## wpiea2020283-print-pdf - 0.1517 and the effect of a 10 percent or more appreciation is an increase in leverage of 0.0265.

## Source details

**Canonical URL:** [wpiea2020283-print-pdf - 0.1517 and the effect of a 10 percent or more appreciation is an increase in leverage of 0.0265.](https://www.imf.org/-/media/files/publications/wp/2020/english/wpiea2020283-print-pdf.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/wp/2020/english/wpiea2020283-print-pdf.pdf.md)
- [Structured JSON version](/-/media/files/publications/wp/2020/english/wpiea2020283-print-pdf.pdf.json)

---

### Data and sample
- Sample period: 2002 to 2015.
- Economies covered (firm-level accounting data): China, Chinese Taipei, Hong Kong SAR, India, Indonesia, Korea, Malaysia, the Philippines, Singapore, Thailand.
- Unit of observation: firm-year.
- Final combined cleaned sample size (ORBIS + exchange rates): 1,661,677 firm-year observations.
- Sample used in regressions (after winsorization and cleaning): 1,373,080 observations.
- Variables and transformations:
  - Dependent variable: Leverage = book value of total financial debt / book value of total assets.
  - Collateral = tangible fixed assets / total assets.
  - Profitability = EBIT / total assets.
  - Size = log(total assets).
  - Sales growth = growth in sales (winsorized at 5 percent).
  - Exchange rate change dummies: Dummy∆e k equals 1 when nominal bilateral exchange rate vs USD changes by more than k where k ∈ {0, 5, 10} percent (sign indicates appreciation/depreciation).
  - FX debt measures: country-level share of FX debt = non-financial sector FX debt / total credit to the non-financial sector (from BIS GLI and Total Credit databases). Also decomposed into FX loans and FX bonds; government FX bonds subtracted in some specifications.
- Winsorization: All variables at 1 percent (sales growth at 5 percent).
- Estimation methods: OLS with standard errors clustered at the firm level; dynamic system GMM also used with similar results.
- Fixed effects and controls: Firm fixed effects, country fixed effects, granular industry (4-digit) × year fixed effects, and lagged controls (size, collateral, profitability, sales growth).

### Key empirical findings — FX exposure, exchange rate movements, and firm leverage
- Aggregate FX debt interaction (mean FX debt ratio 0.1519):
  - A 10 percent or more depreciation decreases firm leverage by 0.0844 (calculation reported in text based on benchmark results).
  - Effects grow with the size of depreciations and appreciations; depreciations generally have larger effects than appreciations.
- FX debt excluding government FX bonds (mean ratio 0.1191):
  - A 10 percent or more depreciation decreases firm leverage by 0.1352.
  - A 10 percent or more appreciation increases firm leverage by 0.0108.
- FX loans drive the results (Table 7; mean FX loans ratio 0.0996):
  - Using mean FX loans share 0.0996:
    - A 10 percent or more depreciation produces a decline in leverage of 0.1517.
    - A 10 percent or more appreciation produces an increase in leverage of 0.0265.
  - Context: mean firm leverage is 0.16, so:
    - A 10 percent or more depreciation (decline of 0.1517) is a change nearly equivalent to the mean leverage (decline ≈ mean).
    - A 10 percent or more appreciation (increase of 0.0265) is a 20 percent increase relative to the mean leverage (0.0265 is 20 percent of 0.16).
- Directional patterns:
  - In countries with higher corporate-sector FX debt shares, currency depreciations are associated with reductions in firm leverage; appreciations are associated with increases in firm leverage.
  - In countries with no FX debt, the (small) effects are opposite in sign: firm leverage increases with depreciations and decreases with appreciations.
- Robustness and additional checks:
  - Results robust to subtracting government FX bonds from non-financial sector FX debt; separating FX loans and FX bonds (FX loans drive main results); using continuous exchange rate changes; using firm-level FX debt proxies and limited direct firm-level FX debt data for 4 countries.
  - Placebo tests: movements less than 1 percent in exchange rates show no effect on firm leverage.
  - Additional sensitivity checks: controlling for firms in non-tradeable sectors; separating long-term vs short-term debt; use of country×time fixed effects via a proxy for firm-level FX debt share.
- Magnitudes and statistical significance (selected coefficients):
  - Table 5 (aggregate FX debt): FXdebt × Dummy∆e k (k≤−10%): -0.5312*** (t-stat -19.0); FXdebt × Dummy∆e k (k≥+10%): -0.2620*** (t-stat -10.8).
  - Table 6 (FX debt excluding government bonds): FXdebt × Dummy∆e k (k≥+10%): -1.1352*** (t-stat -19.1); FXdebt × Dummy∆e k (k≤−10%): 0.0908*** (t-stat 2.8).
  - Table 7 (FX loans): FXloan × Dummy∆e k (k≥+10%): -1.5236*** (t-stat -23.7); FXloan × Dummy∆e k (k≤−10%): 0.2665*** (t-stat 7.4).
  - FX debt level coefficients: FXdebt c,t−1 ≈ 2.38 to 2.99 (highly significant); FXloan c,t−1 ≈ 3.04 to 3.23 (highly significant).
  - Control variable coefficients:
    - Profitability ≈ -0.0444 to -0.0441 (highly significant).
    - Collateral ≈ 0.0650 to 0.0695 (highly significant).
    - Size ≈ 0.0200 to 0.0219 (highly significant).
    - Sales growth ≈ -0.0009 to -0.0010 (significant).
- Interpretation: FX loan exposures substantially amplify the sensitivity of firm leverage to exchange rate fluctuations; large depreciations materially reduce leverage and large appreciations increase leverage, conditional on the country’s FX loan share.

### Identification concerns and strategies
- Threats: omitted time-varying country-level factors correlated with exchange rate fluctuations interacted with corporate FX debt shares (e.g., terms-of-trade shocks, government borrowing responses).
- Strategies used:
  - Granular industry-year fixed effects (4-digit) to capture industry-specific time-varying shocks.
  - Country×time fixed effects by proxying firm-level FX debt (assume firm FX debt share = corporate sector FX debt share) to absorb country-time heterogeneity.
  - Use of actual firm-level FX debt shares where available (4 countries) to validate proxy results.
  - Exclusion of multinational branches and firms with headquarters outside the reporting country.
  - Lagging FX debt shares and control variables by one year.

