Intraday Momentum Strategies From 3 Research Papers, Backtested on ES, NQ, Gold and Bitcoin

00The short answer

One of the three rules passed our tests, on one market only. The noise area breakout of Zarattini, Aziz and Barbon passed all six of our gates on Nasdaq-100 futures (NQ) in the New York session: +$231,557 per contract over 1,504 trades after costs, long and short, and +$152,227 long only. The same rule did not pass on ES, gold or Bitcoin, or in the Asian or London session of any market. It is now in a forward test on new data.

• The noise area breakout paid its costs about ten times over on NQ, and not at all on Bitcoin. In the New York session, a trade made about 10.8 times its cost before costs on NQ, 1.8 times on ES and less than its cost on gold; around the clock on Bitcoin it made about nothing.
• The 5-minute opening range breakout of Zarattini and Aziz reproduced the paper's numbers and still failed. On NQ it won 24.3% of its trades and made +0.107 R a trade before costs, against about 24% and +0.13 R in the paper, and +$70,146 after costs. But the same trades with their directions taken from other days made money too, and without its 10 best trades it lost.
• The intraday momentum of Gao, Han, Li and Zhou lost money even before costs in all 8 versions on ES and NQ futures from 2020 to 2026: in these years, the first half-hour said nothing useful about the last.
• Reproducing a paper is not the same as finding an edge. Two of the three rules made money in at least one version; only one beat its control and passed every other gate as well.

01Who we are, and why you can check these numbers

RIP Backtest is an independent site run by one retail trader. We test published trading rules and popular strategies under one fixed protocol: the rules and the checks are written down and dated before any data is run, costs are charged on every trade, the years are split into periods that are judged separately, and every version we try is published, including the ones that fail.

We sell nothing. No signals, courses, indicators or robots; there are no affiliate links on the site, and this guide names no broker or vendor. It also runs no new test: every number below comes from the seven tests already on the site, each with its specification, code and complete results, and every section links to the full write-up. Read more about the site and our testing protocol.

02The three papers

Each paper argues, in its own way, that the first part of the US trading day carries information about what comes next. Two of the three titles use the words intraday momentum, and all three rules bet that an early move continues.

• Zarattini, Aziz and Barbon (2024), “Beat the Market: An Effective Intraday Momentum Strategy for the S&P500 ETF (SPY)”. For each minute of the session, average how far the price has usually moved from the open at that time of day over the last 14 days; that band around the open is the “noise area”. A close outside it is a breakout: buy above, sell short below, checked every 30 minutes, with a trailing exit at the band or the session's VWAP, and everything closed at the end of the day. The paper tested it on SPY, the S&P 500 ETF. We tested the entry and the exit unchanged, one contract, one ounce or one bitcoin at a time, without the paper's position sizing.
• Zarattini and Aziz (2023), “Can Day Trading Really Be Profitable?”. Take the direction of the first 5-minute candle of the US session, enter at 09:35 New York, put the stop at the other side of that candle and aim for ten times the risk, or get out at the close. The paper tested it on QQQ, the Nasdaq-100 ETF, 2016 to 2023. We added the classic breakout of the same range, the version most traders mean by ORB.
• Gao, Han, Li and Zhou (2018), “Market Intraday Momentum (Journal of Financial Economics)”. On SPY, the S&P 500 ETF, 1993 to 2013, the return of the first half-hour, from the previous close to 10:00 New York, predicted the return of the last half-hour. The paper reports a statistical relation, not a trading rule, so we turned it into one: long or short for the last half-hour, in the direction of the first.

The papers tested their rules on US stock ETFs. We tested them on the data we have, which is not theirs: ES and NQ futures, spot gold and Bitcoin. A rule that only works on the instrument and the years it was found on is a weaker rule than one that works elsewhere too.

03How the seven tests were run

Every version had to pass the six gates of our protocol: a profit factor of at least 1.2 in validation; at least 50 trades in each period; a development t-statistic of at least 2.0, or 2.5 when the idea is tried in more versions (2.5 for the noise area and the half-hour momentum, 2.0 for the opening range breakout); a result above the 95th percentile of a random control; a positive result without the 10 best trades; and a positive holdout. Each of these data files had been used by earlier tests, so every holdout here is flagged as used, not clean.

04The results, side by side

Seven tests, 34 versions, two survivors. The last column but one shows the range of the result per trade after costs, from the worst version to the best.

Futures in dollars per contract, gold in dollars per ounce, Bitcoin in percent of the price. Each test's full write-up explains its version codes.

05The noise area breakout on NQ: the one that passed

In the New York session, 09:30 to 16:00, both versions passed all six gates. Long and short made +$231,557 per contract over 1,504 trades, +$153.96 a trade after costs; long only made +$152,227 over 747 trades, +$203.78 a trade. Their development t-statistics were 2.82 and 2.98 against the 2.5 required, their validation profit factors 1.33 and 1.27, their holdouts +$48,864 and +$47,433, and they ranked at the 100th and 99.6th percentile of the random control.

