## wp17119

## Source details

**Canonical URL:** [wp17119](https://www.imf.org/-/media/files/publications/wp/2017/wp17119.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/wp/2017/wp17119.pdf.md)
- [Structured JSON version](/-/media/files/publications/wp/2017/wp17119.pdf.json)

---

### I. Introduction — challenge to conventional wisdom
- Conventional view: treatment of private creditors in sovereign debt exchanges does not vary by maturity; pari passu clauses imply equal ranking.
- New claim: creditor losses (“haircuts”) systematically vary by maturity; specifically, haircuts on shorter-term debt tend to be larger.
- Evidence basis: a new, comprehensive dataset on instrument-specific haircuts in 28 sovereign debt restructuring episodes (18 external and 10 domestic) over 1999–2015.
- Robustness: finding holds across measures (Sturzenegger and Zettelmeyer (SZ) measure and a price-based “Exchange” measure), across external and domestic episodes, preemptive and post-default restructurings, and across exchange methods (single menu and multiple-menu cases).

### II. Stylized facts and economic mechanism
- Stylized facts:
  - Haircuts on short-term bonds are larger than on longer-term bonds.
  - Short-term bond prices tend to be higher before an exchange than prices of longer-term debt; in the run-up to restructuring, short-term bond prices converge to long-term bond prices from above.
- Economic mechanism (arbitrage-free risk-neutral pricing intuition):
  - If investors anticipate strictly positive default risk after the maturity date of short-term debt, then short-term bond prices exceed longer-term bond prices (other characteristics constant).
  - If all outstanding instruments are exchanged for the same new instrument, higher pre-exchange short-term prices imply larger ex-post haircuts on short-term bonds.
  - If perceived default risk concentrates in the short term in the run-up to restructuring, prices across maturities converge and haircut differentials diminish as the event approaches.
- Empirical confirmation: panel regressions from 6 months before announcement to exchange/default date, controlling for restructuring-specific factors, confirm predicted price convergence driven by increasingly short-term default risk.

### III. Haircut measures (definitions and properties)
- SZ recovery rate measure (Sturzenegger and Zettelmeyer 2006, 2008):
  - SZ R^T_i ≡ NPV_T(e, r^T_e) / NPV_T(i, r^T_e).
  - Numerator: observed price of new instrument immediately after the exchange.
  - Denominator: synthetic present value of the old instrument discounted at the new instrument's exit yield; not directly observable.
- Exchange recovery rate (price-based alternative):
  - R_{t,T}^i ≡ NPV_T(e, r^T_e) / NPV_t(i, r^t_i).
  - Equals ratio of two observed prices: price of new instrument immediately after exchange divided by price of old instrument at date t ≤ T.
  - Advantage: computable from observed market prices; allows measures over varying windows prior to exchange.
  - Relation to SZ: R_{T,T}^i / S Z R^T_i = NPV_T(i, r^T_e) / NPV_T(i, r^T_i) ≈ 1 (empirically minor difference).
  - Comparative dynamics: R_{t,T}^i − R_{t,T}^j across varying T−t illustrates which instrument’s price converged more quickly to the new instrument’s market value.

### IV. Dataset and scope
- Sample: 28 restructuring episodes (18 external, 10 domestic) over 1999–2015.
- Dating convention: start = month of default or announcement of distressed restructuring; end = month of final agreement or implementation of debt exchange.
- Coverage:
  - SZ-measure haircuts: 449 instruments across 18 external and 10 domestic episodes.
  - Exchange-measure haircuts: 111 instruments across 12 external and 6 domestic episodes.
  - For R_{t,T}^i, price data collected monthly from 6–9 months prior to announcement up to exchange.
- Data sources: offering memoranda, government press releases, financial sector databases and reports, IMF staff reports, SZ, Cruces and Trebesch (2013), Asonuma and Papaioannou (2016); market yields/prices from Bloomberg, Datastream, Dealogic, JP Morgan Markit.
- Instrument-level fields collected: maturity, coupon, payment type, options; same fields for new instruments received in exchange.

