## wpiea2020166-print-pdf

## Source details

**Canonical URL:** [wpiea2020166-print-pdf](https://www.imf.org/-/media/files/publications/wp/2020/english/wpiea2020166-print-pdf.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/wp/2020/english/wpiea2020166-print-pdf.pdf.md)
- [Structured JSON version](/-/media/files/publications/wp/2020/english/wpiea2020166-print-pdf.pdf.json)

---

### First-stage regression: linking spreads to fundamentals
- Purpose: link the observed spread level to economic fundamentals X_it via an OLS first-stage regression (equation (1)).
- Fundamentals included:
  - Country-specific: debt-to-GDP ratio, terms of trade, inflation, institutional quality (simple average of “government effectiveness” and “regulatory quality” from the World Bank’s Worldwide Governance Indicators).
  - Global: VIX uncertainty index, S&P500 stock returns.
- Excluded / robustness notes:
  - Credit ratings and IMF/World Bank debt sustainability assessments are not included as regressors (treated as potentially non-fundamental and biased).
  - Federal funds rate, current account deficit, FX reserve levels, and government deficits were tested but did not add materially once VIX and US stock returns were included; government consumption vs investment breakdown not examined due to data limits.
  - Real GDP growth is not in the baseline first-stage (no consensus on its importance), but second-stage results are robust to including it in alternative specifications.
- Interpretation:
  - Remainder variation after fundamentals suggests a significant role for “sentiment” in spreads.
  - Non-fundamental deviations can reflect geopolitics, optimistic views by ratings/IFIs, or investors buying narratives (examples: Argentina, Mozambique, Southern Europe).

### Table 1 — OLS regression (first-stage) key statistics and coefficients
- Dependent variable: natural log of sovereign spread (ln S_it)
- Coefficients (with t-statistics in parentheses; significance markers preserved):
  - debt-to-GDP ratio: 0.0047008 * (1.87)
  - inflation: 0.0061898 ** (2.43)
  - terms of trade: −0.0070032 *** (−3.62)
  - governance index: −1.197624 *** (−15.28)
  - VIX: 0.0501941 *** (10.36)
  - S&P 500 returns: 2.596903 *** (6.43)
  - constant: 4.915916 *** (16.44)
- Goodness of fit and sample:
  - R^2: 0.5694
  - countries: 89
  - observations: 4,651
- Note: t-statistics calculated using cluster-robust standard errors. * denotes 10% significance, ** 5%, *** 1%.

### Implication from first-stage
- Fundamentals explain a substantial share of cross-country/time variation in sovereign spreads (R^2 = 0.5694) but leave meaningful residual “mispricing” (μ_it) attributed to sentiment and other non-fundamental factors.

### Second-stage: effects on real GDP growth
- Objective: estimate the growth impact of the mispricing component μ_it via a second-stage AR(2) regression (equation (2)).
- Second-stage specification:
  - y_it = θ + γ_1 y_it−1 + γ_2 y_it−2 + δ S_it−1 + κ μ_it−` + ε_it
  - Baseline lag of the mispricing term: ` = 8 quarters (2 years); results robust to different lags or moving averages.
  - Control for S_it−1 to mitigate the working capital channel so κ captures the effect of past sentiment.
- Key quantitative findings:
  - κ (coefficient on μ_it−8): 0.206 (implied by the authors’ magnitude calculation).
  - Standard deviation of μ (one standard deviation): 0.873.
  - Impact of a one standard deviation negative mispricing (μ < 0, i.e., borrowing at a discount) on real GDP growth two years later:
    - 0.873 × 0.206 = 0.180 percentage points — being able to borrow at a “discount” of one standard deviation reduces real GDP growth by 0.180 percentage points two years later.
  - For comparison: one standard deviation increase in the spread S_it (491.815 basis points) is expected to reduce real GDP growth by 0.225 percentage points.
- Interpretation of signs:
  - A significantly positive κ implies that optimistic past sentiment (μ < 0) is followed by lower future growth; conversely, pessimistic sentiment predicts higher future growth.
  - This counterintuitive direction is robust and suggests that favorable non-fundamental borrowing conditions do not translate into stronger future growth on average.

