Portfolio Construction
The markets were never independent. They belong to the same system. The purpose of diversification is not to pretend otherwise. It is to make sure no single expression of that system determines our fate.
A lightning conductor does not prevent the storm.
It does not stop lightning from striking.
And it certainly does not make the building independent of the electrical system surrounding it.
Quite the opposite.
The building is already inside the storm.
The conductor simply changes what happens when the connection becomes violent.
It provides a path through the structure that the structure has been built to survive.
That is a useful way to think about portfolio construction.
Markets are not isolated islands. Commodities, currencies, bonds and equities belong to the same global economic system. Capital flows between them. Policy changes propagate through them. Inflation connects them. Liquidity connects them. Expectations connect them. The participants trading one market are often responding to developments originating somewhere else entirely.
The system is coupled.
So the objective of diversification cannot be to find fifty completely independent markets and put them in a portfolio.
They do not exist.
The objective is more interesting.
We want as many meaningfully different ways as possible to encounter what the coupled system does next.
That is portfolio construction.
And for an Outlier Hunter, it is not primarily an exercise in reducing volatility.
It is an exercise in expanding opportunity without allowing any single expression of the system to dominate the programme.
Fifty Markets Can Still Be One Trade
Imagine a portfolio containing twenty equity indices.
S&P 500.
Nasdaq.
DAX.
CAC.
FTSE.
Nikkei.
ASX.
Hang Seng.
And another dozen from around the world.
Twenty markets.
Plenty of geographical diversity.
It looks impressive in a spreadsheet.
Then global risk appetite collapses.
Suddenly those twenty markets begin behaving rather like what they always were: different local expressions of a powerful shared global mechanism.
The tickers were different.
The dominant force was not.
This is the first mistake we can make when thinking about diversification.
We count instruments when we should be thinking about dependencies.
The same problem appears elsewhere.
Ten energy contracts are not necessarily ten independent opportunities.
Several government bond markets can become different expressions of the same inflation shock.
Multiple currencies can suddenly become manifestations of one enormous dollar move.
None of this means we should trade only one equity index, one bond, one currency or one energy market.
Quite the opposite.
Small differences matter.
But the number of lines in the portfolio is not, by itself, a measure of how diversified the programme really is.
Thirty markets can still become one trade.
Correlation Is a Shadow
The conventional way to solve this problem is correlation.
Measure how markets have moved relative to one another.
Prefer the low correlations.
Avoid the high ones.
Correlation is useful.
I use it.
But I do not ask it to tell me more than it knows.
A correlation coefficient is a description of an observed relationship over a particular historical sample.
It tells us what happened together.
It does not necessarily tell us why.
That distinction matters because markets are adaptive.
The forces dominating them change.
Relationships strengthen.
Relationships weaken.
Connections that appeared irrelevant become important.
Connections that dominated for years disappear into the background.
The correlation matrix changes because the system generating the matrix changes.
So correlation is not structure itself.
It is a statistical shadow cast by structure.
Sometimes the shadow tells us a great deal.
Sometimes it tells us remarkably little about what happens when the light moves.
This is why I am uncomfortable using historical correlation as the primary architect of a portfolio.
I want to know something deeper.
What makes these markets move?
Look Beneath the Price
Consider wheat.
Weather matters.
Crop yields matter.
Inventories matter.
Planting decisions matter.
War matters.
Transport infrastructure matters.
Physical supply matters.
Now consider government bonds.
Inflation expectations matter.
Monetary policy matters.
Growth expectations matter.
Fiscal policy matters.
Institutional demand for duration matters.
Capital flows matter.
Those are meaningfully different price-generating mechanisms.
Not independent.
Different.
That word is important.
A severe inflation shock can connect wheat and bonds.
A war can affect agricultural supply, energy, currencies, interest rates and equities at the same time.
Central bank policy can propagate through almost every financial market on Earth.
The world is a network.
We should expect connections.
What we want is diversity in the pathways through which those connections express themselves.
Now add currencies.
Relative interest rates matter.
Trade flows matter.
Capital flows matter.
Policy credibility matters.
Relative economic conditions matter.
Again there is overlap.
Of course there is.
But we have added another way for the global system to move.
That is the objective.
Not perfect independence.
Structural variety.
“Correlation tells us how markets moved together. Structure asks what made them move at all.”
The Markets Were Never Independent
This leads to an important change in language.
I do not think the best mental model is that markets are normally independent and occasionally become correlated during crises.
They were connected before the crisis.
The crisis merely changes which connections dominate.
That is a very different idea.
During ordinary conditions, wheat may be dominated by crop conditions while bonds respond primarily to monetary expectations.
