Overfit cover card on a flat dark gradient, kicker 'Internal Bar Strength': the headline 'No 14-for-14 year. Above random: 4 of 4.' over the line '200 backtests on NQ and ES futures', with a corner badge reading 'NQ, ES, QQQ, SPY'.

April 11, 2025

Internal Bar Strength (IBS) Strategy: 200 Backtests on Nasdaq and S&P 500 Futures

9 min read

Written by Ali Casey, founder of StatOasis and AlgoChef, creator of the Algo Trading Masterclass (ATM), with over 10 years of experience building systematic trading tools - building algorithmic strategies, testing ideas with data, and teaching traders how to build structured, portfolio-based trading workflows.

Published April 11, 2025 · Updated October 28, 2026 · Method

← Back to Research

The short version

I retested my 2025 Internal Bar Strength study on NQ and ES futures, plus QQQ and SPY, buying at the next open. The core idea holds. IBS below 0.25 made $307,587 on one NQ contract over 26 years, with a worst drop of $74,350, and it beat random entries on every market. The headline did not hold. No NQ long setting went 14 for 14 in 2024, and the same rule lost $50,460 on NQ in 2022.

TL;DR: the answer box

IBS worked as a short-term edge on all four markets. All 200 long settings (10 buy levels x 5 exits x 4 markets) made money before costs.
The plain rule did not reproduce the 2024 headline. On NQ in 2024 the rule made $21,110 from 35 trades, 20 winners. Across 5,519 setting-years with 10 or more trades, 0 had no losing trade.
2024 was not a down market. One NQ contract held all year made $63,445.
It had losing years. 7 of 26 full years on NQ lost money. The worst was 2022, when the index fell, at -$50,460.
On NQ, the profit all came after April 2013. From 2000 to 2013 the rule lost $11,393.
0.25 is a fine line, not a special one. The lowest levels earned more per trade than the highest. Higher levels traded more often.
The long side was far stronger. Every long setting made money. Selling a strong close, the mirror trade, made money in 10 to 33 of 50 settings, with a typical Ret/DD of -0.49 to 0.25.

What is Internal Bar Strength (IBS)?

Internal Bar Strength tells you where the day closed inside its own range.

IBS = (close minus low) / (high minus low)

It reads 0 when the market closes at the low of the day. It reads 1 when it closes at the high. That is the whole indicator.

Three drawn daily candles side by side. The first closes near its low, labelled a weak close with IBS near the low. The middle one closes halfway, IBS 0.5. The last closes near its high, IBS near the high. The bottom quarter of each candle's range is shaded as the buy zone, IBS below 0.25. — IBS reads 0 at a close on the low and 1 at a close on the high.

The trade idea is mean reversion: the bet that a market that just closed weak will bounce. So you buy after a low IBS reading and get out fast. A 2023 paper on country ETFs by Pandey and Joshi tested IBS as a short-term signal. The popular open IBS strategy on TradingView ships with lines at 0.2 and 0.8, but no tested results.

Here is one real trade, so you can see it work.

Daily candlestick chart of NQ futures from late November 2024 to early January 2025, with IBS bars underneath and a dashed line at 0.25. On 2024-12-18 IBS reads 0.18. An up-arrow marks the buy at the next open on 2024-12-19 at 22,922.00 and a down-arrow the sell at the open on 2024-12-23 at 22,988.20: +66.20 points, one contract, $1,325. — A typical 2024 trade: IBS 0.18, held two days, $1,325 on one contract.

On 2024-12-18, NQ closed with IBS at 0.18, under the 0.25 line. The trade bought the next open, 2024-12-19, at 22922.00. That day closed below the entry, so the trade held one more day. It sold the open of 2024-12-23 at 22988.20. +66.20 points, $1,325. Of the 35 trades in 2024, this is the one closest to the middle result. A typical trade, not the best one.

How I tested the Internal Bar Strength strategy

I ran every test through the same engine every StatOasis study uses.

