Location-Sector Analysis of International Profit Shifting on a Multilayer Ownership-Tax Network
NNoname manuscript No. (will be inserted by the editor)
Location-Sector Analysis of International ProfitShifting on a Multilayer Ownership-Tax Network
Tembo Nakamoto · Odile Rouhban · Yuichi Ikeda
Received: date / Accepted: date
Abstract
Currently all countries including developing countries are expectedto utilize their own tax revenues and carry out their own development forsolving poverty in their countries. However, developing countries cannot earntax revenues like developed countries partly because they do not have effectivecountermeasures against international tax avoidance. Our analysis focuses ontreaty shopping among various ways to conduct international tax avoidance be-cause tax revenues of developing countries have been heavily damaged throughtreaty shopping. To analyze the location and sector of conduit firms likely to beused for treaty shopping, we constructed a multilayer ownership-tax networkand proposed multilayer centrality. Because multilayer centrality can considernot only the value flowing in the ownership network but also the withholdingtax rate, it is expected to grasp precisely the locations and sectors of conduitfirms established for the purpose of treaty shopping. Our analysis shows thatfirms in the sectors of Finance & Insurance and Wholesale & Retail trade etc.are involved with treaty shopping. We suggest that developing countries makea clause focusing on these sectors in the tax treaties they conclude.
The present study was supported by the Ministry of Education, Science, Sports, and Cul-ture, Grants-in-Aid for Scientific Research (B), Grant No. 17KT0034 (2017-2019) and Ex-ploratory Challenges on Post-K computer (Studies of Multi-level Spatiotemporal Simulationof Socioeconomic Phenomena).Tembo NakamotoGraduate School of Advanced Integrated Studies in Human Survivability1, Yoshida-Nakaadachi-cho, Sakyo-ku, Kyoto, 6068306, JapanTel: +81-75-762-2002E-mail: [email protected] RouhbanE-mail: [email protected] IkedaGraduate School of Advanced Integrated Studies in Human SurvivabilityE-mail: [email protected] a r X i v : . [ ec on . E M ] A p r Tembo Nakamoto et al.
Keywords
Multilayer network · Treaty shopping
JEL Classification
F23 H25
About 800 million people worldwide still cannot enjoy food, safe drinking wa-ter, and clean sanitation. The United Nations has established the SustainableDevelopment Goals (abbreviated as SDGs) with poverty as its first target. Iorder to provide public services required to solve poverty, such as health, edu-cation, and social security, each developing country needs funding equivalentto 4% of their Gross Domestic Product (abbreviated as GDP) (UN 2005). Sofar, such funding has mostly relied on developed countries. Recently, however,it has been emphasized that it is important for developing countries to pro-cure such funds by themselves (G20 2010). It has been confirmed again inthe implementation of SDGs (UN 2015) and it is set as SDG 17.1 to developthe taxing capabilities of developing countries. To reduce poverty, developingcountries are expected to increase their tax revenue.However, developing countries’ ability to raise tax revenues is still lowerthan that of developed countries. While developed countries earn more than30% tax on their GDP, many developing countries do not reach as much as20% except some Asian and Latin American countries. In particular, for morethan half of sub-Saharan African countries, it is less than 15% (UNDP 2010).The reason why developing countries cannot earn enough tax revenue for thescale of their economies is mainly due to a lack of tax collection capacity (Gor-don and Li 2009) and international tax avoidance. In particular, low-incomecountries in sub-Saharan Africa, Latin America, the Caribbean, and SouthAsia have lost much of their tax revenues due to international tax avoidance(Cobham and Jansky 2017). Compared with developed countries, developingcountries are more vulnerable to international tax avoidance because manydeveloping countries do not have effective policies to prevent international taxavoidance (OECD 2014). It is needed now to craft an effective policy towardinternational tax avoidance for developing countries to earn adequate tax rev-enues.We focused on treaty shopping among various ways to avoid taxes. Treatyshopping is a scheme to avoid withholding tax. When a firm pays its dividendsto a firm in another jurisdiction, withholding tax is usually imposed on thedividends. The withholding tax is sometimes reduced or exempted when thepayment is conducted between jurisdictions contracted in a tax treaty. Treatyshopping occurs when a firm in a third jurisdictions and not eligible to receivethe reduction or exemption tries to establish a so-called paper company in tjecontracting jurisdiction in order to enjoy the reductions or exemptions. Manydeveloping countries have reported huge losses in their tax revenue as a resultof treaty shopping (OECD 2014). Our analysis deals with treaty shopping,which heavely damages developing countries by loss of their tax revenues. ocation-Sector Analysis 3
