (2017): Total assets for United States foreign affiliates in 2012 largely exceeded U.S. outward FDI, with up to two-thirds of foreign affiliate assets not financed by FDI.
- Damgaard et al. (2019): Exhaustive account of pass-through capital challenges, proposal to estimate SPE components, and an analytical procedure to estimate ultimate investor distributions.
- Casella (2019) (UNCTAD): Markov chain–based derivation of bilateral FDI stocks by ultimate investors.

*Source: IMF Working Paper — Executive Summary of "Measuring Multinational Production with Foreign Direct Investment Statistics" (contents provided).*

### Section IV takes stock of the main initiatives taking place at the level of the international statistical community

### Section IV takes stock of the main initiatives taking place at the level of the international statistical community

### Main recent progress in FDI statistics and data collections
- Several economies already disseminate FDI statistics that separately identify FDI to and from SPEs resident in their economies as a result of OECD and Eurostat data requests.
- In 2021, the IMF launched a data collection on the cross-border transactions and positions of resident SPEs. The IMF collection includes:
  - FDI statistics for SPEs;
  - portfolio investment and selected trade in services transactions, including financial services, payments for the use of intellectual property, and merchanting.
- Several economies already disseminate inward FDI positions by the ultimate investing economy (UIE). The BPM6 update proposes a recommendation that countries publish:
  - supplemental annual data of inward FDI statistics by UIE;
  - outward FDI statistics “looking through” any directly held SPEs to the first operating unit.
- Proposed BPM6 / BD4-related enhancements under consideration:
  - Breakdown of FDI flows by purpose/type: (1) M&As (purchase/sale of existing equity); (2) greenfield investments (new resources and assets that lead to gross fixed capital formation); (3) extensions of capacity (additional new investments expanding existing affiliates); and (4) financial restructuring. Separate identification of corporate inversions is proposed where relevant.
  - Clarification of coverage of investment funds in FDI and treatment of distributions from retained earnings in affiliates.
  - Fewer recommended methods and more detailed implementation guidance for valuing unlisted equity to reduce bilateral asymmetries in FDI positions.
- Complementary efforts to improve comparability and reduce asymmetries:
  - IMF CDIS requests economies report FDI positions using one specific method (Own Funds at Book Value) to enhance comparability.
  - Bilateral reconciliation exercises and, within Europe, exchange of microdata through the FDI Network.
- Proposed additional SPE-related data to assist analysis:
  - The IMF data collection will help show the types of activities SPEs carry out in different economies, improving understanding of whether SPEs are conduits or have other functions.

### Analytical methods and practical toolkits available to users
- Two broad estimation strategies to separate conduit/SPE investment from other FDI:
  - Fully estimated approach (e.g., Turban et al., 2020): extends estimation to all economies for comparability.
  - Hybrid approach (e.g., Bolwijn et al., 2018; Damgaard et al., 2019): uses reported SPE data when available and estimates only for non-reporters.
- Specific estimation methodologies described:
  - UNCTAD’s implied investment method: relies on a direct linear relationship between inward (or outward) log FDI stock and log GDP; disproportionate FDI relative to GDP is attributed to conduit structures.
  - IMF/DEJ (Damgaard et al., 2019) method: estimates relationship between the share of non-SPE to total FDI and the ratio between total FDI and GDP (“FDI intensity”) using economies reporting non-zero SPEs (variables log-transformed), and extrapolates to others to assign shares of “real” and “phantom” FDI consistent with capacity to absorb real FDI.
  - UNCTAD probabilistic (absorbing Markov chain) approach (Casella, 2019): assigns transition rules to link recipient countries to ultimate investors using bilateral FDI stocks and estimates of conduit FDI.
  - OECD modification in Turban et al. (2020): probabilistic approach where conduit FDI is extrapolated from economies reporting ultimate investors.
  - DEJ firm-level approach: derives economy-pair conversion factors to convert distributions of immediate direct investors into distributions of ultimate investors.