The connection between them exists somewhere in the broader economic network, but it may be weak enough to matter little.
Then inflation surges.
Food prices contribute to the inflation impulse.
Central banks respond.
Rates rise.
Currencies adjust.
Capital moves.
Equities reprice.
What appeared to be separate markets reveal themselves as nodes in the same system.
Nothing magical happened to correlation.
The system changed state.
And with that change, different connections became important.
This matters enormously for portfolio construction because it destroys the fantasy that diversification can somehow place us outside the system.
It cannot.
We trade the system.
The best we can do is avoid making our survival dependent on one particular state of it.
Diversification Is Opportunity Amplification
This is where my view of diversification differs from the standard textbook presentation.
Usually diversification begins with reducing portfolio volatility.
Combine imperfectly correlated assets and the aggregate return stream becomes smoother.
Fine.
That is mathematically useful.
But it is not the primary reason I want breadth.
I want breadth because I have absolutely no idea where the next great trend will occur.
That is the real problem.
Suppose the defining trade of the next two years occurs in cocoa.
If cocoa is not in the portfolio, we miss it.
There is no compensation.
No partial credit.
No amount of sophistication in our bond model somehow captures the cocoa trend for us.
We simply weren’t there.
Or perhaps the next great move comes from Japanese bonds.
A currency nobody has cared about for years.
A grain market after a crop failure.
A metal suddenly caught in a supply squeeze.
A market everyone considered dead.
We don’t know.
That ignorance is not something I am embarrassed about.
It is the reason I diversify.
Every additional robustly tradable market places another sensor somewhere in the global economic system.
Most of those sensors will detect nothing extraordinary most of the time.
Good.
They do not need to.
One of them eventually will.
The Market That Has Done Nothing
This is where diversification becomes psychologically difficult.
There will always be a market in the portfolio that appears useless.
Probably several.
They generate failed signals.
They produce small losses.
Nothing interesting happens.
Year after year.
Eventually someone asks:
Why are we still trading this thing?
It is a perfectly reasonable question.
It is also being asked from the wrong end of time.
We now know that the market did nothing exceptional during the period we just observed.
We did not know that when the period began.
And we certainly do not know what happens tomorrow.
A market’s recent failure to produce an outlier is not evidence that it cannot produce the next one.
In fact, performance-based selection creates a particularly nasty trap for an Outlier Hunter.
Remove markets after they have been unproductive.
Add markets after they have produced large trends.
We end up systematically absent before the event and enthusiastic after it.
Exactly backwards.
The market everyone wants to remove may be the market we most regret removing.
Small Differences Matter
Now consider two markets that appear almost redundant.
Brent crude and WTI crude, for example.
Their prices are often highly correlated.
Why trade both?
Because we do not actually trade correlation coefficients.
We trade systems applied to price paths.
And the paths are not identical.
A small difference in price can determine whether a breakout occurs today or next week.
A different entry creates a different stop.
A different stop creates a different position path.
A retracement may exit one position while leaving the other intact.
The resulting strategy returns are therefore not the same object as the underlying price returns.
This distinction matters.
Price correlation is not trade correlation.
That does not mean highly correlated markets magically become independent when we run a trend-following system over them.
They don’t.
It means that redundancy should be judged at the level at which the portfolio actually operates.
Price relationships matter.
Strategy-return relationships matter.
Structural drivers matter.
Each tells us something different.
I am suspicious whenever one statistic claims to have solved the whole problem.
Markets are rarely that cooperative.
Diversification Has Three Dimensions
So far we have talked about markets.
But markets are only the first dimension of the search.
A trend has a location.
It also has a duration.
And it has a form.
That gives us three dimensions of diversification.
Markets: Where
Markets spread the search across the economic landscape.
Agriculture.
Energy.
Metals.
Currencies.
Rates.
Equity indices.
Different markets expose us to different manifestations of the global system.
This is where we search.
Time Horizons: When
Now imagine a trend that develops and disappears within six weeks.
A very slow system may barely recognise it.
A faster system may capture much of the move.
Then imagine a trend that persists for eighteen months.
The slower system may ride it beautifully while the faster system repeatedly exits and re-enters.
Which horizon is correct?
Neither.
Both.
It depends on a future we have not seen.
Markets have no privileged clock.
What looks like noise at one horizon may contain meaningful directional structure at another.
Time-frame breadth therefore gives us different ways of encountering when trend expresses itself.
Systems: How
Then there is the rule itself.
A channel breakout sees one feature of price.
A moving-average crossover sees another.
A volatility envelope another again.
A retracement system may enter a trend that another system has already held for months.
These systems share a common objective.
They do not share an identical path.
This is how we search.
Markets give us where.
Time horizons give us when.
