## wp1869

## Source details

**Canonical URL:** [wp1869](https://www.imf.org/-/media/files/publications/wp/2018/wp1869.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/wp/2018/wp1869.pdf.md)
- [Structured JSON version](/-/media/files/publications/wp/2018/wp1869.pdf.json)

---

### I. Introduction and motivation
- Tax-motivated profit shifting: MNCs shift income from affiliates in high-tax jurisdictions to affiliates in low-tax jurisdictions to reduce overall tax liability.
- Empirical context cited:
  - German affiliates of MNCs paid on average 27 percent less in taxes than comparable domestic German firms (Finke, 2013).
  - UK subsidiaries of foreign MNCs report taxable profits as a share of total assets on average 12.8 percentage points lower than comparable domestic standalone companies (Habu, 2017).
  - Consensus semi-elasticity of reported profitability by MNCs w.r.t. international tax differentials ≈ -1.2 (Beer, De Mooij, and Liu, 2018).
  - Recent estimates of annual global government revenue loss between $100 and 650 billion (Crivelli, de Mooij, and Keen, 2016; OECD, 2015; UNCTAD, 2015).
- Research objective: explore the effect of Transfer Pricing Regulations (TPRs) on multinational investment using a simple theoretical model and micro-level empirical analysis.

### II. Transfer Pricing Regulations (TPRs) and mechanism
- Transfer prices: prices for related-party transactions; manipulation is a common profit-shifting channel.
- TPR components:
  - Allowed methods to determine arm’s-length prices.
  - Documentation requirements.
  - Penalties and enforcement measures.
  - Probability of transfer price adjustment.
- TPR strictness index (tprisk):
  - Based on 15 regulatory/enforcement features; ranges between 1.26 and 5.17 in the sample.
  - Higher values indicate more stringent TPR.
- Theoretical channel (model):
  - Parent supplies intermediate input x to subsidiary; parent sets transfer price pT possibly deviating from arm’s-length price p.
  - Expected cost per unit of deviation c = β(pT − p)^2, where β is affected by TPR (stricter TPR → higher β).
  - Key model implications:
    - Optimal transfer price: pT = p − (τh − τs)/(2β).
    - Stricter TPR (higher β) reduces manipulation, lowers intermediate supply x and reduces foreign affiliate capital k (∂k/∂β < 0).
    - Effect only for multinational structures with tax differentials (τh ≠ τs).

### III. Data and empirical strategy
- Main unconsolidated sample:
  - Unbalanced panel of 101,079 unique companies in 27 countries, 2006–2014 (ORBIS).
  - MNC affiliate: ultimate parent in different country, ≥ 50% shares.
  - Domestic affiliate: ultimate parent in same country, ≥ 50%, all affiliates in same country.
  - Main regression sample excludes financials/utilities, requires non-missing sales, total assets, fixed assets, industry info, ≥ 3 consecutive observations; excludes MNC affiliates located in same country as parent.
- Consolidated accounts sample (Section VII):
  - 17,638 observations ≈ 2,024 distinct non-financial, non-utility parent companies in > 60 countries, 2006–2014.
- Key firm-level variables:
  - Investment spending It = Kt − Kt−1 + depreciation.
  - Investment rate = It / Kt−1.
  - Intensive margin = log of investment spending.
  - Extensive margin = indicator for positive investment.
  - All ratio variables winsorized at top and bottom 1 percentile.
- Country-level variables:
  - TPR dummy (TPR_kt) and tprisk index from Mescall and Klassen (2014), extended to 2014 via Deloitte’s Transfer Pricing Strategic Matrix, 2014.
  - Thin-capitalization rules (TCRs) from De Mooij and Hebous (2017).
  - Macroeconomic controls from IMF WEO.
  - User cost of capital: r_real + (1 − A)/(1 − CIT), with A net present value of depreciation allowances.
- Identification:
  - Difference-in-difference (DD): compare MNC affiliates to domestic affiliates before/after TPR introduction in host country.
  - Panel regressions for semi-elasticities of MNC investment w.r.t. statutory CIT and cost of capital; interaction terms to capture TPR effects.
  - Event-study tests for parallel trends (pre-reform p-value = 0.228).

### IV. Main DD empirical findings
- Core DD specification: Investment_ikt = ai + dt + β (MNC_i × TPR_kt) + controls + ε_ikt.
- Baseline results (Table 4):
  - Column (1) DD coefficient = -0.049***.
  - Preferred specification (Column (6)) DD estimate = -0.041***.
    - Interpretation: TPR implementation reduces investment rate by multinationals by 4.1 percentage point.
    - Given average gross investment per dollar of fixed asset = 35.9 cents, this corresponds to 11.4 percent reduction in investment for affiliates in the sample.
  - Interaction with tprisk (Column (7)): coefficient = -0.072*** (0.025).
    - For tprisk = 3.0, reduction = 0.216 percentage points.
    - For tprisk = 5.17, reduction = 0.36 percentage points.
- Robustness checks (Table 5):
  - Results stable when excluding affiliates with parents in worldwide tax system; clustering at host-country level; winsorizing dependent variable at 2.5 percentile; Mahalanobis matching on firm characteristics.
  - Dropping Luxembourg affiliates yields DD = -0.043***.

