## wpiea2019200-print-pdf

## Source details

**Canonical URL:** [wpiea2019200-print-pdf](https://www.imf.org/-/media/files/publications/wp/2019/wpiea2019200-print-pdf.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/wp/2019/wpiea2019200-print-pdf.pdf.md)
- [Structured JSON version](/-/media/files/publications/wp/2019/wpiea2019200-print-pdf.pdf.json)

---

### I. Introduction — key findings
- Research question: How informative are IMF-World Economic Outlook (IMF-WEO) output gap estimates for countries in the euro area?
- Core observation: WEO real-time output gap estimates tend to have large and persistently negative means.
- Magnitude: For several large euro area countries the average real-time output gap is close to - 2 percent of potential output over the period 1994–2017.
- Temporal pattern: Negative real-time gaps are even larger before the Great Financial Crisis (GFC) and tend to be systematically “revised away” in later WEO vintages.
- Policy risks:
  - Persistent negative real-time bias can lead to systematic over-optimism about future output and fiscal revenues and to larger-than-planned accumulation of public debt.
  - Can lead to unwarranted monetary stimulus.
- Contribution: Uses the IMF multivariate filter (MVF) to construct artificial real-time output gap estimates based on information available in real time (data and forecasts) to isolate roles of data revisions, forecast uncertainty, and judgment in explaining negative real-time bias.

### Decomposition of real-time output gap bias
- Data revisions:
  - Data revisions play only a minor role in explaining large and persistent negative real-time output gap estimates (consistent with Orphanides and others (2000, 2005)).
  - Quantitative impact: on average less than 0.2 percentage point of potential GDP (section B).
- Forecast errors:
  - WEO growth forecasts tend to be too optimistic on average because the staff baseline forecast is a modal forecast that typically fails to predict recessions (Box 3).
  - Mechanism: Overly optimistic forecasts raise estimated potential output, lowering real-time output gaps.
  - MVF benchmark estimate: forecast errors lowered the real-time output gap estimates by about 0.7 percentage points on average in the euro area.
- Judgment:
  - Defined as staff adjustments relative to the MVF benchmark when using actual WEO data and forecasts available in real time.
  - Effect: Judgment typically increased potential output by almost 1.0 percentage point of potential GDP in the sample.
  - Contribution: Judgment accounts for just above one-half of the overall tendency towards negative output gap estimates in real time.

### Empirical correlations and fiscal implications
- Revisions and debt:
  - Systematic upward revisions to real-time output gaps are positively correlated with both public debt levels and public debt WEO forecast errors in main euro area countries.
- Primary balances:
  - Robust, negative empirical association between primary fiscal balances and revisions to the WEO output gap estimates, controlling for various other variables.
- Policy caution: The paper cautions against an excessive focus on output gap estimates in real time when calibrating counter-cyclical policy.

### Output gaps and inflation predictability
- Real-time output gaps:
  - Generally not robust predictors of inflation in real time; contain at best information on directional changes in business cycles and inflation.
- Final output gap estimates:
  - The Phillips curve relationship is clearly established using final estimates of the output gap.
- Inflation forecasts:
  - Real time WEO inflation forecasts are not significantly revised and therefore are informative for monetary policy purposes.
  - Conclusion: Final or real time inflation is the best predictor of future inflation; real-time output gaps are less reliable for inflation forecasting.

### Estimation methods — overview
- Potential output definitions discussed: statistical trend; utilization of production factors consistent with stable inflation (Okun, 1962); “sustainable” output (IMF, 2015).
- Common approaches:
  - Univariate filters (HP; Baxter and King; Christiano and Fitzgerald): simple and transparent; suffer from end-point problem.
  - Production function estimation: allows input contribution analysis; subject to end-point problem and may ignore labor market and inflation information.
  - Multivariate filters (MVF): incorporate Phillips Curve and Okun’s Law; estimate potential output and NAIRU jointly; flexible to include financial or housing variables for “sustainable” output concepts.
  - Structural DSGE models: joint estimation of structural shocks and potential output; sensitive to specification and parametrization.
- Use of “sustainable” output:
  - Augmenting MVF with real credit and real house price growth can yield a trend closer to sustainable output (IMF, 2015).
  - Caution: Real-time identification of sustainable output is very difficult; best used as a complement or “fire alarm”.

### Desirable properties of output gap estimates
- Stability (Size of Ex-Post Revisions):
  - Revisions tend to be large and mostly upwards; ex-post revisions can be at least as large as the estimates themselves and are persistent.
- Efficiency:
  - Estimates are efficient if they utilize all information available at time of estimation; observed serial correlation in revisions implies inefficiency.
- Asymptotic Zero-Mean:
  - Filter-based methods imply zero-mean asymptotically; one-sided (mostly upward) revisions indicate systemic bias.
- Economic Consistency:
  - Real-time output gaps should correlate with inflation and other slack indicators (unemployment, capacity utilization, labor market tightness).

### Box 1 — Nominal wage rigidities, hysteresis, and output gap estimates (selected findings)
- Asymmetric nominal wage rigidities and hysteresis can generate negative long-run output gaps; conventional filters may spuriously lower potential in deep demand-driven recessions.
- Empirical/modeling evidence cited:
  - Benigno and Ricci (2011): downward wage rigidity can make long-run output gap nonpositive.
  - Aiyar and Voigts (2019): downward nominal wage rigidities lead to negative average output gap; conventional filters show intrinsic upward bias in potential during deep recessions.
  - Coibion and others (2017): potential GDP estimates tend to be sensitive to demand shocks and under-respond to supply shocks.
  - Blanchard and others (2015): 83 percent of recessions associated with supply shocks result in sustained decline in output; disinflation episodes often associated with lower long-term output.
  - Alichi and others (2019): MVF with labor market hysteresis yields higher NAIRU, lower potential, and substantially higher (less negative) output gaps through the GFC and 1980s compared to simulations without hysteresis.
- Definitions and data vintages:
  - Real-time output gap defined as y_i,t | t from Fall WEO of year t.
  - Unbalanced panel for euro area countries starts in 1994 and extends to the 2017 Fall WEO vintage (final).
  - Revisions defined as deviations of real-time from final (2017 WEO) estimates or from WEO estimates up to five years later (y_i,t | t+5).

### Properties of WEO real-time output gap estimates — revisions and bias
- Revisions:
  - For the euro area, real-time estimates are revised upwards by more than 1 percent of potential output on average, with variation across countries and time.
  - Downward revisions constitute only about one-sixth of observations.
  - Revisions are gradual: short-term revisions (one or two years) small and not serially correlated; after two years revisions start to exhibit serial correlation.
- Prevalence of negative real-time gaps:
  - Over WEO history in real time: France and Italy have not recorded a single positive output gap estimate; Germany has recorded four, Spain five, and the euro area only one.
  - Real-time output gaps are predominantly negative on average (p-value<0.1 in most countries); in France, Italy, Finland, Portugal and Spain estimates are close to, or exceed, -2 percent of potential GDP.
- Final vintages:
  - Negative average output gap estimates are largely revised away by the 2017 WEO vintage; most countries’ averaged output gaps over 1994–2017 are substantially more positive and not statistically different from zero.
  - Exceptions: France and Italy and the weighted euro area retain large and statistically significant negative gap estimates in final vintage.

### Okun’s Law and Phillips Curve diagnostics (economic consistency)
- Okun’s Law:
  - Real-time WEO output gap estimates are largely nonpositive even when real-time unemployment is below trend.
  - Using long-run average unemployment as trend, France, Italy, Finland and Portugal still exhibited real-time WEO output gaps around -2 percent of potential GDP when unemployment was below its long-term average.
  - Using NAIRU, inconsistency is less visible, suggesting bias mostly reflects overly high estimates of potential output.
- Phillips Curve / inflation:
  - Despite persistently large negative real-time output gaps, headline inflation has often hovered around or just below 2 percent in many countries; inflation expectations in the euro area are anchored.
  - Revisions comparison (baseline 11 countries, 1994–2012): revisions to Fall WEO real time output gaps average 1.3 percent of potential GDP (unweighted), while revisions to real-time inflation are marginal at -0.04 percentage points (unweighted).