It is not one lucky stretch. Both versions made money in every calendar year from 2020 to 2026; in the long-and-short version the longs and the shorts both made money (+$152,227 and +$79,330); without their 10 best trades they still made +$107,255 and +$58,165; and removing the trades that touch suspect gaps in the data left both verdicts unchanged. They are the only versions on the site that passed every gate and kept that verdict.

Why it is still only a candidate. The same NQ file and years had been used by earlier tests, so the holdout is not clean; 22 versions of this rule were tried across four markets; the same rule failed on ES, gold and Bitcoin; and the Asian and London sessions of NQ itself did not work. A backtest that passes is where testing starts, not where it ends. The forward test, a paper series computed by the backtest's own code on data that came after 25 September 2026, started on 12 October 2026; its first data comes with the November export, and it needs about two years for a fair chance of supporting the result. Read the full NQ test.

06Why the same rule failed on ES, gold and Bitcoin

Outside the New York session of NQ, none of the other 20 noise area versions passed. One number explains most of it: how many times a trade earned its own cost before costs.

Average gross result of a trade divided by the cost of a trade, long-and-short version (Bitcoin: deciding every 30 minutes, as published).

• ES futures. New York came closest: +$38,508 over 1,546 trades long and short, above the control (percentile 97.8), but with a development t-statistic of 1.91 (2.5 needed) and a losing holdout (−$25,692). A trade made +$55.67 before costs, against $30.76 of costs. Read the ES test.
• Gold. Every version ended close to zero after costs, between −$0.34 and +$0.12 an ounce a trade; before costs the New York versions made $0.28 and $0.42 a trade, against a $0.30 round trip. Read the gold test.
• Bitcoin. Around the clock, deciding every 30 minutes as published, the long-and-short version made −0.004% a trade before costs and −0.104% after; deciding every 15 minutes, a second test run after the first result, it made +0.001% and −0.099%. Every version lost after costs in development, validation and the holdout. Read the Bitcoin test.
• The Asian and London sessions failed on every market: none of the 12 versions passed, and in London the rule lost money even before costs on both futures markets.

Read together, the four tests say the rule needs a market that moves far enough inside the US session to pay for its costs many times over. In these years NQ did; ES paid them less than twice, gold less than once and Bitcoin not at all. That describes these results, it is not a law, and the forward test on NQ is the real check.

07The 5-minute opening range breakout: the paper's numbers came back, the edge did not

The rule of Zarattini and Aziz takes the direction of the first 5-minute candle (09:30 to 09:35 New York), enters at 09:35, puts the stop at the other side of the candle and targets ten times the risk, or exits at the close. On NQ from 2020 to 2026, the paper's base numbers came back: 24.3% winners and +0.107 R a trade before costs, against about 24% and +0.13 R in the paper. After $15.76 of costs a trade, one contract made +$70,146 over 1,729 trades.

It still failed five of six gates. The same trades with their directions taken from other days made money too (+$18.66 a trade in the median run of the control), so the profit came from the shape of the trade, a small stop and a big target on a few trend days, more than from the first candle: the rule sat at the 79th percentile of that control. Without its 10 best trades it lost $35,116, and 2026 so far is a losing year for each of the four versions.

The classic breakout of the same range made more, +$93,733, and passed three gates, but it did no better than random directions either (percentile 57.6). Costs were not what failed it: on futures they took about 28% of the paper rule's gross result. Read the full opening range breakout test.

08Intraday momentum: the first half-hour said nothing about the last

Gao, Han, Li and Zhou found that on SPY from 1993 to 2013, the return of the first half-hour, measured from the previous close to 10:00 New York, predicted the return of the last half-hour. We turned it into a rule: at 15:30, go long for the last half-hour if the first half-hour was up, short if it was down. Four versions on each of ES and NQ: the first half-hour measured from the previous close (as in the paper) or from the 09:30 open, long and short or long only.

All eight versions lost money before costs, from −$15.67 to −$6.85 a trade on ES and from −$14.31 to −$1.66 on NQ, and more after costs of $30.76 and $15.76 a trade. None beat the same signals moved to other days (percentiles 20.6 to 43.4). In 2020 to 2026, on these two futures markets, the first half-hour said nothing useful about the last.

That does not refute the paper. It studied a different instrument and earlier years, and it reports a statistical relation, not a profitable rule after costs. What our tests show is narrower: the rule built from it did not make money on ES or NQ futures in the years we have. Read the ES test and the NQ test.

09What the three papers taught us

Reproducing a paper is not the same as finding an edge

The opening range breakout reproduced the paper's winners and its average trade before costs almost exactly, and it made money. Our control then showed that random directions with the same stop and target made money too. Checking a rule against a control with the same trade shape is the step most backtests skip.