### V. Heterogeneity and margins (Table 6)
- Extensive vs intensive margin:
  - Extensive margin (probability of positive investment): coefficient small and insignificant → TPRs do not affect likelihood to invest.
  - Intensive margin (log investment): DD coefficient positive and highly significant → reductions driven by firms already investing.
- Tax differential size:
  - Interaction by tax-differential quartiles: bottom quartile negative/insignificant; 2nd quartile larger and highly significant (MNC_i × TPR_kt × Quartile TaxDiff,2 = -0.052*** (0.015)); responses smaller in 3rd and 4th quartiles and not monotonic.
- Intangible asset intensity:
  - Three-way interaction MNC_i × TPR_kt × IntangShare_i = 0.002*** (0.000).
  - Quantitatively: difference between IntangShare = 0 and IntangShare = 1 is 0.2 percentage points (investment effect drops from -3.2 percentage points to -3.0 percentage points).
- Interaction with thin-capitalization rules (TCRs):
  - Countries without TCR (Column (5)): DD coefficient = -0.013, insignificant.
  - Countries with TCR (Column (6)): DD coefficient ≈ three times larger and significant at 1 percent.
  - Interpretation: TPR effectiveness depends on presence of other anti-avoidance measures; debt shifting and transfer mispricing are likely substitutes.

### VI. TPR-adjusted tax elasticities (Table 7)
- Semi-elasticity specification: ln(Investment_ikt) = ai + dt + βtax CIT_kt + βTPRtax × CIT_kt × TPR_kt + controls + ε_ikt.
- Column (1) (multinational affiliates only):
  - Without TPR: one percentage point lower statutory CIT increases investment (share of total assets) by 0.83 percentage point (CIT_kt = -0.834*** (0.198)).
  - Presence of TPR increases sensitivity by 0.36 percentage point (CIT_kt × TPR_kt = -0.356*** (0.073)) to 1.19 percentage points.
- Column (3) (log fixed tangible assets; semi-elasticity interpretation):
  - Semi-elasticity without TPR slightly larger than one and highly significant.
  - Presence of TPR increases tax effect by 0.24 to overall semi-elasticity = 1.26.
  - Interpretation: a 1 percentage-point increase in CIT reduces MNC investment by ≈ 1 percent without TPR; after TPR, corporate tax rates matter about one quarter more.
  - Following Section V.B interpretation: introduction of TPR corresponds to a “TPR-adjusted” CIT rate that is 23 percent larger than without TPR.
- Replacing CIT with cost of capital:
  - Presence of TPR implies “TPR-adjusted” cost of capital is 15 percent larger than without TPR (Column (4) interpretation).
  - Coefficients: example COC_kt × TPR_kt = -1.736*** (0.425) in Table 7.

### VII. Consolidated (MNC-group) investment (Table 8)
- Question: do affiliate-level reductions reflect total MNC investment decline or relocation across affiliates?
- DD on consolidated parent-company accounts (worldwide investment):
  - Column (1): baseline DD coefficient = 0.056*** (0.019).
  - Column (2): remains significant with country-level characteristics = 0.049** (0.019).
  - Column (3): becomes insignificant when including country-year fixed effects = 0.029 (0.024).
  - Columns (4)-(5): remain insignificant with industry-year and industry-country fixed effects; interacting with parent statutory CIT rate does not change basic finding (MNC_i × TPR_kt × CIT_kt = 0.125 (0.095)).
- Interpretation: absence of a clear negative effect on consolidated investment suggests relocation of investment across affiliates rather than a net global reduction in MNC investment.

### VIII. Key summary statistics (selected from Table 2)
- Firm-level:
  - Investment spending ($1,000): mean 1,725; Std Dev 30,589; Median 70.73; P10 -472,266.
  - Fixed asset ($1,000): mean 11,528; Std Dev 133,200; Median 689.49; P10 2,714; P90 14,167.
  - Investment rate (I_t/K_{t−1}): mean 0.45; Std Dev 1.07; Median 0.15; P10 -0.06; P90 1.06.
  - Operating revenue ($1,000): mean 54,055; Std Dev 440,600; Median 6,812; P10 681; P90 83,028.
- Country-level:
  - CIT rate (%): mean 27.34; Std Dev 5.79; Median 28.00; P10 19.00; P90 33.33.
  - Tax differential (absolute %): mean 4.79; Std Dev 6.22; Median 1.67; P10 0; P90 14.50.
  - Cost of Capital: mean 0.07; Std Dev 0.01; Median 0.07; P10 0.06; P90 0.08.

### IX. Diagnostics and robustness
- Parallel trends / event-study (Table 3):
  - Pre-TPR year coefficients: Pre Year 5 = 0.147 (0.199); Pre Year 4 = 0.191 (0.142); Pre Year 3 = 0.145 (0.129); Pre Year 2 = -0.044 (0.034); Pre Year 1 = 0.008 (0.026).
  - Post-TPR year coefficients: Post Year 1 = -0.049** (0.023); Post Year 2 = 0.001 (0.015); Post Year 3 = -0.036*** (0.013); Post Year 4+ = -0.015** (0.006).
  - Joint test that all pre-reform β_l coefficients are equal: p-value = 0.228 (parallel trends not rejected).
- Standard errors: heteroskedasticity-robust and clustered at firm level unless specified otherwise.
- Significance notation used in tables: ***, **, * denote significance at the 1%, 5% and 10% levels.

