## _wp13221 — 1. Descriptive Statistics

## Source details

**Canonical URL:** [_wp13221 — 1. Descriptive Statistics](https://www.imf.org/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2013/_wp13221.pdf)

## Other formats

- [Markdown version](/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2013/_wp13221.pdf.md)
- [Structured JSON version](/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2013/_wp13221.pdf.json)

---

### I. Introduction — empirical context and motivation
- Debt bias: tax distortion from corporate income tax (CIT) allowing interest deductibility but not equity payments.
- Existing empirical evidence for non-financial companies:
  - Marginal tax effect on the debt-asset ratio ranges between 0.2 and 0.3.
  - Interpretation example: a 10 percentage points higher CIT rate raises the debt-to-asset ratio by somewhere between 2 and 3 percentage points.
- Financial-sector relevance:
  - Debt bias may have larger welfare costs in the financial sector due to externalities associated with systemic risk.
  - Aggregate financial-sector leverage positively affects the probability of financial crisis (cited evidence: Barrell et al. (2010); Kato, Kobayashi, and Saita (2010); De Mooij, Keen and Orihara (2013)).
- Recent empirical work on banks (selected findings):
  - Keen and de Mooij (2012): CIT significantly raises bank leverage ratios; on average effect about as large as for non-banks, with important heterogeneity.
  - Hemmelgarn and Teichmann (2013): similar results.
  - De Mooij, Keen and Orihara (2013): elasticity gets smaller with bank size.
  - Gu, De Mooij and Poghosyan (2012): subsidiaries of 100 largest multinational banks show significant tax effects, including multinational debt shifting.
- Contributions of this paper:
  - Uses Worldscope data containing both banks and non-banks to formally test differences in tax responsiveness.
  - Explores how firm size affects tax responsiveness using micro data on asset size and interactions of tax with polynomials of log assets.
  - Uses quantile regressions to examine variation of tax effects along the conditional leverage distribution.

### II. Methodology — specification and estimation
- Baseline reduced-form specification:
  - Dependent variable: lev (leverage-to-assets ratio).
  - Key explanatory variable: country-specific statutory CIT rate (τ_c).
  - Controls vector (X_isc,t) includes standard capital-structure controls.
  - Fixed effects: time fixed effects, industry fixed effects (for non-banks), and country fixed effects.
- Controls explicitly included:
  - Book value of total assets (in logs) and, in some specifications, its square.
  - Operating income over total assets (profitability).
  - Share of tangible assets in total assets (collateral).
  - Country-level controls: growth rate of GDP and rate of inflation.
- Size interactions:
  - Interaction of tax with first- or second-order polynomial of log assets to allow non-linear size effects (e.g., U-shaped patterns).
- Estimation:
  - All regressions estimated by OLS.
  - Samples: non-banks, banks, combined sample (with bank dummies).
  - Standard errors robust and clustered at the firm level.
- Quantile regressions:
  - Used to approximate conditional median and other quantiles of leverage (Koenker and Bassett, 1978; Koenker and Hallock, 2001).
  - Purpose: examine whether marginal tax effects differ along the conditional leverage distribution.

### III. Data — sample composition and key descriptive statistics
- Data source: Worldscope (Thomson Financial).
- Sample years: 1996 to 2010.
- Final sample after excluding missing variables:
  - 38,867 firms from 69 countries.
  - 336,051 firm-year observations in total.
  - Banks in the sample: 1,603 banks contributing 17,227 bank-year observations.
  - Banks’ share in total number of firms: 4.1 percent.
  - Banks’ share in the overall sample of observations: 5.1 percent.
- Consolidation note:
  - Data reflect consolidated accounts of firms; consolidated statements used rather than unconsolidated.

### IV. Descriptive statistics — selected sample moments (exact figures)
- Sample sizes:
  - Full sample (all firms): 336,051
  - Banks subsample: 17,227
  - Non-banks subsample: 318,824
- Leverage Ratio (total liabilities / total assets)
  - All firms: Mean 0.4863; SD 0.2541; Median 0.4924; Min 0; Max 1
  - Banks: Mean 0.9006; SD 0.0751; Median 0.9130; Min 0; Max 0.9993
  - Non-banks: Mean 0.4639; SD 0.2408; Median 0.4739; Min 0; Max 1
- Corporate income tax rate
  - All firms: Mean 0.3325; SD 0.0830; Median 0.3399; Min 0; Max 0.5725
  - Banks: Mean 0.3665; SD 0.0723; Median 0.4076; Min 0; Max 0.5725
  - Non-banks: Mean 0.3307; SD 0.0831; Median 0.335; Min 0; Max 0.5725
- Log total assets
  - All firms: Mean 18.7271; SD 2.3667; Median 18.6637; Min 6.9078; Max 29.1844
  - Banks: Mean 21.4586; SD 2.2345; Median 21.0481; Min 11.8914; Max 29.1844
  - Non-banks: Mean 18.5795; SD 2.2823; Median 18.5457; Min 6.9078; Max 28.363
- Operating income over assets
  - All firms: Mean 0.00643; SD 0.2283; Median 0.0381; Min -2.5556; Max 0.7459
  - Banks: Mean 0.0095; SD 0.0331; Median 0.0110; Min -1.8278; Max 0.5457
  - Non-banks: Mean 0.0062; SD 0.2342; Median 0.0425; Min -2.5556; Max 0.7459
- Collateral (share of tangible assets)
  - All firms: Mean 0.2920; SD 0.2528; Median 0.2399; Min 0; Max 1
  - Banks: Mean 0.0189; SD 0.0193; Median 0.0158; Min 0; Max 0.9866
  - Non-banks: Mean 0.3067; SD 0.2512; Median 0.2595; Min 0; Max 1
- GDP growth rate
  - All firms: Mean 0.0599; SD 0.0613; Median 0.0546; Min -0.1143; Max 0.4298
  - Banks: Mean 0.0564; SD 0.0544; Median 0.0540; Min -0.1143; Max 0.4298
  - Non-banks: Mean 0.0601; SD 0.0616; Median 0.0546; Min -0.1143; Max 0.4298
- Inflation rate
  - All firms: Mean 0.0250; SD 0.0307; Median 0.0219; Min -0.0623; Max 0.2586
  - Banks: Mean 0.0278; SD 0.0278; Median 0.0244; Min -0.0623; Max 0.2586
  - Non-banks: Mean 0.0249; SD 0.0309; Median 0.0212; Min -0.0623; Max 0.2586

### V. Size distributions and salient quantiles (exact figures)
- Non-banks (log total assets)
  - Mean log assets 18.58 (US$ 117 million)
  - 25th percentile: log 17.16 (US$ 28 million)
  - 75th percentile: log 20 (US$ 482 million)
  - 1 percent quantile: log 12.66 (US$ 316 thousand)
  - 5 percent quantile: log 14.84 (US$ 2.8 million)
  - 95 percent quantile: log 22.47 (US$ 5.8 billion)
  - 99 percent quantile: log 24.29 (US$ 35.4 billion)
  - Distribution: approximately standard normal in log size, symmetric around mean
- Banks (log total assets)
  - Mean log assets 21.45 (US$ 2.1 billion)
  - 1 percent quantile: log 17.5 (US$ 40 million)
  - 5 percent quantile: log 18.47 (US$ 105 million)
  - 90 percent quantile: log 24.5 (US$ 44 billion)
  - 95 percent quantile: log 25.48 (US$ 116 billion)
  - 99 percent quantile: log 27.37 (US$ 771 billion)
  - Distribution: skewed with a long right tail