### Real-time estimates in other advanced economies
- WEO real-time output gap estimates for the US, UK, Canada and Japan have been significantly negative on average; by the 2017 Fall WEO mean estimates center around zero for these countries except Japan.
- 1994–2017 statistics (n=96) — Real-Time means and latest vintage:
  - Real-Time:
    - US: Mean -1.18, SE 0.46, p-value 0.01
    - UK: Mean -1.32, SE 0.31, p-value 0.00
    - Canada: Mean -1.27, SE 0.26, p-value 0.00
    - Japan: Mean -3.07, SE 0.46, p-value 0.00
  - Latest Vintage:
    - US: Mean 0.11, SE 0.44, p-value 0.80
    - UK: Mean -0.31, SE 0.34, p-value 0.37
    - Canada: Mean 0.17, SE 0.24, p-value 0.49
    - Japan: Mean -1.57, SE 0.39, p-value 0.00

### Box 2 — Decomposition using MVF (sample and method)
- Aim: isolate impact of three measurable sources of bias using MVF:
  - (i) Data revisions
  - (ii) Systematic bias in 5-year ahead forecasts used to extend samples
  - (iii) Judgment (deviations from MVF)
- Sample: 1990–2017, 11 countries with full real-time data: Austria, Belgium, France, Finland, Germany, Greece, Italy, Ireland, the Netherlands, Portugal, Spain.
- Caveat: impacts conditional on MVF; “judgment” captures both discretionary expert adjustments and methodological differences.

### Quantitative decomposition — MVF estimates (selected aggregated results, 1994–2012)
- Unweighted average (Mean; p-value):
  - Data revisions: Mean 0.18, p-value 0.00
  - Forecast errors: Mean 0.60, p-value 0.00
  - Judgement: Mean 0.76, p-value 0.00
  - Total bias 1/: Mean 1.53, p-value 0.00
  - WEO revisions 2/: Mean 1.31, p-value 0.00
- Weighted average (GDP weights):
  - Data revisions: Mean 0.11, p-value 0.00
  - Forecast errors: Mean 0.66, p-value 0.00
  - Judgement: Mean 0.95, p-value 0.00
  - Total bias 1/: Mean 1.74, p-value 0.00
  - WEO revisions 2/: Mean 1.42, p-value 0.00
- Country examples (selected means and p-values):
  - Austria: Data revisions 0.19 (p-value 0.00); Forecast errors 0.44 (p-value 0.02); Judgement 0.53 (p-value 0.00); Total bias 1.26 (p-value 0.00); WEO revisions 1.39 (p-value 0.00).
  - France: Data revisions 0.07 (p-value 0.09); Forecast errors 0.71 (p-value 0.00); Judgement 1.37 (p-value 0.00); Total bias 2.15 (p-value 0.00); WEO revisions 1.45 (p-value 0.00).
  - Spain: Data revisions 0.37 (p-value 0.02); Forecast errors 0.55 (p-value 0.05); Judgement 0.64 (p-value 0.01); Total bias 1.75 (p-value 0.00); WEO revisions 2.28 (p-value 0.00).
- Interpretation: the sum of the three measured sources is close to the actual WEO revisions, indicating these channels account for most observed negative bias and subsequent positive revisions.

### Box 3 — GDP growth forecast errors across WEO vintages (summary)
- WEO forecasts for the euro area tend to overpredict real GDP growth; baseline forecasts are “modal” and typically do not incorporate severe recessions.
- Analysis for 12 euro area countries (1994–2017) confirms optimistic bias for nearly all countries and horizons (except nowcasts), though bias is statistically significant in less than half of countries.
- Excluding the crisis period reduces size, incidence, and significance of optimistic bias.
- Optimistic bias stronger in medium-term growth forecasts and stems largely from failure to predict recessions.
- Real-time nowcasts tend to exhibit a small negative bias (GDP growth under-projected in real time).

### Fiscal consequences — magnitudes and simulations
- Thought-experiment translation of real-time output gap bias into cyclical component of fiscal balances yields bias in fiscal policy ranging from 0.7 to 0.9 percent of potential GDP annually.
- Long-run debt implications:
  - Year-after-year overestimation of potential implies significant unplanned deficits and large debt buildup over a 19-year baseline sample.
  - In the exercise the average increase in debt over 19 years is around 15 percent of GDP and can reach close to 25 percent of GDP in countries with larger bias.
- Eyraud and others (2018) and Eyraud and Wu (2015): for 2003–16 the output gap was underestimated in real time by 1.3 percentage points of GDP on average → overestimation of cyclically adjusted balances by 0.5 percentage point on average per year (assuming elasticity of revenue to output = 1, elasticity of expenditure to output = 0, average expenditure to GDP ratio = 45 percent).
- Cross-sectional associations:
  - Positive association between debt-to-GDP ratio in 2017 and average output gap revisions over 1994–2017.
  - Average revision in higher-debt countries: 1.1 percent of potential GDP.
  - Average revision in lower-debt countries: 0.5 percent of potential GDP.
  - WEO output gap revisions over 1994–2012 positively associated with WEO 5-year ahead forecast errors for the public debt ratio.

### Box 4 — Fiscal outcomes and real-time output gap bias (regression and findings)
- Methodology:
  - Adaptation of fiscal reaction functions (Bohn) relating primary balances to lagged primary balance, lagged public debt, detrended real government consumption, WEO output gap (final vintage) and revisions to real-time WEO output gap, MVF simulated sources, and controls.
  - Estimator: bias-corrected LSDV dynamic panel estimator by Bruno (2005) with bootstrapped standard errors.
  - Baseline sample: 11 larger euro area countries for 1994–2012. Observations: 195.
- Selected regression coefficients (preserve numeric entries exactly as reported):
  - Lagged primary balance (Primary balance (t-1)): 0.641***, 0.498***, 0.393***, 0.473***, 0.406***, 0.514***, 0.527***, 0.518***
  - Output gap (final vintage): 0.327***, 0.801***, 0.767***, 0.154, 0.335**, 0.196
  - Revision to real-time output gap against 2017 vintage: -0.714***, -0.849***
  - Output gap (t+5 vintage): 0.834***, 0.832***
  - Revision to real-time output gap against (t+5) vintage: -0.612***, -0.745***
  - MVF components: Data revisions (MVF): 1.023***; Forecast error (MVF): -0.444*; Judgement (MVF): -0.490**
  - Public debt (t-1): 0.039**, 0.035**, 0.082***, 0.041**, 0.079***, 0.082***, 0.086***, 0.085***
  - Current account (in percent of GDP): 0.254***, 0.217**, 0.277***, 0.241***, 0.281***, 0.255***, 0.234**, 0.276***
- Key empirical findings:
  - Revisions to real-time output gaps are statistically highly significant and economically large — nearly as large as the coefficient on the final output gap.
  - Predominantly positive (upwards) revisions averaging 1.3–1.4 percent of potential GDP associated with lower primary balance estimates in the order of 0.8–1.2 percent of GDP annually.
  - MVF-simulated forecast errors and judgement have similar negative associations with primary balances (forecast error (MVF): -0.444*; judgement (MVF): -0.490**).
- Interpretation:
  - Over-estimation of productive capacity and over-optimistic growth forecasts are primary drivers of negative real-time bias.
  - Biases associated with lower primary balances, higher public debt ratios, and faster debt accumulation.

