## wpiea2019141

## Source details

**Canonical URL:** [wpiea2019141](https://www.imf.org/-/media/files/publications/wp/2019/wpiea2019141.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/wp/2019/wpiea2019141.pdf.md)
- [Structured JSON version](/-/media/files/publications/wp/2019/wpiea2019141.pdf.json)

---

### Data: scope and construction
- Sample covers 21 EMDEs over the period 2000 to 2018.
- Daily sovereign spread data: JP Morgan’s Emerging Market Bond Index – Global (EMBI-G), period January 3 2000 – December 31 2018.
- EMBI spreads retrieved from Datastream; spreads measured in basis points and reported as stripped spreads (excluding collateral and guarantees).
- IMF supported program dates obtained from the IMF’s Monitoring of Fund Arrangements (MONA) database.
- 75th percentile of the sample EMBI distribution equals 420 basis points.

### Fiscal austerity announcements database
- Constructed from Dow Jones’ Factiva news articles (English, Spanish, Portuguese translations by source outlets), using text mining (R package “tm”) and manual validation.
- Inclusion criterion: announcements that represent a net fiscal consolidation (an improvement in the fiscal balance); revenue-neutral or offsetting measures excluded.
- Cross-checked against the database constructed by David and Leigh (2018) and endogenous policy actions in that work.
- Total announcements identified: 453.
  - Announcements by executive: 355.
  - Announcements by congress: 98.
- Announcements unevenly clustered, with concentration around 2002-2004 (coinciding with several IMF supported programs and EMBI peaks).

### EMBI spreads: descriptive facts
- EMBI index computed as market-cap weighted average of eligible sovereign and quasi-sovereign bond spreads relative to U.S. government bonds of comparable maturity.
- Spreads vary considerably across sample: lowest realizations negative for some European economies; highest exceeding several percentage points in many Latin American economies.
- Spread volatility concentrated around 2001-03 and 2008 for several sample countries.
- Only 4 countries in the sample did not have an IMF supported program between 2000 and 2018: Chile, India, Slovak Republic, and South Africa.

### Method: local projections (daily horizons)
- LP framework used to project daily changes in EMBI spreads for horizons h = 0,..,30 days.
- Benchmark specification (for each horizon h):
  r_{i,t+h} − r_{i,t−1} = α_{i,h} + γ_{t,h} + β_h D_{i,t} + δ X_{i,t} + ε_{i,t+h}
  - r_{i,t+h}: EMBI sovereign spreads in basis points.
  - D_{i,t}: dummy = 1 on day of fiscal consolidation announcement.
  - X_{i,t}: vector including seven lags of daily changes in EMBI spreads.
  - Country fixed effects α_{i,h} and time fixed effects γ_{t,h} included.
- State-dependent specification splits responses according to indicator S^j_{i,t−1} for scenarios j:
  - Scenario (i) S_level_{i,t−1} = 1 if EMBI spread ≥ 75th percentile (420 basis points).
  - Scenario (ii) S_IMF_{i,t−1} = 1 if country is under an IMF supported program (also 1 for the year before IMF board approval).
  - Scenario (iii) S_level,IMF_{i,t−1} = 1 if both S_level and S_IMF equal 1.
- Confidence bands based on Driscoll-Kraay (1998) standard errors (robust to heteroskedasticity, serial and cross-sectional correlation).

