Rollover Mechanics
The system can only be as honest as the price history it sees. In futures, that history has to be constructed.
Previous: System Anatomy 5: The Ensemble Approach | Next: System Anatomy 7: Execution and Process
A surveyor can work with extraordinary precision and still produce the wrong map.
The measurements can be meticulous.
The calculations flawless.
The boundaries plotted to the millimetre.
But if the chain used to measure the ground is incorrectly calibrated, all that precision simply gives us a more precise version of the wrong answer.
That is the uncomfortable thing about measurement.
Precision and accuracy are not the same.
There is an equivalent problem buried underneath every futures trading system.
We talk casually about “the price history” of crude oil, wheat, gold or bonds as though somebody has been recording one continuous price for the past forty years.
They haven’t.
A futures contract expires.
Another replaces it.
Then another.
And another.
The long price history our system sees has been assembled from those individual contracts.
It is constructed.
That makes the construction part of the system.
Get it wrong and everything sitting above it can still look perfectly respectable.
Signals fire.
Stops move.
Positions are sized.
Backtests produce statistics to several decimal places.
But the chain is wrong.
And precision cannot rescue us from a bad ruler.
There Is No Forty-Year Futures Contract
Take crude oil.
We might show a chart extending across decades and talk about what crude oil “did” over that period.
But no futures contract traded continuously across those decades.
Individual contracts came into existence, traded, approached expiry and disappeared.
The market continued.
The contract did not.
So if we want a continuous history for research, we have to join successive contracts together.
That sounds straightforward until we notice something important.
The expiring contract and the contract replacing it usually have different prices.
Suppose the old contract is trading at 75 and the next contract at 77.
Nothing suddenly happened to the underlying market when we moved from one to the other.
Yet simply joining the two histories creates a two-point jump.
Our data now contain a price move that nobody actually experienced.
And our system cannot know that.
It simply sees price.
So that artificial jump can find its way into everything downstream.
A measure of recent movement.
An entry threshold.
A trailing exit.
A position-sizing calculation.
The machinery does not know the data contain an artefact.
It trusts what we gave it.
That is the first principle of rollover mechanics:
The system cannot distinguish market information from errors introduced by the construction of its own data.
We have to do that for it.
The Market and the Measurement of the Market
This distinction runs deeper than futures.
There is the market.
Then there is our representation of the market.
They are not the same thing.
A price series is already a measurement.
A continuous futures series goes one step further. It is a measurement assembled across time from a succession of instruments.
That doesn’t make it false.
It makes the method of construction important.
Think again about the surveyor’s chain.
The surveyor is interested in the field, not the chain.
But the only way to measure the field is through the instrument.
If the instrument changes length halfway through the survey and nobody accounts for it, the resulting map inherits the error.
Futures research faces much the same problem.
Each contract transition creates a discontinuity that belongs to the structure of the futures market rather than to the directional movement we are trying to measure.
Somehow, we have to preserve continuity without inventing history.
That is what adjustment is trying to accomplish.
Stitching the History Together
There are different ways to join successive futures contracts.
The details matter, but the principle matters more.
When we adjust a historical series across a roll, we want to remove the artificial discontinuity without unnecessarily changing the relationships in the history that our trading rules depend upon.
That qualification is important.
It is easy to make a chart look continuous.
It is harder to make it continuous while preserving the relevant geometry of the price history.
One common approach adjusts history by the absolute difference between the old and new contracts.
Another rescales history according to the ratio between them.
Those methods do not produce the same historical series.
For the type of programme we are discussing, preserving proportional relationships through time is particularly valuable.
Why?
Because much of the architecture we have already built responds to relationships.
How large is recent movement relative to price?
How far has price travelled relative to its historical scale?
Where does the current market sit relative to previous levels?
How far away should an exit sit given the movement of the market?
If our adjustment process changes those relationships, we have changed the object being measured.
This is why the construction method cannot be treated as cosmetic data cleaning.
