## Convergence from Above

## Source details

**Canonical URL:** [Convergence from Above](https://www.imf.org/-/media/files/publications/wp/2026/english/wpiea2026159-source-pdf.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/wp/2026/english/wpiea2026159-source-pdf.pdf.md)
- [Structured JSON version](/-/media/files/publications/wp/2026/english/wpiea2026159-source-pdf.pdf.json)

---

### I. Introduction and central question
- Core idea: "Convergence from above" — an economy lies above the income level implied by its current fundamentals, and subsequent growth is weak because the economy is adjusting toward that lower benchmark.
- Distinctions:
  - Not simply beta-convergence.
  - Not primarily a convergence-club argument.
  - Not merely a boom-bust story (though some overshooting episodes reflect temporary booms, terms-of-trade movements, or credit expansions).
- Conceptual claim: observed convergence can reflect two distinct processes simultaneously — upward mobility from below and downward adjustment from above — and standard convergence regressions can combine these distinct forms of adjustment.

### II. Mechanism and modeling intuition (moving steady states)
- Baseline production and accumulation (preserved notation):
  - Y_it = K_it^α (A_it L_it)^(1−α), 0<α<1.
  - K̇_it = s_it Y_it − δ K_it.
  - L̇_it / L_it = n_it, Ȧ_it / A_it = g_it.
  - k_it = K_it / (A_it L_it).
  - k̇_it = s_it k_it^α − (n_it + g_it + δ) k_it.
  - φ_it ≡ n_it + g_it + δ.
  - Steady-state: k_it^* = (s_it / φ_it)^{1/(1−α)}, y_it^* = A_it (s_it / φ_it)^{α/(1−α)}.
- Overshooting gap and interpretation:
  - ω_it = ln y_it − ln y_it^*.
    - ω_it < 0: economy below steady state (room to catch up).
    - ω_it > 0: economy above steady state (convergence from above).
  - An economy can become “too rich” relative to fundamentals via downward revisions in y_it^* even if actual income does not rise (d ln y_it = 0 and d ln y_it^* < 0 ⇒ d ω_it > 0).
- Asymmetric adjustment implication:
  - Define z_it ≡ k_it / k_it^*.
  - Exact dynamics: k̇_it / k_it = φ_it (z_it^{α−1} − 1).
  - Because 0<α<1, adjustment from below (z_it<1) is faster than adjustment from above (z_it>1) for equal proportional deviations; convergence from above tends to appear as gradual weaker growth rather than abrupt collapse.
- Four empirical predictions:
  1. Declines in predicted steady-state paths increase probability of entering above-steady-state region.
  2. Countries above their steady-state paths subsequently experience persistently weaker growth.
  3. Adjustment from above may be slower than adjustment from below.
  4. Observed convergence in the world income distribution may reflect both upward mobility from below and downward adjustment from above.

### III. Empirical strategy and measures
- Trajectory estimation:
  - Long-run fitted trajectory from ln y_it = β_0 + β_1 ln s_it − β_2 ln(n_it + g + δ) + β_3 ln h_it + μ_i + τ_t + ε_it, with h_it human capital.
  - Baseline macro variables primarily from Penn World Table 11.0; human-capital index from Penn World Table 11.0; Barro–Lee used in some specifications; Maddison Project Database, 2020 vintage for long-run historical series.
  - Non-overlapping five-year intervals; subsequent growth measured over the five years following initial observation.
  - Panel: up to 185 economies and twelve five-year periods. Preferred regressions use 1,305 country–period observations for 137 economies; sample varies by specification.
- Overshooting measure:
  - ω̂_it = ln y_it − ln y_it^*̂ (positive indicates observed income above implied fundamentals).
  - Distributional facts for ω̂_it: standard deviation = 0.334 log points; 10th percentile = −0.387; 90th percentile = 0.378.
  - Positive-gap component ω̂_it^+ = max(ω̂_it, 0); negative-gap component ω̂_it^− = min(ω̂_it, 0).
  - Distributional approach follows Quah (1993, 1996): treat convergence as movement within the full cross-country distribution.

### IV. Main empirical findings and key statistics
- Baseline asymmetric mean-reversion (five-year horizon, preferred specification):
  - Decompose ω̂_it into ω̂_it^− and ω̂_it^+.
  - Key coefficients (preferred five-year specification including initial income):
    - ω̂_it^+ coefficient = −0.511; standard error = 0.184.
    - ω̂_it^− coefficient = −0.444; standard error = 0.211.
  - Interpretation:
    - Both ω̂_it^− and ω̂_it^+ are negative and significant: mean reversion to estimated trajectories from both below and above.
    - A one-standard-deviation increase in positive overshooting predicts approximately 10 log points lower cumulative growth over the subsequent five-year period (roughly 2 percentage points lower annual growth).
  - Table I fit and sample:
    - Observations = 1,532; Countries = 145; R-squared up to 0.093.
- Longer-horizon forward-growth checks (ten-year horizon):
  - Positive-overshooting coefficient preserves negative sign but is smaller and less precisely estimated.
  - Observations reported: 1,387; R-squared up to 0.047.
- Distribution dynamics and compression at the top (transition probabilities):
  - Five-year relative-income transition matrices conditional on being above/below fitted trajectory.
  - Key probability example (Q5 → Q4):
    - If initially above fitted trajectory: 0.2069 (20.69 percent).
    - If initially below fitted trajectory: 0.0769 (7.69 percent).
  - Accounting decomposition (1960–2019 full sample):
    - Compression from above accounts for roughly one-third of the systematic convergence component (point estimate close to one-third).
- Accounting decomposition (Table A5, exact figures):
  - Full sample (1960-2019): Σ Compression from above -0.0201; Σ Catch-up from below +0.4600; Σ Idiosyncratic -2.3246; Net Δσ -1.8846; Share from compression-above 33.7%; Share from catch-up-below 66.3%
  - Pre-1990 (1960-1985): Share from compression-above 49.5%; Share from catch-up-below 50.5%
  - Post-1990 (1990-2019): Share from compression-above 14.5%; Share from catch-up-below 85.5%
  - Decadal examples preserved exactly in source (selected):
    - 2000s (2000-2005): Σ Compression from above -0.1247; Σ Catch-up from below -0.0787; Net Δσ -0.7987; Share from compression-above 100.0%; Share from catch-up-below 0.0%
    - 2010s (2010-2019): Σ Compression from above +0.0473; Σ Catch-up from below +0.3503; Net Δσ -0.0097; Share from compression-above 11.9%; Share from catch-up-below 88.1%
- Fragility, downside risk, and crisis associations (discrete-time hazard models, unconditional ω̂^+ specifications):
  - Stagnation (ω̂^+ alone): +0.8750 (0.3089); unconditional event rate 0.225; at-risk observations 1117.
  - Banking crisis (ω̂^+ alone): +1.5860 (0.3296); unconditional event rate 0.129; at-risk observations 1110.
  - Sovereign-debt crisis (ω̂^+ alone): +0.7417 (0.5635); unconditional event rate 0.091; at-risk observations 1132.
  - Loss of frontier (ω̂^+ alone): +1.9660 (0.3673); unconditional event rate 0.345; at-risk observations 1532.
  - Robustness: banking-crisis and loss-of-frontier relationships remain strong after adding initial income and recent growth momentum; stagnation association weakens with controls; sovereign-debt crisis relationship not robust across specifications.
- Dynamic response profile around initial trajectory gap (Table A7, β⁺ and β⁻ at horizons h):
  - Selected horizons (β⁺; β⁻; N):
    - h = 1: β⁺ -0.0998*** (0.0259); β⁻ -0.1037*** (0.0183); N 7648
    - h = 5: β⁺ -0.3670*** (0.0884); β⁻ -0.3244*** (0.0673); N 7503
    - h = 10: β⁺ -0.3783*** (0.1210); β⁻ -0.3041*** (0.1032); N 6778
    - h = 15: β⁺ -0.2742** (0.1084); β⁻ -0.2060** (0.1036); N 6053
  - Kraay standard errors reported.
- Summary statistics and coverage (selected exact moments from Appendix B):
  - Omega (MRW gap): N 1532; Mean 0.000; Std 0.334; P10 -0.387; Median 0.004; P90 0.378
  - Omega+ (above fitted-trajectory component): N 1532; Mean 0.123; Std 0.196; P10 0.000; Median 0.004; P90 0.378
  - Omega- (below fitted-trajectory): N 1532; Mean -0.123; Std 0.208; P10 -0.387; Median 0.000; P90 0.000
  - Above fitted-trajectory indicator (ω̂ > 0): N 2220; Mean 0.349; Std 0.477; P10 0.000; Median 0.000; P90 1.000

