THE COMPLEXITY OF REPLICATION: Why Simpler Is Not Always What It Seems

A conversation has been circulating in the managed futures community recently, and it deserves closer examination.

The argument runs something like this: broad diversification across hundreds of markets delivers diminishing returns, short-term models are a drag on performance, risk controls can be head-faked, and lower fees leave more alpha for clients. From there, the conclusion follows that simplicity, in the form of a concentrated replication approach using a small number of highly liquid markets, may be superior to the complexity of a fully diversified trend following program.

It is a compelling narrative.

It is also one that needs more careful scrutiny.

Replication Is Not Simple

The first point is straightforward. Replication, as it is actually practiced, is not simple. It contains multiple layers of complexity that the simplicity narrative tends to hide.

The strategies being replicated, usually represented by major CTA indices, are built by some of the largest and most sophisticated quantitative research teams in the world. They run multi-model, multi-market programs across dozens, and often hundreds, of instruments. They apply dynamic risk overlays and continuously invest in research and refinement. That complexity does not disappear simply because a replicator sits downstream of it. It is still there. It has just been outsourced, with the underlying managers bearing the cost of generating it.

Then there is the replicator itself.

A replication algorithm is a quantitative model built on the output of other quantitative models. It attempts to infer the aggregate positioning of a diverse universe of CTAs from daily return data and then express that inferred positioning through a concentrated proxy portfolio. That is not a trivial exercise. It carries model risk, parameter sensitivity, and regime dependence of its own.

This structure will be familiar to anyone who has looked seriously at the quantitative investment strategy space (QIS), where banks and institutions have long offered systematic, rules-based products designed to give investors efficient exposure to strategy returns without the cost of the underlying programs. These products are built on indices, delivered through derivatives, and positioned as low-cost alternatives to the strategies they track. The appeal is real. So are the limitations. In practice, the gap between backtested and live performance in these products has often been substantial, and dispersion across apparently similar offerings can be surprisingly wide. The complexity of specification, model sensitivity, and regime dependence does not disappear simply because the packaging is clean.

Replication sits in the same broad category as QIS. Both are systematic models built on the output of other systematic models. Both compress a complex underlying universe into a more tractable form. Both carry their own engineering challenges that the simplicity narrative tends to leave unexamined.

So when replication is described as simple, what is really meant is that the instrument count is low.

That is a genuine practical advantage. It can improve liquidity, reduce turnover frictions, and lower execution costs.

But a low instrument count is not the same thing as a simple underlying architecture.

The Simplicity Narrative and What It Obscures

I want to be direct about something here, because it matters to how allocators think about this space.

When complexity is positioned as the enemy of returns, a particular kind of flattening occurs. Products that approximate the broad behavior of a fee-heavy, institutionally crowded CTA peer group start to occupy the same mental category as programs that are doing something genuinely different. The distinctions that matter most, about return distributions, about where alpha actually lives, about what happens in stress regimes, get compressed into a single question about instrument count or fee level.

That flattening is not neutral. It makes it harder for allocators to ask the right questions. It substitutes a marketing framework for an analytical one.

A program that hunts outliers across a broad and diverse futures universe, that is built around the empirical reality of fat-tailed, positively skewed return distributions, and that derives its edge from being present across many markets when rare but large moves occur, is not a more expensive or more complicated version of the same thing. It is a different thing. The underlying architecture, the return profile, the behavior in crisis windows, and the source of allocator value are all distinct.

Narratives that treat complexity as inherently wasteful make that distinction harder to see. And when allocators cannot see it clearly, they are less able to make informed decisions about what they actually own and what they are giving up.

Exposure Replication Is Not Distribution Replication

This leads to the deeper question.

The real issue is not whether replication can work under certain conditions. It is whether it can replicate what a genuinely diversified classic trend following program actually does.

That depends entirely on what you are trying to replicate.