### V. Empirical approaches and key results
- Cross-sectional and panel regression findings:
  - Cross-sectional regressions with instrument- and episode-specific effects confirm that shorter maturities face larger haircuts.
  - Panel regressions over the 6 months prior to announcement until exchange/default show price convergence across maturities consistent with rising short-term default probabilities.
- Robustness:
  - SZ measure and Exchange measure yield consistent qualitative results; alternative windows and measures produce similar conclusions.
  - SZ-measure sample in this study: 449 instruments (more than twice SZ’s sample).
- Quantitative summaries (selected):
  - On average, the recovery rate on a 10-year bond is 3–12 percentage points higher than on a 1-year bond.
  - Over 6 months the price difference between short-term and longer-term bonds diminishes by 9 percentage points (relative to face value of 100).
  - Example regression coefficients (Table 2 / columns reported):
    - SZ Recovery Rate coefficients: 0.30*** (0.08); 0.35*** (0.08).
    - Exchange Recovery Rate coefficients: 1.03*** (0.30); 1.24*** (0.29).
  - Price dynamics (Table 3 example coefficients):
    - Time span until exchange coefficient: -1.52*** (0.08).
    - Maturity Differential (S − L, year) coefficient: -0.49*** (0.05).
  - Default probability changes (Table 4 example):
    - Over 6 months the default risk increases by 7–16 percentage points.
    - Default Probability 1-Year Horizon regression: Time Span until the Exchange Date coefficient 1.09*** (0.26); 2.70*** (0.37) in specifications reported.

### VI. Appendices — selected quantitative details
- Appendix II (Recovery rates, maturity, instrument detail):
  - Sample/regression counts (Table 2 excerpts):
    - Number of Countries: 15 (cols 1–2); 12 (cols 3–4).
    - Number of Restructurings: 28 (cols 1–2); 16 (cols 3–4).
    - Observations: 449 (col 1), 447 (col 2), 111 (cols 3–4).
    - Adjusted R-Squared: 0.035 (col 1), 0.262 (col 2), 0.112 (col 3), 0.247 (col 4).
- Appendix IV (Bond price panel; Table A3 selected coefficients):
  - Columns reported:
    - Column (1): coef -1.42*** (0.09)
    - Column (2): coef -1.43*** (0.10)
    - Column (3): coef -1.79*** (0.15)
    - Column (4): coef -1.81*** (0.15)
  - Additional reported coefficient examples: -0.51*** (0.05); -0.40*** (0.05); 0.81*** (0.22); -37.95*** (2.41).
  - Sample design: Number of Restructurings: 6; Observations: Columns 1–2: 487; Columns 3–4: 118.
- Appendix V (Estimation of term structure of default risk — preserved formulas):
  - Monthly interpolation:
    - CDS_t^M,j = CDS_t^A,i + (j−i)/12 (CDS_t^A,i+12 − CDS_t^A,i)
    - RF_t^M,j = RF_t^A,i + (j−i)/12 (RF_t^A,i+12 − RF_t^A,i)
  - No-arbitrage forward condition:
    - (1 + RF_t^M,h,j + CDS_t^M,h,j) = (1 + RF_t^M,j + CDS_t^M,j) / (1 + RF_t^M,h + CDS_t^M,h)
  - Risk-neutral relation for implied conditional default probability P̂_t^h,j:
    - (1 + RF_t^M,h,j) = (1 − P̂_t^h,j) * (1 + RF_t^M,h,j + CDS_t^M,h,j) + P̂_t^h,j * R̂_t
    - R̂_t = argmin_{R̂_t} (1/n) ∑ { p_t^i − p̂_t^i(...) }^2_{i=1}
- Appendix VI (Default probability panel; Table A4 selected coefficients):
  - Column (1) Default Probability 2-Year Horizon: 0.87*** (0.22)
  - Column (2) with controls: 2.16*** (0.31)
  - Column (3) Default Probability 3-Year Horizon: 0.68*** (0.21)
  - Column (4) with controls: 1.87*** (0.30)
  - Sample: Number of Restructurings: 3; Number of Time Periods (months): 12; Observations: 36.

### VII. Asset-pricing model: intuition and formal results
- Setup:
  - Discrete-time model with bonds maturing at date m paying coupon c_m if no restructuring, or recovery value φ_t^m if issuer announces restructuring.
  - Probability of no restructuring between t and s (s>t), given survival until t, equals π_t,s.
  - Investors risk neutral; risk-free gross interest rate equals β^{-1}. Ex-coupon price p_t^m satisfies an asset-pricing identity.
- First result:
  - With identical coupons and recovery values across bonds, long-term restructuring risk implies current price difference Δ_t^{s,l} ≡ p_t^s − p_t^l > 0: short-term bonds are more expensive than longer-term bonds.
- Second result:
  - Over time, Δ_t^{s,l} evolves because time passes and because news about future restructuring risk arrives.
  - Absent new information, price differences tend to diverge further (Δ_{t+1} − Δ_t > 0). If new information raises short-term default risk sufficiently, Δ may fall — i.e., prices converge as short-term default risk rises in the run-up to exchange.
  - Interpretation: selected episodes leading to restructuring are characterized by rising short-term default risk; model predicts positive but decreasing Δ over time, explaining higher haircuts on shorter-term debt and weaker maturity-haircut relation closer to exchange.