### 3.2 Firm-Level FX Debt: Non-Financial Sector Proxy Approach — methodology and main results
- Proxy approach: assume each firm’s share of FX debt equals the aggregate non-financial sector share (after excluding government FX debt); apply this share to each firm’s total debt to obtain firm-level FX debt. FXdebt_i is a time-invariant dummy equal to 1 when the average value of FX debt of a firm is higher than the respective median firm in the same country.
- Direct approach: use actual firm-level FX debt data for Korea, Argentina, Mexico, Colombia for validation.
- Micro specification:
  Leverage_i,j,c,t = β · FXdebt_i × DummyΔe^k_c,t + θ1 · X_i,c,t−1 + θ2 · X_i,c,t−1 × DummyΔe^k_c,t + α_i + γ_c,t + φ_j,t + ε_i,j,c,t
  - α_i = firm fixed effects; γ_c,t = country-year fixed effects; φ_j,t = industry-year fixed effects.
- Estimation: OLS with robust standard errors clustered at the firm level; dynamic system GMM similar; results robust to clustering at the country level.
- Continuous-exchange-rate results (Table 8):
  - FXdebt_c,t−1 × ∆e_c,t: -0.2809*** (t = -3.8)
  - FXdebt_c,t−1: 2.4462*** (t = 68.8)
  - ∆e_c,t: 0.1237*** (t = 30.9)
  - Profitability_i,j,c,t−1: -0.0440*** (t = -23.6)
  - Collateral_i,j,c,t−1: 0.0654*** (t = 34.9)
  - Size_i,j,c,t−1: 0.0196*** (t = 27.3)
  - Sales growth_i,j,c,t−1: -0.0011*** (t = -3.9)
  - Observations: 1,373,080; R^2: 0.7813.
- Proxy FXdebt_i × continuous ∆e_c,t (Table 10):
  - FXdebt_i × ∆e_c,t: -0.0985*** (t = -21.2)
  - Profitability_i,j,c,t−1: -0.0456*** (t = -24.4)
  - Profitability_i,j,c,t−1 × ∆e_c,t: -0.0691*** (t = -3.3)
  - Collateral_i,j,c,t−1: 0.0703*** (t = 37.9)
  - Collateral_i,j,c,t−1 × ∆e_c,t: 0.0260*** (t = 3.1)
  - Size_i,j,c,t−1: 0.0207*** (t = 28.8)
  - Size_i,j,c,t−1 × ∆e_c,t: 0.0173*** (t = 10.4)
  - Sales growth_i,j,c,t−1: -0.0006** (t = -2.0)
  - Observations: 1,372,970; R^2: 0.79.
- Difference-in-Differences interpretation: interaction of firm-level FX debt dummy with appreciation/depreciation dummy; inclusion of firm, industry-year, country-year fixed effects; results smaller once country×year FE included.
- Robustness, placebo, heterogeneity:
  - Table 9 (proxy FXdebt_i; k≥+10%): FXdebt_i × DummyΔe^k_c,t = -0.0058*** (t = -3.7); (k≤−10%): FXdebt_i × DummyΔe^k_c,t = 0.0550*** (t = 26.8). Observations: 1,372,970; R^2: 0.79.
  - Placebo (Table 11): FXdebt_i × DummyΔe^k_c,t for 1%≥k>0%: -0.0629 (t = -1.1) [not significant]; for −1%≤k<−0%: -0.0049* (t = -8.9) with event-sample notes; overall: "Clearly this is insignificant."
  - Tradeable vs non-tradeable (Table 12):
    - Depreciations (k≤−10%): Tradeable FXdebt_i × Dummy = 0.0341*** (t = 10.8); Non-tradeable = 0.0643*** (t = 22.4).
    - Appreciations (k≥+10%): Tradeable = -0.0037 (t = -1.5); Non-tradeable = -0.0048** (t = -2.3).
  - Debt maturity (Table 13):
    - Depreciations (k≤−10%): LT: 0.0501*** (t = 27.4); ST: 0.0052*** (t = 5.9).
    - Appreciations (k≥+10%): LT: -0.0099*** (t = -7.1); ST: 0.0038*** (t = 3.2).
- Firm-level actual FX debt shares (Korea and others):
  - Korea KIS regression (Table 14; Observations: 17,033; R^2: 0.83):
    - FXdebt share_i × DummyΔe^k_c,t:
      - k≥+0%: -0.013*** (t = -4.4)
      - k≥+5%: -0.013*** (t = -4.4)
      - k≥+10%: -0.020*** (t = -7.0)
      - k≤−0%: 0.013*** (t = 4.4)
      - k≤−5%: 0.015*** (t = 3.8)
      - k≤−10%: 0.039*** (t = 12.3)
    - Continuous: FXdebt share_i × ∆e_c,t = -0.038*** (t = -9.3).
  - Additional country evidence (Argentina, Mexico, Brazil; 1990–2005): for large depreciations (90th percentile ≈ 24% depreciation) results consistent; for large appreciations (1st percentile ≈ 16% appreciation) only exporters (≥10 percent of revenues from exports) significantly increase leverage with appreciations.
- Key takeaways from proxy approach:
  - Proxy yields consistent within-sample dynamics and is validated by select firm-level data.
  - Large exchange rate movements (e.g., ≥10% or ≤−10%) drive significant differential leverage responses; minimal changes (≤1–2%) do not.
  - Country×year fixed effects reduce coefficient sizes, indicating other country-year shocks and policies explain part of the relation.
  - Heterogeneity: stronger effects in non-tradeable sector; maturity matters (LT vs ST behavior).