### V. Robustness, inference, and caveats
- Addressed mechanical mean reversion concerns:
  - Controls for initial income.
  - Use of non-overlapping five-year periods.
  - Leave-one-observation-out trajectory construction.
  - Bootstrap and permutation exercises reported in Appendix B.
- Whole-pipeline country-bootstrap inference (Table A8, selected results):
  - β(ω < 0) catch-up slope: Point estimate -0.0760; Bootstrap SE 0.0201; 95% CI lo -0.1180; 95% CI hi -0.0393; p (boot, two-sided) 0.0000; N 1387
  - β(ω ≥ 0) adjustment-from-above slope: Point estimate -0.0419; Bootstrap SE 0.0187; 95% CI lo -0.0772; 95% CI hi -0.0071; p (boot, two-sided) 0.0133; N 1387
  - Asymmetry (β(ω≥0) − β(ω<0)): +0.0340; Bootstrap SE 0.0304; 95% CI lo -0.0226; 95% CI hi +0.0914; p 0.2667; N 1387
- Permutation and Romano–Wolf multiple-testing checks:
  - Permutation inference (Table A11) shows strong evidence for association between ω̂^+ and banking crises and loss of frontier; sovereign-debt and stagnation associations weaker or sample-dependent.
  - Romano–Wolf adjustment (Table A10) preserves significance for stagnation (g<0 next 5y), large decline (g<−5% next 5y), decline next 10y, loss of frontier proximity, and banking crisis next 5y; currency and sovereign-debt crisis do not survive correction.
- Robustness to sample exclusions:
  - Findings robust to excluding major oil exporters, transition economies, small financial centers, and periods with systemic banking crises or large external shocks.
- Caveats:
  - Overshooting may partially reflect omitted fundamentals or temporary cyclical booms rather than persistent deterioration.
  - Trajectory is a generated regressor; bootstrap and alternative specifications mitigate but do not eliminate inference concerns.

### VI. Interpretation, policy implications, and conceptual contribution
- Conceptual reframing:
  - Observed convergence is ambiguous: similar convergence coefficients may reflect upward catch-up, compression from above, or both.
  - Sigma convergence and single-regression metrics do not identify whether progress arises from rising productive capacity among poorer countries or weakening frontier dynamism.
- Empirical summary:
  - Economies above their predicted long-run trajectories subsequently grow more slowly.
  - Adjustment from above tends to be slower and more persistent than adjustment from below.
  - Compression from above contributes materially to observed convergence; over 1960–2019 it accounts for roughly one-third of the systematic convergence component (share varies by sub-period).
- Policy-relevant implications (actionable distinctions preserved from source):
  - Diagnose weak growth by distinguishing:
    - Temporary cyclical slack (appropriate for demand-side or cyclical interventions).
    - Temporary adjustment after a boom.
    - Movement toward a lower long-run trajectory (convergence from above), which calls for structural reforms, productivity-enhancing policies, and measures addressing demographic or diffusion constraints.
  - Monitor overshooting indicator ω̂_it alongside standard convergence metrics to detect heightened downside risk, especially for frontier economies exhibiting positive overshooting and banking-sector fragility.
  - For forecasting and debt-sustainability analysis, account for the possibility that long-run potential estimates may be revised downward over time, implying slower future growth even without absolute income declines.
- Final conceptual claim:
  - Convergence should be read as movement within an evolving world distribution rather than movement toward fixed equilibria; a complete account of global development must encompass both upward catch-up and potential downward adjustment from above.

*IMF Working Paper — "Convergence from Above" (excerpted Sections II–VI; empirical results and appendices summarized as presented in source PDF).*

### References .............................................................................................................

### Convergence from Above

### I. Introduction and central question
- Core idea: "Convergence from above" refers to a specific dynamic in which an economy lies above the income level implied by its current fundamentals, and subsequent growth is weak because the economy is adjusting toward that lower benchmark.
- Distinction emphasized:
  - Not simply beta-convergence.
  - Not primarily a convergence-club argument.
  - Not merely a boom-bust story (though some overshooting episodes reflect temporary booms, terms-of-trade movements, or credit expansions).
- Key conceptual claim: observed convergence may reflect two conceptually distinct processes simultaneously — upward mobility from below and downward adjustment from above — and standard convergence regressions can combine these distinct forms of adjustment.

