What Diversification Actually Does (and What It Cannot Do)
Classical diversification measures correlation and hopes it holds. The Outlier Hunter pursues structural independence, because correlation is the one thing you cannot rely on when you need it most.
The programme positioned in ten markets has ten chances to be present when the next outlier arrives. The programme positioned across a hundred has a hundred. Foundation 2 closed on that line. Foundation 3 begins where it leaves off, by asking what those chances actually are, why more of them is structurally better in a fractal market, and what diversification can and cannot do once it stops being a comfort mechanism and starts being an architecture.
A river delta does not flow toward the sea by choosing its channels. It arrives there by splitting.
From the air, the pattern looks abundant and almost chaotic. Dozens of separate streams spreading across the land, each finding its own path, each carrying a fraction of the total flow. Yet the aggregate movement is purposeful. The delta does not reach the sea faster by concentrating into a single powerful channel. It reaches it more reliably, through more varied terrain, resistant to blockage in any one path, because the flow is distributed.
The Outlier Hunter’s relationship with diversification works the same way. Not as a defence against loss, but as a structural property of how the programme reaches its objective.
That objective is the outlier. The fat-tail event that arrives unexpectedly in one corner of the market landscape and produces a return that dwarfs everything accumulated through the surrounding noise. The programme does not know in advance which market will produce it, which year it will arrive, or how long the current drought will last before it comes. What the programme can control is how many channels are open and waiting when it does.
The Conventional Story and Why It Is Wrong
The conventional story about diversification comes from Modern Portfolio Theory. Spread your capital across uncorrelated assets and you reduce your portfolio’s volatility, because losses in some positions are offset by gains in others. The result is a smoother ride to the same expected return. Risk falls. The Sharpe ratio improves, a conventional measure that rewards smoothness but cannot distinguish between beneficial and destructive volatility. The portfolio becomes more palatable to allocators who measure risk as volatility and reward smooth equity curves.
This story is not false. The mathematical mechanics it describes are real. But for the Outlier Hunter it is the wrong story, because it frames diversification as a comfort mechanism. Something that makes the experience of holding the portfolio more bearable by reducing its roughness.
The Outlier Hunter’s case for diversification is different in kind, not just degree. It is not about reducing the volatility of what already exists in the portfolio. It is about increasing the probability of being present when the rare event that drives the entire long-run performance finally arrives.
True risk is not volatility. True risk is missing the outlier because it was not in your universe.
“The danger is not that the programme will be too volatile. The danger is that the outlier will arrive in a market you are not trading.”
Fractal Markets and the Myth of Dilution
The objection most commonly raised against maximum diversification is that adding more markets dilutes the impact of each individual trade. If the programme is already positioned in crude oil when cocoa produces its extraordinary move, the cocoa position is a smaller fraction of the total portfolio. The outlier’s contribution is diminished.
This objection is correct in a Gaussian world, the neat world of classical finance where returns cluster around an average and extreme events are statistical rarities. In that world, adding more positions does reduce the impact of any individual result, because the results are approximately interchangeable. More bets means the average asserts itself more quickly. The outlier is noise.
Markets are not Gaussian. They are fractal systems governed by power laws, where the distribution of returns has no natural scale. At every level of observation, from individual trades to annual results to decade-long track records, a tiny number of events produce outcomes that dwarf everything else. The Pareto principle is not a metaphor for markets. It is a structural description of how returns are distributed across them.
In a fractal system, the outlier is not diluted by adding more positions. It is amplified by being added to a portfolio that, without it, would have produced far less. The outstanding move in cocoa in 2024, the extraordinary trend in energy in 2020, the sustained currency moves of 2022. Each of these events, captured by a programme that was already positioned and holding, lifted the entire portfolio’s long-run geometric return in a way that no number of ordinary trades could replicate. The outlier is not one outcome among many. It is the engine that makes the many worthwhile.
Adding more markets to the programme does not reduce the outlier’s impact. It increases the number of possible origins from which the next outlier can emerge. That is not dilution. That is how the programme is designed to work.
Variance Drain and the Geometry of Compounding
There is a second reason for maximum diversification that operates through the mathematics of geometric compounding rather than through the mechanics of outlier capture. It concerns what is called variance drain.
The geometric return of a portfolio, the rate at which capital actually compounds, can be approximated as the arithmetic average return minus roughly half its variance. The relationship is not exact. It holds well for typical levels of volatility and breaks down somewhat at the extremes. But the directional message is robust. Higher variance imposes a real cost on the rate of compounding, and that cost accumulates year on year. A programme that runs hotter than necessary pays for that heat in the long-run geometric return, even when its average return on individual trades looks identical to a calmer programme.