### Mechanisms, robustness, and caveats
- Interpretation reasons:
  - Favorable borrowing conditions would be expected to boost future growth via cheaper borrowing and more room for investment; observed results point the opposite way.
  - If spreads reflect market expectations of future performance, μ < 0 would signal optimistic market expectations about future growth — again, observed results point to subsequent growth slowing.
- Methodological notes:
  - The authors place an AR(2) at the core of the growth regression (quarterly real GDP growth described well by AR(2) in prior work).
  - Controlling for S_it−1 is intended to reduce contamination from the working capital channel; remaining estimates are conservative lower bounds for the sentiment effect.
  - Outliers affect the spread standard deviation: excluding the top 5 percentile halves the spread standard deviation to 223.331 basis points (authors note this makes the reported impact conservative).

### Asymmetric effect of sovereign spread mispricing on growth (Section 5)
- Finding: The detrimental effect on future real GDP growth stems primarily from episodes where the mispricing term μ < 0 (countries borrowing at a discount relative to fundamentals).
- Magnitudes and coefficients (selected):
  - μit−8 (baseline, Table 2 column (1)): 0.2063282 (significant at 1%).
  - μit−8 × I{μit−8 < 0} (Table 2 column (2)): 0.2581765 (significant at 1%).
  - Interpretation: When μ < 0, a one standard deviation spread discount is expected to reduce growth by 0.270 percentage points — larger than the effect for the spread level itself (0.225).
- Economic interpretation: Investors’ bullish view (μ < 0) may reflect betting on an upcoming structural break not supported by fundamentals; access to unusually cheap borrowing can induce heavy indebtment that often ends poorly.

### Link with current account deficits and external borrowing
- Interaction result (Table 2 column (3)):
  - μit−8 × (CAD/GDP)it−8 = 0.0238574 (significant at 5%).
  - (CAD/GDP)it−8 alone: −0.0110134 (not significant).
- Correlation: The sample correlation between μ and CAD is 0.0335.
- Finding: The detrimental effects of borrowing at a discount are largest in countries running current account deficits; the main channel appears to be external borrowing rather than domestic government deficits.
- Government deficit channel (Table 2 column (4)):
  - μit−8 × (DEF/GDP)it−8 = 0.0118626 (not significant).
  - (DEF/GDP)it−8 = −0.018155 (not significant).
- Interpretation: Cheap external financing combined with CAD increases risk of future growth losses; current account deficits themselves are not significant unless paired with inexplicably favorable borrowing conditions.

### Out-of-sample predictive performance
- Leave-one-out RMSE results (Table 3):
  - AR(2): RMSE = 2.03; ratio to AR(2) = 1.00.
  - AR(2) + Sit−1 + κμit−8: RMSE = 1.72; ratio to AR(2) = 0.85.
  - AR(2) + Sit−1: RMSE = 2.03; ratio to AR(2) = 1.01.
- Time-based RMSE results (Table 4; averaged over T between 2015q1 and 2016q4):
  - AR(2): RMSE = 2.17; ratio to AR(2) = 1.00.
  - AR(2) + Sit−1 + κμit−8: RMSE = 1.86; ratio to AR(2) = 0.85.
  - AR(2) + Sit−1: RMSE = 2.07; ratio to AR(2) = 0.95.
- Conclusion: Including past mispricing μit−8 reduces forecast RMSE by about 15 percent relative to AR(2); adding only lagged spreads offers negligible improvement.

### Robustness and timing of effects
- Baseline lag: ` = 8 quarters (2 years). Results robust to different lag lengths and less parsimonious first-stage specifications.
- Horizon: Effect maximized at ` = 6 quarters; dynamics operate over about 4 to 14 quarters.
- Moving average robustness: MA-window [μit−14, μit−2] yields κ = 0.1548611 (p-value = 0.001).
- Alternative specifications:
  - Larger first-stage (Table A3) yields high R2 = 0.7811 and similar patterns.
  - Second-stage in Table A4 (alternative first stage): μit−8 = 0.1504278 (significant at 5%); R2 = 0.7568.
- Fiscal crises outcome: μt−` < 0 increases likelihood of future fiscal crisis at longer lags (4–7 years), consistent with average maturity of government debt of 5.6 years.

