## wpiea2021155-print-pdf

## Source details

**Canonical URL:** [wpiea2021155-print-pdf](https://www.imf.org/-/media/files/publications/wp/2021/english/wpiea2021155-print-pdf.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/wp/2021/english/wpiea2021155-print-pdf.pdf.md)
- [Structured JSON version](/-/media/files/publications/wp/2021/english/wpiea2021155-print-pdf.pdf.json)

---

### Rationale for studying bond yields rather than credit spreads
- Individual bond spreads are not directly observable and must be estimated by fitting zero-coupon corporate yield curves, which:
  - limits the sample to firms with multiple bond issuances, and
  - introduces measurement error that biases estimation.
- CDS premiums are difficult to obtain for a large emerging-market sample (examples cited: 200 firms in Harjes (2011); 36 banks in Acharya, Drechsler, and Schnabl (2014); 118 firms in Bedendo and Colla (2015); 226 firms in Augustin et al. (2018); 54 banks in Li and Zinna (2018); Bai and Wei (2012) consider >2,000 firms but only 267 from emerging economies).
- Bond yields permit inclusion of risk-free rates (proxied by U.S. Treasury yields) as a control to mitigate simultaneous influence of risk-free rates on corporate and sovereign yields.
- Results are robust to replacing bond yields with credit spreads (reported in section 4.2).

### Identification strategy and controls
- Controls included at bond-, country-, and global-levels to mitigate omitted-variable concerns and help isolate causal effects.
- Potential reverse causality and robustness to it are discussed in sections 3.2 and 4.2.
- Cross-sectional means of y and x (ȳ_t and x̄_t) are included to account for unobserved common factors f_t in the CS-DL specification.

### Data (coverage, cleaning, and limitations)
- Data source and coverage:
  - Corporate-bond pricing: IHS Markit weekly data between March 2016 and April 2019 (160 weeks).
  - Linked to Bloomberg and Datastream via ISIN for country of risk, industry, issue currency, credit ratings, issuers’ financial data.
  - Sovereign bonds: weekly individual sovereign bonds from Bloomberg (BVAL).
- Sample selection and final sample after cleaning:
  - Kept only emerging-economy country-of-risk bonds and only U.S.-dollar corporate bonds.
  - Bond credit ratings from S&P, Moody’s, Fitch assigned numeric values 1 (AAA) to 22 (defaulted).
  - Final sample: 2,045 U.S.-dollar corporate bonds issued by 1,040 companies; 431 U.S.-dollar sovereign bonds; 46 emerging economies (country must have at least one corporate and one sovereign bond surviving cleaning).
- Data-cleaning steps (applied in order):
  1. Drop defaulted, perpetual, inflation-adjusted bonds.
  2. Drop bonds with maturity < 1 year.
  3. Trim top/bottom 0.5 percent by mid-yields and bid-ask yield spreads; drop non-positive yields.
  4. Drop bonds with < 30 weeks of consecutive data (linear interpolation used to fill gaps; ~5.9 percent of sample prior to this step).
  5. Drop duplicated Rule 144A if Regulation-S counterpart present.
- Notes and limitations:
  - Markit data introduced in 2016; earlier data not used.
  - Weekly frequency used to mitigate measurement error from illiquidity despite daily data availability.
  - Local-currency bonds excluded due to expected depreciation effects that are difficult to measure.
  - Quarterly issuer financial data available for 17 percent of the sample.
  - Conclusions may better apply to emerging economies and to countries where U.S.-dollar bonds are important in financing.

### Empirical models and estimation approach
- Pairing strategy:
  - Each corporate bond i from country j (yield y_{i,j,t}) paired with x_{j,t}, the equal-weighted average yield on sovereign portfolio of country j on same date.
  - Alternative matching by closest maturities yields significant pass-through with smaller magnitude (section 4.2).
- Modeling frameworks:
  - Conditional ARDL panel with unobserved common factors f_t:
    - y_{i,t} = sum_{l=1}^{p_yi} phi_{i,l} y_{i,t-l} + sum_{l=0}^{p_xi} beta_{i,l} x_{i,t-l} + u_{i,t}
    - u_{i,t} = gamma_i' f_t + epsilon_{i,t}
    - Long-run impact theta_i = (sum_{l=0}^{p_xi} beta_{i,l}) / (1 - sum_{l=1}^{p_yi} phi_{i,l})
  - Two estimation approaches for theta_i:
    - ARDL bond-by-bond: consistent under exogeneity/endogeneity of x but requires correct lag orders and stationarity assumptions.
    - Distributed Lag / Cross-Sectionally Augmented DL (DL / CS-DL): truncates lag polynomial and augments with cross-sectional means to control for f_t; allows unit roots and correlation between x and f_t but requires exogeneity of x.
- Baseline estimator and specification:
  - Baseline: CS-DL approach motivated by unit-root presence in yields.
  - Baseline regression (bond-specific, estimated by OLS bond-by-bond) includes theta_i x_{i,t}, lagged differences of x, lags of cross-sectional means ȳ and x̄, bond fixed effects, and observed controls Z_{i,t}.
  - Lag choices: p = integer part of T^{1/3}; baseline p = 4 (robustness with p = 3, 5). Sample T̄ = 121, T_min = 30, T_max = 160.
  - Aggregation: mean-group estimator theta_hat_MG = N^{-1} sum_{i=1}^N theta_hat_i with nonparametric standard errors.
- Robustness: ARDL estimates also computed; authors argue plausibility of sovereign yields x exogenous to corporate yields y in emerging markets but report ARDL results to allow feedback.

### Full-sample empirical results — baseline and trimmed samples
- Baseline (Table 2, column (3) preferred):
  - Positive and highly statistically significant association between corporate bond yields and sovereign bond yields.
  - Long-run pass-through around one.
  - Full-sample estimate: Sovereign yields → Corp. yields = 0.951*** (standard error 0.111); US 10-year Treasury yields coefficient = -0.147** (standard error 0.0665).
  - Observations 190,796; Number of firm bonds 1,987; Number of countries 31; R-squared (mean-group) 0.914.
- Trimmed-full sample (top/bottom 5 percent of individual thetâi removed; 200 bonds dropped):
  - Trimmed-full sample size: 1,787 bonds; Observations 177,890.
  - Main specification (Table 3, column (3)): Sovereign yields → Corp. yields = 0.824*** (standard error 0.0356).
  - Other significant controls: Maturity = 0.197*** (standard error 0.0441); Real effective exchange rate growth = 0.0112*** (standard error 0.00231).
  - R-squared (mean-group) 0.926; Time trend -0.00270*** (standard error 0.000861).
- Individual pass-throughs:
  - Full-sample distribution: Mean = 0.951; median = 0.595; standard deviation = 4.959.
  - Trimmed-full-sample distribution: Mean = 0.824; median = 0.581; standard deviation = 1.506.
- Note: A small fraction of bonds (274 bonds) have negative and significant pass-throughs; >80 percent of these are issued by companies in China. Dropping these 274 bonds does not materially change results.

