VALYNOXSTRATEGY RESEARCH TECHNOLOGY
AlphaPrimeAlphaPrime
Sign inFree 7-day trial
ACADEMY / Strategy generation

Futures algo trading strategies: 5 types and how to backtest each

Most futures algorithms belong to one of five families: breakout, trend-following, mean reversion, time-of-day and volatility. The types that still look sound after costs are those whose average trade is several times the round-trip cost on the contract and bar size chosen, and each type needs its own test settings.

By · Published · updated · Editorial policy

Contents
  1. Which futures algo strategy types hold up in a backtest?
  2. How does a breakout strategy work, and which contracts suit it?
  3. When does trend-following suit ES, NQ and MNQ?
  4. How do you test mean reversion without fooling yourself?
  5. What are time-of-day and volatility strategies, and how are they tested?
  6. How much commission and slippage per contract should the backtest use?
  7. How do you compare strategy types on the same data?
  8. How do you build a small portfolio from different strategy types?
  9. Frequently asked questions
  10. References

Which futures algo strategy types hold up in a backtest?

Five families cover most futures algorithms: breakout, trend-following, mean reversion, time-of-day and volatility strategies. The ones that still look sound after costs are those whose average trade is several times the round-trip cost on the contract and bar size you chose. No single type wins everywhere, so test each on its own terms.

A futures algorithm is a set of written rules that decides when to buy and sell a futures contract without a person choosing each trade. The five families differ in how long they hold, how often they trade and how much a small cost change hurts them:

  • A breakout strategy is a rule that enters when price leaves a defined range, for example the high or low of the first 30 minutes.
  • A trend-following strategy is a rule that enters in the direction of a move that has already started and holds for days or weeks.
  • A mean reversion strategy is a rule that enters against a short move and exits when price returns to an average.
  • A time-of-day strategy is a rule that trades only inside a chosen clock window and is flat the rest of the day.
  • A volatility strategy is a rule that uses a range measure to decide how much to trade or whether to trade at all.

Round-trip cost is the commission plus slippage for one entry and one exit of one contract. Think of each family as a different vehicle: a truck and a bicycle both move, but you do not judge them by the same fuel bill. Each type below is one use of algo trading: written rules that a computer follows without second thoughts.

Five futures strategy types compared: breakout on 5 to 60 minute bars with medium cost sensitivity, trend-following on hourly or daily bars with low cost sensitivity, mean reversion on 1 to 15 minute bars with high cost sensitivity, and time-of-day and volatility strategies whose cost sensitivity varies Five types, five setups Bar size and cost sensitivity Type Bars Cost hit Breakout 5–60 min Medium Trend-following 1 h, 1 day Low Mean reversion 1–15 min High Time-of-day Intraday Varies Volatility Any Varies No single test setup fits all. So: set bars and costs per type.
The five families differ in holding time, bar size and cost sensitivity, so each one needs its own test settings.

How does a breakout strategy work, and which contracts suit it?

A breakout strategy enters when price leaves a defined range, and it suits liquid index futures with a clear session structure, such as ES and NQ on 5- to 60-minute bars. The range can be the first 30 minutes of the US cash session, the previous day's high and low, or a channel of the last N bars.

The entry happens exactly when the market is moving, so the fill is the weak point of the test. Next-bar execution is the rule that a signal seen on the close of one bar is filled at the open of the next bar, never at the close that produced it. A stop order placed at the range edge usually fills a tick or more beyond it in a fast market. Set the backtest to fill on the next bar and add stop-order slippage on every entry. AlphaPrime (Valynox) backtesting applies fees, slippage and next-bar execution to every candidate, so breakout rules are tested with the fill they would really get.

A breakout: bars trade inside a range, a signal bar closes above the range high, the fill comes on the next bar above the range edge, and the stop sits inside the range Breakouts fill beyond the edge Signal on close, fill on next bar Signal Fill Range high Stop Real fills sit past the range edge. So: test next-bar fills with slippage.
A breakout fires on the next bar after the range is broken: the fill sits above the range edge, the stop sits inside it.

A good breakout test shows a similar result when the range length is changed a little (25, 30, 35 minutes). A bad result shows one setting far ahead of its neighbours, which usually means the setting fits a few big days. In AlphaPrime's strategy generation, the range length can be a parameter range, so 25, 30 and 35 minutes are tested together and a setting that leads its neighbours by a wide margin stands out.

When does trend-following suit ES, NQ and MNQ?

Trend-following suits ES, NQ and MNQ when you can hold for days or weeks, test on hourly or daily bars, and fund the overnight margin. The move per trade is large, so costs matter less than for the other families, but three settings decide whether the test is honest.