### Mechanisms and economic channels
- Possible economic channels identified:
  - Debt overhang from heavy borrowing (Myers, 1977; Krugman, 1988).
  - “Investment hangover” after optimistic booms (Beaudry, Galizia, and Portier (2018); Rognlie, Shleifer, and Simsek (2018)).
  - Real exchange rate over-valuation and lost competitiveness (Reis, 2013; Benigno and Fornaro, 2014; Benigno, Converse, and Fornaro, 2015; Gopinath et al., 2017).
- Possible political/institutional channels:
  - Misallocated funds, corruption, or consumption rather than productive investment (institutional Dutch disease; Gelb (1988)).
  - Masking of inefficiencies and postponement of reforms (Fernández-Villaverde, Garicano, and Santos (2013)).
  - Riskier policies or loosening macroprudential regulation under bullish sentiment.

### Policy implications and recommendations
- Monitoring and macroprudential policy:
  - Continuous real-time tracking of the mispricing proxy μ is feasible and may guide policy discussions.
  - Recommendation: Countercyclical macroprudential regulation that tightens during boom periods when foreign borrowing is cheap (μ < 0) — “leaning against sentiment” could yield stability gains (Flemming, L’Huillier, and Piguillem, 2019).
- Caution on anchoring fiscal rules to spreads:
  - Spreads may reflect sentiment and mispricing; anchoring rules solely to spread levels can be misleading.
  - Emphasize fundamentals alongside spread measures when assessing fiscal space (Hatchondo, Martinez, and Roch, 2017).
- Forecasting practice:
  - Incorporating past mispricing μit−8 in forecasting models can materially improve forecast accuracy for one-quarter ahead growth.

### Conclusion (summary)
- Undue optimism (borrowing at spreads lower than fundamentals justify) tends to be followed by lower economic growth and, with a further lag, increased incidence of fiscal crises.
- The harmful effect is asymmetric and concentrated in episodes of spread discounts (μ < 0), particularly when paired with current account deficits.
- Including a sovereign debt mispricing measure improves out-of-sample forecasts and has actionable policy implications for macroprudential regulation and the assessment of fiscal risks.

*Source: wpiea2020166-print-pdf.*

### 3.1    Sovereign spreads and their determinants

### 3.1    Sovereign spreads and their determinants

### First-stage regression: linking spreads to fundamentals
- Purpose: link the observed spread level to economic fundamentals X_it via an OLS first-stage regression (equation (1)).
- Fundamentals included:
  - Country-specific: debt-to-GDP ratio, terms of trade, inflation, institutional quality (simple average of “government effectiveness” and “regulatory quality” from the World Bank’s Worldwide Governance Indicators).
  - Global: VIX uncertainty index, S&P500 stock returns.
- Excluded / robustness notes:
  - Credit ratings and IMF/World Bank debt sustainability assessments are not included as regressors (treated as potentially non-fundamental and biased).
  - Federal funds rate, current account deficit, FX reserve levels, and government deficits were tested but did not add materially once VIX and US stock returns were included; government consumption vs investment breakdown not examined due to data limits.
  - Real GDP growth is not in the baseline first-stage (no consensus on its importance), but second-stage results are robust to including it in alternative specifications.
- Interpretation:
  - Remainder variation after fundamentals suggests a significant role for “sentiment” in spreads, consistent with prior literature.
  - Non-fundamental deviations can reflect geopolitics, optimistic views by ratings/IFIs, or investors buying narratives (examples: Argentina, Mozambique, Southern Europe).

### Table 1 — OLS regression (first-stage) key statistics and coefficients
- Dependent variable: natural log of sovereign spread (ln S_it)
- Coefficients (with t-statistics in parentheses; significance markers preserved):
  - debt-to-GDP ratio: 0.0047008 * (1.87)
  - inflation: 0.0061898 ** (2.43)
  - terms of trade: −0.0070032 *** (−3.62)
  - governance index: −1.197624 *** (−15.28)
  - VIX: 0.0501941 *** (10.36)
  - S&P 500 returns: 2.596903 *** (6.43)
  - constant: 4.915916 *** (16.44)
- Goodness of fit and sample:
  - R^2: 0.5694
  - countries: 89
  - observations: 4,651
- Note: t-statistics calculated using cluster-robust standard errors. * denotes 10% significance, ** 5%, *** 1%.

