Diversification works when you don’t need it and fails when you do. Trend following does the opposite. It performs when everything else is falling apart
During the Global Financial Crisis, the S&P 500 lost 50.9%. The TF Index gained 40.3%. The spread was 91 percentage points. During the dot-com bust, the S&P lost 38.8%. The TF Index gained 53.9%. The spread was 93 percentage points. During the 2022 inflation bear market, the S&P lost 24.6%. The TF Index gained 21.7%. The spread was 46 percentage points.
Episode 7 revealed that both equities and trend following depend on outlier months for terminal wealth. It showed that the trend following process produces a positively skewed distribution in which the right tail extends further than the left. It identified the mechanism: cutting losses truncates the left tail while letting winners run extends the right. And it noted, almost in passing, a fact that demands its own episode: seven of the TF Index’s ten best months occurred when the S&P 500 was negative.
This episode examines that fact. It is not a coincidence. It is the defining property of the trend following process in a portfolio context: the tendency to deliver its strongest performance during the periods of maximum geometric danger for equities. This property has a name. It is called crisis alpha. And its implications for compounding are, by any measure, extraordinary.
The Record
Five equity stress events have occurred since January 2000. Each represents a period in which the S&P 500 experienced a sustained decline of 15% or more. Each tested the geometric resilience of every portfolio that held equities. The following table records what happened.
The TF Index was positive in four of five crisis periods. In the one exception, the 2011 Euro crisis, it lost 4.0% while the S&P 500 lost 16.3%, still delivering a positive spread of 12 percentage points. Across all five crises, spanning 63 months, the aggregate cumulative return of the S&P 500 was −84.8%. The aggregate cumulative return of the TF Index was +159.5%.
Notice the columns for the other benchmarks. Berkshire Hathaway, with its enormous equity exposure, lost 38.1% during the GFC. The 60/40 portfolio lost 27.6% during the GFC and 13.5% during the 2022 bear. Even bonds, the traditional hedge against equity risk, lost 14.0% during 2022 as interest rates surged. Every conventional diversifier either failed or provided only modest protection. The TF Index was the only benchmark in the table that delivered double-digit positive returns during the two largest equity drawdowns in the dataset.
The same crisis that destroyed half of equity wealth created 40.3% of trend following wealth. This is not diversification in the conventional sense. It is geometric insurance that pays a positive premium.
Why Crises Are Trends
The crisis alpha property is not mysterious once the feedback architecture of markets is understood. Episode 5 described how trends emerge from reinforcing loops: local buying aligns into directional flow, liquidity withdraws from the other side of the order book, forced selling cascades through interconnected positions, and the original trigger becomes irrelevant as the feedback loop sustains the movement on its own.
Crises are the most powerful expression of this feedback architecture. During an equity crisis, the reinforcing loops do not merely persist. They intensify. Margin calls force liquidation. Liquidation drives prices lower. Lower prices trigger more margin calls. Risk models recalibrate and demand further selling. Liquidity providers widen spreads or withdraw entirely. Portfolio insurance and hedging activity amplify the decline. Each participant responds rationally to their local conditions, and the aggregate effect is a sustained, directional move of extraordinary magnitude.
A crash is a trend. It is a directional move produced by exactly the same endogenous feedback mechanisms that produce every other trend. The only difference is its magnitude and velocity. The trend following process does not distinguish between a gradual commodity uptrend and a violent equity collapse. Both are sustained directional movements. Both produce the geometry that the process is designed to capture.
During the GFC, the process cut long equity positions as the decline began. As the decline accelerated and established itself as a sustained trend, the process established short positions. Those short positions profited as the crash continued. The same month that inflicted the S&P 500’s worst loss, October 2008, −16.79%, was the TF Index’s third-best month, +10.92%. The process was not hedging. It was not predicting. It was responding to the directional structure of price, which during a crisis happens to point violently downward.
