## wp18257

## Source details

**Canonical URL:** [wp18257](https://www.imf.org/-/media/files/publications/wp/2018/wp18257.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/wp/2018/wp18257.pdf.md)
- [Structured JSON version](/-/media/files/publications/wp/2018/wp18257.pdf.json)

---

### Introduction: debt bias and macrofinancial risks
- Tax systems that allow deduction of interest payments but rarely deduct dividend distributions create a debt bias by making debt artificially cheaper than equity.
- Debt bias encourages higher leverage, which:
  - Can reach "50 percent of GDP" in some countries (post-crisis evolution of NFC debt).
  - May dampen long-term growth by reducing corporate investment incentives.
  - Raises financial stability risks via nonperforming loans, fiscal costs from bailouts, and exchange-rate/contagion risks where corporate debt is in foreign currency.
- Prior literature documents significant debt-bias effects, mostly for large or multinational firms; this paper extends the inquiry to small and medium-sized enterprises (SMEs).

### Scope, data coverage, and sample construction
- Data source and scale:
  - Firm-level data from Bureau van Dijk’s Orbis database (combined vintages per Gal and Hijzen (2016)).
  - Orbis contains around "22.5 million" individual firms overall.
  - The study dataset contains "14 million firm-year observations", "almost 99 percent" of which are SMEs.
- SME definition and sample restrictions:
  - SME defined as firms with fewer than "250 employees".
  - Restricted to advanced economies and nonfinancial sectors.
  - Firms must report number of employees and operating revenues.
  - Truncate the top "1 percent" of the distribution for each financial variable, including leverage.
  - Drop firms with less than "US$1,000" in assets and those that have fewer than "10 employees".
- Reporting basis:
  - Focus on unconsolidated reports; for firms reporting both consolidated and unconsolidated accounts, unconsolidated accounts are selected.
- Tax variable:
  - Statutory (marginal) corporate income tax (CIT) rate on the highest tax bracket used (from an internal IMF Fiscal Affairs Department database).
- Sample coverage:
  - More than 14 million observations.
  - Covers 24 advanced economies over 1995–2013.
  - Sample is unbalanced along sector, time, and country dimensions.

### Hypotheses and theoretical framing
- Hypotheses:
  - Hypothesis 1: The higher the corporate income tax rate the firm faces, the more leveraged its balance sheet.
  - Hypothesis 2: The smaller the firm, the less responsive its leverage is to the corporate income tax rate.
  - Hypothesis 3: The higher a firm’s tangibility, the more responsive its leverage is to the corporate income tax rate.
  - Hypothesis 4: The higher a firm’s revenue and the higher its revenue volatility, the less responsive its leverage is to the corporate income tax rate.
  - Hypothesis 5: The higher a firm’s capital intensity of production, the less responsive its leverage is to the corporate income tax rate.
- Theoretical framing:
  - Firms choose fixed-rate debt against uncertain asset returns; higher investment increases leverage and bankruptcy risk.
  - Interest deductibility lowers the cost of debt and therefore shifts the optimal leverage upward (theoretical model in Appendix A).

### Variables, measurement, and stylized facts
- Preferred leverage indicator: debt-to-asset ratio (chosen for data quality). Alternative measures: debt-to-equity, debt-to-earnings.
- Firm-level controls and definitions:
  - Size proxied by total assets (logarithm used in regressions; square of log included).
  - Collateral proxied by tangible assets; tangible assets also used as a share of total assets (TFA-TA).
  - Turnover and earnings proxied by revenue (REV-TA) and cash flow (CASHFLOW-TA).
  - Turnover volatility proxied by standard deviation of revenue (REV-VOL).
  - Capital intensity proxied by CAPITAL-TA (inverse of labor share).
  - Except for total assets, all series normalized as a share of total assets.
  - Debt defined as total liabilities; assets and liabilities used at nominal/book value (Orbis).
- Stylized bivariate patterns:
  - Leverage increases with firm size overall, but relationship is not linear; largest companies (top quartile; total assets > US$9.4mn) show a negative correlation between leverage and asset size.
  - Firms with more collateralizable/tangible assets are more leveraged.
  - More capital-intensive production associated with higher leverage.
  - Higher cash flows, higher revenue, and higher revenue volatility tend to be associated with lower leverage.
  - Leverage fairly homogenous across 11 sectors; services slightly less leveraged; real estate above-average debt-to-asset ratios.
  - Median leverage by country-year tends to decrease slightly with higher CIT rates (driven by outliers with very high CIT rates), but regressions with fixed effects find a positive relationship between CIT rate and leverage once controls are included.