### Implication from first-stage
- Fundamentals explain a substantial share of cross-country/time variation in sovereign spreads (R^2 = 0.5694) but leave meaningful residual “mispricing” (μ_it) attributed to sentiment and other non-fundamental factors.

---

### Second-stage: effects on real GDP growth
- Objective: estimate the growth impact of the mispricing component μ_it via a second-stage AR(2) regression (equation (2)).
- Second-stage specification:
  - y_it = θ + γ_1 y_it−1 + γ_2 y_it−2 + δ S_it−1 + κ μ_it−` + ε_it
  - Baseline lag of the mispricing term: ` = 8 quarters (2 years); results robust to different lags or moving averages.
  - Control for S_it−1 to mitigate the working capital channel so κ captures the effect of past sentiment.
- Key quantitative findings:
  - κ (coefficient on μ_it−8): 0.206 (implied by the authors’ magnitude calculation).
  - Standard deviation of μ (one standard deviation): 0.873.
  - Impact of a one standard deviation negative mispricing (μ < 0, i.e., borrowing at a discount) on real GDP growth two years later:
    - 0.873 × 0.206 = 0.180 percentage points — being able to borrow at a “discount” of one standard deviation reduces real GDP growth by 0.180 percentage points two years later.
  - For comparison: one standard deviation increase in the spread S_it (491.815 basis points) is expected to reduce real GDP growth by 0.225 percentage points.
- Interpretation of signs:
  - A significantly positive κ implies that optimistic past sentiment (μ < 0) is followed by lower future growth; conversely, pessimistic sentiment predicts higher future growth.
  - This counterintuitive direction is robust and suggests that favorable non-fundamental borrowing conditions do not translate into stronger future growth on average.

### Mechanisms, robustness, and caveats
- The authors interpret the result as a strong signal for at least two reasons:
  - Favorable borrowing conditions would be expected to boost future growth via cheaper borrowing and more room for investment; observed results point the opposite way.
  - If spreads reflect market expectations of future performance, μ < 0 would signal optimistic market expectations about future growth — again, observed results point to subsequent growth slowing.
- Methodological notes:
  - The authors place an AR(2) at the core of the growth regression (quarterly real GDP growth described well by AR(2) in prior work).
  - Controlling for S_it−1 is intended to reduce contamination from the working capital channel; remaining estimates are conservative lower bounds for the sentiment effect.
  - Outliers affect the spread standard deviation: excluding the top 5 percentile halves the spread standard deviation to 223.331 basis points (authors note this makes the reported impact conservative).

*Source: 3.1 Sovereign spreads and their determinants (wpiea2020166-print-pdf).*

### Section 5 shows that this

### Section 5 shows that this

### Asymmetric effect of sovereign spread mispricing on growth
- Finding: The detrimental effect on future real GDP growth stems primarily from episodes where the mispricing term μ < 0 (countries borrowing at a discount relative to fundamentals).
- Magnitudes and coefficients (selected):
  - μit−8 (baseline, Table 2 column (1)): 0.2063282 (significant at 1%).
  - μit−8 × I{μit−8 < 0} (Table 2 column (2)): 0.2581765 (significant at 1%).
  - Interpretation: When μ < 0, a one standard deviation spread discount is expected to reduce growth by 0.270 percentage points — larger than the effect for the spread level itself (0.225).
- Economic interpretation: Investors’ bullish view (μ < 0) may reflect betting on an upcoming structural break not supported by fundamentals; access to unusually cheap borrowing can induce heavy indebtment that often ends poorly.

### Link with current account deficits and external borrowing
- Interaction result (Table 2 column (3)):
  - μit−8 × (CAD/GDP)it−8 = 0.0238574 (significant at 5%).
  - (CAD/GDP)it−8 alone: −0.0110134 (not significant).
- Correlation: The sample correlation between μ and CAD is 0.0335.
- Finding: The detrimental effects of borrowing at a discount are largest in countries running current account deficits; the main channel appears to be external borrowing rather than domestic government deficits.
- Government deficit channel (Table 2 column (4)):
  - μit−8 × (DEF/GDP)it−8 = 0.0118626 (not significant).
  - (DEF/GDP)it−8 = −0.018155 (not significant).
- Interpretation: Cheap external financing combined with CAD increases risk of future growth losses; current account deficits themselves are not significant unless paired with inexplicably favorable borrowing conditions.