### VI. Results — Non-banks (selected regression estimates and interpretations)
- Regression sequence: six specifications (columns (1)–(6)) adding fixed effects, size, profitability, collateral, macro variables, and quadratic size interactions.
- Estimated CIT coefficients (Table 2, columns (1)–(6)):
  - Column (1) CIT rate: 0.3411*** (26.6995)
  - Column (2) CIT rate: 0.2172*** (9.3032)
  - Column (3) CIT rate: 0.1974*** (8.7968)
  - Column (4) CIT rate: -0.0428 (-0.4142)
  - Column (5) CIT rate: -0.0354 (-0.3429)
  - Column (6) CIT rate: 2.9103*** (5.7188)
- Selected size and interaction coefficients:
  - Log assets: reported significant coefficients including 0.0269*** (58.2108), 0.0258*** (13.7773), 0.0259*** (13.7931), 0.1036*** (5.1685)
  - Log assets * CIT: 0.0122** (2.3015); 0.0121** (2.2837); -0.3050*** (-5.5942)
  - Log assets sq.: -0.0021*** (-3.8456)
  - Log assets sq * CIT: 0.0084*** (5.7757)
- Other covariates (columns (4)–(6)):
  - Profitability: -0.1110*** (-34.0219); -0.1111*** (-34.0476); -0.1010*** (-30.5079)
  - Collateral: 0.0629*** (14.3610); 0.0628*** (14.3548); 0.0643*** (14.6414)
  - GDP growth: 0.0324** (2.5546); 0.0307** (2.4233)
  - Inflation: -0.0813*** (-3.7410); -0.0828*** (-3.8085)
- Tax effects at sample mean size (Table 2 reported):
  - Column (1): 0.3411*** (26.6995)
  - Column (2): 0.2172*** (9.3032)
  - Column (3): 0.1974*** (8.7968)
  - Column (4): 0.1845*** (8.25)
  - Column (5): 0.1901*** (8.43)
  - Column (6): 0.1469*** (6.21)
- Goodness-of-fit and sample size:
  - No. of obs: 318,824
  - Adj. R-squared: 0.0139 (col (1)), 0.171 (col (2)), 0.222 (col (3)), 0.234 (col (4)), 0.234 (col (5)), 0.236 (col (6))
- Main interpretations:
  - Columns (1)–(3): tax coefficients statistically significant at the 1 percent level.
  - Column (3) interpretation: leverage rises by 2 percentage points if the CIT rate increases by 10 percentage points (coefficient 0.1974).
  - Size interaction (cols (4)–(6)): larger firms' debt ratios tend to be more responsive to the CIT rate; column (6) marginal tax effect at sample mean log assets is 0.15 (t-value: 6.21).
  - Heterogeneity: convex (U-shaped) marginal tax effect over size distribution for non-banks; 99 percent of firms have tax coefficient below 0.5. Thresholds where marginal tax response exceeds 0.5:
    - Firms with log assets below 11.62 (US$ 111 thousand)
    - Firms with log assets exceeding 24.68 (US$ 52.3 billion)
  - Asset concentration: firms with log assets exceeding 24.68 own approximately 45.8 percent of total assets in the non-banks sample.
  - Non-tax determinants: size positive; profitability negative; collateral positive; GDP growth positive; inflation negative.

### VII. Results — Banks (selected regression findings and interpretations)
- Regression sequence mirrors non-banks.
- Reported magnitude of tax response for banks: between 0.06 and 0.14 (slightly lower than non-banks in Table 2) — (text truncated for full coefficient table).
- Tax–Size interaction (Table 3; regressions (4) and (5)):
  - Interaction of tax and bank size: significantly negative (5 percent confidence) — marginal tax effect on banks’ leverage declines with bank size.
  - Evaluated at sample mean of bank size:
    - Marginal tax effect on leverage ratio is 0.20 in column (4) (t-value: 4.53).
    - Marginal tax effect on leverage ratio is 0.21 in column (5) (t-value: 4.66).
  - Column (6) (second-order polynomial): individual tax coefficients not significant; combined tax effect significant over an important range; quadratic interaction very small suggesting near-linearity.
- Marginal tax effect along bank size distribution:
  - Tax effect largest for small banks, marginal impact close to 0.5.
  - Effect falls and becomes insignificant for banks with log assets of 25 (total assets US$ 72 billion) — applies to the 7.1 percent largest banks.
  - For the median bank (log assets 21.0), marginal coefficient is 0.22 (t-value: 4.53).
- Non-tax determinants for banks vs non-banks (regressions (3)–(6) Table 3):
  - Size: positive coefficient for banks as for non-banks.
  - Profitability: not significant for banks; significant negative for non-banks.
  - Collateral: negative coefficient for banks (contrasts with positive effect for non-banks).
  - GDP growth and inflation: insignificant for banks.

### VIII. Banks vs Non-banks — combined regressions and hypothesis tests
- Table 4 (full-sample regressions with interactions):
  - Column (1): CIT rate 0.1873*** (8.7099)
  - Column (2): Bank * CIT = 0.0695** (2.2107); (1-Bank) * CIT = 0.1933*** (8.6134) — banks’ average response smaller here.
  - Column (3): linear interaction - non-banks: positive interaction (larger non-banks more responsive); banks: negative correlation (larger banks less responsive).
  - Column (4): quadratic interaction non-banks convex; banks coefficient insignificant and very small.
  - Column (5) preferred: non-linear size effect only for non-banks.
- Table 5 — tests of equivalence of tax effects across bank size distribution:
  - Sample means:
    - Mean non-bank log assets: 18.58 (~US$ 117.4 million)
    - Mean bank log assets: 21.45 (~US$ 2.1 billion)
    - Marginal tax effects at sample means:
      - Non-banks: 0.1469*** (t-value: 6.21)
      - Banks: 0.2108*** (t-value: 4.67)
    - F-Statistic: 1.57 (Prob > F: 0.21) — cannot reject equality at sample means.
  - Marginal tax effects at bank-size quantiles (from regression (5) Table 4) — selected entries (all stars and t-values preserved):
    - Bank size Q10 (Log assets: 18.91; ~US$ 163.9 million):
      - Non-banks: 0.1504*** (6.38)
      - Banks: 0.2878*** (3.88)
      - F-Statistic: 3.11 (Prob > F: 0.0778) — borderline.
    - Bank size Q25 (Log assets: 19.76; ~US$ 379.9 million):
      - Non-banks: 0.1674*** (7.13)
      - Banks: 0.2624*** (4.11)
      - F-Statistic: 1.95 (Prob > F: 0.1626)
    - Bank size Q50 (Log assets: 21.05; ~US$ 1.4 billion):
      - Non-banks: 0.2167*** (8.65)
      - Banks: 0.2233*** (4.54)
      - F-Statistic: 0.01 (Prob > F: 0.9055)
    - Bank size Q75 (Log assets: 23.04; ~US$ 10.2 billion):
      - Non-banks: 0.3479*** (9.25)
      - Banks: 0.1630*** (4.69)
      - F-Statistic: 13.03 (Prob > F: 0.0003) — rejects equivalence.
    - Bank size Q90 (Log assets: 24.51; ~US$ 44.1 billion):
      - Non-banks: 0.4872*** (8.58)
      - Banks: 0.1186*** (3.23)
      - F-Statistic: 29.72 (Prob > F: 0.0000) — rejects equivalence.
    - Bank size Q95 (Log assets: 25.48; ~US$ 116.3 billion):
      - Non-banks: 0.5991*** (8.14)
      - Banks: 0.0892** (2.03)
      - F-Statistic: 35.39 (Prob > F: 0.0000) — rejects equivalence.
  - Interpretation:
    - Cannot reject equivalence for Q10, Q25, Q50.
    - For larger firms (Q75 and above), non-banks’ tax responsiveness significantly larger than banks’.