### Appendix — figures and tables (selected numeric highlights)
- Appendix Figure 1 panels use vertical axis tick labels with exact values for countries (examples preserved from panels):
  - France/Germany/Italy/Austria/Belgium: -4, -3, -2, -1, 0, 1, 2, 3 (sometimes extended to 4 or 5).
  - Spain: -8, -6, -4, -2, 0, 2, 4, 6, 8.
  - Greece: -12, -9, -6, -3, 0, 3, 6, 9, 12, 15.
  - Portugal: -7, -6, -5, -4, -3, -2, -1, 0, 1, 2, 3, 4.
  - Ireland: -8, -6, -4, -2, 0, 2, 4, 6, 8.
- Appendix Table 1 (real-time output gaps at times of low real-time unemployment, 1994–2017) — selected preserved table entries:
  - Euro Area -1.10.270.00 Euro Area -0.25
  - Austria -1.00.230.00 Austria -1.440.230.00
  - Belgium -0.80.210.00 Belgium -0.470.210.03
  - France -1.80.240.00 France
  - Germany -0.30.280.34 Germany -0.180.260.48
  - Italy -2.20.270.00 Italy -2.81
  - Luxembourg 1.52.030.45 Luxembourg 0.250.360.50
  - Netherlands -0.30.430.43 Netherlands 0.450.430.30
  - Finland -1.90.430.00 Finland 0.580.210.01
  - Greece 0.80.390.03 Greece 2.310.200.00
  - Ireland 1.10.640.10 Ireland 1.970.690.01
  - Malta 1.20.010.00 Malta
  - Portugal -1.80.440.00 Portugal -1.450.480.00
  - Spain -0.40.200.08 Spain -0.200.290.49
  - Cyprus 2.52.890.39 Cyprus
  - Slovakia -1.40.520.01 Slovakia
  - Estonia -1.00.600.11 Estonia
  - Latvia -0.20.020.00 Latvia
  - Lithuania -0.70.640.30 Lithuania -0.490.250.05
  - Slovenia -0.80.780.33 Slovenia 0.290.910.76
- Appendix Table 3 (Real-Time Output Gaps for High and Low Debt Countries) — preserved entries:
  - Austria 0.70.813.915.4
  - Belgium 0.70.513.110.1
  - France 1.30.924.516.6
  - Germany 0.80.714.813.7
  - Italy 1.40.927.517.0
  - The Netherlands 0.40.47.77.2
  - Finland 1.31.424.826.9
  - Greece 0.10.52.19.3
  - Ireland 0.9-0.116.5-2.5
  - Portugal 0.80.816.015.5
  - Spain 0.91.217.923.4
  - Average 0.90.716.313.9
  - Low debt-0.90.00155 Low debt-0.40.02155
  - High debt-1.60.00194 High debt-0.50.02194
  - Difference-0.750.01349 Difference-0.10.78349
  - Real-time output gaps (unweighted) 2017 WEO vintage (unweighted)
  - Meanp-value DF Meanp-value DF
  - Low debt-0.90.00155 Low debt -0.10.63223
  - High debt-1.60.00194 High debt -0.50.01214
  - Difference-0.750.01349 Difference-0.4 0.17437

### Key numerical findings and summary statistics
- Attribution of WEO revisions: just below half of WEO revisions can be attributed to forecast errors and the remaining half to judgment; data revisions minor.
- Correlation coefficients:
  - 67 percent correlation between changes in real-time and final output gap estimates for 11 baseline countries.
  - 28 percent correlation for all countries.
- Output gap revisions by debt group:
  - Average revision in higher-debt countries: 1.1 percent of potential GDP.
  - Average revision in lower-debt countries: 0.5 percent of potential GDP.
- Fiscal bias magnitudes:
  - Bias in fiscal policy from output gap bias ranges from 0.7 to 0.9 percent of potential GDP annually (thought experiment).
  - Average increase in debt over 19 years in exercise: around 15 percent of GDP; can reach close to 25 percent for countries with larger bias.
  - Eyraud et al. (2003–16): real-time underestimation of output gap by 1.3 percentage points of GDP on average → overestimation of cyclically adjusted balances by 0.5 percentage point on average per year.
- Inflation context:
  - Inflation fluctuating around 2 percent; inflation persistence makes real-time inflation a useful predictor of final inflation.

### Policy recommendations and options to reduce real-time errors
- Fiscal prudence and robustness:
  - Exercise fiscal prudence given uncertainty in estimating potential GDP, particularly for high debt countries.
  - Avoid reliance on point estimates of the output gap; use full distributions, dispersion and skewness to inform policy calibration.
  - Build safety margins in fiscal policy targets; do not take real-time estimates of structural balances at face value.
- Rule design:
  - Expenditure rules that allow automatic stabilizers to work can be more resilient to measurement errors though sensitive to initial conditions.
  - Counterfactual simulations suggest expenditure rules could have led to significantly smaller differences between real-time and ex post public debt outcomes than cyclically adjusted balance rules for France and Italy.
- Estimation improvements:
  - Move toward multivariate filter (MVF) that imposes economic structure while allowing expert judgement integrated via growth accounting and production function approaches.
  - Increase number of indicators (capacity utilization, investment to GDP, labor vacancy rates, house prices, credit) to improve MVF precision and reduce ex-post revisions.
  - Incorporate measures of forecast distribution asymmetry in the central forecast — e.g., rely on the mean instead of the mode to reduce forecast errors and ex-post revisions.
  - Use complementary measures of potential output (for example, IMF (2015) concept of “sustainable” output) to provide timely information when measures diverge.
- Caution on judgement:
  - Exercise caution in incorporating off-model information (structural or fiscal reform impacts); judgement can contribute to negative bias if too sensitive to end-of-sample conditions.
- Overarching guidance:
  - Avoid exclusive focus on real-time output gap estimates when calibrating counter-cyclical policy; adopt an encompassing approach using multiple measures and buffers.

*Source: wpiea2019200-print-pdf*

### REFERENCES ______________________________________________________________30

### wpiea2019200-print-pdf - REFERENCES ______________________________________________________________30

### I. INTRODUCTION — key findings
- Research question: How informative are IMF-World Economic Outlook (IMF-WEO) output gap estimates for countries in the euro area?
- Core observation: WEO real-time output gap estimates tend to have large and persistently negative means.
- Magnitude: For several large euro area countries the average real-time output gap is close to - 2 percent of potential output over the period 1994–2017.
- Temporal pattern: The negative real-time gaps are even larger before the Great Financial Crisis (GFC) and tend to be systematically “revised away” in later WEO vintages.
- Policy risks:
  - Persistent negative real-time bias can lead to systematic over-optimism about future output and fiscal revenues and to larger-than-planned accumulation of public debt.
  - Can lead to unwarranted monetary stimulus.
- Contribution of the paper: Uses the IMF multivariate filter (MVF) to construct artificial real-time output gap estimates based on information available in real time (data and forecasts) to isolate roles of data revisions, forecast uncertainty, and judgment in explaining negative real-time bias.

### Decomposition of real-time output gap bias
- Data revisions:
  - Consistent with Orphanides and others (2000, 2005): data revisions play only a minor role in explaining large and persistent negative real-time output gap estimates.
- Forecast errors:
  - WEO growth forecasts tend to be too optimistic on average because the staff baseline forecast is a modal forecast that typically fails to predict recessions and other tail events (see Box 3).
  - Implication: Overly optimistic forecasts tend to pull up estimates of potential output, lowering real-time output gaps.
  - Estimate from MVF benchmark: forecast errors lowered the real-time output gap estimates by about 0.7 percentage points on average in the euro area.
- Judgment:
  - Defined as staff adjustments relative to the MVF benchmark when using actual WEO data and forecasts available in real time.
  - Effect: Judgment typically increased potential output by almost 1.0 percentage point of potential GDP in the sample.
  - Contribution: Judgment accounts for just above one-half of the overall tendency towards negative output gap estimates in real time.

### Empirical correlations and fiscal implications
- Revisions and debt:
  - Systematic upward revisions to real-time output gaps are positively correlated with both public debt levels and public debt WEO forecast errors in main euro area countries.
- Primary balances:
  - Robust, negative empirical association between primary fiscal balances and revisions to the WEO output gap estimates, controlling for various other variables.
- Policy caution: The paper cautions against an excessive focus on output gap estimates in real time when calibrating counter-cyclical policy.

### Output gaps and inflation predictability
- Real-time output gaps:
  - Generally not robust predictors of inflation in real time.
  - Contain at best information on directional changes in business cycles and inflation.
- Final output gap estimates:
  - The Phillips curve relationship is clearly established using final estimates of the output gap (see also Abdih and others, 2018).
- Inflation forecasts:
  - Real time WEO inflation forecasts are not significantly revised and therefore are informative for monetary policy purposes.
  - Conclusion: Final or real time inflation is the best predictor of future inflation; real-time output gaps are less reliable for inflation forecasting.