Recently, network science has attracted attention because network theoryhas succeeded in highlighting emergent phenomena called ”complex systems,”which describes economic phenomena with a graph consisting of nodes andlinks (Barabasi 2016). However, actual economic phenomena often have toconsider multiple layers, as a result, research on multilayer networks has pro-gressed (Kivela et al. 2014). Multilayer networks consist of layers in which thenodes of one layer are connected to nodes in another.There have been only a few previous studies on treaty shopping. Mintz andWeichenrieder (2008) and Weyzig (2013) investigated treaty shopping takinginto account both the profits and the withholding tax rates, but their targetis only specific countries (Germany and The Netherlands). Garcia-Bernardoet al. (2017) investigate treaty shopping by focusing the profits (value) in theworld and Van’t Riet and Lejour (2018) focuses on the withholding tax ratesof 108 jurisdictions. While Van’t Riet et al. (2015) examined the relationshipbetween treaty shopping and dividends, Hong (2018) and Petkova et al. (2018)studied the effects of double tax treaties on foreign direct Investment. Thereare no studies to analyze treaty shopping taking both profits and withhold-ing tax rates worldwide. In addition, unlike other methods of internationaltax avoidance (Acciari et al. 2015), there are no studies to investigate treatyshopping from the point of view of sectors. This is the first study to analyzetreaty shopping from the perspective of sectors, while taking into account bothprofits (values) and the withholding tax rate worldwide.We construct and analyze a multilayer network consisting of an ownershipnetwork and a withholding tax network. There are three features of our analy-sis. The first point is the use of micro data to record firm information all overthe world. The second point is that we analyze treaty shopping by focusing onsectors for the first time. Third, the multilayer centrality we propose makesit possible to analyze treaty shopping more precisely by considering both theprofit (value) and the withholding tax rate.Section 2 describes the multilayer network we construct and explains mul-tilayer centrality, which we propose for analyzing the locations and sectorsof firms used for treaty shopping. Section 3 outlines the data used to buildthe multilayer network. Section 4 describes the results obtained by applyingour multilayer centrality to the multilayer network and considers effective taxpolicies to prevent treaty shopping. Section 5 summarizes our study.
Although countermeasures against some international tax avoidance methodshave already focused on specific sectors, countermeasures against treaty shop-ping have not focused on the sectors. We supposed that there are some sectorsthat are easy to exploit for treaty shopping like other tax avoidance methodsand examined the locations and the sectors of conduit firms used in treatyshopping. The concept of ”value” is used for our analysis because its effective-ness has been confirmed in previous research (Garcia-Bernardo et al. 2017).
Tembo Nakamoto et al.
However, the purpose for which conduit firms are established is not limited tothe use of treaty shopping. To limit our analysis to conduit firms for treatyshopping, we also considered withholding tax rates imposed on dividends. Oneof the main motivations to do treaty shopping is to recommend withholdingtax rates. We propose multilayer centrality as means to analyze the conduitfirms used in treaty shopping while taking into account both the value intensityand the withholding tax rates.2.1 Multilayer ownership-tax networkThe multilayer ownership-tax network is defined as M = ( g, l ), where g = { G α , G β } is a set of weighted directed graphs (called layers of M ) and l ∈ L αβ is a set of interlayer connections. Figure 1 shows an image of the network M .The first layer of M , defined as G α = ( N α , L α , W α , P α ), is the ownership net-work, where N α is a set of nodes (firms), L α is a set of links (shareholdingrelationships) between pairs of nodes, W α is the link value function (share-holding ratios), and P α is the node value function (operating incomes). Thelinks are directed, going from a shareholder firm to an owned firm. The secondlayer of M , defined as G β = ( N β , L β , W β ), is the withholding tax network,where N β is a set of nodes (jurisdictions), L β is a set of links (direction to paydividends) between pairs of nodes, and W β is the link value function (with-holding tax rates imposed on dividends. A node (firm) n α of the layer G α isconnected with that (jurisdiction) n β of the layer G β according to the locationof n α by L αβ . Withholding Tax NetworkOwnership Network
Fig. 1
Image of the multilayer ownership-tax network. ocation-Sector Analysis 5 firms used for treaty shopping by focusing on value intensity. The sink firmshave to be identified for analyzing the conduit firms because of the conduitfirms’ definition. Therefore, we first identify the locations and sectors of firmsfunctioning as a ”sink” and then analyze the locations and sectors of firmsfunctioning as a ”conduit.”
To identify the sink and the conduit, we use the value flowing in the ownershipnetwork. The value is defined as follows (Vitali et al. 2012): v n i +1 = p k l − (cid:89) i =1 w n i n i +1 (1)Here, p k is an operating income of a firm k located at the end of chains in theownership network and w n i n i +1 is the shareholding ratio from a shareholderfirm n i +1 to an owned firm n i . The value v n ( i +1) enters a firm n i +1 and thevalue v n ( i +2) leaves firm n i +1 .The locations and sectors of firms are expressed by jurisdiction × sectorpairs. For example, The Netherlands × Finance & Insurance indicates a setof firms located in The Netherlands whose sector is financial and insuranceactivities.