### II. Mechanism and modelling intuition
- Standard growth models: economies converge toward steady states determined by productivity, demographics, investment, and human capital.
- Novelty introduced: steady states are allowed to evolve over time (moving steady-state paths).
- Sources of downward shifts in steady-state benchmarks:
  - Productivity growth weakens.
  - Demographic structures deteriorate.
  - Industrial composition changes.
  - Technologies diffuse unevenly.
  - Economies become less effective at using existing technologies.
- Observable implication: an economy can become “too rich” relative to its fundamentals even if actual income does not rise, often signaled by repeated downward revisions in expectations about future productivity and long-run growth.

### III. Empirical strategy and measures
- Empirical implementation:
  - Estimate country-specific steady-state paths as functions of investment, demographics, and human capital.
  - Construct an overshooting measure given by the gap between actual income and predicted steady-state income.
  - Examine how this gap predicts subsequent growth and transition probabilities within the world income distribution.
- Interpretation: the overshooting gap is treated as a disciplined empirical proxy for the distance between observed income and the income level implied by a parsimonious set of long-run fundamentals, not as a direct observation of the true steady state.
- Distributional approach: follows Quah (1993, 1996) in treating convergence as movements within the full cross-country distribution rather than solely as a coefficient in a growth regression.

### IV. Main empirical findings
- Evidence supports the existence of convergence from above:
  - Economies above their predicted long-run trajectories subsequently experience weaker growth.
  - Adjustment from above appears slower and more persistent than convergence from below.
- Quantitative accounting:
  - In the preferred accounting exercise, compression near the top accounts for a material share of observed convergence, with a point estimate close to one-third over the full 1960–2019 sample.

### V. Relationship to existing literature
- Builds on and distinguishes itself from:
  - Barro and Sala-i-Martin (1992); Mankiw, Romer, and Weil (1992); Durlauf, Johnson, and Temple (2005) — by distinguishing convergence from below and convergence from above.
  - Quah (1993, 1996) — by using distribution-dynamics methods.
  - Pritchett (1997); Johnson and Papageorgiou (2020) — by engaging with literature on instability and discontinuity in long-run growth experience.
  - Work on resource-based overshooting, convergence clubs, and multiple steady states — differs by emphasizing evolving country-specific trajectories and coexistence of upward and downward adjustment.

### VI. Policy relevance and implications
- Distinguishing mechanisms matters for:
  - Forecasting.
  - Debt sustainability analysis.
  - Assessments of medium-term potential.
- Practical implication: weak growth may reflect:
  - Temporary cyclical weakness,
  - Temporary adjustment after a boom,
  - Movement toward a lower long-run trajectory (convergence from above).
- Broader interpretation: measured convergence may reflect progress at the bottom, decline at the top, or both; distinguishing these mechanisms changes how global development dynamics are interpreted.

*IMF Working Papers — Convergence from Above*

### Section IV constructs empirical proxies for long-run trajectories, estimates the dynamics of adjustment, and

### wpiea2026159-source-pdf - Section IV constructs empirical proxies for long-run trajectories, estimates the dynamics of adjustment, and

### Motivating facts: convergence, stagnation, and falling back
- Renewed unconditional convergence since the early 1990s coincides with prolonged stagnation and downward revisions in potential growth in parts of the advanced and middle-income world.
- Multiple mechanisms can produce apparent convergence: catch-up from below and downward adjustment from above (convergence from above), including demographic aging, commodity dependence, weak diffusion, premature deindustrialization, institutional rigidities, and financial excess.
- Key interpretive point: standard convergence metrics (single regression coefficients, sigma convergence) do not distinguish upward mobility from below versus compression from above; movements near the top of the distribution can materially affect measured convergence.

### Model: convergence with moving steady states (Section III)
- Baseline production and accumulation setup (preserved notation):
  - Y_it = K_it^α (A_it L_it)^(1−α), 0<α<1.
  - K̇_it = s_it Y_it − δ K_it.
  - L̇_it / L_it = n_it, Ȧ_it / A_it = g_it.
  - k_it = K_it / (A_it L_it).
  - k̇_it = s_it k_it^α − (n_it + g_it + δ) k_it.
  - Define φ_it ≡ n_it + g_it + δ.
  - Steady-state: k_it^* = (s_it / φ_it)^{1/(1−α)}, y_it^* = A_it (s_it / φ_it)^{α/(1−α)}.
- Define gap ω_it = ln y_it − ln y_it^*. Interpretation:
  - ω_it < 0: economy below steady state (room to catch up).
  - ω_it > 0: economy above steady state (convergence from above).
- Asymmetric adjustment implication:
  - With z_it ≡ k_it / k_it^*, dynamics k̇_it / k_it = φ_it (z_it^{α−1} − 1).
  - Because 0<α<1, adjustment from below (z_it<1) is faster than adjustment from above (z_it>1) for equal proportional deviations—adjustment from above tends to be gradual, via weaker relative growth rather than abrupt collapse.
- Empirical objective: use an empirical proxy for ln y_it^* and test whether ω̂_it predicts subsequent growth and distributional mobility, and whether adjustment is asymmetric.

### Empirical framework and key results (Section IV)
- Data, measurement, and sample:
  - Long-run fitted trajectory estimated from ln y_it = β_0 + β_1 ln s_it − β_2 ln(n_it + g + δ) + β_3 ln h_it + μ_i + τ_t + ε_it, where h_it is human capital.
  - Baseline macro variables primarily from Penn World Table 11.0; human-capital index from Penn World Table 11.0; Barro–Lee used in some specifications; long-run historical series from Maddison Project Database, 2020 vintage.
  - Non-overlapping five-year intervals used; subsequent growth measured over the five years following initial observation.
  - Panel: up to 185 economies and twelve five-year periods. Preferred fully specified growth regressions use 1,305 country–period observations for 137 economies; precise sample varies with specification.
- Overshooting gap:
  - ω̂_it = ln y_it − ln y_it^*̂. Positive indicates observed income above implied fundamentals; negative indicates below.
  - Distributional facts: standard deviation = 0.334 log points; 10th percentile = −0.387; 90th percentile = 0.378.
  - Notable upper-tail episodes include Japan after the late 1980s, parts of Southern Europe before the euro-area crisis, and selected commodity exporters during booms.