### Robustness exercises and alternative specifications
- Matching methods and additional controls (Table 4):
  - Alternative maturity matching (MAD, MSD) and matching including ratings and bid-ask spreads yield similar and highly significant pass-throughs, slightly smaller in magnitude than baseline.
  - Additional macro and issuer controls (real GDP growth, CPI, real GDP per capita, current account balance; issuer financials: profit margin, leverage, asset tangibility, interest coverage) produce similar estimates.
  - Sample composition checks: dropping China increases pass-through; keeping only fixed-coupon, senior-unsecured, no-embedded-option bonds yields similar results.
  - Lag specification checks: p = 3 and p = 5 produce similar estimates.
  - Alternative estimation frameworks:
    - Panel ARDL (mean-group) and synthetic time-series ARDL produce pass-throughs similar to baseline. Synthetic ARDL iterations yield pass-through range 0.87 to 1.02; standard error range 0.039 to 0.042.
- Range of mean-group long-run pass-throughs across 12 robustness checks: coefficients range from 0.721*** to 1.059*** (example: MAD maturity 0.721*** (standard error 0.0588); Drop China 1.059*** (standard error 0.0473)).
- Robustness to using spreads:
  - Replacing yields with Z spreads: long-run pass-through from sovereign spreads to corporate spreads = 0.831*** (standard error 0.156), close to 0.824 from yields.
- Additional robustness controls produce similar results (monthly dummies, excess bond premium, Moody’s U.S. Aaa/Baa yields, oil price growth, alternative risk-free rates U.S. 3-, 5-, or 7-year Treasuries).

### Transmission channels — evidence for transfer-risk and liquidity-premium
- Transfer-risk channel (section 5.1):
  - Classification:
    - Investment-grade (IG) if country mean sovereign bond average rating ≥ BBB-.
    - Public external debt / foreign reserves: countries split by pooled median into low-risk (mean < pooled median) and high-risk (mean ≥ pooled median). Resulting groups (for 31 countries with MSCI data): low-risk = 15 countries; high-risk = 16 countries.
  - Key findings (Table 5):
    - Pass-through in high-risk countries almost doubles that in low-risk countries.
    - IG countries: sovereign yields coefficient = 0.591*** (standard error 0.0984); Observations 121,433; corporate bonds 1,218.
    - Non-IG countries: sovereign yields coefficient = 1.078*** (standard error 0.0632); Observations 56,457; corporate bonds 569.
    - Low debt/reserves: sovereign yields = 0.614*** (standard error 0.0604); High debt/reserves: sovereign yields = 1.296*** (standard error 0.0989).
  - Financial firms vs non-financial and government-related bonds:
    - Financial bonds: sovereign yields = 1.197*** (standard error 0.0980).
    - Non-financial bonds: sovereign yields = 0.678*** (standard error 0.0421).
    - Government-related bonds: sovereign yields = 0.474*** (standard error 0.0470).
    - Non-related bonds: sovereign yields = 0.882*** (standard error 0.0463).
    - Double-sort: within non-financial sector, government-related bonds have higher pass-through; within financial sector, non-government-related banks have higher pass-through. Government-related financial companies show small/insignificant pass-through partly due to Chinese banks.
  - Robustness excluding China: dropping Chinese bonds increases pass-throughs; government-related banks remain more resilient than non-related banks.
- Liquidity-premium channel (section 5.2):
  - Sovereign bond liquidity proxied by bid-ask yield spreads; countries split by time-series mean of average sovereign bid-ask spread relative to panel median -> LBA (15 countries) and HBA (16 countries).
  - Double-sort by IG status and LBA/HBA yields four subsamples.
  - Key findings (Table 6):
    - IG & LBA: sovereign yields = 0.496*** (standard error 0.182); Observations 50,007; corporate bonds 442.
    - IG & HBA: sovereign yields = -0.965 (standard error 0.643); Observations 71,426; corporate bonds 776 (insignificant).
    - Non-IG & LBA: sovereign yields = 1.411*** (standard error 0.164); Observations 22,659.
    - Non-IG & HBA: sovereign yields = 1.181*** (standard error 0.0945); Observations 33,798.
  - Interpretation: sensitivity of corporate yields to sovereign yields is smaller where sovereign bid-ask spreads are wider (sovereign bonds less liquid), consistent with liquidity-premium channel; transfer-risk channel confirmed by stronger pass-through in lower sovereign-rating groups irrespective of liquidity grouping.

### Heterogeneity by corporate credit ratings and maturities; nonlinearity
- Credit ratings and maturities (Table 7 and section 6.1):
  - Classification:
    - Corporate bonds IG vs non-IG based on bond time-series mean rating relative to BBB-.
    - Maturity groups: Short-term 1–5 years; Mid-term 5–10 years; Long-term >10 years.
  - Findings (Table 7):
    - IG corporate bonds: sovereign yields = 0.548*** (standard error 0.0546); Observations 92,196; bonds 852.
    - Non-IG corporate bonds: sovereign yields = 1.023*** (standard error 0.0552); Observations 85,694.
    - Short-term corporate bonds: sovereign yields = 0.935*** (standard error 0.0574); Observations 100,526.
    - Medium-term: sovereign yields = 0.816*** (standard error 0.0550); Observations 54,453.
    - Long-term: sovereign yields = 0.488*** (standard error 0.0825); Observations 22,911.
  - Interpretation: pass-through larger for riskier (non-IG) and shorter-maturity bonds; default-premium convexity and declining relevance of transfer effects over long horizons (beyond 5–10 years) explain patterns.
- Nonlinearity (section 6.2):
  - Testing squared sovereign yield term (Table 8):
    - Squared term significant only when control variables omitted; with controls squared term insignificant.
    - In specifications where squared term significant, relation is increasing and convex (pass-through larger when sovereign yields higher), aligning with transfer-risk evidence rather than concave information-provision channel observed in primary issuance literature.

### Panel unit-root and cointegration evidence
- Panel cointegration tests (Appendix A1) — Modified Phillips-Perron, Modified Dickey-Fuller, and Dickey-Fuller tests: p-values 0.00 for Lags 1–5, indicating rejection of no cointegration.
- Phillips–Perron unit-root tests for individual bonds (Appendix A3) indicate a majority of bonds contain unit roots (robust to various lag choices and trend inclusion).

### Summary statistics and sample coverage (Table 1, Appendix A2)
- Summary statistics (selected):
  - Mid yields (%) — #Obs. 263,103; Mean 5.57; St.dev. 3.69; p25 3.77; p50 4.78; p75 6.31.
  - Average ratings (1=AAA, 22=defaulted) — #Obs. 263,103; Mean 11.17; St.dev. 4.39; p25 8; p50 10.33; p75 13.67.
  - Number of years to maturity — #Obs. 263,103; Mean 7.91; St.dev. 8.71; p25 3.02; p50 4.99; p75 8.74.
  - Bid-ask yield spread (%) — #Obs. 263,103; Mean 0.19; St.dev. 0.3; p25 0.06; p50 0.11; p75 0.21.
  - US 10-year Treasury yields (%) — #Obs. 205,140; Mean 2.5; St.dev. 0.45; p25 2.26; p50 2.55; p75 2.87.
- Cross-sectional counts (Appendix A2):
  - Total sovereign bonds: 431; Total corporate bonds: 2,045.
  - Government-related corporate bonds: 1,384; Non-related corporate bonds: 661.
  - Examples: China corporate bonds 862 (government-related 335; non-related 527); Brazil corporate bonds 191 (government-related 35; non-related 156).

