Google AI's Fibonacci Scalping Strategy, Backtested: ES Futures, 1- and 5-Minute Bars, 2020 to 2026

00The verdict in three lines

• Google's AI answer on Fibonacci scalping gives a full “/ES Golden Zone” method: wait for the break of the 09:30 opening range, draw the Fibonacci grid on the push, buy (or sell) the pullback into 50% to 61.8%, stop beyond 78.6% or the swing, target the swing high or the 127.2% extension, best of all with the VWAP nearby. We tested its 1-minute and 5-minute versions, with and without the VWAP, on ES from January 2020 to September 2026 at $30.76 a round trip.
• The 1-minute version lost even before costs. A limit at 61.8% won 19.3% of 1,871 trades and ended at -$107,020, worse than the same trades placed at random on other days; the VWAP filter made it worse.
• The 5-minute version won more often than it lost and still lost money (54.4% winners, -$33,099). With the VWAP it ended at +$2,444 over 251 trades, too few and too fragile to count. All four Buried.

01Where the strategy comes from

The rules are an answer of Google's AI search to a question about Fibonacci retracement scalping on 1- and 5-minute charts, received on 14 October 2026. It has no author and no date, and the AI may answer differently tomorrow; we summarise it here and do not reproduce it. It has three parts:

• A general method (impulse, Golden Zone, a confirmation candle): too vague to test without inventing most of the rules.
• Order-flow advice (the order book, absorption, volume delta, Volume Profile): it needs data we do not have, so it is not tested.
• “The /ES Golden Zone Strategy”, with an anchor, levels, entries, stops and targets for ES: this is what we tested. The answer says its levels work because institutional algorithms use them, and calls the setups where 61.8% lines up with the VWAP “the absolute highest-probability setups”.

02What we tested

• The anchor: the opening range is the first 5 minutes after 09:30 New York. A long setup starts when a bar closes above its high; a short setup when a bar closes below its low. One setup per direction per day.
• The grid: from the session low up to the high of the push after the breakout (longs; shorts the mirror). The push high can keep rising until the entry, and the levels with it: 50%, 61.8%, 78.6% and the 127.2% extension.
• FO-1M-LIM (the “aggressive” version): a buy limit at 61.8% on 1-minute bars, filled only when the price trades at least one tick through it; stop one tick under 78.6%; target the push high (“the previous swing high”).
• FO-5M-CONF (the “conservative” version): on 5-minute bars, a pin bar or an engulfing candle that dips into the 50% to 61.8% zone and closes above 61.8%; entry at the next minute; stop one tick under the swing low (the 100% level); target the 127.2% extension.
• FO-1M-VWAP and FO-5M-VWAP: the same, only when the 61.8% level is within 2 points of the session VWAP.
• For all: a pullback beyond 78.6% cancels the setup (“completely broken”, in the answer's words); entries until 11:30 New York, our reading of “trade the high-volume open”, as the answer gives no time; one position at a time, one trade per direction per day; a position still open at the session close is closed there. Six variants of Fibonacci retracement entries on the site (two earlier ones on EUR/USD), so the development t-statistic bar is 2.5.

03Data and costs

ES (E-mini S&P 500) futures, NinjaTrader's back-adjusted 1-minute history with real contract volume, 1,737 New York sessions from 2 January 2020 to 25 September 2026. Costs: $30.76 per round trip (commission and one tick of slippage on each fill), as in every ES test on the site; we keep them for the limit entries too, whose fills already require a trade-through. Periods: development 2020 to 2023, validation January 2024 to March 2025, holdout April 2025 to September 2026, already used by earlier ES tests. Engine check: a self-test with hand-worked sessions and an independent re-implementation that matched every trade of all four variants (fo_selftest_result.txt).

04Results

Dollars per contract. Gross is before costs. The win rate is the share of trades that ended positive after costs, with its 95% range.

Cumulative net dollars by month. * Holdout already used before this test.

By period

By year (net, trades)

The 5-minute confirmation was positive in 2021, 2022 and 2023 and negative in 2020 and in every year from 2024 on.

05What the answer promised, and what the levels gave

The answer quotes stops and targets in points and risk-to-reward ratios. Its levels give different ones. Medians of our trades, in ES points:

On 1-minute bars the stop came out as the answer says, about 3 points, and the target at the push high was larger than its 4 to 6 points. On 5-minute bars the stop under the swing low was more than twice the answer's 5 to 6 points and larger than the target, so the trade risked more than it aimed to win. We followed the levels the answer names, as fixed before the test, not its ratios.

06The falling knife: why the 1-minute limit lost before costs

The answer warns that a limit at 61.8% “carries the risk of catching a falling knife”. On ES it was the rule rather than the risk: 55% of the 1-minute trades were stopped within five minutes of the fill and 15% in the minute they were filled (with the VWAP filter 63% and 21%). A limit order is filled exactly when the price trades through it, and on 1-minute bars a price that reaches 61.8% often keeps going through 78.6%.

