## onlineannex31 - EXECUTIVE SUMMARY

## Source details

**Canonical URL:** [onlineannex31 - EXECUTIVE SUMMARY](https://www.imf.org/-/media/files/publications/gfsr/2020/october/english/onlineannex31.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/gfsr/2020/october/english/onlineannex31.pdf.md)
- [Structured JSON version](/-/media/files/publications/gfsr/2020/october/english/onlineannex31.pdf.json)

---

### Data Sources
- Aggregate macrofinancial indicators sourced from: IMF, International Financial Statistics; Thomson Reuters EIKON; Haver Analytics; Chicago Board Options Exchange; Bloomberg; DataStream; Federal Reserve; European Central Bank.
- Bank lending survey and aggregate bank loan and deposit indicators sourced from Haver Analytics.
- Corporate loan and bond indicators sourced from Dealogic, S&P LCD, S&P Global Ratings, Bloomberg, Thomson Reuters EIKON, DataStream.
- Excess bond premium (United States) sourced from the Federal Reserve.
- Firm-level financials and market data sourced from S&P Capital IQ, S&P Market Intelligence, Moody's Analytics, Datastream.
- Policy announcements and dates sourced from Yale Program on Financial Stability, IMF COVID Policy Tracker, press releases and press reports.

### Firms’ Choice of Debt Financing Instrument — data and estimation
- Jurisdictions analyzed: Canada, the euro area (France, Germany, Italy), and the United States.
- Sample period: 2000:Q1 to 2020:Q2.
- Number of firms in the analysis: 163 for Canada, 285 for the euro area, and 1,516 for the United States.
- Estimation method: pooled maximum likelihood estimation carried out separately for each jurisdiction.
- Sample inclusion: a firm-quarter is included if the firm issued at least one syndicated loan or one corporate bond during that quarter.
- Indicator variable definition: equals one if the firm issues only bonds in quarter t (or if bonds issued exceed syndicated loans when both are issued); equals zero if only loans are issued.
- Firm characteristics vector (ChIrr): includes size, Tobin’s Q, asset tangibility, rating, profitability, leverage, and liquidity gap.
- Credit-supply conditions vector (CreBit_SISSSSS): includes LIBOR-OIS spread, excess bond premium (EBP — see Online Annex 3.4), Global Financial Crisis dummy (equal to 1 during 2007:Q3-2009:Q2), and two COVID-19 dummies (dummy1 equal to 1 in 2020Q1, and dummy2 equal to 1 in 2020:Q2).
- Syndicated loans composition note: loans labeled “Revolving/Term Credit Facility”, “Revolving Credit”, “Credit Facility”, “L/C Facility”, “Bridge Facility”, “Swingline Facility”, “Reducing Revolving Credit”, “Overdraft Facility”, “Commitment Line”, and term loans labeled “Term loan”, and “Term loan A-H” in the Dealogic database.
- Quarter-firms during 2020:H1: 40 for Canada, 70 for the euro area, and 552 for the United States.

Subsample analyses (United States)
- Two US subsamples:
  - (A) investment grade loan issuance vs investment grade bond issuance.
  - (B) leveraged loan issuance vs high yield bond issuance.
- Inclusion rules: firm-quarter included if at least one investment grade loan or bond issued for (A), and at least one leveraged loan or high-yield bond for (B).
- Crisis-sensitivity analysis: interacts firm characteristics with crisis dummies; uses a single COVID-19 dummy equal to 1 during 2020:H1 for the US-only exercise.