### VIII. Default probability term structure and short-term risk
- Stylized fact 3: The term structure of default probability becomes elevated at the short end in the run-up to the exchange.
- Estimation approach:
  - Inputs: CDS spread curve and US Treasury bill yield curve as risk-free proxy.
  - Construct CDS spread curve for horizons 1–120 months (monthly frequency) via linear interpolation; derive forward default spread curve and implied one-month ahead conditional default probability.
- Evidence:
  - Conditional short-term default probability (1-year horizon) increases in the run-up to exchange for Ecuador 1999–2000 and Greece 2011–2012; similar patterns at 2-year and 3-year horizons.
- Econometric results:
  - Main result: conditional default probability increases in the run-up to the exchange; significant at the 1-percent level.
  - Quantitatively: over 6 months the default risk increases by 7–16 percentage points.
  - Decomposition: restructuring strategy (strictly preemptive vs post-default) is a significant explanatory variable; duration of restructuring reduces conditional default probability in regressions.

### IX. Conclusion and implications
- Main conclusion: creditors are not treated equally in sovereign debt exchanges; shorter-term instruments tend to receive larger haircuts than longer-term instruments.
- Supporting chain: rising short-term default risk → higher prices for short-term instruments pre-run-up and convergence as exchange approaches → larger ex-post haircuts on shorter-term instruments when measured following common approaches.
- Potential implications (authors note further research required before normative conclusions):
  - Relevance to pari passu interpretations and limitations.
  - Negotiation dynamics during debt restructuring episodes.
  - Value-at-risk calculation for debt securities of different maturities and implications for financial institutions holding these instruments.
  - Optimal debt management considerations.

*Source: wp17119 - References and selected appendices excerpts.*

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

### wp17119 - References

### I. Introduction — challenge to conventional wisdom
- Conventional view: treatment of private creditors in sovereign debt exchanges does not vary by maturity; pari passu clauses imply equal ranking (Institute of International Finance 2012; 2015; IMF 2014).
- New claim: creditor losses (“haircuts”) systematically vary by maturity; specifically, haircuts on shorter-term debt tend to be larger.
- Evidence basis: a new, comprehensive dataset on instrument-specific haircuts in 28 sovereign debt restructuring episodes (18 external and 10 domestic) over 1999–2015.
- Robustness: finding holds across measures (Sturzenegger and Zettelmeyer (SZ) measure and a price-based “Exchange” measure), across external and domestic episodes, preemptive and post-default restructurings, and across exchange methods (single menu and multiple-menu cases).

### II. Stylized facts and economic mechanism
- Stylized fact 1: Haircuts on short-term bonds are larger than on longer-term bonds.
- Stylized fact 2: The negative relation between maturity and haircut reflects that, ceteris paribus, short-term bond prices tend to be higher before an exchange than prices of longer-term debt; in the run-up to restructuring, short-term bond prices converge to long-term bond prices from above.
- Theoretical motivation:
  - Using arbitrage-free risk-neutral pricing: if investors anticipate strictly positive default risk after the maturity date of short-term debt, then short-term bond prices exceed longer-term bond prices (other characteristics constant).
  - If a restructuring exchanges all outstanding instruments for the same new instrument, higher pre-exchange short-term prices imply larger ex-post haircuts on short-term bonds.
  - If perceived default risk concentrates in the short term in the run-up to restructuring, prices across maturities converge and haircut differentials diminish as the event approaches.
- Empirical confirmation: panel regressions from 6 months before announcement to exchange/default date, controlling for restructuring-specific factors, confirm predicted price convergence driven by increasingly short-term default risk.

### III. Haircut measures (definitions and properties)
- SZ recovery rate measure (Sturzenegger and Zettelmeyer 2006, 2008):
  - Define NPV_t(i,r) as net present value at time t of cash-flow stream of instrument i discounted at rate r.
  - SZ recovery rate for instrument i exchanged at date T:
    - S Z R^T_i ≡ NPV_T(e, r^T_e) / NPV_T(i, r^T_e)  (equation (1))
  - Numerator equals the observed price of the new instrument immediately after the exchange.
  - Denominator is a synthetic present value of the old instrument discounted at the new instrument's exit yield; it is not directly observable.
- Exchange recovery rate (price-based alternative):
  - R_{t,T}^i ≡ NPV_T(e, r^T_e) / NPV_t(i, r^t_i)  (equation (2))
  - By construction this equals the ratio of two observed prices: price of new instrument immediately after exchange divided by price of old instrument at date t ≤ T.
  - Advantage: computable from observed market prices; allows measures over varying windows prior to exchange.
  - Relation to SZ: R_{T,T}^i / S Z R^T_i = NPV_T(i, r^T_e) / NPV_T(i, r^T_i) ≈ 1 (empirically the difference is minor).
  - Comparative dynamics: R_{t,T}^i − R_{t,T}^j across varying T−t illustrates which instrument’s price converged more quickly to the new instrument’s market value.