### Estimation methods — overview
- Conceptual variation:
  - Potential output definitions: statistical trend, utilization of production factors consistent with stable inflation (Okun, 1962), and “sustainable” output (IMF, 2015).
- Common approaches:
  - Univariate filters (e.g., HP filter; Baxter and King; Christiano and Fitzgerald): Simple and transparent; suffer from end-point problem and may lack economic interpretation.
  - Production function estimation: Supply-side decomposition into labor, capital, and TFP; allows input contribution analysis; still subject to end-point problem and may ignore labor market and inflation information.
  - Multivariate filters (MVF): Incorporate Phillips Curve and Okun’s Law; estimate potential output and NAIRU jointly; interpretable for countercyclical policy; flexible to include financial or housing variables for “sustainable” output concepts.
  - Structural DS GE models: Joint estimation of structural shocks and potential output; theoretically sound but sensitive to specification and parametrization.
- Use of “sustainable” output:
  - Augmenting MVF with real credit and real house price growth can yield a trend closer to sustainable output (IMF, 2015).
  - Caution: Real-time identification of sustainable output is very difficult; best used as a complement or “fire alarm” to conventional output gap estimates.

### Desirable properties of output gap estimates
- Stability (Size of Ex-Post Revisions):
  - Stability interpreted as average size and variation of ex-post revisions of real-time output gaps.
  - Literature documents real-time output gap estimates tend to be too negative and substantially revised upwards; ex-post revisions can be at least as large as the estimates themselves and are persistent (see Box 2).
  - Implication: Unstable and downward-biased estimates may result in involuntary debt buildups or costly adjustments (e.g., in EU fiscal frameworks like the Stability and Growth Pact).
- Efficiency:
  - Output gap estimates efficient if they utilize all information available at time of estimation; future revisions should not be predictable; past revisions should not help explain subsequent revisions (no serial correlation).
- Asymptotic Zero-Mean:
  - Filter-based methods imply estimated output gap has zero-mean asymptotically, reflecting temporary nature of gaps and long-run balanced growth.
  - One-sided revisions indicate systemic bias; negative bias could lead to excessive deficits under structural balance-based fiscal rules.
  - Note: Business cycle asymmetry may arise (e.g., downward nominal wage rigidities, hysteresis, ZLB constraints).
- Economic Consistency:
  - Real-time output gaps should correlate with inflation and other slack indicators (unemployment, capacity utilization, labor market tightness).
  - These indicators can be incorporated into MVF or used separately to assess real-time output gap measures.

*Source: wpiea2019200-print-pdf - REFERENCES ______________________________________________________________30*

### Box 1. Nominal Wage Rigidities, Hysteresis, and Output Gap Estimates

### Box 1. Nominal Wage Rigidities, Hysteresis, and Output Gap Estimates

### Nominal wage rigidities and long-run output gaps
- Some DSGE models with asymmetric nominal wage rigidities produce negative long-run output gaps; the gap is defined as the difference between output under nominal rigidity and output under flexible wages and prices.
- In the framework of Benigno and Ricci (2011) downward wage rigidity dominates the forward-looking reaction of wage setters, so the output gap is always nonpositive in the long run (it is zero without uncertainty).
- Aiyar and Voigts (2019): downward nominal wage rigidities lead to a negative average output gap because negative demand shocks have larger impacts on unemployment and output than positive shocks; conventional filters (HP or MVF) exhibit an intrinsic upward bias in potential output estimates in deep demand-driven recessions because they are bound by the asymptotic zero mean property and thus spuriously lower potential.
- Coibion and others (2017): real-time potential GDP estimates across several institutions tend to be sensitive to demand shocks and under-respond to supply shocks; using a Blanchard and Quah (BQ, 1989) bivariate VAR identifying restriction that only supply shocks have permanent effects, they find post-GFC US potential output tended to be under-estimated and output gaps should be larger and more negative than most institutions estimated.
- Blanchard and others (2015): 83 percent of recessions associated with supply shocks result in sustained decline in output; recessions triggered by demand shocks are frequently followed by lower output or lower output growth and can thus have permanent effects. Of recessions associated with intentional disinflation, almost two-thirds are associated with lower long-term output.
- Blanchard and Summers (1986) and Ball (1999) questioned long-run neutrality of money and argued monetary policy and other demand factors can have permanent effects.
- These findings point to important hysteresis effects: potential output can decline more during recessions, implying output gaps can be higher (or less negative).
- Alichi and others (2019): applying MVF with labor market hysteresis on US data yields significantly higher NAIRU, lower potential, and substantially higher (less negative) output gaps throughout the GFC and 1980s compared to simulations without hysteresis.
- Hamilton (2018) univariate regression can capture persistence and center cycles around zero mean.
- Bashar (2011): in G-7 countries aggregate demand shocks positively affect aggregate supply shocks, causing permanent effects on the output level.

### Definitions and data vintages for real-time output gaps
- Real-time output gap defined as the output gap estimated for year t in the Fall WEO of the same year (y_i,t | t). Example: real-time output gap for year 2000 taken from the 2000 Fall WEO vintage.
- Real-time estimates are defined at a time when about half of the actual data for any particular year is available.
- Unbalanced panel for euro area countries starts in 1994 (when systematic reporting of output gap estimates became available in WEO) and extends to the 2017 Fall WEO vintage, referred to as the “final” WEO output gap estimates.
- Real-time estimates used are contemporaneous real time gaps (estimated at time t conditional on information available at that time) to reflect the measures relevant for fiscal and monetary policy decisions.
- Revisions defined as deviations of real-time output gap from final (2017 WEO) estimates, or from WEO estimates up to five years later (y_i,t | t+5).
- Systematic asymmetry in revisions of real-time estimates (changes) is interpreted as “bias”. Because final estimates broadly satisfy zero-mean, revision bias translates into level bias for real-time estimates.
- Analysis presented for all euro area countries and a baseline sample of 11 countries with full 1994–2017 data; comparisons made with other advanced economies and other international organizations.

### Properties of WEO real-time output gap estimates — revisions and bias
- Revisions of real-time output gap estimates tend to be large and mostly upwards.
- For the euro area, real-time estimates are revised upwards by more than 1 percent of potential output on average, with substantial variation across countries and time periods.
- Downward revisions (compared to the final 2017 estimates) constitute only about one-sixth of all observations and are attributable to very few countries.
- Real-time estimates do not perform well against stability and zero-mean properties.
- Revisions are gradual: short-term revisions (one or two years after) are small and not serially correlated; after two years revisions start to exhibit serial correlation that increases in subsequent years, implying inefficiency in output gap estimates (past revisions could be used to improve future estimates).
- Systematic upward revisions produce large and persistently negative output gap estimates in real time:
  - Over the WEO history in real time: France and Italy have not recorded a single positive output gap estimate; Germany has recorded four, Spain five, and the euro area only one.
- Real-time output gaps are predominantly negative on average (p-value<0.1 in most countries); in France, Italy, Finland, Portugal and Spain estimates are close to, or exceed, -2 percent of potential GDP.
- The negative average output gap estimates are largely revised away over time: by the final (2017) WEO vintage most countries’ averaged output gaps over 1994–2017 are substantially more positive and not statistically different from zero (notwithstanding GFC and irregular business cycle length).
- Exceptions: France and Italy as well as the weighted euro area retain large and statistically significant negative gap estimates also in the final WEO.
- For newer euro area countries (Baltic states, Slovenia, Slovakia, Malta) real-time estimates cover shorter periods (much of which spans the GFC) and therefore are not sufficiently long to assess asymptotic zero-mean property; in the latest WEO their output gap estimates are not statistically significantly different from zero though standard errors are large.
- Over longer history and full business cycles, final vintage weighted output gap estimates for the euro area average -0.3, -0.3, and 0 percent of potential GDP for WEO, OECD and EC respectively; unweighted averages are -0.2, -0.5, and -0.2 respectively (calculations for the OECD cover 17 euro area countries).

