Systematic futures trading

What systematic futures trading actually means

Systematic futures trading uses predefined, coded rules to decide when to enter, how much to hold and when to exit positions in exchange-traded futures contracts. Discretionary trading relies on judgement for each trade. The rules can be simple or elaborate. The defining property is that the same inputs should produce the same decisions. That property makes the approach testable, and it also makes its failure modes specific and traceable.

Before acting on it, separate two questions. The first is whether a set of rules has any merit. The second is whether the infrastructure around those rules can carry them out safely. This article focuses mostly on the second question. Futures add leverage through margin, contract expiry and venue-specific rules, and these mechanics can turn small engineering oversights into large losses. The material is educational and is not a recommendation to trade any instrument.

Futures mechanics every systematic rule must respect

Futures are margined, so a modest deposit controls a much larger notional exposure. A position-sizing bug or a misread contract multiplier can therefore magnify losses well beyond the capital a reader believed was at risk. A sound system calculates exposure in both notional and margin terms. It should reject any order that would push combined exposure past a set limit.

Contracts also expire. Research usually relies on continuous price series stitched together from successive contracts. The stitching method, whether back-adjusted, ratio-adjusted or unadjusted, changes what the history appears to show. Live trading has to roll positions from one contract to the next. Each roll costs spread and slippage, and the next contract may be thinly traded when the roll happens. Exchanges also set trading hours and price limits, and they sometimes halt trading. A rule that assumes it can always exit at the next quoted price will eventually meet a moment when it cannot.

• Confirm the multiplier, tick size and tick value for every contract the rules touch.
• Document the roll schedule and the continuous-series method used in research.
• Plan for margin requirement changes, which exchanges and brokers can raise at short notice.
• Encode session hours, price limits and halt handling explicitly.

Explicit risk limits and human oversight

Risk limits should be written down and enforced in a layer separate from the strategy code, so that a strategy bug cannot quietly switch them off. Typical limits include a maximum position per contract, a maximum gross notional, a maximum loss per day, a maximum size for a single order and a kill switch that stops all new orders. Each limit needs a defined action when it trips: block the order, reduce exposure, flatten everything or alert a person. Decide which in advance, because there is no time to debate it during a fast market.

Human oversight complements automation rather than contradicting it. People set the limits, review exceptions and keep the authority to halt the system. IMRYN's public material describes automation that operates inside guardrails, backed by risk controls and ongoing monitoring. That is why this article treats oversight as an infrastructure requirement rather than an optional extra. That framing is a description of approach, not a promise of outcomes.

Observable execution, including across venues

Execution is observable when every step from signal to settled position leaves a record. That means each signal, order, acknowledgement, fill, partial fill, cancellation and rejection, stored with consistent timestamps. With that record, intended positions can be reconciled against broker and exchange statements. Discrepancies can then be found in minutes rather than discovered at month end.

Observability matters more when orders are routed across several venues or brokers, because each one holds its own view of state, and those views can diverge. IMRYN describes its offering as systematic trading infrastructure capable of execution across multiple venues. Its architecture page outlines that context for readers who want to see how the components are presented. Whatever stack is used, compare realised fill prices against a reference price and track the gap over time. A persistent difference between modelled and actual costs is strong evidence that the research assumptions need revisiting.

Reproducible evaluation and what it cannot tell you

An evaluation is reproducible when another person can rerun it and obtain the same result. That requires versioned data, versioned code, recorded parameters and documented assumptions about costs, slippage and rolls. Without reproducibility, nobody can tell whether a promising result reflects the rules or an accident of data handling.

Even a fully reproducible backtest only shows how rules would have behaved on particular historical data under particular assumptions. Historical results and simulations do not settle what will happen next. Overfitting, structural changes in markets and underestimated costs all erode the link between past and future. Out-of-sample testing, walk-forward analysis and paper trading narrow the gap without closing it. IMRYN's methodology page sets out how it frames evaluation, and it should be read with this boundary in mind.

Example: a pre-launch checklist for a hypothetical rule

Hypothetical example, not a recommendation: a reader has a trend-following rule for an equity index future and a backtest they find encouraging. Before considering live capital, they could work through the checklist below and mark each item yes, no or unknown.

Any answer of no or unknown points to more preparation rather than deployment. The checklist does not judge whether the rule is good. It only judges whether the reader would be able to see, limit and stop what the rule does once it is running.

• Can the backtest be rerun from versioned code and data with identical output?
• Are cost, slippage and roll assumptions written down and conservative?
• Are position, notional, order-size and daily-loss limits enforced outside the strategy?
• Is there a tested kill switch, and does someone know who presses it?
• Are fills reconciled against broker statements automatically and daily?
• Has the rule run in a paper or simulated live environment long enough to compare modelled and realised costs?