### Empirical strategy and identification
- Approach: difference-in-difference style interaction of firm-level characteristics with country-level CIT rate.
- Main regression specification (dependent variable: debt-to-asset ratio):
  - D_{i,t,c} = α D_{i,t−1,c} + β1 X_{i,t−1,c} + β2 τ_{c,t} × x_{i,t−1,c} + FEs_{c,s,t} + ε_{i,t,c}
  - Coefficient of interest: β2 (interaction between CIT rate and a specific firm-level variable x).
- Controls and estimation:
  - Firm characteristics in X: Total assets (log) and its square, share of tangible assets, cash flow, revenue (and volatility), capital intensity — all scaled by assets and lagged by one period.
  - Fixed effects: sector-time, country-time, and country-sector (S*T + C*T + S*C).
  - Pooled OLS with cluster-robust standard errors clustered at country, industry and year levels (C*S + C*T + S*T).
  - Lagged dependent variable included to capture persistence.
  - Difference and system GMM explored but infeasible given large unbalanced dataset.

### Main empirical findings and magnitudes
- General result:
  - Strong evidence that higher CIT rates increase firms’ leverage via an implicit subsidy to debt (interest deductibility).
  - Firms respond to tax incentives by increasing leverage; interaction coefficients largely sizeable and significant.
- Heterogeneous effects (confirming hypotheses):
  - Debt bias stronger for larger firms (Hypothesis 2).
  - Debt bias stronger for firms with higher share of tangible assets (Hypothesis 3).
  - Debt bias weaker for firms with volatile or high revenue (Hypothesis 4).
  - Debt bias stronger for more capital-intensive firms (Hypothesis 5).
- Persistence and controls:
  - Lagged leverage coefficient: LEVERAGE (L) = 0.842*** with standard error [0.003] (Observations 3,494,804; Adj. R-squared 0.762).
  - Significant direct effects: size, cash flow, revenue and volatility, capital intensity, and asset composition on leverage.
- Magnitude and illustrative calculations:
  - Interaction coefficient on total assets: 0.907.
  - Long-run persistence parameter used in examples: 0.842 (1 − α = 1 − 0.842).
  - Using these values, debt bias explains about 20 percentage points of leverage variations among firms in the long run (calculation shown: 0.907/(1−0.842) × 0.28 × 12.74 = 20.47).
  - Debt bias explains about 5 percentage points of variation among firms with different tangible assets.
  - Revenue, revenue volatility, and capital intensity explain about 11, 4, and 11 percentage points of debt bias variation within firms, respectively.
  - At the sample-mean CIT rate of 28 percent, the interaction coefficient on total assets implies debt bias contributes approximately 24 percentage points of leverage in the median firm (alternative long-term calculation result reported as 23.87).
  - Debt bias magnitude noted to explain up to 27 percent of leverage.
- Quantile regressions:
  - Results particularly pronounced for the most leveraged firms (decile quantile regressions).
  - TA x CIT and TFA-TA x CIT positive effects driven by highest-leverage quantiles; REV-TA x CIT and REV-VOL x CIT negative effects driven by most leveraged firms.
  - Example quantile patterns:
    - LEVERAGE (L) ranges across quantiles: 0.673*** [0.001] at 0.1 to 0.976*** [0.001] at 0.75.
    - TA x CIT: -0.204*** [0.018] at 0.1 up to 0.622*** [0.048] at 0.9.
    - TFA-TA x CIT: -11.170*** [0.357] at 0.1 and 28.339*** [1.356] at 0.9.
    - CASH-TA x CIT varies (e.g., 10.123*** [1.134] at 0.1; -2.035 [4.614] at 0.9).
  - Quantile regressions run on an almost-balanced subsample requiring 10 consecutive years, yielding just under 1 million firm-year observations (Observations 1,362,563 in quantile panels).

