The Coupled Spectrum
Sixty-eight markets. One system. The coupling never switches off.
Episode 1 proved that the near-zero autocorrelation is a lie. Beneath it, every market oscillates between periods where moves persist and periods where they chop back and forth. Positive feedback and negative feedback, alternating across four decades.
But Episode 1 looked at markets one at a time. Each chart showed a single contract. A single trace. A single oscillation.
This episode looks at all sixty-eight simultaneously. What emerges is not sixty-eight independent engines running on their own schedules. It is a single coupled system moving along a spectrum, and the coupling never switches off.
The Heatmap
We computed the rolling two-year autocorrelation for every contract in the universe and assembled the results into a single image. Each row is a market. Each column is a date. The colour tells you which force is dominant: blue for positive feedback (trending), red for negative feedback (oscillating), and black for the balanced zone between them.
Figure 2.1 Heatmap of rolling two-year ACF(1) for all sixty-eight contracts, 1986 to 2026. Blue: positive feedback (trending). Red: negative feedback (oscillating). Black: balanced. Read this chart in three layers.
Read this chart in three layers.
First, the texture. At any given date, the column is not a uniform colour. There are always some blue squares, some red squares, and some black. Markets are not all in the same state at the same time. The system has variation. Individual markets retain their own character.
Second, the vertical bands. Despite the variation, there are periods where a single colour dominates. Wide vertical bands of red appear in the mid-2000s and again from 2018 through early 2020. Wide vertical bands of blue appear in the early 1990s and around 2011 to 2013. These bands mean that the majority of markets are being pulled toward the same end of the spectrum simultaneously. Something is coordinating them.
Third, the transitions. Look at how the bands shift. The red band of 2018 to 2020 does not gradually fade. It ruptures in March 2020, replaced within months by a mixed and then blue-leaning pattern. The calm of the mid-2000s does not slowly transition into the GFC. It holds, then breaks. The transitions are sharp, not gradual.
These three layers, variation within coordination, vertical bands of dominance, and sharp transitions, are the visual signature of what we will now prove: a permanently coupled system whose spectral position shifts in response to macro forces.
The Spectrum
Episode 0 introduced the spectrum in words. Now we can see it in data.
At any given date, each of the sixty-eight markets sits somewhere along the axis from strong positive feedback to strong negative feedback. Some are trending. Some are oscillating. Some are balanced. The question is whether these positions are independent or coordinated.
We measured the cross-sectional mean autocorrelation at each date: the average across all sixty-eight markets of their individual rolling ACF values. When this average is positive, the system as a whole leans toward trending. When negative, toward oscillating. When near zero, the system is balanced.
Using the standard deviation of this cross-sectional mean as a threshold, we classified each month into one of three states. Coupled trending: the system leans positive, more than one standard deviation above zero. Coupled mean-reversion: the system leans negative, more than one standard deviation below zero. Noise zone: the system is balanced, within the band.
The system spends thirty-one percent of its time in coupled trending, six percent in coupled mean-reversion, and sixty-three percent in the noise zone. The asymmetry is notable: the system rarely reaches the deep oscillatory end, but when it does, the episodes are intense and prolonged. The 2005 calm and the 2018 to 2020 period are two examples, and they are among the most extreme readings in the dataset.
The Coupling Never Switches Off
This is the central empirical finding of the series.
If markets operated independently, moving to their own rhythms without coordination, we would expect to see a specific pattern in the data. During the noise zone, when no single force dominates the aggregate, individual markets should be scattered across the spectrum. The cross-sectional dispersion, how spread out the sixty-eight ACF values are at any given date, should increase in the balanced state and decrease when one force dominates.
It does not.
The cross-sectional dispersion is 0.071 during coupled trending, 0.072 during the noise zone, and 0.071 during coupled mean-reversion. These numbers are statistically identical. The markets are equally tightly clustered in every state. The coupling is not switching on during stress and off during calm. It is constant.
A second test confirms this. If sixty-eight markets were truly independent, each with a fifty-fifty chance of being in positive or negative feedback, the expected fraction sharing the majority sign would be approximately 0.549. The observed average is 0.599. The observed range runs from 0.500 to 0.823. The fraction exceeds the independence prediction seventy-four percent of the time. In no year since 1987 has the average majority-sign fraction fallen to the level that independence predicts.
A third test: cross-market return correlation by spectral state. During periods when the system leans toward oscillation, the average cross-market return correlation is 0.311. During balanced periods, 0.240. During trending periods, 0.241. Markets are correlated in their oscillation as much as or more than in their trending. The coupling produces different surface expressions depending on the spectral position, but the underlying connection persists in every state.
