427 Regime Changes Analyzed

This research article presents VantMacro's empirical analysis of 427 market regime transitions spanning 23 years (2003-2026). It covers regime durations, transition probabilities, asset performance, and the limitations of regime-based analysis.

Key findings:

• Regime patterns partially persist out-of-sample (average sign consistency 0.80 across 5 assets; equities strong, long Treasuries weak)
• Most regimes last days to weeks, not months
• Equity performance varies significantly by regime; bonds and gold show weaker regime dependence
• Transition probabilities follow recognizable patterns, but timing remains uncertain

---

Methodology

Data Source

• Period analyzed: 2003-2026 (23 years)
• Regime transitions: 427 identified
• Asset data: Twelve Data historical prices (daily)
• Regime classification: VantMacro's 7-state composite model

Regime Classification Model

VantMacro classifies regimes using three dimensions:

1. Real Cycle State — Growth and inflation dynamics (4 states)
2. Liquidity & Policy State — Central bank stance (3 states)
3. Market Risk State — Risk appetite (3 states)

These combine into 7 composite regimes:

• Reflationary Expansion
• Late-Cycle Inflationary Boom
• Stagflationary Squeeze
• Disinflationary Slowdown
• Post-Shock Recovery
• Crisis/Liquidation
• Transitional

Backtesting Methodology

• Returns: Calculated for each asset within each continuous regime period (then aggregated by regime)
• Out-of-sample validation: Time split — in-sample (2003-12-17 to 2014-12-31) vs out-of-sample (2015-01-01 to 2025-12-19), comparing regime-level patterns across the two windows

---

Regime Duration Statistics

One surprising finding: most regimes are short—measured in days or weeks, not months.

Interpretation

Why are durations so short?

VantMacro's model is a high-frequency tactical model. It detects regime shifts quickly—often before traditional economic indicators would flag a change. This captures:

1. Rapid transitions during volatile periods — Crisis and recovery regimes can shift multiple times per month
2. Mixed signal periods — Transitional regimes are inherently brief
3. Sensitivity to real-time data — Daily data allows faster detection than monthly economic releases

Implication: Regimes should be understood as "current conditions" snapshots, not long-term phases. A "Reflationary Expansion" label is descriptive of today, not predictive of next quarter.

---

Transition Probabilities

Regime transitions follow recognizable patterns. The table shows the most likely next regime from each current state.

From Reflationary Expansion

Interpretation: Reflationary environments often transition to recovery-type conditions or slowdown—consistent with mid-cycle dynamics.

From Late-Cycle Inflationary Boom

Interpretation: Late-cycle tends toward either stagflation (if inflation sticks) or transitional (if signals become mixed). Reverting to reflationary is less common.

From Stagflationary Squeeze

Interpretation: Stagflation often resolves back to late-cycle (if growth stabilizes) or disinflationary slowdown (if demand destruction wins).

From Disinflationary Slowdown

Interpretation: Disinflationary conditions typically transition to recovery or expansion—the classic "bad news is good news" (policy response kicks in).

From Post-Shock Recovery

Interpretation: Recovery can go multiple directions—back into crisis (double-dip), into expansion (sustained recovery), or into slowdown (stalled recovery).

From Crisis/Liquidation

Interpretation: Crises almost always transition to recovery. The key question is how quickly, not whether.

From Transitional

Interpretation: Transitional periods typically resolve toward late-cycle dynamics—suggesting mixed signals often precede inflationary pressures.

---

Asset Performance by Regime

This section focuses on what the statistical tests actually support (and where they don’t).

Analysis of Variance (ANOVA): Do Returns Differ Across Regimes?

We run a one-way Analysis of Variance (ANOVA)—a statistical test that checks whether average returns differ across regimes—on daily returns grouped by regime for 8 assets. This tests whether the differences we see are likely real or just random noise. (It does not imply strong predictability.)

Reading the table below:

• p-value: The probability that the observed differences happened by chance. Lower is better. Values below 0.05 are considered statistically significant; values below 0.001 are highly significant.
• Effect Size (η²): How much of the variation in returns is explained by regime. Values around 0.01 mean regimes explain about 1% of daily return variation.

Important context on "small" effects: Daily returns are dominated by noise—news, earnings, random fluctuations. No single factor explains much of daily variance. An effect size of 1% on a daily basis can still compound into meaningful performance differences over weeks or months (the typical regime duration). Think of it like wind on a sailboat: a small but persistent force in one direction adds up over time.

Interpretation: Equities show statistically significant regime dependence (p < .001 means there's less than a 0.1% chance these patterns are random). Effect sizes of ~1% are typical for daily data—daily returns are inherently noisy. What matters is that these small daily differences are consistent and can accumulate over regime periods. Bonds and gold show no meaningful regime effects in this framework.