### Robustness checks and alternative specifications
- Debt maturity:
  - Replacing total debt with long-term debt yields qualitatively similar results; debt bias appears more important for long-term financing decisions.
  - Exception: interaction coefficient on cash flow becomes negative and significant for long-term debt.
- Alternative leverage measures:
  - Using debt-to-equity (LEVERAGE3) does not alter findings qualitatively; sample size increases; overall fit slightly worse; tangibility interaction can flip sign.
- Subsamples and trimming:
  - European-only sample: LEVERAGE (L) = 0.833*** [0.003] (Observations 3,162,831); TA x CIT 0.762*** [0.123].
  - Trimming top two percent instead of one does not qualitatively change results.
  - Two-employee threshold sample: Observations 8,143,290; LEVERAGE (L) 0.828*** [0.004]; notable CASHFLOW-TA x CIT: 16.534*** [3.001].
  - Large firms only: Observations 994,978; LEVERAGE (L) 0.879*** [0.003]; TA (L) -1.456*** [0.406].
  - Balanced sample (minimum 10 consecutive years): Observations 1,362,563; LEVERAGE (L) 0.856*** [0.003]; CASHFLOW-TA x CIT: 17.949*** [4.215].
- Orbis vintages and time coverage:
  - Including more recent observations (Orbis up to 2015/2016) broadly robust though pre-2005 coverage limited.
  - Orbis-up-to-date version noted to cover 1990 to 2016 but sample strongly unbalanced.
- Sectoral and binary mappings:
  - Regressions by each of 11 nonfinancial sectors show results broadly consistent across sectors.
  - Mapping firm characteristics to binary dummies (above/below median) does not qualitatively change conclusions.
- Tables and summary statistics (select):
  - LEVERAGE: mean 25.74; p50 20.25; sd 22.31; # of obs. 7,367,751.
  - LEVERAGE2: mean 21.92; p50 15.48; sd 20.70; # of obs. 5,956,281.
  - TA: mean 14.73; p50 14.85; sd 2.253; # of obs. 10,973,463.
  - TFA-TA: mean 0.251; p50 0.183; sd 0.229; # of obs. 10,951,096.
  - REV-TA: mean 1.886; p50 1.587; sd 1.313; # of obs. 10,973,463.
  - CASHFLOW-TA: mean 0.092; p50 0.07; sd 0.079; # of obs. 8,038,252.
  - REV-VOL: mean 0.642; p50 0.454; sd 0.612; # of obs. 11,097,374.
  - CAPITAL-TA: mean -0.488; p50 -0.316; sd 0.524; # of obs. 9,002,220.
- Country and sector coverage examples (Table 14):
  - Total sample: # firms 3,888,012; # obs. 14,408,051.
  - Country examples: United States: # firms 1,245,865; # obs. 1,400,543; CIT 35. France: # firms 349,669; # obs. 1,726,710; CIT 34.43.
  - Sector examples: Manufacturing 10-33: # firms 752,824; # obs. 3,676,360. Trade, accommodation 49-56: # firms 1,056,270; # obs. 3,776,257.

### Policy implications and recommendations
- Key policy finding:
  - Debt bias induced by corporate taxation increases nonfinancial firm leverage across SMEs and large firms; magnitude economically significant.
- Macro-financial risks:
  - Excessive corporate leverage may lead to rapid deleveraging in adverse environments, curbing investment and potential growth, increasing nonperforming loans and credit tightening, and heightening vulnerability for firms borrowing in foreign currency.
- Policy options to contain debt bias and associated risks:
  - Limit interest deductibility (thin capitalization rules have often only partially addressed debt bias).
  - Introduce an allowance for corporate equity (shown effective in several countries).
  - More radical option: replace CIT with a destination-based cash flow tax with border adjustment to restore neutrality between debt and equity.
  - Tax on borrowing to internalize social cost of excessive leverage (theoretical support cited).
- Expected outcomes from neutral finance taxation:
  - Re-equilibrating tax treatment of debt and equity would increase capital buffers (equity), reduce debt finance, enhance firms’ resilience and macroeconomic stability, and better align tax policy with deleveraging and financial stability objectives.

### Appendix A: model of taxation and corporate leverage (summary)
- Model setup:
  - One-period stylized nonfinancial firm with liabilities equity F_t and debt D_t; assets A_t = F_t + D_t.
  - Return on assets: Θ (random); expected rate of return on assets denoted θ; earnings before tax and interest: θ A_t.
  - Profit tax rate τ, interest rate r_t (debt rolled over), plowback ratio β.
  - Dividends: δ_t = (1−β)(1−τ)[θ A_{t−1} − r_t D_{t−1}] (equation (2) in source).
- Investors maximize expected utility of dividends; utility u non-decreasing and concave; lenders risk-neutral. First-order condition defines optimal debt D* (equation (4)).
- Proposition 1:
  - "The firm hold more debt the higher the tax rate, the shallower equity capital, and the lower the volatility of the return on assets."
- Intuition and formal derivatives:
  - ∂D*/∂τ > 0: higher τ increases tax advantage of debt → higher D*.
  - ∂D*/∂F < 0: lower equity F → higher D*.
  - Increase in variance of Θ reduces optimal debt for risk-averse shareholders (equation (16): marginal increase in variance leads to δD* < 0).
  - Interest rate effect ambiguous (equation (10): ∂D*/∂r ambiguous).

*Source: wp18257 - REFERENCES*

### REFERENCES ___________________________________________________________________________________ 22

### wp18257 - REFERENCES ___________________________________________________________________________________ 22

### Introduction: debt bias and macrofinancial risks
- Tax systems that allow deduction of interest payments but rarely deduct dividend distributions create a debt bias by making debt artificially cheaper than equity.
- Debt bias encourages higher leverage, which:
  - Can reach "50 percent of GDP" in some countries (post-crisis evolution of NFC debt).
  - May dampen long-term growth by reducing corporate investment incentives.
  - Raises financial stability risks via nonperforming loans, fiscal costs from bailouts, and exchange-rate/contagion risks where corporate debt is in foreign currency.
- Prior literature documents significant debt-bias effects, mostly for large or multinational firms; this paper extends the inquiry to small and medium-sized enterprises (SMEs).

