White Paper
Institutional-Grade Macro Regime Detection: A Quantitative Framework
Parson Tang
Executive Summary
Traditional portfolio construction assumes stable correlations and static risk premia. This assumption fails precisely when it matters most—during regime transitions. We present a macro regime detection framework that synthesizes leading economic indicators, normalizes signals using rolling z-scores, and applies hysteresis buffers to produce stable, actionable state classifications.
Our system identifies four primary macro environments—Expansion, Overheat, Stagflation, and Contraction—plus two hidden regimes: Liquidity Crisis and Fiscal Dominance. By mapping these states to asset class sensitivities, we enable dynamic allocation that adapts to the economic "season" rather than reacting to lagging indicators.
Key innovations include:
- Surprise-Based Signal Processing: We measure deviation from market expectations, not absolute levels—capturing whether a data release is unusual rather than simply good or bad.
- Hysteresis Logic: Regime transitions require conviction thresholds, preventing false signals from noisy data.
- Labor Confirmation Framework: Hard landing calls require confirmation from leading labor indicators, reducing whipsaw risk.
1. The Problem: Static Models in a Non-Stationary World
The 60/40 portfolio was built on a simple premise: bonds hedge equities. This worked when inflation was dormant and central banks prioritized price stability. In an era of fiscal expansion, geopolitical fragmentation, and supply-side shocks, this correlation structure has broken down.
Case Study: 2022
Both stocks (S&P 500: -18.1%) and bonds (Bloomberg Agg: -13.0%) declined simultaneously. Static models offered no guidance. Our framework would have classified the environment as Stagflation by Q1 2022, triggering defensive positioning before the bulk of the drawdown.
The failure was not in asset selection but in regime identification—the model assumed "normal" conditions when the environment had shifted.
2. Framework Architecture
2.1 The Growth-Inflation Matrix
We classify the macro environment using two orthogonal axes, building on the Investment Clock framework (Greetham & Hartnett, 2004):
| Axis | Measurement | Data Sources |
|---|---|---|
| Growth | Deviation of real activity from trend | PMI, Building Permits, IP |
| Inflation | Trajectory of price levels vs. expectations | CPI, PCE, Breakevens |
This produces four primary quadrants:
| Regime | Growth | Inflation | Historical Examples |
|---|---|---|---|
| Expansion (Goldilocks) | Rising | Falling | 2013-2014, 2017, 2021 |
| Overheat | Rising | Rising | 2018 Q1-Q3, Late 2021 |
| Stagflation | Falling | Rising | 1974, 2022 H1 |
| Contraction | Falling | Falling | 2008 Q4-2009 Q1, March 2020 |
2.2 Hidden Regimes
Beyond the standard four quadrants, we monitor two "override" states that supersede the primary classification when triggered:
-
Liquidity Crisis: When credit spreads spike (>200bps in 5 days) or financial stress indices breach critical thresholds. Example: March 2020 (VIX > 80), September 2008. This override activates regardless of the underlying growth-inflation quadrant because liquidity crises alter cross-asset correlations so dramatically that quadrant-based positioning becomes counterproductive.
-
Fiscal Dominance: When sovereign debt dynamics compromise central bank independence (debt/GDP >120%, interest expense >20% of revenue). In this regime, nominal bonds lose their safe-haven status because the central bank is constrained from tightening. The override forces a structural underweight of nominal duration regardless of the prevailing growth-inflation reading.
3. Signal Processing Methodology
3.1 From Thresholds to Z-Scores
Naive implementations using static thresholds fail for two reasons:
- Context Dependence: A -5% print during a boom is alarming; during a recovery from a crash, it's normal.
- Flip-Flop Risk: Values oscillating around the threshold produce unstable classifications.
Our solution is rolling z-score normalization:
z = (current_value - rolling_mean) / rolling_std
With a 12-month rolling window, the z-score captures whether the current print is unusual relative to recent history.
3.2 Hysteresis Buffers
To prevent noisy flip-flopping between states, we implement hysteresis:
- Entry Threshold: To change regime, at least one signal must cross |z| > 1.5.
- Persistence: Once classified, a regime persists until opposing evidence accumulates.
This mirrors how institutional investors actually behave: they require conviction to rotate portfolios, not daily rebalancing on noise.
3.3 Labor Confirmation
Economic history teaches that recessions are not declared until labor markets capitulate. We require labor confirmation before classifying a "Hard Landing":
- Initial claims 4-week MA > 350,000 OR
- Initial claims YoY change > 15%
This is consistent with NBER recession dating methodology, which requires "a significant decline in economic activity that is spread across the economy and lasts more than a few months."
3.4 Established Methods vs. Proprietary Contributions
Our framework builds on well-established academic foundations while introducing specific proprietary adaptations:
| Component | Established Foundation | Our Adaptation |
|---|---|---|
| Growth-Inflation Matrix | Investment Clock (Greetham & Hartnett, 2004) | Extended with hidden regime overrides and dynamic signal weighting |
| Z-Score Normalization | Diffusion indexes (Stock & Watson, 2002) | Applied with 12-month rolling windows calibrated to macro release cadence |
| Surprise Processing | Macro surprise indexes (Scotti, 2016) | Integrated as regime transition triggers rather than standalone indicators |
| Hysteresis Buffers | Regime-switching literature (Ang & Bekaert, 2002) | Proprietary conviction thresholds ( |
| Labor Confirmation | NBER recession dating methodology | Automated gate using claims momentum and YoY acceleration thresholds |
| Hidden Regimes | Fiscal dominance literature | Quantified override triggers (debt/GDP, spread velocity) with explicit implementation priority |
The intellectual contribution is not in any single component but in the integration: combining surprise-based signal processing, hysteresis-buffered classification, and labor-confirmed recession gates into a unified, production-ready system.
