2.1 Data sources

The IMF’s CDIS is the most comprehensive source of information on FDI positions. The survey is currently conducted annually in 110–115 economies, including all major economies in the world, under the auspices of the IMF’s Statistics Department. In each of the participating economies, national authorities combine administrative data and survey information about foreign investors and foreign investments from local firms. The survey process follows detailed guidelines based on international macroeconomic statistical manuals to ensure consistent definitions and similar data collection practices across the participating economies. The local authorities aggregate the survey information to produce estimates of inward FDI by the economy of the immediate investor and outward FDI by the economy of the immediate investment.⁹

The OECD FDI Statistics include additional information on FDI positions for a smaller group of OECD economies. First, 30 economies specify how much inward FDI is into SPEs and how much is into other entities (non-SPEs). The statistical manuals describe an SPE as a formally registered legal entity subject to national law that satisfies several criteria: it has few or no employees; it has little or no production in the host economy; it has little or no physical presence; its ultimate owners are foreign residents; its assets and liabilities are mostly vis-a-vis non-residents; and its core business consists of group financing or holding activities (OECD, 2008).¹⁰ FDI into SPEs falls within the standard definition of FDI and is generally included in the FDI reported to CDIS. However, since it has no real economic significance for the host economy, we label it Phantom FDI as opposed to foreign investment into non-SPEs, which we label Real FDI. This decomposition yields a conservative estimate of Phantom FDI as foreign-owned corporations with mixed activities, e.g. production to the local market and passive shareholdings in foreign subsidiaries, are classified as non-SPEs, implying that all the foreign investment they embody is classified as Real FDI.¹¹ Second, 16 economies provide information on inward FDI by the economy of the ultimate owner. The ultimate owner is identified by starting from the immediate owner and moving up the ownership chain through the controlling links. The first enterprise in the chain that is not itself controlled by another enterprise (e.g. because it is a listed firm with no single shareholder owning more than 50 percent) is the ultimate owner. If the immediate owner is not controlled by another enterprise, it is itself the ultimate owner.

We summarize the three FDI data sources and their coverage in 2016 in Table 1.¹² 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 (non-zero) inward FDI in SPEs and non-SPEs separately account for almost 40 percent of the total inward FDI in CDIS whereas the 12 economies further allocating inward Real FDI to ultimate investor economies account for around 20 percent.

Table 1:

Data sources and coverage (2016, FDI in $billions). The table provides an overview of the three macro data sources and the coverage in terms of economies and FDI.

Table 1:

Data sources and coverage (2016, FDI in $billions). The table provides an overview of the three macro data sources and the coverage in terms of economies and FDI.

IMF CDIS		OECD FDI stats		OECD FDI stats
Reporting economies	111	Reporting economies	16	Reporting economies	12
Total inward FDI	30,960	Total inward FDI	11,850	Total inward FDI	6,465
- allocated to IIE	29,372	- to SPEs (“Phantom FDI”)	6,761	- allocated to UIEs	6,384
- unallocated	1,588	- to non-SPEs (“Real FDI”)	5,089	- unallocated	81

View Table

Table 1:

Data sources and coverage (2016, FDI in $billions). The table provides an overview of the three macro data sources and the coverage in terms of economies and FDI.

IMF CDIS		OECD FDI stats		OECD FDI stats
Reporting economies	111	Reporting economies	16	Reporting economies	12
Total inward FDI	30,960	Total inward FDI	11,850	Total inward FDI	6,465
- allocated to IIE	29,372	- to SPEs (“Phantom FDI”)	6,761	- allocated to UIEs	6,384
- unallocated	1,588	- to non-SPEs (“Real FDI”)	5,089	- unallocated	81

Finally, the global firm database Orbis includes accounting and ownership information for millions of corporate entities at the unconsolidated level. To be concrete, consider a multinational group comprising a listed parent company in Canada, an intermediate holding company in Luxembourg and an operating subsidiary in Spain. In this example, we would observe a balance sheet and background information such as the number of employees for each of the three entities. We would also observe the ownership links between them. For the Spanish entity, the database would indicate that the immediate owner is the Luxembourg company and the ultimate owner is the Canadian company (both with a 100 percent shareholding). For the Luxembourg company, it would indicate that the Canadian company is both the immediate and the ultimate owner (with a 100 percent shareholding).

We use the information from Orbis to produce estimates of FDI by ultimate investor economy for economies that do not report such statistics directly.¹³ Intuitively, the concrete example above would suggest that inward FDI in Spain by immediate investors in Canada should be adjusted upwards to arrive at FDI in Spain by ultimate investors in Canada; conversely, FDI in Spain by immediate investors in Luxembourg should be adjusted downwards to arrive at FDI in Spain by ultimate investors in Luxembourg. A key constraint is Orbis’ partial coverage (e.g. Tørsløv et al., 2018): in any given economy only a subset of corporations are represented in the database, and coverage varies considerably across economies and over time. Importantly, we do not attempt to capture levels of FDI with the Orbis data and therefore do not assume that coverage is complete. Rather we use Orbis to measure the ratio between FDI by immediate and ultimate investors in a given economy and thus assume that the ownership patterns observed in the Orbis sample are representative of the wider economy. For this reason, we only rely on Orbis data for pairs of economies where a large part of the bilateral FDI (at least 50 percent) is represented in the database.

2.2 Total FDI by economy of immediate investor

We first estimate the bilateral positions of Total FDI: the direct investment in host economy h by immediate investors in economy i. The estimated network includes all 246 economies in the world and covers the period 2009–2017.

For the 110–115 economies reporting inward FDI positions to the CDIS, we use these reports without any adjustment. For instance, France reports inward FDI from Canada of around $3.8 billion in 2017 and this figure enters our database directly. For the 130–135 economies not reporting to the CDIS, mostly small developing economies and offshore financial centers, we estimate FDI positions based on mirror data. For instance, the Cayman Islands does not report inward FDI from Canada; however, Canada reports outward FDI to the Cayman Islands of around $34 billion and we use this number as an estimate of inward FDI into the Cayman Islands from Canada.¹⁴

Since mirror positions are only available for counterpart economies reporting to the CDIS, we would systematically underestimate economy-level inward FDI for non-reporting economies if we only relied on mirror positions. For instance, when we estimate the inward FDI position of the Cayman Islands against the rest of the world in 2017, we can only create mirror estimates for 53 economies; the subset of the economies participating in the CDIS that report their outward FDI position to the Cayman Islands. The simple sum of these mirror estimates, $611 billion, almost certainly underestimates the sum of all inward FDI to the Cayman Islands from the 246 economies in the world.

We address this gap by applying an adjustment factor to the sum of the mirror estimates when we compute economy-level FDI positions for non-reporting economies. The adjustment factor is computed as the ratio between reported inward FDI positions and the sum of mirror positions for the set of economies R reporting inward FDI:

where $InwardFD{I}_{hi,t}^{reported}$ is inward FDI in h from i as reported by h and $OutwardFD{I}_{ih,t}^{reported}$ is outward FDI from i to h as reported by i. The adjustment factor in 2017 is 1.17 suggesting that, for the aggregate of reporting economies, actual total inward FDI is 17 percent higher than the sum of the mirror positions. We assume that the same is true for each of the non-reporting economies. For the Cayman Islands, this implies that we estimate economy-level inward FDI at $712 billion ($611 billion * 1.17). We are unable to allocate the adjustment $101 billion ($611 billion * 0.17) to any specific counterpart economy and therefore allocate to the category “Not specified.”

We compute the adjustment factor for each year in the sample period separately; the adjustments are in the range 1.15–1.25 in every year. We emphasize that aggregate adjustments only account for around 1.5 percent of global FDI because the largest recipients of FDI tend to report to CDIS and the adjustments only apply to non-reporting economies.