- Baseline asymmetric mean-reversion regressions (Table I, five-year horizon):
  - Specification decomposes ω̂_it into ω̂_it^− = min(ω̂_it, 0) and ω̂_it^+ = max(ω̂_it, 0).
  - Key coefficients in preferred five-year specification including initial income:
    - ω̂_it^+ coefficient = −0.511, standard error = 0.184.
    - ω̂_it^− coefficient = −0.444, standard error = 0.211.
  - Interpretation:
    - Both ω̂_it^− and ω̂_it^+ are negative and significant, consistent with mean reversion to estimated trajectories from both below and above.
    - A one-standard-deviation increase in positive overshooting predicts approximately 10 log points lower cumulative growth over the subsequent five-year period (roughly 2 percentage points lower annual growth).
    - Tests do not reject equality of |γ_2| and |γ_1|; evidence for asymmetric speeds (|γ_2| < |γ_1|) is suggestive but not conclusively established statistically in baseline specifications.
  - Table I observation counts and fit:
    - Observations = 1,532; Countries = 145; R-squared up to 0.093 in presented specifications.

- Longer-horizon forward-growth checks (Table II, ten-year horizon):
  - Forward regressions preserve negative sign on positive-overshooting coefficient but coefficients are smaller and less precisely estimated at longer horizons.
  - Observations reported: 1,387 (in specifications shown); R-squared up to 0.047.

- Distribution dynamics and compression at the top (Table III, transition matrices):
  - Five-year relative-income transition matrices estimated overall and conditional on being initially above or below the fitted trajectory.
  - Persistence high overall; mobility differs by overshooting sign.
  - Key probabilities:
    - Among economies initially in the highest relative-income state (Q5), probability of downward transition (into Q4) if initially above fitted trajectory = 0.2069 (20.69 percent), compared with 0.0769 (7.69 percent) if initially below fitted trajectory.
  - Accounting decomposition (observed convergence = catch-up from below + compression from above):
    - In the presented accounting exercise over the full 1960–2019 sample, compression from above accounts for roughly one-third of the systematic convergence component (share varies across sub-periods).
    - Mean subsequent growth declines monotonically across deciles from the most below-steady-state to the most above-steady-state deciles; economies furthest above predicted trajectories subsequently experience the weakest growth.

- Fragility, adjustment, and downside risk (discrete-time hazard models, Table IV):
  - Discrete-time complementary-log-log hazard specifications link ω̂_it^+ to incidence of stagnation, banking crises, sovereign-debt crises, and loss of frontier.
  - Selected hazard coefficients (unconditional specifications) and standard errors:
    - Stagnation (ω̂^+ alone): +0.8750 (0.3089); unconditional event rate 0.225; at-risk observations 1117.
    - Banking crisis (ω̂^+ alone): +1.5860 (0.3296); unconditional event rate 0.129; at-risk observations 1110.
    - Sovereign-debt crisis (ω̂^+ alone): +0.7417 (0.5635); unconditional event rate 0.091; at-risk observations 1132.
    - Loss of frontier (ω̂^+ alone): +1.9660 (0.3673); unconditional event rate 0.345; at-risk observations 1532.
  - Robustness to controls:
    - Relationships for banking crises and loss-of-frontier remain strong after adding initial income and recent growth momentum.
    - Stagnation association weakens with controls; sovereign-debt crisis relationship not robust across specifications.
  - Interpretation: positive overshooting is associated with greater downside risk, particularly banking-sector distress and loss of frontier position.

- Robustness and interpretation notes:
  - Addressed concerns about mechanical mean reversion via controls for initial income, non-overlapping five-year periods, leave-one-observation-out trajectory construction, and bootstrap/permutation exercises (reported in Appendix B).
  - Main findings robust to excluding major oil exporters, transition economies, small financial centers, and periods with systemic banking crises or large external shocks.
  - Caveats:
    - Overshooting may partially reflect omitted fundamentals or temporary cyclical booms rather than persistent deterioration.
    - Trajectory is a generated regressor; bootstrap and alternative specifications mitigate but do not eliminate inference concerns.

### Interpretation, policy implications, and conceptual contribution (Sections V–VI)
- Conceptual reframing:
  - Observed convergence is ambiguous: similar convergence coefficients may reflect upward catch-up, compression from above, or both.
  - Narrowing dispersion (sigma convergence) alone is insufficient to identify whether development progress is driven by rising productive capacity among poorer countries or weakening frontier dynamism.
- Empirical findings summarized:
  - Economies above their predicted long-run trajectories subsequently grow more slowly.
  - Adjustment from above tends to be slower and more persistent than adjustment from below; economies more often experience prolonged weaker relative growth rather than abrupt collapse.
  - Compression from above contributes materially to observed convergence; in the full 1960–2019 sample it accounts for roughly one-third of the systematic convergence component (varying over time).
- Policy-relevant implications:
  - Diagnose weak growth by distinguishing cyclical slack, temporary shocks, and deterioration in fundamentals supporting long-run income—policy responses differ:
    - Temporary demand-side or cyclical interventions appropriate for recoverable slack.
    - Structural reforms, productivity-enhancing policies, and measures to address demographic or diffusion constraints required when long-run trajectories have shifted downward.
  - Monitor indicators of overshooting (ω̂_it) alongside standard convergence metrics to detect heightened downside risk, especially for frontier economies exhibiting positive overshooting and signs of banking-sector fragility.
  - For forecasting and debt-sustainability analysis, account for the possibility that long-run potential estimates may be revised downward over time, implying slower future growth even without absolute income declines.
- Final interpretive claim:
  - Convergence should be read as movement within an evolving world distribution rather than movement toward fixed equilibria; a complete account of global development must encompass both upward catch-up and potential downward adjustment from above.

*IMF Working Paper — Convergence from Above (excerpted content units: Sections II–VI; empirical results summarized as presented)*