### IX. Quantile regressions — non-banks and banks (selected results)
- Non-banks (Table A-3 — quantiles Q10, Q50, Q75, Q90; No. of obs 318,824)
  - (Pseudo) R-squared: Q10 0.1759; Q50 0.1577; Q75 0.1170; Q90 0.0914
  - CIT rate coefficients:
    - Q10: 0.8622*** (4.42)
    - Q50: 3.766*** (10.70)
    - Q75: 3.045*** (7.18)
    - Q90: 0.6832* (1.77)
  - Interaction pattern:
    - Log assets * CIT negative at Q10, Q50, Q75; Log assets sq * CIT positive at Q10, Q50, Q75 — indicating second-order nonlinear interaction and curvature.
  - Operating income: consistently negative and highly significant across quantiles (e.g., Q90: -0.1863*** (-45.45)).
  - Collateral: positive at lower/mid quantiles, negative at Q90 (e.g., Q10: 0.0909*** (42.92); Q90: -0.0098*** (-3.69)).
  - Joint tax effects across evaluated sizes (selected):
    - Non-bank size Q10 (Log assets: 15.77; US$ 7.1 million):
      - Q10: 0.0340 (1.48)
      - Q50: 0.2103*** (7.43)
      - Q75: 0.2898*** (9.66)
      - Q90: 0.2791*** (11.03)
      - F-test Q10 = Q90: 55.53 (Prob: 0.0000)
    - Non-bank size Q50 (Log assets: 18.55; US$ 113.7 million):
      - Q10: 0.1031*** (4.28)
      - Q50: 0.1410*** (6.05)
      - Q75: 0.1855*** (7.96)
      - Q90: 0.2133*** (9.74)
      - F-test Q10 = Q90: 12.28 (Prob: 0.0005)
    - Non-bank size Q90 (Log assets: 21.50; US$ 2.2 billion):
      - Q10: 0.2452*** (9.44)
      - Q50: 0.2448*** (10.37)
      - Q75: 0.1957*** (8.24)
      - Q90: 0.1449*** (6.61)
      - F-test Q10 = Q90: 9.85 (Prob: 0.0017)
  - Interpretation:
    - For non-banks, tax responsiveness varies by conditional leverage quantile and size; patterns include convexity and monotonic changes depending on the leverage quantile.
- Banks (Table A-4 — quantiles Q10, Q50, Q75, Q90; No. of obs 17,227)
  - (Pseudo) R-squared: Q10 0.2030; Q50 0.2080; Q75 0.2559; Q90 0.2761
  - CIT rate coefficients:
    - Q10: 0.9358*** (5.56)
    - Q50: 0.4185*** (7.74)
    - Q75: 0.3397*** (6.02)
    - Q90: 0.3371*** (6.32)
  - Log assets positive and significant across quantiles.
  - Log assets * CIT negative and significant at Q10, Q50, Q90 — indicating decreasing tax impact with bank size.
  - Operating income: not significant at Q10; negative and significant at higher quantiles (Q50: -0.5557***; Q75: -0.6892***; Q90: -0.7418***).
  - Collateral: negative and significant across quantiles (e.g., Q10: -0.4728***).
  - Joint tax effects (selected):
    - Bank size Q10 (Log assets: 18.91; US$ 163.9 million):
      - Q10: 0.3119*** (5.85)
      - Q50: 0.1434*** (9.65)
      - Q75: 0.1207*** (7.97)
      - Q90: 0.0999*** (7.54)
      - F-test Q10 = Q90: 15.65 (Prob: 0.0001)
    - Bank size Q50 (Log assets: 21.05; US$ 1.4 billion):
      - Q10: 0.2416*** (5.69)
      - Q50: 0.1124*** (9.85)
      - Q75: 0.0960*** (8.42)
      - Q90: 0.0731*** (6.90)
      - F-test Q10 = Q90: 15.41 (Prob: 0.0001)
    - Bank size Q90 (Log assets: 24.51; US$ 44.1 billion):
      - Q10: 0.1274*** (4.26)
      - Q50: 0.062*** (6.64)
      - Q75: 0.0559*** (6.66)
      - Q90: 0.0297*** (2.81)
      - F-test Q10 = Q90: 9.73 (Prob: 0.0018)
  - Interpretation:
    - For banks, tax response declines with asset size for all conditional leverage quantiles.
    - Tax response for banks declines with conditional leverage ratio — capital-tight banks (high conditional leverage) are less responsive.
    - Even largest banks with highest conditional leverage show a positive significant marginal impact of 0.03 (reported).

### X. Conclusions — synthesis of findings
- Main findings:
  - Tax bias toward debt induced by corporate tax systems matters for both banks and non-banks.
  - Preferred specification: an average bank is equally responsive to tax incentives as an average non-bank (cannot reject equality at sample means).
- Differences across size distributions:
  - Non-banks (OLS): U-shaped (convex) pattern between asset size and tax responsiveness — very small and very large non-banks relatively responsive.
  - Banks: linearly declining relationship between asset size and tax responsiveness — large banks relatively unresponsive.
- Quantile-regression insights:
  - Capital-tight banks (high conditional leverage) are significantly less responsive to tax than capital-abundant banks — holds across bank sizes.
  - For non-banks, largest tax responses found for (i) large firms with low conditional leverage and (ii) small firms with high conditional leverage.
  - For large firms and banks, patterns along conditional leverage distribution are similar.

### XI. Appendix I — data by country (selected exact figures)
- Sample coverage:
  - Sample covers 70 countries; for 42 countries data for all years 1996–2010.
  - United States makes up more than 20 percent of all observations.
  - US, Japan, China, Canada, Taiwan and India account for more than 52 percent of observations.
  - Adding Korea, Australia, UK, Hong Kong, Malaysia, Germany and France covers more than 76 percent of observations.
- Table A-1 totals:
  - Total number of years in full panel: 15
  - Total All firms: 39,867
  - Total Banks: 1,603
  - Total Firm-years: 336,051
  - Total Bank-years: 17,227
- Selected country entries (exact)
  - USA: Years 15; All firms 8,369; Banks 878; Firm-years 67,209; Bank-years 9,579
  - Japan: Years 15; All firms 3,587; Banks 93; Firm-years 43,405; Bank-years 1,222
  - China: Years 14; All firms 2,241; Banks 17; Firm-years 17,700; Bank-years 136
  - India: Years 15; All firms 2,399; Banks 40; Firm-years 14,929; Bank-years 386
  - Taiwan: Years 15; All firms 1,689; Banks 20; Firm-years 15,371; Bank-years 230
  - Canada: Years 15; All firms 2,795; Banks 11; Firm-years 17,478; Bank-years 143
  - UK: Years 15; All firms 1,530; Banks 9; Firm-years 13,684; Bank-years 92
  - Hong Kong: Years 15; All firms 1,254; Banks 8; Firm-years 11,839; Bank-years 110
  - Korea: Years 15; All firms 1,695; Banks 15; Firm-years 14,237; Bank-years 150
  - Australia: Years 15; All firms 1,700; Banks 7; Firm-years 12,464; Bank-years 104
  - Germany: Years 15; All firms 997; Banks 11; Firm-years 9,791; Bank-years 141
  - France: Years 15; All firms 788; Banks 21; Firm-years 8,602; Bank-years 274
  - Malaysia: Years 15; All firms 928; Banks 10; Firm-years 9,195; Bank-years 147
  - Switzerland: Years 15; All firms 268; Banks 26; Firm-years 3,142; Bank-years 356
  - Brazil: Years 15; All firms 191; Banks 5; Firm-years 1,497; Bank-years 55
  - Russia: Years 11; All firms 543; Banks 18; Firm-years 3,138; Bank-years 112
  - Poland: Years 15; All firms 440; Banks 16; Firm-years 2,943; Bank-years 186
  - Italy: Years 15; All firms 269; Banks 19; Firm-years 2,999; Bank-years 276
  - Spain: Years 15; All firms 152; Banks 11; Firm-years 1,741; Bank-years 120
  - Luxembourg: Years 15; All firms 35; Banks 1; Firm-years 305; Bank-years 15
  - Iceland: Years 9; All firms 6; Firm-years 42
  - Bahrain: Years 2; All firms 44; Banks 14; Firm-years 87; Bank-years 28
  - Bermuda: Years 2; All firms 1; Firm-years 2
  - Latvia: Years 5; All firms 31; Firm-years 152

### XII. Appendix II & III — regression diagnostics and quantile tables (selected diagnostics)
- Appendix II (full-sample regressions; No. of obs 336,051):
  - Adj. R-squared values reported: (1): 0.341; (2): 0.345; (3): 0.344; (4): 0.345; (5): 0.345
  - Time dummies, industry fixed effects, and country fixed effects: Yes (all specifications)
  - Selected coefficient examples preserved exactly (see main text tables for full listings).
- Appendix III (quantile regressions):
  - Non-banks (Table A-3): No. of obs 318,824; (Pseudo) R-squared Q10 0.1759; Q50 0.1577; Q75 0.1170; Q90 0.0914
  - Banks (Table A-4): No. of obs 17,227; (Pseudo) R-squared Q10 0.2030; Q50 0.2080; Q75 0.2559; Q90 0.2761
  - Key quantile coefficients and t-statistics reported in Tables A-3 and A-4 indicate varying CIT responsiveness across conditional leverage quantiles and size.