### Scope, data coverage, and sample construction
- Data source and scale:
  - Firm-level data from Bureau van Dijk’s Orbis database (combined vintages per Gal and Hijzen (2016)).
  - Orbis contains around "22.5 million" individual firms overall.
  - The study dataset contains "14 million firm-year observations", "almost 99 percent" of which are SMEs.
- SME definition and sample restrictions:
  - SME defined as firms with fewer than "250 employees".
  - Restricted to advanced economies and nonfinancial sectors.
  - Firms must report number of employees and operating revenues.
  - Truncate the top "1 percent" of the distribution for each financial variable, including leverage.
  - Drop firms with less than "US$1,000" in assets and those that have fewer than "10 employees".
- Reporting basis:
  - Focus on unconsolidated reports; for firms reporting both consolidated and unconsolidated accounts, unconsolidated accounts are selected.

### Hypotheses development (five hypotheses)
- Hypothesis 1: The higher the corporate income tax rate the firm faces, the more leveraged its balance sheet.
- Hypothesis 2: The smaller the firm, the less responsive its leverage is to the corporate income tax rate.
- Hypothesis 3: The higher a firm’s tangibility, the more responsive its leverage is to the corporate income tax rate.
- Hypothesis 4: The higher a firm’s revenue and the higher its revenue volatility, the less responsive its leverage is to the corporate income tax rate.
- Hypothesis 5: The higher a firm’s capital intensity of production, the less responsive its leverage is to the corporate income tax rate.
- Theoretical framing:
  - Firms choose fixed-rate debt against uncertain asset returns; higher investment increases leverage and bankruptcy risk.
  - Interest deductibility lowers the cost of debt and therefore shifts the optimal leverage upward (theoretical model in Appendix A).

### Empirical approach and identification
- Strategy:
  - Use a difference-in-difference approach exploiting the interaction between firm characteristics and the corporate income tax (CIT) rate to identify debt bias.
  - Controls include sector-time, country-time, and country-sector fixed effects to mitigate omitted variable bias.
- Rationale for SME focus:
  - SMEs typically have less access to capital markets, more bank-based finance, and are likely less prone to tax planning; results for SMEs are interpreted as a lower bound of economy-wide debt-bias effects.

### Key empirical findings and magnitudes
- Debt bias is a significant driver of leverage even for small firms.
- Magnitude:
  - Debt bias may explain "over 5 percentage points" of leverage, relative to the sample-median leverage of "20 percent of assets".
- Firm characteristics confirmed as important determinants of leverage:
  - Tangibility, liquidity, profitability, and size influence leverage and the sensitivity of leverage to CIT in ways consistent with prior literature.
  - The impact of debt bias on leverage is non-linear with firm size.

### Policy implications (high-level)
- Policies aimed at reducing corporate vulnerability and containing leverage should consider addressing tax-induced debt bias.
- Given that debt bias raises leverage across firms including SMEs, reforming interest deductibility rules or moving toward more neutral taxation of financing decisions warrants consideration.

*Source: wp18257 - REFERENCES ___________________________________________________________________________________ 22*

### introduction of measurement error in the tax variable that may occur when foreign subsidiaries

### Introduction of measurement error in the tax variable that may occur when foreign subsidiaries are consolidated into parent accounts, even though they are subject to a different corporate income tax rate

### Data, variables, and sample
- Preferred leverage indicator: debt-to-asset ratio (chosen for data quality). Alternative measures discussed: debt-to-equity, debt-to-earnings.
- Firm-level controls and definitions:
  - Size proxied by total assets (logarithm used in regressions; square of log included to capture nonlinearity).
  - Collateral proxied by tangible assets; tangible assets also used as a share of total assets (asset tangibility).
  - Turnover and earnings proxied by revenue and cash flow.
  - Turnover volatility proxied by standard deviation of revenue.
  - Capital intensity proxied by the complement of labor costs.
  - Except for total assets, all series normalized as a share of total assets.
  - Debt defined as total liabilities; assets and liabilities used at nominal/book value (Orbis).
- Tax variable:
  - Statutory (marginal) corporate income tax (CIT) rate on the highest tax bracket used (from an internal IMF Fiscal Affairs Department database).
- Sample:
  - More than 14 million observations.
  - Covers 24 advanced economies over 1995–2013.
  - Sample is unbalanced along sector, time, and country dimensions.

### Stylized facts (bivariate patterns)
- Size and leverage:
  - Leverage increases with firm size overall, but relationship is not linear.
  - For largest companies (top quartile of sample; defined as total assets > US$9.4mn) there is a negative correlation between leverage and asset size.
- Tangibility and leverage:
  - Firms with more collateralizable/tangible assets are more leveraged (both absolute tangible assets and tangibility ratio).
- Business model and financing:
  - More capital-intensive production associated with higher leverage.
  - Higher cash flows, higher revenue, and higher revenue volatility tend to be associated with lower leverage.
- Sectoral patterns:
  - Leverage fairly homogenous across 11 sectors; services slightly less leveraged; real estate above-average debt-to-asset ratios.
- Cross-country CIT correlation (bivariate):
  - Median leverage by country-year tends to decrease slightly with higher CIT rates (driven by outliers with very high CIT rates).
  - Empirical regressions with fixed effects will find a positive relationship between CIT rate and leverage once controls are included.