The most parsimonious explanation for permanent coupling is that modern market participants have simultaneous exposure across many markets. A single institution, fund, or systematic strategy touches dozens of instruments at once. When it acts, its actions propagate across every market it touches. The coupling is not a mysterious synchronisation. It is the mechanical consequence of one-to-many participation in a market structure where almost no significant participant is confined to a single instrument.
The sign of the aggregate feedback, whether the coupled system trends or oscillates, reflects the balance between participants who extend moves and those who dampen them. Trend followers, momentum strategies, and breakout traders push prices further from their starting point: positive feedback. Value investors, rebalancers, and mean-reversion traders push prices back: negative feedback. The aggregate balance between these competing forces determines where the system sits on the spectrum at any given moment. The data proves the coupling exists and the spectral position shifts. The agent mechanism is the explanatory framework that makes sense of the findings.
What Each Force Contains
Episode 1 showed that the full-sample zero is an averaging artifact. Two opposing forces, positive and negative feedback, cancel when you compress decades into a single number. That finding dismantled the random walk.
But there is a subtler conclusion embedded in the data, one that Episode 1 approached without stating directly.
Directional memorylessness is not a property of price. It is a property of the average.
When you isolate the positive-feedback regime, price moves carry directional memory. Yesterday’s move predicts today’s direction. The autocorrelation is positive and measurable. The market is not flipping a coin. It is extending. When you isolate the negative-feedback regime, the same is true in reverse. Yesterday’s move predicts a reversal. Directional memory exists here too, pointing the other way. The two memories are real, opposing, and roughly balanced in the long-run aggregate. The cancellation produces the zero. The zero does not mean neither force remembers direction. It means both forces do, and they cancel.
This has an important consequence. When practitioners say that direction is unpredictable, they are describing the full-sample average. They are not describing the system. Inside each spectral regime, direction carries structure. The unpredictability is a population-level artifact of mixing two directionally structured states across time. A coupled system sitting in positive feedback is not a coin flip. It is a system in which the weight of agent behaviour is extending moves. The directional signal is there. It is only invisible from far enough away.
The heatmap introduced earlier shows where each force dominates across time and across markets. Reading it correctly means reading it as a map of directional memory, not its absence.
The Earthquake
The transition between spectral states is not gradual. It follows a pattern that resembles geological earthquakes, with three distinct phases.
Phase one: the rupture. A macro shock arrives. The coupled system, which may have been sitting in the noise zone or deep in the oscillatory state, shifts rapidly toward the trending end. Positive feedback floods through dozens of markets simultaneously. Directional energy is released across the system.
Phase two: the aftershock zone. The initial force exhausts. The system transits through the noise zone. Both feedbacks are active and partially offsetting. The data shows a specific signature during this phase: the noise zone, which normally contains forty-five to fifty percent of all markets, collapses. Markets polarise toward end-member states. During the GFC, the fraction of markets in the balanced state fell from its normal range to twenty to twenty-three percent. Some markets were trending violently. Others were whipsawing. The middle ground emptied. The earthquake destroyed the balanced state.
Phase three: the settling. The system arrives at the mean-reverting end. Negative feedback dominates. Markets oscillate in ranges, absorbing remaining energy through chop. This state can persist for years.
Three case studies illustrate the pattern.
Case Studies
Case Study 1: COVID (the cleanest earthquake)
The COVID transition is the most dramatic in the dataset. In February 2020, eighty-two percent of all markets were coupled in negative feedback, the deepest oscillatory reading in four decades. The cross-sectional mean ACF was negative 0.036, the most negative in the dataset.
Then the virus hit.
By March, the fraction had dropped to fifty-nine percent. By April, forty-six percent. By May, forty-two percent. In two months, the system ruptured from its most extreme oscillatory state to trend dominance. The February-to-April swing, from eighty-two percent oscillating to forty-six percent, is the fastest spectral transition in the sample.
The trailing window introduces a measurement lag: the ACF at any date reflects the prior two years of data, not just the current month. This means the February 2020 reading was dominated by 2018 and 2019 data, the calmest period in the dataset. The speed of the transition is actually understated by the rolling measure. The real-time shift was even more abrupt than the heatmap shows.
After the rupture, the system spent 2021 and 2022 in the aftershock zone and then trend-dominant territory, with the inflation build and rate-hiking cycle providing sustained directional fuel. This is the settling phase playing out in real time.
Case Study 2: The Gulf War (the purest trend earthquake)
The Gulf War period from 1990 to 1992 produced the most sustained trend coupling in the dataset. The system was already trend-leaning entering 1990, with sixty to sixty-five percent of markets showing positive feedback. The invasion of Kuwait in August 1990 intensified the coupling. By February 1992, eighty-one percent of markets were in the trending state, the highest reading in forty years.