Buy: IBS at the close is below a level. Buy the next open. I tested ten levels, 0.05 to 0.50.
The profit exit (my 2025 exit): if the trade is in profit at the close of the day you bought, sell the next open. If not, sell the open one day later. Every trade is out within two days.
Four plain exits beside it: sell the open 1, 2, 3 or 5 days after you buy.
The main rule I judge, fixed before the run: IBS below 0.25 with the profit exit.
Markets: NQ (Nasdaq 100 futures) and ES (S&P 500 futures) from 2000-01-03. QQQ from 1999-03-10. SPY from 1993-02-01. All to 2026-08-20. Futures trade one contract. QQQ and SPY trade $35,000, no compounding.
The short mirror: the same levels short, when IBS closes above 1 minus the level.

The fills: Execution rule v2. The signal is read at the close. The trade buys the next open, and every exit also fills at an open. One trade at a time. You can't buy at a close you have not seen yet, so this test doesn't either.

That makes 400 backtests, 200 of them on the two futures markets. The main figures have no costs. ES, QQQ and SPY also carry a figure after costs. NQ does not: the engine has no tick size for NQ, and it refuses to guess one.

The scores, in plain words:

Win rate: the share of trades that made money.
Drawdown: the biggest drop from a high point. I call it the worst drop.
Ret/DD: net profit divided by the worst drop. Profit per dollar of pain. Higher is better.

Ranked bar chart of Ret/DD for IBS below 0.25 with the profit exit. QQQ 11.50 from 1,226 trades and $117,002. ES 6.98 from 1,130 trades and $219,988. SPY 6.29 from 1,452 trades and $61,946. NQ 4.14 from 1,126 trades and $307,587. — The same rule made money on all four markets. QQQ scored best on risk.

On NQ the rule won 64.4% of 1,126 trades, with a Ret/DD of 4.14. On ES it won 66.4% of 1,130, with a Ret/DD of 6.98. ES after costs still made $187,218. The full table is in the reference section at the end.

Did the IBS strategy really make $45K in 2024?

No, not the plain rule. My 2025 article said one NQ contract made $45,000 in 2024 from 14 trades, all winners. I can't reproduce it.

Bar chart of NQ profit in 2024 with the profit exit at each buy level: below 0.05 $4,815, 0.10 $37,665, 0.15 $37,775, 0.20 $28,585, 0.25 $21,110, 0.30 -$2,450, 0.35 $2,020, 0.40 -$4,500, 0.45 $8,420, 0.50 $18,685. A dashed line at $45,000 marks the old headline. No bar reaches it. — No buy level of the 2025 rule reached $45,000 on NQ in 2024.

IBS below 0.25 took 35 trades on NQ in 2024. 20 won. It made $21,110. The best level with the same exit, below 0.15, made $37,775. Across all 50 long settings, 2024 ran from -$31,615 to $105,795. Not one had 14 trades and 14 winners.

Then I widened the search. Every market, every long setting, every calendar year with at least 10 trades.

Two dial charts. Left: the best win rate in any of 5,519 setting-years with 10 or more trades was 94.4%. Right: IBS below 0.25 on NQ in 2024 won 57.1% of its 35 trades. — Across 5,519 setting-years, the best was 94.4%. Nothing reached 100%.

5,519 setting-years. Zero perfect years. The best was 94.4%. The NQ rule's own best year was 2017, 25 wins in 27 trades.

So where did $45,000 come from? Most likely a filtered version. In 2025 I listed four filters: ADX above 20 or 25, Bollinger Band width, ATR and volume. I gave no lengths for the last three, and this engine has no ADX, so I could not retest them. Either way, 14 trades is too few to tell skill from a lucky run. At a 64.4% win rate, 14 wins in a row has a 0.21% chance in any one run of 14 trades. That is the win rate to the 14th power, so it treats each trade as independent of the last. It sounds rare. But 26 years hold a lot of runs. The longest winning streak on NQ, across the 50 settings, was 21 trades, and it sat inside a history with losing years. That streak came from another setting, though. This rule's own longest on NQ was 11.