### Results: effects of austerity announcements on sovereign spreads (daily LP)
- Baseline (all announcements): austerity announcements tend to reduce spreads, but effect is economically small and barely statistically significant: reduction of 5 basis points after 30 days.
- By branch of government:
  - Announcements by congress: spreads fall significantly, by around 15 basis points within a 30 day window after the announcement.
  - Announcements by executive: spreads barely decline.
  - Interpretation: investor confidence improves materially only when consolidation measures are approved by congress.
- Scenario (i) — high sovereign stress (EMBI ≥ 420 bps):
  - Following a fiscal consolidation announcement (any branch), spreads decline by more than 15 basis points in the two weeks following the announcement.
  - At 30 days, reductions are more persistent and significantly larger when announcements come from congress; executive-origin announcements show small and not statistically significant differences relative to baseline at 30 days.
- Scenario (ii) — under an IMF supported program:
  - Having an IMF program is associated with larger and more persistent reductions in spreads after consolidation announcements compared to announcements without IMF program support.
  - Reductions are significantly larger when announcements are made by congress.
  - Caveat: cannot fully distinguish whether markets react to fiscal measures per se or to perceived likelihood of IMF program approval/continuation.
- Scenario (iii) — high stress combined with IMF program:
  - Spreads decline when announcements are made in episodes of high perceived risk regardless of IMF program participation; reductions are larger and more significant than baseline.
  - Difference in 30-day spread reduction with and without IMF assistance is negligible when considering all announcements or executive announcements.
  - For congress-approved packages, reductions are larger and more persistent when countries have elevated spreads and are under an IMF agreement.

### Key implications from daily analysis
- Fiscal consolidation announcements can ameliorate sovereign default risk perceptions, particularly when:
  - Announcements occur during episodes of elevated sovereign spreads, and
  - Announcements reflect congressional approval of consolidation packages.
- IMF program participation amplifies the spread-reducing effects of consolidation announcements, but reductions in spreads in high-risk episodes occur even without IMF support.
- Motivates assessing whether reduced borrowing costs following successful fiscal announcements translate into smaller output losses; explored in the panel VAR analysis below.

### Method: panel VAR specification and identification (monthly)
- Panel of 21 EMDEs at monthly frequency.
- Monthly variables:
  - D_i,t : fiscal consolidation announcement dates (dummy = 1 in month of announcement).
  - ∆r_i,t : monthly change in EMBI spreads (average over the month).
  - ∆y_i,t : log change in monthly economic activity indicator (industrial production or other monthly volume indicators; seasonally adjusted; Haver Analytics).
- Structural form and timing assumptions:
  - Standard Cholesky timing restriction: fiscal announcement variable ordered first, EMBI spreads ordered second, output ordered third.
  - Assumption a_i,2,3 = 0 (no contemporaneous effect of output on EMBI spreads).
  - Austerity announcement shock denoted ε^1_i,t.
- Counterfactual construction (following Bachmann and Sims (2012)):
  - Hold responses of EMBI spreads to a fiscal announcement fixed at zero at all forecast horizons: IRF_2(1,h) = 0.
  - Compute hypothetical EMBI spread shocks ε^2_i,h recursively to enforce IRF_2(1,h) = 0 and derive modified impulse responses ˜IRF_k(1,h).
  - Difference between IRF_k(1,h) and ˜IRF_k(1,h) quantifies the role of sovereign spread reductions in transmission.

### Estimation and inference procedure (panel VAR)
- Allow coefficients A_i,j to vary with EMBI level: A_i,j = β_i,j + ς_i,j * EMBI level_{i,t} for j = 1,...,p.
- Estimation:
  - Each equation estimated by OLS with country fixed effects.
  - Lag length: 6 lags, following the Schwartz Criterion.
  - Use Runkle (1987) bootstrap to account for panel format and interaction terms.
- Bootstrap algorithm steps:
  - Generate 1,000 bootstrap replications ˆA_i using equation (9) and selected EMBI spread levels.
  - Simulate variables period-by-period interacting with interaction terms; re-estimate system coefficients on artificial sample.
  - Compute IRFs 1,000 times for each variable k* to the first structural shock at horizons h = 1,...,H.
  - Construct 1,000 adjusted impulse responses holding EMBI spread changes fixed at zero at each horizon h.
  - Use bootstrap simulations to calculate empirical distribution for the difference between baseline and counterfactual scenarios. Draw 90 percent confidence intervals from simulated estimates.