A continuous series is not merely a prettier chart. It is part of the mathematical environment in which the system operates.
The Dangerous Beauty of a Clean Backtest
This is where things become uncomfortable.
A poor continuous series does not necessarily announce itself.
The backtest may run perfectly.
The equity curve may look excellent.
The statistics may be impressive.
There may be no error message.
Nothing crashes.
That is precisely why this class of problem is dangerous.
If a trading rule contains an obvious coding error, we have a chance of finding it.
If the historical data have been subtly constructed in a way that changes the behaviour of the rule, the system may simply produce a different answer.
And because the answer arrives wrapped in numbers, we are tempted to trust it.
This is another reason I am suspicious of excessive precision in systematic research.
A backtest returning a result to four decimal places tells us nothing about whether the underlying measurement was appropriate.
Calculation can be precise while the premise is wrong.
The survey still looks magnificent.
The chain is still stretched.
When One Contract Becomes Another
There is another decision hiding in the construction.
When do we switch contracts?
Again, there is no universally correct date handed down by the market.
The expiring contract gradually loses relevance.
Activity migrates toward another contract.
Liquidity shifts.
At some point the programme needs to recognise that transition and move with it.
There are sensible ways to define that point.
But for the anatomy of the system, the most important principle is not the particular trigger.
It is consistency.
If our historical series assumes that the programme moved from one contract to another according to one rule, while our live process makes that transition according to another, we have created two different price histories.
They may differ only slightly.
Most of the time that difference may not matter.
Until it does.
A signal sits close to a threshold.
An exit is nearly triggered.
A measure of recent movement changes just enough to alter a position size.
Now the small difference in data construction becomes a difference in behaviour.
The research programme does one thing.
The live programme does another.
Neither necessarily contains an obvious mistake.
They are simply no longer the same system.
The Backtest Is Not the Rules Alone
This point deserves more attention than it usually receives.
We often think of a backtest as a test of trading rules.
Entry rule.
Exit rule.
Position-sizing rule.
Portfolio rule.
But that is incomplete.
The backtest is the interaction between those rules and the data on which they operated.
Change the data construction and you have changed the experiment.
That means a live programme does not faithfully reproduce its research merely because the same entry and exit formulas appear in the code.
It must also preserve the assumptions that produced the price history on which those formulas were tested.
This is why rollover consistency matters so much.
The question is not:
Did we use the same trading rule?
The question is:
Did the trading rule encounter the market on the same basis?
That is a much higher standard.
The Roll Is Not a New Idea
Suppose we are already holding a trend when the contract approaches expiry.
Nothing about our directional thesis has necessarily changed.
The system has not suddenly detected a new trend.
The market has not instructed us to reconsider the position.
The instrument through which we are expressing that position is simply approaching the end of its life.
So we move from one contract to another.
Conceptually, that is not a new trade.
It is the continuation of an existing process through a change in instrument.
That distinction matters.
We do not want the mechanics of the futures contract to manufacture a signal that the underlying trading logic never generated.
Nor do we want a contract transition to manufacture a profit or loss simply because the new contract trades at a different nominal price.
The system should pass through the roll with its structural state intact.
The trend before the roll should remain the trend after it.
The position before the roll should remain recognisably the same exposure after it.
The measuring chain has changed section.
The measurement must continue.
Carry Is Real. The Gap Is Not a Trend
There is an important subtlety here.
The difference between two futures contract prices is not meaningless.
It can contain information about financing, storage, scarcity, convenience yield and expectations embedded in the term structure.
Contango is real.
Backwardation is real.
Rolling futures positions can have real economic consequences.
We should not pretend otherwise.
But that is different from treating the mechanical price gap between successive contracts as though the underlying market instantaneously moved by that amount on roll day.
It didn’t.
This is why continuous-series construction requires care.
We are not trying to erase the economics of futures.
We are trying to prevent the mechanics of stitching contracts together from masquerading as directional price movement.