### X. Conclusions and policy implications
- Main empirical findings:
  - Introduction of TPRs reduced investment in multinational affiliates by more than 11 percent on average (11.4 percent implied by 4.1 percentage point reduction relative to 35.9 cents average gross investment).
  - Stricter TPR regimes (higher tprisk) induce larger reductions in MNC affiliate investment.
  - Reduction larger for firms less intensive in intangible assets; smaller when tax differential is very small or when host country lacks thin-capitalization rules.
  - Consolidated-level regressions indicate aggregate multinational investment not clearly affected by TPRs → evidence consistent with relocation of investment across affiliates.
- Policy considerations:
  - Unilateral introduction of TPRs may distort international capital allocation and discourage adoption or induce leniency.
  - Binding international coordination can prevent harmful relocation effects but may not be Pareto-improving for all countries.
  - Coordination should cover other anti-avoidance measures (e.g., thin-capitalization rules) because avoidance channels can be substitutes; restricting only one channel can induce substitution toward others.
- Suggested further research:
  - Real effects of other anti-avoidance regulations (rules restricting interest deductibility, provisions against treaty abuse, general anti-avoidance rules).
  - Interaction between anti-tax avoidance rules and other tax policy parameters (e.g., corporate tax rates).

*Source unit — wp1869 - References (pages including Introduction, Sections II–VII, and Tables 1–8) — IMF working paper content provided in the input.*

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

### wp1869 - References

### I. Introduction: scope and motivation
- Tax-motivated profit shifting within multinational corporations (MNCs) has been a central international tax policy issue since the global financial crisis, notably in the G20/OECD initiative on base erosion and profit shifting (OECD 2015).
- Profit shifting defined: MNCs shift income from affiliates in high-tax jurisdictions to affiliates in low-tax jurisdictions to reduce overall tax liability.
- Empirical evidence cited:
  - German affiliates of MNCs have paid on average 27 percent less in taxes than comparable domestic German firms (Finke, 2013).
  - In the UK, taxable profits as a share of total assets reported by subsidiaries of foreign MNCs are on average 12.8 percentage points lower than those of comparable domestic standalone companies (Habu, 2017).
  - A survey finds a consensus semi-elasticity of reported profitability by MNCs with respect to international tax differentials of around -1.2 (Beer, De Mooij, and Liu, 2018).
  - Recent estimates suggest an annual loss in government revenue by between $100 and 650 billion globally, with disproportionately larger losses for developing countries (Crivelli, de Mooij, and Keen, 2016; OECD, 2015; UNCTAD, 2015).

### Transfer pricing, transfer pricing regulations (TPRs), and potential trade-offs
- Transfer prices: prices charged for transactions between related parties; manipulation of these prices is a common profit-shifting channel.
- Tax laws generally prescribe arm’s-length pricing, but information asymmetries enable MNCs to charge artificially low or high intra-group prices.
- Transfer pricing regulations (TPRs):
  - Describe allowed methods to determine arm’s-length prices.
  - Prescribe documentation requirements.
  - Set penalties for non-compliance.
  - Determine the probability of a transfer price adjustment.
- TPRs can raise effective tax burdens on MNCs, protecting domestic revenue and leveling the playing field vis-à-vis domestic companies (Fuest and others, 2013; OECD, 2013).
- From the MNC perspective, TPRs may increase tax uncertainty (IMF and OECD, 2017; Mescall and Klassen, 2014).

### Literature gap and research objective
- The relationship between TPR and MNC investment has received little theoretical and empirical attention; no direct empirical evidence existed regarding the investment effect of TPR at the time of this paper.
- Objective: explore the effect of TPR on multinational investment using theory and micro-level empirical analysis.

### Model and mechanism
- Simple model developed to infer likely impact of TPR on the scale of multinational investment.
- Key channel: TPR makes manipulation of transfer prices costlier, reducing profit shifting into the low-tax country.
  - This leads to a reduction in the optimal supply of intermediate inputs.
  - Return on investment in the foreign affiliate falls; TPR increases the cost of capital, reducing foreign affiliate investment.

### Empirical strategy and data
- Main dataset: micro-level dataset with information on both MNC and purely domestic affiliates covering 27 countries during 2006-2014.
- Supplemented with information on:
  - Introduction date of TPRs.
  - Indicator of TPR strictness.
- Identification strategies:
  - Difference-in-difference (DD) approach: compares change in investment by MNC affiliates to change in investment by purely domestic affiliates following TPR introduction in the local economy.
  - Panel regressions akin to Overesch (2009), Lohse and Riedel (2013), Buettner and Wamser (2013) to estimate changes in tax sensitivity of multinational investment due to TPRs.

### Main empirical findings and heterogeneous effects
- Average effects:
  - Investment in foreign affiliates is, on average, around 11 percent lower following the introduction of TPR, compared to investment in similar wholly domestic firms.
  - The “TPR-adjusted” corporate tax rate is 23 percent larger, i.e., MNC investment responses to tax rates are almost one quarter larger if TPRs are in place.
- Heterogeneity and deeper analysis:
  - Effect size increases with the strictness of the TPR.
  - Effect size decreases with the share of intangible assets of firms.
  - Effect is more robust at the intensive margin than at the extensive margin of investment.
  - Effects are larger if the tax differential grows, but the relationship is not monotonic; responses become smaller at very large tax differences.
  - Effect is larger and more robust in countries that also employ thin-capitalization rules.
- Evidence of relocation:
  - Using a different dataset of consolidated accounts, lower investment in MNC affiliates in TPR countries does not correspond to a similar reduction in total investment by the MNC group, interpreted as diversion of investment toward countries without TPRs.