The 2022 bear market provides an equally instructive example because it refutes the argument that crisis alpha depends on equity short selling. The 2022 decline was driven by inflation, rising interest rates, and a simultaneous collapse in both stocks and bonds. The TF Index gained 21.7% during this period primarily through trends in commodities, energy, and interest rate markets, not through short equity positions. The crisis alpha property is not confined to equity short selling. It is a function of the process responding to directional movements across all markets, and crises tend to produce directional movements everywhere.
A crash is a trend. It is produced by the same feedback architecture that produces every other trend. The process does not need to know a crisis is happening. It only needs to follow the direction of price.
Correlation That Matters
The full-period correlation between the TF Index and the S&P 500 is −0.10. This is weakly negative, which by itself is useful for portfolio construction. But the full-period number conceals the property that matters most.
Correlation between the TF Index and equities is not stable. It oscillates across the full spectrum from meaningfully positive to deeply negative, and the direction of that oscillation is not random. It follows the spectral position of the market system. During sustained bull markets, when the coupled system is in a trending state driven by upward positive feedback, trend following participates in that directional persistence across the asset universe. The rolling 12-month correlation during these periods reaches as high as +0.67, and the bull-market average sits at approximately +0.13. Trend following is not working against equities in a rising market. It is harvesting the same directional momentum that is lifting equities, expressed across commodities, currencies, bonds, and equity indices simultaneously.
During equity crises, the spectral position shifts. Positive feedback still dominates the coupled system, but now in the downward direction. Equities fall persistently. Bonds rally persistently. Commodities and currencies reprice directionally. Trend following, responding to the directional structure of price across dozens of markets simultaneously, captures that directional energy wherever it appears. The rolling 12-month correlation plunges to −0.54 during the dot-com bust, to −0.49 during the GFC, and to −0.31 during the 2022 bear market. In bear markets the average settles at approximately −0.68. The same mechanism that produced +0.67 during the bull produces −0.68 during the bear. The process did not change. The spectral direction did.
Chart 8: Rolling 12-month correlation between the TF Index and the S&P 500. The correlation oscillates between +0.67 and −0.66 across the full period. During bull markets the average is approximately +0.13; during bear markets approximately −0.68. The same mechanism harvests directional persistence in both regimes: the spectral direction changes, the process does not.
This is the opposite of what conventional diversification provides. Most diversifying assets, including bonds, real estate, and international equities, maintain their diversification benefits during calm markets and lose them during crises. Correlations among traditional asset classes converge toward one during stress events. The diversification that existed on paper evaporates at the moment it is needed. The 60/40 portfolio is the most prominent casualty of this convergence. It assumes that bonds will rally when equities fall. During the GFC, this held partially: bonds returned +5.8% while equities suffered their deepest drawdown. But during 2022, when inflation drove both stock and bond prices downward simultaneously, bonds lost 14.0%. The 60/40 portfolio lost 13.5%. The assumed hedge was not merely weak. It was negative.
Trend following exhibits the opposite pattern. Its correlation to equities weakens, then inverts, during the precise periods when correlation matters most. The process becomes a better diversifier as the need for diversification increases. This is not a statistical anomaly. It is a mechanical consequence of the process: when equities fall in a sustained directional manner, the trend following process captures that direction, producing returns that are negatively correlated to equities by construction.
The Population Evidence: 32 of 41
The correlation structure is not unique to the TF Index. It is a population-level property of the trend following process.
Of the 41 trend following managers in the NilssonHedge database with track records exceeding 20 years, 32 have negative full-period correlation to the S&P 500. That is 78%. The median correlation across all 41 managers is −0.08. Only nine managers show positive correlation, and among those nine, the highest is +0.27. None approaches the kind of strong positive correlation that characterises most equity-adjacent strategies.
Chart 15: Full-period correlation to the S&P 500 for 41 long-term trend followers. 32 of 41 (78%) exhibit negative correlation. This is a structural property of the process, not a feature of individual manager positioning.