### Out-of-sample predictive performance
- Leave-one-out RMSE results (Table 3):
  - AR(2): RMSE = 2.03; ratio to AR(2) = 1.00.
  - AR(2) + Sit−1 + κμit−8: RMSE = 1.72; ratio to AR(2) = 0.85.
  - AR(2) + Sit−1: RMSE = 2.03; ratio to AR(2) = 1.01.
- Time-based RMSE results (Table 4; averaged over T between 2015q1 and 2016q4):
  - AR(2): RMSE = 2.17; ratio to AR(2) = 1.00.
  - AR(2) + Sit−1 + κμit−8: RMSE = 1.86; ratio to AR(2) = 0.85.
  - AR(2) + Sit−1: RMSE = 2.07; ratio to AR(2) = 0.95.
- Conclusion: Including past mispricing μit−8 reduces forecast RMSE by about 15 percent relative to AR(2); adding only lagged spreads offers negligible improvement.

### Robustness and timing of effects
- Baseline lag: ` = 8 quarters (2 years). Results robust to different lag lengths and less parsimonious first-stage specifications.
- Horizon: Effect maximized at ` = 6 quarters; dynamics operate over about 4 to 14 quarters.
- Moving average robustness: MA-window [μit−14, μit−2] yields κ = 0.1548611 (p-value = 0.001).
- Alternative specifications:
  - Larger first-stage (Table A3) yields high R2 = 0.7811 and similar patterns.
  - Second-stage in Table A4 (alternative first stage): μit−8 = 0.1504278 (significant at 5%); R2 = 0.7568.
- Fiscal crises outcome: μt−` < 0 increases likelihood of future fiscal crisis at longer lags (4–7 years), consistent with average maturity of government debt of 5.6 years.

### Mechanisms and economic channels
- Possible economic channels identified:
  - Debt overhang from heavy borrowing (Myers, 1977; Krugman, 1988).
  - “Investment hangover” after optimistic booms (Beaudry, Galizia, and Portier (2018); Rognlie, Shleifer, and Simsek (2018)).
  - Real exchange rate over-valuation and lost competitiveness (Reis, 2013; Benigno and Fornaro, 2014; Benigno, Converse, and Fornaro, 2015; Gopinath et al., 2017).
- Possible political/institutional channels:
  - Misallocated funds, corruption, or consumption rather than productive investment (institutional Dutch disease; Gelb (1988)).
  - Masking of inefficiencies and postponement of reforms (Fernández-Villaverde, Garicano, and Santos (2013)).
  - Riskier policies or loosening macroprudential regulation under bullish sentiment.

### Policy implications and recommendations
- Monitoring and macroprudential policy:
  - Continuous real-time tracking of the mispricing proxy μ is feasible and may guide policy discussions.
  - Recommendation: Countercyclical macroprudential regulation that tightens during boom periods when foreign borrowing is cheap (μ < 0) — “leaning against sentiment” could yield stability gains (Flemming, L’Huillier, and Piguillem, 2019).
- Caution on anchoring fiscal rules to spreads:
  - Spreads may reflect sentiment and mispricing; anchoring rules solely to spread levels can be misleading.
  - Emphasize fundamentals alongside spread measures when assessing fiscal space (Hatchondo, Martinez, and Roch, 2017).
- Forecasting practice:
  - Incorporating past mispricing μit−8 in forecasting models can materially improve forecast accuracy for one-quarter ahead growth.

### Conclusion (summary)
- Undue optimism (borrowing at spreads lower than fundamentals justify) tends to be followed by lower economic growth and, with a further lag, increased incidence of fiscal crises.
- The harmful effect is asymmetric and concentrated in episodes of spread discounts (μ < 0), particularly when paired with current account deficits.
- Including a sovereign debt mispricing measure improves out-of-sample forecasts and has actionable policy implications for macroprudential regulation and the assessment of fiscal risks.

*Source: wpiea2020166-print-pdf - Section 5 shows that this*

---


_Source: https://www.imf.org/-/media/files/publications/wp/2020/english/wpiea2020166-print-pdf.pdf_