### Key quantitative results (high-level)
- Baseline long-run sovereign-to-corporate pass-through:
  - Full sample preferred: 0.951*** (standard error 0.111).
  - Trimmed-full sample preferred: 0.824*** (standard error 0.0356).
- Robustness range across 12 checks: coefficients from 0.721*** to 1.059***.
- Credit-spread specification: sovereign spreads → corporate spreads = 0.831*** (standard error 0.156).
- Heterogeneity highlights:
  - Non-IG corporate bonds: 1.023*** (standard error 0.0552).
  - IG corporate bonds: 0.548*** (standard error 0.0546).
  - Short-term (1–5 years): 0.935*** (standard error 0.0574); Long-term (>10 years): 0.488*** (standard error 0.0825).
  - High public external debt / foreign reserves: 1.296*** (standard error 0.0989); Low: 0.614*** (standard error 0.0604).

### Conclusions and policy implications
- Dataset: covers more than 2,400 U.S.-dollar-denominated bonds in 46 emerging markets.
- Main empirical conclusion: on average, the long-run pass-through from sovereign yields to corporate yields is around one and highly significant.
- Both transfer-risk and liquidity-premium channels operate and explain heterogeneity across countries and bonds:
  - Transfer-risk: stronger pass-through in lower-rated and higher public-debt countries; riskier and short-maturity corporate bonds more vulnerable.
  - Liquidity-premium: stronger pass-through where sovereign bonds are more liquid (narrower bid-ask spreads).
- Policy implication: fiscal distress raising sovereign yields can persistently increase private-sector borrowing costs and potentially impede growth for an extended period.
- Limitations and future research:
  - Other mechanisms may link corporate and sovereign yields; subsample evidence consistent with transfer-risk and liquidity-premium channels but does not exclude other channels.
  - Direct study of default and liquidity risks via CDS, issuers’ financial positions, and transaction prices/volumes would be valuable but data are limited for emerging-market bonds.

*Source: wpiea2021155-print-pdf - sections 2–7 and References; https://www.imf.org/-/media/files/publications/wp/2021/english/wpiea2021155-print-pdf.pdf*

### section  2  why  we  choose  to  study  bond  yields  rather  than  credit  spreads.  We  further  show  in  section  4.

### section  2  why  we  choose  to  study  bond  yields  rather  than  credit  spreads.  We  further  show  in  section  4.

### Rationale for studying bond yields rather than credit spreads
- Individual bond spreads are not directly observable; they must be numerically estimated by fitting zero-coupon corporate yield curves (see Durbin and Ng (2005) and Peter and Grandes (2005)), which:
  - limits the sample to firms with multiple bond issuances sufficient for estimating yield curves, and
  - inevitably introduces measurement error in the resulting credit spreads that will bias estimation results.
- CDS premiums are difficult to obtain for a large sample of emerging-market firms; CDS-based studies often have relatively small samples (examples cited):
  - 200 firms in Harjes (2011)
  - 36 banks in Acharya, Drechsler, and Schnabl (2014)
  - 118 firms in Bedendo and Colla (2015)
  - 226 firms in Augustin, Boustanifar, Breckenfelder, and Schnitzler (2018)
  - 54 banks in Li and Zinna (2018)
  - Bai and Wei (2012) consider more than 2,000 firms in 30 different countries, but only 267 firms are from emerging economies.
- Bond yields permit inclusion of risk-free rates (proxied by U.S. Treasury yields) as a control variable to mitigate concerns that risk-free rates can simultaneously influence corporate and sovereign yields.
- The paper shows results are robust to replacing bond yields with credit spreads in section 4.2.

### Identification strategy and controls
- The analysis controls for factors that may simultaneously drive corporate and sovereign bond yields to help isolate causal effects.
- Control variables are included at bond-, country-, and global-levels that are common in previous empirical studies to further mitigate omitted-variable concerns.
- The paper discusses potential reverse causality and how the method is robust to it in sections 3.2 and 4.2.

### Evidence on transmission channels: transfer-risk and liquidity-premium
- Subsample analysis examines the transfer-risk channel and the liquidity-premium channel and finds evidence that they co-exist.
- Transfer-risk channel predictions and findings:
  - Prediction: pass-through from sovereign to corporate yields is stronger in countries with greater sovereign risks—where governments are more likely to transfer risks to domestic firms.
  - Findings:
    - The estimated pass-through is larger in countries with lower sovereign credit ratings.
    - The estimated pass-through is larger in countries with higher public external debt (relative to foreign reserves).
    - Government-related firms (regardless of industry) and banks holding government debt are expected to be more exposed; results support this to a large extent except that government-related banks are more resilient to sovereign risk transmission than banks without government relations. This contrasts with the notion of “moral suasion” but is consistent with Acharya, Eisert, Eufinger, and Hirsch (2018) for European advanced economies.
- Liquidity-premium channel predictions and findings:
  - Prediction: the bond-yield association between corporates and sovereigns is weaker if sovereign bonds are less liquid.
  - Method: double-sort sample based on sovereign bonds’ credit ratings and bid-ask yield spreads (lower-rated sovereign bonds are naturally less liquid).
  - Findings:
    - The sensitivity of corporate yields to sovereign yields is smaller in countries where sovereign bid-ask spreads are wider, i.e., where sovereign bonds are less liquid; this holds regardless of the country’s credit rating.
    - Conditional on either high or low sovereign bid-ask spreads, the pass-through is always stronger in countries where sovereign credit ratings are lower, confirming the transfer-risk mechanism again.
- Conclusion: Both transfer-risk and liquidity-premium channels exist and contribute to the observed corporate-sovereign nexus.

### Heterogeneity by corporate credit ratings and maturities
- Sensitivity of corporate yields to sovereign yields depends on corporate bonds’ credit ratings and maturities:
  - For corporate bonds with non-investment-grade ratings, sensitivity is particularly strong (in fact, larger than one).
  - For investment-grade corporate bonds, sensitivity is about halved.
  - Sensitivity by maturity:
    - strongest for corporate bonds with maturities between 1 to 5 years (around one),
    - about 15 percent lower for maturities between 5 to 10 years,
    - almost 50 percent lower if maturities are longer than 10 years.
  - Interpretation: default premiums are convex in default risks—riskier corporate bonds are more vulnerable to transfer risks; investors care less about transfer risks as time horizons increase.
  - Conceptual note: although governments can transfer current credit risks to the private sector, it is difficult for transfer effects to remain as strong for more than 5 or 10 years.
- Supporting evidence from related work: Arteta and Hale (2008) find the negative impact of sovereign debt crisis on credit availability of emerging-market firms persists for two years; bond-market reaction speed implies 1 to 5 years is a sufficiently “long-run” period in bond markets.

### Relation to existing literature and distinct contributions
- Theoretical literature has discussed linkages between sovereign and private-sector borrowing costs; some models assume a corporate–sovereign relation, others derive it:
  - Examples of models assuming a relation: Corsetti, Kuester, Meier, and Müller (2014); Mendoza and Yue (2011; 2012); Neumeyer and Perri (2005); Uribe and Yue (2006).
  - Examples deriving transfer-risk mechanisms: Bocola (2016); Sosa-Padilla (2018); Acharya, Drechsler, and Schnabl (2014); Cooper and Nikolov (2018); Farhi and Tirole (2018).
- Empirical distinctions:
  - This paper examines bond yields (secondary market) rather than CDS premiums or credit spreads; studies using CDS often have small, advanced-economy samples.
  - Other works using aggregate spreads include Cavallo and Valenzuela (2009) and Zinna (2014) (EMBI and CEMBI), and papers studying sovereign spreads and banks’ lending rates include Neri (2013); Zoli (2013); Albertazzi, Ropele, Sene, and Signoretti (2014).
  - Bevilaqua, Hale, and Tallman (2020) also study bond-yield associations but focus on primary corporate bond yields at issuance and include all local-currency denominated bonds; this paper focuses on secondary market yields and U.S.-dollar denominated bonds given interest in emerging economies.
    - Rationale for U.S.-dollar sample: expected domestic currency depreciation can cause strong co-movement of sovereign and corporate yields for bonds in local currencies, and such expected depreciation is difficult to measure empirically.
  - The paper studies the long-run association with a much larger sample, estimates models at the bond-level, and allows the impact of sovereign yields on corporate yields to differ across bonds even within the same country—helping to distinguish the transfer-risk and liquidity-premium channels and identify which corporate bonds are most vulnerable.
- Prior empirical work on long-run association: Dittmar and Yuan (2008) studied credit spreads of 98 sovereign bonds and 239 corporate bonds in 8 emerging economies using country-specific VARs and cumulative impulse responses; this paper expands the sample and focuses on bond yields to avoid measurement-error biases in credit spreads.