Systems give us how.
That is a much richer conception of diversification than simply holding a long list of contracts.
“Market breadth gives us places to search. Time-frame breadth gives us scales to search. System breadth gives us different ways of seeing what is there.”
The Portfolio Is a Search Network
This is perhaps my favourite way of thinking about the whole thing.
The portfolio is a distributed search network.
Every market is a sensor placed somewhere in the global system.
Most sit quietly.
Some fire and fail.
Others detect modest trends that eventually disappear.
And occasionally one finds something extraordinary.
A crop failure.
A monetary regime change.
A currency dislocation.
A supply shock.
A speculative unwind.
A political rupture.
Something begins to move.
At first we do not know what it is.
Episode 1 already dealt with that problem.
The entry fires anyway.
Now we can see why breadth matters.
The perfect entry rule applied to ten markets cannot detect an outlier occurring in the eleventh.
The system cannot capture what it cannot see.
So we distribute the sensors.
Not because every market deserves equal intellectual conviction.
Because we do not know where conviction will eventually have been justified.
This is the same humility that runs through the entire programme.
We are not trying to know more.
We are trying to build intelligently around what we cannot know.
Liquidity Draws the Boundary
There is, however, an obvious limit.
We cannot trade everything.
A market can offer wonderful theoretical diversification and still be completely unsuitable for the programme.
If we cannot enter without materially affecting price, we have a problem.
If we cannot exit under stress, we have a bigger one.
Liquidity therefore places a hard practical boundary around diversification.
And liquidity is not simply today’s trading volume.
What matters is our position relative to available liquidity.
Contract size matters.
Market depth matters.
Execution matters.
Capital matters.
Stress matters.
A small programme and a very large programme do not necessarily have the same tradable universe.
The smaller programme may be able to participate in markets where its orders barely register, although futures contract granularity can make some markets too large for a small account.
The larger programme solves the granularity problem and eventually encounters capacity instead.
There is no universal deployable market list.
There is a live universe appropriate to the capital, liquidity and execution constraints of the programme.
The principle is simple:
Seek the widest useful diversity available within the real constraints of capital, contract size, liquidity and execution.
Not maximum diversification at any price.
Maximum usable diversification.
Search Broadly. Constrain Later.
There is an important sequencing issue here.
The markets we research do not have to be identical to the markets we can deploy today.
That distinction matters.
Suppose a market has forty years of useful history and represents a genuinely different part of the global economic system, but its current futures contract is too large for a $200,000 account.
Should we remove it from the research universe?
No.
Its current tradeability tells us something about the constraints of the account.
It tells us nothing about whether the market contains useful information about the robustness of the system.
So we separate two questions.
Is this a useful market on which to test the architecture?
And:
Can this particular programme trade it today?
They are not the same question.
During research, we want breadth.
A system designed on one market and then tested across many different markets is being asked a much harder question than one repeatedly refined against the market on which it was born.
Does the logic travel?
Does it survive different sectors, different histories and different price-generating mechanisms?
That evidence remains valuable even if some of those markets cannot currently enter the live portfolio.
Deployment comes later.
Now capital matters.
Contract granularity matters.
Margin matters.
Liquidity matters.
Execution matters.
A market that contributed enormously to the research process may fail one of those tests and never receive a live allocation at the programme’s current size.
That is fine.
It has already done useful work.
This is why I prefer to search broadly and constrain later.
If we allow today’s capital to define the research universe at the beginning, we risk making the architecture itself a product of temporary implementation constraints.
A larger account may be able to trade markets the smaller account could not.
A micro contract may become available.
Liquidity may improve.
The deployable universe changes.
The underlying logic should not have to be rediscovered every time it does.
There is another advantage.
Broad testing makes it harder for a strategy to hide.
A rule that looks wonderful in the market where it was developed may reveal itself very differently when exposed unchanged to markets with completely different histories.
That is exactly what we want to know.
The research universe therefore has one job.
Challenge the logic as widely as we reasonably can.
The deployment universe has another.
Determine what the actual programme can responsibly trade.
Do not confuse the two.
One tests the architecture.
The other respects reality.
Then the Storm Arrives
Everything looks beautifully diversified.
Then the crisis comes.
Markets that appeared loosely related begin moving together.
Correlations rise.
Losses appear in places that were not supposed to lose together.
Diversification looks broken.
But the more useful interpretation is that another coupling mechanism has become dominant.
Often that mechanism is liquidity.
A leveraged participant faces a margin call.
They need cash.
They sell what they can.
Another participant does the same.
Prices fall.
Those falls generate pressure elsewhere.
More selling follows.
Feedback begins feeding on itself.