### Identification of Aggregate Credit Supply Shocks — syndicated loan market
- Dataset: syndicated loan issuance in the euro area (France, Germany, Italy), the United Kingdom, and the United States.
- Sample periods:
  - Euro area and United Kingdom: 2010:Q1 to 2020:Q2.
  - United States: 2005:Q1 to 2020:Q2.
- Sample composition: 5,086 loans provided by 220 banks.
- Estimation: carried out separately for each jurisdiction; bank-level bivariate structural panel VAR (contemporaneous spread and volume) with bank fixed effects.
- Key variables:
  - ISrereI B_it: weighted average of contractual spreads of loans made by bank i in quarter t.
  - vBSSID_e_it: weighted average of loan volume of bank i in quarter t.
  - Predetermined controls: weighted average of log maturity, tranche rating, and bank characteristics (capital ratio, NPL ratio, loan-to-asset ratio, ROAE).
- Interpretation: contemporaneous residuals e_iit and I_iit interpreted as supply and demand shocks when supply price elasticity φ_s > 0 and demand price elasticity φ_d < 0.
- Identification technique: Rigobon’s method of identification through heteroscedasticity (multiple regimes).
- Regimes used:
  - Euro area and United Kingdom: two regimes — (i) 2010:Q1-2019:Q4 (before COVID-19) and (ii) 2020:H1 (COVID-19).
  - United States: five regimes — (i) 2005:Q1-2007:Q2 (before the GFC), (ii) 2007:Q3-2009:Q1 (during the GFC), (iii) 2009:Q2-2009:Q4 (recovery after the GFC), and the two regimes identical to the euro area/UK above for COVID-19 periods.
- Parameterization and estimation:
  - Volatilities transformed exponentially to satisfy sign conditions.
  - φ_d = −exp(ψ_d) and φ_s = exp(ψ_s).
  - Generalized method of moments used to minimize distance between theoretical and empirical moment matrices (one-step GMM with equal weighting used for computational simplicity).
- Aggregate credit supply shock computed as residual of the aggregate supply equation using estimated φ_s.

### Estimation of the Excess Bond Premium (EBP)
- Dataset: monthly secondary market corporate bond yields in the euro area (France, Germany, Italy), Japan, and the United Kingdom.
- Sample period: 2005:M1 to 2020:M6.
- Number of bonds included: 70 for the euro area, 1,286 for Japan, and 53 for the United Kingdom.
- Bond selection: unsecured straight bonds with maturity shorter than 30 years; US EBP data used in the chapter obtained from the Federal Reserve Board.
- Methodology (following Gilchrist and Zakrajšek 2012):
  - Construct GZ spread for each bond i issued by firm j: GZISreI B_ij = Y_ieSS B_ij − SISeeY_ieSS B_ij, where safe yield is constructed using OIS-based zero coupon curve and replicated cash flows.
  - Estimate panel regression for log(GZ spread) with firm fixed effects and controls:
    - Controls include PD (Moody’s KMV 1-year expected default frequency), log(days since issuance), callable bond dummy, log of liquidity, log of credit rating factors, log of market-implied factors, leverage, and interaction terms between firm/bond variables and interest-rate factors extracted from the OIS-based zero-coupon curve via a Diebold and Li (2006) dynamic factor model.
    - Interaction terms: [IBtter II t iB B I with LeveS], [IBtter II t iB B I with SSS BSe], [IBtter II t iB B I with CIRvItIr e], representing interactions between firm/bond-specific variables (except fixed effect) and interest-rate factors.
  - Firm-level EBP is the residual ê_i; economy-level EBP defined as the median over firms j of the average of individual ê_i over a firm’s bonds during each quarter:
    - EEBP_i ≃ Median_j (1/N_j Σ_{i=1}^{N_j} ê_{i}).
  - Note: median used instead of average due to smaller sample size and robustness to outliers.

### Firm-Level Stock Market Performance — event study methodology
- Objective: analyze stock market performance of firms classified by degree of financial vulnerability at end-2019 during phases of the pandemic.
- Method: standard event study using daily abnormal returns from a CAPM-style specification.
- Daily abnormal return formula:
  - AR_it = S_tBIS S Return_it − AASSh_i − Beet_i × MMarketReturn_it, where AASSh_i and Beet_i estimated from regression of daily firm returns on daily domestic market returns during 2019.
- Cumulative abnormal returns (CAR) for firm i between days T1 and T2:
  - CAR_i(T1, T2) = Σ_{t=T1}^{T2} AR_it.
- CAR windows computed for:
  - February 3, 2020 to March 31, 2020.
  - February 3, 2020 to June 30, 2020.
- Vulnerability groupings: firms grouped into portfolios based on end-2019 vulnerability using four indicators — leverage, size, relative cash, and liquidity gap (see Online Annex references for definitions).