### 3.5 The Role of Exchange Rate Policies — FX intervention (FXI) and firm leverage
- FXI proxied by FX reserves-to-GDP ratio; analysis of FXI interactions with firm FX debt and exchange rate movements.
- Firm-level evidence (Table 15: Argentina, Mexico, Brazil; Observations: 9,393 All firms; 4,043 High exporters; R2: 0.77 All firms; 0.75 High exporters):
  - FXdebt_i × Dummy∆e_k_c,t (k≥90th percentile, All firms): -0.017** (0.01).
  - FXdebt_i × Dummy∆e_k_c,t (k≤1st percentile, All firms): 0.020 (0.02).
  - FXdebt_i × Dummy∆e_k_c,t (k≥90th percentile, High exporters): -0.031** (0.01).
  - FXdebt_i × Dummy∆e_k_c,t (k≤1st percentile, High exporters): 0.060** (0.03).
- Benchmark Results with FXI (Table 16; sample observations: 1,373,080 across columns; R2: 0.78):
  - FXI_c,t−1 coefficients:
    - k≥+0%: -0.0036*** (-37.0)
    - k≥+5%: -0.0033*** (-34.9)
    - k≥+10%: -0.0030*** (-30.4)
    - k≤−0%: -0.0036*** (-37.2)
    - k≤−5%: -0.0038*** (-38.3)
    - k≤−10%: -0.0037*** (-40.5)
    - ∆e_c,t specification: -0.0032*** (-30.2)
  - FXdebt_c,t−1 × Dummy∆e_k_c,t:
    - Col (1): -0.0338*** (-4.7)
    - Col (2): -0.0659*** (-7.8)
    - Col (3): -0.4587*** (-17.7)
    - Col (4): 0.0340*** (4.7)
    - Col (5): 0.0495*** (6.8)
    - Col (6): -0.2856*** (-11.9)
    - Col (7) (×∆e_c,t): -0.1663** (-2.4)
  - FXdebt_c,t−1 level coefficients: ≈ 1.9581*** to 2.1096*** across columns.
  - Dummy∆e_k_c,t coefficients vary by threshold; ∆e_c,t in col (7): 0.0683*** (17.1).
  - Selected controls same as benchmark: Profitability_i,j,c,t−1: -0.0439*** (-23.5); Collateral_i,j,c,t−1: 0.0667*** (35.6); Size_i,j,c,t−1: 0.0196*** (27.3); Sales growth_i,j,c,t−1: -0.0011*** (-3.8).
- Robustness (Table 17):
  - Excluding government FX bonds and focusing on FX Loans yield similar patterns; FXloan-based interactions often larger in magnitude.
  - Examples:
    - Excluding government FX bonds, FXdebt_c,t−1 × Dummy∆e_k_c,t Col (3): -0.9817*** (-17.6).
    - FXloan_c,t−1 × Dummy∆e_k_c,t Col (3): -1.3152*** (-20.9).
    - FXloan_c,t−1 × ∆e_c,t Col (7): -2.9864*** (-21.1).
- Larger sample (Table 18: 42 economies; Observations and R2):
  - All countries (col (1)): Observations 7,020,054; R2: 0.78; FXI_c,t−1 = 0.13 (0.09).
  - High FX debt economies (col (2)): Observations 1,284,781; R2: 0.79; FXI_c,t−1 = 0.07* (0.04).
  - Low FX debt economies (col (3)): Observations 4,108,779; R2: 0.82; FXI_c,t−1 = 0.01 (0.10).
- Key empirical conclusions on FXI:
  - In the 10 emerging market economies sample, FXI (foreign reserves-to-GDP ratio) has a strong negative effect on firm leverage (coefficients ~ -0.0036 to -0.0028, highly significant).
  - In the larger 42-economy sample, FXI is largely insignificant overall; weak positive effect in high FX debt economies (0.07*).
  - Interaction results indicate asymmetric responses: depreciations often produce larger adverse leverage effects than appreciations provide positive effects.
  - Effects are attributable primarily to FX loans rather than FX bonds.

### Conclusions and policy implications
- Main empirical conclusion:
  - Corporate FX loan exposure materially affects how firm leverage responds to exchange rate fluctuations:
    - Large depreciations (10 percent or more) in economies with non-trivial FX loan shares reduce firm financial leverage substantially (example: decline of 0.1517 using mean FX loans share 0.0996).
    - Large appreciations (10 percent or more) increase firm leverage (example: increase of 0.0265 using same mean).
  - FX loans, rather than FX bonds, drive the main effects.
- Policy-relevant implications:
  - Monitor firms’ FX exposure and extent of currency mismatch on balance sheets.
  - Currency appreciations combined with higher FX exposure can prompt firms to increase leverage during good times (risk taking), rendering them vulnerable to de-leveraging when currency depreciates.
  - Authorities expected to provide FX liquidity to firms (directly or via banks) when firms face FX valuation losses and funding strains; consider adequate FX safety nets proportional to the size of FX mismatches in aggregate and across firms.
  - It may be difficult for many emerging market economies to provide sufficient FX safety nets, especially during crises.
  - Macroprudential policies to prevent accumulation of un-hedged FX debt during booms might better mitigate vulnerabilities.
  - Using monetary policy to limit exchange rate fluctuations during crises in countries with high FX debt can be counter-productive.

### Appendix — Steps to calculate firm-level FX debt (as described)
1. Total FX debt in an economy.
   - Data source: BIS Global Liquidity Indicator (GLI) database.
   - Includes: debt securities denominated in the US dollar, euro and Japanese yen issued by the non-financial sector; bank loans denominated in the US dollar, euro and Japanese yen extended to the non-bank sector.
   - Bank loans include both cross-border bank loans and locally-extended bank loans.
   - Non-financial sector includes non-financial firms, households and the government; non-bank sector includes non-bank financial firms in addition to the non-financial sector.
   - Debt securities data from BIS International Debt Securities Statistics; bank loan data from BIS Locational Banking Statistics.
   - Subtract value of FX-denominated debt securities issued by the government to focus on non-financial firms’ FX-denominated debt securities.
   - When using country-level total FX debt, values denominated in the euro and Japanese yen are converted into US dollars using the quarter-end exchange rate.
2. Total credit to the non-financial sector.
   - Data source: BIS Total Credit database.
   - Includes all forms of credit (loans and debt securities) extended by banks and non-banks in all currencies.
   - Total credit is broken down into non-financial firms, households and the government.
3. Compute the FX debt share of each country as the ratio of FX debt to total credit to the non-financial sector (both include loans and debt securities).
4. Firm-level total debt.
   - Data source: ORBIS database; obtain total debt outstanding reported in firms’ balance sheets.
5. Estimate firm-level FX debt outstanding by multiplying the country FX debt share by firm-level total debt for all firms located in the economy.