A concentrated proxy set can capture a meaningful share of the common factor exposures across a crowded institutional CTA peer group. When the largest managers in the world are all long similar bond trends, similar equity rotations, and similar FX moves, those exposures can compress reasonably well into a small set of liquid instruments. Under those conditions, a replicator can track the broad behavior of that peer group with acceptable tracking error, especially when the underlying managers are themselves smoothing extremes through position sizing and risk control overlays.

But this is also where the structural limitation begins.

Our empirical research across 68 futures markets and 41 years of daily data points to a different reality for classic trend following. The return distributions are not normal. They exhibit positive skew, excess kurtosis, and statistically significant fat tails. The Hurst exponents confirm that markets have memory, that price paths are persistent rather than random, and that the most valuable opportunities tend to be the largest and least frequent moves.

That matters enormously.

In a genuine outlier-hunting program, alpha does not primarily live in the body of the distribution. It lives in the tails.

A distribution that is approximately normal can be proxied reasonably well because most of the information is concentrated in the mean and the variance. A positively skewed, fat-tailed distribution is a very different problem. The rare, path-dependent moves that define the best years for a classic trend follower are precisely the moves that a concentrated proxy set is most likely to miss. Those moves often emerge in less crowded markets, in instruments that are not in the proxy set, and along price paths that do not map neatly onto the most liquid benchmarks.

That is the key distinction.

A replicator may be able to reproduce broad exposure.

It is far less able to reproduce the distributional character of returns that comes from being present when rare outliers emerge.

The Geometry Problem

A useful way to think about this is through geometry.

Smoothing an ensemble of returns that approaches a normal distribution into a straight line is a tractable problem. Replicating a staircase geometry, where the steps are irregular, infrequent, and of unpredictable height, is a very different one.

You can approximate the average slope.

You cannot reliably position yourself for the next step change.

That is why the distinction between exposure and distribution is not semantic. It goes to the heart of what the strategy actually is.

The Replication Ceiling

There is also a mathematical ceiling to what replication can achieve.

The more a target program depends on idiosyncratic outliers, cross-sectional dispersion across many markets, and crisis-path convexity sourced from less crowded instruments, the more information is lost when that program is compressed into a small proxy set.

This loss is not linear.

Tracking error tends to grow faster as the target moves further from the proxy universe, especially in the stress windows that matter most to allocators. And those are precisely the windows in which a genuinely diversified classic trend following program tends to prove its worth.

This is not just a theoretical objection. It is testable. A replication approach may track a fee-heavy, institutionally crowded, risk-controlled peer group with reasonable fidelity. But it cannot fully replicate a program that is genuinely hunting outliers across a broad and diverse futures universe, because the outliers are exactly what fall outside the proxy set.

What Replication Actually Does Well

None of this means replication lacks merit.

A low-cost, liquid, transparent product that approximates the broad behavior of the large-CTA peer group serves a real allocator need. For fee-sensitive portfolios, where institutional trend following access has often been expensive and opaque, that is a legitimate innovation.

That is the strongest version of the replication thesis: a cheaper, more accessible approximation of large-CTA beta, delivered through highly liquid markets with structural cost advantages.

That case is credible.

The weaker version of the thesis is the one that does not survive scrutiny. It is the suggestion that complexity itself is the enemy, and that a concentrated proxy approach can reproduce the economics of a genuinely diversified, outlier-hunting trend following program.

The evidence suggests otherwise.

The complexity is still there. It has simply been moved. And in the case of the fat-tailed return distribution that defines classic trend following, part of it has been left behind entirely.

For a classic trend follower, diversification across markets does not suffer the same diminishing returns that a distribution-smoothing, risk-controlled approach may face. Each additional market is another opportunity for the outlier that defines the year. The empirical record across decades and dozens of markets supports that view. It is not a theoretical preference for complexity. It is a structural feature of how outlier returns are distributed across a diverse futures universe.

When allocators are told that simplicity is a virtue and complexity is a cost, they deserve to know which complexity is being referred to, and what is being given up in its place.

That is not complexity for its own sake.

It is the geometry of the strategy itself.