### Empirical strategy
- Approach: difference-in-difference style interaction of firm-level characteristics with country-level CIT rate.
- Main regression specification (dependent variable: debt-to-asset ratio):
  - D_{i,t,c} = α D_{i,t−1,c} + β1 X_{i,t−1,c} + β2 τ_{c,t} × x_{i,t−1,c} + FEs_{c,s,t} + ε_{i,t,c}
  - Coefficient of interest: β2 (interaction between CIT rate and a specific firm-level variable x).
- Firm characteristics included in X:
  - Total assets (log) and its square, share of tangible assets, cash flow, revenue (and its volatility), capital intensity — all scaled by assets and lagged by one period.
- Estimation:
  - Pooled OLS with cluster-robust standard errors clustered at country, industry, and year levels (C*S + C*T + S*T).
  - Lagged dependent variable included to capture persistence.
  - Explored difference and system GMM but infeasible given large unbalanced dataset.

### Evidence of debt bias (main findings)
- General result:
  - Strong evidence that higher CIT rates increase firms’ leverage via an implicit subsidy to debt (interest deductibility).
  - Firms respond to tax incentives by increasing leverage; interaction coefficients largely sizeable and significant.
- Heterogeneous effects (confirming hypotheses):
  - Debt bias stronger for larger firms (Hypothesis 2).
  - Debt bias stronger for firms with higher share of tangible assets (Hypothesis 3).
  - Debt bias weaker for firms with volatile or high revenue (Hypothesis 4).
  - Debt bias stronger for more capital-intensive firms (Hypothesis 5).
- Persistence and controls:
  - Strong persistence of leverage (lagged dependent variable significant).
  - Significant direct effects of size, cash flow, revenue and volatility, capital intensity, and asset composition on leverage.
  - Nonlinear relationship between size and leverage confirmed; dominant effect in sample is positive.
- Magnitude and illustrative calculations:
  - Interaction coefficient on total assets: 0.907.
  - Long-run persistence parameter used in examples: 0.842 (1 − α = 1 − 0.842).
  - Using these values, debt bias explains about 20 percentage points of leverage variations among firms in the long run (calculation: 0.907/(1−0.842) × 0.28 × 12.74 = 20.47).
  - Debt bias explains about 5 percentage points of variation among firms with different tangible assets.
  - Revenue, revenue volatility, and capital intensity explain about 11, 4, and 11 percentage points of debt bias variation within firms, respectively.
  - At the sample-mean CIT rate of 28 percent, the interaction coefficient on total assets implies debt bias contributes approximately 24 percentage points of leverage in the median firm (alternative long-term calculation result reported as 23.87).
  - Debt bias magnitude noted to explain up to 27 percent of leverage.
- Quantile regressions:
  - Results particularly pronounced for the most leveraged firms (decile quantile regressions).
  - Positive CIT × (size, tangible assets, capital intensity) interactions driven by most leveraged firms.
  - Negative CIT × (revenue, revenue volatility) interactions also driven by most leveraged firms.
  - CIT × cash flow positive (but insignificant in baseline) mainly driven by least leveraged firms.
  - Quantile regressions run on an almost-balanced subsample requiring 10 consecutive years, yielding just under 1 million firm-year observations.

### Robustness checks and alternative specifications
- Different debt maturity composition:
  - Replacing total debt with long-term debt as dependent variable yields qualitatively similar results; debt bias appears more important for long-term financing decisions.
  - Exception: interaction coefficient on cash flow becomes negative and significant for long-term debt.
- Alternative dependent variable:
  - Using debt-to-equity does not alter findings qualitatively; sample size increases; overall fit slightly worse.
  - Differences: role of size diminished; tangibility interaction becomes negative and significant in debt-to-equity specification.
- Subsamples:
  - European-only sample: results broadly robust; cash flow interaction becomes larger and significant at 5 percent; tangibility interaction smaller and not significantly different from zero.
  - Including more small firms (exclude only single-employee firms): broadly similar; tangibility interaction not significant; cash flow interaction significant at 1 percent in this sample.
  - Largest firms (top quartile by assets): determinants differ—asset size control negative; collateral and revenues explain less; capital intensity explains more; debt bias signs maintained but some interactions insignificant.
- Time coverage and Orbis versions:
  - Including more recent observations (Orbis up to 2015/2016) broadly robust though pre-2005 coverage limited.
  - Orbis up-to-date version noted to cover 1990 to 2016 but sample strongly unbalanced.
- Balanced and binary-variable specifications:
  - Balanced (firms present for minimum 10 consecutive years) sample of 1.3 million observations yields coefficient estimates close to baseline; tangibility and revenue less influential; cash flow influence larger and significant at 1 percent.
  - Mapping firm characteristics to binary dummies (above/below median) does not qualitatively change conclusions.
- Sectoral checks:
  - Regressions by each of 11 nonfinancial sectors show results broadly consistent across sectors (sectoral tables available on request).
- Trimming and sample thresholds:
  - Trimming top two percent instead of one does not qualitatively change results.