This is the second population-level finding in consecutive episodes. Episode 7 showed that 37 of 41 managers (90%) have positive skew. This episode shows that 32 of 41 (78%) have negative equity correlation. The overlap between these two properties is the statistical fingerprint of the trend following process: it produces returns that are positively skewed and negatively correlated to equities, and it does so reliably across dozens of independent implementations. The two properties are not coincidental. They arise from the same mechanism. The cut prevents the process from participating in equity crashes (producing the negative correlation), while the trend ride captures directional movement in all markets, including those that trend during crises (producing the positive skew and crisis alpha).
The Geometric Value of Crisis Alpha
Episode 2 established the convexity of the recovery curve: a 50% loss requires a 100% gain to recover. A 30% loss requires 43%. A 20% loss requires 25%. The deeper the drawdown, the more geometric work is needed to return to the previous peak. This convexity means that avoiding deep drawdowns is worth more, in compounding terms, than producing equivalent gains during calm markets.
The recovery cost column reveals why crisis alpha is geometrically transformative. An equity investor who suffers the GFC’s 50.9% drawdown needs the market to double from its trough just to return to the pre-crisis peak. Every month spent recovering is a month not spent compounding new wealth. The trend follower who gained 40.3% during the same period needs no recovery. They are 40.3% ahead of where they started, compounding from a higher base while the equity investor is still digging out of the hole.
Crisis alpha interacts with this convexity in a way that fundamentally changes the geometric arithmetic. A process that merely avoids losses during crises, that goes to cash or reduces exposure, protects the compounding engine by keeping it out of the convex destruction zone. That alone is valuable. But a process that produces positive returns during crises does something more powerful: it grows the compounding engine at the exact moment when the competition’s compounding engine is being destroyed.
Consider what the TF Index’s crisis months are worth in geometric terms. If the TF Index had earned zero during all 63 crisis months, producing neither gains nor losses, its terminal value would fall from $667,058 to $257,100. The crisis months account for 61.5% of terminal wealth. More than half of the process’s total geometric return was created during the 20% of months when equity markets were in crisis.
The second scenario in the table is even more revealing. If the S&P 500 had earned the TF Index’s returns during those same 63 crisis months, while keeping its own returns for the remaining 250 months, its terminal value would increase from $747,697 to $12,754,068. A 17-fold improvement. The S&P 500’s non-crisis returns are spectacular. Its crisis returns destroy the compounding of those spectacular months. Replacing just the crisis months with trend following’s crisis returns produces a portfolio that is geometrically unrecognisable.
This 17-fold improvement does not come from trend following’s non-crisis returns. It comes entirely from what happens during the 63 months of maximum geometric danger. It comes from the difference between losing half your capital and gaining 40.3%. From the difference between a compounding engine that is shattered and a compounding engine that is accelerating.
Replacing the S&P 500’s crisis returns with trend following’s crisis returns turns $747,697 into $12.75 million. The entire geometric difference lives in 63 months out of 313. Crisis alpha is not a feature of the process. It is the feature.
The Failure of Conventional Diversification
The standard framework for portfolio construction assumes that diversification across asset classes reduces portfolio risk. This assumption is correct during normal markets and catastrophically wrong during crises. The reason is that conventional diversification relies on correlations measured during calm periods, and correlations are not stable.
During the GFC, the 60/40 portfolio lost 27.6%. This was less than the S&P 500’s full GFC drawdown, so the bond allocation provided some cushion. But a 27.6% loss still requires a 38% gain to recover, and the portfolio was down at precisely the moment when every other asset in the investor’s life, their home, their job security, their confidence, was also under stress. The diversification benefit that survived was modest. The psychological cost was enormous.
During 2022, even this modest benefit vanished. The 60/40 portfolio lost 13.5% because bonds, rather than hedging the equity decline, amplified it. Rising inflation pushed interest rates higher, destroying bond prices alongside equity prices. The assumed negative correlation between stocks and bonds reversed. The diversifier became a co-conspirator.
The trend following process did not fail during 2022. It gained 21.7%. It did not depend on bonds to hedge equities, because it does not depend on any assumed relationship between asset classes. It responds to the direction of price in each market independently. When bonds fell, the process captured the bond downtrend. When commodities surged, the process captured the commodity uptrend. When equities declined, the process captured the equity downtrend. The diversification was not assumed. It was produced, in real time, by the process responding to whatever the markets were actually doing.