*Italic: wpiea2020283-print-pdf - 0.1517 and the effect of a 10 percent or more appreciation is an increase in leverage of 0.0265.*

### 0.1517 and the effect of a 10 percent or more appreciation is an increase in leverage of 0.0265.

### wpiea2020283-print-pdf - 0.1517 and the effect of a 10 percent or more appreciation is an increase in leverage of 0.0265.

### Data and sample
- Sample period: 2002 to 2015.
- Economies covered (firm-level accounting data): China, Chinese Taipei, Hong Kong SAR, India, Indonesia, Korea, Malaysia, the Philippines, Singapore, Thailand.
- Unit of observation: firm-year.
- Final combined cleaned sample size (ORBIS + exchange rates): 1,661,677 firm-year observations.
- Sample used in regressions (after winsorization and cleaning): 1,373,080 observations.
- Variables and transformations:
  - Dependent variable: Leverage = book value of total financial debt / book value of total assets.
  - Collateral = tangible fixed assets / total assets.
  - Profitability = EBIT / total assets.
  - Size = log(total assets).
  - Sales growth = growth in sales (winsorized at 5 percent).
  - Exchange rate change dummies: Dummy∆e k equals 1 when nominal bilateral exchange rate vs USD changes by more than k where k ∈ {0, 5, 10} percent (sign indicates appreciation/depreciation).
  - FX debt measures: country-level share of FX debt = non-financial sector FX debt / total credit to the non-financial sector (from BIS GLI and Total Credit databases). Also decomposed into FX loans and FX bonds; government FX bonds subtracted in some specifications.
- Winsorization: All variables at 1 percent (sales growth at 5 percent).
- Estimation methods: OLS with standard errors clustered at the firm level; dynamic system GMM also used with similar results.
- Fixed effects and controls: Firm fixed effects, country fixed effects, granular industry (4-digit) × year fixed effects, and lagged controls (size, collateral, profitability, sales growth).

### Key empirical findings — FX exposure, exchange rate movements, and firm leverage
- Aggregate FX debt interaction:
  - Using the mean ratio of FX debt to total debt for the 10 Asian economies (0.1519), a 10 percent or more depreciation decreases firm leverage by 0.0844 (calculation reported in text based on benchmark results).
  - Effects grow with the size of depreciations and appreciations; depreciations generally have larger effects than appreciations.
  - When government FX bonds are excluded from non-financial sector FX debt (mean ratio 0.1191), a 10 percent or more depreciation decreases firm leverage by 0.1352, whereas a 10 percent or more appreciation increases firm leverage by 0.0108.
- FX loans drive the results (Table 7):
  - Mean ratio of FX loans to total debt: 0.0996.
  - Using that mean, a 10 percent or more depreciation produces a decline in leverage of 0.1517.
  - Using that mean, a 10 percent or more appreciation produces an increase in leverage of 0.0265.
  - Context: mean firm leverage is 0.16, so:
    - A 10 percent or more depreciation (decline in leverage of 0.1517) is a change nearly equivalent to the mean leverage (decline ≈ mean).
    - A 10 percent or more appreciation (increase in leverage of 0.0265) is a 20 percent increase relative to the mean leverage (0.0265 is 20 percent of 0.16).
- Directional patterns (summary of regression results):
  - In countries with higher corporate-sector FX debt shares, currency depreciations are associated with reductions in firm leverage; appreciations are associated with increases in firm leverage.
  - In countries with no FX debt, the (small) effects are opposite in sign: firm leverage increases with depreciations and decreases with appreciations.
- Robustness and additional checks:
  - Results robust to: subtracting government FX bonds from non-financial sector FX debt; separating FX loans and FX bonds (FX loans drive main results); using continuous exchange rate changes (Table 8); using firm-level FX debt proxies and limited direct firm-level FX debt data for 4 countries.
  - Placebo tests: movements less than 1 percent in exchange rates show no effect on firm leverage.
  - Additional sensitivity checks: controlling for firms in non-tradeable sectors; separating long-term vs short-term debt; use of country×time fixed effects via a proxy for firm-level FX debt share (assuming firm FX debt share = corporate sector FX debt share) to absorb country-time shocks such as terms-of-trade shocks.
- Magnitudes and statistical significance (selected coefficients reported in tables):
  - Benchmark interaction coefficients (examples from tables):
    - Table 5 (aggregate FX debt): FXdebt × Dummy∆e k (k≤−10%): -0.5312*** (t-stat -19.0) for depreciations; FXdebt × Dummy∆e k (k≥+10%): -0.2620*** (t-stat -10.8) for appreciations in an alternate ordering of columns.
    - Table 6 (FX debt excluding government bonds): FXdebt × Dummy∆e k (k≥+10%): -1.1352*** (t-stat -19.1); FXdebt × Dummy∆e k (k≤−10%): 0.0908*** (t-stat 2.8) — columns structured to capture different sign/direction conventions in the reported tables.
    - Table 7 (FX loans): FXloan × Dummy∆e k (k≥+10%): -1.5236*** (t-stat -23.7); FXloan × Dummy∆e k (k≤−10%): 0.2665*** (t-stat 7.4).
  - FX debt level coefficients (examples):
    - FXdebt c,t−1 coefficients around 2.38 to 2.99 across specifications (highly significant).
    - FXloan c,t−1 coefficients around 3.04 to 3.23 across specifications (highly significant).
  - Control variable coefficients (consistent and significant across specifications):
    - Profitability coefficient ≈ -0.0444 to -0.0441 (highly significant).
    - Collateral coefficient ≈ 0.0650 to 0.0695 (highly significant).
    - Size coefficient ≈ 0.0200 to 0.0219 (highly significant).
    - Sales growth coefficient small negative ≈ -0.0009 to -0.0010 (significant).
- Interpretation of magnitudes:
  - FX loan exposures substantially amplify the sensitivity of firm leverage to exchange rate fluctuations, with large depreciations materially reducing leverage and large appreciations increasing leverage, conditional on the country’s FX loan share.