2.3 Decomposition: Real vs Phantom investment

In the next step, we decompose each inward FDI position into Real FDI and Phantom FDI. For the sample of OECD countries reporting by counterpart economy how much inward FDI is into SPEs and non-SPEs respectively, we use this information directly.¹⁵ For instance, the Netherlands reports inward FDI from the United States of around $750 billion in SPEs and of around $200 billion in non-SPEs in 2017; these figures enter our database as estimates of Phantom FDI and Real FDI without any adjustments. For the sample of OECD countries specifying the decomposition on SPEs and non-SPEs only in the aggregate, we assume that the decomposition is the same for all counterpart economies. For instance, Luxembourg reports aggregate inward FDI of $3,800 billion in SPEs and less than $200 billion in non-SPEs, implying that Real FDI and Phantom FDI account for around 4 percent and 96 percent of Total FDI respectively in the aggregate. To estimate Real FDI and Phantom FDI at the bilateral level, we assume that these shares apply uniformly to all counterpart economies. For instance, Luxembourg reports Total FDI from Canada of $194 billion and we thus estimate that Real FDI from Canada is $9 billion ($194 billion * 0.04) whereas Phantom FDI is $185 billion ($194 billion * 0.96).

While 30 OECD economies provide information about FDI into SPEs and non-SPEs, around half of these economies simply report that there is no FDI into SPEs. Most notably, this is the case for the United States despite evidence that the state of Delaware is a leading jurisdiction for incorporation of empty shells. It hosts more than 1 million business entities (Delaware Division of Corporations) that serve the tax planning purposes of U.S. as well as non-U.S. firms (Dyreng et al, 2013).¹⁶ To understand the nature of these reports, we have conducted a survey among the divisions of statistical agencies responsible for FDI statistics. The survey responses indicate that FDI into SPEs are not literally zero even when this is reported in the new FDI statistics. Instead, it often indicates that the information is not collected. Given the uncertainty about how to interpret such observations, we recode them to missing in the main analysis and only retain the reported decomposition of FDI into SPEs and non-SPEs for the 16 economies that report a value of FDI into SPEs different from zero. The 16 economies include some of the world’s most important financial centers (e.g. the Netherlands and Luxembourg), advanced economies (e.g. Norway and Spain) and emerging markets (e.g. Poland and Chile), and the sample therefore represents a broad subset of the global economy.

Based on the information from the 16 economies reporting an actual breakdown of Total FDI into Real FDI and Phantom FDI, we estimate a similar decomposition for the remaining economies in the world. The estimation exploits a highly salient correlation within the sample of reporting economies between the ratio of Real to Total FDI and the ratio of Total FDI to GDP. For instance, Sweden where Total FDI stands at around 0.7 times GDP, reports that Real FDI accounts for around 95 percent of Total FDI whereas Luxembourg with Total FDI exceeding 60 times GDP, reports that Real FDI accounts for less than 5 percent of Total FDI. There are strong economic reasons to expect this relation to hold more broadly. While the capacity to absorb real investment into firms with employees and productive assets is limited, there are no economic bounds on phantom investment into empty corporate shells with no ties to the local economy. When the Total FDI position is unusually large relative to the size of the economy, it is therefore reasonable to expect that a smaller share of the FDI is real. Our approach is thus to extrapolate the relationship between the ratio of Real FDI to Total FDI, which is unobserved except for the 16 reporting economies, and the ratio of Total FDI to GDP, which can be calculated for virtually all economies in the world.

We first estimate the following univariate model for the 16 reporting economies and the period 2013–2017:

where Real FDI_ht is inward FDI in non-SPEs in economy h in year t; Total FDI_ht is all inward FDI in economy h in year t; and GDP is the Gross Domestic Product in economy h in year t.¹⁷

Figure 2 (upper panel) illustrates the estimated log-log relation: the estimated slope is around -0.5 suggesting that increasing the ratio of Total FDI to GDP by 1 percent reduces the predicted ratio of Real FDI to Total FDI by around 0.5 percent. The figure highlights the three economies in the sample, which consistently have a low ratio of Real FDI to Total FDI through the sample period: Luxembourg (red), Netherlands (blue) and Hungary (green). The R² of the model is 0.84.

View Full Size

Prediction model and predictions. Top panel: the figure shows the ratio of Total FDI to GDP (x-axis) plotted against the ratio of Real FDI to Total FDI (y-axis) for the OECD countries reporting non-zero Phantom FDI. Bottom Panel: the figure shows the relation between between Total FDI/GDP and Real FDI/Total FDI. For economies reporting non-zero Phantom FDI to the OECD, Real FDI/Total FDI corresponds to the reported values (green dots). For non-reporting economies, Real FDI/Total FDI corresponds to predicted values (blue dots).

Citation: IMF Working Papers 2019, 274; 10.5089/9781513521527.001.A001

• Download Figure
• Download figure as PowerPoint slide

Next, we use the estimated coefficients from Equation (2) to predict the ratio of Real to Total FDI for all other economies in the world. Figure 2 (lower panel) shows the actual ratios of Real FDI to Total FDI for the 16 reporting economies (green) and predicted ratios for the remaining economies (blue), plotted against the ratio of Total FDI to GDP.¹⁸ The figure highlights that the univariate model fits the observations for the reporting economies well and illustrates the implied functional form: the ratio of Real FDI to Total FDI approaches zero as Total FDI to GDP approaches infinity. The predicted ratio of Real FDI to Total FDI is lowest for tiny offshore centers such as the Cayman Islands, Bermuda and Gibraltar.

We probe the robustness of the estimation results in several ways. First, one may be concerned that the strong correlation in the univariate model is spurious and reflects a correlation between the dependent variable and variables outside the model. We document that adding controls has remarkably little influence on the estimates. As shown in Table 2, the coefficient on the ratio of Total FDI to GDP barely changes when the model is augmented with variables capturing market size (Column 2), economic development (Column 3) and openness (Column 4). This suggests that the strong correlation in the univariate model reflects a genuine economic mechanism that can be expected to hold beyond the economies in the estimating sample. Second, the small estimating sample raises the concern that the results are sensitive to observations from a single economy. We address this concern by conducting a simple exercise where we re-estimate the univariate model removing one economy at a time from the estimating sample. Figure 3 (upper panel) illustrates the 16 regression lines while highlighting the estimates that result from omitting each of the three outliers, Hungary, the Netherlands and Luxembourg. The results are almost indistinguishable across samples except that the regression line becomes noticeably flatter when Luxembourg is omitted. This suggests that extrapolating from the baseline results may be misleading if Luxembourg has a markedly different ratio of Real FDI to Total FDI than other economies with a similar ratio of Total FDI to GDP. Third, the baseline model pools the years 2013–2017, which could potentially conceal large differences across years. Figure 3 (lower panel) shows the five regression lines resulting from estimating the baseline model for each of the years separately. The estimates are virtually identical across the period 2013–2016 and only slightly different for 2017 suggesting that the estimated relation is stable over time.

Table 2:

Prediction model. The table shows the results from a pooled OLS regression for the period 2013–2017. The sample is the OECD countries reporting inward Total FDI positions into SPEs and non-SPEs separately. The dependent variable is Real FDI / Total FDI (in logs) where Real FDI is inward FDI into non-SPEs. The explanatory variables are: FDI/GDP is the ratio of the inward Total FDI position to GDP; Trade is total foreign trade (imports plus exports) in goods.

Table 2:

Prediction model. The table shows the results from a pooled OLS regression for the period 2013–2017. The sample is the OECD countries reporting inward Total FDI positions into SPEs and non-SPEs separately. The dependent variable is Real FDI / Total FDI (in logs) where Real FDI is inward FDI into non-SPEs. The explanatory variables are: FDI/GDP is the ratio of the inward Total FDI position to GDP; Trade is total foreign trade (imports plus exports) in goods.

