## _wp16217

## Source details

**Canonical URL:** [_wp16217](https://www.imf.org/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2016/_wp16217.pdf)

## Other formats

- [Markdown version](/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2016/_wp16217.pdf.md)
- [Structured JSON version](/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2016/_wp16217.pdf.json)

---

### Tax bias and debt bias: definitions and distinctions
- Tax systems typically allow a deduction of interest paid on debt but not on equity (Sorensen, 2014; Weichenrieder and Klautke, 2008).
- Debt bias: the incentive created by tax deductibility of interest that leads profit-maximizing firms to take on more debt than they would absent that incentive.
- Debt shifting: distinct distortion where, conditional on total leverage, firms allocate debt across jurisdictions to shift profits; does not change total debt and is primarily a BEPS concern.

### Financial-stability implications and interaction with regulation
- Debt bias increases leverage and raises the probability of distress or bankruptcy for non-financial and financial firms.
- Financial firms differ from non-financial firms:
  - The systemic importance of the financial system is much larger, producing negative externalities from financial distress.
  - Financial firms operate with leverage ratios much higher than non-financial corporates, leaving smaller buffers to absorb shocks.
- Policy inconsistency:
  - Capital and solvency regulation increase equity and reserve buffers to limit leverage.
  - Tax deductibility of interest encourages leverage and reduces buffers, creating an inconsistency between tax and regulatory frameworks.
  - Removing debt bias in the tax system could be a more effective corrective than using leverage ratio regulation alone.

### Scope, data, and sample composition
- Covered firm types:
  - Regular commercial, cooperative and savings banks
  - Investment banks
  - Non-bank finance companies
  - Life insurance companies
  - Non-life insurance companies
- Data sources and coverage:
  - BankScope (Bureau van Dijk): about 151,000 observations ≈ 14,000 distinct banks and finance companies in 131 countries, 2001–15.
    - Sample composition: almost 93 percent regular banks; 5 percent non-bank finance companies; 2 percent investment banks.
  - Orbis Insurance Focus (Bureau van Dijk): over 68,000 observations ≈ 7,500 firms from 85 countries, 2005–2015.
    - Sample composition: 30 percent life insurers; 70 percent non-life insurers.
  - Macroeconomic data: IMF World Economic Outlook database.
  - Statutory CIT rates: IMF Fiscal Affair Department database.
- Data cleaning:
  - Winsorized at the 1 percent level for BankScope and Insurance Focus variables (except tax, inflation, GDP growth).
  - Firms not reporting for a minimum of three consecutive years dropped.
  - Countries with less than 20 observations dropped.
  - Bank sample after cleaning: from 153,896 → 151,463 observations.
  - Insurance sample after cleaning: from 68,233 → 67,368 observations.
- Panel dimensions and estimation:
  - T = 15; N > 14,000 (≈7,500 for banks and insurance companies respectively).
  - Main estimator: system GMM (Blundell and Bond (1998)), cluster-robust SEs at the country level.

### Theoretical model — setup and main comparative-statics (preserved expressions)
- Two-period single-bank model:
  - Equity investment E>0; bank chooses external debt D>0; assets W = E + D.
  - Fraction α ∈ (0,1) in risky assets; payoff of risky asset θ−1 with θ distributed on support (0,∞), distribution Φ(θ), density φ(θ), mean 휃̅.
  - Risk-free return and debt interest rate: R>0.
  - Profit tax rate T ∈ (0,1) with interest deductibility.
  - Capital ratio K = equity / risky asset must exceed minimum requirement K̅ ∈ (0,1); penalty for breach C_K>0 per unit shortfall.
- First-order condition with interest deductibility:
  - dΠ(D)/dD = (1−T) α{(휃̅−1)−R} − ∂S_K/∂D = 0.
- Without interest deductibility, first-order condition:
  - (1−T) α{(휃̅−1)−R} − TR = C_K(1+R)E/(D+E) Φ(θ_K).
- Debt Bias B defined as:
  - B = D_ID − D_NID.
- Proposition 1 condition (preserved verbatim):
  - Φ(θ_K) < (1+R) E / [α(D+E)(1−K̅)] φ(θ_K)
  - Under (14), debt bias increases with the tax rate T:
    - ∂B/∂T = R C_K (1+R) E /(D+E)^2 [ (1+R) E / (α(D+E)(1−K̅)) φ(θ_K) − Φ(θ_K) ] > 0.

### Extensions for insurers and lightly regulated non-bank firms
- Insurance companies:
  - Similar tradeoff: premia collection, investment in risky assets, minimum equity buffer K̅ and penalty C_K; technical reserves affect liability composition (technical reserves not explicitly modeled).
- Lightly or non-regulated non-bank financial firms:
  - Failure condition: (1−α) W (1+R) + α θ W − (1+R) D < 0 ⇔ θ < θ_F(D).
  - Survival probability: 1 − Φ(θ_F(D)).
  - Owner’s objective conditional on survival:
    - Π_Shadow(D) = [1−Φ(θ_F(D))] × [ (1−α)(D+E){1+R(1−T)} + TRD − (1+R)D + α(D+E){휃̅ + T(1−휃̅)} ].
  - Trade-off derivatives:
    - ∂[1−Φ(θ_F(D))]/∂D = −φ(θ_F(D)) (1+R)E / [α(D+E)^2] < 0.
    - ∂[conditional payoff]/∂D = (1−T) α{(휃̅−1)−R} > 0.

### Empirical specification and identification
- Dynamic panel reduced-form (equation (23)):
  - Y_it = α + β_1 Y_it−1 + β_2 Tax_it + β_3 X_it + λ_t + μ_i + ε_it.
  - Y_it: leverage ratio (total liabilities / total assets).
  - Tax_it: top statutory corporate income tax (CIT) rate.
  - Controls X_it: bank size (log total assets, linear and quadratic), pre-tax profitability (share of assets), GDP growth, inflation, riskiness (risk-weighted assets / total assets).
  - Year fixed effects λ_t and firm fixed effects μ_i included.
  - Long-term effect = β_2 / (1 − β_1).
- Heterogeneity specifications:
  - Interaction with firm type (Inv_i, Fin_i) and size (Size_it × Tax_it).
  - Quantile regressions (static) to assess heterogeneous responses across leverage distribution.
- Robustness:
  - Winsorize sample at 10th and 90th percentiles (middle 80 percent) and alternative controls.