### Organization of the paper (as stated)
- Section 2 reviews the existing literature.
- Section 3 discusses data and methodology.
- Section 4 presents main results.
- Section 5 includes subsample analysis to examine the transfer-risk and liquidity-premium channels.
- Section 6 analyzes ratings, maturities, and whether the corporate-sovereign relation is nonlinear (following Bevilaqua, Hale, and Tallman 2020).
- Section 7 concludes the paper.

*Source: wpiea2021155-print-pdf - section  2  why  we  choose  to  study  bond  yields  rather  than  credit  spreads.  We  further  show  in  section  4.; https://www.imf.org/-/media/files/publications/wp/2021/english/wpiea2021155-print-pdf.pdf*

### 3. Data and Methodologies

### 3. Data and Methodologies

### 3.1. Data
- Data source and coverage
  - Corporate-bond pricing data obtained from I Markit covering weekly data between March 2016 and April 2019 (160 weeks in total).
  - Markit data include bid-, ask-, and mid-yields to maturity, maturity dates, issuers, and bond-level features (defaulted, perpetual, inflation-adjusted, government-related, etc.).
  - Linked Markit data to Bloomberg and Datastream via ISIN to obtain additional bond-level characteristics: country of risk, classification description, industry group, issue currency, credit ratings, and issuers’ financial data.
  - Sovereign bonds: weekly data of all individual sovereign bonds with non-missing data collected from Bloomberg in the same format.
- Sample selection
  - Kept only bonds whose country of risk is deemed an emerging economy as per the IMF World Economic Outlook definition.
  - Kept only corporate bonds denominated in U.S. dollars as identified by Bloomberg issue currency.
  - Bond credit ratings by Standard & Poor’s, Moody’s, and Fitch downloaded from Bloomberg at quarterly frequency; each rating assigned a numeric value from 1 (AAA) to 22 (defaulted), assuming a linear relation.
  - Quarterly issuer financial data (profit margin, leverage, asset tangibility, interest coverage) collected from Datastream; available for 17 percent of the sample.
  - Country-level control data collected: stock-market indices from MSCI and S&P500 (via Bloomberg); US Treasury yields from FRED; each country’s nominal U.S. dollar exchange rate, real effective exchange rate, real GDP growth, CPI growth, current account balance, real GDP per capita, public external debt, and foreign reserves from IMF internal databases.
- Data-cleaning procedure (applied in order)
  1. Drop bonds that are defaulted, perpetual, or inflation-adjusted.
  2. Drop bonds with maturity less than a year.
  3. Trim the sample at the top and bottom 0.5 percent by mid-yields and bid-ask yield spreads, and drop observations with non-positive yields.
  4. Drop bonds with less than 30 weeks of consecutive data in the sample (linear interpolation used to fill gaps, accounting for around 5.9 percent of the sample prior to this step).
  5. Drop duplicated bonds issued under Rule 144A if its Regulation-S counterpart is already in the sample.
- Final sample after cleaning
  - 2,045 U.S.-dollar corporate bonds issued by 1,040 companies.
  - 431 U.S.-dollar sovereign bonds.
  - 46 emerging economies included (selection criterion: country must have at least one U.S.-dollar corporate bond and one U.S.-dollar sovereign bond surviving cleaning).
- Additional notes on data choices and limitations
  - Markit introduced this data in 2016; earlier data not used due to inconsistency and different valuation bases.
  - Weekly frequency used (following Acharya, Drechsler, and Schnabl (2014)) to mitigate measurement error from lack of liquidity or noisy trading, despite daily data being available in the sample period.
  - Local-currency bonds excluded because yields are highly influenced by expected depreciation of local exchange rates that is difficult to measure empirically.
  - The Bloomberg “country of risk” assignment follows a 10-point methodology (country of domicile 4 points, country of listing 3 points, country of largest revenue 2 points, reporting currency 1 point); ties go to country of domicile. Red-chip issuers and holding companies are treated differently per Bloomberg documentation.
  - Sovereign pricing source at Bloomberg is the Evaluated Pricing Service (BVAL). Sovereign ratings are long-term foreign currency issuer/debt ratings by S&P, Moody’s, and Fitch, if available.
  - The authors note conclusions may better apply to emerging economies and to countries where U.S.-dollar bonds play an important role in financing.