### Cross-section analysis of cumulative abnormal returns (CARs)
- Specification estimated separately for each G7 economy:
  - CAR_i(T1, T2) = β HighVulnerability_i + γ CBPolicy_i + ε_i
  - The coefficient of interest is β, representing the differential impact of developments between T1 and T2 on more vulnerable firms.
- High vulnerability (HighVulnerability_i) is an indicator variable equal to one when a firm is identified as vulnerable:
  - Vulnerability corresponds to any of:
    - size in the lowest tercile (end-2019),
    - relative cash in the lowest tercile (end-2019),
    - liquidity gap in the highest tercile (end-2019),
    - leverage in the top half of the distribution (end-2019).
- Firm-level controls include:
  - book-to-market ratio (end-2019),
  - Tobin’s Q (end-2019),
  - EBITDA-to-total assets ratio (end-2019),
  - SIC2 industry dummy,
  - change in 12-months ahead sales forecast between 2019Q2 and 2020Q2.
- Additional controls:
  - High-leverage indicator = 1 if total debt-to-asset ratio above median (end-2019 distribution in relevant country).
  - Small-size indicator = 1 if total assets below the first tercile (end-2019 distribution in relevant country).
- Robustness:
  - Results are robust to using abnormal returns calculated based on a three-factor Fama-French model instead of the CAPM model.

### Policy intervention analysis — objectives and approach
- Objective: analyze the impact of various policy announcements on firms’ abnormal returns and whether announcements had stronger effects on firms more vulnerable to adverse credit supply shocks.
- Policy set:
  - Includes monetary, fiscal, and financial policy measures (see Annex Table 3.6.1 in source).
  - Policies can be grouped into twelve fine categories and two coarse categories (direct vs indirect support to firms).
- Pooling intervention days across countries in the period from February 3, 2020 to June 30, 2020, the model estimated uses averaged two-day abnormal returns:
  - Dependent variable = average of firms’ abnormal returns in the two-day period including the date of the policy announcement and the following day.

### Policy categories (coarse and examples)
- Policies providing indirect support to firms:
  - Monetary policy rate cut — A cut in the monetary policy rate
  - Asset purchases – government securities — Central bank government securities purchase programs
  - Other market liquidity — Central bank programs aimed at restoring liquidity in a specific market (excluding corporate funding markets)
  - Bank funding — Targeted Long-Term Refinancing Operations
  - Funding for lending (Central Bank) — Central bank liquidity provision to banks to encourage bank lending
  - Macroprudential — Easing of the macroprudential policy stance, easing of bank capital/liquidity requirements or policy guidance encouraging use of flexibility in regulation
  - Bank dividends — Policies limiting bank dividend distribution
- Policies providing direct support to firms:
  - Guarantees — Government loan guarantees to nonfinancial businesses
  - Corporate collateral — Easing of central bank collateral requirements to include a wider scope of nonfinancial firm debt securities
  - Asset purchases – corporate securities — Corporate bond purchase programs
  - Corporate loans funding by the government — Government programs aimed at providing loans to the nonfinancial corporate sector
  - Fiscal relief — Government support through grants, tax holidays, payroll and employment support
- Note: The classification of the policy measures is based on the Yale Program on Financial Stability (YPFS) COVID-19 Financial Response Tracker.

### Econometric specifications for policy interventions
- Baseline pooled specification (indices: i firm, j sector, k country, t trading day, v vulnerability type):
  - AR_i,j,k,t = Σ_{v=1 to 3} β_v HighVul_{i,j,k,v} + Σ_{v=1 to 3} σ_v HighVul_{i,j,k,v} X_{i} + ρ CBPolicy_{i,j,k,t} + λ CBPolicy_{i,j,k,t} X_{i} + θ_{S,t} + λ_i + ε_{i,j,k,t}
  - Where three vulnerabilities enter simultaneously: either (low relative cash, small size, high leverage) or (high liquidity gap, small size, high leverage).
  - X_i includes controls: book-to-market ratio, Tobin’s Q, cash-flow-to-total assets, and a pandemic-related revenue shock proxy.
  - HighVul X variable for volatility interaction:
    - HighVIX_i is dummy = 1 whenever daily VIX > 80th percentile of VIX distribution (February 2020 to June 2020).
    - Captures time-varying effect of extreme volatility on firms with different characteristics.
  - Fixed effects:
    - θ_{S,t} are country-date fixed effects,
    - λ_i are industry fixed effects (2-digits SIC).
  - Standard errors clustered at industry and country-date levels.
  - Coefficients of interest: β_v.
- Enriched specification studying direct vs indirect interventions:
  - AR_i,j,k,t = Σ_{v=1 to 3} β_v HighVul_{...} + Σ_{v=1 to 3} σ_v HighVul_{...} X_i + Σ_{v=1 to 3} η_v HighVul_{...} DirInterv_{k,t} + Σ_{v=1 to 3} ω_v HighVul_{...} X_i DirInterv_{k,t} + ρ CBPolicy + λ CBPolicy X_i + θ_{k,t} + λ_i + ε
  - DirInterv_{k,t} is a dummy = 1 for policy intervention days when the set of interventions included at least one that targeted the corporate sector directly.
  - Coefficients of interest: η_v, measuring the differential effect of announcements that included direct interventions on firms with high vulnerability relative to announcements with only indirect interventions.