### Main quantitative empirical findings — regular banks and non-bank financial firms
- Leverage levels (sample averages):
  - Regular banks: Average leverage 87 percent.
  - Investment banks: Average leverage 73 percent.
  - Finance companies: Average leverage 77 percent.
  - Non-financial private sector (Euro area sample): average leverage 17.6 percent.
- Tax variation in banking sample:
  - Mean statutory CIT rate: 31.2 percent; Min 8.5 percent; Max 60.0 percent (Libya early decade).
- Main regression findings for regular banks:
  - OLS and system GMM indicate positive and statistically significant effect of statutory tax rate on leverage over full period and pre-crisis.
  - Short-term coefficient pre-crisis: 0.24 (column (4)); double the effect during the full period (column (3)).
  - Post-crisis tax effect: positive but smaller and not significantly different from 0 in many specifications.
  - Long-run coefficients:
    - Regular banks (pre-crisis): 0.38.
    - Regular banks (entire period and post-crisis): 0.17.
    - When including investment banks and finance companies: 0.37 (pre-crisis) and 0.21 (entire/post-crisis).
- Debt-bias contribution estimates (preserved verbatim):
  - For regular banks:
    - Pre-crisis, debt bias may have increased leverage by 9 percentage points.
    - Average leverage among banks in the sample: 87 percent.
    - Post-crisis, the estimated effect decreases to some 5 percentage points.
  - Investment banks:
    - Investment bank leverage may have been increased by almost 15 percentage points pre-crisis.
    - Pre-crisis short-term coefficient for investment banks: 0.37 (= 0.23 + 0.14).
    - Pre-crisis long-term tax coefficient for investment banks: 0.59.
  - Quantitative implication of eliminating debt bias at a 25 percent CIT rate:
    - reduce bank leverage by 9 percent
    - reduce investment bank leverage by some 15 percentage points
    - These reductions correspond to about:
      - 10 percent of ex-ante average leverage of 87 percent for regular banks
      - 20 percent of ex-ante average leverage of 73 percent for investment banks

### Insurance sector findings
- Insurance leverage (net of technical reserves) and technical reserves:
  - Technical reserves average 65 percent of assets across life insurers.
  - Technical reserves average 40 percent of assets for non-life insurers.
  - Stock of liabilities in excess of technical reserves: 11.5 percent for life insurers and 13.9 percent for non-life insurers.
- Insurance regression results:
  - Life insurers:
    - Long-run coefficient for 2005-15 is 0.28, suggesting debt bias adds some 7 percentage points to leverage.
    - Debt bias strongest pre-crisis; post-crisis short-run effect shrinks but remains significant in some specifications.
  - Non-life insurers:
    - Coefficient estimates range from 0.05 (whole sample) to 0.16 (pre-crisis) and 0.06 (post-crisis).
    - Long-run coefficient estimates range between 0.14-0.24, implying debt bias may have increased leverage by between 3½ to 6 percentage points.
    - Some regression results fail standard robustness tests and should be interpreted with caution.

### Heterogeneity by size and leverage quantiles
- Size effects:
  - Banks: positive relation between size and leverage; bigger banks have higher leverage.
  - Life insurers: combined coefficient on log assets is -1.5, indicating larger life insurers associated with lower leverage.
- Size × tax interaction:
  - Average bank long-term tax coefficient: 0.19.
  - Effect is three times larger for the smallest banks and decreases along size quantiles; for the largest banks the coefficient becomes negative (not significantly different from zero).
  - Insurers: size effects less clear; for life insurers the long-run tax coefficient decreases with size; for non-life insurers it remains constant across size quantiles.
- Quantile results (banks):
  - Strictly decreasing tax coefficient across leverage quantiles:
    - Long-run tax coefficient for the least leveraged banks: 0.63.
    - Decreases monotonically to -0.04 for the most leveraged banks.
  - Interpretation: most leveraged banks are close to regulatory minima and have limited scope for tax-driven leverage adjustments.
- Quantile results (insurers):
  - Coefficients increase with leverage, strongest in life companies and visible for non-life insurers.
  - For insurers, the most leveraged firms are most influenced by tax considerations because these firms typically have the lowest technical reserves.

### Robustness and sensitivity
- Winsorizing middle 80 percent (drop top/bottom 10 percent) results:
  - Tax coefficients remain significant and positive.
  - Tax coefficients become significantly different from zero for the post-crisis period for the middle 80 percent.
  - Standard tests for overidentifying restrictions and serial autocorrelation do not indicate misspecification for winsorized sample.
- Some specifications (notably certain post-crisis and short pre-2008 insurer samples) fail Hansen and AR tests; results flagged accordingly.

### Policy implications and recommendations
- Debt bias is present across banks, investment banks, finance companies, and insurers; most pronounced pre-GFC but still relevant post-crisis at lower magnitudes.
- Potential for medium-term resurgence of debt bias once buffers are rebuilt and interest rates normalize; magnitude uncertain.
- Policy recommendation: neutralize tax incentives between debt and equity to reduce debt bias.
  - Reform options include denying or limiting interest deductibility, or introducing an allowance for corporate equity (discussed in IMF (2016b)).
- Align tax policy with prudential regulation:
  - Reducing debt bias would better align tax and prudential policies on financial sector leverage and buffers, yielding pronounced social benefits given higher externalities in the financial sector.

*Source: _wp16217 - IMF working paper content as provided.*

### References .............................................................................................................

### _wp16217 - References ............................................................................................................................... 25

### Tax bias and debt bias: definitions and distinctions
- Tax systems in most countries allow for a deduction of interest paid on debt, while not featuring a similar deduction for equity, implying debt is artificially cheap relative to equity and violating the principle of neutrality of the source of finance (Sorensen, 2014; Weichenrieder and Klautke, 2008).
- Debt bias: the incentive created by tax deductibility of interest that leads profit-maximizing firms to take on more debt than they would absent that incentive.
- Debt shifting: a distinct tax distortion where, having decided on total leverage, firms allocate debt across jurisdictions (typically more debt in high-tax jurisdictions) to shift profits; debt shifting does not change total debt and is primarily a BEPS concern.