### IV. Dataset and scope
- Sample: 28 restructuring episodes (18 external, 10 domestic) over 1999–2015.
- Dating convention: start = month of default or announcement of distressed restructuring; end = month of final agreement or implementation of debt exchange (following Asonuma and Papaioannou (2016) and Asonuma and Trebesch (2016)); distressed restructurings per Standard & Poor’s definition.
- Coverage:
  - SZ-measure haircuts: 449 instruments across 18 external and 10 domestic episodes.
  - Exchange-measure haircuts: 111 instruments across 12 external and 6 domestic episodes (fewer due to limited yields/prices data).
  - For R_{t,T}^i, price data collected in monthly frequency from 6–9 months prior to announcement up to exchange.
- Data sources: offering memoranda, government press releases, financial sector databases and reports, IMF staff reports, SZ, Cruces and Trebesch (2013), Asonuma and Papaioannou (2016); market yields/prices from Bloomberg, Datastream, Dealogic, JP Morgan Markit.
- Instrument-level collected fields: maturity, coupon, payment type, options; same fields for new instruments received in exchange.

### V. Empirical approaches and results
- Cross-sectional and panel regression evidence:
  - Cross-sectional regressions controlling for instrument- and episode-specific effects confirm the stylized fact that shorter maturities face larger haircuts.
  - Panel regressions over the 6 months prior to announcement until exchange/default show price convergence across maturities consistent with rising short-term default probabilities.
- Robustness checks: SZ measure and Exchange measure yield consistent qualitative results; alternative windows and measures produce similar conclusions.
- Dataset size comparison: the SZ-measure sample in this study contains more than twice the number of observations than SZ (449 instruments vs. SZ’s smaller sample).

### VI. Relation to literature and implications
- Empirical literature: extends work on creditor losses (Eichengreen and Portes 1986, 1989; Lindert and Morton 1989; Sturzenegger and Zettelmeyer 2006, 2008; Benjamin and Wright 2009; Cruces and Trebesch 2013; Finger and Mecagni 2007; Bedford et al. 2005; Diaz-Cassou et al. 2008; Hatchondo et al. 2014).
- Theoretical literature: implications for models that assume symmetric haircuts across maturities (Eaton and Gersovitz 1981; Rodrik and Velasco 1999; Jeanne 2009; Broner et al. 2013; Arellano and Ramanarayanan 2012; Chatterjee and Eyigungor 2012; Fernandez and Martin 2014; Aguiar and Amador 2014; Hatchondo and Martinez 2009; Hatchondo et al. 2015; Niepelt 2014).
- Key implication: the common assumption of symmetric haircuts across maturities is not supported by the data; models of sovereign debt maturity composition should account for systematic maturity-dependent haircuts driven by the term structure and time-variation of default risk.

*Source: wp17119 - References (excerpts from the chapter content provided).*

### Appendix II reports two additional episodes: Belize 2006–2007 (external) and Saint Kitts and

### wp17119 - Appendix II reports two additional episodes: Belize 2006–2007 (external) and Saint Kitts and Nevis 2011–2012 (external)

### Recovery rates, maturity, and instrument detail
- Stylized fact: recovery rates on short-term bonds are substantially smaller than those on longer-term bonds across multiple episodes (examples include Russia 1998–2000 external, Greece 2011–2012 external, Uruguay 2003 domestic, Cyprus 2013 domestic).
- Robustness: results hold across debt type (external vs. domestic), restructuring strategy (preemptive vs. post-default), and exchange method (single-menu vs. multi-menu exchanges). Appendix III shows robustness to modified SZ recovery rate measures using exit yield +/-1 percent.
- Alternative recovery measure: the authors obtain the same qualitative finding using their alternative recovery rate 푅푡,푇푖. Figure 2 shows 푅푡,푇푖 by maturity for Pakistan 1999 (external) and Cyprus 2013 (domestic). Appendix II adds Greece 2011–2012 (external) and Uruguay 2003 (domestic). Figures A3 and Table A2 show robustness evaluating measures 6 months before and after announcements.
- Quantitative summary (from Table 2 and text):
  - On average, the recovery rate on a 10-year bond is 3–12 percentage points higher than on a 1-year bond.
  - Instruments with high fixed coupon rates or with floating rates experience smaller recovery rates.
  - The effect of amortization profile is not precisely estimated.
  - Due to sample sizes, the effect of restructuring duration is identified only in specification (2) and is positive and significant.
- Sample and regression details (Table 2):
  - Number of Countries: 15 (columns 1–2); 12 (columns 3–4)
  - Number of Restructurings: 28 (columns 1–2); 16 (columns 3–4)
  - Observations: 449 (column 1), 447 (column 2), 111 (columns 3–4)
  - Adjusted R-Squared: 0.035 (col 1), 0.262 (col 2), 0.112 (col 3), 0.247 (col 4)
  - Example reported coefficients (column headers: SZ Recovery Rate, SZ Recovery Rate with controls, Exchange Recovery Rate (announcement), Exchange Recovery Rate with controls): 0.30*** (0.08), 0.35*** (0.08), 1.03*** (0.30), 1.24*** (0.29). Other coefficients and standard errors are reported in the table.