### References

### References

### Key references
- Abramovitz, Moses. 1986. “Catching Up, Forging Ahead, and Falling Behind.” Journal of Economic History 46 (2): 385–406.
- Barro, Robert J., and Jong-Wha Lee. 2013. “A New Data Set of Educational Attainment in the World, 1950–2010.” Journal of Development Economics 104: 184–98.
- Barro, Robert J., and Xavier Sala-i-Martin. 1992. “Convergence.” Journal of Political Economy 100 (2): 223–51.
- Baumol, William J. 1986. “Productivity Growth, Convergence, and Welfare: What the Long-Run Data Show.” American Economic Review 76 (5): 1072–85.
- Bernard, Andrew B., and Steven N. Durlauf. 1996. “Interpreting Tests of the Convergence Hypothesis.” Journal of Econometrics 71 (1–2): 161–73.
- Bolt, Jutta, and Jan Luiten van Zanden. 2020. “Maddison Style Estimates of the Evolution of the World Economy: A New 2020 Update.” Maddison Project Working Paper 15. Groningen: Groningen Growth and Development Centre, University of Groningen.
- Durlauf, Steven N., and Paul A. Johnson. 1995. “Multiple Regimes and Cross-Country Growth Behaviour.” Journal of Applied Econometrics 10 (4): 365–84.
- Durlauf, Steven N., Paul A. Johnson, and Jonathan R. W. Temple. 2005. “Growth Econometrics.” In Handbook of Economic Growth, vol. 1A, edited by Philippe Aghion and Steven N. Durlauf, 555–677. Amsterdam: Elsevier.
- Feenstra, Robert C., Robert Inklaar, and Marcel P. Timmer. 2015. “The Next Generation of the Penn World Table.” American Economic Review 105 (10): 3150–82.
- Fernald, John, Robert Inklaar, and Dimitrije Ruzic. 2025. “The Productivity Slowdown in Advanced Economies: Common Shocks or Common Trends?” Review of Income and Wealth 71 (1): e12690.
- Galor, Oded. 2011. Unified Growth Theory. Princeton, NJ: Princeton University Press.
- Gordon, Robert J. 2016. The Rise and Fall of American Growth: The U.S. Standard of Living since the Civil War. Princeton, NJ: Princeton University Press.
- Haldane, Andrew G. 2017. “Productivity Puzzles.” Speech given at the London School of Economics, March 20. Bank of England.
- Hayashi, Fumio, and Edward C. Prescott. 2002. “The 1990s in Japan: A Lost Decade.” Review of Economic Dynamics 5 (1): 206–35.
- Imam, Patrick A., and Jonathan R. W. Temple. 2025. “Dynamic Development Accounting and Relative Income Traps.” Economic Inquiry 64 (2): 591–613.
- Imam, Patrick A., and Jonathan R. W. Temple. 2026a. “At the Threshold: The Increasing Relevance of the Middle-Income Trap”. Scandinavian Journal of Economics https://doi.org/10.1111/sjoe.70036
- Imam, Patrick A., and Jonathan R. W. Temple. 2026b. “Growth, Interrupted: How Crises Delay Global Convergence.” Journal of International Money and Finance 163: 103690
- Islam, Nazrul. 1995. “Growth Empirics: A Panel Data Approach.” Quarterly Journal of Economics 110 (4): 1127–70.
- Johnson, Paul A., and Chris Papageorgiou. 2020. “What Remains of Cross-Country Convergence?” Journal of Economic Literature 58 (1): 129–75.
- Johnson, Paul A., Chris Papageorgiou, Maria Grazia Pittau, and Roberto Zelli. 2024. “Lessons from 40 Years of Cross-Country Convergence Empirics.” Prepared for the Oxford Handbook of Income Distribution and Economic Growth.
- Jones, Charles I. 2002. “Sources of U.S. Economic Growth in a World of Ideas.” American Economic Review 92 (1): 220–39.
- Jones, Charles I., and Paul M. Romer. 2010. “The New Kaldor Facts: Ideas, Institutions, Population, and Human Capital.” American Economic Journal: Macroeconomics 2 (1): 224–45.
- Kremer, Michael, Jack Willis, and Yang You. 2021. “Converging to Convergence.” NBER Macroeconomics Annual 36: 337–412.
- Mankiw, N. Gregory, David Romer, and David N. Weil. 1992. “A Contribution to the Empirics of Economic Growth.” Quarterly Journal of Economics 107 (2): 407–37.
- Mathunjwa, Jochonia S., and Jonathan Temple. 2006. “Convergence Behaviour in Exogenous Growth Models.” Bristol Economics Discussion Papers 06/590. Bristol, UK: School of Economics, University of Bristol.
- Müller, Karsten, Chenzi Xu, Mohamed Lehbib, and Ziliang Chen. 2025. “The Global Macro Database: A New International Macroeconomic Dataset.” NBER Working Paper No. 33714. Cambridge, MA: National Bureau of Economic Research.
- Patel, Dev, Justin Sandefur, and Arvind Subramanian. 2021. “The New Era of Unconditional Convergence.” Journal of Development Economics 152: 102687.
- Pritchett, Lant. 1997. “Divergence, Big Time.” Journal of Economic Perspectives 11 (3): 3–17.
- Quah, Danny T. 1993. “Galton’s Fallacy and Tests of the Convergence Hypothesis.” Scandinavian Journal of Economics 95 (4): 427–43.
- Quah, Danny T. 1996. “Twin Peaks: Growth and Convergence in Models of Distribution Dynamics.” Economic Journal 106 (437): 1045–55.
- Rodríguez, Francisco, and Jeffrey D. Sachs. 1999. “Why Do Resource-Abundant Economies Grow More Slowly?” Journal of Economic Growth 4 (3): 277–303.
- Rodrik, Dani. 2016. “Premature Deindustrialization.” Journal of Economic Growth 21 (1): 1–33.
- Solow, Robert M. 1956. “A Contribution to the Theory of Economic Growth.” Quarterly Journal of Economics 70 (1): 65–94.
- Summers, Lawrence H. 2014. “U.S. Economic Prospects: Secular Stagnation, Hysteresis, and the Zero Lower Bound.” Business Economics 49 (2): 65–73.

### Appendix A. Dynamic Convergence with Moving Steady States — main framework
- Purpose: extend the standard neoclassical growth model to allow steady states to evolve over time and introduce a distinction between convergence from below and convergence from above.
- Core production function: Y_it = K_it^α (A_it L_it)^(1−α), with 0<α<1.
- Definitions:
  - k_it ≡ K_it / (A_it L_it) (capital per effective worker).
  - ỹ_it ≡ Y_it / (A_it L_it) = k_it^α (output per effective worker).
  - y_it ≡ Y_it / L_it = A_it k_it^α (output per worker).
- Capital accumulation: K̇_it = s_it Y_it − δ K_it, where s_it is the investment rate and δ is the depreciation rate.
- Labor and efficiency growth: L̇_it / L_it = n_it, Ȧ_it / A_it = g_it.
- Law of motion for capital per effective worker:
  - k̇_it = s_it k_it^α − (n_it + g_it + δ) k_it.
  - Define φ_it ≡ n_it + g_it + δ, so k̇_it = s_it k_it^α − φ_it k_it.
- Steady-state capital per effective worker:
  - s_it (k_it^*)^α = φ_it k_it^* ⇒ k_it^* = (s_it / φ_it)^{1/(1−α)}.
- Balanced-growth output per worker:
  - y_it^* = A_it (k_it^*)^α = A_it (s_it / φ_it)^{α/(1−α)}.
  - Log form: ln y_it^* = ln A_it + (α/(1−α)) ln s_it − (α/(1−α)) ln φ_it.

