Introduction

• A tax evasion strategy often overlooked by the literature is cost overreporting (Almunia and Lopez-Rodriguez 2018)

• Cost overreporting arises when firms acquire false invoices to claim additional tax deductions on value-added (VAT) and corporate income taxes (CIT) (Carrillo et al. 2022)

  • “fake invoicing”, “ghost firms”, “invoice mills”, or “missing traders”

• According to the OECD (“Technology Tools to Tackle Tax Evasion and Tax Fraud” 2017), cost overreporting spreads globally leading to significant tax revenue losses for governments

The problem

• Detecting cost overreporting is a difficult task for tax authorities (Slemrod 2019)
  • Firms have incentives to hide their fraudulent activities
  • Tax authorities have limited resources to audit all firms
  • Tax authorities have limited information about firms’ activities

The question

• How can we identify and measure this type of tax evasion?
• How do firms adjust their misreporting due to changes in fiscal policies?
• What are the tax revenue losses due to cost overreporting?

What am I doing to tackle the question?

• To address these questions I will use production functions (Olley, Pakes, and Steven 1996; Levinsohn and Petrin 2003; Ackerberg, Caves, and Frazer 2015; Gandhi, Navarro, and Rivers 2020)

• Intuitively, output is a function of true inputs; and indirect measure

• If we can somehow invert it, the difference between the true inputs and reported inputs is a measure of misreporting

Where does Productivity come in?

• The problem of using production functions to detect tax evasion dealing with productivity

  • heterogeneous, unobserved

• Why is this a problem? Both productivity and misreporting are unobserved and both affect the observed output

• Intuitively, for a given level of output, high input utilization might be due to cost overreporting or because of a negative productivity shock

Contribution

• I provide a novel strategy to estimate tax evasion through cost overreporting using production functions

• Second, I also formally show that ignoring cost overreporting leads to downward biased productivity and production function parameters

• I demonstrate how to recover productivity in the presence of tax evasion

Today’s talk

• For the sake of time, I will provide evidence of cost overreporting using data from manufacturing firms in Colombia 1981-1986

• Roadmap:

  1. Measure of tax evasion
  2. Minimal context for Colombia
  3. Evidence

Setting

• We observe output \(Y_{it}\), reported inputs \(K_{it},L_{it}, M_{it}\), and output \(P_t\) and intermediate input prices \(\rho_t\)

• Firms overreport their true intermediate inputs \(M_{it}=M^*_{it}\exp(e_{it})\) to evade taxes

• We can’t use \(Y_{it}=G(M_{it},K_{it},L_{it})\exp(\omega_{it}+\varepsilon_{it})\)

Strategy

• To fix ideas, assume the production function is Cobb-Douglas,

• \[ G(M_{it}, K_{it}, L_{it})\exp(\omega_{it}+\varepsilon_{it})=M^{*\beta}_{it}K_{it}^{\alpha_K}L_{it}^{\alpha_L}\exp(\omega_{it}+\varepsilon_{it}) \]

• Using the FOC of the cost-minimization problem of the firm, we can get \[ \ln\left(\frac{\rho_t M_{it}}{P_{t}Y_{it}}\right)+e_{it}=\ln\beta + \ln \mathcal{E}- \varepsilon^Y_{it} \]

FOC

Identifying tax evasion

• In the following, I will use \(s_{it}\equiv \ln\left(\frac{\rho_t M^*_{it}}{P_{t}Y_{it}}\right)+e_{it}\)

• Intuitively, changes in incentives to evade taxes will affect \(e\) but not \(\varepsilon\)

• I use the 1983 tax reform in Colombia as a natural experiment to identify tax evasion

Colombia Minimal Context

• The relevant corporate taxes for input overreporting in Colombia during this period are
  • the Corporate Income Tax (CIT), varies by type of juridical organization, and
  • the Sales Tax, or Value-Added Tax (VAT), varies by industry
• Three main types of JO: Corporations, Limited Liability Companies (LLCs), and Proprietorships
  • Corporations have the lowest incentives to evade: Closely monitored by the Superintendent of Corporations, free-tradable shares valued by the market, and CIT is paid on distributed dividends (not on profits)

J.O. in Colombia

The 1983 Tax Reform

• The 1983 reform changed the VAT rate differently for different industries (Perry and Triana 1990)
  • Some increased (Wearing Apparel, Metal Products, and others)
  • Some exempt industries remained exempted (Food Products)
• The 1983 tax reform also introduced changes in CIT for Proprietorships and LLCs
  • LLC CIT went from 20 to 18%;
  • Proprietorships 8% increase on average, but max tax rate reduced from 56 to 49%

Tax Reforms

What data I’m using

• The Colombian dataset comes from the Annual Survey of Manufacturing (EAM) and contains information about manufacturing firms with more than 10 employees from 1981 to 1991.

• Besides the information on output, intermediates, capital, and labor, the dataset includes the type of juridical organization and the sales taxes, and metropolitan area.

Summary Statistics

Empirical Strategy

• Given the changes in sales taxes, it’s natural to think on a diff-in-diff approach, By industry, using Corporations as the control group

• However, the reform also affected the CIT, so the result would be ambiguous

• Hence, I use a triple-diff approach: Corporations vs Not Corporation vs Exempt Industry

Identification

Discussion

• The results show that firms reacted to the changes in sales taxes by adjusting the log of the inputs cost share of revenues
• Under the model’s assumptions, this implies that firms were evading taxes by overreporting inputs
• In particular, non-corporation firms in industries where the sales tax increased responded by increasing tax evasion by overreporting inputs