The Vault

THE FRACTALS OF FINANCE | Research Series Phase 2 | Episode 7 of 9

The Paradox

The feedback structure persists. Simple trend-following returns do not. This paradox is the most important finding in the series. Its resolution is more nuanced than anyone expected.

This episode was supposed to be a celebration. Six episodes of evidence establishing a persistent, structural, asset-class-spanning system of positive and negative feedback. The logical conclusion: divergent strategies should work, have always worked, and will continue to work. The case for trend has never been stronger.

Then we built the strategy and ran the numbers.

Figure 7.1 Left: feedback amplitude stable or rising across four decades. Right: strategy performance declining. The structure persists. The simple strategy does not.

The left panel is the feedback amplitude from Episode 5: stable or slightly rising across four decades. The right panel is the performance of a simple diversified trend strategy applied to the same sixty-eight markets, declining substantially from the first decade to the most recent.

Over the full sample, trend delivers an annualised return of 3.3 percent with a maximum drawdown of 6.0 percent, producing a MAR ratio of 0.54. The S&P 500 delivers 4.5 percent return with a 54.2 percent maximum drawdown, a MAR of 0.08. The MAR advantage is nearly seven to one: trend earns seven times as much return per unit of worst-case pain as a buy-and-hold equity investor endures. These figures reflect a strategy vol-scaled to a 10% annualised volatility target, a conservative benchmark chosen for comparability across sixty-eight markets. A manager running higher volatility would scale returns proportionally; the MAR ratio, which measures return per unit of worst-case drawdown, is invariant to that choice and is the meaningful comparison here. But this full-sample picture conceals a trend in the wrong direction.

The sub-period analysis that follows uses the Sharpe ratio for regime-by-regime and horizon-by-horizon comparisons, because computing maximum drawdown within short sub-periods produces unstable estimates. The Sharpe ratio treats upside and downside volatility equally, which penalises strategies with positively skewed returns like trend-following. But the bias is consistent across periods and strategies, which means the comparative analysis is valid even if the absolute levels are conservative.

These two charts sit side by side. The structure that produces trending and oscillating behaviour has not decayed. The strategy that harvests trending behaviour has. This is not a contradiction. It is the most important finding in the series.

The Story Everyone Tells

The standard narrative goes like this. Trend-following worked brilliantly in the 1980s and 1990s. Then systematic capital flooded in, crowding compressed the edge, and performance declined monotonically. The golden era is over. Simple trend is dying a slow death from competition.

This narrative is wrong. Or rather, it is incomplete in a way that leads to the wrong conclusions.

We tested it comprehensively. We ran twelve different divergent strategies across the full universe: six lookback horizons from twenty to five hundred days, using both moving-average crossover and Donchian breakout signals. We decomposed performance by policy regime, by asset class, by the long and short sides independently, and by the variance ratio from Episodes 1 through 5. What emerged is a more complex and more honest picture.

Not Monotonic

Figure 7.2 Performance by horizon and macro regime. The decline is not monotonic. The QE era suppressed everything. Post-2020, longer horizons recovered substantially.

When we split performance into three macro regimes, pre-QE, the QE era, and post-2020, the monotonic decline narrative collapses. Every strategy suffered during the QE and ZIRP period from 2009 to 2020. But post-2020, the longer-horizon strategies recovered substantially. The 300-day Donchian breakout delivered a Sharpe of 0.74 in the post-2020 period. The 200-day Donchian reached 0.64.

During the 2021 to mid-2022 inflation build, every strategy in the test recovered to pre-QE quality. The 100-day Donchian reached a Sharpe of 1.50. The 300-day Donchian hit 1.50. The 200-day moving average delivered 1.45. When the macro environment provided genuine trends, driven by rising inflation, commodity supply shocks, and central bank tightening, the feedback structure fired and every divergent strategy harvested it.

The lost decade was not structural death. It was regime suppression. QE and ZIRP flattened the landscape that divergent strategies require. When the landscape returned, so did the edge.

The Short Side

If the decline is regime-dependent rather than structural, what specifically did the QE era suppress? The answer is precise and consequential.