### Identification concerns and strategies
- Threats: omitted time-varying country-level factors correlated with exchange rate fluctuations interacted with corporate FX debt shares (e.g., terms-of-trade shocks, government borrowing responses).
- Strategies used to address threats:
  - Use of granular industry-year fixed effects (4-digit) to capture industry-specific time-varying shocks and many supply/demand shocks.
  - Use of country×time fixed effects by proxying firm-level FX debt (assuming firm FX debt share equals corporate sector FX debt share) to allow inclusion of country×time fixed effects; this is an imperfect assumption but useful for absorbing omitted country-time heterogeneity.
  - Use of actual firm-level FX debt shares where available (4 countries) to validate proxy results.
  - Exclusion of multinational branches and firms with headquarters outside the reporting country to focus on firms whose financial decisions are governed by local conditions.
  - Lagging FX debt shares and control variables by one year.

### Policy-relevant findings on FX intervention (FXI) and moral hazard
- In a sample of 10 EMEs, an increase in FXI is associated with a decrease in firm leverage over time, conditional on other determinants including exchange rate fluctuations (negative effect of FXI on leverage in the 10-EME sample).
- Theoretical literature suggests FXI can create moral hazard by encouraging firms to increase FX debt (e.g., Diamond, Hu and Rajan (2018)); empirical evidence varies by sample.
- In a larger sample of 42 countries, the effect of FXI on firm leverage is largely insignificant.
- Note: the paper observes marginally positive significant FXI effects only in countries with high levels of FX debt at the beginning of the sample; differences from other studies (e.g., Tong and Wei (2019)) may reflect sample selection (listed firms vs broader firm coverage including private firms).

### Overall conclusions
- Corporate FX loan exposure materially affects how firm leverage responds to exchange rate fluctuations:
  - Large depreciations (10 percent or more) in economies with non-trivial FX loan shares reduce firm financial leverage substantially (example: decline of 0.1517 using mean FX loans share 0.0996).
  - Large appreciations (10 percent or more) increase firm leverage (example: increase of 0.0265 using same mean).
- FX loans, rather than FX bonds, drive the main effects.
- Results are robust to multiple specifications, placebo tests, alternative FX debt measures (excluding government bonds), continuous exchange rate changes, and additional fixed-effect structures.
- Identification concerns from country-level time-varying omitted factors are addressed through proxying firm FX debt, use of country×time and granular industry-year fixed effects, and limited direct firm-level FX debt data.

*Italic: wpiea2020283-print-pdf - 0.1517 and the effect of a 10 percent or more appreciation is an increase in leverage of 0.0265.*

### 3.2    Firm-Level FX Debt: Non-Financial Sector Proxy Approach

### 3.2    Firm-Level FX Debt: Non-Financial Sector Proxy Approach

### Methodology and identification
- Threat addressed: omitted country-year shocks, policies and other factors; inclusion of country×year fixed effects.
- Redefinition: FX debt redefined at the firm level using two approaches:
  - Proxy approach: assume each firm’s share of FX debt equals the aggregate non-financial sector share (after excluding government FX debt); apply this share to each firm’s total debt to obtain firm-level FX debt. By construction, larger firms have more debt and hence more FX debt → control for time-varying firm size.
  - Direct approach: use actual firm-level FX debt data for four countries (Korea, Argentina, Mexico, Colombia) for validation.
- Firm-level FX debt operationalization (proxy): FXdebt_i is a time-invariant dummy equal to 1 when the average value of FX debt of a firm is higher than the respective value of the median firm in the same country, and zero otherwise.
- Empirical specification (micro-level):
  Leverage_i,j,c,t = β · FXdebt_i × DummyΔe^k_c,t + θ1 · X_i,c,t−1 + θ2 · X_i,c,t−1 × DummyΔe^k_c,t + α_i + γ_c,t + φ_j,t + ε_i,j,c,t
  - α_i = firm fixed effects; γ_c,t = country-year fixed effects; φ_j,t = industry-year fixed effects.
- Estimation: OLS with robust standard errors clustered at the firm level; dynamic system GMM provided similar results. Results are robust to clustering at the country level.

### Benchmark and continuous-exchange-rate results (selected coefficients preserved exactly)
- Table 8 (continuous ∆e_c,t; dependent variable: Financial debt/assets):
  - FXdebt_c,t−1 × ∆e_c,t: -0.2809*** (t = -3.8)
  - FXdebt_c,t−1: 2.4462*** (t = 68.8)
  - ∆e_c,t: 0.1237*** (t = 30.9)
  - Profitability_i,j,c,t−1: -0.0440*** (t = -23.6)
  - Collateral_i,j,c,t−1: 0.0654*** (t = 34.9)
  - Size_i,j,c,t−1: 0.0196*** (t = 27.3)
  - Sales growth_i,j,c,t−1: -0.0011*** (t = -3.9)
  - Fixed effects: Firm FE Yes; Country FE Yes; Industry-Year FE Yes
  - Observations: 1,373,080
  - R^2: 0.7813

- Table 10 (proxy FXdebt_i × continuous ∆e_c,t; dependent variable: Financial debt/assets):
  - FXdebt_i × ∆e_c,t: -0.0985*** (t = -21.2)
  - Profitability_i,j,c,t−1: -0.0456*** (t = -24.4)
  - Profitability_i,j,c,t−1 × ∆e_c,t: -0.0691*** (t = -3.3)
  - Collateral_i,j,c,t−1: 0.0703*** (t = 37.9)
  - Collateral_i,j,c,t−1 × ∆e_c,t: 0.0260*** (t = 3.1)
  - Size_i,j,c,t−1: 0.0207*** (t = 28.8)
  - Size_i,j,c,t−1 × ∆e_c,t: 0.0173*** (t = 10.4)
  - Sales growth_i,j,c,t−1: -0.0006** (t = -2.0)
  - Observations: 1,372,970
  - R^2: 0.79

### Difference-in-Differences interpretation and main findings (proxy approach)
- Interaction: firm-level FX debt dummy × appreciation/depreciation dummy interpreted in a Difference-in-Differences framework; interaction of firm controls with appreciation/depreciation dummies included.
- Fixed effects included: firm, industry-year, country-year (the appreciation/depreciation dummy is absorbed by country-year FE; firm-level FX debt dummy is absorbed by firm FE).
- Main inference: results similar to earlier country-level findings but with smaller coefficients once country×year fixed effects are included, highlighting the role of other policies and country-level shocks.

