Time Filter vs ADX Filter: The 12:00 to 16:00 UTC Window Beat No Filter in 7 of 12 Runs, ADX Above 25 in 8 of 12, and 0 of 24 Gaps Were Bigger Than Noise

When an indicator gives too many bad signals, there are two popular fixes. One says trade only at the right time of day. The other says trade only when the market is trending, and measures that with ADX. Both are sold as the thing that separates good signals from bad.

I ran both on the same entries and the same bars, next to a third common filter, the 200 EMA. The question is simple: after the filter, is the average trade better than before it?

What exactly is being compared?

• Time filter: take a signal only if it appears between 12:00 and 16:00 UTC, the London and New York overlap. In New York that is 8:00 a.m. to noon in summer.
• ADX filter: take a signal only if ADX 14 is above 25 on the signal bar. This is the classic trend strength reading from my ADX 25 trend filter test.
• EMA 200 filter: take buys only above the 200 EMA and sells only below it.
• No filter: take every signal, all day.

The entries are the three from the session time filter test: EMA 9/21 cross, RSI 14 returning from 30 or 70, and a Stochastic 14, 3, 3 cross from below 20 or above 80. Three entries on four charts gives 12 runs per filter.

How was it tested?

• Data: EURUSD, June 3 to August 26, 2026, 16,926 M5 bars and 5,642 M15 bars; gold, June 16 to August 26, 13,736 M5 bars and 4,577 M15 bars.
• Exit: stop 1.5 ATR, target 3 ATR, 48 bars at most, one trade at a time.
• Cost: 1 pip per EURUSD trade and $0.30 an ounce on gold, from my spread and swap measurement.
• Noise check: for every gap between a filtered and an unfiltered result I resampled the trades 2,000 times. A gap counts as real only if the middle 90% of the resampled gaps does not include zero.

Which filter improved the average trade?

Per trade after cost: pips on EURUSD, dollars an ounce on gold. “Too few” means fewer than 20 trades were left after the filter.

Seven of twelve and eight of twelve sound like a mild win for the first two filters. They are not. A coin gives six. And when I asked whether each gap was larger than the luck of which trades happened to fall inside the filter, the answer was no 24 times out of 24.

Take the best-looking cell: the EMA cross on gold M5 inside the overlap, +$3.09 a trade against +$1.20 unfiltered. The gap is $1.89. Resampled, it ran from -$0.92 to +$4.75, so against the unfiltered run it is inside noise. The unfiltered run contains the filtered trades, which shrinks the gap. Against only the trades the window removed, the gap is $2.98 and just outside noise (+$0.15 to +$5.82). It is one of the two gaps in this article that cleared noise that way. The other is the Stochastic on gold M5 with ADX above 25: +$0.14 against -$1.73 when ADX was 25 or lower.

The one gap that did clear noise went the wrong way. The EMA 200 filter cut the gold M5 EMA cross from +$1.20 to -$0.73 a trade. I saw the same weakness on full systems in trading system vs its own entry signal, where the 200 EMA filter made 8 of 12 runs worse.

Time filter vs ADX filter head to head

Six runs each, and no gap outside noise. The split is not random-looking in one respect: ADX won all three runs on gold M15 and the time filter won two of three on the other charts. With 24 to 51 trades inside the window on gold M15, I would not read much into it.

The two filters also remove very different amounts. The time window kept 14% to 32% of signals, most often about 18%, because four hours is a sixth of the day. ADX above 25 kept 13% to 23% of EMA crosses but 74% to 87% of RSI signals. That is by design: an RSI at 30 or 70 comes after a strong push, which is exactly when ADX is high. So on a reversal entry the ADX filter removes almost nothing.

Do signals inside the window beat signals outside it?

This is the cleanest way to ask the time question. Same entry, same chart, split only by the clock.

• Inside 12:00 to 16:00 was better in 7 of 12 runs.
• One gap cleared noise: the EMA cross on gold M5, +$3.09 inside against +$0.10 outside.
• On EURUSD M5 the three entries made -1.72, -0.14 and -0.82 pips inside the window, and -1.62, -0.63 and -0.89 outside. The most active four hours of the day changed almost nothing.

The overlap has the widest range of the day on EURUSD, 2.5 to 3 pips per M5 bar against about 1.5 in the Asian session. A wider range makes the spread a smaller part of the stop, as little as 22% against as much as 49%. But the stop and target in this test scale with ATR, so a bigger range also means bigger losses when the entry is wrong. The two effects roughly cancel.

Does using both filters together help?

Six runs had at least 30 trades left with both the time window and ADX above 25. Two of the six made money: the Stochastic on gold M5 at +$3.46 over 79 trades, and the RSI on EURUSD M5 at +0.21 pips over 56 trades. Four lost. Stacking filters mostly leaves a sample too small to judge, the same thing that happened with stacked indicators in the combo vs single indicator test.

What does a time filter actually buy?

Fewer trades. That is the one effect that showed up in every run.

The filtered version lost 809 pips less. It did that with a slightly worse average trade. If you stop a losing rule from trading for 20 hours a day, it loses for 4 hours instead of 24. That looks like an improvement on an account statement, and it is the reason time filters get good reviews. It is not evidence that the hours were well chosen. Any four-hour window would have done about the same, and the grid in the trading hours filter guide for MT4 shows it.

Which filter should you use?

• Neither as a fix. No filter here turned a losing entry into a winning one by more than chance.
• A time window is a sensible way to trade less. If you can only watch the screen four hours a day, the overlap costs the least in spread terms. Just do not expect a better trade from it.
• ADX removes most crossover signals and very few RSI signals. It improved neither by more than noise: the EMA cross was better with it in 2 runs of 4, the RSI entry in 3 of 4.
• Changing the timeframe moved one entry more than any filter did. The EURUSD EMA cross went from -1.66 pips a trade on M5 to +0.28 on M15, a gap outside noise. It was the only one of six: the Stochastic lost 0.51 pips more on M15, and nothing on gold changed by more than noise. Compare ADX with two other range filters in Choppiness Index vs ADX. There ADX was ahead in 2 runs of 10, here in 8 of 12, on different entries. Neither count is outside what a coin gives.

Frequently asked questions

Is a time filter better than an ADX filter?

Not measurably. Head to head they won 6 runs each, and no gap was bigger than sampling noise. Against no filter the time window was ahead in 7 of 12 runs and ADX in 8 of 12.

Does trading only the London and New York overlap improve results?

Signals inside 12:00 to 16:00 UTC did better than signals outside in 7 of 12 runs, and only one of those gaps was bigger than noise. On EURUSD M5 the average trade was almost the same inside and outside.

Why does a time filter look good in an account history?

Because it removes about 80% of trades. A losing rule loses less in total when it trades less. On the EURUSD M5 EMA cross the loss fell from 1,005 to 196 pips while the loss per trade went from 1.66 to 1.72.

Does the ADX filter work with RSI signals?

It removes very little: ADX was above 25 on 74% to 87% of RSI 30/70 signals. What it kept did better in 3 runs of 4, never by more than noise.

Should I combine a time filter and an ADX filter?

Only 6 of 12 runs kept 30 trades with both filters on, and 2 of those 6 made money. The sample gets too small to trust.