54 NQ Backtests Inspired by Jake Bernstein: An Independent Study

NQ: the top five for each schedule

This retrospective comparison ranks all 54 intraday configurations by cumulative net profit from January 4, 2016 to December 31, 2025. For single-day trading, first choose each configuration’s highest-profit weekday using that same period, then rank those selected-day profits. Use January 1, 2026 as the analytical cutoff, retain those lists and weekday choices, and evaluate from January 2, 2026 to September 4, 2026. No 2026 result enters either ranking.

NQ futures only. Net includes the $5 round-trip fee and original modeled slippage, at $20 per point for one normalized position, without compounding. Positions merge partial exits. The 2026 drawdowns start from zero at the cutoff, follow realized exit cash flows and exclude open-position losses; they are shown as loss magnitudes. These lists rank historical profit, with no additional robustness filter.

The selections were reconstructed after the fact using only pre-2026 results. The underlying data had already been inspected during research, so 2026 is not an untouched holdout. Excluding it from the ranking does not remove earlier research choices.

Top five trading all weekdays

The training and evaluation columns both include Monday through Friday. The strategy order is fixed by 2016–2025 profit.

Scroll tables sideways to view all columns.

2 of the five were profitable in the 2026 evaluation. The two highest training-profit configurations both lost money afterward.

Top five trading one selected weekday

Both profit columns and all 2026 metrics cover only the chosen weekday. The 2016–2025 data selects Thursday for B29 and B36, Wednesday for B39, and Friday for B07 and B09.

4 of the five were profitable in 2026. The lists contain different configurations, so the difference from the all-weekday table does not measure the effect of weekday selection on the same five strategies. The two breakout variants also have closely related entries; these are individual strategy results, not a combined portfolio.

The same five strategies, with and without weekday filtering

Keep the all-weekday top five and their order fixed by 2016–2025 profit. Compare each configuration’s 2026 results on all weekdays with its best weekday selected using only 2016–2025 results. Both columns cover January 2–September 4, 2026. Change in net is selected-day profit minus all-weekday profit; position counts show the change in trading frequency.

The selected weekday improves 2026 net profit for 3 of these five configurations. This holds the strategy roster constant, but trading one weekday also changes exposure and sample size. The results describe this historical filter effect; they do not establish a persistent weekday advantage or a combined portfolio.

Three hypotheses for further research

The configurations in this research shortlist are B36 on all weekdays, B07 on Fridays, and B09 on Fridays. These schedules define simulation hypotheses, not suggested trading actions. The research order weighs the results now available, including 2026, separately from the pre-2026 profit rankings above; it does not rank suitability for an investor.

Scroll sideways to view all columns.

2024 and 2025 are complete calendar years; 2026 runs through September 4. For each breakout variant, Friday was selected using only the data before that evaluation year: 2016–2023 for 2024, 2016–2024 for 2025, and 2016–2025 for 2026. B36 trades all weekdays in every year. All figures use NQ and the same sizing and modeled costs.

B36: the book’s example, tested on NQ

The crossover method, 30-minute bars and 18-period simple moving average come from the book. In The Compleat Day Trader (1995), Chapter 4, pages 35–37, Bernstein illustrates that combination on T-bond and Swiss franc futures charts. B36 retains it. Our contribution is testing the example on NQ under explicit session, execution and cost assumptions. The implementation used for these results is:

• Market and session. We use NQ futures and build 30-minute candles from one-minute data, anchored to the New York cash session, 09:30–16:00. Indicator history carries across eligible sessions and resets after excluded sessions or contract changes.
• Execution and exits. A completed-candle crossover triggers an order at the next minute’s open. An opposite crossover closes and reverses the position. Remaining positions close at the cash-session end; each day starts flat and waits for a fresh crossover. This version adds no separate stop-loss or profit target.
• Sizing, costs and schedule. We use one normalized position at $20 per point, a $5 round-trip fee and a quarter point of adverse slippage per market fill, without compounding. The shortlisted version trades all eligible weekdays.

Bernstein ultimately advised against this simple crossover method for day trading in that chapter. B36’s place here reflects our NQ backtest results and our choice for further testing; it is not an endorsement by Bernstein.

1. B36 — Price versus its moving average, 30-min: the book’s example, tested on NQ, all weekdays. The first hypothesis in our research order. It was profitable in 2024, 2025 and the partial 2026 period, and all-weekday trading beat its preselected Thursday in each of those years.
2. B07 — Full-target breakout, no early protection: Fridays. It precedes B09 in the research order because the complete-year cost, profit-concentration and weekday-control results are available. It remained positive at the $20 fee and after removing its five best trades, with all six 2020–2025 evaluation years positive. It nevertheless failed both statistical gates described below.
3. B09 — Full-target breakout, no early protection or gap filter: Fridays. An alternative to B07 with higher Friday profit in 2025 and 2026. It was not included in the same six-configuration weekday robustness study, so its relative statistical reliability is unresolved. Removing the gap filter changes which entries qualify, but the two variants still share much of the same breakout exposure. B07’s priority is an editorial testing choice, not an established advantage over B09.