This was almost entirely blue on the heatmap. There was no oscillatory phase preceding it, as there was with COVID. No extreme calm to rupture. The system was already trend-coupled and the geopolitical shock deepened it. The earthquake metaphor still applies, but this was a quake that struck a system already under directional stress, amplifying rather than reversing.
The decay was slow. Trend coupling remained above sixty percent through 1993 and did not return to balanced levels until 1994. The energy dissipated gradually, consistent with the settling phase lasting years when the initial shock is sustained.
Case Study 3: The GFC (the complex earthquake)
The GFC is the most instructive case because it is the most counterintuitive.
Conventional wisdom says the financial crisis synchronised markets. Everything crashed together. This is true of returns. It is not true of feedback structure.
The pre-GFC period, 2005 through 2007, was characterised by elevated oscillatory coupling. Seventy-five percent of markets were in negative feedback in early 2005. This was the Greenspan calm: low volatility, range-bound markets, and negative feedback dominating across the universe. On the heatmap, it shows as a wide red band.
When the crisis hit in 2008, the system did not shift cleanly to one end of the spectrum. The GFC shows only fifty-six percent average oscillatory coupling through the crisis period, lower than the calm that preceded it. The heatmap shows a mixed pattern, not the clean blue band that the Gulf War produced. Some markets were trending violently (bonds rallying, equities falling, certain commodities collapsing). Others were whipsawing on a daily basis. The noise zone collapsed as markets polarised.
The measurement lag compounds this. The ACF reading at September 2008 reflects data from September 2006 to September 2008, a window that is mostly pre-crisis calm. The full crisis impact on the rolling ACF does not appear until 2010, when the window has shifted to capture purely crisis-era data. By then, the system shows fifty-four to fifty-nine percent oscillatory, reflecting the post-crisis settling phase.
The GFC earthquake was complex. A prolonged pre-crisis calm (deep oscillatory coupling). A rupture that polarised markets rather than synchronising them toward one end. Then a multi-year settling phase in moderate oscillatory coupling from 2009 through 2011. By 2011, the system had shifted back toward trend coupling (sixty-eight percent trending by September 2011), as the Euro crisis provided a new directional catalyst.
The lesson: not every earthquake produces the same spectral signature. COVID was a clean rupture from extreme calm to trend dominance. The Gulf War was a sustained deepening of existing trend coupling. The GFC was a complex event that polarised markets rather than aligning them.
What Changes, What Stays
The permanent coupling finding reshapes how we think about several phenomena that the industry treats as separate.
Diversification failure. The standard story is that correlations spike during crises. Independent markets suddenly become connected. This is wrong. The coupling is always present. What changes is the spectral position. During the noise zone, coupling produces offsetting returns across asset classes. Diversification appears to work. When the system shifts to coupled trending, the same permanent coupling produces co-directional moves. Diversification appears to fail. The coupling did not change. The spectral position did.
Regime changes. The industry uses the term “regime change” to describe periods when market behaviour shifts suddenly. The spectral framework provides a precise vocabulary for this. A regime change is a rapid shift in the spectral position of the permanently coupled system. The earthquake analogy describes the mechanics: rupture, aftershock, settling. This is not metaphor. The three phases are visible in the data for every major market event in four decades.
Correlation instability. Portfolio managers spend enormous effort modelling correlations that shift through time. The spectral framework explains why they shift: the permanently coupled system is moving along the spectrum, and the surface correlations are a consequence of the spectral position, not an independent phenomenon. The correlation between any two assets is a downstream measurement. The spectral position of the coupled system is the upstream driver.
What This Means
If you manage money, allocate capital, or think about market structure, here is what this episode has established.
Markets are permanently coupled. The cross-sectional dispersion of autocorrelation is constant across all spectral states. The majority-sign fraction exceeds independence seventy-four percent of the time. Coupling is not an event. It is a property.
Directional memory is regime-specific, not absent. The full-sample zero conceals opposing directional structures inside each feedback state. Practitioners who treat direction as unpredictable are describing the average. The system underneath it is anything but.
The coupling expresses itself along a spectrum. At one end, positive feedback dominates and markets trend together. At the other, negative feedback dominates and markets oscillate together. In the middle, the forces balance and the coupling produces apparent independence.
Transitions between spectral states follow earthquake mechanics. Rupture, aftershock, settling. The pattern is visible in every major market event in four decades.