Figure 2 shows the concept of a ”sink.” We supposed that a jurisdiction × sector pair functioning as a ”sink” has much more value compared with itseconomic scale. SinkvaluecapitalOther jurisdic (cid:1) ons Other jurisdic (cid:1) onsvaluecapital
Fig. 2 Sink concept : more value enters jurisdictions × sector pairs functioning as a sinkand less value leaves the pair. Tembo Nakamoto et al. The sink centrality S js of jurisdiction j × sector s pair is defined as follows.At first, take the difference between the value entering and the value leavingfirms whose location is j and sector is s to calculate the value of the pair. Next,this is divided by the total value flowing in the ownership network to calculatethe value intensity of the pair. Finally, it is normalized by jurisdiction GDPto compare the value intensity with its economic scale: S js = (cid:80) V injs − (cid:80) V outjs V total · (cid:80) i GDP i GDP j (2)Here, (cid:80) V injs is the sum of the values entering the firms whose location is j andsector is s , (cid:80) V outjs is the sum of the values leaving the firms whose location is j and sector is s , V total is the total amount of the value flowing in the ownershipnetwork, and GDP j is the GDP of jurisdiction j . We suppose that the GDPof each jurisdiction represents its economic scale. A conduit is like a tunnel through which value enters or leaves a sink. Weconsider that much more value entering the sink or leaving the sink passesthrough jurisdiction × sector pairs functioning as a conduit compared withits economic scale. Because the conduit plays a key role in treaty shopping,we analyze the location and sectors of firms functioning as conduits. For theanalysis, we define a conduit centrality c js of jurisdiction j × sector s thatconsists of conduit outward centrality c outjs and conduit inward centrality c injs .Figure 3 shows the concept of a conduit. The conduit outward centrality c outjs measures the value entering the sink through firms whose location is j andsector is s while the conduit inward centrality c outjs measures the value leavingthe sink through firms whose location is j and sector is s . SinkOther jurisdic (cid:1) ons ConduitOutwardvaluecapitalvalue ConduitInwardcapital
Fig. 3 Conduit concept : value enters into ”sink” through ”conduit outward” and leaves”sink” through ”conduit inward.”ocation-Sector Analysis 7
The conduit outward centrality c outjs of jurisdiction j × sector s is defined asfollows. At first, it measures the values passing through firms whose location is j and sector is s and entering the sink. Next, it is divided by the total amountof values flowing in the ownership network G α to calculate the value intensity.Lastly, it is normalized by jurisdiction GDP to compare the passing value withits economic scale. c outjs = V sinkout V total · (cid:80) i GDP i GDP j (3)Here, V sinkout is the sum of the value passing through the firms whose location is j and sector is s and entering the sinks, V total is the total amount of the valueflowing in the ownership network G α , and GDP j is the GDP of jurisdiction j representing its economic scale.The conduit outward centrality c outjs is standardized so that its averagevalue and standard deviation is 1.0 because of the calculation of the multilayercentrality described in Sect. 2.4: C outjs = c outjs − c outjs σ c outjs + 1 (4)Here, c outjs is the average of all conduit outward centrality c outjs and σ c outjs isthe standard deviation of all conduit outward centrality c outjs . Therefore, if thestandardized conduit outward centrality C outjs of a jurisdiction j × sector s pare is above 1.0, then it can be said that the value to sink has passed throughthe jurisdiction j × sector s pare while much considering its economic scalecompared with other pairs.On the other hand, the conduit inward centrality c injs of jurisdiction j × sector s is defined as follows. At first, it measures the values leaving from thesink and passing through the firms whose location is j and sector is s . Next,it is divided by the total amount of values flowing in the ownership network G α . Lastly, it is normalized by jurisdiction GDP to compare the passing valuewith its economic scale: c injs = V sinkin V total · (cid:80) i GDP i GDP j (5)Here, (cid:80) V g is the sum of the value from sink through firms whose location isjurisdiction j and sector s , (cid:80) V g is the total amount of the values flowing in theownership network G α , and GDP j is the GDP of jurisdiction j representingits economic scale.The conduit inward centrality c injs is standardized so that its average valueand standard deviation is 1.0 because of the calculation of the multilayercentrality described in Sect. 2.4: C injs = c injs − c injs σ c injs + 1 (6) Tembo Nakamoto et al.