Diversification across genuinely independent return streams reduces the portfolio’s total variance relative to the sum of the individual variances. Each stream added to the portfolio that is not perfectly correlated with what is already there reduces the combined variance below the weighted average of its components. The portfolio becomes less volatile in aggregate than any of its parts.
The direct result is a higher geometric return on the same underlying set of strategies. The programme is not making more money on individual trades. It is losing less to variance drag between them. Over a decade, the difference between a portfolio with well-managed variance and one with poorly-managed variance is not marginal. It is the difference between materially different levels of accumulated capital, from programmes that entered the decade with the same signals and the same starting capital.
Three Levels at Which Diversification Operates
In practice, diversification for an Outlier Hunting programme operates simultaneously at three levels, and confusing them produces a programme that is diversified in name but concentrated in exposure.
The first level is across markets. Trading futures across energy, metals, agricultural commodities, currencies, interest rates, and equity indices. Each market has its own supply and demand dynamics, its own participant base, its own relationship between noise and trend. A move in crude oil driven by geopolitical disruption is structurally independent of a move in Japanese government bonds driven by central bank policy. Being present in both is not redundancy. It is coverage.
The second level is across time frames. Running systems with different lookback periods against the same markets. A system that defines a trend over twenty days captures different moves from one that defines it over two hundred. The return streams from the two time frames have low correlation even when applied to the same instrument, because they respond to different structures within the same price series. A programme running only one time frame is not diversified across time. It is a single lens applied to many markets.
The third level is across systems. Using multiple entry and exit rule sets so that the programme is not entirely dependent on any single signal generating its historical pattern of results. Different entry rules fire at different points in the development of a trend. Different exit rules hold positions for different durations. The combined return stream is less sensitive to the specific mechanical properties of any individual rule, and more reflective of the underlying structural tendency of markets to trend.
Each level is necessary. A programme diversified only across markets but running a single system on a single time frame is far more concentrated than it appears. The three levels compound each other. More markets times more time frames times more systems produces the maximum number of independent chances to be present when the outlier arrives.
Long and Short Are Not Symmetric
There is a fourth structural feature of how a divergent programme should be constructed, one that the three levels of diversification address only implicitly. It concerns the asymmetry between the long side and the short side of the programme.
A symmetric trend following system applies the same lookback to both directions. The same rule that defines a long entry, mirrored, defines a short entry. The same exit rule, mirrored, defines the close. This is mathematically clean and, for a long time, was the default assumption in systematic trend following. The empirical evidence now shows it is wrong.
Bull markets and bear markets do not operate at the same frequency. Across two centuries of US equity data, bull markets have averaged roughly forty-six months in duration. Bear markets have averaged fourteen. The median drift in bear markets has been fifty to one hundred percent larger in absolute terms than during bull markets. Prices take the escalator up and the elevator down. This asymmetry is not a statistical accident. It is a structural consequence of how the global capital base is composed.
The bulk of capital in the world’s markets is structurally long. Pension funds, sovereign wealth funds, endowments, passive index funds, retail investors through retirement accounts, and corporate balance sheets are all long-biased participants. The ratio of long-biased capital to short-biased capital in equity markets sits at twenty to one or higher. During uptrends, this population is comfortable holding, and selling pressure builds slowly through profit-taking, rebalancing, and value rotation. The trend rises gradually because the convergent force pushing against it is dispersed and patient. During downtrends, the same population becomes a coordinated source of forced selling. Margin calls trigger across leveraged longs simultaneously. Stop losses fire across systematic strategies in cascade. Risk management systems hit drawdown limits at thousands of institutions in the same week. The decline accelerates because the convergent force is now overwhelming and synchronised.
A symmetric model, calibrated to a single frequency, is permanently misaligned with at least one side. If the lookback is set for the slower bull-market frequency, the short signals fire too late and exit too late, missing the speed at which bear markets actually develop. If the lookback is set for the faster bear-market frequency, the long signals whipsaw, exiting good trends prematurely and entering on noise.
The empirical evidence supports asymmetric calibration directly. A grid search across long and short lookback combinations, run on a forty-year, sixty-eight-market futures sample, finds that the best risk-adjusted performance pairs long lookbacks of roughly two hundred to three hundred days with short lookbacks of ten to thirty days. The optimal ratio is approximately twenty-five to one. Shorter short lookbacks dominate across virtually every long pairing on the surface. This is not a marginal optimisation result. It is a structural feature of the data, consistent with the asymmetric agent population that produces the price series in the first place.