### Bond price dynamics and convergence prior to exchange
- Stylized fact 2: The difference in prices between short-term and longer-term bonds decreases in the run-up to the debt exchange.
- Evidence: Figures report bond price differentials (short minus long) for Ukraine 2000 external and Greece 2011–2012 external; Appendix IV adds Dominican Republic 2004–2005 external (bonds) and Argentina 2001–2005 external. Grey bars mark announcement dates; prices converge toward the exchange.
- Econometric results (Table 3 and text):
  - Main result: price of short-term debt approaches price of longer-term debt from above during run-up; significant at the 1-percent level.
  - Quantitatively: over 6 months the price difference diminishes by 9 percentage points (relative to face value of 100).
  - Maturity differential effect: a 1-year vs 10-year example — the price of a 1-year bond exceeds the price of a 10-year bond by 4 percentage points (conditional on time).
  - Decomposition: restructuring strategy (strictly preemptive vs post-default) and type of debt (external vs domestic) significantly explain variation; Argentina 2001–2005 appears an outlier with larger bond price differential ceteris paribus.
  - Speed of convergence similar whether exchange involves a single menu or different menus (see Table A3 in Appendix IV).
- Sample and regression details (Table 3):
  - Number of Restructurings: 12
  - Number of Pairs of Instruments: 66
  - Observations: 605
  - Adjusted R-Squared: 0.224 (col 1), 0.429 (cols 2–3)
  - Example reported coefficients (Bond Price Differential regressions): Time span until exchange coefficient -1.52*** (0.08) across specifications; Maturity Differential (S - L, year) coefficient -0.49*** (0.05) across specifications. Other coefficients and SEs shown in the table.

### Asset-pricing model: intuition and formal results
- Setup:
  - Discrete-time model with bonds maturing at date m paying coupon c_m if no restructuring, or recovery value φ_t^m if issuer announces restructuring.
  - Probability of no restructuring between t and s (s>t), given survival until t, equals π_t,s.
  - Investors risk neutral; risk-free gross interest rate equals β^{-1}. Ex-coupon price p_t^m satisfies an asset-pricing identity (equation (5) in text).
- First result:
  - With identical coupons and recovery values across bonds, long-term restructuring risk implies current price difference Δ_t^{s,l} ≡ p_t^s − p_t^l > 0: short-term bonds are more expensive than longer-term bonds (equation (6)).
  - Intuition: expected future positive price difference and long-term exposure to default risk support current positive price difference absent coupon or recovery value offsets.
- Second result:
  - Over time, Δ_t^{s,l} evolves because time passes and because news about future restructuring risk arrives (equations (7) and (8)).
  - Absent new information, price differences tend to diverge further (Δ_{t+1} − Δ_t > 0). If new information raises short-term default risk sufficiently, Δ may fall — i.e., prices converge as short-term default risk rises in the run-up to exchange.
  - Interpretation: selected episodes leading to restructuring are characterized by rising short-term default risk; model predicts positive but decreasing Δ over time, explaining higher haircuts on shorter-term debt and weaker maturity-haircut relation closer to exchange.