### Financial stability implications of debt bias
- Debt bias increases leverage, raising the probability of distress or bankruptcy for both non-financial and financial firms.
- Financial firms differ from non-financial firms in two salient ways:
  - The systemic importance of the financial system is much larger than that of other sectors, producing negative externalities from financial distress that individual firms do not internalize.
  - Financial firms on average operate with leverage ratios much higher than non-financial corporates, leaving smaller buffers to absorb shocks.
- The combination of higher systemic cost of distress and higher average leverage implies debt bias poses larger macro-financial risks in the financial sector.

### Interaction with regulatory capital and solvency policy
- Capital and solvency regulation aim to increase equity and reserve buffers and hence limit leverage to strengthen resilience and reduce negative externalities.
- Tax deductibility of interest works in the opposite direction by encouraging leverage and reducing buffers, creating a policy inconsistency between tax and regulatory frameworks.
- Removing debt bias in the tax system could be a more effective policy response than using leverage ratio regulation as a corrective distortion.

### Prior evidence and gaps in the literature
- Pre-existing empirical work on tax distortions in financial institutions is limited and focused mainly on banks.
  - Early observers: Slemrod 2009.
  - Comprehensive empirical studies: Langedijk et al 2014 (Europe) and De Mooij and Keen 2016 (banks worldwide); these studies focus on “regular” banks and data up to 2011 and 2009 respectively.
- Less attention has been paid to non-bank financial firms (investment banks, finance companies, life and non-life insurers) despite the non-bank sector being large and growing.

### Scope and contribution of this paper
- Extends analysis of corporate taxation’s effect on leverage to:
  - Investment banks
  - Non-bank finance companies
  - Life insurance companies
  - Non-life insurance companies
  - Regular commercial, cooperative, and savings banks
- Includes a significant part of the non-bank financial sector which, in several countries, represents a multiple of GDP and can be comparable in size with the regular banking sector.
- Uses seven years of post-crisis data to contrast pre-crisis, post-crisis, and full-sample findings.
- Analyzes heterogeneity by firm size and degree of leverage to assess whether large (small) or highly (modestly) leveraged firms behave differently from the average.

### Data note on Figure 1 (as presented in source)
- Figure 1: The Non-Bank Financial System.
- Note: Data excludes insurance and pension funds.
- Advanced Economies (AEs) sample: Canada, Hong Kong, Japan, Korea, Singapore, Australia, Switzerland, Chile, and Mexico.
- Emerging Market Economies (EMEs) sample: Argentina, Brazil, China, India, Indonesia, Russia, Saudi Arabia, South Africa, and Turkey.
- Source for figure: Financial Stability Board.
- Figure panels depict variables in "Percent of GDP" and "Percent of banking system assets" across 2002–2014.

### Main quantitative findings (preserved verbatim)
- For regular banks:
  - Pre-crisis, debt bias may have increased leverage by 9 percentage points.
  - Average leverage among banks in the sample: 87 percent.
  - Post-crisis, the estimated effect decreases to some 5 percentage points.
- Interpretation offered in source:
  - Post-GFC, many financial firms focused on rebuilding and increasing buffers in response to tighter regulation and increased supervisory scrutiny, making tax considerations less central to leverage decisions.
  - The post-GFC ultra-low interest rate environment likely reduced incentives associated with tax deductibility of interest payments.

### Policy implications and interpretation (drawn from source text)
- Debt bias remains a relevant consideration driving leverage, even if sensitivity to taxation has declined post-crisis.
- Aligning tax policy with regulatory objectives (reducing tax bias toward debt) could complement regulatory efforts to raise capital buffers and strengthen financial stability.
- Because leverage-related externalities in the financial sector are larger than for non-financial firms, tax reforms addressing debt bias could yield pronounced social benefits.

*Source: _wp16217 - References, IMF working paper content as provided.*

### Section III describes the estimation procedure and data used to estimate the model. Section IV

### _wp16217 - Section III describes the estimation procedure and data used to estimate the model. Section IV

### II. A model of leverage — setup and key parameters
- Two-period model for a single bank:
  - Period 1: fixed equity investment E>0; bank chooses external debt D>0; total assets W = E + D.
  - Bank holds risky assets fraction α ε (0,1); remainder 1−α in risk-free assets.
  - Risky asset payoff: θ−1 with θ distributed on support (0,∞), twice-differentiable distribution Φ(θ), density φ(θ), mean 휃̅.
  - Risk-free return and debt interest rate: R>0 (same rate).
- Profit taxation and regulation:
  - Profit tax rate T ε (0,1) with interest deductibility (deduction for interest paid on debt).
  - Capital ratio K defined as equity over risky asset must exceed minimum capital requirement 퐾̅ ε (0,1).
  - Penalty for breaching minimum capital ratio: C_K>0 per unit shortfall.
- Non-bank financial firms:
  - When lightly or not regulated, failure can occur if returns wipe out equity; failure threshold θ_F(D) defined and survival probability 1−Φ(θ_F(D)).

### A. Bank leverage — objective, constraints, and optimality conditions
- Bank net worth at end of period (state θ) incorporates asset returns, interest on debt, taxes, and penalties for breaking minimum capital ratios.
- Expected net worth Π(D) equals asset-return terms less expected regulatory penalty S_K(D).
- First-order condition for optimal debt D_ID (with interest deductibility):
  - dΠ(D)/dD = (1−T)α{(휃̅−1)−R} − ∂S_K/∂D = 0.
- Without interest deductibility, first-order condition becomes:
  - (1−T)α{(휃̅−1)−R} − TR = C_K(1+R)E/(D+E) Φ(θ_K).
- Debt bias B defined as:
  - Debt Bias ≡ B = D_ID − D_NID.
- Key model assumptions and expressions (preserved exactly as in source):
  - Support and parameters: θ distributed with distribution function Φ(θ) and density φ(θ) on support (0,∞), mean 휃̅.
  - Capital constraint rewritten leading to minimum return θ_K(D) (equation (3) in source).
  - Expected regulatory penalty S_K(D) given by (7) in source.
  - Derivative of θ_K with respect to D: dθ_K(D)/dD = (1+R)E / [α(D+E)^2] (1−K̅) > 0.
  - Assumption ensuring left-hand side of (12) positive:
    - T < α{(휃̅−1)−R}/((1−α)R + α(휃̅−1)) > 0.