*Source: _wp13221 - 1. Descriptive Statistics (PDF chapter).*

### 1. Descriptive Statistics

### 1. Descriptive Statistics

### I. Introduction — empirical context and motivation
- Debt bias: tax distortion from corporate income tax (CIT) allowing interest deductibility but not equity payments.
- Existing empirical evidence for non-financial companies:
  - Marginal tax effect on the debt-asset ratio ranges between 0.2 and 0.3.
  - Interpretation example: a 10 percentage points higher CIT rate raises the debt-to-asset ratio by somewhere between 2 and 3 percentage points.
- Financial-sector relevance:
  - Debt bias may have larger welfare costs in the financial sector due to externalities associated with systemic risk.
  - Aggregate financial-sector leverage positively affects the probability of financial crisis (cited evidence: Barrell et al. (2010); Kato, Kobayashi, and Saita (2010); De Mooij, Keen and Orihara (2013)).
- Recent empirical work on banks:
  - Keen and de Mooij (2012): CIT significantly raises bank leverage ratios; on average effect about as large as for non-banks, with important heterogeneity (smaller effects for the largest banks and for banks with higher leverage).
  - Hemmelgarn and Teichmann (2013): similar results with different specification.
  - De Mooij, Keen and Orihara (2013): elasticity gets smaller with bank size.
  - Gu, De Mooij and Poghosyan (2012): subsidiaries of 100 largest multinational banks show significant tax effects, including multinational debt shifting; effects broadly similar to non-banks.

- Contributions of this paper:
  - Uses Worldscope data containing both banks and non-banks to formally test differences in tax responsiveness.
  - Explores how firm size affects tax responsiveness using micro data on asset size and interactions of tax with polynomials of log assets (following Gordon and Lee (2001)).
  - Uses quantile regressions to examine variation of tax effects along the conditional leverage distribution (conditional on other firm characteristics).

### II. Methodology — specification and estimation
- Baseline reduced-form specification:
  - Dependent variable: lev (leverage-to-assets ratio).
  - Key explanatory variable: country-specific statutory CIT rate (τ_c).
  - Controls vector (X_isc,t) includes standard capital-structure controls.
  - Fixed effects: time fixed effects, industry fixed effects (for non-banks), and country fixed effects.
- Control variables explicitly included:
  - Book value of total assets (in logs) and, in some specifications, its square (to capture size effects).
  - Operating income over total assets (profitability).
  - Share of tangible assets in total assets (collateral).
  - Country-level controls: growth rate of GDP and rate of inflation.
- Rationale for controls and interaction terms:
  - Size: can affect access to credit, governance, and tax responsiveness; interaction of tax with first- or second-order polynomial of log assets is used to allow non-linear size effects (e.g., U-shaped patterns per Gordon and Lee (2001)).
  - Profitability: ambiguous sign (retained earnings reduce leverage but better firms may access credit more easily).
  - Collateral: typically raises leverage for non-banks; relationship for banks may differ.
  - Macroeconomic variables: GDP growth and inflation included to capture country-specific time variation and ambiguous theoretical effects on leverage.
- Estimation approach:
  - All regressions estimated by OLS.
  - Three samples considered: non-banks, banks, and combined sample (with bank dummies to distinguish effects).
  - Standard errors robust and clustered at the firm level.
- Quantile regressions:
  - Used to approximate conditional median and other quantiles of leverage (Koenker and Bassett, 1978; Koenker and Hallock, 2001).
  - Purpose: examine whether marginal tax effects differ along the conditional leverage distribution (e.g., whether highly levered banks respond differently).

### III. Data — sample composition and key descriptive statistics
- Data source: Worldscope (Thomson Financial).
- Sample years: 1996 to 2010.
- Final sample after excluding observations with missing required variables:
  - 38,867 firms from 69 countries.
  - 336,051 firm-year observations in total.
  - Banks in the sample: 1,603 banks contributing 17,227 bank-year observations.
  - Banks’ share in total number of firms: 4.1 percent.
  - Banks’ share in the overall sample of observations: 5.1 percent.

### IV. Empirical considerations highlighted
- Ambiguity about relative tax responsiveness of banks versus non-banks:
  - Theoretical ambiguity arises because banks face capital requirements (potentially limiting tax responsiveness) but often hold buffers above requirements and may exhibit similar debt-choice behavior to non-financial firms.
- Importance of heterogeneity:
  - Size heterogeneity and position in the leverage distribution are central to understanding marginal tax responses.
  - Quantile regressions provide robust estimates in skewed distributions (noted to be the case for banks in this sample).

*Source: _wp13221 - 1. Descriptive Statistics (PDF chapter).*

### Appendix I gives more information about the sample by country.

### _wp13221 - Appendix I gives more information about the sample by country.

### Data and sample
- Data reflect consolidated accounts of firms (consolidated statements used rather than unconsolidated).
- Note on consolidation: inclusion of foreign branches or wholly-owned subsidiaries may mean the tax rate applying to the parent might not properly reflect the actual tax on variables in the accounts; most countries tax branches on a worldwide basis (and US also taxes subsidiaries on a worldwide basis), so the tax rate of the parent matters for foreign source profit.
- Note on relevance: consolidated statements matter for the financial stability of a firm and motivate use of consolidated statements.
- Number of observations:
  - Full sample (all firms): 336,051
  - Banks subsample: 17,227
  - Non-banks subsample: 318,824

### Descriptive statistics (Table 1 highlights)
- Leverage Ratio (ratio of total liabilities to total assets)
  - All firms: Mean 0.4863; SD 0.2541; Median 0.4924; Min 0; Max 1
  - Banks: Mean 0.9006; SD 0.0751; Median 0.9130; Min 0; Max 0.9993
  - Non-banks: Mean 0.4639; SD 0.2408; Median 0.4739; Min 0; Max 1
- Corporate income tax rate
  - All firms: Mean 0.3325; SD 0.0830; Median 0.3399; Min 0; Max 0.5725
  - Banks: Mean 0.3665; SD 0.0723; Median 0.4076; Min 0; Max 0.5725
  - Non-banks: Mean 0.3307; SD 0.0831; Median 0.335; Min 0; Max 0.5725
- Log total assets
  - All firms: Mean 18.7271; SD 2.3667; Median 18.6637; Min 6.9078; Max 29.1844
  - Banks: Mean 21.4586; SD 2.2345; Median 21.0481; Min 11.8914; Max 29.1844
  - Non-banks: Mean 18.5795; SD 2.2823; Median 18.5457; Min 6.9078; Max 28.363
- Operating income over assets
  - All firms: Mean 0.00643; SD 0.2283; Median 0.0381; Min -2.5556; Max 0.7459
  - Banks: Mean 0.0095; SD 0.0331; Median 0.0110; Min -1.8278; Max 0.5457
  - Non-banks: Mean 0.0062; SD 0.2342; Median 0.0425; Min -2.5556; Max 0.7459
- Collateral (share of tangible assets in total assets)
  - All firms: Mean 0.2920; SD 0.2528; Median 0.2399; Min 0; Max 1
  - Banks: Mean 0.0189; SD 0.0193; Median 0.0158; Min 0; Max 0.9866
  - Non-banks: Mean 0.3067; SD 0.2512; Median 0.2595; Min 0; Max 1
- GDP growth rate
  - All firms: Mean 0.0599; SD 0.0613; Median 0.0546; Min -0.1143; Max 0.4298
  - Banks: Mean 0.0564; SD 0.0544; Median 0.0540; Min -0.1143; Max 0.4298
  - Non-banks: Mean 0.0601; SD 0.0616; Median 0.0546; Min -0.1143; Max 0.4298
- Inflation rate
  - All firms: Mean 0.0250; SD 0.0307; Median 0.0219; Min -0.0623; Max 0.2586
  - Banks: Mean 0.0278; SD 0.0278; Median 0.0244; Min -0.0623; Max 0.2586
  - Non-banks: Mean 0.0249; SD 0.0309; Median 0.0212; Min -0.0623; Max 0.2586

### Size distributions and salient quantiles
- Non-banks (log of total assets)
  - Mean log assets 18.58 (US$ 117 million)
  - 25th percentile: log 17.16 (US$ 28 million)
  - 75th percentile: log 20 (US$ 482 million)
  - 1 percent quantile: log 12.66 (US$ 316 thousand)
  - 5 percent quantile: log 14.84 (US$ 2.8 million)
  - 95 percent quantile: log 22.47 (US$ 5.8 billion)
  - 99 percent quantile: log 24.29 (US$ 35.4 billion)
  - Distribution: approximately standard normal in log size, symmetric around mean
- Banks (log of total assets)
  - Mean log assets 21.45 (US$ 2.1 billion)
  - 1 percent quantile: log 17.5 (US$ 40 million)
  - 5 percent quantile: log 18.47 (US$ 105 million)
  - 90 percent quantile: log 24.5 (US$ 44 billion)
  - 95 percent quantile: log 25.48 (US$ 116 billion)
  - 99 percent quantile: log 27.37 (US$ 771 billion)
  - Distribution: skewed with a long right tail (few small banks, many very large banks)