07Long, short, time of day and the size of the push

By entry time (New York, net and trades)

By the size of the push from the swing to its high (net, trades, share won)

These splits are descriptions, not new rules: none of them was fixed before the test, and we do not trade them.

08Is it the signal or chance?

The moved-entry control: every real entry moved by one random offset of whole sessions, at least 5% of them, to the same minute after the open in another session, same direction, entered at that minute's open with the same stop and target distances, the same exits and costs; 500 runs with fixed seeds. The bar is the 95th percentile of the average net result per trade.

The 1-minute limits did worse than every run of the control: the same stops and targets, entered at the same time on other days, lost less. The 5-minute version with the VWAP sat just above the bar; it fails four other gates.

09The gates

10Verdict by variant

No proven edge (both 1-minute versions and the 5-minute confirmation): negative before costs. Fragile (5-minute with the VWAP): positive and just above the control, but only 156 / 61 / 34 trades per period, a development t-statistic of 1.05 (2.5 needed), negative without its 10 best trades and negative in the holdout. Not tested: the order-flow parts of the answer, its EMA and RSI filters, other markets and other session times.

TWhat this shows, and what to take into your own tests

What the results show

• Taken exactly as Google's AI describes it, the 1-minute version lost before costs (-$26.44 a trade) and -$107,020 after, and its entries did worse than every run of the control.
• The VWAP filter the answer calls the best setup made the 1-minute version worse: 17.9% winners instead of 19.3%.
• The 5-minute version won 54.4% of its trades and still lost $33,099: its stop under the swing low was larger than its target.

What remains unknown

• The order-flow parts of the answer (the order book, absorption, volume delta, Volume Profile), which need data we do not have, and its EMA and RSI filters.
• Other readings of a vague rule: another end of the entry window, a minimum size of the push, or an exit at the 4 to 6 points the answer quotes instead of the levels it names.

Why it failed

A limit order at 61.8% is filled exactly when the price trades through it, and on 1-minute ES bars the price often kept going: 55% of the trades were stopped within five minutes, 15% in the minute they were filled. With a stop of about 3 points, the $30.76 round trip is 20% of the risk on every trade. On 5-minute bars the confirmation candle cut the falling-knife problem, but a stop under the swing low is wider than a target at the 127.2% extension, so a win rate above 50% was not enough.

What you can apply to your own tests

• Treat an AI answer like any other trading tip: it states rules with confidence, but it does not test them.
• Check that the numbers of a strategy fit together: here the stated risk-to-reward ratios did not match the levels the same answer named.
• With limit orders, test fills honestly: a fill on a touch, without the price trading through, makes a backtest look better than trading.

A note on the numbers: drawdown figures are measured on closed trades unless a table says it includes open losses. Open losses inside a trade are not counted in a closed-trade drawdown, so real account risk is higher, especially for systems with no stop. Backtests are not live trading.

FAQCommon questions

Does Fibonacci scalping work on ES futures?

Not in our backtest from 2020 to 2026. A buy limit at 61.8% on 1-minute bars won 19.3% of its trades and lost $107,020 per contract after costs; the 5-minute version with a confirmation candle won 54.4% and lost $33,099.

Does the VWAP make Fibonacci setups better?

Not here. Keeping only the 1-minute setups where 61.8% was within 2 points of the VWAP, which the AI answer calls the highest-probability setups, lowered the win rate from 19.3% to 17.9%.

How was the Fibonacci scalping strategy tested?

On ES futures 1-minute data from 2020 to 2026, with the rules fixed before the test, $30.76 per round trip, limit fills only when the price traded through the level and 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.

11Check our arithmetic

Code provenance: our own code, written from the rules of the AI answer as fixed in the specification. No third-party source code is used or published.

12Data notes

Holes. No trade of any variant is open across a known hole in the ES file; every verdict stands as computed.

Back-adjustment. The file is back-adjusted, which shifts older prices by a constant at each roll; Fibonacci levels, the VWAP and every result in points move with the prices, so nothing changes.

Fills. One-minute bars do not show the order of prices inside a minute. We assumed the worst where it matters: a limit is filled only on a trade-through of one tick, a stop can be hit in the entry minute while a target cannot, and if a minute touches both, the stop is taken. A trader with tick data could get a slightly different result, mainly in the minute of the fill.

Source. An AI answer has no fixed text: asked again, Google's AI may give other levels or other advice. The rules tested are the ones in our specification, written from the answer received on 14 October 2026.

13Corrections

22 Oct 2026, navigation: added a link to the new guide comparing trading costs across all our tests. No result changed. 22 Oct 2026, navigation: added a link to the new guide comparing trend following and mean reversion across all our tests. No result changed.

+More tests and guides

• Fibonacci retracement strategy backtest
• Vwap strategy backtest
• 9 ema strategy backtest
• Noise area breakout es
• Trend following vs mean reversion: this test next to the others of its style
• 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