- Main comparative-static result (Proposition 1):
  - Condition (14) (preserved verbatim):
    - Φ(θ_K) < (1+R) E / [α(D+E)(1−K̅)] φ(θ_K)
  - Under (14), debt bias increases with the tax rate T:
    - ∂B/∂T = R C_K (1+R) E /(D+E)^2 [ (1+R) E / (α(D+E)(1−K̅)) φ(θ_K) − Φ(θ_K) ] > 0.
  - Intuition summarized in the model:
    - A higher corporate income tax rate leads to higher debt bias; banks borrow more to exploit interest deductibility, lowering buffers.

### B. Leverage in insurance and non-bank financial firms — extensions
- Insurance companies:
  - Similar tradeoff as banks: collect premia, invest in risky assets, face regulatory equity buffer requirement K̅ and penalty C_K, plus regulation on technical reserves (not explicitly modeled here).
  - Differences: minimum equity buffer 퐾̅ and punishment 퐶퐾 differ; technical reserves affect liability composition.
- Non-bank financial firms (lightly or non-regulated):
  - Tradeoff between higher return on equity via leverage and probability of failure (bankruptcy).
  - Firm fails when:
    - (1−α) W (1+R) + α θ W − (1+R) D < 0, equivalently θ < θ_F(D) where θ_F(D) ≡ (1+R)D − (1−α)W / (α W) = (1+R)/αW − E/αW ≡ θ_F(D).
  - Survival probability into period 2: 1 − Φ(θ_F(D)).
  - Owner maximizes expected payoff conditional on survival:
    - Π_Shadow(D) = [1−Φ(θ_F(D))] × [ (1−α)(D+E){1+R(1−T)} + TRD − (1+R)D + α(D+E){휃̅ + T(1−휃̅)} ].
  - Trade-off derivatives:
    - ∂[1−Φ(θ_F(D))]/∂D = −φ(θ_F(D)) (1+R)E / [α(D+E)^2] < 0.
    - ∂[conditional payoff]/∂D = (1−T) α{(휃̅−1)−R} > 0.

### III. Regression analysis — specification and identification
- Reduced-form dynamic panel specification estimated (equation (23)):
  - Y_it = α + β_1 Y_it−1 + β_2 Tax_it + β_3 X_it + λ_t + μ_i + ε_it.
  - Y_it: leverage ratio (total liabilities / total assets).
  - Tax_it: top statutory corporate income tax (CIT) rate in firm i’s country.
  - X_it: controls including bank size (log total assets, linear and quadratic), pre-tax profitability (share of assets), GDP growth, inflation, riskiness (risk-weighted assets / total assets).
  - Year fixed effects λ_t and firm fixed effects μ_i included.
  - Coefficient of interest β_2 expected positive; long-term effect = β_2 / (1 − β_1).
- Heterogeneity by firm type (equation (24)):
  - Y_it = α + β_1 Y_it−1 + β_2 Tax_it + β_3 X_it + β_4(Inv_i × Tax_it) + β_5(Fin_i × Tax_it) + λ_t + μ_i + ε_it.
  - β_4 and β_5 capture differential tax effects for investment banks (Inv_i) and non-bank financial corporations (Fin_i).
- Size interaction (equation (25)):
  - Y_it = α + β_1 Y_it−1 + β_2 Tax_it + β_3 X_it + β_4(Size_it × Tax_it) + λ_t + μ_i + ε_it.
- Quantile regressions (static model, equation (26)) to assess heterogeneous responses across leverage distribution:
  - Y_it = α + β_1 Tax_it + β_2 X_it + λ_t + ε_it.
  - Used to examine whether taxation has less impact for institutions with lower buffers (more constrained).
- Robustness checks:
  - Winsorize sample at 10th and 90th percentiles (middle 80 percent).
  - Alternative control variables employed.

### III.B Data (briefly)
- Financial institution-level unconsolidated balance sheet and profit & loss data used to align taxation and regulation effects within national borders.
- Estimation method:
  - System GMM dynamic panel estimator (Blundell and Bond (1998)) to correct biases from lagged dependent variable and endogenous risk variable.
  - Panel dimensions: T = 15; N > 14,000 (7,500 for banks and insurance companies respectively).
  - Cluster-robust standard errors at the country level.

*Italic: Source — _wp16217 (excerpt provided).*

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

### _wp16217 - 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 and samples
- Banks and non-bank finance company data:
  - Source: BankScope database by Bureau van Dijk.
  - Contains about 151,000 observations corresponding to almost 14,000 distinct banks and finance companies in 131 countries over the period 2001–15.
  - Sample composition: almost 93 percent of firms are “regular” banks; 5 percent are non-bank finance companies; 2 percent are investment banks.
- Insurance company data:
  - Source: Orbis Insurance Focus database by Bureau van Dijk.
  - Contains over 68,000 observations corresponding to more than 7,500 firms from 85 countries over the period 2005-2015.
  - Sample composition: 30 percent life insurers; 70 percent non-life insurers.
- Macroeconomic and tax data:
  - Macroeconomic data: IMF’s World Economic Outlook database.
  - Statutory CIT rates: IMF’s Fiscal Affair Department database.

### Data cleaning and consistency
- Winsorization:
  - All variables in BankScope and Insurance Focus data have been winsorized at the 1 percent level, except for tax, inflation, and GDP growth rates.
  - Note on winsorization: I.e., all data below the 1st / 10th (above the 99th / 90th) percentile is bunched at the 1st / 10th (99th / 90th) percentile value.
- Minimum reporting requirements and country coverage:
  - Firms not reporting financial results for a minimum of three consecutive years were dropped.
  - Countries with less than 20 observations were dropped.
  - Bank sample: from the original 153,896 observations, drop 2,087 observations corresponding to firms reporting for less than three consecutive years and 346 observations corresponding to countries with less than 20 observations; resulting sample: 151,463 bank observations.
  - Insurance sample: from the original 68,233 observations, drop 676 and 189 observations respectively; resulting sample: 67,368 insurance company observations.