### Economic consistency: Okun’s Law and Phillips Curve diagnostics
- Okun’s Law:
  - Real-time WEO output gap estimates are largely nonpositive even at times of below-trend real-time unemployment when output would be expected to exceed potential.
  - Using two measures of trend (long-run average unemployment rate and NAIRU): with the long-run average unemployment rate, France, Italy, Finland and Portugal still exhibited real-time WEO output gaps around -2 percent of potential GDP when unemployment was below its long-term average, pointing to inconsistency in the gap estimate.
  - Using NAIRU, inconsistency is less visible, suggesting the bias mostly reflects overly high estimates of potential output; nevertheless, when unemployment is below NAIRU, real-time output gap estimates are rarely positive.
- Phillips Curve / inflation relationship:
  - Despite persistently large negative real-time output gaps, headline inflation has often hovered around or just below 2 percent in many countries, with inflation expectations in the euro area anchored.
  - Revisions comparison (baseline sample of 11 countries, years 1994–2012): revisions to Fall WEO real time output gaps average 1.3 percent of potential GDP (unweighted), while revisions to real-time inflation are marginal at -0.04 percentage points (unweighted) and on average downwards.
  - Implication: if final vintage output gap estimates are useful for forecasting inflation, real-time estimates are unlikely to be useful for level forecasts of inflation and at best may inform direction.

### Real-time estimates in other advanced economies
- WEO real-time output gap estimates for the US, UK, Canada and Japan have been significantly negative on average; by the 2017 Fall WEO mean estimates center around zero for these countries except Japan.
- From 1994 to 2017: Japan has not recorded a single positive output gap estimate in real time; Canada has recorded two positive real-time gap estimates; the UK four.

*Source: wpiea2019200-print-pdf - Box 1. Nominal Wage Rigidities, Hysteresis, and Output Gap Estimates*

### Box 2. Real-Time Output Gap Assessments in International Organizations

### Box 2. Real-Time Output Gap Assessments in International Organizations

### Documented negative bias and literature findings
- Real-time output gap estimates by international organizations for the euro area countries are characterized by a high degree of instability, with large and predominantly positive revisions (ECB (2005, 2011); Marcellino and Musso (2011); Rünstler (2002)).
- Kempkes (2014) finds a negative bias in real-time estimates for EU 15 over 1996–2011:
  - Bias is present (i) irrespective of the data source, (ii) in all real-time vintages, and (iii) across the cross-section of countries.
  - The bias is estimated on average 0.5 percentage points of potential GDP per year.
- Hernández de Cos and others (2016) and Ademmer and others (2019) find asymmetric revisions:
  - Revisions are upward in expansions and downward in recessions (Hernández de Cos et al., 2016 for EU 15 over 2004–14; Ademmer et al., 2019 for EU 28 over 2004–17).
- Turner and others (2016) and others show that additional cyclical indicators (manufacturing capacity utilization, investment share in GDP, house prices, credit) improve reliability of output gap estimates.
- Edge and Rudd (2016) and Champagne and others (2018) find that output gap revisions have become smaller in more recent samples (U.S. and Canada).

### Evidence from IMF WEO: Real-Time WEO Output Gaps in Major Non-Euro Area Economies (1994–2017)
- Table 3: Mean estimates and significance (n=96)
  - Real-Time (n=96):
    - US: Mean -1.18, SE 0.46, p-value 0.01
    - UK: Mean -1.32, SE 0.31, p-value 0.00
    - Canada: Mean -1.27, SE 0.26, p-value 0.00
    - Japan: Mean -3.07, SE 0.46, p-value 0.00
  - Latest Vintage (n=96):
    - US: Mean 0.11, SE 0.44, p-value 0.80
    - UK: Mean -0.31, SE 0.34, p-value 0.37
    - Canada: Mean 0.17, SE 0.24, p-value 0.49
    - Japan: Mean -1.57, SE 0.39, p-value 0.00

### IV. Explaining the negative bias in real-time output gaps — decomposition approach
- Aim: isolate impact of three measurable sources of bias on real-time output gap estimates using IMF’s multivariate filter (MVF):
  - (i) Data revisions
  - (ii) Systematic bias in 5-year ahead forecasts used to extend samples
  - (iii) Judgment (deviations from MVF estimates, capturing model/parameter uncertainty and expert knowledge)
- Sample: 1990–2017, 11 countries with full real-time data: Austria, Belgium, France, Finland, Germany, Greece, Italy, Ireland, the Netherlands, Portugal, Spain.
- Caveat: impacts are conditional on using the MVF; “judgment” captures both discretionary expert adjustments and methodological differences between MVF and other filters.

### B. Data revisions (impact)
- Method: compare quasi-real-time MVF estimates (final data truncated at year T) with real-time estimates (vintage WEO data from year T).
- Finding: data revisions caused real-time estimates to be generally below final estimates, but on average by less than 0.2 percentage point of potential GDP.

### C. Forecast accuracy (impact)
- WEO/GDP forecasts tend to be too optimistic on average; medium-term forecasts show stronger optimistic bias.
- Mechanism: MVF real-time estimates use a five-year forecast of output, inflation, unemployment (sample extended to T+5). Overly optimistic forecasts raise estimated potential and thus make contemporaneous output gaps more negative.
- Quantitative finding: forecast inaccuracy tends to lower output gap estimates (increase potential output) by around 0.7 percentage points on average.
- Perfect-foresight versus vintage-forecast concept:
  - Vintage forecast estimate: based on year-T vintage data and five-year ahead WEO vintage forecast.
  - Perfect-foresight estimate: extends sample using ex-post correct growth rates for T+1 to T+5 from final data (fall 2017 WEO).
  - The two differ by the inaccuracy of the vintage forecast.

### Box 3 — GDP growth forecast errors across WEO vintages (summary)
- WEO forecasts for the euro area tend to overpredict real GDP growth; main reason is baseline forecasts typically do not incorporate severe recessions (they are “modal,” not “average” forecasts).
- Analysis for 12 euro area countries (1994–2017) confirms optimistic bias for nearly all countries and horizons (except nowcasts), though bias is statistically significant in less than half of countries.
- Excluding the crisis period reduces size, incidence, and significance of optimistic bias.
- Optimistic bias is stronger in medium-term growth forecasts and stems largely from failure to predict recessions.
- Real-time nowcasts tend to exhibit a small negative bias (GDP growth under-projected in real time).
- Institutional tendencies (e.g., implicit desire to close the output gap by a five-year horizon) can exacerbate optimistic bias effects on potential output estimates.

### E. The role of judgment (staff adjustments versus MVF)
- Definition: historical judgment = deviations of WEO real-time estimates from pure MVF-based vintage estimates (MVF uses all real-time information available to staff, excluding discretionary judgment).
- Interpretation: judgment captures staff discretion, methodological differences, and other external factors (e.g., over-optimism about convergence).
- Finding: real-time WEO estimates are on average about 1 percentage point below MVF-based vintage estimates (i.e., predominantly negative judgment relative to MVF benchmark).
- Note: MVF-based vintage estimates are not claimed superior to WEO real-time estimates — filters complement but do not substitute staff assessment.

### D. Quantitative decomposition — MVF estimates of real-time output gap biases (Table 4 summary for 1994–2012)
- Table 4 reports mean and p-values for contributions of Data revisions, Forecast errors, Judgment, Total bias (sum of three), and WEO revisions (difference between 2017 Fall WEO and real-time estimates) for 11 countries and averages.
- Selected aggregated results:
  - Unweighted average:
    - Data revisions: Mean 0.18, p-value 0.00
    - Forecast errors: Mean 0.60, p-value 0.00
    - Judgement: Mean 0.76, p-value 0.00
    - Total bias 1/: Mean 1.53, p-value 0.00
    - WEO revisions 2/: Mean 1.31, p-value 0.00
  - Weighted average (GDP weights):
    - Data revisions: Mean 0.11, p-value 0.00
    - Forecast errors: Mean 0.66, p-value 0.00
    - Judgement: Mean 0.95, p-value 0.00
    - Total bias 1/: Mean 1.74, p-value 0.00
    - WEO revisions 2/: Mean 1.42, p-value 0.00
- Country examples (selected means and significance):
  - Austria:
    - Data revisions: Mean 0.19, p-value 0.00
    - Forecast errors: Mean 0.44, p-value 0.02
    - Judgement: Mean 0.53, p-value 0.00
    - Total bias: Mean 1.26, p-value 0.00
    - WEO revisions: Mean 1.39, p-value 0.00
  - France:
    - Data revisions: Mean 0.07, p-value 0.09
    - Forecast errors: Mean 0.71, p-value 0.00
    - Judgement: Mean 1.37, p-value 0.00
    - Total bias: Mean 2.15, p-value 0.00
    - WEO revisions: Mean 1.45, p-value 0.00
  - Spain:
    - Data revisions: Mean 0.37, p-value 0.02
    - Forecast errors: Mean 0.55, p-value 0.05
    - Judgement: Mean 0.64, p-value 0.01
    - Total bias: Mean 1.75, p-value 0.00
    - WEO revisions: Mean 2.28, p-value 0.00
- Interpretation: the sum of the three measured sources of bias (data revisions, forecast errors, judgment) is close to the actual WEO revisions, indicating these channels account for most of the observed negative bias and subsequent positive revisions.