### Contribution to literature
- Adds to cross-sectional studies of anti-avoidance legislation effects on firm behavior (reported profits, transfer prices, capital structure).
- Complements studies on profit-shifting opportunities and the broader literature on taxation and business investment by offering a new perspective on TPR effects on MNC investment.
- Relates to prior work on thin capitalization rules and investment but focuses specifically on transfer pricing regulation impacts.

### Policy implications and considerations
- Negative investment effects from TPRs may:
  - Make governments reluctant to introduce TPRs unilaterally.
  - Encourage adoption of more lenient regulations to mitigate adverse investment effects.
- Global coordination:
  - Binding global coordination can prevent unilateral backtracking on TPRs.
  - Restricting countries’ ability to set their own anti-avoidance rules could reinforce tax competition over corporate tax rates.
  - Coordination should cover other anti-avoidance rules (such as thin-capitalization rules) because otherwise TPRs might cause substitution into other avoidance channels.

### Paper structure (as presented)
- Section II: overview of TPR across countries.
- Section III: model illustrating how TPR can affect MNC investment into an affiliate.
- Section IV: data and sample selection for empirical analysis.
- Section V: research designs.
- Section VI: main results.

*Italic: Source unit — wp1869 - References (pages including Introduction) — IMF working paper content provided in the input.*

### Section VII elaborates on the results for total investment by the MNC group, based on consol-

### wp1869 - Section VII elaborates on the results for total investment by the MNC group, based on consol-

### II. TRANSFER PRICING REGULATION
- Current international taxation largely uses separate accounting; unconsolidated affiliate accounts terminate at the border and require transfer prices for related-party transactions.
- Arm’s-length principle: related-party prices should mimic prices between unrelated parties.
- Empirical evidence documents tax-motivated transfer mispricing; studies for the US, UK and France find significant responses of the price wedge to statutory CIT rate differentials (Bernard, Jensen, and Schott, 2006; Clausing, 2003; Cristea and Nguyen, 2016; Davies and others, 2018; Flaaen, 2016; Liu, Schmidt-Eisenlohr, and Guo, 2017; Vicard, 2015).
- Transfer Pricing Regulations (TPRs) aim to limit mispricing by:
  - Limiting allowable methods to establish arm’s-length prices;
  - Specifying documentation requirements;
  - Imposing transfer-pricing specific penalties and enforcement measures.
- Adoption history and scope:
  - Sweden introduced some form of TPR in 1928; modern TPRs began in early 1980s in Australia.
  - Today, almost 70 countries have TPRs in place.
  - Since 1995 many OECD countries base TPR on the OECD Transfer Pricing Guidelines.
  - In the study sample, 27 countries are considered; countries that introduced TPR between 2006 and 2014 include Bosnia and Herzegovina (2008), Finland (2007), Greece (2008), Luxembourg (2011), Norway (2008), and Slovenia (2007).
- Two policy variables used in empirical analysis:
  - Discrete dummy TPR_kt: equals 1 in years after country k introduced some TPR (data from Deloitte’s Strategy Matrix for Global Transfer Pricing, summarized in Mescall and Klassen (2014)); Panel A in Figure 1 gives 1928–2015 overview.
  - Index of TPR strictness tprisk: developed by Mescall and Klassen (2014), based on 15 detailed regulatory/enforcement features (12 regulatory, 3 enforcement). tprisk ranges between 1.26 and 5.17 in the sample; higher values = more stringent TPR.
    - The 15 features include: (1) advance pricing agreements allowed, (2) benchmark data available, (3) contemporaneous documentation required, (4) cost-contribution arrangement allowed, (5) commissionaire arrangement allowed, (6) foreign comparables allowed, (7) related party setoffs allowed, (8) taxpayer required to pay tax assessment before going to competent authority, (9) government identifies order of transfer pricing methods, (10) disclosure on tax return required, (11) self-initiated adjustment allowed, (12) transfer pricing documentation required; enforcement: (13) government discretion over penalty reduction, (14) government uses proprietary tax data to calculate a “revised” transfer price, (15) assessed degree of transfer pricing enforcement as a percentage based on experts’ 1 to 5 assessment (1.0 (5/5) most strict, 0.2 (1/5) least strict).
    - Perceived transfer pricing risk regression (from source): t_prisk = 1.27*** + 0.262** SecretComparables − 0.437*** APA + 0.614*** NoForeignComps + 0.102 NoSetoffs + 0.319** NoCCA + 0.062 PayTaxFirst − 0.326*** BenchmarkData + 0.008 SelfInitiatedAdj + 0.321** NoCommissionaire + 0.075 RelatedParty + 0.39*** ContemporaryDoc + 0.035 TPDoc + 0.296 Priority + 0.533*** PenaltyUncertainty + 2.46*** TPEnforceSvy + 0.011*** AgeofRules (***,**, * denote significance at 1%, 5%, 10%).
  - Panel B of Figure 1 shows cross-country and over-time variation in tprisk: dispersion has become smaller in recent years while the median remained similar.
- Alternative interpretation: tprisk can be viewed as a measure of tax uncertainty induced by TPR.