Here, c injs is the average of all conduit inward centrality c injs and σ c injs is thestandard deviation of all conduit inward centrality c injs . Therefore, if the stan-dardized conduit inward centrality C injs of a jurisdiction j × sector s pare isabove 1.0, then it can be said that the value from the sink has passed throughthe jurisdiction j × sector s pare while considering its economic scale comparedto other pairs.To make it easy to compare between jurisdiction × sector pairs, the Eu-clidean distance between the standardized conduit outward centrality C outjs andthe standardized conduit inward centrality C injs is calculated. The distance isadjusted to 1.0 when the standardized conduit outward centrality C outjs andthe standardized conduit inward centrality C injs are both 1.0: C js = (cid:113) ( C injs ) + ( C outjs ) / √ l j to measure the attraction of treaty shoppingquantitatively. Load centrality l j is the total amount of the packet passingthrough a node when all pairs of nodes send and receive a data packet betweenthem (Goh et al. 2001; Brandes 2008). We think of a node as a jurisdiction,a data packet as a dividend, put the withholding tax rates as the weight, andcalculate the load centrality l j (Nakamoto and Ikeda 2018). Therefore, thehigher the load centrality l j , the more likely it will be for the centrality to beused for treaty shopping. The load centrality l j of a node j is calculated asfollows: l j = (cid:88) o (cid:54) = d (cid:54) = j w o,d (8)Here, j , o , and d are nodes (jurisdictions), w o,d is the amount of the dividendpassing through j when a firm located in o sends a dividend to a firm locatedin d .The load centrality l i is standardized so that both its average value andstandard deviation are 1.0: L j = l j − l j σ l j + 1 (9)Here, l j is the average of all load centrality l j and σ l j is the standard deviationof all load centrality l j . Therefore, if the standardized load centrality L j of a ocation-Sector Analysis 9 given jurisdiction j is above 1.0, then it can be said that jurisdiction j hasmore possibility to be chosen as the location of conduit firms compared withother jurisdictions.2.4 Multilayer centralityMultilayer centrality M js consists of the multilayer outward centrality M outjs and the multilayer inward centrality M injs like the conduit centralities c js . Themultilayer outward centrality M outjs and the multilayer inward centrality M injs are, respectively, the weighted geometry average of the values of the standard-ized conduit outward centrality C outjs and the standardized conduit inwardcentrality C injs in the ownership network G α multiplied by the standardizedload centrality L j in the withholding tax network G β . The multilayer out-ward centrality M outjs and the multilayer inward centrality M injs are defined asfollows: M outjs = α + β (cid:113) ( C outjs ) α · ( L w ) β (10) M injs = α + β (cid:113) ( C injs ) α · ( L w ) β (11)Here, α and β determine the ratio to consider the value intensity and thewithholding tax rate to analyze the conduit firms.To make it easy to compare between jurisdiction × sector pairs, the Eu-clidean distance between the multilayer outward centrality M outjs and the mul-tilayer inward centrality M injs is calculated. The distance is adjusted to 1.0when the multilayer outward centrality M outjs and the multilayer inward cen-trality M injs are both 1.0: M js = (cid:113) ( M injs ) + ( M outjs ) / √ We used the Orbis 2015 database (Bureau van Dijk 2015) for the ownershipnetwork G α . The database comprises shareholding ratio, operating income,and sector information of about more than 30 million firms across more than20 jurisdictions. Sectors are defined following the statistical classification ofeconomic activities in the European community (NACE Rev. 2).This database is based on information that each firm reported to their lo-cal Chamber of Commerce. Because each jurisdiction requires different criteriafor the Ministry of Commerce to submit financial statements, data availabilityvaries greatly depending on jurisdictions. For example, Kosovo includes 99.1%while Seychelles includes only 0.1%. In particular, it does not include informa-tion on small-scale firms (Kalemi-Ozan et al. 2015) and information on firmslocated in jurisdictions where financial secrets are high. Therefore, our analysishas a certain bias due to data availability. The shareholding ratio, operating income, and sector information are necessary to calculate the value (see Sect.2.2.1). We removed the nodes v α (firms) that do not have such informationfrom the ownership network G α .The withholding tax rates imposed on dividends that a firm pays to a firmin other jurisdiction are defined in the domestic laws of each jurisdiction orthe tax treaties concluded between jurisdictions. We used the reduced with-holding tax rates because the purpose of our analysis is treaty shopping. Thedata has been extracted from Ernst & Young (2017), which summarizes suchwithholding tax rates, for the withholding tax network G β . At first, 25 jurisdiction × sector pairs are identified as ”sinks.” Next, thestandardized conduit centrality C js clarified which jurisdictions × sector pairsthrough which more value passes and the standardized load centrality L j re-veals which jurisdiction is attractive for treaty shopping. Finally, we calculatedthe standardized multilayer centrality M js by combining the standardized con-duit centrality C js and the standardized load centrality L j and found thatfirms in certain sectors are often used for treaty shopping. Based on the re-sult, we suggest countermeasures against treaty shopping to focus