### 3.2. Empirical models
- Pairing strategy
  - Each corporate bond I from country j (yield denoted y_{i,j,t}) is paired with x_{j,t}, the equal-weighted average yield on the sovereign portfolio of country j on the same date.
  - Rationale: the study aims to characterize time-series association between corporate and sovereign yields; theoretical channels (e.g., transfer-risk) imply associations need not rely on maturity matching.
  - Authors also note alternative matching on closest maturities (or matching further on ratings and bid-ask spreads) yields a highly significant sovereign-to-corporate pass-through but with smaller magnitude (see section 4.2).
- Modeling framework
  - Conditional ARDL panel-data model with unobserved common factors f_t:
    - y_{i,t} = sum_{l=1}^{p_yi} phi_{i,l} y_{i,t-l} + sum_{l=0}^{p_xi} beta_{i,l} x_{i,t-l} + u_{i,t}  (Equation (1))
    - u_{i,t} = gamma_i' f_t + epsilon_{i,t}  (Equation (2))
  - Long-run impact for bond i:
    - theta_i = (sum_{l=0}^{p_xi} beta_{i,l}) / (1 - sum_{l=1}^{p_yi} phi_{i,l})  (Equation (3))
    - Interpretation: cointegrating coefficient if y and x are unit-rooted; long-run impact of a permanent change in mean of x on y if stationary.
  - Two estimation approaches for bond-specific theta_i:
    - ARDL approach: estimate equations (1) and (2) bond-by-bond and compute theta_i via (3). Pros: consistent whether x is exogenous or endogenous. Cons: requires correct specification of lag orders p_{y_i} and p_{x_i}; panel must be stationary for ARDL to account for unobserved common factors except under restricted conditions.
    - Distributed Lag (DL) / Cross-Sectionally Augmented DL (CS-DL) approach: rewrite to DL form and estimate theta_i directly by truncating lag polynomial. CS-DL augments model with cross-sectional means of dependent and independent variables (ȳ_t and x̄_t) and their lags to control for unobserved common factors. Pros: allows unit roots in y, x, and f_t; allows correlation between x and f_t; does not require knowledge of individual lag orders because truncated order p set as increasing function of T. Cons: requires x to be exogenous (no feedback from lagged y into x).
  - Choice of baseline estimator
    - Baseline: CS-DL approach (motivated by typical unit-root presence in bond yields).
    - Authors argue plausibility of sovereign yields x being exogenous to corporate yields y in emerging markets (citing Dittmar and Yuan (2008) and other literature).
    - Robustness: also estimate via ARDL (which allows bidirectional causality) and find similar results; ARDL requires assuming unobserved common factors f_t uncorrelated with country-level average sovereign yields x_{i,t}.
- Baseline regression specification (bond-specific, estimated by OLS bond-by-bond)
  - y_{i,t} = c_{y_i} + theta_i x_{i,t} + sum_{l=0}^{p-1} delta_{i,l} Delta x_{i,t-l} + sum_{l=0}^{p_y_bar} omega_{y,i,l} ȳ_{t-l} + sum_{l=0}^{p_x_bar} omega_{x,i,l} x̄_{t-l} + Z_{i,t}' gamma + e_{i,t}  (Equation (8))
    - y_{i,t}: corporate bond I yield on date t.
    - c_{y_i}: bond-specific constant term (bond fixed effect).
    - x_{i,t-l}: l-th (weekly) lag of the average sovereign yield of the corporate bond’s home country.
    - ȳ_{t-l} = N^{-1} sum_{i=1}^N y_{i,t-l} and x̄_{t-l} = N^{-1} sum_{i=1}^N x_{i,t-l} are cross-sectional means to account for effects of unobserved common factors f_t.
    - Z_{i,t}: additional observed control variables.
    - Lag orders: p and p_x̄ set to integer part of T^{1/3}; p_ȳ set to 0 following Chudik, Mohaddes, Pesaran, and Raissi (2017).
- Choice of lag orders and sample time dimensions
  - In the sample: average number of periods T̄ = 121, shortest period T_min = 30, longest period T_max = 160.
  - Baseline number of lags chosen to be p = 4; robustness checks performed for p = 3 or 5.
  - Note: choice of 4 weekly lags aligns with choice of 20 daily lags (trading days) in Dittmar and Yuan (2008).
- Aggregation and inference
  - Mean-group estimator for average pass-through:
    - theta_hat_MG = N^{-1} sum_{i=1}^N theta_hat_i  (Equation (7))
    - Standard error calculated nonparametrically.
  - Estimation of theta_i via DL/truncated DL (Equation (6)) is consistent under usual root assumptions on phi_i(L) and absence of feedback from lagged y onto x; CS-DL addresses omitted-variable bias and cross-sectional dependence by including cross-sectional means and their lags.
- Additional methodological remarks
  - ARDL approach robust to exogeneity/endogeneity of x but requires correct lag-order specification and stationarity conditions to handle unobserved factors.
  - CS-DL preferred for baseline because it handles unit roots and allows x and f_t to be correlated; bias of CS-DL shown by simulations to be small even when feedback effects exist.
  - The authors match weekly bond yields with most recent quarterly ratings and quarterly financial data when constructing controls.

*Italic: Source — 3. Data and Methodologies (wpiea2021155-print-pdf)*

### 4. Full-sample empirical results

### 4. Full-sample empirical results

### 4.1. Baseline estimates
- Regression specification chosen as main model: column (3) of Table 2.
- Key empirical findings from baseline (columns (1)–(4) of Table 2):
  - The association between corporate bond yields and sovereign bond yields is positive and highly statistically significant.
  - The long-run pass-through from sovereign yields to corporate yields is around one (i.e., approximately full pass-through).
  - There appears to exist a slight downward trend.
  - The CS-DL approach accounts for possible unobserved common factors 풇풇t that may drive both corporate and sovereign bond yields.
- Controls added in columns (3) and (4):
  - Bond-level: average credit rating (average across Standard & Poor’s, Moody’s, and Fitch ratings), number of years to maturity, bid-ask yield spreads.
  - Country-level: MSCI stock-market returns and volatilities (30-day rolling window), growth rates of real and nominal exchange rates.
  - Global-level: S&P 500 stock returns and volatilities (30-day rolling window), U.S. 10-year Treasury yields as credit risk-free rate.
- Data coverage and sample trimming:
  - MSCI stock indices available for 31 countries (listed in Appendix A2).
  - Columns (3) and (4) cover 1,987 individual corporate bonds, accounting for more than 97 percent of the entire sample.
  - Based on estimates in column (3) of Table 2, individual pass-throughs 휃̂i display extreme values for some bonds (e.g., below -50 or almost reaching 100), possibly due to CS-DL technical issues (roots of 휑i(L) near the unit circle).
  - Trimmed sample: top and bottom 5 percent of 휃̂i removed; 200 bonds dropped; remaining 1,787 bonds constitute the trimmed-full sample.
- Results on trimmed-full sample (Table 3):
  - Re-estimation yields similar results with slightly decreased average pass-through and much smaller standard error.
  - Main specification in column (3) of Table 3: pass-through estimated to be 0.824 and statistically significant at the 1 percent level.
  - Corporate bond yields increase with:
    - bond’s maturity,
    - home country’s stock-market volatility,
    - real exchange rate appreciation.
  - Distribution of individual pass-throughs reported in Figure 2.
- Note on negative individual pass-throughs:
  - For a small fraction of bonds (274 bonds), pass-throughs are negative and significantly nonzero.
  - More than 80 percent of these 274 bonds are issued by companies in China.
  - Dropping these 274 bonds does not materially change results (available upon request).
- Choice for subsequent analysis:
  - Focus on the trimmed-full sample (1,787 bonds) to mitigate outlier concerns, with all results robust to the original full sample.

### 4.2. Robustness
- Summary of robustness exercises reported in Table 4 (mean-group estimates of long-run pass-throughs):
  - Alternative sovereign-corporate matching methods:
    - Column (1): pair corporate bond with sovereign bond minimizing mean-absolute distance (MAD) in maturities.
    - Column (2): pair by minimizing mean-squared distance (MSD) in maturities.
    - Columns (3)–(4): consider maturities, credit ratings, and bid-ask yield spreads simultaneously (MAD and MSD variants).
    - Estimated pass-throughs in these four cases are similar and highly significant, with magnitudes slightly smaller than baseline.
  - Additional controls:
    - Column (5): control for real GDP growth, CPI inflation, real GDP per capita, ratio of current account balance to GDP (annual frequency; weekly observations paired with previous year).
    - Column (6): control for issuers’ financial variables: profit margin, leverage, asset tangibility, interest coverage (quarterly frequency; weekly observations matched to most recent quarterly data). Data available for only a small fraction of bonds.
  - Sample composition checks:
    - Column (7): drop bonds issued by companies from China.
    - Column (8): keep only bonds with most common features: fixed-coupon, senior-unsecured, without embedded options.
  - Lag specification checks:
    - Columns (9)–(10): distributed lag order equal to 3 and 5, respectively.
  - Alternative estimation frameworks:
    - Column (11): panel ARDL approach (Pesaran and Smith 1995; Pesaran, Shin, and Smith 1999) allowing feedback from corporate to sovereign yields; report mean-group estimates.
    - Column (12): time-series ARDL approach (Pesaran, Shin, and Smith 2001) with synthetic time series pooled from panel data following Bevilaqua, Hale, and Tallman (2020). Estimation repeated with random shuffling of bond order 2000 times; reported long-run pass-through and standard error are means across iterations. Iteration ranges: pass-through from 0.87 to 1.02; standard error between 0.039 and 0.042.
  - Overall robustness conclusion:
    - In each case, a highly significant long-run pass-through from sovereign yields to corporate yields is found, and for most specifications the estimates are similar to baseline.
- Robustness to using credit spreads:
  - Replace corporate and sovereign bond yields with Z spreads (corporate: IHS Markit; sovereign: Bloomberg).
  - Baseline specification: long-run pass-through from sovereign spreads to corporate spreads estimated to be 0.831, close to the 0.824 pass-through from sovereign yields to corporate yields.
  - Interpretation: corporate-sovereign association unlikely driven by risk-free rates.
- Additional robustness checks (summarized; results in Appendix A4 and A5):
  - Results robust to controlling for:
    - time fixed effects (monthly frequency),
    - monthly excess bond premium in Gilchrist and Zakrajsek (2012),
    - ratio of portfolio investment to GDP (net, asset, liability flows separately),
    - weekly Moody’s U.S. Aaa corporate index yields,
    - weekly Moody’s U.S. Baa corporate index yields,
    - oil price weekly growth,
    - all variables in the list above simultaneously.
  - Robustness to alternative choices of risk-free rate: using U.S. 3-, 5-, or 7-year Treasury yields yields similar results.
- Notes on matching and distance metrics:
  - MAD and MSD formulas defined for comparing maturities, and combined MAD/MSD formulas incorporate maturities, average ratings (S&P, Moody’s, Fitch), and bid-ask yield spreads as shown in the source text.
- General conclusion from robustness section:
  - The positive and statistically significant long-run sovereign-to-corporate pass-through is robust across matching procedures, additional macroeconomic and issuer-level controls, alternative lag structures, alternative estimation methods, sample composition checks, and when using spreads instead of yields.