### Key empirical findings, validations, and quantitative results
- Global estimates of share of SPEs in total FDI stock (2016 replication):
  - Total FDI stock, Billion $: 34,183
  - UNCTAD estimated SPEs, Billion $: 12,332 (Share SPEs, Per Cent: 36%)
  - DEJ estimated SPEs, Billion $: 12,543 (Share SPEs, Per Cent: 37%)
- Countries reporting SPEs (2016):
  - Total FDI stock, Billion $: 20,453
  - UNCTAD estimated SPEs, Billion $: 9,382 (Share SPEs, Per Cent: 46%)
  - DEJ estimated SPEs, Billion $: 8,917 (Share SPEs, Per Cent: 44%)
  - Reported SPEs, Billion $: 8,819 (Share SPEs, Per Cent: 43%)
- Top 10 recipient countries by size of inward FDI stock into SPEs (selected figures, 2016):
  - Luxembourg: Total FDI stock, Billion $: 3,775 — UNCTAD estimated SPEs, Billion $: 3,626 (Share SPEs, Per Cent: 96%) — DEJ estimated SPEs, Billion $: 3,480 (Share SPEs, Per Cent: 92%) — Reported SPEs, Billion $: 3,505 (Share SPEs, Per Cent: 93%)
  - Netherlands: Total FDI stock, Billion $: 4,185 — UNCTAD estimated SPEs, Billion $: 3,588 (Share SPEs, Per Cent: 86%) — DEJ estimated SPEs, Billion $: 3,044 (Share SPEs, Per Cent: 73%) — Reported SPEs, Billion $: 3,269 (Share SPEs, Per Cent: 78%)
  - Hong Kong, China: Total FDI stock, Billion $: 1,419 — UNCTAD estimated SPEs, Billion $: 1,048 (Share SPEs, Per Cent: 74%) — DEJ estimated SPEs, Billion $: 989 (Share SPEs, Per Cent: 70%) — Reported SPEs, Billion $: 1,097 (Share SPEs, Per Cent: 77%)
  - British Virgin Islands: Total FDI stock, Billion $: 796 — UNCTAD estimated SPEs, Billion $: 778 (Share SPEs, Per Cent: 98%) — DEJ estimated SPEs, Billion $: 779 (Share SPEs, Per Cent: 98%)
  - Switzerland: Total FDI stock, Billion $: 1,280 — UNCTAD estimated SPEs, Billion $: 727 (Share SPEs, Per Cent: 57%) — DEJ estimated SPEs, Billion $: 689 (Share SPEs, Per Cent: 54%) — Reported SPEs, Billion $: 249 (Share SPEs, Per Cent: 19%)
  - Singapore: Total FDI stock, Billion $: 1,001 — UNCTAD estimated SPEs, Billion $: 647 (Share SPEs, Per Cent: 65%) — DEJ estimated SPEs, Billion $: 654 (Share SPEs, Per Cent: 65%)
  - Ireland: Total FDI stock, Billion $: 841 — UNCTAD estimated SPEs, Billion $: 487 (Share SPEs, Per Cent: 58%) — DEJ estimated SPEs, Billion $: 523 (Share SPEs, Per Cent: 62%)
  - Cayman Islands: Total FDI stock, Billion $: 441 — UNCTAD estimated SPEs, Billion $: 404 (Share SPEs, Per Cent: 92%) — DEJ estimated SPEs, Billion $: 415 (Share SPEs, Per Cent: 94%)
  - United Kingdom: Total FDI stock, Billion $: 1,476 — UNCTAD estimated SPEs, Billion $: 297 (Share SPEs, Per Cent: 20%) — DEJ estimated SPEs, Billion $: 217 (Share SPEs, Per Cent: 15%) — Reported SPEs, Billion $: 415 (Share SPEs, Per Cent: 28%)
  - Mauritius: Total FDI stock, Billion $: 283 — UNCTAD estimated SPEs, Billion $: 218 (Share SPEs, Per Cent: 77%) — DEJ estimated SPEs, Billion $: 247 (Share SPEs, Per Cent: 87%)
  - Total top 10 (UNCTAD): Total FDI stock, Billion $: 11,722 — Estimated SPEs, Billion $: 8,194 (Share SPEs, Per Cent: 70%)
  - Total top 10 (DEJ): Total FDI stock, Billion $: 12,134 — Estimated SPEs, Billion $: 9,286 (Share SPEs, Per Cent: 77%)
  - Total top 10 (Reported): Total FDI stock, Billion $: 11,722 — Reported SPEs, Billion $: 8,536 (Share SPEs, Per Cent: 70%)
- Concordance between FDI stock and operational indicators after removing SPEs:
  - Removal of FDI stock into SPEs from total inward FDI stock increases the correlation coefficient (R²) between inward FDI stock and foreign affiliates’ turnover from 0.44 to 0.87 (OECD country sample, 2016).
  - Near-unity elasticity between foreign affiliates’ turnover and FDI stock without SPEs: an x percent increase in the FDI stock is associated with an equivalent x percent increase in sales (as observed in OECD sample).
- BEA-based results for U.S. MNE affiliates (2018):
  - Estimated elasticity of affiliate sales with respect to US outward FDI position: 0.94, with a robust standard error of 0.04, R squared: 86 percent.
  - Estimated elasticity of affiliate total assets with respect to US outward FDI position: 0.95, with a robust standard error of 0.03, R squared: 91 percent.
  - Observations with residuals ≥2 when regressing ln(affiliate sales) on ln(FDI): 11 observations; six of those are on the IMF list of OFCs (Antigua and Barbuda, Singapore, Luxembourg, Malta, St. Lucia, and Anguilla); also UK Pacific Islands and others noted.
- On bilateral distributions and ultimate investor allocations:
  - Around 40 percent of foreign affiliates have an economy of the immediate direct investor that differs from the economy of the ultimate investor (UNCTAD, 2016; Alabrese and Casella, 2019).
  - UNCTAD and DEJ methods provide substantial improvements in approximating distributions by ultimate investors compared with immediate-investor bilateral FDI statistics.
  - High correlation between immediate and ultimate ownership bilateral FDI observations in OECD data: correlation coefficient of 83 percent (Wacker, 2020); correlation remains high (81% to 80%) after trimming large observations and zeroes as described.
- Gravity estimation comparison (re-estimation of Wacker (2020), OECD data, PPML estimator, dropping sum of GDPs):
  - Table of estimates (robust standard errors in parentheses; significance stars preserved):
    - ln(GDP investor): OECD immediate: 0.384*** (0.100) — OECD ultimate: 0.668*** (0.109) — p-value for equality: 0.009***
    - ln(GDP recipient): OECD immediate: 1.047*** (0.159) — OECD ultimate: 0.900*** (0.178) — p-value for equality: 0.408
    - ln(distance): OECD immediate: -0.624*** (0.126) — OECD ultimate: -0.334*** (0.134) — p-value for equality: 0.030**
    - Relative skill endowment: OECD immediate: 2.709*** (0.613) — OECD ultimate: 2.462*** (0.560) — p-value for equality: 0.659
    - Constant: OECD immediate: -25.739*** (5.187) — OECD ultimate: -31.740*** (5.008) — p-value for equality: 0.231
    - Observations: 1,324 — R-squared: 0.169 (immediate), 0.192 (ultimate)
  - Interpretation highlights:
    - No qualitative differences in FDI determinants between immediate and ultimate ownership data.
    - Investor GDP coefficient doubles for ultimate vs. immediate investors; distance coefficient halves for ultimate vs. immediate investors — consistent with OFCs appearing as small immediate investors.
    - Recipient GDP parameter is near unity in both cases, consistent with horizontal gravity model predictions for foreign affiliates’ sales.