### Robustness checks, placebo, and additional tests
- Table 9 (proxy FXdebt_i; dependent variable: Financial debt/assets; k≥+10% and k≤−10% specifications):
  - For k≥+10%:
    - FXdebt_i × DummyΔe^k_c,t: -0.0058*** (t = -3.7)
    - Profitability_i,j,c,t−1: -0.0442*** (t = -23.9)
    - Profitability_i,j,c,t−1 × DummyΔe^k_c,t: -0.0235*** (t = -2.7)
    - Collateral_i,j,c,t−1: 0.0690*** (t = 37.1)
    - Collateral_i,j,c,t−1 × DummyΔe^k_c,t: 0.0364*** (t = 12.7)
    - Size_i,j,c,t−1: 0.0202*** (t = 28.2)
    - Size_i,j,c,t−1 × DummyΔe^k_c,t: 0.0040*** (t = 6.9)
    - Sales growth_i,j,c,t−1: -0.0006* (t = -1.9)
    - Observations: 1,372,970; R^2: 0.79
  - For k≤−10%:
    - FXdebt_i × DummyΔe^k_c,t: 0.0550*** (t = 26.8)
    - Profitability_i,j,c,t−1: -0.0427*** (t = -23.2)
    - Profitability_i,j,c,t−1 × DummyΔe^k_c,t: -0.0151*** (t = -2.7)
    - Collateral_i,j,c,t−1: 0.0704*** (t = 38.1)
    - Collateral_i,j,c,t−1 × DummyΔe^k_c,t: -0.0114*** (t = -3.6)
    - Size_i,j,c,t−1: 0.0212*** (t = 29.5)
    - Size_i,j,c,t−1 × DummyΔe^k_c,t: -0.0106*** (t = -19.7)
    - Sales growth_i,j,c,t−1: -0.0005* (t = -1.8)
    - Observations: 1,372,970; R^2: 0.79

- Placebo test (Table 11): very small exchange rate changes (appreciations/depreciations less than 1% and alternative thresholds up to 2%) show insignificant interaction coefficients, indicating the effects require sufficiently large exchange rate movements.
  - Examples from Table 11:
    - FXdebt_i × DummyΔe^k_c,t for 1%≥k>0%: -0.0629 (t = -1.1) [not significant]
    - For −1%≤k<−0%: -0.0049* (t = -8.9) with fewer observations in event sample notes; overall interpretation in text: "Clearly this is insignificant." (exercise shows importance of size of exchange rate movements)

- Tradeable vs non-tradeable sector (Table 12):
  - Depreciations (k≤−10%):
    - Tradeable: FXdebt_i × DummyΔe^k_c,t = 0.0341*** (t = 10.8) — insignificant in text interpretation for depreciations (effect stronger in non-tradeable)
    - Non-tradeable: FXdebt_i × DummyΔe^k_c,t = 0.0643*** (t = 22.4)
  - Appreciations (k≥+10%):
    - Tradeable: FXdebt_i × DummyΔe^k,t = -0.0037 (t = -1.5)
    - Non-tradeable: FXdebt_i × DummyΔe^k,t = -0.0048** (t = -2.3)
  - Interpretation: depreciations significant in the non-tradeable sector while insignificant in tradeable sector; appreciations significant in both but stronger in non-tradeable sector (coefficient roughly twice that of tradeable).

- Debt maturity (Table 13: LT and ST debt/assets):
  - Depreciations (k≤−10%):
    - FXdebt_i × DummyΔe^k_c,t on LT: 0.0501*** (t = 27.4)
    - FXdebt_i × DummyΔe^k_c,t on ST: 0.0052*** (t = 5.9)
    - Text interpretation: during large depreciations firms with higher FX debt increase short-term based leverage and decrease long-term based leverage (roll-over behavior). For appreciations, effects are stronger for long-term debt-based leverage with negative coefficients (improving balance sheets).
  - Appreciations (k≥+10%):
    - FXdebt_i × DummyΔe^k_c,t on LT: -0.0099*** (t = -7.1)
    - FXdebt_i × DummyΔe^k_c,t on ST: 0.0038*** (t = 3.2)
  - Observations: 1,372,970 each; R^2 around 0.74–0.75.

### Firm-level actual FX debt shares: evidence from select countries (Korea and others)
- Concern addressed: proxy approach validity → compare with actual firm-level FX debt share data for Korea, Argentina, Mexico, Colombia (data availability limited; listed firms only).
- Korea comparison (BIS aggregate vs KIS firm data, 1994–1999):
  - Aggregate BIS country-level FX debt/Total debt ratio and KIS firm-level patterns follow similar time variation; 1997 spike in BIS aggregate larger than KIS average (Asian financial crisis) — shows importance of representative firm samples (ORBIS vs KIS differences highlighted).
  - ORBIS sample vs KIS sample differences (illustrative numbers):
    - ORBIS-based sample of Korean firms: 151,684 firms (median total assets ≈ 3 billion won; median total liabilities ≈ 1.5 billion won).
    - KIS sample: 6,987 firms (median total assets ≈ 13 billion won; median total liabilities ≈ 9 billion won).
  - Conversion note: to compare, authors estimate FX debt-to-total debt ratio from KIS by dividing foreign debt-to-liability ratio by 0.6 (the ORBIS sample average debt-to-liability ratio).

- Regression using Korea KIS firm-level FX debt share (Table 14; dependent variable: Liability/assets; firm fixed effects and industry-year FE included; Observations: 17,033; R^2: 0.83):
  - Columns (1)–(6) (thresholds k≥+0%, k≥+5%, k≥+10%, k≤−0%, k≤−5%, k≤−10%):
    - FXdebt share_i × DummyΔe^k_c,t:
      - k≥+0%: -0.013*** (t = -4.4)
      - k≥+5%: -0.013*** (t = -4.4)
      - k≥+10%: -0.020*** (t = -7.0)
      - k≤−0%: 0.013*** (t = 4.4)
      - k≤−5%: 0.015*** (t = 3.8)
      - k≤−10%: 0.039*** (t = 12.3)
  - Column (7) (continuous ∆e_c,t interaction): FXdebt share_i × ∆e_c,t = -0.038*** (t = -9.3)
  - Interpretation: signs and significance of interaction term align with benchmark Table 9 results.