### Policy implications and recommendations
- Key policy finding:
  - Debt bias induced by corporate taxation increases nonfinancial firm leverage across SMEs and large firms; magnitude economically significant.
- Macro-financial risks:
  - Excessive corporate leverage may lead to rapid deleveraging in adverse environments, curbing investment and potential growth, increasing nonperforming loans and credit tightening, and heightening vulnerability for firms borrowing in foreign currency.
- Policy options to contain debt bias and associated risks:
  - Limit interest deductibility (thin capitalization rules have often only partially addressed debt bias).
  - Introduce an allowance for corporate equity (shown effective in several countries).
  - More radical option: replace CIT with a destination-based cash flow tax with border adjustment to restore neutrality between debt and equity (discussion referenced).
  - Tax on borrowing to internalize social cost of excessive leverage (cited theoretical support).
- Expected outcomes from neutral finance taxation:
  - Re-equilibrating tax treatment of debt and equity would increase capital buffers (equity), reduce debt finance, enhance firms’ resilience and macroeconomic stability, and better align tax policy with deleveraging and financial stability objectives.

*Source: IMF working paper (excerpt provided).*

### REFERENCES

### wp18257 - REFERENCES

### Key bibliographic sources cited
- Angrist, J. D., and J. -S. Pischke, 2009, “Mostly Harmless Econometrics: An Empiricist's Companion,” (Princeton, New Jersey: Princeton University Press).
- Auerbach, Alan, Michael P. Devereux, Michael Keen, and John Vella, 2017, “Destination-Based Cash Flow Taxation,” Working Paper 17/01 (Oxford, United Kingdom: Oxford University Centre for Business Taxation).
- De Mooij, R.A., 2011, “The Tax Elasticity of Corporate Debt: A Synthesis of Size and Variations,” IMF Working Paper, 11/95 (Washington, DC: International Monetary Fund).
- International Monetary Fund, 2016b, “Tax Policy, Leverage and Macroeconomic Stability,” IMF Policy Paper, October 2016, (Washington, DC: International Monetary Fund).
- Modigliani, Franco and M. Miller, 1963, "Corporate Income Taxes and The Cost of Capital: A Correction", American Economic Review, 53 (3): pp. 433–43.
- (Additional academic and working-paper sources on capital structure, taxation, firm-level leverage, FDI and Orbis-based firm data are listed in full in the source.)

### Variable definitions (Table 1)
- LEVERAGE: (Loans & Long-term debt)-to-total assets; Unit: Percent; Source: Orbis.
- LEVERAGE2: Long-term debt-to-total assets; Unit: Percent; Source: Orbis.
- LEVERAGE3: (Loans & Long-term debt)-to-equity; Unit: Percent; Source: Orbis.
- LEVERAGE4: (Loans & Long-term debt)-to-earnings; Unit: Percent; Source: Orbis.
- TA: Total assets [log]; Unit: USD; Source: Orbis.
- TA-SQ: Total assets [log] squared; Source: Orbis.
- TFA-TA: Tangible fixed assets-to-total assets ratio; Unit: Ratio; Source: Orbis.
- REV-TA: Revenue-to-total assets ratio; Unit: Ratio; Source: Orbis.
- CASHFLOW-TA: Cash flow-to-total assets; Unit: Ratio; Source: Orbis.
- REV-VOL: Revenue-to-assets standard deviation; Unit: Ratio; Source: Orbis.
- CAPITAL-TA: Capital-to-total assets (inverse of labor-to-total assets); Unit: Ratio; Source: Orbis.
- CIT: Top central statutory CIT rate; Unit: Ratio; Source: IMF.
- Note: ratio as unit means for instance that CIT = 0.3 for a country where 30 percent of income is levied.

### Micro-level summary statistics (Table 2)
- Sample sizes and descriptive statistics (min, mean, p25, p50, p75, max, sd, # of obs.):
  - LEVERAGE: min ≈0; mean 25.74; p25 6.839; p50 20.25; p75 39.78; max 104.1; sd 22.31; # of obs. 7,367,751.
  - LEVERAGE2: min ≈0; mean 21.92; p25 5.333; p50 15.48; p75 32.94; max 93.35; sd 20.70; # of obs. 5,956,281.
  - LEVERAGE3: min -144.8; mean 39.95; p25 0.361; p50 13.60; p75 48.81; max 564.0; sd 68.63; # of obs. 9,246,447.
  - LEVERAGE4: min -3,695; mean 225.4; p25 0; p50 61.14; p75 306.7; max 5,453; sd 659.2; # of obs. 7,962,715.
  - TA: min 0; mean 14.73; p25 13.78; p50 14.85; p75 16.00; max 20.57; sd 2.253; # of obs. 10,973,463.
  - TA-SQ: min 0; mean 222.1; p25 189.8; p50 220.4; p75 256.1; max 423.2; sd 61.47; # of obs. 10,973,463.
  - TFA-TA: min 0; mean 0.251; p25 0.06; p50 0.183; p75 0.390; max 0.926; sd 0.229; # of obs. 10,951,096.
  - REV-TA: min 0; mean 1.886; p25 1.003; p50 1.587; p75 2.411; max 8.656; sd 1.313; # of obs. 10,973,463.
  - CASHFLOW-TA: min 0; mean 0.092; p25 0.034; p50 0.07; p75 0.127; max 0.465; sd 0.079; # of obs. 8,038,252.
  - REV-VOL: min 0; mean 0.642; p25 0.232; p50 0.454; p75 0.839; max 4.257; sd 0.612; # of obs. 11,097,374.
  - CAPITAL-TA: min -3.571; mean -0.488; p25 -0.619; p50 -0.316; p75 -0.153; max 0; sd 0.524; # of obs. 9,002,220.