### III. THEORY
- Model setup:
  - Parent in home country h decides capital k to invest in foreign subsidiary in country s; investment financed by equity at cost r (exogenous).
  - Parent supplies intermediate input x to subsidiary; subsidiary production f(k,x) has decreasing returns in k and x (fk, fx > 0; fkk, fxx < 0), and fkx > 0.
  - Local market price for intermediate p; parent can set transfer price pT that may deviate from p.
  - Tax rates: τs in subsidiary country, τh in parent country. If τs < τh and repatriation is exempt, parent has incentive to shift income to subsidiary.
  - Expected cost per unit of deviation is c = β(pT − p)^2; parameter β is influenced by government via TPR (stricter TPR → higher β).
- Income expressions (as in source):
  - Subsidiary income: (1 − τs)[f(k,x) − pT x]. (Equation (1))
  - Parent earnings: (1 − τh)(pT − p)x + (1 − τs)[f(k,x) − pT x] − rk − β[pT − p]^2 x. (Equation (2))
- First-order conditions (from source):
  - (1 − τs) fk = r. (Equation (3)) — higher host tax increases cost of capital and reduces investment under decreasing returns.
  - fx = p + (τh − τs)(pT − p) + β(pT − p)^2/(1 − τs). (Equation (4)) — marginal product equals marginal cost adjusted for tax differential and shifting cost.
  - pT = p − (τh − τs)/(2β). (Equation (5)) — optimal transfer price is lower than arm’s-length price when τs < τh; magnitude depends on β.
  - Combining (4) and (5): fx = p − (τh − τs)^2/(4β(1 − τs)). (Equation (6))
- Implications:
  - Any tax rate differential τh − τs leads to lower required marginal return to x (∂fx/∂(τh − τs) < 0), implying a higher supply of intermediates x (∂x/∂(τh − τs) > 0) and, since fkx > 0, higher capital investment (∂k/∂(τh − τs) > 0).
  - Stricter TPR (higher β) reduces supply of x (∂x/∂β < 0), reduces marginal product of capital fk and reduces investment (∂k/∂β < 0).
  - Effect arises only for multinational structures with tax differentials; if τh = τs or parent and subsidiary in same country, increase in β has no implication for x or k.
- Main hypothesis: stricter TPR will reduce investment by multinational parents in foreign subsidiaries, but not by purely single-national parents in their domestic subsidiaries.

### IV. DATA
- Primary dataset:
  - Unbalanced panel of 101,079 unique companies in 27 countries for years 2006 to 2014, constructed from unconsolidated financial statements in ORBIS.
  - MNC affiliate definition: ultimate parent in different country and owns at least 50% shares.
  - Domestic affiliate definition: ultimate parent in same country owning at least 50% and all other affiliates in same country.
  - Main regression sample criteria: non-financial, non-utility affiliates with non-missing and non-zero sales, total assets and fixed asset values; exclude firms with missing industry info, less than three consecutive observations, countries with less than 1,000 observations; eliminate MNC affiliates located in same country as parent.
- Consolidated accounts sample (for Section VII analysis on spillovers):
  - 17,638 observations corresponding to about 2,024 distinct non-financial, non-utility parent companies in more than 60 countries for 2006–2014.
  - Consolidated accounts reflect total investment of the company group.
- Key firm-level variables and definitions:
  - Investment spending It = Kt − Kt−1 + depreciation, where Kt is reported book value of fixed tangible assets in year t.
  - Investment rate = It / Kt−1.
  - Intensive margin = log of investment spending.
  - Extensive margin = indicator for positive investment.
  - Sales = operating revenue; sales growth = ratio current-year to previous-year operating revenue minus 1.
  - Cash flow rate = current-year cash flow / lagged capital stock.
  - Profit margin = EBIT / sales.
  - All ratio variables winsorized at top and bottom 1 percentile.
- Country-level variables:
  - TPR (binary) and tprisk index constructed from Mescall and Klassen (2014) and extended to 2014 using Deloitte’s Transfer Pricing Strategic Matrix, 2014.
  - Thin capitalization rules (TCRs) info from De Mooij and Hebous (2017).
  - Macroeconomic controls from IMF WEO: GDP per capita, growth rate of GDP per capita, population, unemployment rate.
  - User cost of capital computed as r_real + (1 − A)/(1 − CIT), where r_real is real interest rate and A is net present value of depreciation allowances; statutory CIT rates and A from Oxford University Centre for Business Taxation.
  - Tax differential = absolute difference between host country and parent country statutory CIT rate.
  - Summary statistics presented in Table 2 (as cited).
- Alternative regression samples:
  - Section VI.D: smaller dataset excluding domestic affiliates to focus on tax sensitivity of multinational investment.
  - Section VI.B: matched sample of multinational and domestic affiliates matched on average turnover, turnover growth rate, number of workers, and total assets.