### Results — Non-banks (summary of Table 2 and main findings)
- Regression approach: regress leverage ratio on statutory CIT rates across six specifications:
  - Column (1): simple regression excluding fixed effects.
  - Column (2): adds time, country and industry fixed effects.
  - Column (3): adds firm size (log total assets).
  - Column (4): adds operating income over total assets, collateral (tangibility), and interaction log assets * CIT.
  - Column (5): adds GDP growth and inflation.
  - Column (6): adds quadratic polynomial of firm size and interaction of that with CIT.
- Estimated CIT coefficients (from Table 2, columns (1)–(6)):
  - Column (1) CIT rate: 0.3411***
    - t-statistic (in parentheses): (26.6995)
  - Column (2) CIT rate: 0.2172***
    - (9.3032)
  - Column (3) CIT rate: 0.1974***
    - (8.7968)
  - Column (4) CIT rate: -0.0428
    - (-0.4142)
  - Column (5) CIT rate: -0.0354
    - (-0.3429)
  - Column (6) CIT rate: 2.9103***
    - (5.7188)
- Log assets and interactions (selected coefficients and t-statistics):
  - Log assets: 0.0269*** (column (1) shown), 0.0258***, 0.0259***, 0.1036*** (associated t-statistics shown: (58.2108), (13.7773), (13.7931), (5.1685))
  - Log assets * CIT: 0.0122**; 0.0121**; -0.3050*** (t-statistics: (2.3015), (2.2837), (-5.5942))
  - Log assets sq.: -0.0021*** (t-statistic: (-3.8456))
  - Log assets sq * CIT: 0.0084*** (t-statistic: (5.7757))
- Other covariates (columns (4)–(6)):
  - Profitability (Operating income over assets): -0.1110***; -0.1111***; -0.1010*** (t-statistics: (-34.0219), (-34.0476), (-30.5079))
  - Collateral: 0.0629***; 0.0628***; 0.0643*** (t-statistics: (14.3610), (14.3548), (14.6414))
  - GDP growth: 0.0324**; 0.0307** (t-statistics: (2.5546), (2.4233))
  - Inflation: -0.0813***; -0.0828*** (t-statistics: (-3.7410), (-3.8085))
- Tax effects at sample mean size (reported in Table 2)
  - Column (1): 0.3411*** (26.6995)
  - Column (2): 0.2172*** (9.3032)
  - Column (3): 0.1974*** (8.7968)
  - Column (4): 0.1845*** (8.25)
  - Column (5): 0.1901*** (8.43)
  - Column (6): 0.1469*** (6.21)
- Goodness-of-fit and sample size
  - No. of obs: 318,824 (for all columns)
  - Adj. R-squared values shown: 0.0139 (column (1)), 0.171 (column (2)), 0.222 (column (3)), 0.234 (column (4)), 0.234 (column (5)), 0.236 (column (6))
- Main interpretations and patterns
  - All tax coefficients in columns (1)–(3) statistically significant at the 1 percent level.
  - Column (3) interpretation: the leverage ratio rises by 2 percentage points if the CIT rate increases by 10 percentage points (based on coefficient 0.1974).
  - Size interaction:
    - In columns (4) and (5) the direct tax coefficient is negative while the interaction of CIT rate with log assets is positive and significant at 5 percent, implying larger firms' debt ratios tend to be more responsive to the CIT rate than smaller firms.
    - Column (6) (quadratic interaction) — all tax terms significant at 1 percent; marginal tax effect at sample mean log assets is 0.15 (t-value: 6.21).
  - Heterogeneity across firm size (Figure 3 description):
    - The quadratic interaction implies a convex shape of the marginal tax effect along the size distribution: both very small and very large firms show relatively large tax responses; intermediate firms show relatively small tax responses.
    - Over the most relevant range, the tax response is always smaller than 0.5; 99 percent of firms have a tax coefficient below 0.5.
    - Thresholds where marginal tax response exceeds 0.5:
      - Firms with log assets below 11.62 (asset value US$ 111 thousand)
      - Firms with log assets exceeding 24.68 (asset value US$ 52.3 billion)
    - Asset concentration: firms with log assets exceeding 24.68 own approximately 45.8 percent of the total assets in the non-banks sample.
  - Non-tax determinants:
    - Firm size exerts a significant positive effect on leverage in columns (3)–(6).
    - Profitability exerts a significant negative effect on leverage (consistent with higher retained earnings increasing equity).
    - Collateral has a positive and significant coefficient (1 percent).
    - GDP growth is associated with an increase in corporate debt (columns (5) and (6)), while inflation has a negative effect.

### Results — Banks (summary excerpt)
- Regression sequence for banks mirrors that for non-banks.
- Tax coefficients reported in columns (1)–(3) are statistically significant at the 1 percent level.
- Magnitude of tax response for banks: between 0.06 and 0.14 (i.e., slightly lower than for non-banks in Table 2).
- (Text is truncated after reporting the range; no further bank regression coefficients are provided in the supplied content.)

*Source: _wp13221 - Appendix I gives more information about the sample by country.*

### 1.3 percentage points if the marginal tax rate increases by 10 percentage points.

### _wp13221 - 1.3 percentage points if the marginal tax rate increases by 10 percentage points.

### Tax–Size Interaction for Banks (Table 3; regression (4) and (5))
- Interaction of tax and bank size:
  - The interaction term is significantly negative (at 5 percent confidence), implying the marginal tax effect on banks’ leverage ratio is a decreasing function of bank size.
  - Evaluated at the sample mean of bank size:
    - Estimated marginal tax effect on the leverage ratio is 0.20 in column (4) (t-value: 4.53).
    - Estimated marginal tax effect on the leverage ratio is 0.21 in column (5) (t-value: 4.66).
  - These marginal effects are larger than in columns (1)–(3) and very similar to the marginal impact for an average non-bank.
- Column (6) (second-order polynomial of size interacted with tax):
  - None of the individual tax coefficients are significant, including the interaction with the log of size squared (t-value shown as (-0.1212) for that term).
  - The combined tax effect of the three tax terms together is significant over an important range.
  - Over the relevant range of the size distribution, the quadratic interaction coefficient is very small, suggesting near-linearity; the linear interaction (regression (5)) is preferred to illustrate the relationship.

### Marginal Tax Effect along Bank Size Distribution (Figure 4 & text)
- The negative interaction coefficient implies a downward-sloping marginal tax effect across bank size:
  - Tax effect is largest for small banks, with a marginal impact close to 0.5.
  - The effect gradually falls and becomes insignificant for banks with log assets of 25 (total assets of US$ 72 billion) — this applies to the 7.1 percent largest banks.
  - For the 7.1 percent largest banks, the point estimate becomes negative but is insignificant due to very large standard errors.
  - For the median bank (log assets 21.0), the marginal coefficient is 0.22 (t-value of 4.53).

### Non-tax Determinants for Banks versus Non-Banks (regressions (3)–(6) in Table 3)
- Differences in non-tax factor effects:
  - Size enters with a positive coefficient for banks, as for non-banks.
  - Profitability:
    - Not significant for banks.
    - Significant negative coefficient for non-banks.
  - Collateral:
    - Negative coefficient for banks (does not necessarily reflect usual role for non-banks where collateral enters positively).
  - GDP growth and inflation:
    - Insignificant coefficients for banks.