### Main regression results (Table 3) — Dependent variable: LEVERAGE
- Baseline persistent dynamics:
  - LEVERAGE (L): coefficient 0.842*** with standard error [0.003] in all specifications (Observations 3,494,804; Adj. R-squared 0.762).
- Key micro-level coefficients (selected columns, lagged where noted):
  - TA (L): coefficients vary, e.g., -0.283*** [0.078] in column (1); 0.117** [0.050] in column (2); -0.269*** [0.077] in column (7).
  - TFA-TA (L): 1.880*** [0.153] in column (1); 0.945** [0.474] in column (2); 1.334*** [0.462] in column (7).
  - REV-TA (L): 0.146*** [0.017] in column (1); 0.433*** [0.072] in column (3); 0.364*** [0.086] in column (7).
  - CASHFLOW-TA (L): -6.750*** [0.232] in column (1); -8.311*** [1.068] in column (4); -9.554*** [1.014] in column (7).
  - REV-VOL: -0.379*** [0.028] in column (1); -0.391*** [0.029] in many columns; -0.575*** [0.132] in column (7).
  - CAPITAL-TA (L): 0.353*** [0.043] in column (1); note sign flip in column (6): -0.385** [0.176].
- Micro × CIT interaction effects (examples):
  - TA x CIT: 0.907*** [0.134] in column (2); 0.882*** [0.136] in column (7).
  - REV-TA x CIT: -0.908*** [0.214] in column (3); -0.702*** [0.258] in column (7).
  - CASHFLOW-TA x CIT: 8.994*** [3.329] in column (7) (column (5) shows 5.203 [3.532] not significant).
- Notes: Standard errors clustered by sector, country and year. Significance: ***, **, * denote 1%, 5%, 10%. Fixed effects S*T + S*C + C*T. (L) stands for lag. Lagged micro-level variable interacted with contemporaneous CIT rate.

### Quantile regressions (Table 4) — Dependent variable: LEVERAGE
- Panel (a) Total Assets (quantiles 0.1, 0.25, 0.5, 0.75, 0.9; Observations 1,362,563):
  - LEVERAGE (L) increases across quantiles: 0.673*** [0.001] at 0.1 to 0.976*** [0.001] at 0.75, then 0.934*** [0.001] at 0.9.
  - TA (L) coefficients decline from 0.065*** [0.015] at 0.1 to -0.227*** [0.060] at 0.9.
  - TA x CIT: -0.204*** [0.018] at 0.1 up to 0.622*** [0.048] at 0.9.
- Panel (b) Tangible Fixed Assets: TFA-TA (L) shows large variation across quantiles (e.g., 4.665*** [0.118] at 0.1; -7.940*** [0.424] at 0.9) with TFA-TA x CIT: -11.170*** [0.357] at 0.1 and 28.339*** [1.356] at 0.9.
- Other panels (c) Revenue, (d) Cash Flow, (e) Revenue Volatility, (f) Capital Intensity display differing coefficient profiles across leverage quantiles; see table for exact coefficients and standard errors. Key example:
  - CASH-TA x CIT in panel (d): 10.123*** [1.134] at 0.1; 11.885*** [2.054] at 0.75; -2.035 [4.614] at 0.9.

