part 1: What have we done with the OECD DATA: Estimating profit shifting

Javier Garcia-Bernardo (TJN/UvA)

Petr Janský (Charles University)

Method 1: Hines-Rice alternatives

  • Data:
    •  Sub-groups with positive profits, for 13 reporting countries with enough data quality
  • Method:
    • Calculate profit according to Hines-Rice
    • Compare with hypothetical tax rate of 25%
    • Redistribute the remaining profits based on:
      • 50% unrelated party revenue , 25% employees, 25% wages
      • Revenues/Employees/Wages estimated based on model
    • Scale up using the share of GDP (~50% of total)

Method 2: Misalignment

  • Data:
    •  All sub-groups, all countries
  • Method (1000 times):
    • Estimate revenues and employees for:
      • Domestic operations (simple linear model) - R-square ~95%
      • Each pair of countries  (complicated model based on Gradient Boosting) - R-square ~50% (out-of-sample prediction)
      • --> Those numbers used to scale up the revenues/employees/profits in a country
      • --> The models are fitted using different bootstrap samples
    • The median estimation is that ~50% of the revenue/employees are missing, but more information is missing for small countries, because they often get groupped together (this is key!)
    • Calculate profit misalignment, using the same equation (50% sales, 25% emp, 25% wages)


High level:

- Extreme non-linearity of the semi-elasticity of tax on profits

- Level of profit shifting: Around $1 trillion

- Level of tax revenue loss: Around $250 billion

- Large countries lose 15-75% of their profits

Results by income group

Results by income group and region

PARt 2: discussion about methodology/results

general questions about data

What are the potential error margins of the imputation of missing data?

Especially for developing countries.

Janský,  Garcia-Bernardo & Tørsløv "Multinational Corporations and Tax Havens: Evidence from Country-by-Country Reporting"(ITAX, 2020)

Potential impact: 5-12 billion (2-5% of total)

- Why only 10-30% reallocation (instead of 100%)

- Why based only on sales? (instead of adding also employees)


Pillar 1

Pillar 2

  • Direct effect: 23-43 billion.
    • Why redistributed to shareholders instead to market jurisdictions?
  • Indirect effect through reduced profit shifted: 19-28 billion.
    • Based on assumption of linear semi-elasticity?
    • This is expected to increase profits in investment hubs and foster a race to the minimum tax. How to fix this?

Estimating profit shifting

By Javier GB

Estimating profit shifting

  • 871