This is the fundamental distinction. Conventional diversification assumes that asset classes will behave differently during stress. Trend following does not assume anything about how asset classes will behave. It responds to how they actually behave. When the assumption holds, conventional diversification and trend following both provide protection. When the assumption fails, conventional diversification fails with it. Trend following does not, because it was never depending on the assumption in the first place.
The 60/40 portfolio assumes bonds will hedge equities. In 2022, bonds lost 14% alongside equities. The assumption failed. Trend following does not depend on assumptions about asset class relationships. It responds to what is actually happening.
Convergent and Divergent
The crisis alpha property clarifies a distinction that runs through the entire architecture of financial markets. Most investment strategies are convergent. They bet on stability. They profit when markets remain orderly, when correlations hold, when volatility stays contained, when prices revert to means. They extract small, frequent profits from the assumption that tomorrow will resemble today.
Trend following is divergent. It profits from change. It captures directional movement that departs from equilibrium and persists. It does not need markets to be orderly. It needs markets to move. And the largest, most sustained movements in financial history have occurred during crises, when the convergent strategies that depend on stability are being destroyed.
This is why crisis alpha is structural rather than coincidental. Convergent strategies and divergent strategies are, in a fundamental sense, on opposite sides of the same trade. When the convergent strategies fail, the capital they release, the forced liquidation, the margin calls, the panic selling, produces the sustained directional movements that divergent strategies capture. The trend follower’s crisis return is funded, in part, by the convergent strategies’ crisis losses. The feedback loop described in Episode 5 is the transmission mechanism. The convergent strategies’ forced selling amplifies the trend. The trend follower rides the amplified trend.
But this framing is incomplete if it suggests that trend following only earns during crises and merely endures the rest of the time. The coupled spectrum reveals a fuller picture. Trend following is not a crash insurance policy. It is a feedback harvester that aligns itself with whichever direction the dominant regime is running. During a sustained bull market, positive feedback produces upward directional persistence across many markets simultaneously. A divergent strategy captures that persistence just as it captures downward persistence during a crash. The mechanism is identical. The direction differs. The industry label “crisis alpha” is accurate but narrow: it names the most dramatic expression of a property that is operating at all times.
The practical implication is significant. An investor who frames the trend following allocation as “protection that costs during bull markets” is working with an incomplete model. During coupled trending in a bull market, the allocation contributes positively, with a weakly positive equity correlation that makes it genuinely additive rather than merely dilutive. During a crash, the same allocation flips to strongly negative correlation and delivers crisis alpha. The allocation is not switching between contributing and waiting. It is expressing the same underlying mechanism in whichever direction the feedback is running.
Every portfolio is, whether its owner realises it or not, a blend of convergent and divergent exposures. A portfolio that is entirely convergent will perform well during calm markets and catastrophically during crises. A portfolio that includes a divergent component has structural protection against the moments when convergent strategies fail. The divergent component does not merely reduce risk. It converts crisis into opportunity. It turns the geometric cost of a drawdown into the geometric fuel of a gain.
What Crisis Alpha Feels Like
The data on crisis alpha is clean and compelling. The experience of living through it is neither. This deserves honest discussion because the behavioural challenge of holding a divergent strategy is the primary reason most investors never capture the crisis alpha that the data promises.
Between crises, which is most of the time, the trend following allocation will underperform equities during sustained bull markets. It will generate whipsaws and flat returns while the S&P 500 compounds steadily upward. The investor will question why they are paying for a strategy that appears to be doing nothing. The temptation to reduce the allocation or eliminate it entirely will be intense, and it will feel rational. Every month of underperformance will feel like evidence that the strategy is broken.
Then the crisis arrives. The equity allocation loses 30%, 40%, 50%. The trend following allocation, quietly maligned for years, delivers the returns that rescue the portfolio. The investor who maintained the allocation is protected. The investor who cut it is not.