*Source: Box 2. Real-Time Output Gap Assessments in International Organizations (wpiea2019200-print-pdf).*

### 0.2 percentage points of potential GDP. According to this exercise, just below half of WEO

### wpiea2019200-print-pdf - 0.2 percentage points of potential GDP. According to this exercise, just below half of WEO

### Relevance to Monetary Policy and Inflation Forecast
- Monetary policy reaction functions usually include a measure of slack (most often captured by the output gap) and a measure of inflation expectations, with leads and lags to denote forward-looking policy and gradual adjustment through interest rate persistence.
- Real-time biases in output gap estimates could lead to policy errors through:
  - direct impact on the policy instrument (Orphanides, 2003), or
  - lower predictability of inflation (Orphanides and van Norden, 2005).
- Empirical tests reported:
  - Phillips curve relationships are strongly confirmed based on final 2017 WEO data.
  - The coefficient for the real-time output gap is small and generally non-significant statistically.
  - The real-time WEO output gap does not provide additional information beyond past inflation and inflation expectations.
  - Only when sample size is reduced to the years when countries entered the monetary union can the real-time output gap be statistically associated with inflation in some specifications; such associations are rare and not robust.
- Correlation and informational content:
  - Simple correlation coefficients between changes in real-time and final output gap estimates are positive at 67 percent for 11 baseline countries and 28 percent for all countries, and are highly significant.
  - Real-time output gaps can be informative about the direction or turning points of final output gaps and hence inflation, despite large revisions.
- Consequences for inflation forecasts and policy:
  - Lack of a “real-time” Phillips curve and negligible ex-post revisions to WEO inflation forecasts mean the real-time output gap bias is not reflected in inflation forecasts.
  - There is an important disconnect between inflation (fluctuating around 2 percent, close to the euro area inflation expectations) and persistently negative real-time output gaps.
  - Due to high inflation persistence in Europe and small revisions, inflation in real time is a useful predictor of final inflation and informative for monetary policy.
- Caution:
  - To the extent that real-time output gap bias leads to loss in inflation predictability, or policy rules rely on real-time estimates of cyclical position, monetary policy can suffer from information uncertainty.

### Relevance to Fiscal Policy Outcomes
- Role of output gap in fiscal policy:
  - Output gaps are used to separate fiscal balances into cyclical and structural components; structural balance informs medium-to-long-term fiscal stance and debt consolidation objectives.
  - If real-time output gaps fail to disentangle trend from cycle, long-term public debt implications of a given deficit are inaccurately estimated.
  - A negative output gap bias in real time would result in an excessively optimistic view of the long-term level of output and fiscal revenues.
- Empirical magnitudes and simulation evidence:
  - Ley and Misch (2014) using WEO data for 175 countries over 17 years find that in more than one-fifth of cases the implied revisions of overall and structural fiscal balances exceed 1 percent of GDP.
  - Thought-experiment translation of real-time output gap bias into cyclical component of fiscal balances yields a resulting bias in fiscal policy ranging from 0.7 to 0.9 percent of potential GDP annually.
    - Year after year overestimation of potential would imply significant unplanned deficits and large debt buildup over a 19-year baseline sample.
    - In the authors’ exercise the average increase in debt over 19 years is around 15 percent of GDP and can reach close to 25 percent of GDP in countries with larger bias.
  - Eyraud and others (2018) and Eyraud and Wu (2015) find that for the period 2003–16 the output gap was underestimated in real time by 1.3 percentage points of GDP on average.
    - Assuming elasticity of revenue to output = 1, elasticity of expenditure to output = 0, and average expenditure to GDP ratio = 45 percent, the underestimation implies an overestimation of cyclically adjusted balances by 0.5 percentage point on average per year.
    - Given euro area countries are subject to cyclically-adjusted balance rules, negative bias in authorities’ real-time output gap estimates may have led to unplanned deficits.
- Cross-sectional associations:
  - Drawing on WEO 1994–2017 data, there is a positive association between debt-to-GDP ratio in 2017 and average output gap revisions over the past 24 years.
  - Average revision in higher-debt countries amounts to 1.1 percent of potential GDP, while real-time estimates in lower-debt countries are on average revised by 0.5 percent.
  - WEO output gap revisions over 1994–2012 are positively associated with WEO 5-year ahead forecast errors for the public debt ratio.
- Box 4 findings (summarized):
  - The final output gap estimate is positively associated with primary fiscal balances.
  - Revisions to real-time output gap estimates are strongly negatively associated with primary fiscal balances: larger upward revisions to WEO output gaps are associated with lower primary fiscal balances.
  - Excess slack in WEO real-time estimates is related to both overly optimistic forecasts and overoptimistic potential growth incorporated through judgment; both are associated with lower primary balances.
  - These associations suggest over-optimism is strongly linked to lower-than-expected primary balances and debt accumulation.

### Key Findings and Quantitative Statistics
- Attribution of WEO revisions: just below half of WEO revisions can be attributed to forecast errors and the remaining half to judgment, with data revisions having a minor role.
- Correlation coefficients:
  - 67 percent correlation between changes in real-time and final output gap estimates for 11 baseline countries.
  - 28 percent correlation for all countries.
- Output gap revisions by debt group:
  - Average revision in higher-debt countries: 1.1 percent of potential GDP.
  - Average revision in lower-debt countries: 0.5 percent of potential GDP.
- Fiscal bias magnitudes:
  - Bias in fiscal policy from output gap bias ranges from 0.7 to 0.9 percent of potential GDP annually (thought experiment).
  - Average increase in debt over 19 years in exercise: around 15 percent of GDP; can reach close to 25 percent for countries with larger bias.
  - Eyraud et al. (2003–16): real-time underestimation of output gap by 1.3 percentage points of GDP on average → overestimation of cyclically adjusted balances by 0.5 percentage point on average per year.
- Inflation context:
  - Inflation fluctuating around 2 percent; inflation persistence makes real-time inflation a useful predictor of final inflation.

### Policy Recommendations and Implications
- Fiscal prudence given uncertainty:
  - The uncertainty in estimating potential GDP calls for fiscal policy prudence, particularly for high debt countries that lack fiscal space.
  - Avoid reliance on point estimates of the output gap; instead rely more on the whole distribution of output gap estimates.
  - Use dispersion and skewness of estimated distributions to inform policy calibration.
- Rule design and resilience to measurement error:
  - Build safety margins in fiscal policy targets; do not take real-time estimates of structural balances at face value.
  - Expenditure rules that allow automatic stabilizers to work can be more resilient to measurement errors, though more sensitive to initial conditions (Eyraud and others, 2018).
  - Counterfactual simulations (Andrle and others, 2015) show expenditure rules could have led to significantly smaller differences between real-time and ex post public debt outcomes than cyclically adjusted balance rules for France and Italy.
- Caution for monetary policy:
  - Given limited usefulness of real-time output gaps for inflation forecasting and potential for information uncertainty, monetary policy should be cautious about relying heavily on real-time output gap estimates when setting policy.