### Summary statistics and capital structure
- Leverage:
  - Average leverage ratio across the regular banks sample is 87 percent.
  - Investment banks: average leverage 73 percent.
  - Finance companies: average leverage 77 percent.
  - By comparison, average leverage in the non-financial private sector (excluding extractive industries) in a sample of firm-level data from a number of Euro area countries was 17.6 percent.
- Insurance companies (leverage measured net of technical reserves):
  - Technical reserves average 65 percent of assets across life insurers.
  - Technical reserves average 40 percent of assets for non-life insurers.
  - The stock of liabilities in excess of technical reserves (where tax considerations act) comprises 11.5 percent for life insurers and 13.9 percent for non-life insurers.
  - Note: Technical reserves are tightly regulated and presumably not influenced by taxation.
- Distributional patterns:
  - Insurance companies: concentration of companies with low leverage.
  - Banks, investment banks, and finance companies: concentration with higher leverage.
  - Asset size distributions: investment banks and finance companies show a two-humped distribution (bunching at moderately low and moderately high size); banks and insurers show unimodal distributions.

### Corporate income tax (CIT) variation and sample composition
- CIT statistics:
  - Mean statutory CIT rate in the banking sample is 31.2 percent.
  - Minimum CIT rate observed: 8.5 percent.
  - Maximum CIT rate observed: 60 percent (due to Libya in the early years of the decade).
  - Mean CIT rate in the insurance sample is 29.4 percent.
  - Over the analyzed period there is a decreasing trend in CIT rates in all regions; little variation in the rate for the United States.
- Sample concentration by country:
  - Samples are unbalanced and dominated by the United States: over 60 percent of the banking observations and over 40 percent of the insurance observations.
  - For banking, the next 20 percent of observations are distributed in descending order across Germany, Russia, and Japan.
  - For insurance companies, the next 20 percent of observations are from Germany, UK, Canada, Spain, and France.
  - Basic patterns are similar when the US is excluded.

### Basic bivariate patterns
- Binned scatter plots (Figure 6) show:
  - Positive correlation between leverage ratio and statutory tax rate for both banks and insurance companies.
  - For banks: clear positive correlation between leverage and firm size (assets).
  - For insurance companies: no clear linear relation between size and leverage.

### Empirical strategy and estimations
- Baseline and extended samples:
  - Baseline estimations use the sample of regular banks.
  - Extended sample includes investment banks and finance companies.
  - Separate sample and analysis for insurance companies.
- Estimation methods:
  - OLS regressions with and without fixed effects.
  - System GMM specifications on full sample and pre- and post-2008 subsamples.
  - Robust standard errors clustered at the country level.

### Main empirical findings — regular banks
- Tax effect on leverage:
  - OLS regressions (columns (1)–(2)) suggest the tax rate has a strongly significant impact on leverage.
  - System GMM (columns (3)–(5)) confirms a positive and, over the full period and the period up to the crisis, statistically significant effect of taxes on leverage.
  - Short-term coefficient estimate for the tax effect in the period before the crisis (column (4)) is 0.24, double the effect during the full period (column (3)).
- Pre- vs post-crisis:
  - Post-crisis, the tax effect remains positive but is smaller and not significantly different from 0 for the post-crisis period.
  - Possible reasons: tighter regulation requiring banks to build up additional capital; market pressure for larger bank buffers; low interest rates reducing tax benefits from debt.
- Long-run effects:
  - Long-run coefficient for regular banks (pre-crisis): 0.38.
  - Long-run coefficient for regular banks (entire period and post-crisis): 0.17.
  - When including investment banks and finance companies, coefficients remain stable at 0.37 (pre-crisis) and 0.21 (entire/post-crisis).

### Investment banks and finance companies — differential behavior
- Investment banks:
  - Over the whole sample period, the CIT rate is not a significant driver of leverage for investment banks (interaction dummy negative and comparable in size to the CIT coefficient).
  - Pre-crisis: investment banks show much higher sensitivity to tax rates than regular banks.
    - Short-term coefficient estimate for investment banks: 0.37 (= 0.23 + 0.14).
    - This is about 60 percent higher than the coefficient estimate for regular banks.
    - Translates into a long-term tax coefficient of 0.59 for investment banks in the pre-crisis period.
  - Post-crisis: tax considerations are not a significant driver for investment banks; regulatory constraints and buffer rebuilding dominate.
  - Statistical caveats: some statistical tests (Hansen and autocorrelation) fail in certain post-crisis specifications.
- Finance companies:
  - Exhibit tax sensitivity similar to regular banks in both pre- and post-crisis periods.
  - Interaction dummy for finance companies is not significantly different from zero in both periods.
  - Pre-crisis net effect (CIT coefficient plus interaction) is significant; long-term coefficient of 0.28.
  - Post-crisis: no significant tax effect.
  - Formal F tests confirm that for the full period and post-crisis periods the hypothesis of a zero net effect of tax on leverage for finance companies cannot be rejected.

### Quantitative implications and policy-relevant magnitudes
- Debt bias contribution to leverage:
  - Results imply that debt bias may have increased pre-crisis bank leverage by as much as 9 percentage points.
  - Investment banks’ leverage may have been increased by almost 15 percentage points pre-crisis.
- Interpretation:
  - Pre-crisis tax incentives appear to have materially increased leverage, especially for investment banks.
  - Post-crisis regulatory changes and low interest rates attenuated tax-driven leverage adjustments across financial institutions.

*Source: _wp16217 - 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.*

### 0.37 for regular banks and 0.59 for

### _wp16217 - 0.37 for regular banks and 0.59 for

### Evidence on tax-induced debt bias and leverage
- Estimated tax coefficients imply that eliminating the debt bias in a country with a 25 percent corporate income tax rate would:
  - reduce bank leverage by 9 percent
  - reduce investment bank leverage by some 15 percentage points
- These reductions correspond to about:
  - 10 percent of ex-ante average leverage of 87 percent for regular banks
  - 20 percent of ex-ante average leverage of 73 percent for investment banks
- Results are consistent with similar analyses in the literature (e.g., a synthetic control experiment for Belgium in IMF (2016b)).

### Robustness and sensitivity to outliers
- Winsorizing the highest and lowest 10 percent of observations yields:
  - tax coefficients that remain significant and positive
  - tax coefficients that are now significantly different from zero for the post-crisis period
  - standard tests for overidentifying restrictions and serial autocorrelation that do not indicate misspecification
- Conclusion: For the middle 80 percent of banks, tax bias remained significantly relevant post crisis despite increased regulatory focus on capital.