*Source: 4. Full-sample empirical results (from wpiea2021155-print-pdf)*

### 5. Transmission channels

### 5. Transmission channels

### 5.1. Transfer-risk channel
- Hypothesis: governments with greater sovereign risks are more likely to offload them onto the private sector; thus the pass-through from sovereign yields to corporate yields should be larger in higher sovereign-risk countries.
- Sample split and classification:
  - Countries classified as investment-grade (IG) if the mean of its sovereign bonds’ average ratings is better than or equal to BBB-.
  - “Average rating” = average across Standard & Poor’s, Moody’s, and Fitch for a given sovereign bond/date; country mean taken across all sovereign bonds and dates.
  - Public external debt to foreign reserves ratio: constructed on a country-year panel of the 31 countries between 2015 and 2018; time-series mean for each country compared to pooled median to classify low-risk (mean below pooled median) versus high-risk.
  - Resulting groups: low-risk group = 15 countries; high-risk group = 16 countries. (31 countries used here because only these 31 have MSCI stock data available.)
- Empirical findings (Table 5, columns (1)-(4)):
  - The pass-through from sovereign yields to corporate yields in high-risk countries almost doubles that in low-risk countries, irrespective of whether groups are defined by sovereign ratings or debt-to-reserve ratios.
  - Supports the transfer-risk mechanism.
- Financial firms and government-related issuers:
  - Financial firms identified using Bloomberg variable industrygroup.
  - Government-related issuers defined as “government-related” by IHS Markit, or as “government-guaranteed”, “government-owned”, or “government-sponsored” by Bloomberg.
  - Findings (Table 5, columns (5)-(8)):
    - Pass-through for financial firms’ bonds is much larger than for non-financial firms’ bonds.
    - Pass-through for government-related bonds is only half of that for firms without government relations, contradicting initial conjecture.
- Double-sorted analysis (financial vs non-financial and government-related vs non-government-related; Table 5, columns (9)-(12)):
  - Non-financial sector: pass-through is higher for government-related bonds than for non-government-related bonds (consistent with prediction).
  - Financial sector: pass-through is higher for non-government-related bonds.
  - Pass-through for government-related financial companies is small and insignificant from zero; part of this insignificant result is attributable to Chinese banks in the sample.
- Robustness check excluding Chinese bonds (Table 5, columns (13)-(16)):
  - Dropping bonds from China increases pass-throughs across columns.
  - Outside the financial sector: pass-through for government-related bonds remains larger than for non-government-related bonds.
  - Within financial sector: pass-through for government-related bonds becomes significant but still lower than for non-government-related bonds.
- Interpretation and literature relation:
  - Evidence indicates that in sovereign distress, banks without government relations are more likely to be vulnerable than government-related banks.
  - This contradicts the “moral suasion” notion (sovereigns induce related banks to increase sovereign bond holdings), but is consistent with evidence in Acharya et al. (2018) for certain European countries.
- Caveat:
  - Sample-split creates uneven bond counts because low-risk countries have more corporate bonds.
  - Chinese state-owned banks are atypical: gigantic size, strong government guarantees, large foreign reserves → their bond yields may be insensitive to sovereign yields.

### 5.2. Liquidity-premium channel
- Theory and hypothesis:
  - Sovereign U.S. dollar-denominated bonds can act as benchmark securities and hedging instruments for a country’s U.S. dollar-denominated corporate bonds; more liquid sovereign bonds enhance this role, increasing corporate-sovereign yield association.
  - Prediction: sensitivity of corporate yields to sovereign yields larger in countries where sovereign bonds are relatively more liquid.
- Measurement and sample split:
  - Sovereign bond liquidity measured by bid-ask yield spreads.
  - For each country/date: average bid-ask yield spread of equal-weighted sovereign bond portfolio computed, producing a weekly panel of 31 countries.
  - Panel median computed; time-series mean of average bid-ask spreads for each country compared to panel median.
  - Low-bid-ask (LBA) group: time-series mean lower than panel median; High-bid-ask (HBA) group: time-series mean higher than panel median.
  - Resulting groups: 15 LBA countries and 16 HBA countries.
  - Double-sort by LBA/HBA and investment-grade versus non-investment-grade (as in section 5.1) to obtain four subsamples.
- Empirical findings (Table 6):
  - Within investment-grade countries (columns (1) and (2)):
    - Sensitivity of corporate bond yields to sovereign bond yields is significant in LBA countries, but insignificant in HBA countries.
  - Within non-investment-grade countries (columns (3) and (4)):
    - Association between corporate and sovereign yields is larger for LBA countries than for HBA countries.
  - Comparison across IG status:
    - Within LBA countries, pass-through from sovereign to corporate yields is larger in non-investment-grade countries than in investment-grade countries.
    - Same pattern within HBA countries.
  - These results are consistent with both the liquidity-premium channel and the transfer-risk channel operating simultaneously.

### 6. Further exploration

### 6.1. Credit ratings and maturities
- Purpose: identify which corporate bonds are most vulnerable to sovereign risks by bond credit rating and maturity.
- Classification:
  - Corporate bonds assigned to investment-grade or non-investment-grade groups based on whether the time-series mean of the bond’s ratings is better than BBB- or not.
  - Maturity categories based on time-series mean years to maturity:
    - Short-term: between 1 year and 5 years.
    - Mid-term: between 5 years and 10 years.
    - Long-term: over 10 years.
- Empirical findings (Table 7):
  - From columns (1) and (2):
    - Pass-through is larger for non-investment-grade corporate bonds than for investment-grade corporate bonds.
    - Interpretation: default premium convexity implies riskier corporate bonds are more vulnerable to transfer risks.
  - From columns (3)-(5):
    - Sensitivity of corporate yields to sovereign yields is highest for short-term corporate bonds, lower for mid-term bonds, and much lower for long-term bonds.
    - Speculation: for mid- and long-term bonds a large fraction of repayment is not due in 5 or 10 years; sovereign transfer effects (increased taxation, reduced government guarantees, impaired intermediation) are less likely to affect repayments far in the future, reducing sensitivity.
  - Robustness: using bond durations (Macauley, modified, effective) in place of maturities yields similar results.