on certainsectors.4.1 Value intensity We calculated standardized sink centrality S js for 1,704 jurisdiction × sectorpairs for which data can be obtained. The maximum value of the sink centralitywas 1784.38 and the minimum value of that was -251.82. The value of sinkcentrality is wide, but most of the value of the sink centrality is concentratedin parts. Figure 4 shows its frequency distribution. The sink centrality of 1,598pairs (about 93.8% of the total pairs) is 1 or less. We can suppose that thepairs whose standardized sink centrality S js is high are very unique and arelikely to function as a sink.Table 1 shows 25 jurisdiction × sector pairs whose standardized sink cen-tralities S js are higher than 10.0. Nine sectors are represented among the pairswith the highest sink centrality. Financial and insurance activity (”Finance &Insurance”) represents more than half of this list and professional, scientific,and technical activities (”Professional activities etc.”) represents about 1/8 ofthis list, showing that much value remains in these sectors. For the calculationof standardized sink centrality S js , we regarded the 25 pairs accounting forabout 1.5% of all pairs as sinks. ocation-Sector Analysis 11 Sink centrality F r equen cy Fig. 4 Frequency distribution of sink centrality S js . The sink centrality of mostjurisdictions × sector pairs is less than 1.0. The width of the bins is 81.5. The standardized conduit outward centrality C injs was calculated for 636 juris-diction × sector pairs and the standardized conduit inward centrality C injs wascalculated for 461 jurisdiction × sector pairs. Finally, we obtained 389 pairshaving both standardized conduit outward centralities C outjs and standardizedconduit inward centralities C injs . The maximum value of the standardized con-duit outward centrality C outjs was 19.37 and the minimum value was 0.78. Themaximum value of the standardized conduit inward centrality C injs is 10.79and the minimum value is 0.81. Even though the differences between theirmaximum value and their minimum value are not small, both centralities weredistributed in certain parts. Figure 5 and Figure 6 show frequency distribu-tions of the standardized conduit outward centrality C outjs and the standardizedconduit inward centrality C injs respectively. 608 pairs, accounting for 95.6% ofthe total pairs, have standardized conduits outward centrality C outjs less than1.0. Similarly, 449 pairs, accounting for 97.4% of the total pairs, have stan-dardized conduit inward centrality C injs less than 1.0. We can suppose that thepairs whose standardized conduit centrality C js is high are very unique andare likely to function as conduits.Table 2 shows 15 jurisdiction × sector pairs whose standardized conduitcentrality C js is over 2.0. Seven sectors across 13 jurisdictions are representedamong the 15 pairs. Most of the 13 jurisdictions are developed countries andthere are no small island jurisdictions except Bermuda. Wholesale and retailtrade and repair of motor vehicles and motorcycles (”Wholesale & Retail tradeetc.”) represents more than 1/4 of this list and Finance & Insurance represents Table 1
Jurisdiction × Sector Pairs Where Sink Centrality S js is above 10Jurisdiction Sector S js Malta Finance & Insurance 1784.38Luxembourg Professional activities etc. 511.86Luxembourg Administrative & Support service 369.72Bermuda Construction 134.63Bermuda Finance & Insurance 87.75British Virgin Islands Manufacturing 57.03Cayman Islands Finance & Insurance 39.75Curacao Finance & Insurance 34.47France Finance & Insurance 34.10Marshall Islands Transportation & Storage 30.36Sweden Finance & Insurance 28.00British Virgin Islands Wholesale & Retail trade 23.13Cyprus Finance & Insurance 19.56Spain Finance & Insurance 19.04Curacao Wholesale & Retail trade 16.43UK Mining & Quarrying 16.18Portugal Finance & Insurance 15.50Norway Finance & Insurance 14.30Belgium Finance & Insurance 13.07the UK Finance & Insurance 12.95Austria Professional activities etc. 12.71Iceland Finance & Insurance 12.69South Africa Manufacturing 10.81Singapore Other service etc. 10.48UK Professional activities etc. 10.25 about 1/5 of this list, showing that considerable value, either from or towardsinks, passes through firms in these sectors.4.2 Withholding tax rateWe calculated the standardized load centrality L j for 165 jurisdictions to findwhich jurisdictions are likely to be used for treaty shopping. The standardizedload centrality L j considers all withholding tax liabilities imposed on dividendsmade between 165 jurisdictions (27,060 pairs in total). The maximum valueof the standardized load centrality L j was 7.87 and the minimum value was0.59. The value of the standardized load centrality L j is wide, but most of thevalue of the standardized load centrality L j is concentrated in parts. Figure7 shows its relative frequency distribution. The standardized load centrality L j of 147 jurisdictions (about 89.0% of 165 jurisdictions) is 1 or less and thejurisdictions with high standardized load centralities L j are limited. We cansuppose that the jurisdictions whose standardized load centrality L j is highare likely to be used for treaty shopping.Table 3 shows jurisdictions whose standardized load centrality L j is in thetop 15. It is confirmed that focusing on withholding tax rates imposed ondividends and evaluating jurisdictions by the load centralities l j is meaningful ocation-Sector Analysis 13 Conduit outward centrality F r equen cy Fig. 5 Frequency distribution of the standardized conduit outward centrality C outjs . Most of the centralities are less than 1.0. The width of the bins is 1.1.
