Trend following is a way to define a trading rule, not evidence that a trend exists or will continue. Before testing one, decide what counts as a trend, when the signal becomes knowable, what order the test assumes, and what would close the position. Those choices separate a repeatable experiment from a chart that looks persuasive after the move.

The label covers several different rules. A moving-average crossover, a close above a previous range, and a positive return over a fixed lookback all express direction differently. They can respond at different speeds and fail in different markets. Choose one family first; keep the others as separately named alternatives.

moving-average crossover timingIllustrative hourly sequence: wait for the completed bar that confirms the fast average crosses the slow average, then model entry on a later bar; the sketch is not market data.Fast SMASlow SMACompleted signal barLater bar · modeled entryA crossover is a delayed rule event, not a prediction or guaranteed fill
Illustrative hourly sequence: wait for the completed bar that confirms the fast average crosses the slow average, then model entry on a later bar; the sketch is not market data. Illustrative teaching diagram · not live market data
four trend and momentum signal familiesMoving-average crossover, single-instrument time-series momentum, previous-range breakout and cross-sectional ranking turn a directional thesis into different calculations. They need distinct timing, universe and turnover rules.FOUR WAYS TO DEFINE A DIRECTIONAL RULE01 · Averagesfast crosses slowsame asset · past closesState or new crossing?02 · Own returntime-series momentumone asset · lookback returnLong · flat · or short?03 · Prior rangecompleted close breaks outexclude signal barHow is a false break handled?04 · Relative rankcross-sectional momentumuniverse · rank · rebalanceTurnover and delistings?Related ideas · different calculations · freeze the market, clock and rule before comparing results
Moving-average crossover, single-instrument time-series momentum, previous-range breakout and cross-sectional ranking turn a directional thesis into different calculations. They need distinct timing, universe and turnover rules. Illustrative teaching graphic · not live market data

Choose the signal family before its settings

Rule family Example definition Question to freeze
Time-series momentum Go long, stay flat or short according to the same asset’s return over a completed lookback Which return horizon, threshold and position states?
Moving-average crossover A fast average crosses a slower average of completed prices Does entry require a new cross or only that the fast line is above the slow one?
Price-channel breakout A completed close moves above a range made from earlier bars Is the current bar excluded from the range, and what confirms a failed break?
Cross-sectional momentum Rank a defined group of assets by their earlier returns and allocate by rank Which assets qualify, when is the list refreshed, and how are delisted or missing assets handled?

These families overlap in purpose but are not interchangeable. Time-series rules judge one instrument against its own history. Cross-sectional rules compare a changing set of instruments; their outcome can depend on universe construction, ranking, turnover and rebalancing. A chart indicator may also be a filter or exit condition rather than the entry signal itself.

Recent evidence is a reason to test the horizon, not to copy a parameter. A Bitcoin study published in Risk Management in July 2026 compares signal speeds and its publisher abstract reports stronger risk-adjusted performance for a slower signal with a 12-week baseline horizon than for the faster alternatives it studied. That result belongs to the paper’s Bitcoin sample and specification. It does not validate this article’s hourly moving-average example, another coin, or a live execution model. The full paper is subscription-only, so this guide relies on the publisher’s abstract and does not infer unreported details. Read the paper abstract at Springer Nature.

Make a crossover causal

A simple moving average is the arithmetic mean of a declared number of closes. A fast average uses fewer observations than a slow one. “Fast above slow” is a state; “fast crosses above slow” is an event that occurs once when the relationship changes. A test must say which event it uses.

Here is a deliberately small, fictional 3-bar/5-bar example. It is arithmetic for learning, not market data or a recommended setting. Before the final bar, the three-bar average is (95 + 95 + 96) ÷ 3 = 95.33, below the five-bar average (110 + 95 + 95 + 95 + 96) ÷ 5 = 98.20. After a fictional close of 105, the fast average becomes (95 + 96 + 105) ÷ 3 = 98.67 and the slow average becomes (95 + 95 + 95 + 96 + 105) ÷ 5 = 97.20. The crossover is only known when that last bar has completed.