	(1)	(2)	(3)	(4)	(5)
	Real FDI / Total FDI (in logs)
FDI / GDP (in logs)	-0.5044*** (0.0357)	-0.5013*** (0.0346)	-0.4884*** (0.0339)	-0.5217*** (0.0338)	-0.5059*** (0.0355)
GDP (in logs)		0.0105 (0.0274)			0.0094 (0.0227)
GDP / capita (in logs)			-0.0934* (0.0474)		-0.0573 (0.0598)
Trade / GDP (in logs)				0.1943** (0.0820)	0.1581 (0.0988)
Constant	-0.4431*** (0.0344)	-0.5743* (0.3390)	0.5389 (0.4973)	-0.3948*** (0.0398)	0.0815 (0.7091)
Observations	75	75	75	75	75
R-squared	0.8449	0.8452	0.8513	0.8561	0.8581

View Table

Table 2:

Prediction model. The table shows the results from a pooled OLS regression for the period 2013–2017. The sample is the OECD countries reporting inward Total FDI positions into SPEs and non-SPEs separately. The dependent variable is Real FDI / Total FDI (in logs) where Real FDI is inward FDI into non-SPEs. The explanatory variables are: FDI/GDP is the ratio of the inward Total FDI position to GDP; Trade is total foreign trade (imports plus exports) in goods.

	(1)	(2)	(3)	(4)	(5)
	Real FDI / Total FDI (in logs)
FDI / GDP (in logs)	-0.5044*** (0.0357)	-0.5013*** (0.0346)	-0.4884*** (0.0339)	-0.5217*** (0.0338)	-0.5059*** (0.0355)
GDP (in logs)		0.0105 (0.0274)			0.0094 (0.0227)
GDP / capita (in logs)			-0.0934* (0.0474)		-0.0573 (0.0598)
Trade / GDP (in logs)				0.1943** (0.0820)	0.1581 (0.0988)
Constant	-0.4431*** (0.0344)	-0.5743* (0.3390)	0.5389 (0.4973)	-0.3948*** (0.0398)	0.0815 (0.7091)
Observations	75	75	75	75	75
R-squared	0.8449	0.8452	0.8513	0.8561	0.8581

View Full Size

Model sensitivity. Top panel: the figure illustrates the sensitivity of the estimated univariate relation between Total FDI / GDP and Real FDI / Total FDI to observations from individual economies. Each regression line corresponds to a regression where one economy is removed from the sample. The figure highlights the regression lines corresponding to regressions where Hungary (green), Netherlands (blue) and Luxembourg (red) are removed. Bottom Panel: the figure illustrates the sensitivity of the estimated univariate relation between Total FDI / GDP and Real FDI / Total FDI to sample period. Each regression line corresponds to a regression with observations from only one year.

Citation: IMF Working Papers 2019, 274; 10.5089/9781513521527.001.A001

• Download Figure
• Download figure as PowerPoint slide

Figure 4 (upper panel) summarizes the sensitivity analysis by plotting the baseline predictions against the predictions that result from augmenting the model with a full set of controls (blue), excluding observations for Luxembourg (red) and using only observations from 2017 (green). Adding controls occasionally changes the ranking of economies with a similar ratio of Total FDI to GDP relative to the baseline, but does not systematically change the predicted ratio of Real FDI to Total FDI across the distribution. Excluding observations for Luxembourg (mechanically) retains the ranking of economies, but raises (lowers) the predicted ratio of Real FDI to Total FDI for economies with high (low) ratios of Total FDI to GDP. Relying only on observations from 2017 has the opposite effect although somewhat less pronounced: it slightly lowers (raises) the predicted ratio of Real FDI to Total FDI for economies with a high (low) ratio of Total FDI to GDP.

View Full Size

Sensitivity of predictions. Top panel: the figure shows how the predicted values of Real FDI/Total FDI depend on the model used to make the predictions. The figure plots Real FDI/Total FDI predicted by the baseline univariate model on the x-axis against Real FDI/Total FDI predicted by the multivariate model with a full set of controls (blue dots), the univariate model excluding observations for Luxembourg from the estimation sample (red dots) and the univariate model using only observations from 2017 (green dots) on the y-axis. Bottom panel: the figure shows the estimated Real FDI/Total FDI by economy. Economies where the ratio is unity are not shown. The figure distinguishes between economies belonging to a broad set of offshore financial centers (red) and other economies (blue).

Citation: IMF Working Papers 2019, 274; 10.5089/9781513521527.001.A001

• Download Figure
• Download figure as PowerPoint slide

Finally, we use the baseline predictions to decompose inward FDI positions into Real FDI and Phantom FDI for economies where this decomposition is not directly available in reported statistics. For a given economy, we apply the same prediction uniformly to all counterpart economies. For instance, we predict that Total FDI into the Cayman Islands is 95 percent Phantom FDI and 5 percent Real FDI. Applying these shares to our estimate of Total FDI into the Cayman Islands from Canada of $34 billion, we set Phantom FDI at $32.5 billion and Real FDI at $1.5 billion. Applying the same ratios to FDI into the Cayman Islands from other economies, we decompose the entire inward FDI position of the Cayman Islands. We proceed similarly to decompose the inward FDI positions of other economies.¹⁹ When no information on GDP is available, we are unable to calculate the ratio of Total FDI to GDP and therefore cannot predict the ratio of Real to Total FDI. In these rare cases, we apply the ratio corresponding to the simple average across the reporting OECD economies (i.e. 78 percent of Total FDI is Real FDI). Figure 4 (lower panel) shows the estimated share of Real FDI in Total FDI for all economies where this share is not unity. We distinguish between a broad set of offshore financial centers (red bars) and other economies (blue bars).²⁰ Strikingly, almost all the economies with the lowest ratio are offshore financial centers and largely overlap with the list of sink and conduit offshore financial centers identified in Garcia-Bernardo et al. (2017) although the methodologies are quite different.

2.4 Real FDI by economy of ultimate investor

The final step in the estimation of the FDI database is to break down inward Real FDI by the economy of the ultimate investors. This completes the attempt to look through chains of Phantom FDI and capture the genuine cross-border investment links: from real investment in productive assets in one economy to the investors who ultimately control the assets in another economy.

2.4.2 Orbis database

For all other economies, we start from the estimates of Real FDI by immediate investor economy obtained in the previous step and apply a conversion factor δ_hi obtained from the global firm database Orbis.²² The conversion factor expresses, for a given host economy h and a given investor economy i, the estimated ratio between real investment in h by ultimate investors in i and real investment in h by immediate investors in i. Intuitively, the conversion factor captures how much a given counterpart economy is used as a conduit for investments originating from other economies (which decreases δ_hi) relative to how much it uses other economies as conduits for its own foreign investments (which increases δ_hi). With these conversion factors, Real FDI by ultimate investor economy follows from the following simple formula:

$\begin{array}{cc}\mathrm{Re}alFD{I}_{hi}^{UIE}={\delta }_{hi}\mathrm{Re}alFD{I}_{hi}^{IIE}& \left(3\right)\end{array}$

To estimate the conversion factors, we rely on information from Orbis. For each corporation in the database owned by foreign investors, we first approximate the value of the investment and then aggregate to obtain economy-level measures of FDI by immediate and ultimate investor economy.

First, we construct the sample of corporate entities in Orbis receiving Real FDI. Specifically, we extract accounting and ownership information for all corporations where at least one immediate investor with an ownership share of at least 10 percent is located in a foreign economy. The 10 percent threshold is conventionally used to delineate direct investment and portfolio investment in external statistics. To focus on Real FDI, we drop corporations with characteristics resembling SPEs: at most five employees and assets per employee exceeding $10 million.²³

Second, we estimate the value of the shareholders’ investment in each of the corporations. Conceptually, FDI can take the form of equity or debt. We measure equity investments in a corporation as shareholders’ capital. We cannot directly infer the value of investments by the owners in the form of debt in Orbis since the database does not distinguish between external debt (from banks and other unrelated investors) and internal debt (from affiliates). We overcome this difficulty by assuming that the ratio of internal debt to total debt is 0.2 in all non-financial corporations and that internal debt comes from the owners in proportion to their shareholdings in the corporation. The ratio 0.2 is calibrated so that our aggregates roughly match the ratio between FDI investment in the form of debt and equity in official FDI statistics, which is around 0.25. Hence, for each non-financial corporation we estimate the value of the shareholders’ investment as the sum of shareholders’ capital and 20 percent of the total debt. For financial corporations, we only include shareholders’ capital as most debt between affiliated financial corporations is not classified as FDI according to macroeconomic statistical manuals.