Figure 7.3 Long vs short side decomposition. The long side dipped and recovered. The short side collapsed and has not returned. Post-2020, 93% of returns come from the long side.

We decomposed the 200-day MA strategy into its long-side and short-side contributions independently across all three regimes. The finding is stark.

The long side has been remarkably stable. It produced a Sharpe of 1.49 pre-QE, dipped to 0.59 during the QE era, and recovered to 0.67 post-2020. This is a regime-driven cycle, not structural erosion. The long side of trend-following is alive.

The short side appears dead. It produced a Sharpe of 0.76 pre-QE, turned negative at minus 0.08 during the QE era, and has barely registered at 0.05 post-2020. Pre-QE, the long and short sides contributed sixty-nine and thirty-one percent of total returns respectively. Post-2020, the split is ninety-three to seven. Under a symmetric model, the strategy has become a long-only proposition with short-side decoration.

The “trend-following decline” is almost entirely a short-side story. Central bank backstops, passive fund flows, and systematic dip-buying have structurally impaired short-side standalone alpha. However, short-side standalone alpha and short-side portfolio utility are different things. The crisis alpha documented in Episode 3, the negative bear-market correlation that defines trend’s value to equity portfolios, is produced by the short side. Remove shorts entirely and a trend portfolio becomes long-only, losing the correlation structure, the convex tail payoff during crises, and the regime robustness that prevents catastrophic outcomes during sustained bear markets. The short side may deliver negligible standalone returns. It provides the asymmetric payoff that makes trend investable as a portfolio component.

The Feedback Connection

Episodes 1 through 5 established the spectral structure: positive and negative feedback alternating across four decades. Episode 5 showed that the oscillation amplitude has not decayed. But we never tested whether the structure’s output, measured as the raw variance ratio itself, varies by regime in a way that explains strategy performance.

It does.

Figure 7.4 Variance ratio by sub-period. 2020–2022: the only period since the 1980s with VR above 1.0. 2023–2026: the strongest oscillating reading in the sample. The VR explains about a third of strategy variation.

The 2020 to 2022 inflation period produced a VR of 1.042, the only reading above 1.0 since the 1986 to 1995 period. This is the variance ratio telling us that positive feedback was dominant: trends were persisting, multi-period variance was growing faster than linear. And this is exactly the period when every divergent strategy recovered to pre-QE quality.

The correlation between annual VR and annual strategy Sharpe is 0.38 for the 200-day MA. The VR tells you whether the raw material for trends exists. It does not tell you whether crowding, transaction costs, or signal choice will capture or miss them. The feedback structure explains about a third of strategy variation. The rest is implementation.

The Horizon Gradient

Within any regime, shorter lookback horizons deliver less edge than longer ones. The twenty-day Donchian is dead in every regime. The 300-day Donchian thrives when conditions allow. This gradient is real and it has steepened over time.

Figure 7.5 The horizon gradient. Short horizons are dead. Long horizons recovered. The correlation between them has been declining, meaning ensemble diversification benefit is increasing.

There is a silver lining. The correlation between the shortest and longest horizons has been declining, from a pre-QE mean of 0.29 to a post-2020 level of 0.20. They have become increasingly independent return streams. The ensemble case for combining multiple horizons is stronger than ever.

Where Trend Works Now

Figure 7.6 Post-2020 VR by asset class. The recovery is selective. Grains and fixed income returned to trending. Metals collapsed to the most extreme oscillating reading in the sample.

The recovery is selective, not universal. The variance ratio explains which asset classes recovered and which did not. Grains shifted to a trending VR of 1.178 post-2020, and delivered strong divergent strategy performance. Fixed income and livestock also showed VR above 1.0. Metals collapsed to a VR of 0.588, the most extreme oscillating reading of any class in the sample. This is the fingerprint from Episode 6 operating in real time.

Figure 7.7 Diversified vs individual class returns. The asynchronous timing of spectral shifts across asset classes produces smoother portfolio-level returns than any single class.

Three Forces

The evidence supports three forces operating on divergent strategy returns, properly ordered by importance.

