Regime Research is a statistical monitoring platform that classifies the structural health of the S&P 500 every trading day. The engine applies mathematics from critical transitions theory — originally developed in ecology and climate science — to detect when the statistical properties of market returns begin exhibiting signatures associated with increasing structural instability.

These signatures have been documented in peer-reviewed research published in Nature (2009), Science, PLOS ONE, and dozens of domain-specific journals over the past two decades.

The system does not predict market direction. It classifies structural conditions. Markets can rise while structurally weakening. They can fall while structurally sound. The regime state and the price trend are measuring different things.

The system classifies every trading day into one of four directional states:

Clear

No structural deterioration detected. The system is absorbing shocks normally. Statistical properties of returns are within historical baseline ranges.

Historical context: During CLEAR periods, the S&P 500 has exhibited an annualized return of approximately 15.5% with a Sharpe ratio of 1.25.*

Deteriorating

Early statistical signatures of structural weakening. The system is losing resilience — perturbations take longer to dissipate, or the distribution of outcomes is becoming asymmetric.

This does not mean a drawdown is imminent. Many DETERIORATING periods resolve without significant market decline.

Historical context: During DETERIORATING periods, the S&P 500 has exhibited an annualized return of approximately 0.3% with a Sharpe ratio of 0.02.*

Alert

Confirmed structural breakdown. Multiple statistical signatures of systemic fragility are active. This is the system's highest-severity classification for active deterioration.

Not every ALERT precedes a major drawdown. The system generates false positives — periods where structural deterioration is detected but the market recovers without a significant decline.

Historical context: During ALERT periods, the S&P 500 has exhibited an annualized return of approximately −6.4% with a Sharpe ratio of −0.28.*

Recovering

Post-ALERT normalization. The system previously identified structural breakdown, and conditions are now stabilizing. Statistical properties are returning toward baseline.

Historical context: During RECOVERING periods, the S&P 500 has exhibited an annualized return of approximately 21.7% with a Sharpe ratio of 1.05.*

* Based on hypothetical backtested analysis of the S&P 500 from 1995–2025 using a 60-day recovery window. These results were not achieved by any actual investor and do not reflect actual trading. Past structural conditions are not indicative of future results.

The engine monitors multiple independent channels of structural health simultaneously. Each channel tracks a different dimension of market behavior. The detection phase describes where the system is in the detection sequence:

Quiet

Neither channel is in an elevated state. Structural conditions are within normal ranges across all monitored dimensions.

Early Detection

One channel has identified distributional stress — the shape of the return distribution is becoming asymmetric in a way associated with increasing instability. The other channel has not yet confirmed structural breakdown.

This phase represents the system detecting something that may or may not develop further. Many Early Detection events resolve without progressing.

Confirmed Regime

The stability channel has identified structural breakdown — patterns indicate reduced system resilience. This can occur with or without prior distributional stress.

Escalation

The highest-conviction classification. Distributional stress was identified first (Early Detection), and structural breakdown subsequently confirmed — typically days or weeks later.

This sequential confirmation pattern — where one channel detects something early and a second channel independently confirms — has historically been associated with the most significant structural episodes in the system's backtest.

The engine tracks recurrence — how many times structural deterioration has been detected within a rolling window. First-time detections have different statistical properties than repeated detections.

When the system detects deterioration, resolves, and then detects it again in short succession, the second detection carries higher conviction.

The system has never gone directly from a quiet state to its highest alert level without at least one prior episode in the lookback window. Flickering precedes breaking.

Current State

The large display at the top shows today's directional state and detection phase. The seismograph animation reflects the current structural conditions.

Channel Health

Three gauges showing real-time readings for each detection channel plus the Market Breadth indicator, normalized to a 0–100 scale. Green (0–69) indicates baseline conditions. Amber (70–89) indicates elevation. Red (90–100) indicates critical levels.

The Market Breadth gauge is a context indicator — it does not trigger state changes, but elevated readings alongside channel activity warrant attention.

Regime Timeline

The main chart shows SPY price with color-coded backgrounds representing the directional state on each trading day. Use the time range selector (YTD, 1Y, 3Y, ALL) to zoom in or out.

State Distribution

A donut chart showing what percentage of the selected time period was spent in each directional state.

Drawdown Overlay

A bar chart showing the forward maximum drawdown following each ALERT entry. Red bars indicate episodes where a significant drawdown followed (HIT). Gray bars indicate episodes where the market recovered without a major decline (MISS). This chart shows both successes and failures transparently.

Episode Tracker

Details of the current active episode, including duration, state progression, and detection phase.

Episode History

A complete table of all historical episodes with duration, trigger type, detection phase, maximum drawdown, and outcome.

Detection Phase Log

A timeline showing every detection phase transition — when the system entered Early Detection, when it escalated, and when it returned to Quiet.

Historical Episodes

Past ALERT episodes for reference, shown in reverse chronological order, with their outcomes.

Applied to the S&P 500, the engine identified structural deterioration preceding 10 of 11 major drawdowns since 1997. Median ALERT lead time: 47 calendar days.

•• Both channels active · • Single channel detection

The only miss: September 11, 2001 — a sudden external attack with no statistical precursors in any domain.

Important: Not every detection precedes a major drawdown. The system generates false positives. False alarms cost attention. Missed detections are irreversible.

All detection results are based on hypothetical backtested analysis developed with the benefit of hindsight.