  1. Full-session data. Trend positions are held through the night session, so the backtest needs bars that cover it, otherwise overnight gaps appear as free jumps in your favour or against you.
  2. Roll handling. A roll is the point where a continuous price series switches from the expiring contract to the next one. If the price gap at the roll is not adjusted, a trend strategy may book a gain or loss that no trader could have had.
  3. Overnight margin. Overnight margin is the money the exchange and broker require to hold one contract past the daily close. It is higher than intraday margin, so the account in the backtest must cover it on the worst day.

MNQ is one tenth of NQ: $2 per index point against $20. The same rules can be tested on a small account with one tenth of the point risk per contract, which makes the overnight margin and drawdown fit the account you really have.

How do you test mean reversion without fooling yourself?

Mean reversion wins often and loses big, so the stop is part of the strategy and the backtest must prove the result does not rest on a few days. The typical trade is small and quick: buy a dip below a moving average on 1- to 15-minute ES bars, exit at the average. One day that keeps falling can give back many such trades.

A hard stop is an exit at a fixed distance that closes the trade no matter what the rule says. Run the test with and without it: if the version without the stop looks better, the difference comes from a few trades that happened to recover, and the next one may not.

The same mean reversion trades with and without a hard stop: eight small wins, then one trend day whose loss stays capped with the stop and runs far larger without it The stop is the strategy Same trades, with and without a stop Small wins Hard stop Loss capped No stop Loss runs Without a stop, one day erases wins. So: always test with a hard stop.
Many small moves back to the average, then one trend day: the hard stop decides how large the one loss becomes.

Three checks keep a mean reversion test honest:

  1. Limit order fills. Count a limit order as filled only when price trades through it, not when it merely touches the level.
  2. Remove the worst days. Rerun the test with the five worst and five best trading days removed. A sound result changes a little; a fragile one changes its sign.
  3. Crash periods inside the sample. Keep fast sell-offs in the date range. A strategy that was only tested on calm years has not been tested.

What are time-of-day and volatility strategies, and how are they tested?

Time-of-day strategies trade only in a chosen window, volatility strategies size or filter trades with a range measure, and both need session times set in one time zone and a test of neighbouring settings. A common window is the first two hours after the US cash open, 08:30 to 10:30 Central Time, with a flat-by time before the 15:00 cash close. The futures market hours tool shows these sessions in your own time zone.

The time zone is the most common mistake. If the data file is stamped in exchange time and the rule is written in your local time, the backtest trades different hours than the live strategy, and the result is for a strategy you will never run. Pick one zone, convert the data once, and write every window in that zone.

The ES cash day from 08:30 to 15:00 Central Time: the strategy trades only from 08:30 to 10:30 and is flat for the rest of the day, with every time written in the same zone One window, one time zone ES cash day: trade only the window All times in CT Trade Flat 08:30 10:30 15:00 Mixed zones trade the wrong hours. So: convert the data once to one zone.
A time-of-day window is one slice of the ES session, placed in one time zone so the backtest trades the same hours as live.

A volatility strategy uses a range measure such as the ATR (Average True Range) to skip quiet days, to set the stop distance, or to trade fewer contracts when the range is wide. Test the thresholds next to the chosen one as well: if ATR above 20 points works and ATR above 18 or 22 does not, the threshold fits noise. The same applies to windows: compare 08:30–10:00, 08:30–10:30 and 08:30–11:00 side by side.

How much commission and slippage per contract should the backtest use?

Set the round-trip cost per contract from the tick value plus at least one tick of slippage per side, then compare it with the average trade of each strategy type. One tick is 0.25 index points on ES, NQ and MNQ; it is worth $12.50 on ES, $5.00 on NQ and $0.50 on MNQ. The futures contract calculator shows tick size, tick value and the profit or loss of a trade for each contract. Exchanges can change contract values; check the exchange's contract page for the current specification.

Hypothetical example

An ES account pays $2.50 commission per side, so $5.00 per round trip. Add one tick of slippage per side, two ticks in total, worth $25.00. The round-trip cost is $30.00 per contract. A breakout strategy with a gross average trade of $90 keeps $60 after costs. A trend-following strategy with a gross average trade of $300 keeps $270. A mean reversion strategy with a gross average trade of $35 keeps $5, and one extra tick of slippage turns it negative.

A hypothetical 30 dollar ES round-trip cost taken from each type's average trade: trend-following keeps 270 dollars of 300, breakout keeps 60 of 90, mean reversion keeps only 5 of 35 Cost decides who survives ES average trade, $30 round trip Net kept Cost $30 Illustration Trend $270 Breakout $60 Mean reversion $5 Mean reversion keeps only $5. So: add costs before the first search.
The same round-trip cost takes a different share of each type's average trade, so cost decides which types survive on which contract.

The smallest average trades, intraday mean reversion and fast breakouts, are the first to turn negative. Put the costs in before the first search run, because a search ranked without costs fills the top of the list with exactly these strategies.