### Default probability term structure and short-term risk
- Stylized fact 3: The term structure of default probability becomes elevated at the short end in the run-up to the exchange.
- Estimation approach:
  - Based on Ranciere (2002): inputs are (a) CDS spread curve, and (b) US Treasury bill yield curve as risk-free proxy.
  - Assumptions: no arbitrage, risk-neutrality, exchange involves a single menu of new instruments.
  - Construct CDS spread curve for horizons 1–120 months (monthly frequency) via annual yield curve data with linear interpolation; derive forward default spread curve (one-month ahead conditional default risk) and implied one-month ahead conditional default probability (Appendix V details).
- Evidence:
  - Figure 4 reports conditional default probability estimates at the 1-year horizon for Ecuador 1999–2000 external and Greece 2011–2012 external; Appendix VI shows Argentina 2001–2005 external. Grey bars mark announcement dates; conditional short-term default probability (1 year) increases in run-up to exchange.
  - This rise is observed also at the 2-year and 3-year horizons (Table A4 in Appendix VI) and is not specific to restructuring strategy.
- Econometric results (Table 4 and text):
  - Main result: conditional default probability increases in the run-up to the exchange; significant at the 1-percent level.
  - Quantitatively: over 6 months the default risk increases by 7–16 percentage points.
  - Decomposition: restructuring strategy (strictly preemptive vs post-default) is a significant explanatory variable. Duration of restructuring reduces conditional default probability (possibly because default probability gradually increases for restructurings with longer duration).
- Sample and regression details (Table 4):
  - Number of Restructurings: 3
  - Number of Time Periods (months): 12
  - Observations: 36
  - Adjusted R-Squared: 0.373 (col 1), 0.628 (cols 2–3)
  - Example reported coefficients (Default Probability 1-Year Horizon regressions): Time Span until the Exchange Date coefficient 1.09*** (0.26) (col 1), 2.70*** (0.37) (cols 2–3). Other coefficients and SEs reported in the table.

### Conclusion and implications
- Main conclusion: creditors are not treated equally in sovereign debt exchanges; shorter-term instruments tend to receive larger haircuts than longer-term instruments.
- Supporting chain: rising short-term default risk → higher prices for short-term instruments pre-run-up and convergence as exchange approaches → larger ex-post haircuts on shorter-term instruments when measured following common approaches.
- Potential implications (authors note further research required before normative conclusions):
  - Relevance to pari passu interpretations and limitations.
  - Negotiation dynamics during debt restructuring episodes.
  - Value-at-risk calculation for debt securities of different maturities and implications for financial institutions holding these instruments.
  - Optimal debt management considerations.

*Source: IMF Working Paper wp17119 (Appendices and main text excerpts provided).*

### Appendix I. Dataset: Selected Recent Restructurings

### Appendix I. Dataset: Selected Recent Restructurings

### Selected Recent Sovereign Debt Restructurings (1998–2015) — Key Entries
- External restructurings (examples with exact fields preserved):
  - Pakistan: Episode Start Jan-99, Exchange (End) Dec-99, Restructuring Strategy Strictly Preemptive, SZ Haircut Differential (S- and L-term, percent) 25.7, Maturity differential: 0 and 2.5
  - Russia (PRINs & IANs): Episode Start Nov-98, Exchange Aug-00, Restructuring Strategy Post-Default, SZ Haircut Differential 19.2, Maturity differential: -0.1 and 10.9
  - Ecuador: Episode Start Jan-99, Exchange Aug-00, Restructuring Strategy Post-Default, SZ Haircut Differential 28.2, Maturity differential: 1.7 and 24.5
  - Uruguay (external): Episode Start Mar-03, Exchange May-03, Restructuring Strategy Strictly Preemptive, SZ Haircut Differential 8.4, Maturity differential: 0.5 and 24.2
  - Argentina (Global Exchange): Episode Start Nov-01, Exchange Jun-05, Restructuring Strategy Post-Default, SZ Haircut Differential 13.4, Maturity differential: 0.1 and 22.6
  - Grenada: Episode Start Oct-04, Exchange Nov-05, Restructuring Strategy Weakly Preemptive, SZ Haircut Differential 21.4, Maturity differential: 0 and 13.0
  - Belize: Episode Start Aug-06, Exchange Feb-07, Restructuring Strategy Weakly Preemptive, SZ Haircut Differential 22.4, Maturity differential: 0 and 8.0
  - Greece: Episode Start Jul-11, Exchange Mar-12, Restructuring Strategy Preemptive, SZ Haircut Differential 45.4, Maturity differential: 0 and 28.5
  - Saint Kitts and Nevis: Episode Start Jun-11, Exchange Apr-12, Restructuring Strategy Weakly Preemptive, SZ Haircut Differential 10.3, Maturity differential: 0 and 8

- Domestic restructurings (examples with exact fields preserved):
  - Russia (GKOs resident): Episode Start Aug-98, Exchange May-99, Restructuring Strategy Post-Default, SZ Haircut Differential 2.4, Maturity differential: -0.3 and 0.3
  - Argentina (Phase I): Episode Start Oct-01, Exchange Nov-01, Restructuring Strategy Strictly Preemptive, SZ Haircut Differential 7.8, Maturity differential: 0.5 and 29.6
  - Uruguay (domestic): Episode Start Mar-03, Exchange May-03, Restructuring Strategy Strictly Preemptive, SZ Haircut Differential 10.4, Maturity differential: 0.0 and 8.8
  - Jamaica: Episode Start Feb-13, Exchange Mar-13, Restructuring Strategy Strictly Preemptive, SZ Haircut Differential 0.9, Maturity differential: 0 and 33.2
  - Cyprus: Episode Start May-13, Exchange Jul-13, Restructuring Strategy Strictly Preemptive, SZ Haircut Differential 14.7, Maturity differential: 1 and 2.7

