91 countries · 225 years · 1,656 observations · 4-phase independent audit · 2026-02-08
Mean reversion k ≈ 0 at every stage. AR(1) equilibrium at L*=81.6 suggests three attractor basins (tyranny, hybrid trap, democratic plateau), not the original bistable two-basin model. System is tristable. Bimodal distribution is a persistence artifact.
Three methods converge on L≈52-55 as the Critical Instability Zone. Recovery rate below this threshold: 3.0% (95% CI: 0.7-6.0%). The original 12% figure approximately matches L=50 in 1995-2025 only.
Mean across 7 indices: 76.6. Credible range: 57-84. V-Dem reclassification supports decline direction. TCF Democracy Meter at 57 is closest independent measurement.
Confirmed in 2-year window (2023-2025). Standardized: -4.2/yr at 10yr. At 10yr window with credible L values: -2 to -4/yr. Still the fastest-declining consolidated democracy even at moderate estimate.
Data-driven MC: P(tyranny) ≈ 0%. Thesis σ values were stipulated without empirical basis, inflated 2-7x. AR(1) mean reversion dominates. The original 62% is a phantom created by wrong parameters. However, post-2006 structural break dynamics yield P(L<50|15yr)=69%.
Slope and R² reproduce exactly on the n=32 core sample. Intercept corrected to 33.05 (was 18.7). Log-linear specification fits better (R²=0.51). The relationship is real and economically meaningful. Note: on the expanded n=64 sample, Liberty-only R²=0.209; with GDP controls, the Liberty coefficient attenuates to ~4.6bp/point. The 35bp figure should be treated as an upper bound.
Reserve currency premium of 2,080bp explains most of the US yield anomaly. Model with reserve: 3.8% predicted vs 4.5% actual. Defensible claim: 200-580bp over 5-10 years.
r dropped from 0.79 to 0.57. 39 capable autocracies identified. Capability ≠ freedom is empirically supported. One of the thesis's strongest contributions.
Original +21.4% confirmed as overall average, but misleading. Structural break: +40% pre-1971, pre-2006: +38%, post-2006: -23.3%. Historical averages mask a regime change in Stage 5 dynamics.
Path dependence significant at Stages 2, 5, 6. Stage 6: declining arrivals -77.8% vs improving +25.5%. Direction of travel matters more than current state.
Data: σ = 0.45-4.45. Thesis overstates by 2-7x at every stage. Stage 1: thesis 3, actual 0.45 (6.7x). All distributions non-normal with heavy tails.
Persistence baseline achieves 73%. Stage model adds +5pp at best. Out-of-sample backtesting (3 windows) confirms persistence beats all models.
A 1 SD improvement in rule of law predicts 3× (OLS) to 8× (IV) income gains. Seven natural experiments (Georgia, Rwanda, Estonia, Botswana, Chile, South Korea, Singapore) show ROI timelines from 1–2 years (anti-corruption → tax revenue) to 25+ years (democratisation → sustained GDP). Global corruption costs $2.6T/yr (5% of GDP); $9.3T in dead capital locked by weak property rights. Justice correlates with HCI at r = 0.84. See A11: The Justice Dividend.