### Data, timing, and implementation details
- Period for pooled intervention-day analysis: February 3, 2020 to June 30, 2020.
- Dependent variable construction: average abnormal returns over the announcement day and the following trading day.
- Controls and interactions:
  - Book-to-market ratio, Tobin’s Q, cash-flow-to-total assets, pandemic-related revenue shock proxy.
  - VIX interaction using 80th percentile threshold (February 2020–June 2020).
- Fixed effects and clustering:
  - Country-date fixed effects (θ_{k,t}),
  - Industry fixed effects (2-digit SIC, λ_i),
  - Standard errors clustered at industry and country-date levels.
- The source provides Online Annex Table 3.6.2 listing announcement dates and main policy interventions used in the econometric analysis of Chapter 3.

### Key parameters and inference targets
- Cross-section CAR analysis: coefficient β (differential impact on vulnerable firms).
- Policy intervention pooled models: coefficients β_v, σ_v, η_v, ω_v, ρ, λ capture:
  - β_v: baseline differential effect of vulnerabilities on announcement-day abnormal returns.
  - η_v: additional differential effect when announcements include at least one direct corporate-sector intervention.
  - σ_v and ω_v: interactions with firm characteristics X_i.
  - ρ and λ: effects of CBPolicy and its interaction with firm controls.
- Robustness: abnormal returns computed with three-factor Fama-French model instead of CAPM produce robust results.

*Source: onlineannex31 - EXECUTIVE SUMMARY*

### EXECUTIVE SUMMARY

### onlineannex31 - EXECUTIVE SUMMARY

### Data Sources
- Aggregate macrofinancial indicators sourced from: IMF, International Financial Statistics; Thomson Reuters EIKON; Haver Analytics; Chicago Board Options Exchange; Bloomberg; DataStream; Federal Reserve; European Central Bank.
- Bank lending survey and aggregate bank loan and deposit indicators sourced from Haver Analytics.
- Corporate loan and bond indicators sourced from Dealogic, S&P LCD, S&P Global Ratings, Bloomberg, Thomson Reuters EIKON, DataStream.
- Excess bond premium (United States) sourced from the Federal Reserve.
- Firm-level financials and market data sourced from S&P Capital IQ, S&P Market Intelligence, Moody's Analytics, Datastream.
- Policy announcements and dates sourced from Yale Program on Financial Stability, IMF COVID Policy Tracker, press releases and press reports.

### Firms’ Choice of Debt Financing Instrument — data and estimation
- Jurisdictions analyzed: Canada, the euro area (France, Germany, Italy), and the United States.
- Sample period: 2000:Q1 to 2020:Q2.
- Number of firms in the analysis: 163 for Canada, 285 for the euro area, and 1,516 for the United States.
- Estimation method: pooled maximum likelihood estimation carried out separately for each jurisdiction.
- Sample inclusion: a firm-quarter is included if the firm issued at least one syndicated loan or one corporate bond during that quarter.
- Indicator variable definition: equals one if the firm issues only bonds in quarter t (or if bonds issued exceed syndicated loans when both are issued); equals zero if only loans are issued.
- Firm characteristics vector (ChIrr): includes size, Tobin’s Q, asset tangibility, rating, profitability, leverage, and liquidity gap.
- Credit-supply conditions vector (CreBit_SISSSSS): includes LIBOR-OIS spread, excess bond premium (EBP — see Online Annex 3.4), Global Financial Crisis dummy (equal to 1 during 2007:Q3-2009:Q2), and two COVID-19 dummies (dummy1 equal to 1 in 2020Q1, and dummy2 equal to 1 in 2020:Q2).
- Syndicated loans composition note: loans labeled “Revolving/Term Credit Facility”, “Revolving Credit”, “Credit Facility”, “L/C Facility”, “Bridge Facility”, “Swingline Facility”, “Reducing Revolving Credit”, “Overdraft Facility”, “Commitment Line”, and term loans labeled “Term loan”, and “Term loan A-H” in the Dealogic database.
- Quarter-firms during 2020:H1: 40 for Canada, 70 for the euro area, and 552 for the United States.