What B07 passed and failed

In the complete 2020–2025 NQ evaluation, choosing the best weekday from expanding prior-year data selected Friday each year. B07 made +$89,270, stayed positive at the $20 round-trip fee (+$84,185), and remained positive after removing its five largest winning positions (+$68,620). It was profitable in all 6 years, meeting the screen’s minimum of four positive years.

Two statistical requirements failed. The 95% interval for its dollar advantage over random annual weekday selection was −$3,322 to +$111,441, crossing zero. Its Holm-adjusted tail probability was 0.7615, above the screen’s 0.05 limit. The control chooses one weekday per year at random; its expected profit is 20% of all-weekday profit. The interval resamples four-week blocks within each year, and the Holm adjustment covers 24 primary comparisons across the two historical feeds. These checks do not remove earlier research selection.

B09 has no result in that same robustness study. Its higher recent Friday profit warrants comparison in the next test, while B07’s documented cost and concentration checks explain why we start there. Neither has an established statistical advantage from this screen.

Execution costs: higher-slippage reruns

These are full strategy re-simulations on all eligible weekdays from January 4, 2016 through September 4, 2026. Each column keeps the $5 round-trip fee and increases adverse slippage per market or stop fill. Target orders still fill at touch without queue friction. Changed fills can change later exits and trade counts. The B07 row here uses all weekdays; the shortlist uses Fridays.

The base breakout turns negative at 1.00 point per side. B36, the full-target breakout and the hourly 8 Open/Close approximation retain positive historical totals in these scenarios. This measures sensitivity to the modeled execution costs; live fills remain unverified.

The $20 round-trip column in the full configuration table changes fees only, holding fills and slippage fixed. Both checks matter: higher commissions and worse fills affect results differently.

The index

An index of the methods covered by the reviewed materials, with published context and our test coverage. A named technique may lack a complete mechanical specification. Similarity between book and course methods does not establish a direct historical lineage. S1–S4 refer to the reviewed course sessions; book references use printed page numbers.

All 54 configurations on NQ

Ranked by full-period NQ net profit, January 2016–September 2026, after $5 per round trip and a quarter-point of slippage on every market fill. Shaded rows beat the benchmark. t is the unadjusted t-statistic of the mean position profit. It does not account for dependence between trades or the many configurations inspected. The last column tests a $20 round-trip fee with fills and slippage unchanged.

Positions merge the legs of a scaled position into one trade. 34 of 54 finished positive; 12 beat the benchmark. These are full-period descriptive results; the top-five tables above rank only pre-2026 profit.

The 132-setting single-moving-average grid

The 1995 book tests price crossing its own moving average and tells the reader to stay away. B36 tests the book’s 30-minute example on NQ, with our backtest assumptions described in the B36 explanation. We also tested the existing grid on NQ: six bar sizes from 5 to 60 minutes, eleven average lengths from 5 to 50, simple and exponential, long above and short below, reversing on the cross and flat at the close. 93 of 132 settings were profitable. The grid covers 2016 through September 4, 2026. Each cell shows net profit in thousands and positive calendar periods out of eleven.

0 settings were positive in all eleven calendar periods. In the completely covered 2016–2017 period, 0 of 132 settings were profitable after costs. Here are the top eight by full-period net profit; this ranking uses all years, including 2026, and is descriptive.

The profitable settings cluster toward slower signals, but every setting lost money across 2016–2017. These related settings are not independent confirmations, and choosing the strongest after inspecting the whole period does not establish a future trading edge.

The 240-setting dual-moving-average grid

Here we cross a fast moving average with a slow moving average of closing prices. This additional search uses the same 2,659 NQ sessions from January 4, 2016 through September 4, 2026. It expands B35 and is separate from the 54 main configurations and the single-MA grid above.

The grid combines six bar sizes (5, 10, 15, 20, 30 and 60 minutes), four fast lengths (3, 5, 8 and 12 bars), five slow lengths (18, 24, 36, 54 and 72 bars), and two average types. Both averages use the same type: simple (SMA) or exponential (EMA). These ranges are our research choices.

A completed-bar crossover opens or reverses a position at the next minute’s open. Each cash session begins flat and any remaining position closes at session end. There is no separate stop or profit target. Costs remain $5 per round trip plus a quarter point of adverse slippage on each fill, at $20 per NQ point with one position and no compounding.