Third, using the ownership information from Orbis, we define ${\theta }_c^i$ as the share of corporation c controlled by ultimate investors in economy i and ${\lambda }_c^h$ as the share of corporation c owned by immediate investors in economy i. With these definitions, we can estimate Orbis-based measures of Real FDI by the economy of the immediate and ultimate owners respectively:

$\begin{array}{c}\mathrm{Re}alOrbisFD{I}_{hi}^{UIE}=\underset{c\in ℍ}{\mathrm{\Sigma }}{\theta }_c^{i}investmen{t}_c\\ \mathrm{Re}alOrbisFD{I}_{hi}^{IIE}=\underset{c\in ℍ}{\mathrm{\Sigma }}{\lambda }_c^{i}invest{m}_c\end{array}$

where ℍ is the set of corporations in economy h in our Orbis sample and investment_c is the estimated value of the shareholders’ investment in corporation c.

To provide a sense of the overall comparability of the firm data in Orbis and official FDI statistics, we note that aggregating the Orbis-based FDI measures across all economies yields a measure of global Real FDI of around $8.4 trillion in 2016. This compares to the estimate of global Real FDI of around $22 trillion that emerges from our analysis of official FDI statistics in the previous subsection. Presumably, most of the discrepancy is due to the incomplete coverage of the Orbis database; however, it may also partly derive from differences in valuation principles: our Orbis-based measures of investment use book values of equity and debt from firms’ financial reports whereas FDI statistics use market values whenever possible.²⁴

Finally, we use the Orbis-based measures of FDI to produce the conversion factors δ_hi. For pairs of economies where the Orbis data deviate less than 50 percent from our estimate of Real FDI and where we therefore expect ownership links in Orbis to be reasonably representative of all FDI, we compute conversion factors that are specific to the pair:²⁵

$\begin{array}{cc}{\delta }_{hi}=\frac{\mathrm{Re}alOrbisFD{I}_{hi}^{UIE}}{\mathrm{Re}alOrbisFD{I}_{hi}^{IIE}}if|\frac{\mathrm{Re}alOrbisFD{I}_{hi}}{\mathrm{Re}alFD{I}_{hi}}-1|<0.5andh\ne i& \left(4\right)\end{array}$

The conversion factor simply expresses the ratio, as observed in Orbis, between Real FDI in h from ultimate and immediate owners in i respectively. Multiplying the conversion factor on Real FDI in h by immediate owners in i, as estimated from the official FDI statistics, yields an estimate of Real FDI in h by ultimate owners in i. The implicit assumption is that the ratio between Real FDI in h by ultimate and immediate owners in i is the same for investment covered and not covered in Orbis. We will probe this assumption below.²⁶

We illustrate the mechanics with a concrete example. We estimated above, based on official FDI statistics, that Spain’s Real FDI from immediate owners in the United States is around $23 billion. We observe around $13 billion of this investment in Orbis; hence, the coverage ratio exceeds 50 percent and we proceed to compute a pair-specific conversion factor. The Orbis sample includes investment in Spain from ultimate owners in the United States of around $29 billion; hence, Equation (4) yields a conversion factor of 2.2 (i.e. $29 billion/$13 billion). Applying Equation (3), we arrive at an estimate of Real FDI in Spain from ultimate owners in the United States of just over $50 billion (i.e. $23 billion*2.2).

We validate the approach by comparing the conversion factors δ_hi emerging from Orbis with the actual ratio between Real FDI in h by ultimate owners in i and Real FDI in h by immediate owners in i for pairs of economies where both are available: when the host economy belongs to the OECD countries reporting Real FDI by ultimate investor economy and Orbis has sufficient coverage. Figure 5 (upper panel) plots the ranks of the two variables. The rank-rank correlation when observations are weighted with Real FDI is 0.72, suggesting that our approach creates sensible measures of FDI by ultimate investor economies in the aggregate. The unweighted correlation is 0.43 (not shown) suggesting that the approach yields larger errors for pairs of economies with lower levels of bilateral FDI.

View Full Size

Validation of Orbis approach. The figure illustrates the results from a test of our approach to estimating FDI by ultimate investor economy using micro-data from Orbis. For each pair of economies where the value of real investments in Orbis accounts for at least 50 percent of the Real FDI position and where the host country self-reports Real FDI by ultimate investor economy the figure plots the conversion factor computed with Orbis data (i.e. the ratio of Real Orbis FDI in h by ultimate investors in i to Real Orbis FDI in h by immediate investors in i as defined in Equation (4)) against the equivalent ratio in self-reported FDI data. Observations are weighted by the value of Real FDI. The regression line illustrates the parameters of the (weighted) univariate regression.

Citation: IMF Working Papers 2019, 274; 10.5089/9781513521527.001.A001

• Download Figure
• Download figure as PowerPoint slide

For pairs of economies where Orbis data deviates more than 50 percent from Real FDI, it is an important concern that ownership links in the Orbis sample may not be representative of foreign investment more broadly. To address this concern, we do not use conversion factors that are specific to the pairs of economies when coverage is low, but rather use information from all host economies to compute conversion factors that are specific to the investor economy (and uniform across host economies with low coverage):²⁷

$\begin{array}{cc}{\delta }_{hi}=\frac{{\mathrm{\Sigma }}_{h\in 𝕊}\mathrm{Re}alOrbisFD{I}_{hi}^{UIE}}{{\mathrm{\Sigma }}_{h\in 𝕊}\mathrm{Re}alOrbisFD{I}_{hi}^{IIE}}if|\frac{\mathrm{Re}alOrbisFD{I}_{hi}}{\mathrm{Re}alFD{I}_{hi}}-1|>0.5andh\ne i& \left(5\right)\end{array}$

where 𝕊 is the set of host economies in Orbis. The conversion factor simply expresses the ratio, as observed in Orbis, between global Real FDI from ultimate and immediate investors in i respectively. Again, the conversion factor is applied to Real FDI in h by immediate investors in i, as estimated from the official FDI statistics, to obtain an estimate of Real FDI in h by ultimate investors in h.

Once again, we illustrate the mechanics with a concrete example. We estimated above, based on official FDI statistics, that Russia’s Real FDI from immediate owners in the Netherlands is around $41 billion. We observe around $19 billion of this investment in Orbis. Hence, the coverage ratio falls below 50 percent and we therefore do not apply a conversion factor specific to the pair Russia-Netherlands, but rather one that is specific to the investor economy the Netherlands. In Orbis, we observe global Real FDI from ultimate investors in the Netherlands of around $294 billion and global Real FDI from immediate investors in the Netherlands of around $597 billion. Hence, Equation (5) yields a conversion factor of around 0.5 (i.e. $294 billion/$597 billion). Applying Equation (3), we arrive at an estimate of Real FDI in Russia from ultimate owners in the Netherlands of $21 billion (i.e. $41 billion*0.5).²⁸

2.5 Summary statistics

We summarize the data sources ultimately used to estimate the database in Table 3. Around 90 percent of the inward Total FDI in the database is reported to the CDIS while the remaining 10 percent is inferred from mirror data or based on our own estimates (first panel). The decomposition into Phantom FDI and Real FDI relies much more on our own estimates: around 50 percent of the global Phantom FDI and 80 percent of Real FDI is estimated from the prediction model. Finally, around one quarter of the allocation of Real FDI to ultimate investor economies derives from reports to the OECD FDI statistics and three quarters are estimated by applying conversion factors based on Orbis data.³³

Table 3:

Summary statistics (2017). The table provides an overview of the data sources ultimately used to construct the database. The first panel breaks down observations of inward Total FDI by immediate investor economy (IIE) into pairs where the value is reported to CDIS and pairs where the value is inferred from mirror data or based on our estimates. The second and third panels break down observations of inward Phantom FDI and inward Real FDI by immediate investor economy (IIE) into pairs where the value is reported to the OECD and pairs where the value is estimated from the prediction model. The fourth panel breaks down observations of inward Real FDI by ultimate investor economy (UIE) into pairs where the value is reported to the OECD and pairs where the value is estimated from the prediction model. In all cases, we report both the number of economy pairs and aggregate inward FDI (in $billion).