### Appendix II — Haircuts/Recovery Rates: SZ Recovery and Maturity (selected notes)
- Figures reported: SZ Recovery and Maturity for selected cases including:
  - Belize 2006–2007 External
  - Saint Kitts and Nevis 2011–2012 Domestic
- Figure A2 examples:
  - Greece 2011–2012 External
  - Uruguay 2003 Domestic
- Figure A3 presents Exchange Recovery and Maturity at different points in time for:
  - Pakistan 1999 External — 6 months before and 6 months after the announcement
  - Greece 2011–2012 External — 6 months before and 6 months after the announcement

### Table A2 — Cross-sectional Regression Results (Exchange Recovery Rate)
- Dependent variable: Exchange recovery rate (in %). Main explanatory variable: Maturity of instrument at the time of exchange (years).
- Columns and reported estimates (coef/se) — exact values preserved:
  - Column (1): coef 0.54** (0.24)
  - Column (2): coef 0.54** (0.24)
  - Column (3): coef 5.74*** (1.74)
  - Column (4): coef 5.79*** (1.81)
  - Additional coefficients reported (exact pairs preserved):
    - -0.76 (0.64)
    - 0.86 (4.95)
    - -11.31 (9.18)
    - 4.53 (74.22)
    - -13.61* (7.12)
    - 33.03 (56.42)
    - 68.70*** (1.89)
    - 76.68*** (5.64)
    - 80.04*** (15.31)
    - 71.25 (42.88)
- Regression design and sample:
  - Episode Fixed Effects: Yes (all columns)
  - Number of Countries: 12 (all columns)
  - Number of Restructurings: 16 (all columns)
  - Observations: Column 1: 104; Column 2: 104; Column 3: 111; Column 4: 111
  - Adjusted R-Squared: Column 1: 0.054; Column 2: 0.118; Column 3: 0.104; Column 4: 0.108
- Notes:
  - Columns 1 and 2: Exchange recovery rate 6 months before the announcement.
  - Columns 3 and 4: Exchange recovery rate 6 months after the announcement.
  - Control variables: instrument-specific controls and restructuring-specific controls.
  - Significance: *** p<0.01, ** p<0.05, * p<0.10.
  - All regressions include restructuring-specific fixed effects. Robust standard errors in parentheses.

### Appendix III — SZ Haircuts Robustness Check (method and outcomes)
- Method:
  - Recompute SZ haircuts using exit yield adjustments: exit yield +1 percent and exit yield −1 percent.
- Reported outcome (examples):
  - Figure A4: Dominica 2003–2004 Domestic — SZ recovery rate with exit yield +1 percent and exit yield −1 percent.
  - Figure A5: Belize 2006–2007 External — SZ recovery rate with exit yield +1 percent and exit yield −1 percent.
- Key empirical statement (preserved text):
  - "In both cases, SZ recovery rates on short-term debt are lower than those on long-term debt whichever yield is used."

### Appendix IV — Bond Prices; Table A3 Panel Regression Results (Bond Price Differential)
- Dependent variable: Bond price differential (price of short-term bond − price of long-term bond).
- Main explanatory variables: Time span until the exchange date (months) and maturity differential (S − L, years).
- Columns and reported estimates (coef/se) — exact values preserved:
  - Column (1): coef -1.42*** (0.09)
  - Column (2): coef -1.43*** (0.10)
  - Column (3): coef -1.79*** (0.15)
  - Column (4): coef -1.81*** (0.15)
  - Additional coefficients and exact pairs:
    - -0.51*** (0.05)
    - -0.40*** (0.05)
    - 0.34 (0.24)
    - 0.81*** (0.22)
    - -37.95*** (2.41)
    - -29.37*** (4.25)
    - -114.63*** (7.23)
    - 19.75*** (2.02)
    - -15.74*** (5.98)
    - 17.43*** (1.56)
    - 48.69*** (5.11)
    - 7.81** (3.70)
    - 127.09*** (7.72)
    - 1.05 (5.46)
    - -18.02*** (2.74)
- Regression design and sample:
  - Episode Fixed Effects: No (all columns)
  - Number of Restructurings: 6 (all columns)
  - Number of Pairs of Instruments: 5 (columns 1–2), 15 (columns 3–4)
  - Observations: Columns 1–2: 487; Columns 3–4: 118
  - Adjusted R-Squared: Column 1: 0.190; Column 2: 0.403; Column 3: 0.289; Column 4: 0.787
- Notes:
  - Columns 1 and 2: restructurings with exchanges involving a single menu of new instruments.
  - Columns 3 and 4: restructurings with exchanges involving different menus of new instruments.
  - Control variables: episode-specific controls.
  - Significance: *** p<0.01, ** p<0.05, * p<0.10. Robust standard errors in parentheses.