First, regime. This is the dominant driver. QE and ZIRP suppressed the variance ratio across the full universe, starving divergent strategies of the raw material they need. The lost decade from 2009 to 2020 was primarily a VR drought. When macro conditions delivered genuine trends post-2020, every strategy recovered. This force is cyclical, not structural, and it can reverse.

Second, short-side impairment. Central bank backstops, passive fund flows, and systematic dip-buying have severely impaired short-side standalone alpha under symmetric calibration. However, short-side portfolio utility remains mechanistically intact and requires maintaining short exposure.

Third, horizon gradient. Within any regime, shorter lookback horizons deliver less edge than longer ones. The edge has redistributed from shorter to longer horizons, and an ensemble that spans both provides diversification benefit that is increasing, not decreasing, over time.

The interaction of these three forces created the illusion of monotonic structural decline. The QE era was a triple penalty: the regime suppressed all strategies, the short side was structurally impaired, and the most visible horizons lost signal quality. Decade-level bucketing conflated all three effects and attributed the result to a single cause. The data tells a more honest story: one force is cyclical, one is structural, and one is a redistribution. The edge is not dying. It has relocated.

But there is a question embedded in these findings. The analysis above used symmetric models: the same lookback for long and short signals. Bull markets last three to five times longer than bear markets. Bear markets deliver intensity per unit time that is fifty to one hundred percent greater. Prices take the escalator up and the elevator down. If the two directions operate at different frequencies, the symmetric model is structurally misaligned with at least one side at all times. Is the short side dead, or is it invisible to a symmetric lens?

Episode 8 investigates.

Endnotes

Methodology

Strategy construction: 200-day simple moving average applied to each of 68 contracts. Long when price > MA, short when below. Daily returns vol-scaled to 10% annualised target using 63-day rolling volatility, capped at 3.0x. Portfolio: equal-weighted average of all available contract returns each day. No transaction costs, no slippage. Full-sample performance: Trend annualised return 3.3%, max drawdown 6.0%, MAR 0.54. S&P 500 annualised return 4.5%, max drawdown 54.2%, MAR 0.08. Donchian breakout strategies tested at 20, 50, 100, 200, and 300-day lookbacks with exit channel at half the entry lookback.
Three-regime decomposition (Sharpe ratios): Pre-QE (1986–2009), QE Era (2009–2020), Post-2020 (2020–2026). MA 200d Sharpe: 1.71 > 0.39 > 0.59. Don 300d: 1.61 > 0.28 > 0.74. Don 20d: 1.40 > 0.14 > 0.17. During the 2021–mid-2022 inflation build: MA 200d SR = 1.45, Don 100d SR = 1.50, Don 300d SR = 1.50.
Long/short decomposition: MA 200d long-side Sharpe by regime: Pre-QE 1.49, QE Era 0.59, Post-2020 0.67. Short-side: Pre-QE 0.76, QE Era −0.08, Post-2020 0.05. Return contribution: Pre-QE 69%/31%, Post-2020 93%/7%.
VR by sub-period: universe-mean VR(20) by sub-period: 1986–1995: 1.003, 1996–2005: 0.935, 2006–2009: 0.982, 2010–2014: 0.962, 2015–2019: 0.889, 2020–2022: 1.042, 2023–2026: 0.837. Annual VR(20) correlated with annual MA 200d Sharpe: r = 0.38.
Cross-horizon correlation (Don 20d vs Don 300d): Pre-QE 0.286, QE Era 0.070, Post-2020 0.197.
Post-2020 VR(3m) by asset class: Grains 1.178, Livestock 1.054, Fixed Income 1.040, Softs 0.981, Energy 0.861, FX 0.850, Equity Index 0.776, Metals 0.588.

Data and Figures

Same dataset as Episodes 1 through 6. 68 contracts, 8 asset classes.
Figure 7.1: side-by-side paradox.
Figure 7.2: V-shape by horizon and regime.
Figure 7.3: long vs short decomposition.
Figure 7.4: VR by sub-period.
Figure 7.5: horizon gradient and cross-correlation.
Figure 7.6: asset class VR recovery.
Figure 7.7: diversified vs individual class cumulative returns.

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.