- Additional firm-level FX debt data (Argentina, Mexico, Brazil; 1990–2005):
  - For large depreciations, threshold set at the 90th percentile of exchange rate changes (~24% depreciation).
  - For large appreciations, threshold set at the 1st percentile (~16% appreciation).
  - Key finding: results consistent with previous results for depreciations; for appreciations only exporters (defined as firms with ≥10 percent of revenues from exports) significantly increase leverage with appreciations (Table 15 referenced in text).

### Key takeaways
- The proxy approach to construct firm-level FX debt (using country non-financial sector FX debt share applied to firm debt) yields consistent within-sample dynamics once validated against firm-level FX debt data for select countries.
- Large exchange rate movements matter: appreciations and depreciations of sufficient magnitude (e.g., ≥10% or ≤−10%) drive significant differential leverage responses for firms with high FX debt exposure; minimal exchange rate changes (≤1–2%) do not produce significant effects.
- Country×year fixed effects materially reduce coefficient sizes relative to specifications without these fixed effects, indicating other country-year shocks and policies help explain part of the relation between exchange rate movements and firm leverage.
- Heterogeneity:
  - Non-tradeable sector firms experience stronger balance-sheet effects from exchange rate movements compared with tradeable sector firms.
  - Maturity matters: long-term debt-based leverage exhibits the main effects for appreciations; during large depreciations, firms with high FX debt increase short-term leverage and reduce long-term leverage (consistent with roll-over behavior during stress).
- Validation using actual firm-level FX debt shares (Korea and other countries) corroborates the proxy-based findings in sign and significance.

*Italic source: 3.2 Firm-Level FX Debt: Non-Financial Sector Proxy Approach — wpiea2020283-print-pdf*

### 3.5    The Role of Exchange Rate Policies

### 3.5    The Role of Exchange Rate Policies

### Overview
- Literature provides rationale for foreign exchange intervention (FXI) to limit appreciations (lean against the wind) to restrain firms’ risk-taking and leverage; parallel literature highlights moral hazard costs of FXI during depreciations.
- FXI proxied by the FX reserves-to-GDP ratio; authors examine how FXI interacts with firms’ foreign-currency (FX) debt to affect firm leverage.

### Firm-level evidence (selected results)
- Table 15: Firm-Level FX Debt: Evidence from Argentina, Mexico and Brazil
  - Dependent variable: Liability/assets i,j,c,t (All firms / High exporters)
  - Coefficient on FXdebt_i × Dummy∆e_k_c,t (k≥90th percentile, All firms): -0.017** (0.01)
  - Coefficient on FXdebt_i × Dummy∆e_k_c,t (k≤1st percentile, All firms): 0.020 (0.02)
  - Coefficient on FXdebt_i × Dummy∆e_k_c,t (k≥90th percentile, High exporters): -0.031** (0.01)
  - Coefficient on FXdebt_i × Dummy∆e_k_c,t (k≤1st percentile, High exporters): 0.060** (0.03)
  - Observations: 9,393 (All firms), 4,043 (High exporters)
  - R2: 0.77 (All firms), 0.75 (High exporters)
  - Notes: FXdebt_i represents the short-term dollar debt share of a given firm. Standard errors clustered at the firm level. ∗ p<0.10, ∗∗ p<0.05, ∗∗∗ p<0.01.

- Table 16: Benchmark Results with FXI
  - Dependent variable: Financial debt/assets i,j,c,t; sample observations: 1,373,080 across columns (1)–(7); R2: 0.78 in all columns.
  - FXI_c,t−1 (foreign reserves-to-GDP ratio in year t−1) coefficients:
    - Column (1) k≥+0%: -0.0036*** (-37.0)
    - Column (2) k≥+5%: -0.0033*** (-34.9)
    - Column (3) k≥+10%: -0.0030*** (-30.4)
    - Column (4) k≤−0%: -0.0036*** (-37.2)
    - Column (5) k≤−5%: -0.0038*** (-38.3)
    - Column (6) k≤−10%: -0.0037*** (-40.5)
    - Column (7) k (∆e_c,t for col (7)): -0.0032*** (-30.2)
  - FXdebt_c,t−1 × Dummy∆e_k_c,t coefficients:
    - Col (1): -0.0338*** (-4.7)
    - Col (2): -0.0659*** (-7.8)
    - Col (3): -0.4587*** (-17.7)
    - Col (4): 0.0340*** (4.7)
    - Col (5): 0.0495*** (6.8)
    - Col (6): -0.2856*** (-11.9)
    - Col (7) (×∆e_c,t for col (7)): -0.1663** (-2.4)
  - FXdebt_c,t−1 level coefficients:
    - Col (1): 1.9942*** (63.1)
    - Col (2): 2.0263*** (63.0)
    - Col (3): 2.1096*** (64.9)
    - Col (4): 1.9581*** (58.1)
    - Col (5): 1.9042*** (55.3)
    - Col (6): 2.0642*** (61.0)
    - Col (7): 2.0640*** (54.6)
  - Dummy∆e_k_c,t coefficients:
    - Col (1): 0.0104*** (12.2)
    - Col (2): 0.0246*** (26.9)
    - Col (3): 0.0529*** (34.2)
    - Col (4): -0.0102*** (-12.1)
    - Col (5): -0.0020*** (-3.4)
    - Col (6): 0.0113*** (6.6)
    - Col (7) (∆e_c,t for col (7)): 0.0683*** (17.1)
  - Selected controls (coefficients identical across columns as reported):
    - Profitability_i,j,c,t−1: -0.0439*** (-23.5)
    - Collateral_i,j,c,t−1: 0.0667*** (35.6)
    - Size_i,j,c,t−1: 0.0196*** (27.3)
    - Sales growth_i,j,c,t−1: -0.0011*** (-3.8)
  - Firm FE, Country FE, Industry-Year FE included. t-stats based on clustered standard errors. ∗ p<0.10, ∗∗ p<0.05, ∗∗∗ p<0.01.