This is the fundamental challenge: crisis alpha is most valuable when it has been least visible. The years of patient holding, of enduring underperformance, of watching the equity market rally without participation, are the price of being positioned when the crisis arrives. The investor must accept the cost in advance and maintain the allocation through the long stretches when it appears to deliver nothing. There is no way to time the entry. There is no signal that says the crisis begins next month. The allocation must be permanent, or the crisis alpha is theoretical rather than actual.
The investor who holds trend following during bull markets and cuts it before the crash has paid the full cost of participation and captured none of the benefit. Crisis alpha requires permanent allocation. There is no shortcut.
The Running Ledger
Our $100,000 continues. This episode adds the crisis alpha dimension, revealing the moment-of-maximum-danger performance of each benchmark.
The GFC column encapsulates the crisis alpha argument. Every benchmark except the TF Index lost money, most of them catastrophically. The 60/40 portfolio, designed specifically to provide diversification during equity stress, lost 27.6%. Berkshire, run by arguably the greatest investor in history, lost 38.1%. The TF Index gained 40.3%. The S&P correlation column confirms the structural explanation: the TF Index is the only benchmark with negative correlation to equities.
The Bridge
Trend following produces positive returns during equity crises because crises are trends. The same feedback architecture that produces ordinary trends produces crisis trends, and the process captures them with the same mechanism it uses in every other market environment. But the story is larger than crises. The rolling correlation between trend following and equities oscillates from +0.67 to −0.68 not because the process is sometimes working and sometimes not. It oscillates because the coupled market system is always expressing directional persistence somewhere, and the process is always aligned with it. In bull markets that directional persistence runs upward; in bear markets it runs downward. Crisis alpha is the bear-market expression of feedback harvesting. Bull-market participation is the same mechanism in the opposite regime. The result is a return stream that is positively skewed, regime-responsive, and worth more in geometric terms than any other single property a portfolio component can possess.
The natural question that follows is: what happens when we combine these properties into a portfolio? If the trend following process produces positive skew, crisis alpha, and negative equity correlation, what does a blended portfolio of equities and trend following look like across the full 25-year dataset?
Episode 9 will build that portfolio. It will show that the combination of trend following and equities is not merely additive. It is geometrically synergistic: the blended portfolio produces a terminal value greater than either component alone, with a maximum drawdown smaller than either component alone. It will construct the geometric efficient frontier and demonstrate that the optimal allocation to trend following is not a small, token position. It is substantial, and the data will show exactly why.
Data and Sources
All performance data from the NilssonHedge Trend Following Performance Database (January 2000 to January 2026). All returns are net of management and performance fees. 313 monthly observations. Crisis periods defined as sustained S&P 500 declines exceeding 15%: dot-com bust (April 2000–October 2002), GFC (November 2007–February 2009), Euro crisis (May–September 2011), COVID (February–March 2020), and 2022 bear (January–September 2022). Cumulative returns computed as the product of (1 + monthly return) across each period. Correlations are Pearson product-moment coefficients of monthly returns. Rolling correlations use trailing 12-month windows. Bull-market and bear-market conditional correlations (+0.13 and −0.68 respectively) are drawn from the Fractals of Finance Phase 2 research series, which computed rolling one-year correlations between a diversified trend strategy across 68 futures markets and the S&P 500 from 1986 to 2026, classifying regimes by the sign of the trailing 252-day equity cumulative return. The rolling correlation range (+0.67 to −0.66) is the observed extremum of that rolling series. The 60/40 benchmark is the Vanguard Balanced Index (VBIAX, 60% equity / 40% bond). Bond returns are the Vanguard Total Bond Market Index Fund (VBMFX). The counterfactual S&P-with-TF-crisis-returns substitutes the TF Index’s monthly returns during the 63 identified crisis months while retaining the S&P 500’s returns for all remaining months. Crisis alpha concept formalised by Kathryn Kaminski in “Crisis Alpha” (2014). 41 managers met the 20-year track record threshold for the correlation analysis.
Want the theoretical foundation for why trend following works?
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.