Conduit inward centrality F r equen cy Fig. 6 Frequency distribution of the standardized conduit inward centrality C injs . Most of the centralities are less than 1.0. The width of the bins is 1.1. to the analysis of treaty shopping because the list contains jurisdictions thatmultinationals usually use for treaty shopping (Diamond et al. 2017; Nakamotoand Ikeda 2018). The 15 jurisdictions are scattered around Europe, the MiddleEast, East Asia, Africa, and America. European jurisdictions especially havehigh standardized load centrality L j because the withholding tax imposed Table 2
Jurisdiction × Sector Pairs Whose Standardized Conduit Centrality C js is over2.0Jurisdiction Sector C outjs C injs C js The Netherlands Finance & Insurance 19.37 6.13 14.36Luxembourg Wholesale and Retail trade etc. 6.18 10.79 8.79Bermuda Mining & Quarrying 1.33 9.49 7.48Luxembourg Finance & Insurance 6.55 8.08 7.36Sweden Electricity & Gas supply etc. 9.03 0.91 6.41Austria Wholesale & Retail trade etc. 3.21 7.87 6.01Bermuda Wholesale & Retail trade etc. 4.12 7.42 6.00Portugal Professional activities etc. 6.35 1.24 4.57Malaysia Manufacturing 5.62 1.03 4.04Switzerland Wholesale & Retail trade etc. 3.85 0.93 2.80Germany Manufacturing 3.08 2.19 2.46UK Administrative & Support services 1.86 2.94 2.46France Professional activities etc. 3.23 1.05 2.40Ireland Finance & Insurance 3.12 1.18 2.36Austria Mining & Quarrying 3.08 0.84 2.26Austria Manufacturing 2.53 1.61 2.12Portugal Information & Communication 2.16 1.94 2.05
Load centrality F r equen cy Fig. 7 Distribution of standardized load centrality L j . Most of the centralities arebelow 1.0. on dividends made between EU member states is exempted by the EuropeanUnion directive (Directive 90/435/EC). ocation-Sector Analysis 15
Table 3
Jurisdictions Whose Standardized Load Centralities L j are in the Top 15Jurisdiction L j Jurisdiction L j Jurisdiction L j UK 7.87 Singapore 2.94 Saint Lucia 2.46UAE 6.59 Switzerland 2.93 Bahrain 2.46Kuwait 5.22 Mauritius 2.91 Malaysia 2.45The Netherlands 3.44 Spain 2.86 Ireland 2.39Cyprus 3.11 Luxembourg 2.65 Estonia 2.24Hong Kong 3.05 Qatar 2.44 Malta 2.21 M js can be calculated regarding 389pairs because the calculation needs both the standardized conduit outwardcentrality C outjs and the standardized conduit inward centrality C injs . In addi-tion, for the calculation of the standardized multilayer centrality M js , α and β need to be set as in Eqs. (10) and (11). α and β determine how much valueintensity and withholding tax rate are considered, respectively. We tried tofind the appropriate value of β by comparing the results obtained by fixing α to 1.0 and fluctuating β . Table 4 contains the jurisdiction × sector pairswhose standardized multilayer centrality M js is above 2.0 when β is set to0.8. Many listed pairs are in the United Kingdom and we cannot see the dif-ference between sectors because the standardized load centrality L j does nottake the difference of sectors into account. In other words, it shows that setting β = 0 . M js is above 2.0when β is set to 0.1. The listed pairs include jurisdictions such as Germanyand France, which are unlikely to be used for treaty shopping because thesejurisdictions are not listed in Table 2. There may be another reason (excepttreaty shopping) why more value passes through such jurisdictions.Table 6 shows pairs whose standardized multilayer centrality M js is above2.0 when β is set to 0.5 and includes the jurisdictions well-known as conduits(Diamond et al. 2017) and, therefore, likely to be used for treaty shopping.When we set β to 0.3 for our analysis of conduit firms, 179 pairs had standard-ized multilayer centrality M js above 1.0, accounting for 46.02% of total pairs;48 pairs had standardized multilayer centrality M js above 1.5, accounting for12.34%; and 16 pairs had standardized multilayer centrality M js above 2.0,accounting for 4.11%. Finally, we decided to consider the pairs whose stan-dardized multilayer centrality M js is above 2.0 as the pairs that are likely tobe used for treaty shopping.The 13 pairs listed in Table 6 include eight sectors across 11 jurisdictionsand Finance & Insurance and Wholesale & Retail trade etc., firms are remark-able in Table 6 compared with other sectors. This implies that these sectorshave greater potential to be used for treaty shopping. It is possible that treatyshopping can be prevented by making new tax rules focusing on these sectors.In addition, no Middle East Asian jurisdictions are ranked highly in the stan- Table 4