And 2024 was not a down market. One NQ contract bought at the start of 2024 and held to the end made $63,445. I put a good year in a headline and called it the strategy. Early on I lost roughly $270K to overfitting and leverage. One great year in a headline is that same trap.

Does the IBS strategy work in every market regime?

No. My 2025 article said it "works very well in all market regimes". The NQ calendar years say otherwise: the rule lost money in 7 of 26, including the falling year 2022.

Bar chart of NQ profit by calendar year from 2000 to 2026 for IBS below 0.25 with the profit exit, with white dashes for one contract held all year. Losing years are orange. 2022 lost $50,460 while holding lost $111,280. 2024 made $21,110 while holding made $63,445. — 7 losing years in 26 on NQ. The worst came when the index itself fell.

The NQ rule lost money in 7 of 26 full years. The worst was 2022: 56 trades, 32 winners, -$50,460. One contract held through 2022 lost $111,280. So IBS lost less than holding. But it lost. From 2000 to 2002 it netted -$22,803, against -$61,320 for holding. I entered the markets in 1998, at the dot-com peak, so these were my first years too.

The typical NQ year made just $2,755. The best was 2025, at $88,185. ES was steadier: 3 losing years in 26, the worst 2000 at -$16,425.

Here is the whole run in one picture, with the pain underneath.

Two-panel chart of NQ futures from 2000 to 2026. Top: running profit of IBS below 0.25, ending at $307,587, against one contract bought and held, ending at $430,325. Bottom: how far the IBS account sat below its last high, with the worst drop of $74,350. — Buy and hold made more. IBS made $307,587 while in a trade 24.7% of the time.

Buy and hold made $430,325 on one NQ contract. IBS made $307,587. The worst drop on the IBS account was $74,350, and the largest single losing trade was -$19,900. There was no stop loss.

The IBS money came from a small slice of the calendar.

Donut chart: IBS below 0.25 with the profit exit on NQ was in a trade on 24.7% of days and in cash on the rest. — In a trade on 24.7% of days. Cash the rest of the time.

That is the real trade-off. Holding is in the market every day. IBS sat in cash about three days out of four.

Did the rule hold in both halves?

Not on NQ. I split each market's years at the middle day and scored the rule on each half.

Dumbbell chart of Ret/DD for IBS below 0.25 in the first and second half of each market's years. NQ -0.30 then 4.29. ES 0.91 then 6.26. QQQ 6.83 then 6.82. SPY 3.39 then 3.34. — On NQ the IBS below 0.25 rule's profit arrived late. On QQQ and SPY it held level.

On NQ, from 2000 to April 2013, the rule lost $11,393, a Ret/DD of -0.30. From 2013 on it made $318,980, Ret/DD 4.29. ES was thin early too, 0.91, then 6.26. QQQ held level: 6.83 and 6.82. SPY too: 3.39 and 3.34.

NQ traded one contract and QQQ a fixed $35,000, so the sizing differs. This shows the gap. It does not explain it.

What IBS level should you buy below?

Any level from 0.10 to 0.30 made money with all five exits, on all four markets. 0.25 is a fine line, not a special one. Picture it as a pivot table: buy level down the side, exit across the top, Ret/DD in each cell.

Two heatmap tables of Ret/DD for 10 buy levels from below 0.05 to below 0.50 and 5 exits, on NQ and ES. Every cell is positive. A white box marks below 0.25 with the profit exit: NQ 4.14, ES 6.98. A dashed box marks the best cell: NQ below 0.10 with the profit exit at 10.31, ES below 0.15 with the profit exit at 10.00. — Every cell made money. The best cell sits in a different place on each market.

Every one of the 50 long settings made money before costs, on every market. On ES and the ETFs, all 25 settings from 0.10 to 0.30 still made money after costs.