### Practical guidance and recommended steps for applied researchers (from concluding Section V)
- Researchers using IIP-based FDI statistics should consider practical steps (summary of recommendations from the paper):
  - Where possible, use FDI statistics that separately identify SPEs and remove SPE investment when the objective is to measure multinational production.
  - Prefer inward-side FDI data where consistency is an issue; avoid mixing inward and outward sides in applications to prevent mismatches due to different valuation practices.
  - Use statistics by ultimate investing economy (UIE) where available; where not available, apply established estimation techniques (UNCTAD probabilistic method, DEJ extrapolation, OECD/EIA methods) to approximate UIE distributions or exclude/control for country pairs known to be heavily affected by conduit FDI and round-tripping.
  - When performing gravity or determinant analyses, recognize that use of ultimate-investor statistics can change magnitudes (e.g., investor GDP and distance coefficients) but typically does not overturn qualitative conclusions about FDI determinants.
  - Consider hybrid approaches combining reported SPE data with estimation for non-reporters to prioritize empirical accuracy while maintaining broad coverage.
  - Where valuation heterogeneity exists (market vs. book values), consider using book values (Own Funds at Book Value) for cross-country samples to reduce heterogeneity.
  - Where feasible, supplement macro FDI statistics with enterprise-level or project-level information (FATS, greenfield announcements, M&A project data) while acknowledging their coverage limitations.