Jurisdiction × Sector Pairs Whose Standardized Multilayer Centrality M js isabove 2.0 ( β = 0 . M outjs M injs M js The Netherlands Finance & Insurance 8.98 4.74 7.18Luxembourg Wholesale & Retail trade etc. 4.24 5.78 5.07Luxembourg Finance & Insurance 4.38 4.93 4.66UK Administrative activities etc. 3.53 4.55 4.08UK Manufacturing 3.13 3.18 3.15Malaysia Manufacturing 3.89 1.52 2.95Switzerland Wholesale & Retail trade etc. 3.41 1.55 2.65UK Information & Communication 2.40 2.44 2.42UK Electricity & Gas supply etc. 2.37 2.41 2.39Bermuda Mining & Quarrying 1.01 3.19 2.37UK Wholesale & Retail trade etc. 2.29 2.42 2.35Sweden Electricity & Gas supply etc. 3.18 0.89 2.33Bermuda Wholesale & Retail trade etc. 1.90 2.63 2.29UK Construction 2.22 2.35 2.29UK Transportation & Storage 2.23 2.32 2.27Ireland Finance & Insurance 2.77 1.62 2.27UK Water supply etc. 2.20 2.30 2.25UK Other service 2.20 2.30 2.25UK Real estate activities 2.23 2.25 2.24UK Arts & Entertainment etc. 2.20 2.28 2.24UK Public Administration etc. 2.24 2.23 2.24Singapore Wholesale & Retail trade etc. 2.41 2.05 2.23UK Human health etc. 2.18 2.24 2.21UK Accommodation and Food service etc. 2.19 2.24 2.21UK Education 2.18 2.24 2.21UK Agriculture etc. 2.18 2.23 2.21Austria Wholesale & Retail trade etc. 1.60 2.64 2.18Singapore Manufacturing 1.87 2.44 2.17Ireland Professional activities etc. 2.40 1.80 2.12Spain Professional activities etc. 2.07 1.94 2.00 dardized multilayer centrality M js , although their standardized load centrali-ties L j are ranked highly (see Table 3). Other regulations of the jurisdictionsimposed may cause the results.At present, the mainstream countermeasures against treaty shopping in-volve introducing a limitation of benefit clause or a principal purpose test intotax treaties. The limitation of benefit clause limits firms that can receive thewithholding tax reduction or exemption by certain criteria. On the other hand,the principal purpose test deprives firms whose main purpose is to enjoy thereduction or exemption of withholding tax liabilities. Developing countries pre-fer the principal purpose test because they are easy to enforce compared withthe limitation of benefit clause, whose application criteria are complicated.On the other hand, the business community is concerned with the principalpurpose test because the test is unclear as to the main purpose and tends toprefer the limitation of benefit clause whose application criteria are clearer.Our analysis shows that some sectors, such as Manufacturing, Wholesale& Retail trade etc., Professional activities etc., are likely to be used for treaty ocation-Sector Analysis 17 Table 5
Jurisdiction × Sector Pairs Whose Standardized Multilayer Centrality M js isabove 2.0 ( β = 0 . M outjs M injs M js The Netherlands Finance & Insurance 13.00 5.37 9.94Luxembourg Wholesale & Retail trade etc. 5.08 7.81 6.59Luxembourg Finance & Insurance 5.32 6.25 5.80Bermuda Mining & Quarrying 1.16 5.65 4.08Sweden Electricity & Gas supply etc. 5.25 0.90 3.77Bermuda Wholesale & Retail trade etc. 2.75 4.33 3.63Austria Wholesale & Retail trade etc. 2.24 4.46 2.20Malaysia Manufacturing 4.64 1.26 3.40UK Administrative & Support service 2.60 3.69 3.19Switzerland Wholesale & Retail trade etc. 3.62 1.21 2.70Portugal Professional activities etc. 3.66 1.04 2.69Ireland Financial & Insurance 2.93 1.39 2.29UK Manufacturing 2.20 2.24 2.22Germany Manufacturing 2.43 1.87 2.17France Professional activities etc. 2.72 1.14 2.08Ireland Professional activities etc. 2.41 1.61 2.05Singapore Wholesale & Retail trade etc. 2.23 1.78 2.02
Table 6