*Italic: Sources: IMF WEO and staff calculations; IMF WEO and staff estimates.*

### Box 4. Fiscal Outcomes and Real-Time Output Gap Bias

### Box 4. Fiscal Outcomes and Real-Time Output Gap Bias

### Methodology
- Framework: Adaptation of fiscal reaction functions popularized by Bohn (1998, 2008), relating primary fiscal balances to:
  - lagged primary balance (persistence),
  - lagged public debt,
  - detrended real government consumption (Barro, 1979),
  - WEO output gap (final vintage) and revisions to real-time WEO output gap,
  - MVF simulated sources of real-time bias (data revisions, forecast errors, judgement).
- Controls: institutional and economic variables, 5-year ahead debt forecast errors, real-time inflation and unemployment gaps.
- Estimator: bias-corrected Least Squares Dummy Variable (LSDV) dynamic panel estimator by Bruno (2005) with bootstrapped standard errors.
- Baseline sample: 11 larger euro area countries for the period 1994–2012.
- Observations: 195.

### Regression results (selected coefficients from Table 7)
- Lagged primary balance (Primary balance (t-1)):
  - 0.641***, 0.498***, 0.393***, 0.473***, 0.406***, 0.514***, 0.527***, 0.518***
- Output gap (final vintage):
  - 0.327***, 0.801***, 0.767***, 0.154, 0.335**, 0.196
- Revision to real-time output gap against 2017 vintage:
  - -0.714***, -0.849***
- Output gap (t+5 vintage):
  - 0.834***, 0.832***
- Revision to real-time output gap against (t+5) vintage:
  - -0.612***, -0.745***
- MVF components:
  - Data revisions (MVF): 1.023***
  - Forecast error (MVF): -0.444*
  - Judgement (MVF): -0.490**
- Public debt (t-1):
  - 0.039**, 0.035**, 0.082***, 0.041**, 0.079***, 0.082***, 0.086***, 0.085***
- Current account (in percent of GDP):
  - 0.254***, 0.217**, 0.277***, 0.241***, 0.281***, 0.255***, 0.234**, 0.276***
- Trade openness:
  - -0.060***, -0.084***, -0.111***, -0.089***, -0.107***, -0.073***, -0.074***, -0.096***
- Inflation gap:
  - 0.656***, 0.656***, 0.784***, 0.783***, 0.783***
- Unemployment gap:
  - 0.004, 0.065, -0.228, -0.227, -0.290*
- NAIRU gap:
  - -0.294**, -0.251*, -0.160, -0.190, -0.284*

### Key empirical findings
- Real-time output gap revision effects:
  - Revisions to real-time output gaps are statistically highly significant and economically large — nearly as large as the coefficient on the final output gap.
  - Predominantly positive (upwards) revisions to WEO real-time output gaps averaging to 1.3–1.4 percent of potential GDP are associated with lower primary balance estimates in the order of 0.8–1.2 percent of GDP annually.
  - The association operates through both MVF-simulated forecast errors and judgement with similar coefficients.
- MVF components:
  - Data revisions have a statistically significant positive coefficient (but small economic magnitude, driven mostly by one country).
  - Forecast errors and judgement components show negative associations with primary balances (forecast error (MVF): -0.444*; judgement (MVF): -0.490**).
- Other determinants:
  - High persistence in primary balances (large, significant autoregressive component).
  - Positive coefficient on lagged public debt, consistent with governments attempting higher primary balances when debt is high (weak sustainability condition).
  - Current account positively associated with primary balance, capturing twin deficits and international spillovers.
- Robustness:
  - Baseline bias-corrected LSDV estimates robust to time effects (capturing basic cross-country dependence).
  - Results robust to sample size and real-time debt (results not reported).

### Interpretation and mechanisms
- Direction of bias:
  - WEO real-time output gap estimates for main euro area countries are large and systematically upwards in revisions; staff’s real-time output gap estimates are persistently downward biased in real time.
  - Counterfactual simulations: over-estimation of productive capacity and over-optimistic growth forecasts are primary drivers of the negative real-time bias.
- Forecast behavior:
  - WEO forecasters tend to predict the mode of the growth distribution; in presence of downwardly skewed risks this mode exceeds average growth, contributing to forecast errors.
- Policy linkage:
  - Overestimation of slack and potential growth in WEO data is associated with lower primary balances, higher public debt ratios, and faster debt accumulation than projected.
  - Biases in output gap estimates may reflect an underappreciation of a slowdown in potential growth; cyclical revenue losses initially judged as cyclical became structural losses, leading to debt buildup.
  - National authorities’ real-time output gap estimates show similar negative bias and may have influenced fiscal calibration toward excessive deficits.

### Implications for inflation and monetary policy
- Real-time output gaps:
  - Real-time output gaps are not useful to predict inflation in a reliable real-time Phillips curve sense.
  - Real-time output gaps at best contain information on directional changes in business cycles and inflation.
- Inflation as indicator:
  - Because inflation is persistent and not revised, real-time inflation remains useful to forecast inflation.

### Policy recommendations and options to reduce real-time errors
- Estimation approach:
  - Move toward a multivariate filter (MVF) that imposes economic structure to produce more desirable real-time potential output estimates while allowing for expert judgement integrated via growth accounting and production function approaches.
- Enhancements under study:
  - Increase the number of indicators (capacity utilization, investment to GDP, labor vacancy rates, etc.) to improve MVF precision and reduce ex-post revisions.
  - Incorporate measures of forecast distribution asymmetry in the central forecast — e.g., rely on the mean instead of the mode of forecasts to reduce forecast errors and ex-post revisions.
  - Use complementary measures of potential output (for example, IMF (2015) concept of “sustainable” output) to provide timely information when measures diverge.
  - Policymakers could build safety margins or robust measures — build buffers during upswings to increase resources for counter-cyclical fiscal policy in downturns.
- Caution:
  - Exercise caution in incorporating off-model information (structural or fiscal reform impacts); judgement can contribute to negative bias if too sensitive to end-of-sample conditions.
- Overarching guidance:
  - Avoid exclusive focus on real-time output gap estimates when calibrating counter-cyclical policy; adopt an encompassing approach using multiple measures and buffers.

*Source: Box 4. Fiscal Outcomes and Real-Time Output Gap Bias, IMF working paper content provided in the supplied PDF excerpt.*

### Appendix Figure 1. WEO Real Time Output Gap Estimates and Subsequent Revisions

### Appendix Figure 1. WEO Real Time Output Gap Estimates and Subsequent Revisions

### Real-time versus Final WEO output gap revisions (panel summary)
- Chart panels compare "Real-time" and "Final WEO" output gap estimates across countries and vintages for the sample years shown (e.g., 19 94 through 20 15 as labeled in figures for multiple countries).
- Countries included in the figure panels: France, Germany, Italy, Spain, Greece, Portugal, Austria, Belgium, Finland, Netherlands, Ireland.
- Vertical axis tick labels shown in panels include exact values such as:
  - France/Germany/Italy/Austria/Belgium: -4, -3, -2, -1, 0, 1, 2, 3 (and in some panels extended to 4 or 5).
  - Spain: -8, -6, -4, -2, 0, 2, 4, 6, 8.
  - Greece: -12, -9, -6, -3, 0, 3, 6, 9, 12, 15.
  - Portugal: -7, -6, -5, -4, -3, -2, -1, 0, 1, 2, 3, 4.
  - Ireland: -8, -6, -4, -2, 0, 2, 4, 6, 8.
- Time axis labels in the panels use year fragments as shown (for example: 19 94, 19 97, 20 00, 20 03, 20 06, 20 09, 20 12, 20 15).

### Continued panels and figure set context
- Figures are followed by Appendix Figure 2, Appendix Figure 3, and Appendix Figure 4 presenting related vintage and methodology comparisons:
  - Appendix Figure 2: "WEO Real Time, 2-Year Ahead and 2-Year Back Estimates" — panels for the same set of countries with vertical axis tick labels repeated (e.g., -5 to 2 for many countries; -8 to 6 for Spain; -12 to 9 for Greece; etc.) and time axis labels 1994 1997 2000 2003 2006 2009 2012 2015.
  - Appendix Figure 3: "Real-Time and Quasi Real-Time MVF Estimates" — panels spanning 1990 through 2017 with vertical axis tick labels such as -3 to 2, -4 to 2, -10 to 6, -16 to 4, -6 to 4, -4 to 2, -8 to 4, -3 to 3, -8 to 8.
  - Appendix Figure 4: "Vintage Forecast, Perfect Foresight, and WEO Real-Time Estimates" — panels spanning 1990 through 2017 with vertical axis tick labels varying by country (e.g., -4 to 3; -5 to 3; -6 to 4; -7 to 7; -14 to 13; -6 to 6; -3 to 4; -10 to 10; -6 to 6; -10 to 10).