### Banks and Non-Banks Comparisons (Table 4)
- Full-sample regressions (banks + non-banks) focus on tax variables and their interactions with size.
- Column (1):
  - CIT rate coefficient: 0.1873*** (t-value: 8.7099) — dominated by non-bank observations.
- Column (2):
  - Bank dummy interactions:
    - Bank * CIT = 0.0695** (t-value: 2.2107) — average response by banks.
    - (1-Bank) * CIT = 0.1933*** (t-value: 8.6134) — average response for non-banks.
  - Hence, in this regression, banks’ average response is smaller than non-banks’.
- Column (3):
  - Linear interaction of asset size and tax:
    - Non-banks: significant positive interaction (larger non-bank firms are more responsive).
    - Banks: negative correlation (larger banks are less responsive).
- Column (4):
  - Adds quadratic interaction of tax with asset size for both banks and non-banks:
    - Non-banks: interaction is convex (significant positive coefficient for quadratic size term).
    - Banks: coefficient insignificant and very small.
- Column (5):
  - Preferred specification: non-linear size effect only for non-banks.

### Testing Equivalence of Tax Effects across Bank Size Distribution (Table 5)
- Sample means:
  - Mean non-bank log assets: 18.58 (~US$ 117.4 million).
  - Mean bank log assets: 21.45 (~US$ 2.1 billion).
  - Marginal tax effects at sample means:
    - Non-banks: 0.1469*** (t-value: 6.21).
    - Banks: 0.2108*** (t-value: 4.67).
    - F-Statistic: 1.57 (Prob > F: 0.21) — cannot reject equality of coefficients between banks and non-banks at sample means.
- Marginal tax effects evaluated at bank-size quantiles (joint effects from regression (5) of Table 4):
  - Bank size Q10 (Log assets: 18.91; ~US$ 163.9 million; Non-bank size Q57):
    - Non-banks: 0.1504*** (t-value: 6.38).
    - Banks: 0.2878*** (t-value: 3.88).
    - F-Statistic: 3.11 (Prob > F: 0.0778) — borderline rejection at conventional levels.
  - Bank size Q25 (Log assets: 19.76; ~US$ 379.9 million; Non-bank size Q71.9):
    - Non-banks: 0.1674*** (t-value: 7.13).
    - Banks: 0.2624*** (t-value: 4.11).
    - F-Statistic: 1.95 (Prob > F: 0.1626).
  - Bank size Q50 (Log assets: 21.05; ~US$ 1.4 billion; Non-bank size Q86.9):
    - Non-banks: 0.2167*** (t-value: 8.65).
    - Banks: 0.2233*** (t-value: 4.54).
    - F-Statistic: 0.01 (Prob > F: 0.9055).
  - Bank size Q75 (Log assets: 23.04; ~US$ 10.2 billion; Non-bank size Q96.9):
    - Non-banks: 0.3479*** (t-value: 9.25).
    - Banks: 0.1630*** (t-value: 4.69).
    - F-Statistic: 13.03 (Prob > F: 0.0003) — rejects equivalence.
  - Bank size Q90 (Log assets: 24.51; ~US$ 44.1 billion; Non-bank size Q99.3):
    - Non-banks: 0.4872*** (t-value: 8.58).
    - Banks: 0.1186*** (t-value: 3.23).
    - F-Statistic: 29.72 (Prob > F: 0.0000) — rejects equivalence.
  - Bank size Q95 (Log assets: 25.48; ~US$ 116.3 billion; Non-bank size Q99.8):
    - Non-banks: 0.5991*** (t-value: 8.14).
    - Banks: 0.0892** (t-value: 2.03).
    - F-Statistic: 35.39 (Prob > F: 0.0000) — rejects equivalence.
- Interpretation:
  - Table 5 cannot reject equivalence for Q10, Q25, and Q50.
  - For larger firms (Q75 and above), non-banks’ tax responsiveness becomes significantly larger than banks’.
  - At Q95, non-bank tax effect is 0.5991*** versus 0.0892** for banks.

### Quantile Regressions — Non-banks (Table 6)
- Setup:
  - Quantiles of conditional leverage distribution: Q10, Q50, Q75, Q90.
  - Size evaluation at non-bank size Q10 (Log assets: 15.77; ~US$ 7.1 million), Q50 (Log assets: 18.55; ~US$ 113.7 million), Q90 (Log assets: 21.50; ~US$ 2.2 billion).
  - Interaction: CIT rate with a second-order polynomial of size.
- Joint tax effects (selected values and t-values):
  - Non-bank size Q10:
    - Q10: 0.0340 (t-value: 1.48).
    - Q50: 0.2103*** (t-value: 7.43).
    - Q75: 0.2898*** (t-value: 9.66).
    - Q90: 0.2791*** (t-value: 11.03).
    - F-test Q10 = Q90: 55.53 (Prob: 0.0000) — rejects equality across leverage quantiles.
  - Non-bank size Q50:
    - Q10: 0.1031*** (t-value: 4.28).
    - Q50: 0.1410*** (t-value: 6.05).
    - Q75: 0.1855*** (t-value: 7.96).
    - Q90: 0.2133*** (t-value: 9.74).
    - F-test Q10 = Q90: 12.28 (Prob: 0.0005).
  - Non-bank size Q90:
    - Q10: 0.2452*** (t-value: 9.44).
    - Q50: 0.2448*** (t-value: 10.37).
    - Q75: 0.1957*** (t-value: 8.24).
    - Q90: 0.1449*** (t-value: 6.61).
    - F-test Q10 = Q90: 9.85 (Prob: 0.0017).
- Additional F-tests of equality across sizes:
  - Q10 = Q90 (size): 175.71 (Prob: 0.0000) for Q10 leverage quantile; other statistics given for other leverage quantiles (2.75 (0.0975); 14.65 (0.0001); 40.64 (0.0000)).
- Interpretation:
  - Patterns vary by conditional leverage quantile:
    - For Q50 and Q75 of conditional leverage, a convex relationship between asset size and tax coefficient (smallest at Q50 of size distribution).
    - For Q10 conditional leverage, tax responsiveness monotonically increases with size.
    - For Q90 conditional leverage, tax responsiveness monotonically decreases with size.
  - F-tests reject equivalence of tax effects between size Q10 and Q90 for all conditional quantiles at 10 percent; for Q50 of conditional leverage, equivalence not rejected at 5 percent.

### Quantile Regressions — Banks (Table 7)
- Setup:
  - Quantiles of conditional leverage distribution: Q10, Q50, Q75, Q90.
  - Size evaluation at bank size Q10 (Log assets: 18.91; ~US$ 163.9 million), Q50 (Log assets: 21.05; ~US$ 1.4 billion), Q90 (Log assets: 24.51; ~US$ 44.1 billion).
  - Interaction: CIT rate with a first-order polynomial of size (preferred for banks).
- Joint tax effects (selected values and t-values):
  - Bank size Q10:
    - Q10: 0.3119*** (t-value: 5.85).
    - Q50: 0.1434*** (t-value: 9.65).
    - Q75: 0.1207*** (t-value: 7.97).
    - Q90: 0.0999*** (t-value: 7.54).
    - F-test Q10 = Q90: 15.65 (Prob: 0.0001).
  - Bank size Q50:
    - Q10: 0.2416*** (t-value: 5.69).
    - Q50: 0.1124*** (t-value: 9.85).
    - Q75: 0.0960*** (t-value: 8.42).
    - Q90: 0.0731*** (t-value: 6.90).
    - F-test Q10 = Q90: 15.41 (Prob: 0.0001).
  - Bank size Q90:
    - Q10: 0.1274*** (t-value: 4.26).
    - Q50: 0.062*** (t-value: 6.64).
    - Q75: 0.0559*** (t-value: 6.66).
    - Q90: 0.0297*** (t-value: 2.81).
    - F-test Q10 = Q90: 9.73 (Prob: 0.0018).
- F-tests of equality across bank sizes (Q10 = Q90) for each leverage quantile yield large test statistics and Prob = 0.0000, rejecting equality across size groups for all conditional quantiles.
- Interpretation:
  - Tax response for banks declines with asset size for all conditional leverage quantiles.
  - Tax response for banks declines with the conditional leverage ratio (banks with high conditional leverage are less responsive), irrespective of bank size.
  - Even the largest banks with the highest conditional leverage ratio (Q90) show a positive and significant marginal impact of 0.03 (reported in text).