Table 3:

Summary statistics (2017). The table provides an overview of the data sources ultimately used to construct the database. The first panel breaks down observations of inward Total FDI by immediate investor economy (IIE) into pairs where the value is reported to CDIS and pairs where the value is inferred from mirror data or based on our estimates. The second and third panels break down observations of inward Phantom FDI and inward Real FDI by immediate investor economy (IIE) into pairs where the value is reported to the OECD and pairs where the value is estimated from the prediction model. The fourth panel breaks down observations of inward Real FDI by ultimate investor economy (UIE) into pairs where the value is reported to the OECD and pairs where the value is estimated from the prediction model. In all cases, we report both the number of economy pairs and aggregate inward FDI (in $billion).

	Pairs of economies	Inward FDI
Inward Total FDI by IIE	23,062	38,566
- Reported to CDIS	16,230	34,563
- Mirror data	6,693	3,435
- Own estimate	139	568
Inward Phantom FDI by IIE	23,062	14,653
- Reported to OECD	2,393	7,436
- Estimated	20,669	7,217
Inward Real FDI by IIE	23,062	23,912
- Reported to OECD	2,393	4,610
- Estimated	20,669	19,302
Inward Real FDI by UIE	23,032	23,913
- Reported to OECD	1,598	5,220
- Detailed Orbis	625	4,296
- Aggregate Orbis	14,128	12,965
- No adjustment	6,681	1,432

View Table

Table 3:

Summary statistics (2017). The table provides an overview of the data sources ultimately used to construct the database. The first panel breaks down observations of inward Total FDI by immediate investor economy (IIE) into pairs where the value is reported to CDIS and pairs where the value is inferred from mirror data or based on our estimates. The second and third panels break down observations of inward Phantom FDI and inward Real FDI by immediate investor economy (IIE) into pairs where the value is reported to the OECD and pairs where the value is estimated from the prediction model. The fourth panel breaks down observations of inward Real FDI by ultimate investor economy (UIE) into pairs where the value is reported to the OECD and pairs where the value is estimated from the prediction model. In all cases, we report both the number of economy pairs and aggregate inward FDI (in $billion).

	Pairs of economies	Inward FDI
Inward Total FDI by IIE	23,062	38,566
- Reported to CDIS	16,230	34,563
- Mirror data	6,693	3,435
- Own estimate	139	568
Inward Phantom FDI by IIE	23,062	14,653
- Reported to OECD	2,393	7,436
- Estimated	20,669	7,217
Inward Real FDI by IIE	23,062	23,912
- Reported to OECD	2,393	4,610
- Estimated	20,669	19,302
Inward Real FDI by UIE	23,032	23,913
- Reported to OECD	1,598	5,220
- Detailed Orbis	625	4,296
- Aggregate Orbis	14,128	12,965
- No adjustment	6,681	1,432

3.1 The nature of global FDI

We first aggregate inward FDI positions across all economies in the world and show, in Figure 6, Total FDI at the global level for the period 2009–2017 with a breakdown on Real FDI (blue bars) and Phantom FDI (red bars). In 2017, global FDI was almost $40 trillion with Real FDI accounting for around $25 trillion and Phantom FDI accounting for around $15 trillion. Over the period 2009–2017, Total FDI increased by around 70 percent reflecting that Phantom FDI almost doubled whereas Real FDI increased by a more modest 50 percent. The share of Phantom FDI in Total FDI increased from just above 30 percent in 2009 to just below 40 percent in 2017.³⁴

We conduct a number of sensitivity tests of the estimate of global Phantom FDI in Figure 7 (upper panel). First, we estimate Phantom FDI for economies that do not self-report FDI in SPEs and non-SPEs separately based on a model that does not include Luxembourg (green line). Within the sample of reporting economies, Luxembourg exhibited by far the largest share of Phantom FDI in Total FDI and we noted in Section 2.3 that the parameters of the prediction model were quite sensitive to whether Luxembourg was included or not. Our estimates of global Phantom FDI, however, are very robust to excluding Luxembourg from the prediction model.³⁵ Second, we aggregate Phantom FDI across the economies that self-report FDI in SPEs only (blue line). Only relying on self-reported statistics effectively makes the extreme assumption that Phantom FDI is zero in all economies that do not self-report FDI in SPEs (e.g. the British Virgin Islands, Bermuda and the Cayman Islands) and thus provides a lower bound. Even in this extreme scenario, Phantom FDI amounted to around 7.4 trillion globally in 2017. Third, we adopt an ad hoc approach to approximating Phantom FDI in economies that do not self-report FDI in SPEs by assuming that Phantom FDI constitutes 90 percent of Total FDI in offshore centers and 0 percent in other economies (red line). This yields an estimate of global Phantom FDI of around $20 trillion in 2017, considerably higher than the baseline estimate of around $15 trillion.

View Full Size

Global Real and Phantom FDI. The figure shows the evolution of inward FDI positions aggregated over all economies in the world over the period 2009–2017. Total FDI is decomposed into Real FDI (blue bars) and Phantom FDI (red bars).

Citation: IMF Working Papers 2019, 274; 10.5089/9781513521527.001.A001

• Download Figure
• Download figure as PowerPoint slide

View Full Size

Phantom FDI Top panel: the figure compares our baseline estimates of global Phantom FDI over the period 2009–2017 (black line) to three alternative scenarios: assuming zero Phantom FDI in economies that do not self-report FDI in SPEs and non-SPEs separately so that the Phantom FDI estimate only reflects self-reported FDI in SPEs (blue line); estimating the share of Phantom FDI in Total FDI based on a model that does not include Luxembourg (green line); assuming that Phantom FDI in non-reporting economies constitutes 90 percent of Total FDI in offshore centers and that Phantom FDI is zero in other economies. Bottom panel: the figure shows inward Phantom FDI positions in 2017 for the 20 economies in the world with the largest such positions. For OECD economies reporting these statistics, the figure shows the reported numbers. For non-reporting economies, the figure shows our estimates (explained in the main text).

Citation: IMF Working Papers 2019, 274; 10.5089/9781513521527.001.A001

• Download Figure
• Download figure as PowerPoint slide

While the sensitivity analysis highlights that the incomplete coverage of reported data on FDI in SPEs creates significant uncertainty about the true magnitude of Phantom FDI at the global level, it also illustrates that we do not necessarily err on the high side. On the one hand, if one believes that all of the FDI in non-reporting offshore centers like the British Virgin Islands, Bermuda and the Cayman Islands represent real investment with substantial links to the local economies, one may think that $7.4 trillion, or around 20 percent of Total FDI, is the most sensible global estimate of Phantom FDI. One the other hand, if one holds the view that the vast majority of the FDI positions of offshore centers reflect pass-through investment, $20 trillion, or around 50 percent of Total FDI, may be a more accurate number than our baseline estimate of $15 trillion.

Finally, we note that our estimates of Phantom FDI are sensitive to the particular definition of SPEs applied by national statistical authorities. Since there is currently no strictly harmonized international SPE definition, national compilers have some measure of flexibility (as described in Section 2.1). To illustrate the issue, Ireland does not currently report a breakdown of Total FDI into SPEs and non-SPEs in official OECD statistics, but a recent publication from Ireland’s Central Statistics Office (2019) estimates that FDI into SPEs only accounts for 5 percent of Irish FDI. By contrast, Galstyan (2019) estimates that distortions to Irish FDI – primarily from redomiciled public limited companies, IP transfers and aircraft leasing – account for 47 percent of Ireland’s inward FDI, which is roughly consistent with our estimate that 62 percent of Total FDI into Ireland is in SPEs. The low estimate produced by the Central Statistics Office (2019) is based on the SPE definition recently proposed by the international statistical community (IMF, 2018). This definition is more restrictive than the description in the current manuals, for instance by applying a cap of five employees, and therefore categorizes fewer entities as SPEs. In the case of Switzerland, our model generates an SPE share of 50 percent compared to only 13 percent in official Swiss statistics. In this context, the Swiss National Bank (2017) underscores that a narrow definition of SPEs leads to underestimation of pass-through FDI as significant FDI will pass through entities that are categorized as non-SPEs.