The rule I judge ranked 5 of 50 on ES and 3 of 50 on QQQ. But it ranked 29 of 50 on NQ and 33 of 50 on SPY. The best single cells were below 0.10 on NQ (Ret/DD 10.31), below 0.15 on ES (10.00) and QQQ (12.62), and below 0.15 with a 3-day hold on SPY (19.35).

I'm not using any of them. Picking the brightest cell after the test is the hindsight trap the probability of backtest overfitting paper by Bailey, Borwein, López de Prado and Zhu warns about. Think of the grid as a landscape. A lone spike rests on one cell, and you cannot tell it apart from luck. A wide plateau, where nearby settings all made money in the same test, does not hang on one lucky cell. IBS is a plateau: every level from 0.10 to 0.30, with every exit, made money before costs on all four markets.

Does a lower IBS close earn more per trade?

The lowest levels beat the highest on every market. But the steps are not even: 0.05 earned less than 0.10 on all four. To measure it fairly, I take away the market's normal drift: what you would earn buying any day and holding just as long. What is left is the edge over drift: how much more the trade made than a random day held just as long.

Heatmap table of the extra gain per trade over a random day by buy level and market. The top rows are brightest: below 0.10 gives NQ 0.134%, ES 0.151%, QQQ 0.344% and SPY 0.206%. Below 0.50 gives NQ 0.048%, ES 0.029%, QQQ 0.130% and SPY 0.047%. Every cell is above zero. — Closes below 0.05 and 0.10 were followed by bigger bounces than closes below 0.40 to 0.50, on every market.

On ES the typical edge was 0.139% a trade at levels 0.05 and 0.10, and 0.030% at 0.40 to 0.50. On NQ, 0.113% against 0.051%. On QQQ, 0.330% against 0.140%.

So the level is a trade-off, not a magic number. Buy at 0.05 or 0.10 and you trade less but earn more each time than at 0.40 to 0.50. Go higher and you trade more for less each time.

Does the profit exit help?

Mostly, yes, against a plain two-day hold. It is not the best exit everywhere.

Dot grid of four rows, one per market, with ten dots for the ten buy levels. A filled dot means the profit exit scored a higher Ret/DD than a plain 2-day hold at that level. NQ 10 of 10, ES 8 of 10, QQQ 6 of 10, SPY 6 of 10. — The quick profit exit beat a 2-day hold at most levels, but not all.

The profit exit beat a 2-day hold at 10 of 10 levels on NQ and 8 of 10 on ES. It also cut time in the market on NQ from 33.0% to 26.6%. On NQ, 591 of 1,126 trades were out after just one day.

But on NQ a plain one-day hold scored a higher typical Ret/DD, and on SPY a three-day hold did. No single exit led on all four markets.

Does IBS beat a random entry?

Yes. This is the part of my 2025 article that holds up.

First, drift. Buy any NQ open and sell the next open, and you earned 0.024% on average. Hold two days, 0.046%. All 50 NQ long settings beat that. So did 49 of 50 on ES, and 50 of 50 on QQQ and SPY.

Second, the coin flip. I bought on random days, as many times as IBS did, and sold 1 or 2 days later. I ran it 10 times with different random draws for each hold.

Dot plot of single random runs on NQ, ES, QQQ and SPY, each as a share of what IBS below 0.25 made. Random buys on random days, as many as IBS made, sold after 1 or 2 days, 10 runs each. On NQ the best random run made $245,955, 80% of the $307,587 IBS made. On SPY 1 of 20 runs made more than IBS: $68,536 against $61,946. — The random runs' average fell short on every market. Only 1 SPY run in 20 beat IBS.

On NQ, the random one-day runs made $81,388 on average. IBS made $307,587, 3.5 spreads above them (a spread is how much the random runs swung from one draw to the next). Against the two-day runs it sat 2.7 spreads above. ES: 3.9 and 2.5. QQQ: 6.1 and 3.5.

SPY is the weak spot. IBS sat only 1.5 spreads above the two-day random runs, and 1 of the 20 SPY runs beat it outright. On NQ, ES and QQQ, none of the 20 runs did. The best NQ random run made 80% of what IBS made.

