## sovereignannex

## Source details

**Canonical URL:** [sovereignannex](https://www.imf.org/-/media/files/publications/reo/whd/2019/october/english/sovereignannex.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/reo/whd/2019/october/english/sovereignannex.pdf.md)
- [Structured JSON version](/-/media/files/publications/reo/whd/2019/october/english/sovereignannex.pdf.json)

---

### Estimation using Local Projections (Annex 1)
- Benchmark LP specification for horizons (h = 0, . . , 30) in days:
  - Dependent variable: r_{i,t+h} − r_{i,t−1} (EMBIG sovereign spreads in basis points).
  - Key regressor: D_{i,t} (dummy = 1 on day of fiscal consolidation announcement).
  - Controls: seven lags of daily changes in EMBIG spreads (X_{i,t}).
  - Fixed effects: country (α_{i,h}) and time (γ_{t,h}).
  - Impulse responses constructed from estimated β_h coefficients; confidence bands from estimated standard errors.
- Annual-data note:
  - Dummy D_{i,t} takes the value of one in the year of the announcement and the horizon (h) is equal to three years.
- Advantages and uses:
  - LP flexible for panel structure and unconstrained impulse response shapes; less sensitive to misspecification.
  - Can handle non-linearities and state dependency.
- State-dependent LP specification (Equation (2)):
  - Indicator S_{i,t−1}^j ∈ {0,1} conditions responses on state j.
  - Scenario (i): S_{i,t−1}^{high} = 1 if EMBIG spread is at or above the 75th percentile of the sample distribution (420 basis points).
  - Scenario (ii): S_{i,t−1}^{IMF} = 1 if the country is under an IMF supported program; also set to 1 for the year before the board approval date of the IMF program.

### Data and sample (Annex 1)
- Daily data on sovereign bond spreads for 11 Latin American and Caribbean economies between January 3, 2000, and December 31, 2018.
- Spread series: JP Morgan’s Emerging Market Bond Index – Global (EMBIG) database; only stripped spreads used.
  - Spread definition: premium paid by an emerging market over a U.S. government bond with comparable maturity features.
  - Country spread index: average of spreads of included bonds weighted by market capitalization.
- Data sources:
  - EMBIG data retrieved from Datastream.
  - Fiscal announcement dates from David, Guajardo, and Yépez (2019).
  - IMF program dates from the IMF’s Monitoring of Fund Arrangements (MONA) database.
  - Real GDP and debt-to-GDP ratios from IMF’s World Economic Outlook (WEO) database.

### Estimation using a Panel VAR (Annex 2)
- PVAR setup:
  - Panel of 11 EMDEs with three monthly variables: fiscal consolidation announcement dummy (D_{i,t}), ∆r_{i,t} (monthly change in EMBIG spreads), ∆y_{i,t} (log change in monthly economic activity indicator).
  - Fiscal announcement dummy equals one in the month of the announcement; EMBIG spreads averaged over the month.
  - Economic activity indicators: monthly industrial production or other volume indicators, seasonally adjusted, from Haver Analytics.
  - Jamaica exception: monthly interpolation of quarterly GDP series is used.
- Identification assumptions:
  - Cholesky ordering with fiscal policy variable ordered first (fiscal announcements do not react within the month to other shocks).
  - Additional restriction: α_{i,2,3} set to zero so changes in EMBIG spreads ordered second and output third.
  - Structural shocks: ε_{i,t,1} (fiscal announcement shock), ε_{i,t,2} (sovereign spread shock), ε_{i,t,3} (residual output shock).
- Counterfactual “shut-off” analysis:
  - To isolate role of sovereign spreads as transmission mechanism, construct hypothetical impulse response of output to a fiscal consolidation announcement by holding changes in EMBIG spreads fixed at zero at all forecast horizons.
  - Compare hypothetical response to baseline response to quantify importance of changes in sovereign borrowing costs for transmission of fiscal consolidation announcement shocks.

### Nonlinearity, initial conditions, and varying-coefficient PVAR
- Allow coefficients in A_{i,j} matrices to vary depending on:
  - Level of EMBIG spreads (full empirical distribution of EMBIG spread levels considered).
  - Whether consolidation announcement was done under IMF program support.
- Estimation details:
  - Each equation estimated by ordinary least squares (OLS) with country fixed effects and 6 lags, following the Schwartz Criterion.
  - Because impulse responses are non-linear functions of OLS estimates, Runkle (1987) bootstrapping method employed for inference to adjust for panel format and interaction terms.
  - References to methodological parallels: Burnside, Eichenbaum, and Fisher (2004); Cavallo (2005); Blanchard and Perotti (2002); Bachmann and Sims (2012).

*International Monetary Fund | October 2019 — Annex 1 and Annex 2, SOVEREIGN SPREADS AND FISCAL CONSOLIDATIONS*

---


_Source: https://www.imf.org/-/media/files/publications/reo/whd/2019/october/english/sovereignannex.pdf_