How do you compare strategy types on the same data?

A fair comparison fixes every setting before the first run and then ranks by net result relative to drawdown, not by gross profit or win rate. Drawdown is the fall from an equity high to the following low; relating net result to it rewards strategies that earn without deep holes, and it does not favour the type that happens to trade most.

  1. Choose one contract and one bar size for all types.
  2. Choose one date range and one in-sample/out-of-sample split. In-sample is the data used to pick settings; out-of-sample is the data kept aside to check them.
  3. Set the same commission, slippage and next-bar execution for every run.
  4. Run every strategy type with these settings.
  5. Rank by net result after costs relative to drawdown.
  6. Walk forward the top candidates: re-tune on one window, test on the next, repeat.
A fair comparison in three steps: fix all settings, run every strategy type, rank by net result relative to drawdown; changing a setting sends you back to step one Fix first, rank last Same costs, bars and data for all 1 2 3 Fix all settings Run every type Rank by net ÷ drawdown Change a setting? Restart. Changed settings break the comparison. So: lock settings before the first run.
Fix every setting first, run all types on it, then rank. Changing a setting after seeing a result restarts the comparison.

The more variants you compare, the more the best one is inflated by chance; Bailey, Borwein, López de Prado and Zhu (2014) show how quickly this happens. Keep the out-of-sample part for the final check, and expect the winner to lose some ground there. In AlphaPrime's strategy generation, the rules, parameter ranges and search constraints for each type are set before the search, so every candidate is produced under the same conditions.

How do you build a small portfolio from different strategy types?

Two or three types that lose at different times, such as trend-following and mean reversion, usually give a combined drawdown smaller than the sum of the parts, and the portfolio backtest must check daily return correlation and total margin on the worst day. Trend-following tends to bleed in ranging years and mean reversion tends to break in trend years, so their bad periods rarely line up.

Correlation is a number between −1 and +1 that says how closely two series move together; near zero means their daily gains and losses are mostly independent. Compute it on daily returns of each strategy, not on equity curves, because two rising curves always look correlated. Then line up the daily returns, add them, and measure the combined drawdown with the same tool you used for each strategy alone. The portfolio backtesting guide covers the steps in detail.

Illustrative equity curves: trend-following has its drawdown early, mean reversion has its drawdown late, and the combined curve has only shallow drawdowns because the bad periods do not line up Lose at different times Combined drawdown is shallower Equity Drawdown Trend Reversion Together Their bad periods rarely line up. So: pair types by when they lose.
Two strategies that lose at different times combine into one curve whose deepest drawdown is shallower than either alone.

The margin check is simple: find the day when both strategies held their largest positions at the same time, add the overnight margin of every contract, and compare it with the account. A portfolio that needs more margin than the account on one day is not tradable, even if its combined curve looks smooth.

What this means if you use AlphaPrime: define each strategy type with its own rules and parameter ranges, run them all with the same costs, bar size and execution settings, and read the net result relative to drawdown before anything else. Export the survivors to your trading platform through platform export.

Frequently asked questions

Which futures algo strategy type is easiest to test first?

Trend-following on daily or hourly bars is the simplest to test correctly. Trades are few and long, so one tick of slippage barely changes the result, and the rules are easy to check by hand. Mean reversion and fast breakouts need more care with fills and stops, so start with them after the test setup is proven.

Can I run the same algorithm on ES, NQ and MNQ?

Yes, but keep the stop and target in index points or in ATR units, not in dollars, because one point is $50 on ES, $20 on NQ and $2 on MNQ. Test each contract on its own, since NQ moves more points per day than ES. MNQ follows NQ tick for tick, so NQ rules usually transfer, with the cost per contract changed.

How many trades does a backtest need to compare strategy types?

Enough that the average trade is not driven by a handful of outliers. Many traders want at least a few hundred trades for intraday types and as many years as the data allows for trend-following, which trades rarely. If removing the five best trades flips the sign of the result, the sample is too small to compare.

Do I need a GPU to backtest futures algorithms?

No. The same tests run on a CPU; a graphics card only changes how long a large search takes. AlphaPrime runs on a CPU or an NVIDIA GPU with the same features, and the speed depends on the processor or graphics card, the data length and the research settings.

Should the backtest use day-session or full-session data?

Use full-session data for trend-following and for any strategy that holds overnight, so gaps and the night move are inside the test. A time-of-day strategy that is always flat before the close can use day-session bars, but the opening range must then start at the exact cash open in the time zone of the data.

References

  • Bailey, D. H., Borwein, J., López de Prado, M., & Zhu, Q. J. (2014). Pseudo-Mathematics and Financial Charlatanism: The Effects of Backtest Overfitting on Out-of-Sample Performance. Notices of the American Mathematical Society, 61(5), 458–471.