### Overall assessment of progress and outlook
- Statistical compilers and international organizations have developed data collections, reporting recommendations, and analytical methods that materially improve the ability to reconcile FDI statistics with measures of multinational production.
- The most severe distortions in using FDI as a proxy for multinational production are concentrated in specific OFC jurisdictions; available methods and increased reporting (SPE reporting, UIE statistics) make it feasible to mitigate these distortions substantially.
- While FDI statistics represent sources of funds and not direct measures of use of funds, empirical evidence (elasticities near unity, high R squared in BEA-based comparisons) supports using FDI stocks as appropriate measures of multinational production in most empirical settings, provided analysts account for SPEs, UIE issues, valuation differences, and known bilateral asymmetries.

*IMF Working Papers — Section IV summary as provided in the source content*

### 1.    Explore the standardized reporting as well as the metadata to understand the coverage and valuation

### 1.    Explore the standardized reporting as well as the metadata to understand the coverage and valuation of FDI statistics

### Coverage and valuation: metadata and standardized reporting
- Explore standardized reporting and metadata to understand whether FDI statistics:
  - include SPEs or not,
  - report SPEs separately, and
  - use a particular method for valuing unlisted equity in FDI positions.
- All of the databases mentioned in footnote 23 include metadata information, and many economies publish such information on their websites.

### Institutional setting and potential distortions
- Take into account the institutional setting of the economies being studied to detect factors that can distort the relationship between FDI positions and multinational production:
  - If the economy offers access to capital markets or sophisticated financial services, reduced regulatory and/or tax burdens, isolates the ultimate owner from financial risk, or ensures confidentiality, it may host SPEs (see Section II).
  - If the host economy has advanced financial markets, the affiliate can access external financing, which could lead to an understatement of multinational production (see Section III).

### Industry-level analysis: cautions and limitations
- Analysts should be careful with FDI data on the industry level:
  - Conduit FDI and FDI as a source of funds are more likely to be present in financial service activities.
  - An industry-level analysis of the appropriateness of FDI as a proxy for multinational production is beyond the scope of this paper.
  - Lipsey (2007) highlights that the cross-country distribution of FDI provides a much more accurate representation than the distribution across industries.

### Adjustments for SPEs and ultimate investors
- Aggregate analysis:
  - Adjust for SPEs using published data combined with one of the available analytical methods as described in Section II of this paper.
- Bilateral analysis:
  - Adjust for ultimate investors using published data combined with one of the available analytical methods discussed in Section II.
  - If adjustment for ultimate investors is not possible, FDI statistics by immediate partner country can be used for some analyses, especially if country pairs most liable to tax optimization and round-tipping are either excluded or controlled for (e.g., with specific pair fixed effects).

### Complementary data sources
- Compare and complement FDI data with other relevant data sources when available:
  - project-based data,
  - firm-level data,
  - survey-level data, and
  - trade and value-added trade data.
- Examples of data sources noted in the text include:
  - project-based data from fDi Markets for announced greenfield projects and from Thomson Reuter for cross-border M&As;
  - firm-level data from ORBIS;
  - survey-level data from Eurostat or US BEA Foreign Affiliates Statistics;
  - trade in value added data from UNCTAD-EORA GVC database or OECD TiVA database.

### Role and value of IIP-based FDI statistics
- With care taken in their use, IIP-based FDI statistics remain an important and instrumental data source to study and understand multinational production.
- FDI statistics capture financial considerations that play a key role in how multinational production networks are structured (see also Davies and Markusen, 2021).
- Exclusive focus on “real” activities risks failing to capture financial motives that differentiate multinational enterprises from domestic counterparts.
- FDI statistics that separately identify resident SPEs and ultimate vs. immediate ownership provide consistent data to study the interconnection of ‘real’ and ‘financial’ motives.
- FDI statistics could better reflect important aspects of production such as intangible capital that are difficult to measure.
- FDI statistics allow separation of income from value added in the context of multinational production, which is essential to understanding the welfare effects of globalization (e.g., Bohn et al. 2021; Wang et al., 2021).

*IMF Working Paper — Measuring Multinational Production with Foreign Direct Investment Statistics*

---


_Source: https://www.imf.org/-/media/files/publications/wp/2023/english/wpiea2023113-print-pdf.pdf_