Does IBS work on the short side?

Not well. Selling a strong close is the mirror of the long rule, and it mostly failed.

Grouped bar chart of how many of 50 settings made money before costs. Long, buying a weak close: 50 of 50 on NQ, ES, QQQ and SPY. Short, selling a strong close: 24 on NQ, 33 on ES, 29 on QQQ and 10 on SPY. — Buying weak closes worked everywhere. Selling strong closes mostly did not.

The short side made money in 24 of 50 settings on NQ, 33 on ES, 29 on QQQ and just 10 on SPY. Its typical Ret/DD ran from -0.49 to 0.25, against 4.36 to 7.84 for the long side. The mirror of the main rule lost $11,497 on SPY over 1,939 trades. These four markets rose over their years. A strong close was not much of a sell signal.

The honest limits

Where my 2025 research was right: buy a weak close, get out fast. IBS below 0.25 made $307,587 on NQ and $219,988 on ES over 26 years. Every long setting made money before costs, and the rule beat random entries on every market.

Where it was wrong: the 2024 headline, and "all market regimes". No plain NQ long setting went 14 for 14 in 2024, and 2024 was a rising year, not a down one. The rule lost in 2022, in 2000 and 2001, and over the whole first half of NQ's years.

What this test does not cover:

NQ has no figure after costs. ES after costs made $187,218.
No stop loss. The largest single loss on one NQ contract was -$19,900.
Futures prices are back-adjusted (old prices shifted so the contract rolls join up). Point changes are exact. Percent figures on futures use the adjusted prices.
The four 2025 filters were not retested. I gave no settings for them in 2025.
2026 is a part year, to 2026-08-20.
A backtest is not a live edge. It is the best evidence about the past.

What this means for you

Look at the long side first. Over these test windows, on four rising index markets, the long settings scored a typical Ret/DD of 4.36 to 7.84. The short mirrors scored -0.49 to 0.25.
Pick your level for how often you want to trade. Any level from 0.10 to 0.30 made money on all four markets. The lowest levels earn more per trade than the highest. Higher levels trade more.
Don't judge it on one year. 14 trades could not tell an edge from a streak. The full NQ record, 1,126 trades, finished 3.5 and 2.7 spreads above random entries.
Size for the bad years, not for 2024. The NQ rule had 7 losing years in 26, the worst -$50,460 in 2022, and a worst drop of $74,350.
Check it before you trade it. Put the trade lists of a few nearby levels into AlgoChef and confirm they behave alike before money goes on any one of them.

Reference tables

IBS below 0.25 with the profit exit, full windows, no costs

	NQ futures	ES futures	QQQ	SPY
Trades	1,126	1,130	1,226	1,452
Win rate	64.4%	66.4%	62.2%	59.8%
Net profit	$307,587	$219,988	$117,002	$61,946
Worst drop (max drawdown)	$74,350	$31,512	$10,172	$9,845
Ret/DD	4.14	6.98	11.50	6.29
Rank among 50 long settings	29	5	3	33

Levels 0.10 to 0.30, all five exits (25 settings per market)

Market	Made money	After costs	Typical Ret/DD	Lowest	Highest
NQ	25 of 25	no costed view	4.14	1.43	10.31
ES	25 of 25	25 of 25	5.69	2.82	10.00
QQQ	25 of 25	25 of 25	5.91	3.05	12.62
SPY	25 of 25	25 of 25	8.87	4.99	19.35

The two halves, IBS below 0.25 with the profit exit

Market	First half	Second half
NQ	2000 to 2013: -$11,393, Ret/DD -0.30	2013 to 2026: $318,980, Ret/DD 4.29
ES	2000 to 2013: $22,700, Ret/DD 0.91	2013 to 2026: $197,288, Ret/DD 6.26
QQQ	1999 to 2012: $69,455, Ret/DD 6.83	2012 to 2026: $47,547, Ret/DD 6.82
SPY	1993 to 2009: $33,350, Ret/DD 3.39	2009 to 2026: $28,596, Ret/DD 3.34

Internal Bar Strength is one of the simplest edges I have tested that still holds up. Buy a weak close. Get out fast. Just don't believe a single year of it.