---

Out-of-Sample Validation (Time Split)

We evaluate persistence by comparing regime-level performance patterns across time:

• In-sample: 2003-12-17 to 2014-12-31
• Out-of-sample: 2015-01-01 to 2025-12-19

Two metrics are reported:

• Correlation: correlation between in-sample vs out-of-sample regime CAGRs (across regimes present in both windows)
• Sign consistency: share of regimes whose in-sample and out-of-sample CAGRs have the same sign

Interpretation by asset class:

• Equities (SPX, QQQ): Strong persistence. Correlations above 0.9 and high sign consistency mean regime patterns held up well out-of-sample.
• Risk assets (HYG, GLD): Moderate persistence. Patterns generally held but with more variability.
• Long-duration Treasuries (TLT): Patterns reversed. The strongly negative correlation (-0.958) means regimes that were good for TLT in-sample became bad out-of-sample (and vice versa). This is a key finding—regime analysis doesn't work for bond allocation in this framework.

Note on averages: The "overall" average (0.418 correlation, 0.80 sign consistency) mixes equities with TLT, which is misleading since TLT's patterns are inverted. For equities alone, persistence is much stronger.

Bottom line: Use regimes for equity context and risk framing. Don't use them for bond allocation decisions.

---

Verified Historical Examples

Only periods that meet a 50%+ verification threshold are listed. Most regimes lack verified examples because classifications are data-driven and may differ from narrative interpretations.

Post-Shock Recovery

• 2009-04-01 to 2009-12-31: Post-GFC recovery initiation (64% verified)
• 2020-04-01 to 2020-09-30: Post-COVID shutdown recovery (61% verified)

Other Regimes

Most other regimes do not have examples that meet the verification threshold. This reflects the model's sensitivity and the difficulty of matching algorithmic classifications to human-identified "regime periods."

---

Limitations

1. Short Durations Limit Statistical Power

With median durations of 3-31 days, individual regime periods have limited data. Aggregating across many periods helps, but statistical significance varies by regime.

2. Look-Ahead Bias Risk

All analysis is conducted on historical data. Real-time regime classification may differ due to:

• Data revisions (especially GDP, employment)
• Publication lags (some indicators release with delay)
• Model updates

3. Metric Interpretation (Important)

Several metrics used in regime research are easy to misunderstand:

• Short regimes → unstable annualization: many regime periods are short, so annualized CAGRs can swing wildly from small samples.
• Regime-level vs time-series: out-of-sample “correlation” here is computed across regimes (comparing regime CAGRs in two time windows), not correlation of daily returns.
• Statistical significance ≠ predictability: even when p-values are tiny, effect sizes are small (η² ≈ 0.01 for equities).

4. Structural Changes

The regime-asset relationships from 2003-2026 may not hold in the future due to:

• Central bank policy regime shifts
• Market structure changes (passive indexing, algorithmic trading)
• Monetary policy framework changes

5. Proxy & Data Caveats

Some instruments and proxies have idiosyncrasies (e.g., futures roll yield for certain commodity ETFs). Treat any single ticker as a proxy, not a perfect representation of an asset class.

---

Practical Implications

Use Regimes For:

1. Understanding current conditions — "Are we in a supportive or challenging macro environment?"
2. Portfolio positioning tilts — "Should I lean more defensive or more risk-on?"
3. Avoiding fighting the trend — "The regime is crisis—don't buy the dip yet"
4. Cross-checking other signals — "Technicals say buy, but regime says crisis—be cautious"

Don't Use Regimes For:

1. Precise return forecasts — Out-of-sample results are mixed and not a substitute for a forecasting model
2. Timing decisions — Regimes tell you where you are, not when to act
3. Bond allocation — Bond returns aren't regime-dependent enough
4. Gold allocation — Same issue as bonds

---

Conclusion

The analysis of 427 regime transitions over 23 years reveals that:

1. Out-of-sample persistence is mixed — average sign consistency 0.80 across 5 assets; correlation averages 0.418 (equities strong, TLT weak)
2. Regimes are short — Median durations of 3-31 days, not months
3. Equities show statistically significant regime effects — but effect sizes are small (η² ≈ 0.01)
4. Bonds and gold show weak regime dependence in this framework’s tests
5. Use regimes for context and risk framing — not mechanical timing or point forecasts

VantMacro's regime model provides a systematic framework for understanding macro conditions, but it is not a timing tool or crystal ball. Use it as context for decision-making, not as investment advice.

---

Data Sources

• VantMacro dashboard methodology — /dashboard/methodology
• FRED data (macro inputs and stress indicators) — https://fred.stlouisfed.org/

Further Reading

• Market Regimes Explained — Overview of the 7 composite regimes
• The Complete Guide to Global Liquidity — How liquidity feeds into regime classification

---

Explore Regime Analysis on VantMacro

• Real-time regime detection with confidence scores
• Empirical performance data for current regime
• Transition monitoring and historical comparisons

View Dashboard →