If the test uses the next available bar’s open as its entry convention, that later price is the modeled entry. Using the 105 close both to confirm the signal and to fill the order would need separate evidence that the rule and fill could occur in that sequence. The next-open convention is still an assumption: hourly or daily OHLC data cannot reveal queue position or guarantee an executable fill.

fictional moving-average crossover calculation and later modeled entryInvented closes create a three-bar average of 95.33 below a five-bar average of 98.20 before the last bar, then 98.67 above 97.20 after its close. The signal is known at that completed close; an illustrative next-bar open of 106 is a later modeled entry, not a live fill.FICTIONAL CLOSES · THE CROSS IS NOT KNOWN UNTIL THE BAR CLOSESBEFORE FINAL BARFast 3-bar SMA · 95.33Slow 5-bar SMA · 98.20Fast is below slowAFTER CLOSE = 105Fast 3-bar SMA · 98.67Slow 5-bar SMA · 97.20Cross confirmed at bar completionsignal close · 105next open · 106 assumedThe later price is a model convention · every value shown is invented, not market data
Invented closes create a three-bar average of 95.33 below a five-bar average of 98.20 before the last bar, then 98.67 above 97.20 after its close. The signal is known at that completed close; an illustrative next-bar open of 106 is a later modeled entry, not a live fill. Illustrative teaching graphic · not live market data

For a real test, define the instrument, venue, bar interval and timestamp convention first. Record whether the series is spot or perpetual, what happens with missing bars, and whether the signal uses a completed candle. Write the exit before running the test: for example, an opposite completed crossover or a predeclared maximum holding period, whichever happens first. Specify the order type, price reference, stop behavior, position size and fill convention. The entry average alone is not a complete strategy.

The current SuperQuant trend strategy emits a signal when its configured fast SMA crosses its slow SMA at the previous closed hourly bar. The backtest evaluates that signal for a modeled entry at the next bar’s open, subject to the engine’s shared position and exit rules. This describes the product’s current implementation; it is not a universal definition of trend following.

trend strategy signal to execution sequenceA completed-bar trend condition is evaluated first, then an order is modeled on a later bar. The rule records the exit condition and execution assumptions before judging a result; the drawn path is not price data.SEPARATE WHEN THE RULE IS KNOWN FROM WHEN IT CAN ACT1 · Observeclosed bar only2 · Confirmrule becomes true3 · Model entrylater executable point4 · Invalidatewritten exit rule5 · Reviewcosts + errorsFreeze thresholds, bar timing, fill rule, stop and cost assumptions firstTrend-following can lag reversals and lose during sideways conditions · not a forecast
A completed-bar trend condition is evaluated first, then an order is modeled on a later bar. The rule records the exit condition and execution assumptions before judging a result; the drawn path is not price data. Illustrative teaching graphic · not live market data

Measure the rough patches as well as the long moves

Trend rules react to past prices, so some delay is inherent. A reversal can occur before the signal turns; a sideways market can cross the same thresholds repeatedly. Count those entries, exits and costs instead of showing only a clean run-up. Compare the rule with a relevant passive or flat benchmark on the same dates, and report time in market, trade count, drawdown path, longest losing sequence and the share of the result contributed by the largest trades.

Split the history chronologically. Freeze a later interval before choosing settings; if you inspect it and change the rule, record that the interval has been used. Show the parameter choices you actually compared, including nearby values that did poorly. Do not present the winning pair as if it had been specified before the search. Test more than one market condition and preserve the version, data source and code or configuration for each run.

Put costs and exposure beside the signal

Suppose a separate fictional test uses about $10,000 of entry notional and assumes a 0.10% fee on each side. That is about $10 on entry and, near the same exit notional, about $10 on exit. A price move of roughly 0.20% of the starting notional would only offset those two assumed fees; spread, slippage and any funding remain outside that simple estimate. Actual fees depend on the venue, account tier, order role and executed notional. Treat every cost as an input to stress, not a footnote to remove after seeing the result.

For a perpetual, include the funding settlements that actually fall within each position, with source, sign and timestamp. A rate observed now does not fix a later payment. For derivatives, a quantity chosen from a stop distance is not a guarantee against a gap or liquidation. The CFTC’s virtual-currency advisory warns that leverage amplifies losses and can lead to margin calls or forced closure; its discussion is a general futures-risk warning, not a venue rule. Check the exact contract and current venue terms for a particular test. CFTC virtual-currency risk advisory.

Match the conclusion to the evidence

SuperQuant’s bundled histories are synthetic, and its simplified engine does not reproduce every venue’s spreads, order-book depth, funding history, tiered liquidation or fill behavior. A repeatable run can show how the software applies the selected rule to its input. It cannot establish that the same result occurred in Bitcoin history or will occur in a live account. Replace the example data only with a documented source whose timestamps, gaps and contract definitions you have checked.

Before extending a test, save the rule version, data provenance, cost assumptions, benchmark, chronological split and failure condition in a journal. A result that disappears under a small cost increase or a later window is useful evidence when it is recorded honestly. Change the rule only as a new version with a new question.

Continue with the crypto breakout rule guide, the backtesting workflow, and the cost model guide. The backtest audit checklist helps preserve the assumptions. Trend following can lose money; this article is an educational research method, not a forecast or personal recommendation.