### Sector-specific findings
- Investment banks: leverage decisions sensitive to the CIT rate pre-crisis, but not for the post-crisis period.
- Finance companies: behave much like regular banks in sensitivity to tax bias.
- Life insurers:
  - Long-run coefficient for 2005-15 is 0.28, suggesting debt bias adds some 7 percentage points to leverage.
  - Debt bias strongest pre-crisis; post-crisis short-run effect shrinks but remains significant.
  - Note: some specifications fail AR(1) and AR(2) autocorrelation tests due to short pre-2008 sample.
- Non-life insurers:
  - Coefficient estimates range from 0.05 (whole sample) to 0.16 (pre-crisis) and 0.06 (post-crisis).
  - Long-run coefficient estimates range between 0.14-0.24, implying debt bias may have increased leverage by between 3½ to 6 percentage points.
  - Regression results do not meet standard robustness tests and should be interpreted with caution.

### Firm size and heterogeneity in tax responsiveness
- Life insurers: combined coefficient on log assets is -1.5, indicating larger life insurers are associated with lower leverage.
- Banks (regular, investment, finance companies): relation between size and leverage is strongly positive—bigger banks have higher leverage and hence lower buffers.
- Large banks are less responsive to taxation:
  - Interaction of size with the tax rate shows the long-term tax coefficient for the average bank is 0.19.
  - The long-term effect is three times larger for the smallest banks and decreases along size quantiles; for the largest banks the coefficient would become negative (not significantly different from zero).
  - Explanation: largest banks hold lower buffers in excess of regulatory minima, leaving less room for tax considerations to influence leverage.
- Insurers: size effects are less clear; insignificant size effects formally confirm lack of clear relation between size and leverage in scatter plots.
  - For life insurers the long-run tax coefficient decreases with size; for non-life insurers it remains constant across size quantiles.

### Leverage level and tax responsiveness (quantile analysis)
- Banks:
  - Strictly decreasing relationship between tax coefficient estimate and leverage across quantiles.
  - Long-run tax coefficient for the least leveraged banks: 0.63.
  - Tax coefficient decreases monotonically to -0.04 for the most leveraged banks.
  - Interpretation: most leveraged banks are close to regulatory minima and have limited scope for tax-driven leverage adjustments.
- Insurers:
  - Quantile regressions show coefficient estimates increasing with leverage, strongest in life companies and visible for non-life insurers.
  - For insurers, the most leveraged firms are most influenced by tax considerations because these firms typically have the lowest technical reserves and are less bound by regulatory constraints.

### Macro and policy implications
- Debt bias is evident for banks, investment banks, finance companies, and insurance companies.
- Bias is most prevalent pre-GFC but remains present post-crisis, typically at lower magnitudes.
- Debt bias may become more prevalent over the medium-term once buffers are rebuilt and interest rates normalize; whether it will exceed pre-crisis levels is an empirical question.
- Policy recommendation: Tax policy should aim to neutralize incentives between debt and equity finance to reduce debt bias.
  - Reform options include denying or limiting interest deductibility, or introducing an allowance for corporate equity (both discussed in IMF (2016b)).
- Aligning tax policy with prudential regulation:
  - Current regulation focuses on increasing buffers and limiting leverage, while the tax system incentivizes leverage.
  - Reducing debt bias would better align tax and prudential policies on financial sector leverage and buffers, even if perfect coordination is elusive.

*Source: IMF Staff estimates presented in Tables 5, 7 and 8*

### REFERENCES

### _wp16217 - REFERENCES

### Key bibliographic sources cited
- Acharya, V., R. Engle, and M. Richardson. 2012. “Capital Shortfall: A New Approach to Ranking and Regulating Systemic Risks.” American Economic Review 102 (3): pp. 59–64.
- Admati, A., and M. Hellwig. 2014. The Bankers’ New Clothes: What’s Wrong with Banking and What to Do about It. Princeton University Press.
- Auerbach, Alan J., 2002, “Taxation and Corporate Financial Policy”, in Handbook of Public Economics, Auerbach and Feldstein (eds.), Vol 3, pp. 1251-92, Elsevier, Amsterdam, Netherlands.
- Bluedorn, John, and C. Ebeke, 2016, “Investment, Firm Size, and the Corporate Debt Burden: A Firm-Level Analysis of the Euro Area,” in Euro Area Policies Selected Issues Papers, IMF Country Report 16/220, International Monetary Fund: Washington, DC, USA.
- Claessens, S. 2014. “An Overview of Macroprudential Policy Tools.” Annual Review of Financial Economics 7 (1): pp. 397-422.
- Dagher, Jihad, Giovanni Dell’Ariccia, Luc Laeven, Lev Ratnovski, and Hui Tong, 2016, “Benefits and Costs of Bank Capital,” Staff discussion Note 16/04, International Monetary Fund: Washington, DC, USA
- De Mooij, Ruud A., 2011, “The Tax Elasticity of Corporate Debt: A Synthesis of Size and Variations”, Working Paper 11/95, International Monetary Fund: Washington, DC, USA.
- De Mooij, Ruud A, and Michael Keen, 2016, “Debt, Taxes, and Banks,” Journal of Money, Credit, and Banking, vol 48, pp. 5 – 33.
- Diamond, D. W., and R. G. Rajan. 2000. “A Theory of Bank Capital.” Journal of Finance 55 (6): 2431–65.
- Feld, Lars P., J.H. Heckemeyer, and M. Overesch, 2013, “Capital Structure Choice and Company Taxation: A Meta-study”, Journal of Banking & Finance, Vol. 37, pp. 2850-66.
- Graham, John R., 2006, “A Review of Taxes and Corporate Finance”, Foundations and Trends in Finance, Vol 1. pp 587-622.
- International Monetary Fund, 2014, “Shadow Banking Around the Globe: How Large and How Risky?” Global Financial Stability Report, October 2014, IMF: Washington, DC, USA
- International Monetary Fund, 2016a, “Debt – Use It Wisely,” Fiscal Monitor, October 2016, IMF: Washington, DC, USA
- International Monetary Fund, 2016b, “Tax Policy, Leverage and Macroeconomic Stability,” IMF Policy Paper 16/151, International Monetary Fund: Washington, DC, USA
- Langedijk, Sven, G. Nicodème, A. Pagano and A. Rossi, 2014, “Debt Bias in Corporate Taxation and the Costs of Banking Crises in the EU”, Working Paper N. 50-2014, European Commission: Brussels, Belgium
- Modigliani, Franco, and M. Miller, 1958, "The Cost of Capital, Corporation Finance and the Theory of Investment", American Economic Review. 48 (3): pp. 261–297.
- Modigliani, France and M. Miller, 1963, "Corporate income taxes and the cost of capital: a correction". American Economic Review. 53 (3): pp. 433–443.
- Slemrod, Joel, 2009, “Lessons for Tax policy in the Great Recessions,” National Tax Journal, 62, pp. 387-97.
- Sorensen, Peter B., 2014, “Taxation and the Optimal Constraint on Corporate Debt Finance,” Working Paper no. 5101, CESifo: Munich, Germany.
- Thimann, Christian, 2014, “How Insurers Differ from Banks: A Primer on Systemic Regulation,” SRC Special Paper No. 3, London School of Economics: London, UK.
- Weichenrieder, Alfons J., and T. Klautke, 2008, “Taxes and the Efficiency Costs of Capital Distortions,” Working Paper No. 2431, CESifo: Munich, Germany.