The 3/24 and 3/54 reference settings

Book I, Chapter 4, pages 37–39 discusses 3/24 averages and illustrates them on a 30-minute T-bond chart. Its separate performance summary uses 3/54 averages on 30-minute S&P futures, with an $800 stop and no deduction for costs. The NQ rows below adapt those average lengths under our assumptions; they do not reproduce that historical S&P test. The 5-minute SMA 3/24 row reproduces all 13,457 B35 positions and its −$97,040 net result.

Choose a bar size to explore the settings

Rows are fast-average lengths; columns are slow-average lengths. Each cell shows net profit rounded to the nearest $1,000 and positive calendar periods out of eleven. Green is positive and red is negative. The 2026 period ends September 4. Scroll each table sideways on smaller screens.

224 of 240 dual-MA settings finished positive after base costs. 0 were profitable in all eleven calendar periods; 39 were profitable across 2016–2017. These counts describe related variants on the same data.

Top eight by full-period net profit

This ranking uses every year, including partial 2026. It is an exploratory comparison on previously inspected data, with no untouched holdout or correction for searching 240 settings. The highest historical profit is not a validated setting for future trading.

Average history carries across eligible sessions within the same contract segment, including available pre-2016 warm-up, and resets after excluded sessions or contract changes. EMAs use a recursive weight of 2 / (length + 1), starting at the first segment close and waiting for the required number of bars. Bars start at the cash-session open; the shortened final bars for 20- and 60-minute settings enter indicator history but cannot open a new position. All 240 trade ledgers were reconciled against source minute prices and summary statistics, with separate crossover-event replays for eight reference settings.

Method, data and caveats

• Data. Databento CME Globex MDP 3.0 data (GLBX.MDP3), using the ohlcv-1m schema and NQ.v.0 continuous symbol. The full-period results cover January 4, 2016–September 4, 2026: 2,659 complete New York cash sessions, containing 1,037,010 eligible minute bars. The period starts in 2016 because earlier history has sparse session coverage. Within this period, 21 scheduled early closes and 4 incomplete sessions were excluded.
• Session and contract history. Weekdays follow the New York entry session date. The continuous series selects the highest-volume contract using the previous day’s trading volume. Prices are unadjusted across futures contracts; prior-close prices and indicators reset after excluded sessions and actual contract changes. Mixed-contract sessions are excluded. Weekday filtering retains the original indicator history and exits. See Databento’s symbology documentation.
• Validation. The NQ review reconciles positions, weighted exit legs, annual and monthly profits, the benchmark and the moving-average reference. Full-period metrics use the 2016 onward sample. The weekday tables and shortlist use the same NQ ledger and cost assumptions.
• Fills. Completed-bar market signals fill at the next minute’s open with a quarter point of adverse slippage. Stops fill at their level or a worse opening price if the minute gaps through, plus the same slippage. Targets fill at touch without queue friction. Ambiguous stop/target touches generally take the stop first; the shared routine permits a target first when the minute opens beyond a resting target, while channel tests retain stop-first handling. $5 per round trip, $20 per point, one normalized position, no compounding.
• Statistics. A position is the whole scaled trade with its weighted legs merged. Fractions of normalized exposure are accounting weights; partial exits require multiple whole NQ contracts in practice. Scaling changes dollar P&L and capital needs, so these figures are not account returns. Drawdown follows realized exit cash flows from zero at the start of each reported period and excludes open-position losses. The t-statistic is unadjusted. Years positive counts calendar periods, with 2026 partial.
• Educational and research purposes only. These are hypothetical simulations on one instrument and one period. This article does not provide investment, financial or trading advice and does not assess anyone’s financial circumstances, objectives or risk tolerance. No instrument, strategy, setting or schedule is recommended. See Section 12 (Disclaimer and Limitation of Liability) of our Terms of Service and the hypothetical performance disclosure at the top of this article.

Sources and attribution

• Course. Jake Bernstein, 3 Days – 5 Ways: Short-Term and Day Trading, the 2025 recordings reviewed for this study, sessions 1–4. S1–S4 in the index identify those sessions and timestamps. Missing follow-up instructions and proprietary formulas limit the test coverage. The author’s official website provides information about his teaching and publications. This study does not verify his live trading record.
• Book I. Jake Bernstein, The Compleat Day Trader (1995). Chapter and printed-page references appear in the index; the B36 example is discussed in Chapter 4, pages 35–37. Published examples and our NQ execution assumptions are identified separately.
• Book II. Jake Bernstein, The Compleat Day Trader II (1998). The index cites the relevant chapters and printed pages. Some published overnight, weekday-specific or exit variants are outside the tested implementation.
• Market data and contract size. Databento’s continuous symbology describes the contract-selection convention. CME’s NQ specifications give the $20-per-point multiplier and 0.25-point minimum tick. The method section records the data schema, session exclusions and contract resets used here.