Want the next study when it is finished, including the ones that overturn something I believed? That is what the newsletter is for: StatOasis.com/Overfit

Methodology and honesty footer: 400 backtests on NQ and ES futures (2000 to 2026), QQQ (1999 to 2026) and SPY (1993 to 2026) daily bars, plus the long grid on each half of every window. Execution rule v2 (signal at the close, fill at the next open, every exit at an open). One trade at a time per setting, futures one contract, QQQ and SPY $35,000 in whole shares, no compounding. Main figures have no costs. ES, QQQ and SPY are also shown after costs; NQ has no costed view. Buy-and-hold and seeded, frequency-matched random controls computed on the same bars. Every figure here comes from the study's own results tables. This is research, not investment advice.

Methodology

Data source: Daily bars from TradingView exports read through engine.market_universe.load_tv: E-mini Nasdaq 100 (NQ) and E-mini S&P 500 (ES) continuous front-month futures, back-adjusted by the additive method so point changes are exact, and QQQ and SPY price-only series. IBS from engine.oscillators.ibs, an admitted engine indicator.
Date range: NQ and ES 2000-01-03 to 2026-08-20 (26.63 years); QQQ 1999-03-10 to 2026-08-20 (27.45 years); SPY 1993-02-01 to 2026-08-20 (33.55 years). Halves split each window at its middle bar.
Entry / exit rules: Long when IBS at the close is below one of ten levels (0.05 to 0.50); buy the next open. Five exits: the profit exit (sell the next open if the entry bar closed above the entry price, otherwise the open two bars after entry) and plain time exits of 1, 2, 3 and 5 bars, all filled at the open. Short mirror: IBS above 1 minus the level. Reference: IBS below 0.25 with the profit exit. No stop, no target.
Sizing: Futures one contract (BigPointValue NQ 20, ES 50); QQQ and SPY $35,000 in whole shares; no compounding. Frictionless headline figures. ES, QQQ and SPY rows also carry a costed net profit under the engine's atr-handbook-2026-09-24 schedule. NQ has no costed view: its engine row has no tick size.
Overlap mode: Flat-only within each setting: one position at a time; a setup while a position is open is ignored, including on the bar the exit order is placed (Execution rule v2).
Look-ahead: IBS uses only the signal bar's own high, low and close, read at that close; the trade fills at the next open. The profit test is read at the entry bar's close and acts at the next open.
Minimum sample: 50 trades for the honesty baselines (the engine default). Rankings count a setting as reliable at 100 trades or more; every long setting cleared it.
Buy-and-hold benchmark: Each market held over the same window, computed by tools/controls_report.py. NQ one contract: $430,325.00 net, CAGR 10.20%. ES one contract: $268,312.50 net, CAGR 8.45%. QQQ on $35,000: $451,306.62 net, CAGR 10.06%. SPY on $35,000: $572,030.47 net, CAGR 8.88%.
Random control: Seeded random long entries, exits at the open under Execution rule v2, 10 seeds from base seed 20260803. tools/controls_report.py matches the median setting (NQ 952 entries, ES 940, QQQ 1012, SPY 1218, 2-bar hold). The article's comparison uses controls matched to the reference's own trade count with 1-bar and 2-bar holds, computed in the facts sheet with the same engine function.

Parameter scopeParameters swept

The study searched the parameter space and reports the spread, not one tuned setting.

400 backtests on the full windows (10 levels x 5 exits x 2 sides x 4 markets), 200 of them on NQ and ES, plus the long grid on each half of every window. The reference (IBS below 0.25, profit exit) was fixed before the run from my 2025 rules. The stable region (levels 0.10 to 0.30, every exit) was read after the run; the halves, drift and random controls are the checks. The single best cells are named and not used.