### Summary Statistics — Banking (selected exact figures from Table 1)
- Sample counts and coverage:
  - All banks [13,990 bank IDs; 131 countries]
  - Regular banks (commercial, cooperative and savings banks) [13,047 bank IDs; 131 countries]
  - Investment banks [308 bank IDs; 67 countries]
  - Finance companies [635 bank IDs; 85 countries]
- Aggregated sample-level descriptive statistics (multiple rows provided in table; selected figures):
  - Leverage, pct total as sets (winsorized fraction .01): Obs 151,463; Mean 86.51; Std. Dev 13.21; Min 10.99; Max 98.5
  - CIT rate (pct): Obs 151,174; Mean 31.2; Std. Dev 6.5; Min 8.5; Max 60.0
  - Log assets (winsorized fraction .01): Obs 151,463; Mean 5.2; Std. Dev 1.7; Min 1.8; Max 9.9
  - GDP growth (pct): Obs 151,449; Mean 1.9; Std. Dev 2.6; Min -62.1; Max 104.5
  - Total assets, million USD (winsorized fraction .01): Obs 151,463; Mean 1,119; Std. Dev 3,194; Min 620; Max 753
  - Total equity, million USD (winsorized fraction .01): Obs 151,463; Mean 932; Std. Dev 701; Min 1; Max 1,971

- Subsamples also reported (counts and summary stats repeated for different groupings and aggregation levels), including:
  - 144,552; 2,477; 4,434 observations rows shown for varied groupings with corresponding means and ranges for leverage, CIT rate, log assets, GDP growth, CPI inflation, pre-tax profit, risk, total assets, total equity, pretax profits.

### Summary Statistics — Insurance (selected exact figures from Table 2)
- Life and non-life insurance sample sizes:
  - Life insurance companies [2,294 firm IDs; 80 countries]
  - Non-life insurance companies [5,348 firm IDs; 85 countries]
- Key statistics (panel-level and pooled):
  - Leverage, pct total assets (winsorized fraction .01): Obs 19,469; Mean 11.5; Std. Dev 16.1; Min 0.0; Max 73.0
  - CIT rate (pct): Obs 19,419; Mean 28.3; Std. Dev 7.5; Min 8.5; Max 39.0
  - Log assets (winsorized fraction .01): Obs 19,469; Mean 6.1; Std. Dev 2.7; Min -0.4; Max 11.0
  - Total assets, million USD (winsorized fraction .01): Obs 19,469; Mean 5,654; Std. Dev 13,240; Min 1; Max 62,574
  - Total equity, million USD (winsorized fraction .01): Obs 19,469; Mean 406; Std. Dev 1,080; Min 0; Max 6,633

- Larger pooled counts appear (e.g., Obs 47,899) with associated means: Leverage Mean 13.9; CIT rate Mean 29.9; Log assets Mean 4.7; Total assets Mean 878; Total equity Mean 245; Pretax profits Mean 30.

### Country distribution (Table 3) — top contributors by frequency (exact figures)
- Financial companies (excluding insurance) — top countries by Frequency, Percent, Cumulative:
  - 1 United States: 95,791; 63.24; 63.2
  - 2 Germany: 13,946; 9.21; 72.5
  - 3 Russian Federation: 7,515; 4.96; 77.4
  - 4 Japan: 4,657; 3.07; 80.5
  - 5 Italy: 3,594; 2.37; 82.9
  - Other countries: 6,363; 4.21; 100.0
- Insurance companies — top countries by Frequency, Percent, Cumulative:
  - 1 United States: 31,162; 46.26; 46.3
  - 2 Germany: 4,651; 6.9; 53.2
  - 3 United Kingdom: 2,707; 4.02; 57.2
  - 4 Canada: 1,843; 2.74; 59.9
  - 5 Spain: 1,664; 2.47; 62.4
  - Other countries: 3,635; 5.36; 100.0

### Regression results — tax rate and bank leverage (selected coefficients and diagnostics)
- Table 4: Tax Rate and Bank Leverage: Regular Banks (Data sample: Regular banks, unconsolidated accounts. 2001–15.)
  - Estimation methods: Columns (1)-(2) OLS; Columns (3)-(5) two-step system GMM with lagged dependent variable and risk instrumented. Lag limits (2,1).
  - Lagged leverage coefficient examples:
    - Column (1) noFE: 0.81*** [0.01]
    - Column (2) FE: 0.37*** [0.01]
    - Column (3) full period: 0.34*** [0.02]
    - Column (5) 2008 on: 0.69*** [0.08]
  - CIT rate (pct) coefficients:
    - Column (1): 0.04*** [0.01]
    - Column (2): 0.06* [0.03]
    - Column (3): 0.11** [0.05]
    - Column (4) before 2008: 0.24*** [0.07]
    - Column (5) 2008 on: 0.05 [0.04]
  - Other reported coefficients (all with standard errors in brackets) include Log assets, Log assets sq, Pre-tax profit (pct total assets), GDP growth, CPI inflation, Risk (winsorized fraction .01).
  - Observations:
    - Column (1): 131,118
    - Column (3): 91,950
    - Column (4): 38,372
    - Column (5): 47,104
  - Number of id: ranges shown (e.g., 13,024; 9,970; 7,405; 8,563)
  - Long-term effect examples:
    - Column (1) OLS: 0.23*** [0.05]
    - Column (2) FE: 0.09* [0.05]
    - Column (3) full period: 0.17** [0.07]
    - Column (4) before 2008: 0.38*** [0.11]
    - Column (5) 2008 on: 0.17 [0.12]