The implication for the Outlier Hunter is significant. Asymmetric calibration is not an exotic refinement to be considered after the symmetric programme is established. It is foundational. A programme that runs symmetric long and short signals is operating with a structural defect that no amount of additional market or time-frame diversification will correct, because the defect is not in the breadth of the universe. It is in the calibration of the signals being applied to it.
There is one further consequence worth naming clearly. The post-2020 era has seen the apparent collapse of standalone short-side returns under symmetric models. Central bank intervention, passive flows, and systematic dip-buying have structurally compressed the standalone alpha of trading shorts as a profit-generating activity. But this does not mean shorts can be removed from the programme. The short side carries the negative bear-market correlation, the convex tail payoff during sustained declines, and the regime robustness that prevents catastrophic outcomes when long-only portfolios are exposed. Removing shorts to chase recent performance turns a divergent programme into a long-only proposition that has lost its portfolio reason for existing. The correct response to short-side impairment is asymmetric calibration that reduces the cost of carrying the short side, not the elimination of the short side itself.
For the depth of the empirical case underlying this section, including the full grid search and the historical evidence on the asymmetric agent population, see The Fractals of Finance research series, particularly the episodes on the paradox and on the escalator and the elevator.
“Maximum diversification is not the ambition of a cautious programme. It is the ambition of one that understands where its edge actually lives.”
What Diversification Cannot Do, and What the Outlier Hunter Does Instead
The correlation problem runs deeper than a single crisis observation. In a complex adaptive system, correlation is not a stable property that occasionally breaks down under stress. It is a dynamic property that reflects the current distribution of participant behaviour, crowding, and leverage.
Under normal conditions, participants are heterogeneous. Different time horizons, different risk tolerances, different information sets. That heterogeneity produces the low measured correlations that conventional portfolio construction relies on.
Under stress, heterogeneity collapses. Forced selling dominates every decision regardless of the seller’s normal strategy. The mechanism driving prices in each market is no longer the varied set of structural forces that produced independence during calm conditions. It is a single force. The need to raise cash. Correlations do not drift toward one during a crisis. They snap there, rapidly, as the cascade of forced liquidations propagates across asset classes that were statistically unrelated the day before.
The Outlier Hunter does not take measured correlation seriously as a portfolio construction tool for this reason. A correlation matrix built on historical returns describes how markets moved together during the period in which participant behaviour was heterogeneous. It says nothing reliable about how they will move together during the period when it matters most, which is precisely the period when that heterogeneity collapses.
The response is to pursue structural independence rather than statistical independence. Two return streams are structurally independent when the forces that generate them are genuinely different in kind, not merely different in recent history.
A sustained trend in agricultural commodities driven by supply disruption and a sustained trend in fixed income driven by central bank policy are structurally independent because their generative mechanisms have no common cause. A portfolio of equities spread across twenty geographies is not structurally independent because its generative mechanism, the global risk appetite of institutional participants, is shared across all twenty. The geographic spread produces statistical independence during normal conditions and reveals its structural dependence the moment conditions become abnormal.
System diversification reinforces this. Different entry and exit rule sets applied to the same markets produce return streams whose correlation is lower than the correlation of the underlying price series, because they respond to different structural features of the same market at different moments. The combined return stream is less sensitive to the correlated liquidation event that destroys conventional diversification because it is capturing independent structural tendencies rather than variations on the same underlying exposure.
The Outlier Hunter’s diversification is therefore not a portfolio construction technique borrowed from conventional finance and applied more broadly. It is a fundamentally different objective. Maximum coverage of genuinely independent structural tendencies, so that the programme is present wherever the next outlier originates and is not destroyed by the correlation cascade that will eventually accompany it.
Correlation in the Price Series, Independence in the Outliers
One further point from the site’s own research on this topic is worth making explicitly, because it contradicts the most common reason practitioners give for limiting their market universe.
The assumption is that trading highly correlated markets, Brent crude and WTI crude for instance, or two closely tracked equity indices, adds nothing to the portfolio because the price series move together. This assumption looks at price correlation and concludes that trade correlation must be similarly high.
It is wrong, and the reason it is wrong cuts to the heart of what an Outlier Hunting programme actually is.