Run to v1 of the StatOasis research standard - the rules every study here has to meet before it is published. The version is the study's own: a standard that gained a rule later never reaches back and claims this one met it.

Historical backtest results are not a guarantee of future returns. This content is for educational purposes only and is not investment advice. Hypothetical performance disclosure (CFTC Rule 4.41).

Frequently asked questions

What does IBS mean in trading?

IBS stands for Internal Bar Strength. It is (close minus low) divided by (high minus low). It reads 0 when the day closes at its low and 1 when it closes at its high. A low reading means the day closed weak inside its own range. That is the setup a mean-reversion trader buys, betting on a bounce.

What IBS level is a buy signal?

There is no magic line. On NQ, ES, QQQ and SPY, every level from 0.05 to 0.50 made money long: 200 of 200 settings before costs. The lowest levels earned more per trade than the highest. On ES the typical extra gain over a random day was 0.139% a trade at 0.05 and 0.10, against 0.030% at 0.40 to 0.50. IBS below 0.25 ranked 5 of 50 settings on ES and 29 of 50 on NQ by Ret/DD (profit divided by the worst drop).

Does the IBS strategy work on ES and NQ futures?

From 2000 to 2026, yes. IBS below 0.25, out at the next open if in profit, otherwise one day later: 1,126 trades on NQ, $307,587, worst drop $74,350, Ret/DD 4.14. On ES: 1,130 trades, $219,988, worst drop $31,512, Ret/DD 6.98, or $187,218 after costs. Both beat random entries. But on NQ the same rule lost $11,393 from 2000 to 2013 and made its money after that.

How long are IBS trades held?

With the exit tested here, one or two days. The trade buys the next open after a weak close. If it is in profit at the close of that day, it sells the next open. If not, it sells the open one day later. On NQ, 591 of 1,126 trades were out after one day. The average hold was 1.48 bars, and the rule was in a trade 24.7% of the time.

Is a 100% win-rate year evidence of an edge?

Not on its own. Across every market, long setting and calendar year with 10 or more trades, 5,519 setting-years, not one went without a losing trade. The best was 94.4%. At the NQ rule's 64.4% win rate, treating trades as independent, 14 wins in a row has a 0.21% chance in any one run of 14 trades, yet streaks that long did happen inside 26 years across the 50 NQ settings, with losing years around them. This rule's own longest on NQ was 11.

Does IBS work in a down market?

Not on NQ in 2022, or over 2000 to 2002. The rule lost $50,460 in 2022, while one NQ contract held all year lost $111,280. It netted -$22,803 over 2000 to 2002. It lost less than holding, but it lost. And 2024, the year in my old headline, was a rising year: one NQ contract held all year made $63,445.

Read the Strategies, Backtested hub

Back to all research

#147

Mean Reversion: 40 Ways to Buy the Dip, and the 16 That Held Up

Nov 5, 2026 · 7 min read

Forty buy-the-dip entries tested on 33.5 years of SPY, 17,385 trades, every one measured against what entering on any day and holding as long paid. 35 of 37 rankable entries beat it. Then the window was cut into three periods: 16 held their edge, 11 decayed to under half, and only 4 of the top 10 held up.

#155

Donchian Channel: What It Is, How to Trade It, and 900 Backtests on 15 Markets

Oct 29, 2026 · 9 min read

A Donchian channel is the highest high and the lowest low of the last N bars, with a middle line halfway between. I traded it both ways, as a breakout and as a fade, in 900 variants on 15 markets from 2004 to 2026. The breakout made money on the side each market drifted (283 of 300 variants) and rarely against it (43 of 300). Set beside random entries with the same exit, 111 of 900 variants beat the random 95th percentile per bar held.

Overfit - the newsletter

Skip the hype. Trust the data.

One practical, evidence-driven takeaway per issue - strategy testing, portfolio construction, market structure, trading psychology, tactical asset allocation.

Delivered +2 times a month - when the work is ready, not on a calendar.I'll never sell your address.Unsubscribe in one click, any time.