Now the fundamental reason somebody owned wheat or equities or bonds matters less than the fact that they need liquidity.
Different markets have become connected through the same urgent constraint.
This is what complex adaptive systems do.
Relationships that were weak become strong.
Dormant pathways become dominant.
Local events propagate.
Feedback amplifies them.
The structure reorganises.
Portfolio construction cannot prevent this.
Nothing can.
The question is what happens to the programme while it is happening.
The Portfolio Does Not Prevent the Storm
This brings us back to the hero image.
The portfolio does not prevent the storm.
It determines whether the structure survives it.
But portfolio construction cannot do that alone.
This is where the earlier episodes begin fitting together.
Position sizes are small.
So individual failures are contained.
The portfolio is broad.
So the programme does not depend on one market complex.
The Cut Back Rule can reduce exposure as closed equity deteriorates.
The exit machinery continues to operate.
And the programme remains capable of participating in whatever trends emerge from the dislocation.
That last part matters.
A crisis does not merely destroy old structures.
It creates new ones.
Interest-rate regimes change.
Currencies reprice.
Supply chains reorganise.
Commodity shortages appear.
Policy responses create second-order effects nobody anticipated.
Capital moves somewhere else.
The aftermath of one outlier often contains the seeds of another.
So survival is not merely defensive.
Survival preserves access to what happens next.
This is why I do not want a portfolio engineered to look wonderful during the historical crisis we already know about.
I want an architecture capable of remaining functional during the one we don’t.
Diversification Is Supposed to Look Inefficient
There is another thing nobody tells you about genuine diversification.
It is annoying.
At almost every point in time, part of the portfolio looks stupid.
One market is losing.
Another has gone nowhere for three years.
One system appears completely unsuited to the current environment.
One time horizon seems obviously inferior to another.
And sitting right there in hindsight is a cleaner portfolio containing only the things that worked.
Of course there is.
Hindsight is an extraordinary portfolio manager.
It never owns dead weight.
Unfortunately, it starts trading tomorrow with yesterday’s answers.
Real diversification cannot do that.
It must carry things whose future usefulness is unknown.
That means inefficiency is not necessarily evidence that diversification has failed.
Sometimes it is evidence that diversification is real.
If everything in the portfolio is working for the same reason at the same time, I become nervous.
Because there is a reasonable chance I do not own many different things at all.
I own one thing wearing several costumes.
We Are Diversifying Dependencies
This is the deeper point.
We are not really diversifying tickers.
We are diversifying dependencies.
Different markets.
Different economic drivers.
Different time scales.
Different entry mechanisms.
Different exit points.
Different directions.
Different paths.
None is perfectly independent.
That is impossible inside a coupled global system.
But neither are they identical.
And those differences give the programme multiple ways to encounter uncertainty.
That is enough.
We do not need every component to succeed.
We do not even expect it.
We need the architecture to avoid depending on any single component being the one that succeeds.
That is a very different standard.
And I think it is a much more robust one.
From Portfolio to Ensemble
We now have the landscape.
Many markets.
Different structural drivers.
Different scales.
A broad search network spread across the global economic system.
But there is still a problem.
Suppose we chose the right market.
The great outlier occurs there.
But the particular trading rule we chose encounters that trend badly.
Perhaps it enters too late.
Perhaps it exits during a retracement.
Perhaps the move develops in a form that another simple rule would have captured much better.
We solved where to search.
We have not yet fully solved how to search there.
That is the next layer of the architecture.
Instead of asking one system to become clever enough to recognise every possible manifestation of trend, we do something much simpler.
We use several.
Different rules.
Different responses.
Different weaknesses.
Different strengths.
No master system.
No requirement that one of them be right all the time.
Once again, we distribute the problem.
That is the ensemble.
And that is Episode 5.
READ DEEPER
→ Casting a Wider Net: The Power of Maximum Diversification for Outlier Hunters
→ Fractals, Diversification, and the Myth of Dilution
→ Are Markets Fractal? The Case for Maximum Diversification
→ Diversification for Trend Following Models: The Small Variations Matter
Previous: System Anatomy 3: Position Sizing | Next: System Anatomy 5: The Ensemble Approach
Richard Brennan writes on systematic trading, complex adaptive markets, and the philosophical foundations of trend following at atstradingsolutions.com. His books include The Fractals of Finance, Complex Adaptive Markets, Carved by Impossibility and The Aussie Turtles Trend Following Guide.
Want to explore why structure exists at all?
Carved by Impossibility: What Remains When Everything Else Is Eliminated
The book explores the architecture of constraint, emergence, and reality itself, and what it means for how we understand markets, life, and the universe.
Available now on Amazon in paperback, hardcover, and Kindle.
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.