### V. EMPIRICAL SPECIFICATIONS
- Two identification strategies:
  - Difference-in-difference (DD) comparing MNC affiliates to domestic affiliates in same host country.
  - Traditional panel regression estimating change in tax sensitivity of multinational investment before/after TPR introduction.
- Main DD specification (as in source):
  - Investment_ikt = ai + dt + β (MNC_i × TPR_kt) + βx x_ikt + βz z_kt + ε_ikt. (Equation (7))
    - Investment_ikt is current-year investment spending divided by lagged capital stock.
    - MNC_i = 1 if firm i is part of a multinational group.
    - TPR_kt = 1 for years following TPR introduction in country k.
    - β_TPR is expected negative per theoretical prediction.
  - Controls and fixed effects:
    - Firm fixed effects ai; time dummies dt.
    - Firm-level controls x_ikt (size, growth prospects, financial constraints, profitability).
    - Country-level time-varying controls z_kt (GDP per capita, population, unemployment).
    - Preferred specification includes industry-year fixed effects, country-year fixed effects, and country-industry fixed effects to absorb industry-year shocks, macro shocks, and country-industry specific shocks.
- Additional modeling choices:
  - Include statutory corporate tax rate in host country or country-year fixed effects to control for concurrent tax reforms.
- Identification and placebo testing:
  - Parallel trends assumption tested via event-study specification:
    - Investment_ikt = ai + dt + sum_{l=-5}^{-1} β_l MNC_i × TPR_kt × Pre-TPR_l + sum_{n≥1} β_n MNC_i × TPR_kt × Post-TPR_n + βx x_ikt + βz z_kt + ε_ikt. (Equation (9))
    - Pre-TPR_l = 1 for l-th year before TPR; Post-TPR_n = 1 for n-th year after TPR; β_0 normalized to 0.
    - Parallel trends implies all pre-TPR β_l equal to each other (no differential pre-trend).
    - Table 3 presents full regression results (as cited).
  - Dynamics: coefficients on MNC_i × TPR_kt × Post-TPR_n indicate TPR has a large negative effect on investment in the first year after adoption; effect smaller but persists in later years, consistent with forward-looking investment decisions and lasting impact.
- Notes and caveats:
  - DD assumes away general equilibrium effects; if domestic investment expands in response to reduced multinational investment, β_TPR could underestimate TPR effect on multinational investment.
  - If TPRs also apply to domestic group transactions, TPR may affect domestic investment where domestic groups engage in mispricing via domestic tax-rate differentials (losses, loss carryforwards); then β_TPR captures differential effect and serves as lower bound on total business investment impact.

*Source: https://www.imf.org/-/media/files/publications/wp/2018/wp1869.pdf*

### 0.23 does not reject the null hypothesis; our parallel trends assumption therefore passes the

### wp1869 - 0.23 does not reject the null hypothesis; our parallel trends assumption therefore passes the

### B. Panel Regression — identification strategy
- Goal: infer the impact of the introduction of TPR on MNC investment by estimating tax-sensitivity of MNC investment with and without TPR.
- Key regression for semi-elasticities (Eq. (10)):
  - ln(Investmentikt) = ai + dt + βtax CITkt + βTPRtax × CITkt × TPRkt + βx xikt + εikt
  - Interpretation:
    - βtax ≡ ∂lnInvestment/∂CIT is the semi-elasticity of MNC investment w.r.t. CIT in the absence of TPR.
    - After TPR, semi-elasticity becomes γtax = βtax . (Text preserves original notation: γtax = βtax .)
    - A change in semi-elasticity can be interpreted as TPR increasing the effective CIT in proportion to βTPRtax/βtax.
    - The “TPR-adjusted” corporate tax rate (or cost of capital) is (1 + βTPRtax/βtax) × CITt.

### VI. RESULTS — overview
- Structure:
  - Direct DD evidence on reduction in MNC investment after TPR introduction.
  - Robustness checks and heterogeneity.
  - Estimation of the “TPR-adjusted” semi-elasticity of multinational investment.

### A. Baseline DD findings (Table 4)
- Main estimates:
  - Column (1): DD coefficient = -0.049, significant at the 1 percent level.
  - Column (2): After adding host country controls, DD coefficient = -0.041.
  - Preferred specification (Column (6)): DD estimate = -0.041, significant at the 1 percent level.
    - Interpretation: implementation of TPR reduces the investment rate (investment as a percentage of fixed assets) by multinationals by 4.1 percentage point.
    - Given average gross investment per dollar of fixed asset = 35.9 cents, this corresponds to 11.4 percent reduction in investment for affiliates in the sample.
  - Column (7): interaction between MNCi and tpriskkt (strictness index) yields coefficient = -0.072, significant at the 1 percent level.
    - For a country with tprisk = 3.0, reduction = 0.216 percentage points.
    - For a country with tprisk = 5.17, reduction = 0.36 percentage points.
- Additional note: tprisk not available for countries without TPRs; Column (7) uses variation only across countries with TPRs.

### B. Robustness (Table 5)
- Alternative specifications and samples:
  - Column (1): exclude affiliates with parent in a country with worldwide tax system — TPR result unchanged.
  - Column (2): cluster standard errors at host country level — TPR result unchanged.
  - Column (3): dependent variable = investment rate winsorized at top and bottom 2.5 percentile — DD estimate ≈ 0.018, significant at the 1 percent level; not statistically different from Column (6) of Table 4 (winsorized at 1 percentile).
  - Column (4): matching DD (Mahalanobis) on average turnover, turnover growth, employment, total assets — estimate remains positive and significant at the 1 percent level; coefficient size similar.
  - Additional check (text): dropping Luxembourg affiliates yields DD = -0.043, significant at the 1 percent level.