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

We describe how the $15 trillion of Phantom FDI is distributed across economies in Figure 7 (lower panel). The global leader is Luxembourg with a Phantom FDI position of around $3.8 trillion, closely followed by the Netherlands with around $3.3 trillion. These positions are self-reported by the national statistical authorities (indicated with dark red bars) and therefore not subject to the uncertainties associated with our estimation method. The following five economies are known offshore financial centers – Hong Kong SAR, the British Virgin Islands, Bermuda, Singapore and the Cayman Islands – each accounting for Phantom FDI around $0.5–1 trillion. These positions are estimated using the method developed above (indicated with light red bars) and therefore subject to some uncertainty.

Next, we turn to the distribution of the global $25 trillion of Real FDI across economies in Figure 8 (upper panel). We plot both Real FDI hosted in the economy (blue bars) and Real FDI ultimately controlled by investors in the economy (green bars) for the 20 economies with the largest inward Real FDI position. The United States has the largest inward and outward positions; it hosts Real FDI of around $4 trillion and ultimately owns Real FDI of around $3.6 trillion. Many large economies have highly unbalanced Real FDI positions: China’s inward position exceeds its outward position by around $1.2 trillion whereas the three largest European economies (i.e. Germany, the United Kingdom and France) have a joint outward position that exceeds the joint inward position by around $1.5 trillion. A handful of the economies with the largest Real FDI positions also appeared among the economies with the largest Phantom FDI positions. It is noteworthy that at least some of these economies (e.g. the Netherlands, Singapore and Hong Kong SAR) have strikingly balanced Real FDI positions, which is suggestive that the Real FDI estimates for these economies include an important element of pass-through investment. Possibly, some pass-through investment is handled by corporate entities that also have real activities, in which case the entities are categorized as non-SPEs and the investment as real.

View Full Size

Real FDI and ultimate ownership. Top panel: the figure shows inward Real FDI positions in 2017 for the 20 economies in the world with the largest such positions (blue bars). For OECD economies reporting these statistics, the figure shows the reported numbers. For non-reporting economies, the figure shows our estimates (explained in the main text). For each economy, the figure also shows the outward Real FDI position, i.e. sum of Real FDI in foreign economies ultimately owned by investors in the economy. Bottom Panel: the figure shows, for each of the 20 largest economies in the world with the largest outward FDI positions, the ratio between ultimately owned Real FDI and immediately owned Total FDI.

Citation: IMF Working Papers 2019, 274; 10.5089/9781513521527.001.A001

• Download Figure
• Download figure as PowerPoint slide

We highlight the important distinction between ultimate and immediate ownership in Figure 8 (lower panel). For each of the 20 largest investor economies in the world, the figure shows the ratio between ultimately owned Real FDI, the measure of an economy’s real aggregrate outward FDI position emerging from our analysis, and immediately owned Total FDI, the measure of the aggregate outward FDI position in standard FDI statistics. Generally, the estimated ratios are close to unity for large developed economies such as Italy, France, Spain, Germany and Japan, reflecting that standard FDI statistics provide an accurate picture of their aggregate outward FDI position. In the United States and the United Kingdom, the estimated ratios are below unity reflecting that standard FDI statistics overstate their aggregate outward FDI position. A similar pattern emerges, but much more pronounced, for offshore financial centers like Luxembourg, Hong Kong SAR, Bermuda and the British Virgin Islands where much of the outward FDI is ultimately owned by investors in other economies. The opposite holds for China where an estimated ratio above two suggests that Chinese investors are the ultimate owners of more than twice as much FDI as indicated by their outward FDI position in standard FDI statistics.

Finally, we illustrate the economic significance of round-tripping in Figure 9 (upper panel). For each of the 20 largest economies in the world, the figure shows the share of real inward FDI ultimately owned by investors in the economy itself. For some economies, the round-tripping share is based on FDI data by ultimate investor economies reported by the national statistical authorities (indicated with dark red bars). For others, it is based on the Orbis database and therefore subject to more uncertainty (indicated with light red bars). There is considerable variation in the significance of round-tripping across economies. In transition economies like China and Russia, around 25 percent of Real FDI is ultimately owned by domestic investors whereas the share is far below 10 percent, and sometimes completely negligible, in most developed economies.

View Full Size

Round-tripping and Real FDI links. Top panel: the figure shows the round-tripping factors (i.e. the share of inward Real FDI in economy h ultimately owned by investors in economy h) for the 20 largest economies in the world (by GDP). Bottom panel: the figure also shows for each combination of the five largest economies in the world by GDP, United States (USA), China (CHN), Japan (JPN), Germany (DEU) and United Kingdom (GBR), the estimated Real FDI position capturing links between ultimate investors and real investments (blue bars) and the FDI position observed in standard FDI statistics (green bars).

Citation: IMF Working Papers 2019, 274; 10.5089/9781513521527.001.A001

• Download Figure
• Download figure as PowerPoint slide

3.3 Real bilateral investment links

Standard FDI statistics may misrepresent the strength of cross-border investment links by confounding real and phantom investment and by defining investment ties in terms of immediate rather than ultimate ownership. We illustrate this point in Figure 9 (lower panel) for the five largest economies in world: the United States, China, Japan, Germany and the United Kingdom. For each pair of economies, we plot both our measure of the real link, the value of Real FDI in one economy ultimately owned by investors in the other economy (blue bars), and the measure of the FDI link available in standard statistics, the value of Total FDI in one economy immediately owned by investors in the other economy (green bars). For almost all of the 20 pairs, the real FDI link is underestimated in standard FDI statistics and sometimes very substantially so. For instance, FDI in Germany from the United States is around $90 billion according to CDIS whereas our analysis suggests a Real FDI link of almost $170 billion. The likely explanation is that a large share of the direct investments made by U.S. multinationals in Germany are channeled through economies like Luxembourg and the Netherlands, which conceals the real strength of the investment links.

We make this point more generally by showing how gravity factors shape different types of bilateral FDI positions in Table 4. We first estimate the following simple gravity model with OLS:

where X is a vector of dummies including contiguity, common language, colonial links etc. and where h denotes the host economy, i the investor economy and t the year. The model yields elasticities of Total FDI with respect to the GDPs of the host economy and the immediate investor economy far below unity, suggesting that the relative importance of foreign investment decreases quickly with market sizes, and an R-squared of 0.266 (Column 1). Estimating the same model for Phantom FDI and Real FDI separately yields very different results. Most strikingly, the elasticity of Phantom FDI with respect to host economy GDP is negative, suggesting that smaller economies attract more Phantom FDI in absolute terms, whereas the elasticity of Real FDI with respect to host economy GDP is positive and somewhat larger than for Total FDI (Columns 2–3). Finally, including the characteristics of the ultimate investor economy rather than the immediate investor economy yields larger coefficients than the baseline model on real variables such as market size and distance and an R-squared of 0.334 (Column 4).

Table 4:

Gravity models. The table shows the results from pooled OLS regressions (Columns 1–4) and pooled Poisson regressions (Columns 4–8) for the period 2013–2017. The dependent variable is a measure of inward FDI: Total FDI (Columns 1 and 5); Phantom FDI (Columns 2 and 6); and Real FDI (Columns 3–4 and 7–8). The explanatory variables are: Host economy GDP is the GDP of the economy hosting the investment; Immediate investor economy GDP is the GDP of the economy of the immediate investor; Ultimate investor economy GDP is the GDP of the economy of the ultimate investor; Bilateral distance is the Euclidean distance between the economy of the investor and the economy of the investment; Contiguity is a dummy indicating whether the economy of the investment and the economy of the investor share a border; Common language is a dummy indicating whether the economy of the investment and the economy of the investor share an official language; Former colony is a dummy indicating whether the economy of the investment was formerly a colony of the economy of the investor or vice versa; Former same country is a dummy indicating whether the economy of the investment and the economy of the investor were formerly part of the same country. For the OECD economies reporting inward Total FDI positions into SPEs and non-SPEs separately, Phantom FDI is equal to the former and Real FDI to the latter. For other economies, Phantom FDI and Real FDI are estimated (explained above).

