## Online Annex 2.1

## Source details

**Canonical URL:** [Online Annex 2.1](https://www.imf.org/-/media/files/publications/esr/2021/english/onlineannex21.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/esr/2021/english/onlineannex21.pdf.md)
- [Structured JSON version](/-/media/files/publications/esr/2021/english/onlineannex21.pdf.json)

---

### Data Sources
- Primary macroeconomic data: July 2021 World Economic Outlook Update.
- Net international investment position: Lane and Milesi-Ferretti (2018).
- VIX/VXO: Federal Reserve Economic Data.
- Financial crisis occurrences: Laeven and Valencia (2020).
- Terms-of-trade index: Gruss and Kebhaj (2019).
- Sudden stops: David and Gonçalves (2019).
- Narrative fiscal shocks:
  - Advanced economies: Alesina and others (2018) for 16 OECD economies for 1978–2014, with The Netherlands added and the sample extended through 2019 for 17 advanced economies.
  - Emerging market and developing economies: Carriere-Swallow, David, and Leigh (2021) for 14 countries in Latin America and the Caribbean for 1989–2016, with China and India added and the sample extended through 2019 for 16 EMDEs.
- Additional variables and computation notes:
  - Cyclical component of real GDP computed using an HP filter (λ = 100), following Jordà and Taylor (2016).
  - Terms-of-trade index from Gruss and Kebhaj (2019).
  - VXO used as proxy for global risk appetite (measure similar to VIX).
  - Crisis dummies from Laeven and Valencia (2020); sudden stop dummy from David and Gonçalves (2019).

### Robustness Checks and Extensions
- Focus: impact of a fiscal consolidation (fiscal shock of 1 percent of GDP) on the current account balance at the three-year horizon (h = 3).
- Baseline and alternative estimators reported; all specifications include country and time fixed effects.
- Key table estimates (point estimates for β at h = 3 and accompanying standard errors):
  - β h=3: 0.60*** 0.55*** 1.11*** 1.10*** 0.58*** 0.67*** 0.60*** 0.69***
  - Standard errors (parentheses): (0.12)(0.08)(0.20)(0.24)(0.13)(0.26)(0.15)(0.18)
- Observations (as reported): 1199 1120 724 842 1199 1199 1260 1260
- First Stage F-statistic: 85.22 (5)
- Notes on estimation methods and inference:
  - Outlier robustness: observations with Cook’s distance > 4/N removed; removal does not significantly alter results (column 2).
  - Instrumental Variable (IV) approach (Ramey and Zubairy (2018)): narrative fiscal shocks used as instruments for changes in the cyclically adjusted primary balance.
    - First stage is strong; Kleibergen-Paap Wald F-statistics reported exceed the rule-of-thumb for instrument strength.
    - Second-stage IV point estimates are significant and larger than baseline.
  - Augmented inverse propensity score weighting (AIPW) estimator (Jordà and Taylor (2016)) to address predictability of narrative shocks:
    - Narrative shock converted to binary treatment (1 for fiscal consolidation, 0 otherwise).
    - Probit model for treatment probability uses rich set of determinants: two lags of treatment, two lags of current account balance, two lags of net international investment position, two lags of government-debt-to-GDP ratio, cyclical component of real GDP, change in terms-of-trade index, lagged VIX/VXO, crisis and sudden stop dummies, and country fixed effects.
    - AIPW yields a significant and slightly higher point estimate of 1.1 percent, versus 0.6 percent in the baseline specification (noting the shock is transformed into a binary treatment and results are rescaled for comparability).
- Standard error methods:
  - Driscoll-Kraay standard errors for columns (1)–(3) and (5)–(7).
  - Kleibergen-Paap Wald F-statistic for IV estimator.
  - Empirical sandwich standard errors clustered by country for AIPW estimator.

### Results for Sample Splits and Heterogeneity Tests
- Time-split analysis specification (equation (2.1.1)):
  - Fiscal shocks split into two periods: first sample 1978–2009 (pre-global financial crisis) and second sample 2010–19 (post-2009).
  - Tests for equality of estimated coefficients across periods show they are statistically distinguishable at conventional levels; p-values reported in Online Annex Figure 2.1.1.
  - Findings:
    - Evidence of a larger effect of fiscal consolidation on the current account after the global financial crisis.
    - Reestimating with different components of aggregate demand as dependent variables finds significantly larger effects of fiscal consolidations in the most recent decade on real investment and real GDP (over the three- to five-year horizon).
    - The effect of fiscal consolidations on the real effective exchange rate is not statistically distinguishable over the two sample periods.
- Cross-group heterogeneity (AEs vs EMDEs):
  - Equation (2.1.1) reestimated splitting fiscal shocks by group (advanced economies and emerging market and developing economies).
  - Estimated response of the current account balance is larger for emerging market and developing economies, but estimates are less precise and not statistically distinguishable from the result for advanced economies (column 5).
- Composition of consolidation (spending-based vs tax-based):
  - Spending-based adjustments defined as those where the budgetary impact of spending changes exceeds that of tax changes.
  - Separate estimates using spending-based and tax-based narrative shocks (columns (6) and (7)) indicate both tax-based and spending-based fiscal consolidations raise the current account balance by roughly the same magnitude as the baseline specification.
- Graphical test results (Online Annex Figure 2.1.1):
  - Pre– and post–Global Financial Crisis (GFC) estimated impulse responses are statistically distinguishable over most time horizons for:
    - Current account
    - GDP
    - Real investment
    - Real imports
  - Dashed line in figure indicates 10 percent significance level; p-values shown for null hypothesis of equality in estimated responses.

### Model Simulations and Global Feedbacks (summary of figure captions)
- G20 Model simulations (Online Annex Figure 2.1.2) illustrate:
  - When all economies implement fiscal consolidation at the same time:
    - The exchange rate for an individual economy depreciates by less or even appreciates.
    - The global real interest rate declines.
  - Figure panels:
    - 1. Real Exchange Rate (Individual fiscal consolidation)
    - 2. World Real Interest Rate

### Key Quantitative Findings (preserved exactly)
- Baseline point estimate of a fiscal shock of 1 percent of GDP on the current account at h = 3: 0.60*** (standard error (0.12)).
- AIPW estimator point estimate (rescaled for comparability): 1.10*** (standard error (0.24)).
- IV estimator point estimate: 1.11*** (standard error (0.20)).
- Other reported point estimates for variants: 0.55*** (0.08); 0.58*** (0.13); 0.67*** (0.26); 0.60*** (0.15); 0.69*** (0.18).
- Observations reported across specifications: 1199 1120 724 842 1199 1199 1260 1260.
- First Stage F-statistic reported: 85.22 (5).

*Source: Online Annex 2.1, "Empirical Analysis: Data Sources, Robustness Checks, and Tests of Changes in Results over Time," IMF 2021.*

---


_Source: https://www.imf.org/-/media/files/publications/esr/2021/english/onlineannex21.pdf_