### Appendix V — Estimation of Term Structure of Default Risk (methodology)
- Overview: Follow Ranciere (2002). Three-step procedure to estimate term structure of conditional default probabilities and implied recovery.
- 1st Step: Constructing the monthly yield curve
  - Linear interpolation formulae (preserved exactly):
    - CDS_t^M,j = CDS_t^A,i + (j−i)/12 (CDS_t^A,i+12 − CDS_t^A,i)
    - RF_t^M,j = RF_t^A,i + (j−i)/12 (RF_t^A,i+12 − RF_t^A,i)
    - for i = 0,12,24,36,...,108 and j = i+1, i+2,...., i+11
    - Annual yield for 0-year horizon assumed zero: CDS_t^A,0 = RF_t^A,0 = 0.
- 2nd Step: Computing a forward default spread curve
  - No arbitrage condition (preserved exactly):
    - (1 + RF_t^M,h,j + CDS_t^M,h,j) = (1 + RF_t^M,j + CDS_t^M,j) / (1 + RF_t^M,h + CDS_t^M,h)
    - for j = 1,2,3,....,120 and h = j−1
  - Forward default spread measures the implied one-period ahead conditional default risk.
- 3rd Step: Computing the term structure of conditional default probabilities
  - Risk-neutral relation (preserved exactly):
    - (1 + RF_t^M,h,j) = (1 − P̂_t^h,j) * (1 + RF_t^M,h,j + CDS_t^M,h,j) + P̂_t^h,j * R̂_t
    - subject to R̂_t = argmin_{R̂_t} (1/n) ∑ { p_t^i − p̂_t^i(...) }^2_{i=1}
    - for j = 1,2,3,....,120 and h = j−1
  - Where P̂_t^h,j is implied one-period ahead conditional default probability and R̂_t is time-variant recovery.
  - Implied time-variant recovery estimated to minimize mean squared deviations of estimated bond prices p̂_t^i from observed price p_t^i.
  - "This implied time-variant recovery converges to the observed nominal recovery at the time of exchange as reported in Figure A7."
- Figures referenced:
  - Figure A7: Implied Nominal Recovery Rate for Ecuador 1999–2000 External and Greece 2011–2012 External.

### Appendix VI — Default Probability Term Structure and Table A4 (Panel Regression Results)
- Figure A8: Default Probability Term Structure example: Argentina 2001–2005 External.
- Table A4 — Panel regressions (Dependent variable: conditional default probability at specified horizons)
  - Columns and reported estimates (coef/se) — exact values preserved:
    - Column (1) Default Probability 2-Year Horizon (%): 0.87*** (0.22)
    - Column (2) Default Probability 2-Year Horizon with controls: 2.16*** (0.31)
    - Column (3) Default Probability 3-Year Horizon (%): 0.68*** (0.21)
    - Column (4) Default Probability 3-Year Horizon with controls: 1.87*** (0.30)
    - Additional coefficients and exact pairs:
      - -1.86 (1.19)
      - -12.01*** (2.23)
      - -1.40 (1.14)
      - -10.72*** (2.19)
      - -30.69*** (6.17)
      - -28.20*** (6.06)
      - 52.56*** (3.83)
      - 96.28*** (9.27)
      - 48.28*** (3.65)
      - 88.45*** (9.09)
  - Regression design and sample:
    - Episode Fixed Effects: No (all columns)
    - Number of Restructurings: 3 (all columns)
    - Number of Time Periods (months): 12 (all columns)
    - Observations: 36 (all columns)
    - Adjusted R-Squared: Column 1: 0.336; Column 2: 0.606; Column 3: 0.252; Column 4: 0.543
  - Notes:
    - Main explanatory variable: Time span until the exchange date (months).
    - Columns 1 and 2: 2-year horizon results; Columns 3 and 4: 3-year horizon results.
    - Control variables: episode-specific controls.
    - Significance: *** p<0.01, ** p<0.05, * p<0.10. Robust standard errors in parentheses.

- Reported covariates in tables (preserved labels):
  - Time Span until the Exchange Date (months)
  - Duration of Restructuring (years)
  - Strictly Preemptive Restructuring (dummy)
  - Other episode- and instrument-specific controls cited across appendices.

*Source: wp17119 - Appendix I. Dataset: Selected Recent Restructurings (IMF Working Paper appendices).*

---


_Source: https://www.imf.org/-/media/files/publications/wp/2017/wp17119.pdf_