### Robustness and alternative specifications (Tables 5–13)
- Table 5 (Dependent: LEVERAGE2; Observations 2,420,610; Adj. R-squared 0.738):
  - LEVERAGE2 (L): 0.818*** [0.004].
  - TA x CIT: 1.528*** [0.225] in column (2); 1.586*** [0.254] in column (7).
  - TFA-TA x CIT: 3.862*** [1.281] in column (2).
- Table 6 (Dependent: LEVERAGE3; Observations 4,513,192; Adj. R-squared 0.679):
  - LEVERAGE3 (L): 0.791*** [0.004].
  - Large positive TFA-TA (L): 7.758*** [0.325] in column (1).
- Table 7 (European countries subset; Observations 3,162,831; Adj. R-squared 0.741):
  - LEVERAGE (L): 0.833*** [0.003].
  - TA x CIT: 0.762*** [0.123] in column (2); 0.717*** [0.121] in column (7).
- Table 8 (Trimming top 2%): LEVERAGE (L) 0.836*** [0.003]; REV-VOL shows positive and large coefficients in this trimmed sample (e.g., 0.868*** [0.051]).
- Table 9 (Two-employee threshold): Observations 8,143,290; LEVERAGE (L) 0.828*** [0.004]; notable CASHFLOW-TA x CIT: 16.534*** [3.001] in column (5).
- Table 10 (Orbis 2005-2015 vintage): Observations 5,954,763; LEVERAGE (L) 0.813*** [0.003]; TFA-TA x CIT: 6.582*** [1.719] in column (2).
- Table 11 (Large firms only): Observations 994,978; LEVERAGE (L) 0.879*** [0.003]; TA (L) -1.456*** [0.406] in column (1).
- Table 12 (Balanced sample): Observations 1,362,563; LEVERAGE (L) 0.856*** [0.003]; CASHFLOW-TA x CIT: 17.949*** [4.215] in column (6).
- Table 13 (Binary micro variables): Various binary dummy × CIT interactions significant (e.g., TA dummy x CIT 0.698*** [0.072]); sample sizes vary by column (observations listed per column).

### Coverage by country and sector (Table 14)
- (a) By Country — sample counts and CIT (top statutory CIT rate in 2013 expressed in percent):
  - Austria: # firms 32,759; # obs. 122,882; CIT 25.
  - Belgium: # firms 34,056; # obs. 243,481; CIT 33.99.
  - Canada: # firms 194,198; # obs. 584,242; CIT 15.
  - France: # firms 349,669; # obs. 1,726,710; CIT 34.43.
  - Germany: # firms 330,649; # obs. 1,698,813; CIT 15.825.
  - Italy: # firms 293,585; # obs. 1,511,880; CIT 27.5.
  - Japan: # firms 358,416; # obs. 1,628,051; CIT 28.05.
  - United States: # firms 1,245,865; # obs. 1,400,543; CIT 35.
  - Total: # firms 3,888,012; # obs. 14,408,051.
- (b) By Sector — NACE codes, # firms and # obs.:
  - Agriculture 01-03: # firms 51,363; # obs. 188,758.
  - Manufacturing 10-33: # firms 752,824; # obs. 3,676,360.
  - Construction 41-43: # firms 583,255; # obs. 2,197,592.
  - Trade, accommodation 49-56: # firms 1,056,270; # obs. 3,776,257.
  - Professional & administrative activities 69-82: # firms 498,741; # obs. 1,586,696.
  - Total: # firms 3,888,012; # obs. 14,408,051.
- Notes: Financial sector is dropped. NACE 2nd revision used. CIT reported is top statutory CIT rate in 2013.

### Appendix: A model of taxation and corporate leverage (summary of structure and formal results)
- Model setup:
  - One-period stylized nonfinancial firm with liabilities equity F_t and debt D_t; assets A_t = F_t + D_t.
  - Return on assets: Θ (random); expected rate of return on assets denoted θ; earnings before tax and interest: θ A_t.
  - Profit tax rate τ, interest rate r_t (debt rolled over), plowback ratio β. Dividends:
    - δ_t = (1−β)(1−τ)[θ A_{t−1} − r_t D_{t−1}] (equation (2) in source).
- Investors maximize expected utility of dividends; utility u is non-decreasing and concave; lenders are risk-neutral. First-order condition defines optimal debt D* (equation (4)).
- Proposition 1 (formal statement):
  - "The firm hold more debt the higher the tax rate, the shallower equity capital, and the lower the volatility of the return on assets."
- Intuition and formal derivatives:
  - Higher τ increases tax advantage of debt → higher D* (equation (8): ∂D*/∂τ > 0).
  - Lower equity F (shallower equity capital) → higher D* (equation (11): ∂D*/∂F < 0).
  - Higher return volatility → shareholders deleverage; thus lower volatility → more debt (discussion and equation (16): marginal increase in variance leads to δD* < 0).
  - Interest rate effect ambiguous (equation (10): ∂D*/∂r ambiguous because higher r increases financing cost but also increases interest deductibility).
- Lemmas and technical points:
  - Lemma 1: Optimal leverage only if expected return on assets exceeds cost of capital: r ≤ E(Θ).
  - Lemma 2: Certain integral terms g_1, g_2 (proportional to moments of u'') are negative.
  - Functional derivative analysis (Euler-Lagrange) shows how changes in the distribution f of Θ affect D* (equations (12)–(16)): an increase in variance holding mean constant reduces optimal debt for risk-averse shareholders.

*Italicized: Source: wp18257 - REFERENCES (IMF working paper material provided in the content unit).*

---


_Source: https://www.imf.org/-/media/files/publications/wp/2018/wp18257.pdf_