### Sources and calculations
- Sources: IMF WEO and staff calculations (as labeled in the figures).

---

### Appendix Table 1. Real-Time Output Gaps: Mean Estimates and Significance at Times of Low Real-Time Unemployment (1994–2017)

- Table columns appear to present "Mean", "SE", and "p-value" for two conditions:
  - "Output gap estimates when real time unemployment is below sample average (n=212)"
  - "Output gap estimates when unemployment is below NAIRU in real-time (n=80)"
- Reported values as presented in the table (row by row, preserving text and numeric formatting exactly):
  - Euro Area -1.10.270.00 Euro Area -0.25
  - Austria -1.00.230.00 Austria -1.440.230.00
  - Belgium -0.80.210.00 Belgium -0.470.210.03
  - France -1.80.240.00 France
  - Germany -0.30.280.34 Germany -0.180.260.48
  - Italy -2.20.270.00 Italy -2.81
  - Luxembourg 1.52.030.45 Luxembourg 0.250.360.50
  - Netherlands -0.30.430.43 Netherlands 0.450.430.30
  - Finland -1.90.430.00 Finland 0.580.210.01
  - Greece 0.80.390.03 Greece 2.310.200.00
  - Ireland 1.10.640.10 Ireland 1.970.690.01
  - Malta 1.20.010.00 Malta
  - Portugal -1.80.440.00 Portugal -1.450.480.00
  - Spain -0.40.200.08 Spain -0.200.290.49
  - Cyprus 2.52.890.39 Cyprus
  - Slovakia -1.40.520.01 Slovakia
  - Estonia -1.00.600.11 Estonia
  - Latvia -0.20.020.00 Latvia
  - Lithuania -0.70.640.30 Lithuania -0.490.250.05
  - Slovenia -0.80.780.33 Slovenia 0.290.910.76

---

### Appendix Table 2. Potential Fiscal Implications of Negative Real Time Output Gap Bias (Percent of Potential GDP)

- Sources: IMF WEO and staff estimates.
- Note: Computed using the EC’s semi-elasticity of budget balance to changes in the output gap (see Mourre and others, 2014).
- (No numeric detail lines appear in the supplied excerpt beyond the title and note.)

---

### Appendix Table 3. Real-Time Output Gaps for High and Low Debt Countries: Mean Estimates and Significance (1994–2017)

- Table structure and reported values preserved exactly as shown:
  - Headings: "Common Sample" and "Full Sample"; subcolumns include "Total MVF Bias", "WEO Final Revisions", "Total MVF Bias", "WEO Final Revisions".
  - Country rows with numeric entries:
    - Austria 0.70.813.915.4
    - Belgium 0.70.513.110.1
    - France 1.30.924.516.6
    - Germany 0.80.714.813.7
    - Italy 1.40.927.517.0
    - The Netherlands 0.40.47.77.2
    - Finland 1.31.424.826.9
    - Greece 0.10.52.19.3
    - Ireland 0.9-0.116.5-2.5
    - Portugal 0.80.816.015.5
    - Spain 0.91.217.923.4
  - Average 0.90.716.313.9
  - Additional labeled lines (preserving formatting):
    - "Average annual bias in fiscal balances"
    - "Potential debt buildup over a sample of 19 years"
  - Reported means and significance lines shown later in the table excerpt:
    - Meanp-valueDF Meanp-valueDF
    - Low debt-0.90.00155 Low debt-0.40.02155
    - High debt-1.60.00194 High debt-0.50.02194
    - Difference-0.750.01349 Difference-0.10.78349
    - Real-time output gaps (unweighted) 2017 WEO vintage (unweighted)
    - Meanp-value DF Meanp-value DF
    - Low debt-0.90.00155 Low debt -0.10.63223
    - High debt-1.60.00194 High debt -0.50.01214
    - Difference-0.750.01349 Difference-0.4 0.17437
    - Real-time output gaps (unweighted) 2017 WEO vintage (unweighted)

---

### Appendix Table 4. Robustness of Primary Balance Regressions

- Sources: IMF WEO and staff estimates.
- Note: Standard errors in parentheses. ***, **, and * denote significance at 1, 5, and 10 percent, respectively. Baseline sample for 11 larger euro area countries for the period 1994–2012. The bias-corrected LSDV dynamic panel data estimates are reported with bootstrapped standard errors.
- Regression results presented across columns labeled (1) through (8) and model types OLS, OLSFELSDV, OLS, OLSFELSDV with "Revisions agains 2017 WEO" and "Revisions against (t+5) vintage" context. Key coefficient entries and standard errors preserved exactly:
  - Primary balance (t-1) 0.595***0.2200.263***0.577***0.2430.286***
    - (0.12)(0.16)(0.07)(0.12)(0.15)(0.07)
  - Output gap (final vintage) 0.655***0.625***0.3010.296**
    - (0.20)(0.16)(0.24)(0.15)
  - Revision to real time output- -0.593***-0.387**-0.548*-0.531*** gap against 2017 vintage
    - (0.22)(0.18)(0.32)(0.15)
  - Output gap (t+5) vintage 0.718***0.568***0.2780.274*
    - (0.19)(0.14)(0.23)(0.15)
  - Revision to real time output- -0.507*-0.394*-0.547*-0.531*** gap against (t+5) vintage
    - (0.27)(0.24)(0.32)(0.15)
  - Public debt (t-1) 0.038***0.021***0.100***0.097***0.037***0.019***0.092***0.090***
    - (0.01)(0.01)(0.03)(0.02)(0.01)(0.01)(0.03)(0.02)
  - Current account 0.426***0.162***0.293***0.276***0.423***0.174***0.286***0.268*** (in percent of GDP)
    - (0.06)(0.05)(0.08)(0.08)(0.06)(0.05)(0.08)(0.08)
  - Government real 0.023-0.041-0.017-0.0140.020-0.036-0.022-0.019 consumption gap
    - (0.05)(0.04)(0.04)(0.05)(0.05)(0.04)(0.04)(0.05)
  - Trade openness -0.018*-0.010-0.110**-0.103***-0.018*-0.011-0.106**-0.099*** 
    - (0.01)(0.01)(0.05)(0.03)(0.01)(0.01)(0.05)(0.03)
  - IMF arrangement -1.3540.233-0.267-0.231-1.3780.339-0.333-0.303
    - (1.17)(0.93)(1.05)(0.76)(1.19)(0.93)(1.06)(0.77)
  - Election year -0.507-0.184-0.153-0.156-0.505-0.176-0.177-0.178
    - (0.47)(0.34)(0.32)(0.33)(0.46)(0.34)(0.32)(0.33)
  - 5Y ahead public debt 0.011-0.0180.0340.032*0.005-0.0110.0270.025 forecast error
    - (0.02)(0.02)(0.03)(0.02)(0.02)(0.02)(0.02)(0.02)
  - Inflation gap 0.767***0.715***0.615*0.602***0.744**0.671**0.685*0.672***
    - (0.28)(0.26)(0.33)(0.20)(0.30)(0.27)(0.36)(0.20)
  - Unemployment gap -0.335*0.169-0.325*-0.305*-0.2770.133-0.294*-0.274
    - (0.18)(0.13)(0.19)(0.17)(0.20)(0.14)(0.17)(0.17)
  - NAIRU gap -0.1590.138-0.483***-0.461***-0.1430.115-0.447***-0.425***
    - (0.15)(0.14)(0.14)(0.13)(0.15)(0.14)(0.13)(0.13)
  - Fixed effects nonoyesyesnonoyesyes
  - Time effects nonoyesyesnonoyesyes
  - Constant -0.357-0.054-9.188**-0.3710.121-8.414**
    - (1.13)(1.07)(3.74)(1.12)(1.04)(3.44)
  - R-squared 0.490.690.840.500.680.84
  - Adjusted R-squared 0.460.660.800.460.660.80
  - Observations 197195195195197195195195

---

*Sources: IMF WEO and staff calculations.*

---


_Source: https://www.imf.org/-/media/files/publications/wp/2019/wpiea2019200-print-pdf.pdf_