### Moving steady states and overshooting
- Time variation in determinants (A_it, s_it, φ_it) allows the steady-state path to change.
- Differential of steady-state log output:
  - d ln y_it^* = d ln A_it + (α/(1−α)) d ln s_it − (α/(1−α)) d ln φ_it.
- Condition for steady-state path to decline:
  - d ln A_it + (α/(1−α)) d ln s_it < (α/(1−α)) d ln φ_it.
- Define overshooting deviation:
  - ω_it ≡ ln y_it − ln y_it^*.
  - ω_it < 0 ⇒ below steady state; ω_it > 0 ⇒ above steady state.
- Key implication: an economy can enter the above-steady-state region without actual income growth if d ln y_it = 0 and d ln y_it^* < 0 ⇒ d ω_it > 0. Overshooting may therefore arise from downward revisions to long-run potential rather than exceptional booms.

### Adjustment dynamics and asymmetry
- From the law of motion:
  - k̇_it / k_it = s_it k_it^{α−1} − φ_it.
- Using steady-state condition s_it (k_it^*)^{α−1} = φ_it, define z_it ≡ k_it / k_it^*.
- Exact adjustment dynamics:
  - k̇_it / k_it = φ_it (z_it^{α−1} − 1).
- Interpretation:
  - If z_it < 1 then z_it^{α−1} > 1 and capital per effective worker grows (adjustment from below).
  - If z_it > 1 then z_it^{α−1} < 1 and capital per effective worker falls relative to steady state (adjustment from above).
- Asymmetry example for equal proportional deviations:
  - Let z_it = e^x above and z_it = e^{−x} below, with x>0.
  - Adjustment from below: (k̇_it / k_it)|_{z_it=e^{−x}} = φ_it (e^{(1−α) x} − 1).
  - Absolute magnitude of adjustment from above: −(k̇_it / k_it)|_{z_it=e^{x}} = φ_it (1 − e^{−(1−α) x}).
  - Since e^{(1−α) x} − 1 > 1 − e^{−(1−α) x}, adjustment from below is stronger than adjustment from above.
- Implication: convergence from above likely appears empirically as prolonged stagnation or persistent underperformance rather than abrupt collapse.

### Convergence regressions and decomposition
- Standard convergence regression:
  - Δ ln y_{i,t+T} = a + b ln y_it + ε_it.
  - Negative b conventionally interpreted as catch-up by poorer economies.
- Decomposition using overshooting parts:
  - Define ω_it^− = min(ω_it, 0) and ω_it^+ = max(ω_it, 0).
  - Growth process representation:
    - Δ ln y_{i,t+T} = β_1 ω_it^− + β_2 ω_it^+ + ε_it, with β_1 < 0, β_2 < 0, |β_1| > |β_2|.
  - Observed convergence coefficient b combines two mechanisms:
    - b = b_below + b_above.
- Proposition: When steady-state paths evolve over time, standard convergence regressions confound:
  1. upward adjustment toward the steady state from below; and
  2. downward adjustment toward the steady state from above.
- Policy-relevant implication: Observed convergence is not sufficient evidence of broad-based catch-up or development progress when steady-state paths are moving.

### Distributional implications and empirical predictions
- States and transitions:
  - Countries occupy relative-income states S_1,...,S_J.
  - Transition probabilities depend on overshooting gap: P_jk(ω_it) = Pr(S_{i,t+T}=k | S_it=j, ω_it).
  - Countries below steady-state paths should have higher upward mobility probabilities; countries above should face greater stagnation or downward transition probabilities.
- Observed convergence in cross-country distribution may arise from:
  1. upward mobility from below; and
  2. compression at the top through adjustment from above.
- Four empirical predictions:
  1. Declines in predicted steady-state paths should increase the probability that economies enter the above-steady-state region.
  2. Countries above their steady-state paths should subsequently experience persistently weaker growth.
  3. Under the asymmetry conditions, adjustment from above may be slower than adjustment from below.
  4. Observed convergence in the world income distribution may reflect both upward mobility from below and downward adjustment from above.

*Source: References and Appendix A from the IMF Working Paper "Convergence from Above" (content unit: wpiea2026159-source-pdf - References).*

### Appendix B: Supplementary Empirical Results

### Appendix B: Supplementary Empirical Results and Robustness Analysis

### Summary statistics (Table A1)
- Sample: Pooled five-year panel, 1960–2019.
- Observations and descriptive moments:
  - Log output per worker (initial): N 1805; Mean 9.957; Std 1.147; P10 8.282; Median 10.086; P90 11.361
  - Annualized five-year log growth in output per worker: N 1805; Mean 0.065; Std 0.246; P10 -0.168; Median 0.079; P90 0.295
  - Forward 5-yr growth: output per worker: N 1621; Mean 0.027; Std 0.107; P10 -0.055; Median 0.027; P90 0.123
  - Annualized ten-year forward growth in output per worker: N 1621; Mean 0.160; Std 0.386; P10 -0.238; Median 0.177; P90 0.562
  - Omega (MRW gap = residual from SS regression): N 1532; Mean 0.000; Std 0.334; P10 -0.387; Median 0.004; P90 0.378
  - Omega+ (above the fitted trajectory component): N 1532; Mean 0.123; Std 0.196; P10 0.000; Median 0.004; P90 0.378
  - Omega- (below the fitted trajectory): N 1532; Mean -0.123; Std 0.208; P10 -0.387; Median 0.000; P90 0.000
  - Above fitted-trajectory indicator (ω̂ > 0): N 2220; Mean 0.349; Std 0.477; P10 0.000; Median 0.000; P90 1.000
  - Investment share (initial, period avg): N 1962; Mean 0.214; Std 0.114; P10 0.080; Median 0.201; P90 0.355
  - Population growth rate (initial): N 1960; Mean 0.018; Std 0.014; P10 0.001; Median 0.019; P90 0.034
  - Human capital index (initial): N 1601; Mean 2.076; Std 0.725; P10 1.162; Median 1.982; P90 3.118
  - Relative income: output/wkr vs US: N 1805; Mean 0.394; Std 0.551; P10 0.044; Median 0.256; P90 0.834
  - Log gap to US frontier: N 1805; Mean -1.495; Std 1.112; P10 -3.135; Median -1.363; P90 -0.182
  - Above US indicator: N 2220; Mean 0.039; Std 0.193; P10 0.000; Median 0.000; P90 0.000
- Note: Derived gap indicators and positive- and negative-gap components are left missing whenever the underlying trajectory gap is unavailable.