### C. Heterogeneous responses (Table 6)
- (a) Extensive vs Intensive margin
  - Column (1): Linear probability for positive investment — coefficient small and insignificant → TPRs do not affect the likelihood to invest.
  - Column (2): Intensive margin (log of investment, excluding negative investment) — DD coefficient positive and highly significant → reductions driven by firms that were already investing.
- (b) Size of tax differential
  - Specification: interact MNC × TPR × quartile indicators of tax differential (Eq. (11)).
  - Results (Column (3)):
    - Bottom quartile: response negative but insignificant.
    - 2nd quartile: response larger and highly significant.
    - 3rd and 4th quartiles: coefficient smaller and less significant, but remains negative and significant at 10 percent in higher quartiles.
- (c) Intangible asset intensity
  - Specification (Eq. (12)) introduces IntangSharei (average intangible fixed assets / total assets).
  - Table 6 Column (4):
    - Main interaction MNCi × TPRkt negative.
    - Three-way interaction with IntangShare small, positive, and highly significant.
    - Quantitative magnitude: difference between IntangShare = 0 and IntangShare = 1 is 0.2 percentage points (investment effect drops from -3.2 percentage points to -3.0 percentage points).
- (d) Interaction between TPRs and TCRs (thin-capitalization rules)
  - Split sample by presence of TCR in host country:
    - Countries without TCR (Column (5)): DD coefficient = -0.013, insignificant.
    - Countries with TCR (Column (6)): DD coefficient ≈ three times larger and significant at 1 percent.
  - Interpretation: transfer mispricing and debt shifting are likely substitutes; effectiveness of TPR depends critically on presence of other anti-avoidance measures.

### D. TPR-adjusted tax elasticity (Table 7)
- Sample: multinational affiliates only; regression based on Eq. (10).
- Column (1): without TPR, a one percentage point lower statutory CIT rate increases investment (share of total assets) by multinationals by 0.83 percentage point.
  - In presence of TPR, sensitivity increases by 0.36 percentage point to 1.19 (absolute terms).
- Column (2): replacing CIT with cost of capital (COC) yields qualitatively similar result, though COC coefficient imprecise without TPR.
- Column (3): dependent variable = log fixed tangible assets (semi-elasticity interpretation):
  - Estimated semi-elasticity of fixed capital assets slightly larger than one without TPR and highly significant.
  - Presence of TPR increases tax effect by 0.24 to overall semi-elasticity = 1.26.
  - Interpretation: a 1 percentage-point increase in CIT reduces MNC investment by ≈ 1 percent without TPR; after TPR, corporate tax rates matter about one quarter more.
  - Following interpretation in Section V.B: introduction of TPR corresponds to a “TPR-adjusted” CIT rate that is 23 percent larger than without TPR.
- Column (4): replacing CIT with cost of capital implies “TPR-adjusted” cost of capital is 15 percent larger than without TPR.

### VII. Effect on total MNC investment (Table 8)
- Question: do reductions in affiliate fixed capital investment reflect (i) reduction in total MNC investment or (ii) relocation of investment within MNC group?
- Approach: DD on consolidated parent-company accounts; Investmentikt = worldwide investment by MNC group; TPRkt = dummy if parent country has TPR.
- Results:
  - Column (1): baseline DD coefficient positive and significant at 1 percent.
  - Column (2): remains significant with country-level characteristics.
  - Column (3): becomes insignificant when including country-year fixed effects.
  - Columns (4) and (5): remain insignificant with industry-year and industry-country fixed effects; interacting with parent statutory CIT rate does not change basic finding.
- Interpretation: absence of clear negative effect on consolidated investment suggests negative effect on foreign affiliates is likely due to relocation of investment to other affiliates rather than a global reduction in MNC investment.

### VIII. Conclusions — main findings and policy implications
- Empirical contributions:
  - Data: unconsolidated accounts 2006–2014; seven of 27 sample countries introduced TPR during period.
  - Core estimate: introduction of TPRs reduced investment in multinational affiliates by more than 11 percent on average.
  - Stricter TPR regimes induce larger reductions in MNC affiliate investment.
  - Reduction larger for firms less intensive in intangible assets; smaller when tax differential is very small or when host country lacks thin-capitalization rules.
  - Consolidated-level regressions indicate aggregate multinational investment not affected by TPRs → evidence consistent with relocation of investment across affiliates.
- Policy implications highlighted in the source:
  - Unilateral introduction of transfer pricing regulation will distort international allocation of capital; negative investment effects can discourage adoption or induce leniency.
  - Binding international coordination can prevent harmful relocation effects, but may not be Pareto-improving for all countries.
  - Broad coverage of anti-avoidance measures is important because avoidance channels may be substitutes; restricting only one channel can induce substitution toward others.
- Suggested further research (left by authors):
  - Real effects of other anti-avoidance regulations (rules restricting interest deductibility, provisions against treaty abuse, general anti-avoidance rules).
  - Interaction between anti-tax avoidance rules and other tax policy parameters (e.g., corporate tax rates).