Table 4:

Gravity models. The table shows the results from pooled OLS regressions (Columns 1–4) and pooled Poisson regressions (Columns 4–8) for the period 2013–2017. The dependent variable is a measure of inward FDI: Total FDI (Columns 1 and 5); Phantom FDI (Columns 2 and 6); and Real FDI (Columns 3–4 and 7–8). The explanatory variables are: Host economy GDP is the GDP of the economy hosting the investment; Immediate investor economy GDP is the GDP of the economy of the immediate investor; Ultimate investor economy GDP is the GDP of the economy of the ultimate investor; Bilateral distance is the Euclidean distance between the economy of the investor and the economy of the investment; Contiguity is a dummy indicating whether the economy of the investment and the economy of the investor share a border; Common language is a dummy indicating whether the economy of the investment and the economy of the investor share an official language; Former colony is a dummy indicating whether the economy of the investment was formerly a colony of the economy of the investor or vice versa; Former same country is a dummy indicating whether the economy of the investment and the economy of the investor were formerly part of the same country. For the OECD economies reporting inward Total FDI positions into SPEs and non-SPEs separately, Phantom FDI is equal to the former and Real FDI to the latter. For other economies, Phantom FDI and Real FDI are estimated (explained above).

	(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)
	OLS pooled cross-section				Poisson pooled cross-section
	All FDI	Phantom FDI	Real FDI	Real FDI	All FDI	Phantom FDI	Real FDI	Real FDI
Host economy GDP	0.524*** (0.00895)	-0.119*** (0.0262)	0.615*** (0.00863)	0.665*** (0.00862)	0.528*** (4.40e-05)	0.144*** (6.58e-05)	0.821*** (6.45e-05)	0.813*** (6.73e-05)
Immediate investor economy GDP	0.652*** (0.00835)	0.718*** (0.0218)	0.667*** (0.00806)		0.561*** (4.66e-05)	0.552*** (7.87e-05)	0.575*** (5.93e-05)
Ultimate investor economy GDP				0.699*** (0.00807)				0.774*** (6.85e-05)
Bilateral distance (in logs)	-0.824*** (0.0230)	-0.728*** (0.0577)	-0.877*** (0.0222)	-0.857*** (0.0220)	-0.678*** (8.98e-05)	-0.785*** (0.000145)	-0.662*** (0.000117)	-0.623*** (0.000125)
Contiguity	0.153 (0.107)	0.437 (0.309)	0.191* (0.103)	0.277*** (0.101)	-0.575*** (0.000318)	-1.771*** (0.000668)	-0.0999*** (0.000372)	-0.156*** (0.000408)
Common language (official)	1.979*** (0.0552)	3.847*** (0.150)	1.900*** (0.0532)	1.833*** (0.0525)	1.122*** (0.000207)	1.129*** (0.000327)	0.967*** (0.000267)	1.001*** (0.000285)
Former colony	1.300*** (0.113)	1.238*** (0.299)	1.338*** (0.109)	1.291*** (0.107)	0.492*** (0.000287)	0.464*** (0.000500)	0.534*** (0.000352)	0.498*** (0.000363)
Formerly same country	-0.102 (0.152)	0.147 (0.410)	0.0143 (0.147)	-0.0187 (0.144)	1.321*** (0.000327)	1.710*** (0.000565)	1.178*** (0.000395)	0.855*** (0.000499)
Constant	-4.116*** (0.240)	-1.208* (0.650)	-5.167*** (0.231)	-6.293*** (0.229)	-0.533*** (0.00106)	4.280*** (0.00171)	-5.404*** (0.00146)	-8.408*** (0.00161)
Observations	30,383	13,536	30,369	29,575	75,361	75,361	75,361	74,502
R-squared	0.266	0.140	0.305	0.334	-	-	-	-

View Table

Table 4:

Gravity models. The table shows the results from pooled OLS regressions (Columns 1–4) and pooled Poisson regressions (Columns 4–8) for the period 2013–2017. The dependent variable is a measure of inward FDI: Total FDI (Columns 1 and 5); Phantom FDI (Columns 2 and 6); and Real FDI (Columns 3–4 and 7–8). The explanatory variables are: Host economy GDP is the GDP of the economy hosting the investment; Immediate investor economy GDP is the GDP of the economy of the immediate investor; Ultimate investor economy GDP is the GDP of the economy of the ultimate investor; Bilateral distance is the Euclidean distance between the economy of the investor and the economy of the investment; Contiguity is a dummy indicating whether the economy of the investment and the economy of the investor share a border; Common language is a dummy indicating whether the economy of the investment and the economy of the investor share an official language; Former colony is a dummy indicating whether the economy of the investment was formerly a colony of the economy of the investor or vice versa; Former same country is a dummy indicating whether the economy of the investment and the economy of the investor were formerly part of the same country. For the OECD economies reporting inward Total FDI positions into SPEs and non-SPEs separately, Phantom FDI is equal to the former and Real FDI to the latter. For other economies, Phantom FDI and Real FDI are estimated (explained above).

	(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)
	OLS pooled cross-section				Poisson pooled cross-section
	All FDI	Phantom FDI	Real FDI	Real FDI	All FDI	Phantom FDI	Real FDI	Real FDI
Host economy GDP	0.524*** (0.00895)	-0.119*** (0.0262)	0.615*** (0.00863)	0.665*** (0.00862)	0.528*** (4.40e-05)	0.144*** (6.58e-05)	0.821*** (6.45e-05)	0.813*** (6.73e-05)
Immediate investor economy GDP	0.652*** (0.00835)	0.718*** (0.0218)	0.667*** (0.00806)		0.561*** (4.66e-05)	0.552*** (7.87e-05)	0.575*** (5.93e-05)
Ultimate investor economy GDP				0.699*** (0.00807)				0.774*** (6.85e-05)
Bilateral distance (in logs)	-0.824*** (0.0230)	-0.728*** (0.0577)	-0.877*** (0.0222)	-0.857*** (0.0220)	-0.678*** (8.98e-05)	-0.785*** (0.000145)	-0.662*** (0.000117)	-0.623*** (0.000125)
Contiguity	0.153 (0.107)	0.437 (0.309)	0.191* (0.103)	0.277*** (0.101)	-0.575*** (0.000318)	-1.771*** (0.000668)	-0.0999*** (0.000372)	-0.156*** (0.000408)
Common language (official)	1.979*** (0.0552)	3.847*** (0.150)	1.900*** (0.0532)	1.833*** (0.0525)	1.122*** (0.000207)	1.129*** (0.000327)	0.967*** (0.000267)	1.001*** (0.000285)
Former colony	1.300*** (0.113)	1.238*** (0.299)	1.338*** (0.109)	1.291*** (0.107)	0.492*** (0.000287)	0.464*** (0.000500)	0.534*** (0.000352)	0.498*** (0.000363)
Formerly same country	-0.102 (0.152)	0.147 (0.410)	0.0143 (0.147)	-0.0187 (0.144)	1.321*** (0.000327)	1.710*** (0.000565)	1.178*** (0.000395)	0.855*** (0.000499)
Constant	-4.116*** (0.240)	-1.208* (0.650)	-5.167*** (0.231)	-6.293*** (0.229)	-0.533*** (0.00106)	4.280*** (0.00171)	-5.404*** (0.00146)	-8.408*** (0.00161)
Observations	30,383	13,536	30,369	29,575	75,361	75,361	75,361	74,502
R-squared	0.266	0.140	0.305	0.334	-	-	-	-

Estimating gravity models in log-levels with OLS may lead to biased estimates because observations with zero FDI are dropped (Silva and Tenreyro, 2006). When we estimate the model on its multiplicative form with a Poisson estimator to retain zero observations, we obtain even more striking results: the coefficient on the market size variables increases from around 0.55 when considering Total FDI by immediate investor economy (Column 5) to around 0.80 when considering our estimates of Real FDI by ultimate investor economy (Column 8). This is consistent with the notion that standard FDI statistics systematically underestimate the real investment links between large economies because much of the investment is channeled through small offshore financial centers.

