ICT Killzone Strategy: Does the New York Time Filter Help?
A common claim in ICT-style trading is that the New York killzone is where the best sweep and fair value gap setups appear. The idea is that restricting entries to a short morning window filters out noise and lifts the odds. This article tests that claim on one specific rule set: a liquidity sweep, then a fair value gap, inside 07:00 to 10:00 New York time.
The headline finding is that the time filter did not produce a profitable edge on the three markets tested. NQ, XAU/USD and EUR/USD all ended with a profit factor at or below 1 after costs, and NQ and EUR/USD both lost money at 1% risk per trade. The 2023 results look strong, but that year has only a handful of trades and does not carry the weight of the later years.
The rules we tested
Between 07:00 and 10:00 New York time, wait for a liquidity sweep, then wait for a fair value gap. Enter on the FVG. Stop above the sweep. Target liquidity, at least 2R.
Results
| Market | Trades | Win rate | Profit factor | Net | Max DD | Period |
|---|---|---|---|---|---|---|
| NQ 15m | 157 | 28.0% | 0.98 | -4.6% | -19.0% | 3y |
| XAU/USD 15m | 132 | 30.3% | 1.00 | -2.0% | -20.9% | 3y |
| EUR/USD 15m | 121 | 28.9% | 0.98 | -3.3% | -16.6% | 3y |
1% risk per trade, spread and slippage included, entries at the next candle open, stop checked before target. Periods: NQ 2023-11-09 to 2026-11-13 · XAU/USD 2023-11-09 to 2026-11-13 · EUR/USD 2023-11-09 to 2026-11-13.
Equity (% at 1% risk per trade) on XAU/USD 15m, 132 trades.
What was tested and why
The rule set is narrow on purpose. Between 07:00 and 10:00 New York time, the engine waits for a liquidity sweep. A sweep here means price takes an untaken high or low, which can be a swing point, an equal high or low, the previous day's extreme, or a session extreme. After the sweep, the engine waits for a fair value gap, and then places a limit order at the edge of that gap.
The stop sits beyond the sweep wick. The target is the nearest liquidity at least 2R away. Both long and short trades are included. Each step must occur within 20 bars of the one before it. Entries are at the next candle open, costs for spread and slippage are included, and risk is 1% per trade. The test ran on 15-minute charts from 2023-11-09 to 2026-11-13, a period the data describes as three years. The point of the test is not to find the best version of the idea. It is to see what the time window alone does to a setup that many traders describe in similar terms.
Results by market
NQ produced 157 trades with a 28% win rate, a profit factor of 0.98 and an average result of -0.02R. At 1% risk per trade the net return was -4.6%, with a maximum drawdown of 19%. The trade split was 70 longs and 87 shorts.
XAU/USD produced 132 trades with a 30.3% win rate and a profit factor of 1. The average result was 0R and the net return at 1% risk was -2%, with a maximum drawdown of 20.9%. The split was 67 longs and 65 shorts. EUR/USD produced 121 trades with a 28.9% win rate, a profit factor of 0.98 and an average result of -0.01R. The net return at 1% risk was -3.3%, with a maximum drawdown of 16.6%. The split was 66 longs and 55 shorts.
On all three markets the trade count is above 30, so the samples are not tiny. Even so, a profit factor of 0.98 or 1 after costs means the setup did not clearly make money. The XAU/USD figure of exactly 1 is break-even at best, and the net return is still negative.
What stands out
The year-by-year numbers show a clear pattern. On NQ, 2023 has 6 trades and a total of 16.4R, which is the source of most of the strength in the three-year figure. The 2024, 2025 and 2026 years each have around 50 trades and are negative in total R, at -5.1R, -7.9R and -6R. The early result rests on a very small number of trades and should not be read as evidence of an edge.
XAU/USD is the most even of the three. Its 2025 and 2026 years are positive, at 1R and 7.3R, after a weak 2024 at -8.4R. That is the only market in the set where the later years look better than the earlier ones, though the margin is modest and the total still nets to a flat result. EUR/USD is close to flat in every year, with 2024 at 1R and 2026 at 2R offset by losses elsewhere. On direction, the split between longs and shorts is fairly balanced across all three markets. The data does not break out the long and short results in R, so it is not possible to say from this test which side carried the results.
Why it might behave this way
One explanation is that the sweep and FVG steps already capture most of what the model is looking for, and the time window adds little. If a sweep is more often a random stop run than a reversal, a fixed morning window will not change that. The window narrows the number of trades but does not change the quality of each one.
A second explanation is the structure of the rules. Each step must occur within 20 bars, and the stop sits beyond the wick. With a 2R minimum target, the trade needs a meaningful move to pay. A 28% win rate at 2R breakeven is around 33%, so the setup needs to be right more often than it was here to break even before costs. The data does not show how many trades reached the 2R target versus stopped out, so this is a reasoning point rather than a measured result. A third point is that a time window is one filter among many. The test does not show what happens without the window, so it cannot say whether the killzone helped relative to a version that trades all day. That comparison would be the cleanest next step.
What to try next
The most useful follow-up is a control test. Run the same sweep, FVG and 2R rules across the full trading day with no time filter, then compare the results with the New York window. If the window does not improve profit factor or drawdown, the time filter is not the source of any edge. Other tests could vary one rule at a time. Changing the 20-bar limit between steps, or limiting the sweep types to swings only, would show how sensitive the result is to those choices. Looking at the 2023 NQ trades in more detail would also help, since six trades is too few to draw conclusions from. Readers can run the same test themselves on WickCheck, using the same rules, markets and period, and adjust any setting to see how the numbers move.
Takeaway. Restricting a sweep and FVG model to the New York 07:00 to 10:00 window did not produce a profitable result on NQ, XAU/USD or EUR/USD in this test. Profit factors sat at 0.98 to 1 after costs, and two of the three markets lost money at 1% risk. The early 2023 NQ result rests on six trades and does not outweigh the losing years that follow. The time filter needs a control test against the full trading day before anyone can say whether it adds value.
Same rules on XAU/USD. Change any word, a market or the dates and see what moves.
Terms used here
Questions
- Does the ICT killzone strategy work on NQ?
- In this test, NQ produced 157 trades with a profit factor of 0.98 and a net return of -4.6% at 1% risk per trade. That is a losing result after costs. The 2023 year was strong, but it had only 6 trades, and the 2024 to 2026 years were negative in total R.
- Is the New York killzone better than trading all day?
- This test does not answer that. It only measured the 07:00 to 10:00 New York window. A control run with the same sweep, FVG and 2R rules across the full day would be needed to compare the two.
- Can I test the New York killzone myself?
- Yes. WickCheck lets you run the same rules on the same markets and period, and change settings such as the time window or the bar limit between steps, to see how the results change.
More tests
We tested the 9/21 EMA crossover with a 200 EMA filter on gold, EUR/USD and the Nasdaq, with costs included. Gold lost money after costs over 10 years.
Previous Day High Low Strategy: A Backtest Across 4 MarketsWe backtested a previous day high low strategy with an FVG entry on ES, XAU/USD, EUR/USD and NQ. Two markets lost money after costs. See the numbers.
Asian Range Strategy: Does the London Sweep Hold Up?We backtested an Asian range sweep and fair value gap setup on EUR/USD, USD/JPY, GBP/USD and XAU/USD. Results were mixed, with one market profitable.
Past results, real or simulated, do not guarantee future results. Not investment advice.