### Key empirical results (panel VAR, unconditioned)
- Sovereign spreads:
  - Decline immediately following congressional announcements by around 20 basis points on impact.
  - Confidence bands widen after three months; response not significantly different from zero thereafter.
- Output response:
  - Output contracts by around 1 percent 6 months after the announcement and stabilizes thereafter.
  - Cumulative response of output is significantly different from zero from the sixth month onward.
- Role of spreads:
  - Holding EMBI spreads fixed at zero yields a larger contraction in output at most horizons.
  - Difference between baseline and counterfactual output responses is statistically significant but economically small: around 0.3 percentage points in a 12 month window.

### State-dependent (conditioned) results by EMBI spread level (panel VAR)
- High perceived sovereign risk (EMBI at or above 75th percentile):
  - Spreads significantly decline by around 100 basis points in a 12 month window after announcements by congress.
  - Consolidation announcement followed by a large and protracted output loss of around 4 percent.
  - Holding the response of sovereign spreads fixed at zero is equivalent to having a fiscal shock that is 50 percent more intense than in the baseline.
  - Reduction in the actual output response is almost 50 percent lower than the counterfactual response (i.e., confidence effects cut the output decline by almost 50 percent).
- Low perceived sovereign risk (EMBI below 25th percentile):
  - Austerity announcements do not have a significant effect on spreads.

### Simulations and scenario comparisons
- Contribution of increased investor confidence (lower sovereign spreads) increases with the level of EMBI spreads.
- Countries with higher spread levels benefit the most from credible austerity packages.
- Slope of the output response curve becomes steeper above the 75th percentile, indicating larger confidence effects at very high spread levels.
- Under IMF-supported programs, the increased importance of confidence effects with higher spread levels becomes even more pronounced; the change in the slope at high spread levels is remarkable under IMF support.
- Overall interpretation: results do not indicate “expansionary” fiscal consolidations. Contraction in GDP is less pronounced when sovereign spreads are high, especially if the country is undertaking an IMF-supported adjustment program.

### Main conclusions and policy-relevant implications
- New dataset on fiscal consolidation announcements for 21 EMDEs since the early 2000s using local newspaper articles.
  - Findings: (i) obtaining congressional approval for consolidation is difficult; (ii) consolidation packages tend to occur during episodes of high global or regional financial stress; (iii) consolidation efforts frequently coincided with periods when countries had an agreement for IMF program assistance.
- Fiscal announcements aimed at improving the budget balance tend to reduce EMBI spreads, particularly when approved by congress; investors mainly react to austerity proposals after legislative approval.
- Initial conditions matter: reductions in spreads are larger and more persistent when announcements occur during elevated sovereign spreads or under an IMF program.
- Approval of austerity packages by congress in periods of increased sovereign risk yields payoffs via reduced spreads.
- Confidence effects (lower sovereign spreads) substantially reduce output losses from fiscal consolidations compared to counterfactuals without spread reductions; the role of confidence effects increases with the level of spreads, particularly under IMF-supported adjustment programs.
- Policy takeaway: credible consolidation efforts in conditions of severe fiscal stress can be rewarded by financial markets, and confidence effects are crucial in lowering the drag on economic activity following austerity measures.

*Source: wpiea2019141 - Section 5 concludes.*

### Section 5 concludes.

### Section 5 concludes.

### Data: scope and construction
- Sample covers 21 EMDEs over the period 2000 to 2018.
- Daily sovereign spread data: JP Morgan’s Emerging Market Bond Index – Global (EMBI-G), period January 3 2000 – December 31 2018.
- EMBI spreads retrieved from Datastream; spreads measured in basis points and reported as stripped spreads (excluding collateral and guarantees).
- IMF supported program dates obtained from the IMF’s Monitoring of Fund Arrangements (MONA) database.
- 75th percentile of the sample EMBI distribution equals 420 basis points.

