Raffael Speitmann - Corporate Tax Microsimulation for the EU

  • Presenting author: Raffael Speitmann (European Commission, Joint Research Centre)

  • Authors: Raffael Speitmann, Andrzej Stasio, Fotis Delis

  • Session: A02B - Tax-Benefit - Monday 16:30-18:00 - Marietta-Blau Hall

This paper introduces a novel corporate tax microsimulation model designed for policy analyses, offering solutions to the challenges posed by limited micro-tax data and the underrepresentation of small firms. While microsimulation models have been widely used in tax-benefit analysis (e.g. EUROMOD), those focusing specifically on firms have been less common (noteworthy exceptions include ZEW TaxCOMM, the University of Göttingen ASSERT model and the ISTAT-MATIS corporate tax model). However, the growing interest from policymakers in firm-level models has prompted the development of this framework, which leverages financial accounting data to estimate the tax base for corporate income tax at the firm-level. One of the key contributions of this microsimulation model is its ability to address the discrepancies between profit and tax expense as presented in financial statements and what is reported to tax authorities. While accounting standards ensure consistency in financial reporting, tax treatment can vary significantly for similar activities. By estimating taxable profits at the firm level, the model captures temporal and permanent book-tax differences, enabling a deeper understanding of the heterogeneous impacts of corporate tax reforms on different firms. Our microsimulation methodology takes inspiration from existing microsimulation models but at the same time presenting also new developments. First, we rely on data imputation methods to generate firm-specific accounts when these are not directly observable. The scarce availability of certain firm-level tax data represents a severe constraint in the literature so far. For example, ZEW TaxCOMM includes 80.000 German firms from which only 20 percent provide data for all the years of the simulation. To impute missing observations of our target variables, we rely on advanced imputation methods, such as Predictive Mean Matching (PMM). PMM introduces some random component into the estimation so that the model simulations provide a (narrow) range of statistically plausible results. The imputation process also utilizes nearest-neighbour matching to preserve desirable properties and avoid implausible values, allowing for the estimation of taxable profits for a broader range of firms in multiple countries. This approach allows us to model tax provisions related to tax depreciations of fixed assets, exempt dividend income, limitations to interest deductibility, tax loss carry-forwards, and tax group consolidation. In particular, we derive the distribution of tangible and intangible fixed assets by type of asset, the distribution of financial revenue by its source, and accumulated profit or losses. Second, the model adopts a data-driven approach to accurately model historical investments and estimate current period depreciations. Unlike previous models, which made assumptions about constant past capital stock or growth rates, this approach aligns the simulation period with the available historical firm-level asset stock data. Investments are derived backward from the current period to the oldest available period, allowing for precise calculation of investment amounts for firms with complete time-series data. For firms with incomplete or unavailable historical data, a non-parametric simulation approach based on Blouin et al. (2010) is utilized to model historical investments. Third, we use entropy balancing and reweighting to ensure that our sample matches the underlying population of corporations. Publicly available accounting data suffers from underrepresentation of small firms. A recent study from the OECD suggests that firms in ORBIS are disproportionally large, old and productive, even within their size class (Bajgar et al., 2020). By using these techniques, the microsimulation results from the observed sample are extrapolated to the total business population, effectively smoothing out any structural differences. The paper describes preliminary results for selected EU Member States. Specifically, we validate the model against official tax revenue statistics at the country-year level and separately using country-by-country reporting data (CbCR) from the OECD to validate our results for large multinational groups. The model’s flexibility allows for its extension to other countries, offering valuable insights for corporate tax policy analysis and serving as a foundation for further research in this field.