How thesis conclusions change across the credible range of US Liberty scores:
| Metric | L=48 (Thesis) | L=57 (TCF) | L=65 (V-Dem low) | L=70 (V-Dem mid) | L=75 (Cross-index) | L=84 (FH official) |
|---|---|---|---|---|---|---|
| Stage | S6: Soft Dictatorship | S5: Electoral Autocracy | S4: Competitive Auth. | S3: Democratic Erosion | S3: Democratic Erosion | S2: Early Warning |
| Velocity | -18.0/yr (2yr) | -3.7/yr (10yr) | -2.9/yr (10yr) | -2.4/yr (10yr) | -1.9/yr (10yr) | -1.0/yr (10yr) |
| Event Horizon | BELOW | BELOW | Above | Above | Above | Above |
| Hist. reversal to L≥70 | 54% | 71% | 86% | 82% | 91% | 99% |
| Predicted yield | 16.2% | 13.1% | 10.3% | 8.5% | 6.8% | 3.6% |
| P(L<25 by 2040) | 0.0% (data σ) | 0.0% | 0.0% | 0.0% | 0.0% | 0.0% |
| Narrative | Critical instability zone | Crisis zone | Serious erosion | Declining democracy | Declining democracy | Stressed but intact |
| Task | Finding | Severity |
|---|---|---|
| 1.1 Crosswalk | 67% match (18/27). Large deviations: South Africa (-14), Turkey (-10), Venezuela (-9) | CRITICAL |
| 1.2 Event Horizon | Threshold at L≈52-55 (Critical Instability Zone). Recovery: 3.0% (95% CI: 0.7-6.0%). | CRITICAL |
| 1.3 Stage 5 | pre-2006: +38%, post-2006: -23.3%. Overall +21.4% confirmed but misleading. Global break ~2000 (F=21.2); stage-specific 2006. | CRITICAL |
| 1.4 Velocity | -18/yr confirmed (2yr window; standardized: -4.2/yr at 10yr). 15yr: -3.07. 0th percentile | CRITICAL |
| 1.5 Holdout | Persistence baseline 73%. Thesis 78%. Marginal +5pp skill | HIGH |
| Task | Finding | Severity |
|---|---|---|
| 2.1 Shock σ | ALL thesis σ stipulated without empirical basis, wrong by 2-7x. Stage 1: 0.45 (thesis 3). Stage 5: 2.45 (thesis 7) | CRITICAL |
| 2.2 Markov | REJECTED at Stages 2, 5, 6. Stage 6: -78% declining vs +26% improving | CRITICAL |
| 2.3 Regression | β=-0.35 and R²=0.37 confirmed (n=32 core sample). Intercept corrected: 33.05 (was 18.7). On expanded n=64 with GDP controls: β attenuates to ~4.6bp/point | HIGH |
| 2.4 AIC/BIC | AR(1) beats ALL stage models. ΔAIC > 300. R²=0.872 | CRITICAL |
| 2.5 Mean reversion | k ≈ 0 everywhere. "Two deep wells" not detectable | CRITICAL |
| Task | Finding | Severity |
|---|---|---|
| 3.1 Cross-validation | Mean across 7 indices: 76.6. L=48 is 22-36pts below every measure | FATAL |
| 3.2 Institutions | 56/100. Military (80), Fed (65) strong. Congress (32) compromised | HIGH |
| 3.3 Matched comparison | 71% of countries at L=45-55 recovered. Best US analogue: France 1958 | HIGH |
| 3.4 Elections | Base rate ~7% reversal in electoral autocracies. US has structural advantages | MEDIUM |
| 3.5 Reserve currency | Reserve status explains most of US yield anomaly. Model: 3.8% vs actual 4.5% | HIGH |
| Task | Finding | Severity |
|---|---|---|
| 4.1 Recalibration | V-Dem reclassified US as electoral autocracy Sep 2025. TCF: 57. Range: 57-84 | CRITICAL |
| 4.2 Monte Carlo | Data-driven σ → P(tyranny) ≈ 0%. Thesis σ inflates by 185x | CRITICAL |
| 4.3 Backtesting | Persistence beats all models in 3/3 windows. US 84→48 = only 30+ pt miss | HIGH |
| 4.4 Counter-args | 3/7 strong (CA5, CA6, CA7). Thesis 2-3x too aggressive | HIGH |
| # | Counter-Argument | Strength | Disposition |
|---|---|---|---|
| CA1 | US institutions uniquely resilient | Medium-Strong | Acknowledged with caveats. Military (80/100) and Fed (65/100) are genuine stabilizers the thesis ignores. |