Subsample analyses (United States)
- Two US subsamples:
  - (A) investment grade loan issuance vs investment grade bond issuance.
  - (B) leveraged loan issuance vs high yield bond issuance.
- Inclusion rules: firm-quarter included if at least one investment grade loan or bond issued for (A), and at least one leveraged loan or high-yield bond for (B).
- Crisis-sensitivity analysis: interacts firm characteristics with crisis dummies; uses a single COVID-19 dummy equal to 1 during 2020:H1 for the US-only exercise.

### Identification of Aggregate Credit Supply Shocks — syndicated loan market
- Dataset: syndicated loan issuance in the euro area (France, Germany, Italy), the United Kingdom, and the United States.
- Sample periods:
  - Euro area and United Kingdom: 2010:Q1 to 2020:Q2.
  - United States: 2005:Q1 to 2020:Q2.
- Sample composition: 5,086 loans provided by 220 banks.
- Estimation: carried out separately for each jurisdiction; bank-level bivariate structural panel VAR (contemporaneous spread and volume) with bank fixed effects.
- Key variables:
  - ISrereI B_it: weighted average of contractual spreads of loans made by bank i in quarter t.
  - vBSSID_e_it: weighted average of loan volume of bank i in quarter t.
  - Predetermined controls: weighted average of log maturity, tranche rating, and bank characteristics (capital ratio, NPL ratio, loan-to-asset ratio, ROAE).
- Interpretation: contemporaneous residuals e_iit and I_iit interpreted as supply and demand shocks when supply price elasticity φ_s > 0 and demand price elasticity φ_d < 0.
- Identification technique: Rigobon’s method of identification through heteroscedasticity (multiple regimes).
- Regimes used:
  - Euro area and United Kingdom: two regimes — (i) 2010:Q1-2019:Q4 (before COVID-19) and (ii) 2020:H1 (COVID-19).
  - United States: five regimes — (i) 2005:Q1-2007:Q2 (before the GFC), (ii) 2007:Q3-2009:Q1 (during the GFC), (iii) 2009:Q2-2009:Q4 (recovery after the GFC), and the two regimes identical to the euro area/UK above for COVID-19 periods.
- Parameterization and estimation:
  - Volatilities transformed exponentially to satisfy sign conditions.
  - φ_d = −exp(ψ_d) and φ_s = exp(ψ_s).
  - Generalized method of moments used to minimize distance between theoretical and empirical moment matrices (one-step GMM with equal weighting used for computational simplicity).
- Aggregate credit supply shock computed as residual of the aggregate supply equation using estimated φ_s.

### Estimation of the Excess Bond Premium (EBP)
- Dataset: monthly secondary market corporate bond yields in the euro area (France, Germany, Italy), Japan, and the United Kingdom.
- Sample period: 2005:M1 to 2020:M6.
- Number of bonds included: 70 for the euro area, 1,286 for Japan, and 53 for the United Kingdom.
- Bond selection: unsecured straight bonds with maturity shorter than 30 years; US EBP data used in the chapter obtained from the Federal Reserve Board.
- Methodology (following Gilchrist and Zakrajšek 2012):
  - Construct GZ spread for each bond i issued by firm j: GZISreI B_ij = Y_ieSS B_ij − SISeeY_ieSS B_ij, where safe yield is constructed using OIS-based zero coupon curve and replicated cash flows.
  - Estimate panel regression for log(GZ spread) with firm fixed effects and controls:
    - Controls include PD (Moody’s KMV 1-year expected default frequency), log(days since issuance), callable bond dummy, log of liquidity, log of credit rating factors, log of market-implied factors, leverage, and interaction terms between firm/bond variables and interest-rate factors extracted from the OIS-based zero-coupon curve via a Diebold and Li (2006) dynamic factor model.
    - Interaction terms: [IBtter II t iB B I with LeveS], [IBtter II t iB B I with SSS BSe], [IBtter II t iB B I with CIRvItIr e], representing interactions between firm/bond-specific variables (except fixed effect) and interest-rate factors.
  - Firm-level EBP is the residual ê_i; economy-level EBP defined as the median over firms j of the average of individual ê_i over a firm’s bonds during each quarter:
    - EEBP_i ≃ Median_j (1/N_j Σ_{i=1}^{N_j} ê_{i}).
  - Note: median used instead of average due to smaller sample size and robustness to outliers.