- Table 17: Benchmark Results with FXI: Robustness
  - Sample variations: Benchmark Results excluding government FX bonds; Benchmark Results using FX Loans.
  - FXI_c,t−1 coefficients (selected):
    - Excluding government FX bonds, Col (1) k≥+0%: -0.0036*** (-38.4)
    - FX Loans, Col (1) k≥+0%: -0.0037*** (-40.2)
  - FXdebt_c,t−1 × Dummy∆e_k_c,t (excluding government FX bonds):
    - Col (1): -0.1385*** (-10.5)
    - Col (2): -0.3135*** (-19.6)
    - Col (3): -0.9817*** (-17.6)
    - Col (7) (×∆e_c,t): -1.7370*** (-14.6)
  - FXloan_c,t−1 × Dummy∆e_k_c,t (FX Loans specification):
    - Col (1): -0.0868*** (-5.5)
    - Col (2): -0.5032*** (-25.1)
    - Col (3): -1.3152*** (-20.9)
    - Col (7) (×∆e_c,t): -2.9864*** (-21.1)
  - Dummy∆e_k_c,t and other controls reported; Firm controls, Firm FE, Country FE, Industry-Year FE included.

- Table 18: The effect of FXI: Larger Sample of Economies
  - Dependent variable: Financial debt/assets i,j,c,t; Observations:
    - All countries (column (1)): 7,020,054; R2: 0.78
    - High FX debt economies (column (2)): 1,284,781; R2: 0.79
    - Low FX debt economies (column (3)): 4,108,779; R2: 0.82
  - FXI_c,t−1 coefficients:
    - All countries: 0.13 (0.09)
    - High FX debt economies: 0.07* (0.04)
    - Low FX debt economies: 0.01 (0.10)
  - Firm controls, Firm FE, Country FE, Industry-Year FE included. Firm clustered standard errors in parentheses. ∗ p<0.10, ∗∗ p<0.05, ∗∗∗ p<0.01.

### Key empirical findings and magnitudes
- FXI (foreign reserves-to-GDP ratio) has a strong negative effect on firm leverage in the 10 emerging market economies sample (coefficients ~ -0.0036 to -0.0028, all significant at ∗∗∗ levels with large t-statistics).
- Interaction of firm FX exposure (FXdebt) with exchange rate movements:
  - Positive FXdebt × appreciation dummy coefficients are negative in high FX exposure percentiles (showing larger leverage increases when currency appreciates for FX-exposed firms), with magnitudes varying across specifications (e.g., -0.4587*** in Table 16, col (3)).
  - Interactions indicate asymmetric responses: depreciations often produce larger adverse leverage effects than appreciations produce positive effects, especially for depreciations larger than 10 per cent (noted in text and reflected in larger-magnitude coefficients for large depreciation dummies).
- Robustness:
  - Results hold when excluding government FX bonds and when focusing on FX loans (effects driven more by FX loans than FX bonds).
  - In a larger sample of 42 economies (over 7 million firms), FXI is insignificant overall, with a weak positive effect in high FX debt economies (FXI coefficient 0.07* in high FX debt economies).

### Conclusions (as stated)
- When home currency appreciates, firms with FX debt and local currency assets see higher net worth (appreciation lowers the value of FX debt) and can borrow more; when home currency depreciates, firms deleverage.
- Firms in countries with higher aggregate non-financial sector FX debt increase (decrease) leverage relatively more after appreciations (depreciations).
- The effect of depreciation is quantitatively larger than that of appreciation, especially for depreciations larger than 10 per cent.
- Results are attributable to FX loans rather than FX bonds.
- FXI policies designed to limit exchange rate fluctuations are associated with a negative significant effect on firm leverage in the 10 emerging market economies sample but show no effect in the larger 42-economy sample.

### Policy implications (as stated)
- Monitor firms’ FX exposure and extent of currency mismatch on balance sheets.
- Currency appreciations combined with higher FX exposure can prompt firms to increase leverage during good times (risk taking), rendering them vulnerable to de-leveraging when currency depreciates.
- Authorities often expected to provide FX liquidity to firms (directly or via banks) when firms face FX valuation losses and funding strains; this implies considering adequate FX safety nets proportional to the size of FX mismatches in aggregate and across firms.
- It may be difficult for many emerging market economies to provide sufficient FX safety nets, especially during crises.
- Macroprudential policies to prevent accumulation of un-hedged FX debt during booms might be a better approach to mitigate these vulnerabilities.
- Kalemli-Ozcan (2019) noted that using monetary policy to limit exchange rate fluctuations during crises in countries with high FX debt can be counter-productive.

### Appendix — Steps to calculate firm-level FX debt (as described)
1. Total FX debt in an economy.
   - Data source: BIS Global Liquidity Indicator (GLI) database.
   - Includes: debt securities denominated in the US dollar, euro and Japanese yen issued by the non-financial sector; bank loans denominated in the US dollar, euro and Japanese yen extended to the non-bank sector.
   - Bank loans include both cross-border bank loans and locally-extended bank loans.
   - Non-financial sector includes non-financial firms, households and the government; non-bank sector includes non-bank financial firms in addition to the non-financial sector.
   - Debt securities data from BIS International Debt Securities Statistics; bank loan data from BIS Locational Banking Statistics.
   - Subtract value of FX-denominated debt securities issued by the government to focus on non-financial firms’ FX-denominated debt securities.
   - When using country-level total FX debt, values denominated in the euro and Japanese yen are converted into US dollars using the quarter-end exchange rate.
2. Total credit to the non-financial sector.
   - Data source: BIS Total Credit database.
   - Includes all forms of credit (loans and debt securities) extended by banks and non-banks in all currencies.
   - Total credit is broken down into non-financial firms, households and the government.
3. Compute the FX debt share of each country as the ratio of FX debt to total credit to the non-financial sector (both include loans and debt securities).
4. Firm-level total debt.
   - Data source: ORBIS database; obtain total debt outstanding reported in firms’ balance sheets.
5. Estimate firm-level FX debt outstanding by multiplying the country FX debt share by firm-level total debt for all firms located in the economy.

*Source: wpiea2020283-print-pdf - 3.5    The Role of Exchange Rate Policies*

---


_Source: https://www.imf.org/-/media/files/publications/wp/2020/english/wpiea2020283-print-pdf.pdf_