Jurisdiction × Sector Pairs Whose Standardized Multilayer Centrality M js isabove 2.0 ( β = 0 . M outjs M injs M js The Netherlands Finance & Insurance 10.88 5.06 8.49Luxembourg Wholesale & Retail trade 4.66 6.76 5.81Luxembourg Finance & Insurance 4.85 5.57 5.22UK Administrative & Support services 3.01 4.08 3.59Malaysia Manufacturing 4.26 1.38 3.17Bermuda Mining & Quarrying 1.09 4.30 3.13Sweden Electricity & Gas supply etc. 4.13 0.89 2.99Bermuda Wholesale & Retail trade 2.30 3.41 2.91Austria Wholesale & Retail trade 1.90 3.46 2.80Switzerland Wholesale & Retail trade 3.52 1.36 2.67UK Manufacturing 2.60 2.65 2.63Ireland Finance & Insurance 2.85 1.49 2.28Singapore Manufacturing 1.71 2.35 2.05 shopping. We think that it is effective for preventing treaty shopping to focuson such sectors as Controlled Foreign Company rules of some jurisdictions,which is a countermeasure against another scheme of international tax avoid-ance, already focused on sectors. The introduction of rules focusing on somesectors may not only prevent treaty shopping effectively but also reduce thecomplexity of application criteria developing countries are concerned with andimprove taxpayer predictability.The size of each jurisdictions in Figure 8 indicates the size of the standard-ized multilayer centrality M js regarding Finance & Insurance and Wholesale& Retail trade etc., respectively. It should be noted that jurisdictions withhigh standardized multilayer centrality M js are limited. Figure 8 (a) indicates the standardized multilayer centrality M js of Finance & Insurance and showsthat the centrality of The Netherlands, Luxembourg, and other financial cen-ters such as the United Kingdom, Bahrain, Hong Kong, and Mauritius is high.Figure 8 (b) indicates the standardized multilayer centrality M js of Wholesale& retail trade etc., and shows that the centrality of the jurisdictions of Europeand South East Asia is high. We suggest that the new clauses focusing on cer-tain sectors are introduced to tax treaties already concluded with jurisdictionshaving high standardized multilayer centrality M js . (a) Finance & Insurance(b) Wholesale & Retail trade etc. Fig. 8 Cartogram of multilayer centralities M js . The size of each jurisdiction indicatesthe degree of multilayer centrality M js . Developing countries conclude tax treaties with developed countries to in-crease investment from developed countries. With globalization of economy,the number of tax treaties has increased and about 1,000 tax treaties are re-lated to developing countries among about 4,000 tax treaties in the world. Itshould be noted that tax treaties not only increase foreign direct investmentfrom developed countries, but also increase the possibility of treaty shoppingand the loss of their own tax sources. ocation-Sector Analysis 19
Developing countries have mainly relied on assistance from developed countriesto reduce poverty, but they are now required to carry out their own develop-ment through their own tax revenues. International tax avoidance is one ofthe reasons why developing countries cannot earn tax revenues compared withdeveloped countries. Even though there are various ways to conduct interna-tional tax avoidance, treaty shopping is focused on in this paper due to taxrevenues of developing countries. To analyze the location and sector of conduitfirms that are likely to be used for treaty shopping, we constructed the multi-layer ownership-tax network and proposed the multilayer centrality. Becausemultilayer centrality can consider not only the value flowing in the ownershipnetwork but also the withholding tax rate, it is expected to grasp preciselythe locations and the sectors of conduit firms established for the purpose oftreaty shopping. The results of our analysis suggest that firms in the sectors ofFinance & Insurance and Wholesale & Retail trade etc. may be conduit firmsthat plays an important role in treaty shopping. Therefore, we suggest that toprevent treaty shopping, developing countries should introduce a clause focus-ing on certain sectors in their tax treaty, especially with developed countrieswith high multilayer centrality. This is because the countermeasures to treatyshopping that focus on some sectors is not complicated, developing countriesfind it easier to use them, and the predictability of taxpayers is not harmedmuch. A further quantitative study of treaty shopping is needed that takes notonly withholding tax but also corporate tax into consideration. Such findingswould contribute to our understanding of the effects of treaty shopping towardeach jurisdiction’s tax revenue.
Acknowledgements
I would like to thank Tadao Okamura and Hiroaki Takashima forvaluable comments.
Conflict of interest statement
On behalf of all authors, the corresponding author statesthat there is no conflict of interest.