### Firm-Level Stock Market Performance — event study methodology
- Objective: analyze stock market performance of firms classified by degree of financial vulnerability at end-2019 during phases of the pandemic.
- Method: standard event study using daily abnormal returns from a CAPM-style specification.
- Daily abnormal return formula:
  - AR_it = S_tBIS S Return_it − AASSh_i − Beet_i × MMarketReturn_it, where AASSh_i and Beet_i estimated from regression of daily firm returns on daily domestic market returns during 2019.
- Cumulative abnormal returns (CAR) for firm i between days T1 and T2:
  - CAR_i(T1, T2) = Σ_{t=T1}^{T2} AR_it.
- CAR windows computed for:
  - February 3, 2020 to March 31, 2020.
  - February 3, 2020 to June 30, 2020.
- Vulnerability groupings: firms grouped into portfolios based on end-2019 vulnerability using four indicators — leverage, size, relative cash, and liquidity gap (see Online Annex references for definitions).

*Source: onlineannex31 - EXECUTIVE SUMMARY*

### 3.1 for definitions).

### onlineannex31 - 3.1 for definitions)

### Cross-section analysis of cumulative abnormal returns (CARs)
- Specification estimated separately for each G7 economy:
  - CAR_i(T1, T2) = β HighVulnerability_i + γ CBPolicy_i + ε_i
  - The coefficient of interest is β, representing the differential impact of developments between T1 and T2 on more vulnerable firms.
- High vulnerability (HighVulnerability_i) is an indicator variable equal to one when a firm is identified as vulnerable:
  - Vulnerability corresponds to any of:
    - size in the lowest tercile (end-2019),
    - relative cash in the lowest tercile (end-2019),
    - liquidity gap in the highest tercile (end-2019),
    - leverage in the top half of the distribution (end-2019).
- Firm-level controls include:
  - book-to-market ratio (end-2019),
  - Tobin’s Q (end-2019),
  - EBITDA-to-total assets ratio (end-2019),
  - SIC2 industry dummy,
  - change in 12-months ahead sales forecast between 2019Q2 and 2020Q2.
- Additional controls:
  - High-leverage indicator = 1 if total debt-to-asset ratio above median (end-2019 distribution in relevant country).
  - Small-size indicator = 1 if total assets below the first tercile (end-2019 distribution in relevant country).
- Robustness:
  - Results are robust to using abnormal returns calculated based on a three-factor Fama-French model instead of the CAPM model.

### Policy intervention analysis — objectives and approach
- Objective: analyze the impact of various policy announcements on firms’ abnormal returns and whether announcements had stronger effects on firms more vulnerable to adverse credit supply shocks.
- Policy set:
  - Includes monetary, fiscal, and financial policy measures (see Annex Table 3.6.1 in source).
  - Policies can be grouped into twelve fine categories and two coarse categories (direct vs indirect support to firms).
- Pooling intervention days across countries in the period from February 3, 2020 to June 30, 2020, the model estimated uses averaged two-day abnormal returns:
  - Dependent variable = average of firms’ abnormal returns in the two-day period including the date of the policy announcement and the following day.

### Policy categories (coarse and examples from Annex Table 3.6.1)
- Policies providing indirect support to firms:
  - Monetary policy rate cut — A cut in the monetary policy rate
  - Asset purchases – government securities — Central bank government securities purchase programs
  - Other market liquidity — Central bank programs aimed at restoring liquidity in a specific market (excluding corporate funding markets)
  - Bank funding — Targeted Long-Term Refinancing Operations
  - Funding for lending (Central Bank) — Central bank liquidity provision to banks to encourage bank lending
  - Macroprudential — Easing of the macroprudential policy stance, easing of bank capital/liquidity requirements or policy guidance encouraging use of flexibility in regulation
  - Bank dividends — Policies limiting bank dividend distribution
- Policies providing direct support to firms:
  - Guarantees — Government loan guarantees to nonfinancial businesses
  - Corporate collateral — Easing of central bank collateral requirements to include a wider scope of nonfinancial firm debt securities
  - Asset purchases – corporate securities — Corporate bond purchase programs
  - Corporate loans funding by the government — Government programs aimed at providing loans to the nonfinancial corporate sector
  - Fiscal relief — Government support through grants, tax holidays, payroll and employment support
- Note: The classification of the policy measures is based on the Yale Program on Financial Stability (YPFS) COVID-19 Financial Response Tracker.