### Fiscal austerity announcements database
- Database constructed from Dow Jones’ Factiva news articles (English, Spanish, Portuguese translations by source outlets), using text mining (R package “tm”) and manual validation.
- Inclusion criterion: announcements that represent a net fiscal consolidation (an improvement in the fiscal balance); revenue-neutral or offsetting measures excluded.
- Cross-checked against the database constructed by David and Leigh (2018) and endogenous policy actions in that work.
- Total announcements identified: 453.
  - Announcements by executive: 355.
  - Announcements by congress: 98.
- Announcements unevenly clustered, with concentration around 2002-2004 (coinciding with several IMF supported programs and EMBI peaks).

### EMBI spreads: descriptive facts
- EMBI index computed as market-cap weighted average of eligible sovereign and quasi-sovereign bond spreads relative to U.S. government bonds of comparable maturity.
- Spreads vary considerably across sample: lowest realizations negative for some European economies; highest exceeding several percentage points in many Latin American economies.
- Spread volatility concentrated around 2001-03 and 2008 for several sample countries.
- Only 4 countries in the sample did not have an IMF supported program between 2000 and 2018: Chile, India, Slovak Republic, and South Africa.

### Method: local projections (LP) for daily horizons
- LP framework used to project daily changes in EMBI spreads for horizons h = 0,..,30 days.
- Benchmark specification (for each horizon h):
  r_{i,t+h} − r_{i,t−1} = α_{i,h} + γ_{t,h} + β_h D_{i,t} + δ X_{i,t} + ε_{i,t+h}
  - r_{i,t+h}: EMBI sovereign spreads in basis points.
  - D_{i,t}: dummy = 1 on day of fiscal consolidation announcement.
  - X_{i,t}: vector including seven lags of daily changes in EMBI spreads.
  - Country fixed effects α_{i,h} and time fixed effects γ_{t,h} included.
- State-dependent specification splits responses according to indicator S^j_{i,t−1} for scenarios j:
  - Scenario (i) S_level_{i,t−1} = 1 if EMBI spread ≥ 75th percentile (420 basis points).
  - Scenario (ii) S_IMF_{i,t−1} = 1 if country is under an IMF supported program (also 1 for the year before IMF board approval).
  - Scenario (iii) S_level,IMF_{i,t−1} = 1 if both S_level and S_IMF equal 1.
- Confidence bands based on Driscoll-Kraay (1998) standard errors (robust to heteroskedasticity, serial and cross-sectional correlation).

### Results: effects of austerity announcements on sovereign spreads
- Baseline (all announcements): austerity announcements tend to reduce spreads, but effect is economically small and barely statistically significant: reduction of 5 basis points after 30 days.
- By branch of government:
  - Announcements by congress: spreads fall significantly, by around 15 basis points within a 30 day window after the announcement.
  - Announcements by executive: spreads barely decline.
  - Interpretation: investor confidence improves materially only when consolidation measures are approved by congress.
- Scenario (i) — high sovereign stress (EMBI ≥ 420 bps):
  - Following a fiscal consolidation announcement (any branch), spreads decline by more than 15 basis points in the two weeks following the announcement.
  - At 30 days, reductions are more persistent and significantly larger when announcements come from congress; executive-origin announcements show small and not statistically significant differences relative to baseline at 30 days.
- Scenario (ii) — under an IMF supported program:
  - Having an IMF program is associated with larger and more persistent reductions in spreads after consolidation announcements compared to announcements without IMF program support.
  - Reductions are significantly larger when announcements are made by congress.
  - Caveat noted: cannot fully distinguish whether markets react to fiscal measures per se or to perceived likelihood of IMF program approval/continuation.
- Scenario (iii) — high stress combined with IMF program:
  - Spreads decline when announcements are made in episodes of high perceived risk regardless of IMF program participation; reductions are larger and more significant than baseline.
  - Difference in 30-day spread reduction with and without IMF assistance is negligible when considering all announcements or executive announcements.
  - However, for congress-approved packages, reductions are larger and more persistent when countries have elevated spreads and are under an IMF agreement.