The spectral position, not the coupling itself, is what changes. This means that the questions practitioners ask about correlation, diversification, and regime change are downstream questions. The upstream question is: where is the coupled system sitting on the spectrum? Episode 3 shows that this question has a direct, measurable consequence for portfolio construction.
Next
Episode 2 established that sixty-eight markets form a permanently coupled system whose spectral position shifts between trending and oscillating behaviour. Episode 3 asks the question that matters most to anyone holding equities: what does this coupling mechanism mean for the relationship between trend-following and equity portfolios? The answer, it turns out, reframes the most commercially important finding in managed futures.
The coupling is permanent. Episode 3 shows what it produces.
Endnotes
Methodology
- Heatmap construction: rolling 504-day ACF(1) computed for each of sixty-eight contracts at 21-day steps (monthly), producing a matrix of 68 rows by approximately 474 columns spanning 1986 to 2026. Colour mapping: positive ACF (trending) in blue, negative ACF (oscillating) in red, near-zero in black. The transition between colours is continuous, not binary.
- Cross-sectional mean ACF: at each date, the arithmetic mean of all available contract-level rolling ACF(1) values. The standard deviation of this cross-sectional mean series is 0.0173. States classified using one-sigma bands: coupled trending (mean ACF > +0.0173), coupled mean-reversion (mean ACF < -0.0173), noise zone (within band). Fraction of time in each state: coupled trending 30.9%, noise 63.0%, coupled mean-reversion 6.1%.
- Cross-sectional dispersion: the standard deviation of ACF values across all available contracts at each date. Mean dispersion: coupled trending 0.0710, noise zone 0.0719, coupled mean-reversion 0.0710. The constancy of dispersion across states is the primary evidence for permanent coupling. Under independence, dispersion should increase during the noise zone as markets scatter across the spectrum.
- Majority-sign fraction: at each date, the fraction of contracts with ACF of the same sign as the majority. Under true independence with p = 0.5, the expected majority fraction for N = 65 markets is approximately 0.5 + sqrt(1/(2*pi*N)) = 0.549. Observed average: 0.599. Observed range: 0.500 to 0.823. Fraction of months exceeding independence prediction: 74.0%.
- Cross-market return correlation by spectral state: using ten representative contracts (ES, TY, GC, CL, EC, S, AD, HG, W, NK), rolling 63-day average absolute pairwise correlation was computed and merged with the spectral state classification. Mean return correlation: MR-leaning 0.311, balanced 0.240, trend-leaning 0.241. The elevated correlation during oscillatory periods confirms that markets are correlated in their oscillation, not just in their trending.
- COVID transition: cross-sectional fraction of markets with negative ACF, monthly. Feb 2020: 82% (peak oscillatory coupling). Mar 2020: 59%. Apr 2020: 46%. May 2020: 42%. The transition from 82% to 46% in two months is the fastest spectral shift in the sample.
- Gulf War period: fraction of markets with positive ACF. 1989-1990: 60-65%. Peak: Feb 1992 at 81% (highest trend coupling in dataset). Decay to balanced levels by 1994.
- GFC period: average oscillatory fraction during crisis (2007-2009): 55.9%. This is lower than the pre-crisis calm (2005: 67.9%) because the crisis polarised markets rather than uniformly shifting them to one spectral end. Noise-zone fraction collapsed from ~45-50% pre-crisis to 20-23% during the crash, confirming market polarisation.
- Trailing window effect: the rolling 504-day ACF at date X reflects data from X minus 504 trading days to X. At September 2008, this window spans approximately September 2006 to September 2008, capturing mostly pre-crisis data. The full crisis impact does not dominate the ACF until approximately mid-2009 to mid-2010.
Data
- Same dataset as Episode 1: sixty-eight CSI ratio-adjusted continuous futures, September 1984 to January 2026. Returns computed as daily log returns. Rolling windows: 504 trading days, 21-day steps.
Figures
- Figure 2.1: Full heatmap of rolling ACF(1) for all sixty-eight contracts, 1986-2026. Blue = positive feedback (trending), red = negative feedback (oscillating), black = balanced. Rows sorted by asset class.
This research series is drawn from The Fractals of Finance: Determinism, Adaptation and the Geometry of Markets
The book explores the full architecture of feedback, fat tails, and fractal structure in financial markets, and what it means for how we trade, invest, and understand risk.
Available now on Amazon in paperback, hardcover, and Kindle.
Want a practical field manual for trading trends and capturing outliers?
The Aussie Turtles Trend Following Guide: A Field Manual for Hunting Outliers adapts the timeless principles of the original Turtle traders into a systematic, rules-based approach for modern markets. Co-authored with Adam Havryliv.
Available now on Amazon in paperback, hardcover, and Kindle.