| CA2 | 2024 election was free and fair | Medium | Reframed. Correct for current state, but thesis is about trajectory. The question is 2028/2032. |
| CA3 | Freedom House Western bias | Weak | Dismissed. If FH has pro-Western bias, it overstates the US score. All 7 indices confirm decline direction. |
| CA4 | Markets are efficient | Medium | Acknowledged with lag model. Historical yield lags: 3-12 years. Current pricing tells us nothing about 2030. |
| CA5 | Confuses regime type with policy | Strong | Must operationalize taxonomy: policy reversals vs. institutional degradation vs. democratic infrastructure damage. |
| CA6 | Mean reversion in 248-yr democracies | Strong | 98% of decline episodes from L≥80 in long-standing democracies recovered. Must model democratic tenure. |
| CA7 | Economic growth stabilizes democracy | Medium-Strong | No democracy above $15K GDP/capita has collapsed. US at $85K is historically unprecedented territory. |
| # | Gap | Severity | Status |
|---|---|---|---|
| G1 | US FH 48 is author's estimate | CRITICAL | Partially Resolved Credible range 57-84 |
| G2 | -18/yr velocity over 2 years | CRITICAL | Resolved Bootstrap CIs; 10yr: -2 to -4/yr |
| G3 | Stage 5 contradiction | HIGH | Resolved Structural break identified |
| G4 | Event Horizon revised to L≈52-55 | HIGH | Revised 3.0% recovery (95% CI: 0.7-6.0%); three methods converge |
| G5 | Shock σ stipulated (without empirical basis) | HIGH | Resolved Data-driven σ estimated |
| G6 | Ternary constraint axiomatic | MODERATE | Open |
| G7 | Markov assumption | MODERATE | Rejected Path dependence confirmed |
| G8 | No country >100M comparison | MODERATE | Open |
| G9 | Selection bias in comparators | MODERATE | Resolved Matched comparison built |
| G10 | Cross-sectional regression | MODERATE | Partial Slope confirmed; panel FE not run |
| G11 | Reserve currency structural break | MODERATE | Resolved |
| G12 | Pre-1972 crosswalk | LOW | Open |
| G13 | 78% holdout unsubstantiated | LOW | Resolved Marginal +5pp over persistence |
All analysis used base Python (csv, math, statistics, random). No numpy, scipy, or statsmodels. Bootstrap CIs approximate. Regression SEs assume homoskedasticity.
Did not access raw FH or V-Dem microdata. Did not merge external economic data. 67% crosswalk match suggests data quality issues for ~1/3 of observations.
The AR(1) that "beats" stage models is itself a low-dimensional linear model. It cannot capture regime changes, structural breaks, or nonlinear dynamics.
The US has only 13 observations in the dataset (1800-2025). Any US-specific statistical claim has very wide confidence intervals.
| # | Action | Priority | Effort |
|---|---|---|---|
| 1 | Wait for FH 2026 and V-Dem 2026 official reports | CRITICAL | 1-2 months |
| 2 | Replace stipulated σ (note: all values were stipulated without empirical basis) with data-driven estimates in Monte Carlo | CRITICAL | 1 day |
| 3 | Add AR(1) mean-reversion force to projections | CRITICAL | 1 day |
| 4 | Present recalibration table as core sensitivity analysis | HIGH | 1 day |
| 5 | Operationalize policy vs. structural erosion taxonomy (CA5) | HIGH | 1 week |
| 6 | Model democratic tenure as stabilizing factor (CA6) | HIGH | 1 week |
| 7 | Add GDP per capita as covariate (CA7) | HIGH | 3 days |
| 8 | Reframe Monte Carlo as distribution + scenarios | HIGH | 2 days |
| 9 | Seek external peer review (pol sci + fin econ) | MEDIUM | 2-4 weeks |
| 10 | Publish crosswalk formula and replication code | MEDIUM | 1 day |