### Key implications drawn
- Fiscal consolidation announcements can ameliorate sovereign default risk perceptions, particularly when:
  - Announcements occur during episodes of elevated sovereign spreads, and
  - Announcements reflect congressional approval of consolidation packages.
- IMF program participation amplifies the spread-reducing effects of consolidation announcements, but reductions in spreads in high-risk episodes occur even without IMF support.
- The results motivate assessing whether reduced borrowing costs following successful fiscal announcements translate into smaller output losses; that transmission (confidence → lower spreads → economic activity) is explored in subsequent analysis.

*Source: wpiea2019141 - Section 5 concludes.*

### 4.1  Identifying the effects of austerity announcement shocks:

### 4.1 Identifying the effects of austerity announcement shocks: a panel VAR approach

### Methodology: panel VAR specification and identification
- Sample and frequency:
  - Panel of 21 EMDEs.
  - Monthly frequency.
- Variables included (monthly):
  - D_i,t : fiscal consolidation announcement dates (dummy = 1 in month of announcement).
  - ∆r_i,t : monthly change in EMBI spreads (average over the month).
  - ∆y_i,t : log change in monthly economic activity indicator (industrial production or other monthly volume indicators; seasonally adjusted; Haver Analytics).
- Structural form and timing assumptions:
  - Standard Cholesky timing restriction: fiscal announcement variable ordered first, EMBI spreads ordered second, output ordered third.
  - Assumption a_i,2,3 = 0 (no contemporaneous effect of output on EMBI spreads) to identify a_i,3,2 and a_i,2,3.
  - Austerity announcement shock denoted ε^1_i,t.
- Channels of transmission:
  - Direct contemporaneous effect on output via a_i,3,1 and dynamic effects via A_i,j matrices.
  - Indirect effect through contemporaneous movement in spreads via a_i,2,1 and subsequent effect of spreads on output via a_i,3,2.
  - Spreads can propagate fiscal shocks if they respond at any horizon and lagged spread coefficients in the output equation are significant.
- Counterfactual construction (following Bachmann and Sims (2012)):
  - Hold responses of EMBI spreads to a fiscal announcement fixed at zero at all forecast horizons: IRF_2(1,h) = 0.
  - Construct hypothetical sequence of EMBI spread shocks ε^2_i,h that force IRF_2(1,h) = 0, using:
    - On impact (h = 1): ε^2_i,1 = − A^{-1}_{i,0}(2,1) / A^{-1}_{i,0}(2,2).
    - Recursively for h = 2,...,H via the stated formula.
  - Modified impulse responses computed as ˜IRF_k(1,h) = IRF_k(1,h) + Σ_{j=1}^h Λ^{h−j} A^{-1}_{i,0}(2,1) ε^2_{i,j}, for k = 1,2,3.
  - Difference between IRF_k(1,h) and ˜IRF_k(1,h) quantifies the role of sovereign spread reductions in transmission.

### Estimation and inference procedure
- Allow coefficients A_i,j to vary with EMBI level: A_i,j = β_i,j + ς_i,j * EMBI level_{i,t} for j = 1,...,p.
- Estimation details:
  - Each equation estimated by OLS with country fixed effects.
  - Lag length: 6 lags, following the Schwartz Criterion.
  - Use Runkle (1987) bootstrap to account for panel format and interaction terms.
- Bootstrap algorithm (procedural steps):
  - 1) Estimate PVAR(p) in equation (4) and generate 1,000 bootstrap replications ˆA_i using equation (9) and selected EMBI spread levels.
  - 2) After simulating first period for all variables, interact variables with interaction terms in (9) and repeat step 1 for t = 2,...,T and i = 1,...,N.
  - 3) Use artificial sample with interaction variables to re-estimate system coefficients. Compute IRFs 1,000 times for each generated variable k* to the first structural shock at horizons h = 1,...,H.
  - 4) Construct 1,000 adjusted impulse responses holding EMBI spread changes fixed at zero at each horizon h.
  - 5) Use bootstrap simulations to calculate empirical distribution for the difference between baseline and counterfactual scenarios. Draw 90 percent confidence intervals from simulated estimates.