*Source: wp1869 - 0.23 does not reject the null hypothesis; our parallel trends assumption therefore passes the*

### REFERENCES

### REFERENCES

### Key empirical findings from regression tables and figures
- Transfer pricing regulations (TPRs) are associated with negative investment responses by multinational affiliates:
  - Baseline DD estimates (Table 4): MNC_i × TPR_kt coefficients range: -0.049***, -0.041***, -0.024***, -0.025***, -0.025***, -0.041*** (standard errors reported in table).
  - In Column (2) of Table 4, MNC_i × tprisk_kt = -0.072*** (0.025).
- Robustness checks (Table 5) confirm negative investment effects:
  - MNC_i × TPR_kt = -0.041*** (0.012), -0.041*** (0.010), -0.018*** (0.006), -0.036*** (0.013) across robustness specifications.
- Heterogeneous effects (Table 6):
  - The baseline interaction MNC_i × TPR_kt shows mixed magnitudes across subsamples: 0.002; -0.037**; -0.032***; -0.013; -0.036*** (standard errors in table).
  - Interaction with intangible asset share: MNC_i × TPR_kt × IntangShare_i = 0.002*** (0.000).
  - Tax-differential quartiles produce varying estimates, including one significant negative estimate: MNC_i × TPR_kt × Quartile TaxDiff,2 = -0.052*** (0.015).
  - Columns (5)-(6) report responses in countries without and with thin-capitalization rules (TCR): results reported in table (including significance at 1%, 5%, 10% levels).
- Taxes and cost of capital effects (Table 7):
  - CIT_kt = -0.834*** (0.198) and -1.023*** (0.141) in Columns (1)-(2).
  - CIT_kt × TPR_kt = -0.356*** (0.073) and -0.238*** (0.060) in Columns (1)-(2).
  - COC_kt = -0.462 (1.412) and -8.594*** (1.211) in Columns (1)-(2).
  - COC_kt × TPR_kt = -1.736*** (0.425) and -1.260*** (0.306) in Columns (1)-(2).
- Worldwide (MNC-group level) investment responses (Table 8):
  - MNC_i × TPR_kt coefficients reported as 0.056*** (0.019), 0.049** (0.019), 0.029 (0.024), 0.031 (0.024) across columns (1)-(4).
  - Interaction with CIT: MNC_i × TPR_kt × CIT_kt = 0.125 (0.095) in Column (5).
  - R2 values and sample sizes reported per column in table.

### Key summary statistics (Table 2)
- Firm-level statistics (mean, Std Dev, Median, P10, P90):
  - Investment spending ($1,000): 1,725; 30,589; 70.73; -472,266.
  - Fixed asset ($1,000): 11,528; 133,200; 689.49; 2,714; 14,167.
  - Investment rate (I_t/K_{t−1}): 0.45; 1.07; 0.15; -0.06; 1.06.
  - Operating revenue ($1,000): 54,055; 440,600; 6,812; 681; 83,028.
  - Cash flow rate: 2.12; 7.08; 0.39; -0.25; 5.23.
  - Profitability: 0.08; 0.16; 0.06; -0.03; 0.23.
  - Sales Growth Rate: 0.06; 0.30; 0.03; -0.26; 0.41.
- Country-level statistics:
  - CIT rate (%): 27.34; 5.79; 28.00; 19.00; 33.33.
  - Tax differential (in absolute %): 4.79; 6.22; 1.67; 0; 14.50.
  - Cost of Capital: 0.07; 0.01; 0.07; 0.06; 0.08.
  - Population (million): 35.01; 125.98; 38.14; 5.40; 63.38.
  - Unemployment rate (%): 9.34; 4.88; 8.10; 5.33; 16.18.
  - Exchange rate (rel to USD): 29.21; 154.93; 0.75; 0.68; 7.65.
  - GDP per capita (constant USD): 40,579; 20,855; 42,249; 12,977; 60,944.
  - GDP growth rate (%): 1.02; 2.92; 1.26; -2.94; 4.18.

### Additional empirical checks and diagnostics
- Common trends test between treated and control groups (Table 3):
  - Pre TPR Year coefficients: Pre TPR Year 5 = 0.147 (0.199); Pre TPR Year 4 = 0.191 (0.142); Pre TPR Year 3 = 0.145 (0.129); Pre TPR Year 2 = -0.044 (0.034); Pre TPR Year 1 = 0.008 (0.026).
  - Post TPR Year coefficients: Post TPR Year 1 = -0.049** (0.023); Post TPR Year 2 = 0.001 (0.015); Post TPR Year 3 = -0.036*** (0.013); Post TPR Year 4 and more = -0.015** (0.006).
  - Joint test that all pre-reform β_l coefficients are equal: p−value = 0.228.
- Sample composition (Table 1):
  - Number of companies in main estimation sample (2006–2014) by ownership types for selected countries (Total / MNC / Domestic Company Group):
    - Austria: 5,643 / 4,565 / 1,078
    - Belgium: 37,417 / 25,695 / 11,722
    - France: 144,662 / 70,158 / 74,504
    - Germany: 27,752 / 19,588 / 8,164
    - Spain: 100,403 / 39,720 / 60,683
    - Sweden: 91,067 / 20,446 / 70,621
    - United Kingdom: 63,053 / 44,894 / 18,159
    - (Full country list and counts presented in table.)

### Notes on measures and estimation (extracted from table notes)
- Investment measure: gross investment scaled by book value of fixed capital asset in (end of) previous year.
- Main DD specification: interaction MNC_i × TPR_kt with affiliate fixed effects, year fixed effects, and additional controls.
- Affiliate-level controls: lagged turnover, lagged turnover growth rate, cash flow scaled by lagged asset, and lagged profit margin.
- All firm-level ratio variables are winsorized at top and bottom 1 percentile.
- Standard errors: heteroskedasticity-robust and clustered at firm level (unless otherwise specified).
- Significance notation: ***, **, * denote significance at the 1%, 5% and 10% levels, respectively.

*Source: wp1869 - REFERENCES (pages and tables as provided).*

---


_Source: https://www.imf.org/-/media/files/publications/wp/2018/wp1869.pdf_