One market is not a pattern

The noise area breakout passed on NQ and failed on three other markets and in two other sessions. That can mean the edge is specific to NQ's US session, or that NQ's strong trend in these years flattered it. Only new data can tell those apart, which is why a forward test, not a better backtest, comes next.

Costs decide where a day-trading rule can live

The same entries made about 10.8 times their cost on NQ and less than their cost on gold. Futures costs are small against the size of an index move inside a day; a fixed spread on a CFD or a percentage fee on crypto can be as large as the whole edge.

Sessions matter

All three papers are about the US session. The noise area versions in the Asian and London sessions failed on every market we tried, and London lost even before costs on both futures markets. A rule found in one session is not a rule for the whole day.

Data gets reused

Every one of these files had been used by earlier tests, so no holdout here is clean, and 34 versions on overlapping data leave room for luck. That is why we count and publish every version, and why a pass is called a candidate.

10How to read a day-trading paper before you trade it

1. Which instrument, which years? An ETF from 1993 to 2013 is not a futures contract in 2026.
2. Are costs in, at your size and with your broker? Compare the average trade before costs with your cost per trade; we needed several times the cost to survive.
3. Is the rule fully mechanical? Entry, exit, stop and session times must be fixed before you test, or the test fits itself to the data.
4. How many versions were tried, and which were picked after looking? The opening range paper also shows versions with other stops and targets chosen after seeing results; we tested its base rule.
5. Is there a period the rule was not fitted on? Judge it on years it has not seen.
6. Does it beat random entries with the same trade shape? A small stop and a big target can make money on trend days whatever the signal.
7. Does it work on a second market or session? If not, treat it as a hypothesis.
8. What is left without the 10 best days? A result that lives on a handful of trend days is fragile.
9. Test it forward, on paper or a demo account, before you risk money.

The full method is in our guide how to backtest a trading strategy without fooling yourself, and every result on the site is summed up in what the backtests found so far.

11What this guide cannot tell you

• Our data, not the papers'. The papers used US stock ETFs; we used futures, spot gold and Bitcoin, so our results test the rules, not the papers' own numbers.
• Rules as published. One contract, ounce or bitcoin at a time, without the papers' position sizing or leverage, and without discretion.
• Reused data, many versions. 34 versions on files used before; some results look better than they are by luck, some worse.
• The future. A rule that worked from 2020 to 2026 can stop working, and the forward test may still contradict the survivor.

This is educational content, not financial advice, and not a recommendation to trade any of these rules.

FAQCommon questions

What is an intraday momentum strategy?

A day-trading rule that bets an early move of the session continues: buy after the price breaks out of its usual range, or after a strong first half-hour, and close the trade before the session ends. Two research papers use the name: Zarattini, Aziz and Barbon (2024) for the noise area breakout, and Gao, Han, Li and Zhou (2018) for the first half-hour predicting the last.

Does the noise area breakout strategy work?

In our backtest it passed every gate on NQ futures in the New York session, long and short (+$231,557 per contract over 1,504 trades) and long only (+$152,227), from 2020 to 2026 after costs. It did not pass on ES, gold or Bitcoin, or in the Asian and London sessions. The NQ result is a candidate: a forward test on new data started on 12 October 2026.

Does the Beat the Market strategy work on ES futures?

Not in our test. In the New York session it made +$38,508 per contract over 1,546 trades from 2020 to 2026 after costs, but its development t-statistic was 1.91 (2.5 needed) and it lost in the holdout (-$25,692).

Does the 5-minute opening range breakout work?

It reproduced the paper on NQ futures (24.3% winners, +0.107 R a trade before costs) and made +$70,146 on one contract from 2020 to 2026, but the same trades with directions taken from other days made money too, so it was buried. Read the test.

Does the first half-hour predict the last half-hour?

Not on ES or NQ futures from 2020 to 2026 in our tests: all 8 versions of the rule lost money even before costs, and none beat the same signals moved to other days.

Which market is best for intraday momentum strategies?

In our tests of the noise area breakout, NQ futures: a New York trade made about 10.8 times its cost before costs, against 1.8 times on ES, less than once on gold and nothing on Bitcoin. The bigger the move inside the session compared with the cost of a trade, the better a day-trading rule's chances.

How were these research papers tested?

With the rules as published, one contract, ounce or bitcoin at a time, on 1-minute data: ES and NQ futures 2020 to 2026, gold 2004 to 2026, Bitcoin 2017 to 2026, with rules written and dated before the data was run, costs included on every trade, results split into development, validation and holdout periods, a random-entry control as the benchmark, and every variant reported. Read the protocol.

12Check our arithmetic

Code provenance: our own code. No new test was run for this guide.

13Corrections

None since publication.

+More tests and guides

• Noise area breakout nq
• Opening range breakout strategy
• Intraday momentum strategy nq
• Noise area breakout es
• Free prop firm challenge calculator
• What all the backtests found so far
• The backtesting protocol
• How to backtest a trading strategy without fooling yourself
• Backtesting glossary