### 6.2. Nonlinearity
- Context: Bevilaqua, Hale, and Tallman (2020) find an increasing but concave relation between primary corporate yields at issuance and 10-year sovereign yields across issuances; they propose an information-provision channel where sovereign yields convey noisy public information.
- Test for nonlinear relation in secondary markets:
  - Table 8: columns (1) and (3) mirror Table 3; columns (2) and (4) add a squared sovereign yield term to capture potential nonlinearity.
  - Findings:
    - Squared term significant only when control variables are omitted (column (2)).
    - When control variables are added (column (4)), squared term becomes much smaller and insignificant.
    - Even in column (2) estimates, nonlinear relation is increasing and convex (pass-through larger when sovereign yields are higher), not concave.
    - This convexity aligns with earlier results that pass-through is larger in non-investment-grade countries or those with higher external debt, supporting the transfer-risk channel.
- Interpretation relative to prior literature:
  - Differences with Bevilaqua, Hale, and Tallman (2020) may reflect market segment differences:
    - Their results pertain to primary corporate yields at issuance (cross-section across issuances).
    - Current study examines co-dynamics of secondary corporate and sovereign yields in time series.
  - Possible conclusion: information-provision channel dominates in cross-section/primary markets; transfer-risk channel more prominent in time series/secondary markets.

### 7. Concluding remarks
- Dataset and main result:
  - Dataset covers more than 2,400 U.S. dollar-denominated bonds in 46 emerging markets.
  - On average, the long-run pass-through from sovereign yields to corporate yields is around one and highly significant.
- Heterogeneity and channels:
  - Pass-through displays great cross-country and cross-bond heterogeneity consistent with:
    - Governments transferring credit risks to domestic firms (transfer-risk channel).
    - Sovereign bonds affecting corporate bond pricing through liquidity effects (liquidity-premium channel).
  - Both channels have theoretical support but limited empirical attention; this work provides empirical support.
- Policy implication:
  - Fiscal distress causing high sovereign yields has a persistent spillover effect on private sector borrowing costs, potentially impeding growth for an extended period.
  - This perspective is related to literature documenting persistence of sovereign risks and long-lasting declines in foreign credit to domestic firms following sovereign debt crises.
- Limitations and avenues for future research:
  - Authors do not claim transfer-risk and liquidity-premium channels are the only mechanisms linking corporate and sovereign yields.
  - Despite subsample evidence consistent with both channels, other channels correlated with sovereign risks and liquidity could produce similar results.
  - Directly studying default risks and liquidity risks via CDS data, issuers’ financial positions, and bond transaction prices/volumes would help but such data are difficult to obtain for emerging-market bonds; left for future research.

*Source: wpiea2021155-print-pdf - 5. Transmission channels*

### References

### References

### Key literature on sovereign-to-corporate linkages
- Acharya, V., T. Eisert, C. Eufinger, and C. Hirsch. 2018. Real effects of the sovereign debt crisis in Europe: Evidence from syndicated loans. The Review of Financial Studies 31:2855-96.
- Acharya, V., I. Drechsler, and P. Schnabl. 2014. A pyrrhic victory? Bank bailouts and sovereign credit risk. The Journal of Finance 69:22-1082.
- Adelino, M., and M. A. Ferreira. 2016. Bank ratings and lending supply: Evidence from sovereign downgrades. The Review of Financial Studies 29(7): 1709-1746.
- Ağca, S., and O. Celasun. 2012. Sovereign debt and corporate borrowing costs in emerging markets. Journal of International Economics 88:198-208.
- Albertazzi, U., T. Ropele, G. Sene, and F. M. Signoretti. 2014. The impact of the sovereign debt crisis on the activity of Italian banks. Journal of Banking & Finance 46:387-402.
- Bocola, L. 2016. The pass-through of sovereign risk. Journal of Political Economy 124:879-926.
- Farhi, E., and J. Tirole. 2018. Deadly embrace: Sovereign and financial balance sheets doom loops. The Review of Economic Studies 85:1781-823.
- Gennaioli, N., A. Martin, and S. Rossi. 2013. Sovereign default, domestic banks, and financial institutions. The Journal of Finance 69:819-66.
- Pesaran, M.H., Y. Shin, and R. Smith. 1999. Pooled mean group estimation of dynamic heterogeneous panels. Journal of the American Statistical Association, 94(446): 621-634.
- Zinna, G. 2014. Identifying risks in emerging market sovereign and corporate bond spreads. Emerging Markets Review 20:1-22.
- (Full list of references provided in source.)

### Summary statistics (Table 1)
- Mid yields (%) — #Obs. 263,103; Mean 5.57; St.dev. 3.69; p25 3.77; p50 4.78; p75 6.31
- Average ratings (AAA=1, defaulted=22) — #Obs. 263,103; Mean 11.17; St.dev. 4.39; p25 8; p50 10.33; p75 13.67
- Number of years to maturity — #Obs. 263,103; Mean 7.91; St.dev. 8.71; p25 3.02; p50 4.99; p75 8.74
- Bid-ask yield spread (%) — #Obs. 263,103; Mean 0.19; St.dev. 0.3; p25 0.06; p50 0.11; p75 0.21
- Country stock returns (30-day %) — #Obs. 198,744; Mean 0.74; St.dev. 5.99; p25 -3.13; p50 0.7; p75 4.5
- Country stock volatilities (30-day %) — #Obs. 198,744; Mean 5.74; St.dev. 2.31; p25 4.21; p50 5.25; p75 6.64
- Nominal USD exchange rate growth (quarterly %) — #Obs. 205,140; Mean 1.12; St.dev. 5.22; p25 -1.44; p50 0.41; p75 2.52
- Real effective exchange rate growth (monthly %) — #Obs. 205,140; Mean 0.04; St.dev. 2.23; p25 -0.97; p50 0.04; p75 0.87
- Real GDP growth (annual %) — #Obs. 205,140; Mean 3.79; St.dev. 3.14; p25 1.73; p50 3.36; p75 6.9
- CPI growth (annual %) — #Obs. 201,676; Mean 4.43; St.dev. 5.45; p25 1.80; p50 2.52; p75 4.88
- Current account balance / GDP (annual %) — #Obs. 205,140; Mean -0.44; St.dev. 2.94; p25 -2.34; p50 -0.73; p75 1.8
- Real GDP per capita (2010 USD) — #Obs. 205,140; Mean 9971; St.dev. 7966; p25 6884; p50 7755; p75 10990
- Total external debt / foreign reserves (annual %) — #Obs. 205,074; Mean 236.2; St.dev. 344.5; p25 55.65; p50 179; p75 277.8
- S&P 500 returns (30-day %) — #Obs. 205,140; Mean 0.89; St.dev. 3.28; p25 -0.35; p50 1.25; p75 2.73
- S&P 500 volatilities (30-day %) — #Obs. 205,140; Mean 3.34; St.dev. 1.82; p25 2.03; p50 2.62; p75 4.14
- US 10-year Treasury yields (%) — #Obs. 205,140; Mean 2.5; St.dev. 0.45; p25 2.26; p50 2.55; p75 2.87

Note: Average rating assigned 23 when unrated by all three agencies; P25, p50, and p75 denote 25-, 50-, and 75-percentiles. Summary statistics for mid yields, average ratings, maturity, and bid-ask spreads cover all bonds; other statistics pertain to corporate bonds.