The Scientific Foundation

In complex systems — from lakes to power grids to financial markets — catastrophic transitions are preceded by measurable statistical signatures. The system takes longer to recover from small perturbations. Fluctuations grow in magnitude. The distribution of outcomes becomes asymmetric.

Whether these signatures reflect classical critical slowing down (proximity to a bifurcation point) or increasing stochastic instability (growing noise intensity), the practical detection value is the same: the system is becoming structurally fragile.

The Engine

The engine monitors multiple independent statistical channels simultaneously. Each channel tracks a different dimension of market structure using distinct mathematical properties of the return distribution.

The engine processes the full available history after each market close. Classifications are not based on fixed thresholds — they adapt as new data arrives, comparing current conditions against the expanding historical record.

When channels independently confirm deterioration, conviction is highest. This dual-channel architecture was validated through extensive out-of-sample testing.

What We Publish

The current regime state, detection phase, historical context for each state, conviction level, episode tracking with historical parallels, and channel health readings.

What We Don't Publish

The specific indicators, parameters, weightings, and thresholds that drive the classification. The methodology is proprietary. The science it's built on is not.

The foundational research was published in Nature (2009) and has since been validated in Science, PLOS ONE, and dozens of domain-specific journals.

Our Market Breadth channel monitors the relative health of the broader market versus the headline index.

When smaller companies begin underperforming significantly, it often signals internal market deterioration before it appears in the S&P 500's own statistics. This divergence can emerge weeks before structural deterioration becomes visible in the primary detection channels.

This channel provides early context — it does not trigger state changes on its own, but elevated readings alongside other channel activity warrant attention.

Readings are normalized on the same 0–100 scale as the Stability and Distribution channels. Green (0–69) indicates healthy broad participation. Amber (70–89) indicates narrowing. Red (90–100) indicates significant stress.

What This System Does Not Do

• It does not predict market direction or magnitude of moves
• It does not provide buy, sell, or hold recommendations
• It does not replace professional financial advice
• It does not account for individual financial circumstances
• It does not guarantee detection of future market events

False Positives

The system generates false positives — periods where structural deterioration is detected but the market recovers without a significant decline. This is inherent to any early warning system.

False positives are not system failures. They indicate that structural conditions deteriorated and then self-corrected — a real phenomenon, just not one that resulted in a major market event.

Backtested Performance

All historical data, detection records, and state return profiles presented in this dashboard are based on backtested analysis using data from 1995–2025. These results:

• Were not achieved by any actual investor
• Do not reflect actual trading
• Do not account for transaction costs, slippage, or taxes
• Were developed with the benefit of hindsight
• Are not indicative of future results

There are frequently material differences between hypothetical backtested results and actual results.

Out-of-Sample Validation

The system has been computing live daily classifications since 2025. The live track record is documented in the Episode History and can be verified against the daily blog posts published at regime-pulse.ghost.io.

Structural Deterioration

A measurable change in the statistical properties of market returns indicating reduced system resilience.

Regime State / Directional State

The system's daily classification of structural conditions (CLEAR, DETERIORATING, ALERT, RECOVERING).

Detection Phase

Where the system is in the sequential detection process (QUIET, EARLY DETECTION, CONFIRMED REGIME, ESCALATION).

Episode

A continuous period of elevated structural deterioration, from the first ALERT entry through recovery to CLEAR.

Recurrence

How many times the system has entered ALERT within a rolling window. Higher recurrence indicates persistent structural instability.

Stability Channel

One of the engine's two independent detection channels. Monitors structural resilience — how quickly the system recovers from perturbations.

Distribution Channel

The second independent detection channel. Monitors the symmetry of market outcomes — whether the return distribution is becoming skewed.

Market Breadth

A context indicator monitoring the relative health of the broader market versus the headline index. When broad market participation narrows significantly, it can signal internal deterioration before it appears in the S&P 500's own statistics. This indicator does not trigger state changes on its own.

Channel Health

A normalized 0–100 reading for each channel. Green (0–69) is baseline. Amber (70–89) is elevated. Red (90–100) is critical.

Historical Episode

A past ALERT episode shown for reference, with its duration, peak state, and outcome.

Conviction Tier

A classification of alert reliability based on recurrence patterns and detection phase sequencing.

Forward Max Drawdown

The largest peak-to-trough decline in SPY price within a specified window following an ALERT entry. Used to assess whether a detection was followed by a significant market decline.

Hit / Miss

An episode outcome classification. HIT = a drawdown exceeding 7% occurred within 90 days of ALERT entry. MISS = it did not.

Regime Research is a statistical monitoring platform built on peer-reviewed complexity science methodology applied to financial market data. It provides scientific observations about structural market conditions for educational and informational purposes only.

Regime Research does not provide investment advice, investment recommendations, or solicitations to buy, sell, or hold any security. The classification of market regimes is a scientific observation about structural conditions, not a direction to act on.

The operator is not a registered investment adviser, broker-dealer, or financial planner. No fiduciary relationship is created by use of this service.

All historical detection data is based on backtested analysis applied to historical market data. These results were not achieved by any investor, do not reflect actual trading, and do not account for transaction costs, slippage, or taxes. Results were developed with the benefit of hindsight. Past structural conditions are not indicative of future results.

The system generates false positives — periods where structural deterioration is detected but no significant market decline follows. False positives are inherent to early warning systems. Users should not interpret any regime classification as a guarantee of future market direction.

Users should consult qualified financial professionals before making investment decisions. There are frequently material differences between hypothetical backtested results and actual results.