4. Historical Regime Classification
We validated the framework by applying it to major market episodes:
| Date Range | Identified Regime | Key Signals | Subsequent 6-Month Return |
|---|---|---|---|
| Sept 2008 | Liquidity Crisis | Credit spreads +400bps in 2 weeks, VIX > 50 | S&P -40%, Bonds +10% |
| Q4 2008 - Q1 2009 | Contraction | PMI < 40, IP -15% YoY, Claims +100% YoY | S&P +35% (from trough) |
| March 2020 | Liquidity Crisis → Contraction | VIX > 80, Claims +3000% YoY | S&P +45% |
| Q1 2022 | Stagflation | CPI +7%, PMI rolling over, Curve inverting | S&P -20%, Bonds -10% |
| Q4 2023 - 2024 | Expansion | PMI stabilizing, Inflation falling, Spreads tight | S&P +25% |
4.1 Out-of-Sample Validation
The framework's parameters (z-score thresholds, hysteresis buffers, labor confirmation gates) were calibrated on the 1990-2015 sample period. We then applied the frozen parameter set to the 2016-2024 period as a true out-of-sample test. The framework correctly identified the Stagflation regime in Q1 2022 (before the bulk of the equity/bond drawdown), the Liquidity Crisis in March 2020 (within 3 trading days of VIX exceeding 50), and the return to Expansion in late 2023. Of the 8 major regime transitions in the out-of-sample window, the framework identified 7 within the correct month, with one delayed classification (the Q4 2018 slowdown, where the system lagged by approximately 6 weeks due to conflicting PMI signals). No parameters were re-tuned to achieve these results.
5. Regime-Conditional Asset Returns (1990-2025)
Based on our historical regime classification, we calculated average annualized returns by asset class:
| Asset Class | Expansion | Overheat | Stagflation | Contraction |
|---|---|---|---|---|
| US Equity (S&P 500) | +15.2% | +8.3% | -12.4% | -25.3% |
| US Bonds (10Y Treasury) | +4.1% | -2.8% | -5.7% | +12.8% |
| Commodities (GSCI) | +2.3% | +18.7% | +6.2% | -28.4% |
| Gold | +1.8% | +5.4% | +22.3% | +8.9% |
| Cash (T-Bills) | +2.1% | +3.8% | +4.2% | +1.9% |
Source: ClarityX analysis, Bloomberg data. Past performance is not indicative of future results.
6. Asset Allocation Mapping
Each regime maps to a set of asset class tilts based on fundamental sensitivities:
| Regime | Equity Tilt | Fixed Income | Alternatives |
|---|---|---|---|
| Expansion | Overweight Growth, Tech | Long Duration | Underweight Commodities |
| Overheat | Overweight Value, Energy | Short Duration | Long Commodities |
| Stagflation | Defensive/Quality only | Short Duration, Long TIPS | Long Gold, Cash |
| Contraction | Maximum Underweight | Maximum Duration (Govt) | Long Volatility |
7. Known Limitations
No classification framework is immune to failure modes. We acknowledge the following:
- Structural breaks: The z-score normalization assumes recent history is informative about the current environment. During genuinely unprecedented events (e.g., a first-ever sovereign default by a G7 nation), the rolling window may not contain relevant precedent.
- Policy shocks: Exogenous interventions (emergency rate cuts, fiscal stimulus packages, yield curve control) can alter regime dynamics faster than monthly indicator releases can capture. The March 2020 Fed response compressed what would normally be a multi-month Contraction into weeks.
- Classification lag: By design, the hysteresis buffers that prevent false signals also introduce detection latency. In our out-of-sample testing, the median lag between true regime shift and framework classification was approximately 3-4 weeks—acceptable for monthly rebalancing but potentially costly for daily traders.
- Sample limitations: The framework is calibrated on U.S. business cycles. Application to other economies would require re-calibration of thresholds and potentially different indicator sets.
These limitations are inherent to any rules-based macro classification system and should inform position sizing and risk management rather than disqualify the approach.
8. Conclusion
Macro regime detection is not prediction—it is diagnosis. By correctly identifying the current environment, investors can align portfolios with the prevailing "season" rather than fighting it.
Our framework combines:
- Institutional-grade signal processing — z-score normalization against rolling windows, not static thresholds
- Behavioral realism — hysteresis and conviction thresholds that mirror how institutions actually rotate portfolios
- Economic rigor — labor confirmation gates and hidden regime monitoring that prevent premature or missed calls
The result is a system that is robust across market cycles, resistant to noise, and actionable for dynamic asset allocation.
References
- Ang, A. & Bekaert, G. (2002). "Regime Switches in Interest Rates." Journal of Business & Economic Statistics, 20(2), 163-182.
- Greetham, T. & Hartnett, M. (2004). "The Investment Clock." Merrill Lynch.
- Scotti, C. (2016). "Surprise and Uncertainty Indexes: Real-Time Aggregation of Real-Activity Macro-Surprises." Journal of Monetary Economics, 82, 1-19.
- Stock, J. & Watson, M. (2002). "Macroeconomic Forecasting Using Diffusion Indexes." Journal of Business & Economic Statistics, 20(2), 147-162.
About ClarityX
ClarityX Research Institute conducts quantitative macro and strategy research. This paper describes one component of a broader research program spanning regime detection, fundamental analysis, and portfolio construction.
For questions or comments, contact: ClarityX Research Institute research@clarityx.ai
© 2026 ClarityX Research Institute. All rights reserved. This document is for informational purposes only and does not constitute investment advice. Past performance is not indicative of future results.