### Key empirical results (unconditioned)
- Sovereign spreads:
  - Decline immediately following congressional announcements by around 20 basis points on impact.
  - Confidence bands widen after three months; response not significantly different from zero thereafter.
- Output response:
  - Output contracts by around 1 percent 6 months after the announcement and stabilizes thereafter.
  - Cumulative response of output is significantly different from zero from the sixth month onward.
- Role of spreads:
  - Holding EMBI spreads fixed at zero yields a larger contraction in output at most horizons.
  - Difference between baseline and counterfactual output responses is statistically significant but economically small: around 0.3 percentage points in a 12 month window.

### State-dependent (conditioned) results by EMBI spread level
- High perceived sovereign risk (EMBI at or above 75th percentile):
  - Spreads significantly decline by around 100 basis points in a 12 month window after announcements by congress.
  - Consolidation announcement followed by a large and protracted output loss of around 4 percent.
  - Holding the response of sovereign spreads fixed at zero is equivalent to having a fiscal shock that is 50 percent more intense than in the baseline.
  - Reduction in the actual output response is almost 50 percent lower than the counterfactual response (i.e., confidence effects cut the output decline by almost 50 percent).
- Low perceived sovereign risk (EMBI below 25th percentile):
  - Austerity announcements do not have a significant effect on spreads.

### Simulations and scenario comparisons
- Figure 9 summary (ratio of counterfactual to baseline output response across EMBI percentiles):
  - Contribution of increased investor confidence (lower sovereign spreads) increases with the level of EMBI spreads.
  - Countries with higher spread levels benefit the most from credible austerity packages.
  - Slope of the output response curve becomes steeper above the 75th percentile, indicating larger confidence effects at very high spread levels.
- IMF-supported program scenario:
  - Under IMF-supported programs, the increased importance of confidence effects with higher spread levels becomes even more pronounced.
  - The change in the slope at high spread levels is remarkable under IMF support.
- Overall interpretation:
  - Results do not indicate “expansionary” fiscal consolidations.
  - Contraction in GDP is less pronounced when sovereign spreads are high, especially if the country is undertaking an IMF-supported adjustment program.

### Main conclusions and contributions
- Data contribution:
  - New dataset on fiscal consolidation announcements for 21 EMDEs since the early 2000s using local newspaper articles.
  - Findings from dataset: (i) obtaining congressional approval for consolidation is difficult; (ii) consolidation packages tend to occur during episodes of high global or regional financial stress; (iii) consolidation efforts frequently coincided with periods when countries had an agreement for IMF program assistance.
- Empirical findings on spreads:
  - Fiscal announcements aimed at improving the budget balance tend to reduce EMBI spreads, particularly when approved by congress.
  - Investors mainly react to austerity proposals after legislative approval.
  - Initial conditions matter: reductions in spreads are larger and more persistent when announcements occur during elevated sovereign spreads or under an IMF program.
- Policy-relevant implication on output losses:
  - Approval of austerity packages by congress in periods of increased sovereign risk yields payoffs via reduced spreads.
  - Confidence effects (lower sovereign spreads) substantially reduce output losses from fiscal consolidations compared to counterfactuals without spread reductions.
  - The role of confidence effects increases with the level of spreads, particularly under IMF-supported adjustment programs.
- Comparison with other literature:
  - Results contrast with Born, Müller, and Pfeifer (2019) for a sample of 38 countries; alternative identification using announcement dates and higher-frequency timing restrictions yields different conclusions about how fiscal consolidations affect sovereign spreads.
- Policy takeaway:
  - Credible consolidation efforts in conditions of severe fiscal stress can be rewarded by financial markets, and confidence effects are crucial in lowering the drag on economic activity following austerity measures.

*Source: wpiea2019141 - 4.1 Identifying the effects of austerity announcement shocks: a panel VAR approach (IMF PDF).*

---


_Source: https://www.imf.org/-/media/files/publications/wp/2019/wpiea2019141.pdf_