Those are different objectives.
Confusing them contaminates the signal.
The Same Instrument Must Reach Every Component
Now think back through the series.
Episode 1 built the entry.
What does the entry observe?
Price.
Episode 2 built the exit.
What determines whether that exit is reached?
Price.
Episode 3 built position sizing.
What feeds the measure of recent movement used to scale the position?
Price.
Episode 4 distributed the programme across markets.
What history do we use to understand and test those markets?
Price.
Episode 5 distributed the programme across different structural lenses.
What are all those lenses looking at?
The same underlying price history.
Suddenly rollover mechanics does not look very administrative.
It sits underneath almost everything.
If the data are inconsistent, every component above them inherits the inconsistency.
Not necessarily catastrophically.
Quietly.
That is worse.
Data Are Part of the Architecture
There is a temptation in systematic trading to divide the world into two categories.
There is the interesting stuff:
Signals.
Systems.
Portfolio construction.
Sizing.
And then there is the plumbing:
Data.
Contract specifications.
Rolls.
Execution.
I think that distinction is a mistake.
The plumbing is part of the machine.
A magnificent engine connected to a contaminated fuel line is not a magnificent car.
A perfectly calibrated survey technique using the wrong chain does not produce the right boundary.
And a beautifully designed trading system operating on a price history constructed inconsistently with its live implementation is not the system we think we tested.
This is not an argument for obsessive complexity.
Quite the opposite.
It is an argument for knowing exactly which details matter.
Many things in markets are unknowable.
This isn’t one of them.
We can decide how our historical series is constructed.
We can document it.
We can apply it consistently.
We can make sure the live process respects the same logic.
There is enough uncertainty in trading without manufacturing additional uncertainty ourselves.
Precision Is Not Integrity
This brings us back to the hero image.
The ruler looks precise.
The chain looks precise.
The surveyor can record the result precisely.
But precision is downstream of calibration.
That is the larger lesson.
Systematic trading gives us enormous computational precision.
We can calculate indicators to extraordinary accuracy.
Simulate thousands of trades.
Measure decades of performance.
Optimise portfolios.
Generate statistics to as many decimal places as we please.
None of that establishes integrity.
Integrity comes from maintaining consistency all the way through the chain.
From raw market data.
To continuous history.
To signals.
To positions.
To the live instrument.
Every link has to describe the same process.
Otherwise the apparent precision of the final result is telling us more about our calculator than about the market.
“The survey looks precise. Whether it is accurate depends entirely on the calibration of the chain.”
The System We Tested Must Be the System We Trade
That is ultimately what rollover mechanics is about.
Not contract codes.
Not administrative schedules.
Not remembering expiry dates.
Those things matter operationally, but they belong elsewhere.
The architectural principle is simpler.
The system we trade should be the system we tested.
Same logic.
Same assumptions.
Same treatment of the market.
Same method of maintaining continuity as contracts come and go.
We will never reproduce history.
We do not want to.
The future should be different. That is the whole problem the architecture has been designed to face.
But we should not allow differences in our own implementation to masquerade as differences in the market.
The future is allowed to surprise us.
Our plumbing isn’t.
And that takes us to the final episode.
Because even with clean data, robust rules, controlled sizing, broad diversification and multiple ways of encountering trend, the architecture still has one final point at which intention has to become reality.
Execution.
A system that exists only in research has never taken a position.
A signal that is never faithfully implemented has never encountered the market.
The final link in the chain is where the machine meets the world.
And where the human being operating it can still change everything.
READ DEEPER
→ Who Holds the Risk: Why Markets Survive and Traders Do Not
→ The Paradox of Simplicity: Why the Best Trading Rules Are Counterintuitive
→ Selection, Not Skill: Why Simple Strategies Outlive Brilliant Ones
Previous: System Anatomy 5: The Ensemble Approach | Next: System Anatomy 7: Execution and Process
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.