3.4 Exposure to tax avoidance opportunities

Phantom corporations in low-tax economies give multinational firms a number of opportunities to avoid taxes in the high-tax economies where the real investments and the ultimate investors are located. First, phantom corporations may be the notional owners of financial assets (Clifford, 2019), intangible assets (Dischinger and Riedel, 2011) and patents (Karkinsky and Riedel, 2012), which generate income that would otherwise be booked and taxed in affiliates with real activities. Second, they may facilitate the use of hybrid financial instruments (Johannesen, 2014; Hardeck and Wittenstein, 2018), which artificially create deductible interest expenses to offset taxable business income. Third, they allow firms to accumulate and manage profits earned on foreign activities and thus postpone repatriation taxes in the ultimate investor economies (Foley et al., 2007). Fourth, they may serve to reduce or eliminate taxes on cross-border capital flows inside the firm by exploiting tax treaty networks (Weyzig, 2013; Arel-Bundock, 2017) and (Johannesen, 2012).

We attempt to quantify the exposure to these types of tax avoidance opportunities, by estimating, for each economy in the world, the share of its outward FDI where a foreign phantom corporation receives the investment and the share of its inward FDI where a foreign phantom corporation is the investor.³⁶

To estimate exposure to tax avoidance through outward FDI, we often need to assume that, for a given economy, the share of inward Total FDI into phantom corporations is the same for all counterpart economies. For instance, when Luxembourg reports that around 95 percent of inward Total FDI is Phantom FDI, we assume that this share applies to all economies’ outward FDI to Luxembourg. Similarly, when Sweden reports that only around 5 percent of inward Total FDI is Phantom FDI, we assume that this share applies to all economies’ outward FDI to Sweden. Cross-sectional differences in our measure of exposure therefore, to a large extent, reflect differences in the counterpart economies of outward FDI. For instance, we will estimate a higher exposure for economies investing relatively more in Luxembourg than for economies investing relatively more in Sweden.

We illustrate estimated exposure to tax avoidance through outward FDI in Figure 10 (upper panel) for the 20 largest economies in the world (by GDP). The cross-sectional variation is very significant: exposure ranges from less than 10 percent in Spain to more than 80 percent in Brazil. The relatively high estimated exposure for the United States (dark bar) is consistent with the argument that U.S. firms face particularly strong incentives to channel foreign investments through low-tax economies (Altshuler and Grubert, 2005).³⁷

View Full Size

Exposure to tax avoidance opportunities. Top panel: the figure shows for the 20 largest economies in the world (by GDP) the estimated share of outward Total FDI where the immediate investment is a foreign phantom corporation. Bottom Panel: the figure shows for the 20 largest economies in the world (by GDP) the estimated share of inward Total FDI where the immediate investor is a foreign phantom corporation.

Citation: IMF Working Papers 2019, 274; 10.5089/9781513521527.001.A001

• Download Figure
• Download figure as PowerPoint slide

To estimate exposure to tax avoidance through inward FDI, we need to make two types of assumptions. First, for economies not self-reporting FDI in SPEs and non-SPEs separately, we need an estimate of outward Phantom FDI. In self-reporting economies, we observe that the ratio between outward and inward Phantom FDI is close to unity in the aggregate.³⁸ Presumably, this reflects that the vast majority of phantom corporations are “pass-through” entities with inward and outward investment positions matching almost perfectly. We assume that the pattern is the same for non-reporting economies. Therefore, starting from the estimates of inward Phantom FDI obtained in section 3.3, we produce an estimate of outward Phantom FDI by applying the observed ratio between outward and inward Phantom FDI in the aggregate of reporting economies. Second, mirroring the assumption made above for inward FDI, we assume that, for a given economy, the share of outward Total FDI coming from phantom corporations is uniform across counterpart economies. For instance, Luxembourg reports that 95 percent of outward Total FDI is Phantom FDI, and we assume that this share applies uniformly to all economies’ inward FDI from Luxembourg.³⁹

We illustrate estimated exposure to tax avoidance through inward FDI in Figure 10 (lower panel) for the 20 largest economies in the world (by GDP). The cross-sectional variation is somewhat smaller than for exposure through outward FDI: the estimates range from around 15 percent in Australia to around 60 percent in Russia. For the United States (dark bar), the contrast to Figure 9 (upper panel) is striking: while more than half of its outward FDI is channeled through phantom corporations, only a bit more than one quarter of its inward FDI derives from such corporations.

Finally, we estimate how the estimated exposure to tax avoidance correlates with corporate tax rates and levels of economic development in the cross-section of economies in 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). This is consistent with earlier literature highlighting the importance of taxation in the investor economy for the decision to operate through low-tax economies (Desai et al., 2006). The tax gradient is significantly 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). This is consistent with evidence that profit shifting responds more strongly to tax incentives in less developed economies where tax enforcement is weaker (Johannesen et al., 2016). Similar relationships emerge, but somewhat less clearly, for exposure through inward FDI (Columns 3–4).⁴⁰

Table 5:

Exposure to tax avoidance opportunities. The table shows the results from OLS regressions for 2016. The dependent variable is the share of outward Total FDI where the immediate investment is a foreign phantom corporation (Columns 1–2) and the share of inward Total FDI where the immediate investor is a foreign phantom corporation (Columns 3–4). The explanatory variables are the headline corporate tax (Corp tax rate) and GDP per capita (GDP per cap).

Table 5:

Exposure to tax avoidance opportunities. The table shows the results from OLS regressions for 2016. The dependent variable is the share of outward Total FDI where the immediate investment is a foreign phantom corporation (Columns 1–2) and the share of inward Total FDI where the immediate investor is a foreign phantom corporation (Columns 3–4). The explanatory variables are the headline corporate tax (Corp tax rate) and GDP per capita (GDP per cap).

	(1)	(2)	(3)	(4)
	Share of outward FDI to SPE		Share of inward FDI from SPE
Corp tax rate	0.00598** (0.00256)	0.0381** (0.0179)	-8.38e-05 (0.00164)	0.0203* (0.0117)
GDP per cap (in logs)		0.0725 (0.0520)		0.0559 (0.0352)
Corp tax rate × GDP per cap (in logs)		-0.00352* (0.00193)		-0.00226* (0.00128)
Constant	0.145** (0.0633)	-0.519 (0.479)	0.256*** (0.0412)	-0.249 (0.322)
Observations	106	105	124	123
R-squared	0.050	0.093	0.000	0.027

View Table

Table 5:

Exposure to tax avoidance opportunities. The table shows the results from OLS regressions for 2016. The dependent variable is the share of outward Total FDI where the immediate investment is a foreign phantom corporation (Columns 1–2) and the share of inward Total FDI where the immediate investor is a foreign phantom corporation (Columns 3–4). The explanatory variables are the headline corporate tax (Corp tax rate) and GDP per capita (GDP per cap).

	(1)	(2)	(3)	(4)
	Share of outward FDI to SPE		Share of inward FDI from SPE
Corp tax rate	0.00598** (0.00256)	0.0381** (0.0179)	-8.38e-05 (0.00164)	0.0203* (0.0117)
GDP per cap (in logs)		0.0725 (0.0520)		0.0559 (0.0352)
Corp tax rate × GDP per cap (in logs)		-0.00352* (0.00193)		-0.00226* (0.00128)
Constant	0.145** (0.0633)	-0.519 (0.479)	0.256*** (0.0412)	-0.249 (0.322)
Observations	106	105	124	123
R-squared	0.050	0.093	0.000	0.027