- Table 5: Tax Rate and Bank Leverage: Regular Banks, Investment Banks and Finance Companies (System GMM)
  - Lagged leverage coefficients (examples):
    - Column (1) full period: 0.34*** [0.02]
    - Column (3) 2008 on: 0.65*** [0.09]
  - CIT rate (pct):
    - Column (1): 0.12** [0.05]
    - Column (2) before 2008: 0.23*** [0.06]
    - Column (3) 2008 on: 0.07 [0.05]
  - Interaction terms:
    - Interaction (CIT * Investment bank): -0.17** [0.08] (in one specification)
    - Interaction (CIT * Finance companies): -0.06* [0.04] (in one specification)
  - Observations: 92,818 total; subsamples 38,449 and 47,832
  - Long-term effect examples:
    - Full period: 0.19** [0.07]
    - Before 2008: 0.36*** [0.09]
    - 2008 on: 0.21* [0.11]
    - Long-term effect: Investment bank 0.59*** [0.11] (in one cell)

- Table 6: Alternative treatment of outliers (Sample of Regular and Investment Banks and Finance Companies)
  - Lagged leverage coefficients:
    - Column (1) full period: 0.80*** [0.02]
    - Column (3) 2008 on: 0.77*** [0.05]
  - CIT rate (pct): 0.03*** [0.00] (Column (1)); 0.06*** [0.01] (Column (2)); 0.04*** [0.01] (Column (3))
  - Risk coefficient shows sign differences across subsamples (e.g., -2.12*** [0.47]; 0.74*** [0.22]; -4.72*** [0.66])
  - Long-term effect: 0.13*** [0.03] (full period)

- Table 7: Life Insurance Companies (2005-2015)
  - Lagged leverage examples:
    - Static OLS: 0.93*** [0.02]
    - Full period (GMM): 0.71*** [0.04]
    - Before 2008: 0.41*** [0.15]
  - CIT rate (pct) examples:
    - OLS: 0.39*** [0.13]
    - Full period: 0.03* [0.02]
    - 2008 on: 0.08*** [0.03]
  - Log assets coefficients vary (negative signs reported in several specifications).
  - Observations: 17,890 (some columns 17,015; panel sizes reported)
  - Long-term effect examples: 0.4 [0.25]; 0.28***; 0.43***; 0.48*** (standard errors shown in table)

- Table 8: Non-Life Insurance Companies (2005-2015)
  - Lagged leverage examples:
    - OLS: 0.86*** [0.01]
    - Full period (GMM): 0.64*** [0.03]
    - Before 2008: 0.34*** [0.12]
  - CIT rate (pct):
    - OLS: 0.11* [0.06]
    - 2008 on: 0.05** [0.03]
    - 2008 on in another column: 0.16** [0.08]
  - GDP growth and Pre-tax profit (pct total assets) are reported with signs and significance (e.g., GDP growth 0.27*** [0.08] in OLS).
  - Observations: 43,887; subsamples 42,243; 5,307; 32,502
  - Long-term effect examples: 0.14; 0.14**; 0.24**; 0.15* (with standard errors)

### Leverage and size; quantile results (selected exact coefficients)
- Table 9: Leverage and Size (system GMM; combined samples)
  - Lagged leverage examples:
    - Columns with basic specification: 0.34*** [0.02]; 0.71*** [0.04]; 0.64*** [0.03]
  - CIT rate (pct): examples 0.12** [0.05]; 0.08*** [0.03]; 0.05** [0.03]
  - CIT * log assets interaction: -0.09** [0.04] (one specification)
  - Quantiles of asset size — estimated tax coefficients at quantiles:
    - Banks: Q10 Log assets 3.32; Est. tax coef 0.57**
    - Banks: Q25 Log assets 4.04; Est. tax coef 0.47**
    - Banks: Q50 Log assets 4.81; Est. tax coef 0.37**
    - Banks: Q75 Log assets 5.69; Est. tax coef 0.25**
    - Banks: Q90 Log assets 7.90; Est. tax coef -0.05
    - Non-Life insurance and Life insurance quantiles also displayed with exact log assets and estimated tax coefficients.

- Table 10: Bank Leverage — Quantile Regressions (2001-2015; OLS and quantiles Q(0.05) to Q(0.95))
  - CIT rate (pct) coefficients across quantiles:
    - OLS: 0.18*** [0.01]
    - Q(0.05): 0.63*** [0.08]
    - Q(0.10): 0.41*** [0.02]
    - Q(0.25): 0.13*** [0.01]
    - Q(0.50): 0.02*** [0.01]
    - Q(0.75): -0.02*** [0.00]
    - Q(0.90): -0.03*** [0.00]
    - Q(0.95): -0.04*** [0.01]
  - Log assets and log assets sq coefficients show large positive effects at lower quantiles and declining (even negative) at upper quantiles (exact coefficients reported in table).
  - Profitability (winsorized fraction .01) is consistently negative and significant across OLS and quantiles (e.g., OLS: -2.21*** [0.06]).

- Table 11: Life Insurers Leverage — Quantile Regressions (2005-2015)
  - Table header present; specific coefficient cells not reproduced in the supplied excerpt.

- Table 12: Non-Life Insurers Leverage — Quantile Regressions (2005-2015)
  - CIT rate (pct) across quantiles:
    - OLS: 0.40*** [0.02]
    - Q(0.05): 0.02*** [0.00]
    - Q(0.10): 0.02*** [0.00]
    - Q(0.25): 0.03*** [0.00]
    - Q(0.50): 0.08*** [0.01]
    - Q(0.75): 0.39*** [0.02]
    - Q(0.90): 0.93*** [0.04]
    - Q(0.95): 1.33*** [0.07]
  - Log assets and log assets sq show varying signs across quantiles (exact coefficients provided in the table).
  - GDP growth (pct) and Inflation (pct) show positive and significant coefficients across many quantiles (exact values shown in table).
  - Observations: 17,890 for life/non-life tables; R-squared reported (e.g., 0.09 for one table).

*Source: _wp16217 - REFERENCES (selected references, summary statistics tables, and regression tables as provided in the source PDF).*

---


_Source: https://www.imf.org/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2016/_wp16217.pdf_