### Country-level coverage and trajectory-gap statistics (Table A2)
- Coverage based on country-(regional) period observations with a nonmissing fitted trajectory gap; first and last periods refer to usable observations.
- Examples of country-level statistics (selected lines preserved exactly as in source):
  - ABW Aruba: First period 1960; Last period 2015; Periods (N) 12; Periods ω>0 0; Ever ω≥0 no
  - ARE United Arab Emirates: First period 1960; Last period 2015; Periods (N) 12; S ω  -0.000; ω S 0; Periods ω>0 0.849 7; Ever ω≥0 yes
  - CHN China: First period 1960; Last period 2015; Periods (N) 12; S ω  +0.000 0; ω S 0.721 9; Periods ω>0 5; Ever ω≥0 yes
  - USA United States: First period 1960; Last period 2015; Periods (N) 12; S ω  +0.00 00; ω S 0.087 4; Periods ω>0 5; Ever ω≥0 no
  - ZWE Zimbabwe: First period 1960; Last period 2015; Periods (N) 12; S ω  +0.00 00; ω S 0.826 7; Periods ω>0 9; Ever ω≥0 yes
- Note: Countries with no valid trajectory-gap observations are excluded.

### Variable definitions and data sources (Table A3)
- Outcome variables:
  - growth_5yr_output_pw: Annualized growth in output per worker over the 5-year period. Source: PWT 11.0 (rgdpo / emp)
  - growth_fwd_output_pw: Annualized five-year log growth in output per worker, calculated as [ln(yᵢ,ₜ₊₅) − ln(yᵢₜ)]/5. Source: PWT 11.0
  - growth_10yr_output_pw: Annualized growth in output per worker over the next 10 years. Source: PWT 11.0
- MRW inputs and controls:
  - avg_inv_share: Mean investment-to-GDP ratio over period (csh_i). Source: PWT 11.0
  - avg_pop_growth: Mean population growth rate over period. Source: PWT 11.0 (pop)
  - avg_hc: Mean Penn World Table human-capital index over the period. Source: PWT 11.0 (hc)
  - init_ln_output_pw: Initial-period log output per worker. Source: PWT 11.0
- Omega measures (constructed):
  - omega: ω = ln(y_observed / y_predicted_MRW); deviation from MRW steady-state.
  - omega_positive: max(ω, 0): above-steady-state component.
  - omega_negative: min(ω, 0): below-steady-state component.
  - above_ss: Indicator: 1 if ω > 0.
- Frontier measures:
  - rel_income_us: Output per worker relative to US contemporaneous level. Source: PWT 11.0
  - ln_rel_income_us: Log of rel_income_us. Source: PWT 11.0
  - rel_income_top5: Output per worker relative to mean of top-5-OECD frontier. Source: PWT 11.0
  - ln_rel_income_top5: Log of rel_income_top5. Source: PWT 11.0
- Channels and vulnerability indicators:
  - g_rtfpna: Annualized TFP growth over period. Source: PWT 11.0 (rtfpna)
  - g_rnna: Annualized capital-stock growth over period. Source: PWT 11.0 (rnna)
  - init_labsh: Initial period labor share of GDP. Source: PWT 11.0 (labsh)
  - govdebt_gdp: Government debt-to-GDP ratio. Source: Global Macro Database
  - CA_GDP: Current account balance as share of GDP. Source: Global Macro Database
  - REER: Real effective exchange rate (index). Source: Global Macro Database
  - M2: M2 money stock (used for credit-boom proxy via Δ log M2). Source: Global Macro Database
  - infl: Annual CPI inflation rate. Source: Global Macro Database
  - banking_crisis, currency_crisis, sovdebt_crisis: Crisis onset indicators. Source: GMD (Laeven-Valencia) and GMD.

### First-stage trajectory regression (Table A4)
- Dependent variable: init_ln_output_pw.
- Specifications and key output (selected coefficients and metrics preserved exactly):
  - Columns reported: (1) MRW+HC g+d=0.05; (2) MRW w/o HC; (3) MRW+HC g+d=0.04; (4) MRW+HC g+d=0.06; (5) MRW init-yr controls.
  - Coefficient ln_s: 0.0580 (column 1); 0.0580 (2); 0.0583 (3); 0.0577 (4).
  - Coefficient ln_ngd: 0.0763 (1); 0.0813 (2).
  - Coefficient ln_hc: 0.5597 (1); 0.5552 (3); 0.5633 (4).
  - Coefficient ln_s_init (col 5): 0.0930* (SE 0.0534).
  - Observations: 1532 (cols 1,3,4,5); 1804 (col 2).
  - Countries: 145 (cols 1,3,4,5); 184 (col 2).
  - R-squared: 0.016 (cols 1,3,4); 0.004 (col 2); 0.025 (col 5).
  - Country FE: Yes; Time FE: Yes.

### Accounting decomposition of convergence (Table A5)
- Decomposition separates changes in cross-country income dispersion into:
  - Compression from above, Catch-up from below, Idiosyncratic residual.
- Full sample (1960-2019): Sub-periods (k) 11; Σ Compression from above -0.0201; Σ Catch-up from below +0.4600; Σ Idiosyncratic -2.3246; Net Δσ -1.8846; Net convergence? yes; Share from compression-above 33.7%; Share from catch-up-below 66.3%
- Pre-1990 (1960-1985): k 6; Σ Compression from above +0.1134; Σ Catch-up from below +0.2488; Σ Idiosyncratic -0.9567; Net Δσ -0.5946; Net convergence? yes; Share from compression-above 49.5%; Share from catch-up-below 50.5%
- Post-1990 (1990-2019): k 5; Σ Compression from above -0.1335; Σ Catch-up from below +0.2113; Σ Idiosyncratic -1.3678; Net Δσ -1.2900; Net convergence? yes; Share from compression-above 14.5%; Share from catch-up-below 85.5%
- Decadal examples:
  - 1990s (1990-1995): k 2; Σ Compression from above -0.0560; Σ Catch-up from below -0.0603; Σ Idiosyncratic -0.3653; Net Δσ -0.4816; Net convergence? yes; Share from compression-above 7.4%; Share from catch-up-below 92.6%
  - 2000s (2000-2005): k 2; Σ Compression from above -0.1247; Σ Catch-up from below -0.0787; Σ Idiosyncratic -0.5953; Net Δσ -0.7987; Net convergence? yes; Share from compression-above 100.0%; Share from catch-up-below 0.0%
  - 2010s (2010-2019): k 1; Σ Compression from above +0.0473; Σ Catch-up from below +0.3503; Σ Idiosyncratic -0.4073; Net Δσ -0.0097; Net convergence? yes; Share from compression-above 11.9%; Share from catch-up-below 88.1%
- Note: Shares computed over the two systematic convergence components; idiosyncratic residual excluded from share calculation.