### Econometric specifications for policy interventions
- Baseline pooled specification (indices: i firm, j sector, k country, t trading day, v vulnerability type):
  - AR_i,j,k,t = Σ_{v=1 to 3} β_v HighVul_{i,j,k,v} + Σ_{v=1 to 3} σ_v HighVul_{i,j,k,v} X_{i} + ρ CBPolicy_{i,j,k,t} + λ CBPolicy_{i,j,k,t} X_{i} + θ_{S,t} + λ_i + ε_{i,j,k,t}
  - Where three vulnerabilities enter simultaneously: either (low relative cash, small size, high leverage) or (high liquidity gap, small size, high leverage).
  - X_i includes controls: book-to-market ratio, Tobin’s Q, cash-flow-to-total assets, and a pandemic-related revenue shock proxy.
  - HighVul X variable for volatility interaction:
    - HighVIX_i is dummy = 1 whenever daily VIX > 80th percentile of VIX distribution (February 2020 to June 2020).
    - Captures time-varying effect of extreme volatility on firms with different characteristics.
  - Fixed effects:
    - θ_{S,t} are country-date fixed effects,
    - λ_i are industry fixed effects (2-digits SIC).
  - Standard errors clustered at industry and country-date levels.
  - Coefficients of interest: β_v.
- Enriched specification studying direct vs indirect interventions:
  - AR_i,j,k,t = Σ_{v=1 to 3} β_v HighVul_{...} + Σ_{v=1 to 3} σ_v HighVul_{...} X_i + Σ_{v=1 to 3} η_v HighVul_{...} DirInterv_{k,t} + Σ_{v=1 to 3} ω_v HighVul_{...} X_i DirInterv_{k,t} + ρ CBPolicy + λ CBPolicy X_i + θ_{k,t} + λ_i + ε
  - DirInterv_{k,t} is a dummy = 1 for policy intervention days when the set of interventions included at least one that targeted the corporate sector directly.
  - Coefficients of interest: η_v, measuring the differential effect of announcements that included direct interventions on firms with high vulnerability relative to announcements with only indirect interventions.

### Data, timing, and implementation details
- Period for pooled intervention-day analysis: February 3, 2020 to June 30, 2020.
- Dependent variable construction: average abnormal returns over the announcement day and the following trading day.
- Controls and interactions:
  - Book-to-market ratio, Tobin’s Q, cash-flow-to-total assets, pandemic-related revenue shock proxy.
  - VIX interaction using 80th percentile threshold (February 2020–June 2020).
- Fixed effects and clustering:
  - Country-date fixed effects (θ_{k,t}),
  - Industry fixed effects (2-digit SIC, λ_i),
  - Standard errors clustered at industry and country-date levels.
- The source provides Online Annex Table 3.6.2 listing announcement dates and main policy interventions used in the econometric analysis of Chapter 3.

### Key parameters and inference targets
- Cross-section CAR analysis: coefficient β (differential impact on vulnerable firms).
- Policy intervention pooled models: coefficients β_v, σ_v, η_v, ω_v, ρ, λ capture:
  - β_v: baseline differential effect of vulnerabilities on announcement-day abnormal returns.
  - η_v: additional differential effect when announcements include at least one direct corporate-sector intervention.
  - σ_v and ω_v: interactions with firm characteristics X_i.
  - ρ and λ: effects of CBPolicy and its interaction with firm controls.
- Robustness: abnormal returns computed with three-factor Fama-French model instead of CAPM produce robust results.

*Online Annex 3.6, Global Financial Stability Report — Corporate Funding (PDF chapter content).*

---


_Source: https://www.imf.org/-/media/files/publications/gfsr/2020/october/english/onlineannex31.pdf_