### Baseline pass-through estimates (Tables 2 and 3)
- Full sample (Table 2, column (3) preferred specification):
  - Sovereign yields → Corp. yields: 0.951*** (standard error 0.111)
  - US 10-year Treasury yields coefficient: -0.147** (standard error 0.0665)
  - Real effective exchange rate growth: 0.0189*** (standard error 0.00446)
  - S&P500 volatilities: 0.00714** (standard error 0.00306)
  - Observations 190,796; Number of firm bonds 1,987; Number of countries 31; R-squared (mean-group) 0.914
  - Time trend in some specifications: -0.00237* (standard error 0.00138)

- Trimmed-full sample (Table 3, column (3) preferred specification):
  - Sovereign yields → Corp. yields: 0.824*** (standard error 0.0356)
  - Maturity: 0.197*** (standard error 0.0441)
  - Real effective exchange rate growth: 0.0112*** (standard error 0.00231)
  - Observations 177,890; Number of corporate bonds 1,787; Number of countries 31; R-squared (mean-group) 0.926
  - Time trend: -0.00270*** (standard error 0.000861)

- Table 2 and Table 3 specifications include bond fixed effects, distributed lags, and cross-sectional means. Standard errors reported in parentheses; significance levels: *** p<0.01, ** p<0.05, * p<0.1.

### Robustness checks (Table 4 and Appendix A4–A5)
- Mean-group long-run sovereign-to-corporate pass-through across 12 robustness checks (Table 4, columns (1)-(12)):
  - Estimated sovereign bond yields coefficients range from 0.721*** to 1.059*** in reported checks.
  - Examples: MAD maturity 0.721*** (standard error 0.0588); Extra macro controls 0.839*** (standard error 0.0368); Drop China 1.059*** (standard error 0.0473); Lag = 5 0.885*** (standard error 0.0427); Synthetic time-series ARDL 0.959*** (standard error 0.0404).
  - R-squared (mean-group) values range from 0.134 to 0.937 depending on method.

- Additional controls (Appendix A4):
  - Including monthly dummies: sovereign yields 0.906*** (standard error 0.0536); Observations 177,890; R-squared 0.97
  - Excess bond premium: sovereign yields 0.861*** (standard error 0.0373); Observations 176,757; R-squared 0.929
  - Moody's U.S. Aaa yields: sovereign yields 0.872*** (standard error 0.0376)
  - Oil price growth: sovereign yields 0.832*** (standard error 0.0364)
  - All variables (2)-(8): sovereign yields 0.949*** (standard error 0.152); Observations 176,433; R-squared 0.941

- Credit spreads robustness (Appendix A5):
  - Corporate yields specification: sovereign yields 0.824*** (standard error 0.0356); Observations 177,890; R-squared 0.926
  - Corporate spreads specification: sovereign spreads 0.831*** (standard error 0.156); Observations 147,729; R-squared 0.937

### Heterogeneity and subsample results (Tables 5–7)
- Transfer risk subsamples (Table 5):
  - IG countries: sovereign yields 0.591*** (standard error 0.0984); Observations 121,433; Number of corporate bonds 1,218; R-squared 0.902
  - Non-IG countries: sovereign yields 1.078*** (standard error 0.0632); Observations 56,457; Number of corporate bonds 569; R-squared 0.954
  - Low debt/reserves: sovereign yields 0.614*** (standard error 0.0604); Observations 133,097
  - High debt/reserves: sovereign yields 1.296*** (standard error 0.0989); Observations 44,793
  - Financial bonds: sovereign yields 1.197*** (standard error 0.0980)
  - Non-financial bonds: sovereign yields 0.678*** (standard error 0.0421)
  - Government-related bonds: sovereign yields 0.474*** (standard error 0.0470)
  - Non-related bonds: sovereign yields 0.882*** (standard error 0.0463)

- Liquidity-based subsamples (Table 6):
  - IG & LBA countries: sovereign yields 0.496*** (standard error 0.182); Observations 50,007; Number of corporate bonds 442
  - IG & HBA countries: sovereign yields -0.965 (standard error 0.643); Observations 71,426; Number of corporate bonds 776
  - Non-IG & LBA countries: sovereign yields 1.411*** (standard error 0.164); Observations 22,659
  - Non-IG & HBA countries: sovereign yields 1.181*** (standard error 0.0945); Observations 33,798

- Corporate ratings and maturities (Table 7):
  - IG corporate bonds: sovereign yields 0.548*** (standard error 0.0546); Observations 92,196; Number of corporate bonds 852
  - Non-IG corporate bonds: sovereign yields 1.023*** (standard error 0.0552); Observations 85,694
  - Short-term (ST) corporate bonds: sovereign yields 0.935*** (standard error 0.0574); Observations 100,526
  - Medium-term (MT) corporate bonds: sovereign yields 0.816*** (standard error 0.0550); Observations 54,453
  - Long-term (LT) corporate bonds: sovereign yields 0.488*** (standard error 0.0825); Observations 22,911

### Nonlinearity and fitted relation (Table 8 and Figure 3)
- Nonlinear specification (Table 8, column (2)):
  - Sovereign yields coefficient: -2.975*** (standard error 0.969)
  - Sovereign yields squared coefficient: 0.478*** (standard error 0.114)
  - Fitted curve (from column (2) estimates): ŷ = −2.975 x + 0.478 x^2 + 8.199 (8.199 is the estimated constant)

- Individual pass-through distributions:
  - Full-sample (Figure 1): Mean = 0.951; median = 0.595; standard deviation = 4.959
  - Trimmed-full-sample (Figure 2): Mean = 0.824; median = 0.581; standard deviation = 1.506

### Panel cointegration and unit-root evidence (Appendix A1 and A3)
- Panel cointegration tests (Appendix A1) — p-values for Modified Phillips-Perron, Modified Dickey-Fuller, and Dickey-Fuller tests reported as 0.00 for Lag=1 through Lag=5, indicating rejection of no cointegration (common null) in favor of cointegration across panel individuals.
- Phillips–Perron unit-root tests for individual bonds (Appendix A3) — p-values plotted; test results indicate a majority of bonds contain unit roots (robust to lag choices of 3, 4, or 5 and inclusion/exclusion of trend).

### Sample coverage and country-level bond counts (Appendix A2)
- Cross-sectional size totals:
  - Total sovereign bonds: 431
  - Total corporate bonds: 2,045
  - Government-related corporate bonds: 1,384
  - Non-related corporate bonds: 661
- Examples of country counts (selected entries as presented):
  - Argentina: Sovereign 34; Corporate 44 (Government-related 8; Non-related 36)
  - Brazil: Sovereign 15; Corporate 191 (Government-related 35; Non-related 156)
  - China: Sovereign 4; Corporate 862 (Government-related 335; Non-related 527)
  - Mexico: Sovereign 18; Corporate 210 (Government-related 88; Non-related 122)
  - Russia (Russian Federation): Sovereign 10; Corporate 119 (Government-related 36; Non-related 83)
  - Total sample includes 46 countries in some specifications and 31 countries in trimmed/full-sample regressions (as reported in Tables 2–3).

*Italic: Source — wpiea2021155-print-pdf - References (IMF PDF).*

---


_Source: https://www.imf.org/-/media/files/publications/wp/2021/english/wpiea2021155-print-pdf.pdf_