### Crisis-episode counts and at-risk sample (Table A6)
- At-risk country-periods per decade: 370 each decade (1960s–2010s).
- Crisis onsets and stagnation episodes by decade:
  - 1960s: Banking crises onsets 1; Currency crises onsets 40; Sovereign-debt crises onsets 13; Stagnation episodes (g<0) 21
  - 1970s: Banking 6; Currency 26; Sovereign-debt 7; Stagnation 57
  - 1980s: Banking 49; Currency 73; Sovereign-debt 47; Stagnation 126
  - 1990s: Banking 86; Currency 88; Sovereign-debt 46; Stagnation 141
  - 2000s: Banking 33; Currency 28; Sovereign-debt 14; Stagnation 95
  - 2010s: Banking 6; Currency 28; Sovereign-debt 11; Stagnation 83
- Note: Counts refer to event onsets; country-periods already within the relevant event are excluded from the event-specific risk set.

### Dynamic response profile around initial trajectory gap (Table A7)
- Descriptive β coefficients for positive (β⁺) and negative (β⁻) initial gaps at horizons h (years):
  - h = -3: β⁺ +0.8450*** (SE 0.0374); β⁻ +0.7039*** (SE 0.0527); N 7417
  - h = -2: β⁺ +0.6168*** (0.0209); β⁻ +0.5439*** (0.0310); N 7495
  - h = -1: β⁺ +0.3470*** (0.0168); β⁻ +0.3248*** (0.0129); N 7573
  - h = 1: β⁺ -0.0998*** (0.0259); β⁻ -0.1037*** (0.0183); N 7648
  - h = 2: β⁺ -0.2030*** (0.0505); β⁻ -0.1875*** (0.0303); N 7648
  - h = 3: β⁺ -0.2810*** (0.0723); β⁻ -0.2641*** (0.0470); N 7648
  - h = 5: β⁺ -0.3670*** (0.0884); β⁻ -0.3244*** (0.0673); N 7503
  - h = 7: β⁺ -0.4079*** (0.0796); β⁻ -0.3466*** (0.0877); N 7213
  - h = 10: β⁺ -0.3783*** (0.1210); β⁻ -0.3041*** (0.1032); N 6778
  - h = 15: β⁺ -0.2742** (0.1084); β⁻ -0.2060** (0.1036); N 6053
- Note: Kraay standard errors reported.

### Whole-pipeline country-bootstrap inference (Table A8)
- Procedure: Complete countries resampled with replacement; first-stage and second-stage re-estimated within each replication.
- Estimates:
  - β(ω < 0) catch-up slope: Point estimate -0.0760; Bootstrap SE 0.0201; 95% CI lo -0.1180; 95% CI hi -0.0393; p (boot, two-sided) 0.0000; N 1387
  - β(ω ≥ 0) adjustment-from-above slope: Point estimate -0.0419; Bootstrap SE 0.0187; 95% CI lo -0.0772; 95% CI hi -0.0071; p (boot, two-sided) 0.0133; N 1387
  - Asymmetry: β(ω≥0) − β(ω<0): +0.0340; Bootstrap SE 0.0304; 95% CI lo -0.0226; 95% CI hi +0.0914; p 0.2667; N 1387

### Robustness checks (Tables A9–A11)
- Quantile-conditional robustness (Table A9): Q τ ∈ {0.25, 0.50, 0.75} of the conditional forward-growth distribution; coefficients for ω⁻ and ω⁺:
  - τ = 0.25: β ω<0 -0.0429 (SE 0.0111); β ω≥0 -0.0638 (SE 0.0116); Δβ -0.0209; N 1387
  - τ = 0.50: β ω<0 -0.0643 (SE 0.0101); β ω≥0 -0.0103 (SE 0.0106); Δβ +0.0539; N 1387
  - τ = 0.75: β ω<0 -0.1069 (SE 0.0127); β ω≥0 +0.0129 (SE 0.0135); Δβ +0.1198; N 1387
- Romano–Wolf multiple-testing adjustment (Table A10) for preferred downside-risk specifications:
  - Stagnation (g<0 next 5y): Naive p 0.0000; Romano-Wolf adjusted p 0.0000; Survives WE yes; Survives FWE yes
  - Large decline (g<−5% next 5y): Naive p 0.0000; Romano-Wolf adjusted p 0.0000; Survives WE yes; Survives FWE yes
  - Decline next 10y: Naive p 0.0000; Romano-Wolf adjusted p 0.0000; Survives WE yes; Survives FWE yes
  - Loss of frontier proximity: Naive p 0.0000; Romano-Wolf adjusted p 0.0000; Survives WE yes; Survives FWE yes
  - Banking crisis next 5y: Naive p 0.0001; Romano-Wolf adjusted p 0.0000; Survives WE yes; Survives FWE yes
  - Currency crisis next 5y: Naive p 0.2502; Romano-Wolf adjusted p 0.4375; Survives WE no; Survives FWE no
  - Sovereign-debt crisis next 5y: Naive p 0.2170; Romano-Wolf adjusted p 0.4375; Survives WE no; Survives FWE no
  - Note: Step-down correction with 1000 bootstrap replications.
- Within-sample permutation inference (Table A11) for association between positive trajectory gap (ω⁺) and adverse outcomes:
  - Stagnation: Observed β ω⁺ +0.4319; Permutation mean -0.1877; Permutation SD 0.3517; Permutation p (one-sided) 0.0300; Permutation 95th pct +0.3900; N_obs 1117; B (valid perms) 500
  - Banking crisis: Observed β ω⁺ +1.3807; Permutation mean +0.1666; Permutation SD 0.4321; Permutation p (one-sided) 0.0020; Permutation 95th pct +0.8111; N_obs 1110; B 500
  - Sovereign-debt crisis: Observed β ω⁺ +0.8522; Permutation mean +0.1144; Permutation SD 0.4914; Permutation p (one-sided) 0.0520; Permutation 95th pct +0.8576; N_obs 1132; B 500
  - Loss of frontier: Observed β ω⁺ +1.3205; Permutation mean +0.0128; Permutation SD 0.3398; Permutation p (one-sided) 0.0000; Permutation 95th pct +0.5606; N_obs 1532; B 500
  - Note: Randomization inference under the null that the positive trajectory gap has no association with the specified subsequent adverse outcome.

*Appendix B: Supplementary Empirical Results and Robustness Analysis, Convergence from Above (Working Paper No. WP/2026/159).*

---


_Source: https://www.imf.org/-/media/files/publications/wp/2026/english/wpiea2026159-source-pdf.pdf_