The conventional analysis treats a correlation coefficient between Brent and WTI as a measure of how the two markets behave. Over multi-year windows, that coefficient is typically high, often 0.95 or above on daily returns. The implication, in the conventional frame, is that exposure to one is functionally exposure to the other. Adding both to the portfolio is redundant.
This analysis would be correct if the programme were continuously exposed to the full price series of each market. It is not. The Outlier Hunter is not in these markets all the time. The programme is in them only during the structural episodes that develop into fat-tail moves, the periods when a market enters extreme tail territory and produces the kind of return the strategy exists to capture. Those episodes occupy a small fraction of any market’s total history.
The relevant question, therefore, is not how Brent and WTI prices correlate across the full data set. It is how the outliers in Brent and the outliers in WTI correlate with each other. Those are very different numbers.
Outliers are, by their statistical nature, rare and structural events. Two outliers in correlated markets coincide far less often than their price series would predict. The Brent move that breaks out of multi-year compression in March may be followed by a WTI move that does the same in November, driven by partially different supply dynamics, geopolitical inputs, or refining margins. The full price series remained tightly correlated through both events. The outlier exposures of a programme trading both markets were not. The programme captured one outlier in March and a separate outlier in November, and the contribution of each to the year’s geometric return was independent of the other.
This is the correlation that matters for an Outlier Hunting programme. Not the correlation of the markets, but the correlation of the moments within those markets that actually contribute to the P&L. Because the programme draws its returns from the tails rather than from continuous exposure to the body of the distribution, the diversification benefit of trading correlated markets is structurally larger than any analysis of the price series could reveal.
The argument is reinforced at the system level. Different systems applied to highly correlated markets produce trade distributions that are meaningfully less correlated than their price series would suggest, because different entry rules fire at different points in the two series and different exit rules close the two positions at different moments. A programme running multiple systems against both Brent and WTI captures further independence beyond what the outlier frame alone provides. The two effects compound. The outlier-level independence is the deeper structural reason. The trade-level independence from system diversity is the additional gain layered on top.
This is why maximum diversification means maximum. Not the markets that appear independent on a correlation matrix, but every liquid market where the structural tendency to trend can be captured and the execution costs of doing so are justified. The programme does not choose between Brent and WTI on the grounds of price correlation, because price correlation is the wrong measurement. It trades both, because the edge applies to both, the outliers in the two markets coincide far less often than their prices, and the systems running over those markets produce further independence at the trade level.
The conventional analysis answers the wrong question. The correct question is which markets contribute outliers, and the answer is that every market with a genuine structural tendency to trend will eventually contribute one. This is the deepest argument the essay carries, and the one to keep when the others fade. The programme’s job is to be there when the outlier arrives.
What Comes Next
Diversification, in this framework, is not the search for comfort. It is the search for presence. Presence across markets, across time frames, across systems, across signal directions, and across structural sources of return. The Outlier Hunter cannot know where the next fat-tail event will begin, but it can design the programme so that as many doors as possible are open when it does. That is the entire case, and everything in this essay was a way of making it from a different direction.
Foundation 2 addressed how much to risk on each position. Foundation 3 has addressed how broadly to distribute that risk across the market landscape. Both are portfolio construction questions, answered structurally rather than by optimisation.
The next question is different in character. Foundations 2 and 3 concern the architecture of the programme. Foundation 4 concerns the nature of the edge the architecture is designed to capture. Specifically, what edge actually is, how most traders measure it incorrectly, and why the Outlier Hunter’s relationship with expectancy is different from the one most systematic traders carry.
The answer begins with a formula that feels scientific and is, in the context of markets, a dangerous fiction.
READ DEEPER
→ Casting a Wider Net: The Power of Maximum Diversification for Outlier Hunters
→ Fractals, Diversification, and the Myth of Dilution
→ The Dangers of Diversification: When Spreading Risk Kills Your Edge (Unless You Do It Right)
→ Diversification for Trend Following Models: The Small Variations Matter
→ The Fractals of Finance: The Paradox
→ The Fractals of Finance: The Escalator and the Elevator
Previous: Foundation 2: Why Position Sizing Is the Most Important Decision You Make | Next: Foundation 4: Expectancy, Edge, and Why Most Traders Think About This Wrong
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.
Lorem ipsum dolor sit amet, consectetur adipiscing elit. Ut elit tellus, luctus nec ullamcorper mattis, pulvinar dapibus leo.Lorem ipsum dolor sit amet consectetur adipiscing elit dolor