### Conclusions (Section IV)
- Main findings:
  - Tax bias toward debt induced by corporate tax systems matters for both banks and non-banks.
  - Preferred specification: an average bank is equally responsive to tax incentives as an average non-bank.
- Differences across size distributions:
  - Non-banks (OLS): U-shaped pattern between asset size and tax responsiveness — large non-bank firms are relatively responsive.
  - Banks: linearly declining relationship between asset size and tax responsiveness — large banks are relatively unresponsive to tax.
- Quantile regression insights:
  - Capital-tight banks (high conditional leverage) are significantly less responsive to tax than capital-abundant banks — holds for banks of all sizes.
  - Even the largest banks with high conditional leverage have a significant, positive tax response.
  - For non-banks, tax responsiveness varies with conditional leverage: largest responses found for (i) large firms with low conditional leverage and (ii) small firms with high conditional leverage. For large firms and banks, patterns along conditional leverage distribution are similar.

*Source: _wp13221 - 1.3 percentage points if the marginal tax rate increases by 10 percentage points.*

### Appendix I. Data by Country

### Appendix I. Data by Country

### Sample coverage and geographic distribution
- Sample covers 70 countries with firm-level data spanning up to 1996–2010.
- For 42 of the 70 countries covered, there is data for all years from 1996 to 2010.
- In 28 countries, the number of years of data falls below the full sample period.
- The United States makes up more than 20 percent of all observations.
- Taken together the US, Japan, China, Canada, Taiwan and India account for more than 52 percent of observations.
- Adding firm-years observed for Korea, Australia, UK, Hong Kong, Malaysia, Germany and France covers more than 76 percent of the observations.

### Table A-1: Sample Composition by Countries — key statistics and totals
- Total number of years in full panel: 15
- Total All firms: 39,867
- Total Banks: 1,603
- Total Firm-years: 336,051
- Total Bank-years: 17,227

- Selected country entries (preserve exact figures as presented):
  - USA: Years 15; All firms 8,369; Banks 878; Firm-years 67,209; Bank-years 9,579
  - Japan: Years 15; All firms 3,587; Banks 93; Firm-years 43,405; Bank-years 1,222
  - China: Years 14; All firms 2,241; Banks 17; Firm-years 17,700; Bank-years 136
  - India: Years 15; All firms 2,399; Banks 40; Firm-years 14,929; Bank-years 386
  - Taiwan: Years 15; All firms 1,689; Banks 20; Firm-years 15,371; Bank-years 230
  - Canada: Years 15; All firms 2,795; Banks 11; Firm-years 17,478; Bank-years 143
  - UK: Years 15; All firms 1,530; Banks 9; Firm-years 13,684; Bank-years 92
  - Hong Kong: Years 15; All firms 1,254; Banks 8; Firm-years 11,839; Bank-years 110
  - Korea: Years 15; All firms 1,695; Banks 15; Firm-years 14,237; Bank-years 150
  - Australia: Years 15; All firms 1,700; Banks 7; Firm-years 12,464; Bank-years 104
  - Germany: Years 15; All firms 997; Banks 11; Firm-years 9,791; Bank-years 141
  - France: Years 15; All firms 788; Banks 21; Firm-years 8,602; Bank-years 274
  - Malaysia: Years 15; All firms 928; Banks 10; Firm-years 9,195; Bank-years 147
  - Switzerland: Years 15; All firms 268; Banks 26; Firm-years 3,142; Bank-years 356
  - Brazil: Years 15; All firms 191; Banks 5; Firm-years 1,497; Bank-years 55
  - Russia: Years 11; All firms 543; Banks 18; Firm-years 3,138; Bank-years 112
  - Poland: Years 15; All firms 440; Banks 16; Firm-years 2,943; Bank-years 186
  - Italy: Years 15; All firms 269; Banks 19; Firm-years 2,999; Bank-years 276
  - Spain: Years 15; All firms 152; Banks 11; Firm-years 1,741; Bank-years 120
  - Mexico: (not listed — omitted from source table entries)

- Examples of smaller-sample countries (preserve exact figures):
  - Luxembourg: Years 15; All firms 35; Banks 1; Firm-years 305; Bank-years 15
  - Iceland: Years 9; All firms 6; Firm-years 42
  - Bahrain: Years 2; All firms 44; Banks 14; Firm-years 87; Bank-years 28
  - Bermuda: Years 2; All firms 1; Firm-years 2
  - Latvia: Years 5; All firms 31; Firm-years 152

### Notes on table construction
- Table A-1 columns: Country; Number of years; All firms; Banks; Firm-years; Bank-years.
- Some country rows contain missing entries for certain columns in the reproduced excerpt; where absent, values are not reported.

---

### Appendix II. Regressions on the Full Sample of Non-banks and Banks

### Regression setup and coverage
- Table A-2 reports capital structure regressions for the full sample of banks and non-banks.
- Number of observations: 336,051 (for all reported specifications).
- Adjusted R-squared values reported:
  - (1): 0.341
  - (2): 0.345
  - (3): 0.344
  - (4): 0.345
  - (5): 0.345
- Time dummies: Yes (all specifications)
- Industry fixed effects: Yes (all specifications)
- Country fixed effects: Yes (all specifications)

### Key estimated coefficients and inference (selected coefficients preserved exactly)
- CIT rate:
  - Specification (1): CIT rate 0.1873*** (8.7099)
- Interaction terms with bank indicator and CIT:
  - (1-Bank) * CIT 0.1933*** (8.6134) in (1)
  - (1-Bank) * CIT -0.0354 (-0.3429) in (2)
  - (1-Bank) * CIT 2.9103*** (5.7181) in (3) and (4)
  - Bank * CIT 0.0695** (2.2107) in (1)
  - Bank * CIT 0.8598** (2.5466) in (2) and (5)
  - Bank * CIT 1.6657 (0.7317) in (3)
- Log assets:
  - Log assets 0.0197*** (5.4182) in (1)
  - (1-Bank) * Log assets -0.0061 (-1.4739) in (2)
  - (1-Bank) * Log assets 0.0259*** (13.7915) in (3)
  - (1-Bank) * Log assets 0.1036*** (5.1679) in (4) and (5)
  - Bank * Log assets 0.1447*** (8.8900) in (1)
  - Bank * Log assets 0.0255*** (4.4553) in (2) and (5)
  - Bank * Log assets 0.1840** (2.4253) in (3)
- Log assets squared:
  - Log assets sq. 0.0002** (2.4538) in (1)
  - (1-Bank) * Log assets sq. 0.0010*** (8.6538) in (2)
  - (1-Bank) * Log assets sq. -0.0021*** (-3.8451) in (3) and (4)
  - Bank * Log assets sq. -0.0030*** (-8.2326) in (2)
  - Bank * Log assets sq. -0.0032** (-1.9931) in (3)
- Triple interactions (Log assets * CIT and Log assets sq * CIT):
  - (1-Bank) * Log assets * CIT 0.0121** (2.2834) in (4)
  - (1-Bank) * Log assets * CIT -0.3050*** (-5.5936) in (5)
  - Bank * Log assets * CIT -0.0302** (-2.1194) in (1) and (4)
  - Bank * Log assets * CIT -0.0562 (-0.2886) in (2)
  - (1-Bank) * Log assets sq * CIT 0.0084*** (5.7750) in (4) and (5)
  - Bank * Log ass. sq * CIT -0.0005 (-0.1215) in (4)
- Operating Income:
  - Operating Income -0.1012*** (-30.5968) in (1)
  - (1-Bank) * Operating Income -0.1009*** (-30.5312) in (1)
  - (1-Bank) * Operating Income -0.1111*** (-34.0437) in (2)
  - (1-Bank) * Operating Income -0.1010*** (-30.5043) in (3)–(5)
  - Bank * Operating Income -0.0729 (-0.6979) in (1)
  - Bank * Operating Income 0.0484 (0.4225) in (2) and (5)
  - Bank * Operating Income -0.0794 (-0.7640) in (3)
- Collateral:
  - Collateral 0.0651*** (14.8541) in (1)
  - (1-Bank) * Collateral 0.0645*** (14.6757) in (1)
  - (1-Bank) * Collateral 0.0628*** (14.3532) in (2)
  - (1-Bank) * Collateral 0.0643*** (14.6397) in (3)–(5)
  - Bank * Collateral -0.2952** (-2.0447) in (1)
  - Bank * Collateral -0.3945*** (-2.6881) in (2) and (5)
  - Bank * Collateral -0.2533* (-1.8001) in (3)
- Macroeconomic variables:
  - GDP growth 0.0303** (2.4907) in (1)
  - (1-Bank) * GDP growth 0.0312** (2.4621) in (1)
  - (1-Bank) * GDP growth 0.0324** (2.5543) in (2)
  - (1-Bank) * GDP growth 0.0307** (2.4230) in (3)–(5)
  - Bank * GDP growth 0.0447* (1.7624) in (1)
  - Bank * GDP growth -0.0004 (-0.0152) in (2) and (5)
  - Bank * GDP growth 0.0253 (1.0005) in (3)
  - Inflation -0.0766*** (-3.6833) in (1)
  - (1-Bank) * Inflation -0.0836*** (-3.8402) in (1)
  - (1-Bank) * Inflation -0.0813*** (-3.7406) in (2)
  - (1-Bank) * Inflation -0.0828*** (-3.8081) in (3)–(5)
  - Bank * Inflation -0.0556 (-1.4012) in (1)
  - Bank * Inflation -0.0590 (-1.3898) in (2) and (5)
  - Bank * Inflation -0.0733* (-1.8391) in (3)

### Model controls and diagnostics
- Constant reported as -0.0393 (-1.1008) in specification (1).
- All specifications include time dummies, industry fixed effects, and country fixed effects.
- No. of obs for all specifications: 336,051.
- Adj. R-squared across specifications ranges from 0.341 to 0.345.

