Multi-Timeframe Confluence Tested: An Honest Higher-Timeframe Filter Helped 67 of 144 Setups, a Repainting One Showed 114

“Trade in the direction of the higher timeframe” is the most repeated rule in forex education. Check the H1 or H4 trend, then take only the 5-minute entries that agree with it. I have used a trend filter inside many of my tests, from the gold filter on an EMA cross to the EMA 200 filter in the MACD crossover test. This time I tested the rule itself: four entries, four higher timeframes, three trend lengths, three exits — and three ways of reading the higher-timeframe chart, because that turned out to matter more than anything else.

How was the higher-timeframe filter tested?

• Data: 16,926 EURUSD 5-minute bars, June 3 to August 26, 2026. Results split at July 15 to check both halves.
• Entries on M5: EMA 9/21 cross (888 signals), MACD 12/26/9 signal-line cross (1,482), RSI 14 leaving 30 or 70 (1,016), Stochastic 14/3/3 %K/%D cross below 20 or above 80 (1,338). The first two follow the move, the last two fade it.
• Filter: take a buy only when the higher-timeframe close is above its EMA, a sell only when it is below. Higher timeframes M15, M30, H1 and H4, built from the M5 bars. EMA 20, 50 and 100 on that timeframe.
• Exits: after 1, 2 or 4 hours. One trade at a time. 1 pip of cost, the figure from my spread and swap measurement.

That is 4 × 3 × 4 × 3 = 144 filtered setups per reading, each compared with the same entry and exit without a filter.

What are the three ways to read a higher timeframe?

When your 5-minute bar closes at 10:25, the 11:00 H1 bar is still forming. You can do three things with it:

• Closed: ignore it and use the last finished H1 bar, the 10:00 one. Honest, but up to 55 minutes late.
• Live: use the forming bar with the current price as its close. Honest — this is exactly what you see on a live H1 chart.
• Final: use the forming bar with the close it will have at 11:00. Impossible in live trading, but it is what many multi-timeframe indicators draw on history: once the H1 bar closes, they repaint the whole hour of M5 bars with its final value. I described how to catch this in the repainting arrows test.

Did the higher-timeframe filter improve the trades?

EURUSD M5, 1 pip cost per trade. A setup counts only with 30 or more trades; all 432 had at least 30.

Read honestly, the filter helped a little under half the time and hurt a little over half. The median effect was six hundredths of a pip — noise. With the live bar it was nine hundredths the other way. Out of 288 honest filtered setups, 9 made money after costs, and 4 were positive in both halves.

Read with the final bar, the same filter looked excellent: better in 114 of 144 setups, a median gain of 0.67 pips per trade, and 32 profitable systems. None of that is tradeable.

How can 6% of bars make that much difference?

For the H1 EMA 50, the live and final readings pointed the same way on 94.06% of M5 bars. They disagreed on 5.94%. But those are not random bars. They are the hours where the price crossed the H1 EMA — exactly the moments when a 5-minute entry is most likely to fire, because something is moving.

The final reading knows where the hour ends. When a buy fires at 10:25 and the hour closes lower, the repainting filter has already switched to “down trend” and removes the losing buy from history. The best final-bar setup — H1 EMA 20 on the Stochastic cross, 4-hour exit — showed +1.63 pips per trade. The same filter with the live bar made -0.11.

If a multi-timeframe indicator’s history looks much better than its live results, this is the first thing to check.

Which entries did the filter help?

The filter did almost nothing for MACD: 1 improvement in 36. MACD is already a trend-following line, so a second trend opinion mostly agrees with it and removes a random half of the trades.

It helped the Stochastic cross in 30 of 36 setups. That makes sense: a Stochastic cross below 20 in an uptrend is buying a dip, while the same cross in a downtrend is catching a falling knife. The filter separates the two. RSI leaving 30/70 is also a fade, but it improved only 16 of 36 — RSI reaches its extremes less often and mostly during strong moves, as I found in the Williams %R, Stochastic and RSI comparison.

Which higher timeframe worked best?

The usual advice — “check H4 before trading M5” — came last. H4 flipped trend only 32 times in twelve weeks; a two-hour trade barely notices a trend that slow, and the closed H4 bar can be almost four hours old. M15 was closest to the trade’s own horizon and helped most often, though even there only 3 of 36 setups made money after costs.

What was the one combination that held up?

Buying Stochastic dips only while the M15 close is above its EMA 50 (and selling rallies only below it) turned -1.40 pips per trade into +0.48, and both halves stayed positive with either the closed or the live bar. It kept 28% of the signals.

I would not trade it on this evidence alone. It is the best of 288 honest setups, +0.48 pips is less than half the cost of a trade, and picking the winner from a big grid is how optimized settings fall apart out of sample. What it does show is where a trend filter belongs: on entries that fade short moves, one step above the chart you trade.

So should you use multi-timeframe confluence?

• Do not expect it to rescue an entry. Honest filters improved fewer than half the setups, and 4 of 144 became profitable.
• Use it on dip-buying entries, not on trend entries. It helped the Stochastic fade 30 times in 36 and MACD once.
• Stay close to your timeframe. M15 over M5 beat H4 over M5 in this test.
• Never judge an MTF indicator by its history. A 6% repaint turned 4 profitable setups into 32.

The comparison with a plain slow moving average is in higher-timeframe filter vs moving average, and the setup steps for MetaTrader, including gold, are in multi-timeframe analysis on MT4.

Frequently asked questions

Does multi-timeframe analysis improve win rate?

Only for some entries. On EURUSD M5, an honest higher-timeframe filter improved results in 67 of 144 setups. The Stochastic cross gained the most: its win rate went from 42.7% to 48.3% with an M15 EMA 50 filter.

What is the best higher timeframe for a 5-minute chart?

In this test, M15. It improved 22 of 36 setups against 16 for H1 and 9 for H4. The closer timeframe reacts in time for trades that last one to four hours.

Do multi-timeframe indicators repaint?

Many do. If an indicator uses the current, unfinished higher-timeframe bar and redraws history when that bar closes, its past signals look better than the live ones. Here that effect raised profitable setups from 4 to 32.

Is H4 trend confirmation worth it for scalping?

Not on this data. An H4 EMA 50 filter improved 9 of 36 EURUSD M5 setups and made none profitable. It changed direction only 32 times in twelve weeks.