*Source: _wp13221 - Appendix I. Data by Country*

### Appendix III. Quantile Regressions

### Appendix III. Quantile Regressions

### Table A-3 — Quantile Regression for Non-banks: Interaction of Tax and Second-order Polynomial of Size (Q10, Q50, Q75, Q90)
- Sample size: No. of obs 318,824
- (Pseudo) R-squared:
  - Q10: 0.1759
  - Q50: 0.1577
  - Q75: 0.1170
  - Q90: 0.0914
- Coefficients and significance (t-statistics in parentheses) by quantile:
  - CIT rate
    - Q10: 0.8622*** (4.42)
    - Q50: 3.766*** (10.70)
    - Q75: 3.045*** (7.18)
    - Q90: 0.6832* (1.77)
  - Log assets
    - Q10: -0.0117 (-1.52)
    - Q50: 0.1650*** (13.28)
    - Q75: 0.1309*** (8.92)
    - Q90: 0.0074 (0.50)
  - Log assets sq.
    - Q10: 0.0009*** (4.13)
    - Q50: -0.0036*** (-11.26)
    - Q75: -0.0027*** (-7.28)
    - Q90: 0.0004 (0.99)
  - Log assets * CIT
    - Q10: -0.1185*** (-5.58)
    - Q50: -0.3952*** (-11.15)
    - Q75: -0.2905*** (-6.89)
    - Q90: -0.0269 (-0.69)
  - Log assets sq * CIT
    - Q10: 0.0042*** (7.35)
    - Q50: 0.0108*** (12.04)
    - Q75: 0.0073*** (7.04)
    - Q90: 0.0001 (0.09)
  - Operating income
    - Q10: -0.0389*** (-20.33)
    - Q50: -0.1089*** (-37.82)
    - Q75: -0.1726*** (-43.47)
    - Q90: -0.1863*** (-45.45)
  - Collateral
    - Q10: 0.0909*** (42.92)
    - Q50: 0.0623*** (23.35)
    - Q75: 0.0148*** (5.22)
    - Q90: -0.0098*** (-3.69)
  - GDP growth
    - Q10: 0.0203 (1.19)
    - Q50: 0.0762*** (4.12)
    - Q75: 0.0292 (1.35)
    - Q90: 0.0039 (0.17)
  - Inflation
    - Q10: -0.1862*** (-6.60)
    - Q50: -0.0727** (-2.46)
    - Q75: 0.0196 (0.63)
    - Q90: 0.0209 (0.63)
  - Constant
    - Q10: 0.0809 (1.13)
    - Q50: -1.4482*** (-11.64)
    - Q75: -0.9817*** (-6.62)
    - Q90: 0.3996*** (2.67)
- Fixed effects:
  - Time dummies: Yes (all quantiles)
  - Country Dummies: Yes (all quantiles)
  - Industry fixed effects: Yes (all quantiles)

### Key findings from Table A-3 (Non-banks)
- The interaction terms indicate that the effect of CIT rate on leverage for non-banks varies with firm size:
  - Positive CIT rate coefficients at Q10, Q50, and Q75, with the largest point estimate at Q50: 3.766***.
  - Negative coefficients on Log assets * CIT at Q10, Q50, and Q75, indicating diminishing tax effect on leverage as size increases in these quantiles.
  - Positive Log assets sq * CIT coefficients at Q10, Q50, and Q75, suggesting a second-order nonlinear interaction between size and tax with statistically significant curvature at these quantiles.
- Operating income is consistently negative and highly significant across all quantiles (e.g., Q90: -0.1863***).
- Collateral has a positive and significant association with the dependent variable at lower and middle quantiles (Q10: 0.0909***; Q50: 0.0623***; Q75: 0.0148***), turning negative at Q90: -0.0098***.
- Inflation is negative and significant at lower quantiles (Q10: -0.1862***; Q50: -0.0727**) but not significant at higher quantiles.

### Table A-4 — Quantile Regression for Banks: Interaction of Tax and First-order Polynomial of Size (Q10, Q50, Q75, Q90)
- Sample size: No. of obs 17,227
- (Pseudo) R-squared:
  - Q10: 0.2030
  - Q50: 0.2080
  - Q75: 0.2559
  - Q90: 0.2761
- Coefficients and significance (t-statistics in parentheses) by quantile:
  - CIT rate
    - Q10: 0.9358*** (5.56)
    - Q50: 0.4185*** (7.74)
    - Q75: 0.3397*** (6.02)
    - Q90: 0.3371*** (6.32)
  - Log assets
    - Q10: 0.0237*** (9.09)
    - Q50: 0.0094*** (10.25)
    - Q75: 0.0073*** (8.12)
    - Q90: 0.0070*** (7.86)
  - Log assets * CIT
    - Q10: -0.0330*** (-5.17)
    - Q50: -0.0145*** (-6.27)
    - Q75: -0.0116 (-5.07)
    - Q90: -0.0125*** (-5.46)
  - Operating income
    - Q10: 0.0458 (0.31)
    - Q50: -0.5557*** (-7.58)
    - Q75: -0.6892*** (-13.51)
    - Q90: -0.7418*** (-15.42)
  - Collateral
    - Q10: -0.4728*** (-3.31)
    - Q50: -0.1837*** (-5.12)
    - Q75: -0.1483*** (-4.89)
    - Q90: -0.1476*** (-7.22)
  - GDP growth
    - Q10: 0.0623* (1.81)
    - Q50: 0.0469*** (3.96)
    - Q75: 0.0267*** (3.09)
    - Q90: 0.0249** (2.44)
  - Inflation
    - Q10: -0.1480** (-2.45)
    - Q50: -0.1041*** (-6.09)
    - Q75: -0.0401** (-2.27)
    - Q90: -0.0279* (-1.71)
  - Constant
    - Q10: 0.1830** (2.22)
    - Q50: 0.6764*** (30.09)
    - Q75: 0.7397*** (33.19)
    - Q90: 0.7683*** (36.44)
- Fixed effects:
  - Time dummies: Yes (all quantiles)
  - Country Dummies: Yes (all quantiles)

### Key findings from Table A-4 (Banks)
- CIT rate is positive and statistically significant across all reported bank quantiles, with the largest point estimate at Q10: 0.9358***.
- Log assets are positive and highly significant across quantiles, indicating larger banks are associated with higher values of the dependent variable in these quantiles.
- The interaction Log assets * CIT is negative and statistically significant at Q10, Q50, and Q90, indicating that the positive association of CIT rate with the dependent variable weakens with bank size in these quantiles.
- Operating income is negative and significant at higher quantiles (Q50: -0.5557***; Q75: -0.6892***; Q90: -0.7418***) but not at Q10.
- Collateral is negative and significant across all quantiles for banks (e.g., Q10: -0.4728***).
- GDP growth is positive and statistically significant at Q50, Q75, and Q90; weakly significant at Q10 (0.0623*).
- Inflation has a negative and significant association at all quantiles, with decreasing magnitude at higher quantiles.

*Appendix III. Quantile Regressions — Tables A-3 and A-4, as provided in the source PDF.*

---


_Source: https://www.imf.org/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2013/_wp13221.pdf_