Candlestick Patterns vs MACD: Confirmation Agreed 46% and Contradicted 46%

My candlestick pattern test found the shapes losing money on their own, which surprises nobody who trades them. The standard defence is that patterns were never meant to be traded naked — you confirm them with an indicator, and the combination is what works. That is a testable claim, and it is a more interesting one than the first, because it is about whether two things you already own tell you anything different.

So I ran engulfing bars, pin bars and stars against MACD 12/26/9, an EMA 12/26 crossover and RSI 14 over the same 16,926 EURUSD 5-minute bars, June 3 to August 26, 2026. Not just the returns — the overlap. When does the indicator actually say the same thing as the candle?

How often do a candle and an indicator agree?

Far less often than the advice implies, and the number depends entirely on how generous you are with the timing.

Read the bottom row twice. Give MACD an hour on either side of an engulfing bar and it will have fired 92.1% of the time — and the direction it fired in is a coin flip. Confirmation that is available in both directions on demand is not confirmation. It is a search, and the trader doing the searching already knows which answer they want.

The same-bar row is the honest one. Genuine simultaneous agreement between an engulfing bar and MACD happened 110 times in 1,644 patterns. With an EMA 12/26 crossover it happened 35 times — 2.1%. These are not two views of one event. They are two mostly unrelated events that occasionally coincide, exactly as my MACD against moving average crossover comparison found for two indicators that are the same event.

Do they perform differently on their own?

Four of the five lost before costs; all five lost after. The spread between the best and worst win rate in this table is 4.6 percentage points across a hundred years of combined trader folklore. A shape you eyeball off the chart and a momentum oscillator with three parameters landed within noise of each other.

The engulfing bar and the EMA crossover tying at 44.8% is rounding, not destiny — but the raw signal counts are worth noticing too. 505 tradable engulfing entries against 499 MACD entries over the same window. Two systems that share almost no bars produce almost the same number of trades and almost the same result.

Does confirmation help or hurt?

Here is the test the advice deserves. Take the engulfing bar, and trade it only when the indicator agreed within the previous 30 minutes.

MACD confirmation nearly doubled the loss. It cut the trade count roughly in half and kept the wrong half — the win rate fell 2.3 points and the average fell 0.61 pips. At a four-hour hold the damage is larger: -1.85 pips confirmed against -0.89 unconfirmed, and the win rate drops from 47.0% to 39.4%.

The EMA crossover did the opposite. Same idea, same 30-minute window, and it turned the only losing rule in the table into the only profitable one. It also threw away three quarters of the signals to do it: 125 trades in three months, roughly one every eighteen hours.

Why did the slow filter work and the fast one fail?

Because they are filtering on different things, and only one of them is filtering.

MACD’s signal line fired 1,467 times on these bars — once an hour. A filter that is true roughly half the time is not selecting anything; it is splitting your trades into two arbitrary piles, and my MACD crossover backtest showed why: 63% of those signals were cancelled by the opposite signal within the hour. Asking a coin whether to take a trade will cost you money the moment the coin is correlated with anything, and here it is negatively correlated with the outcome.

The EMA 12/26 crossover fired 672 times — less than half as often — and, unlike the signal line, it does not un-cross casually. It marks that the two averages have actually swapped places. Requiring an engulfing bar to arrive inside that window is a genuine restriction: 125 of 1,644 patterns qualified, 7.6%.

This is the same pattern I keep running into on this dataset. In the moving average crossover test the only profitable pair was the slowest one. In the ADX above 25 test the filter helped precisely because it removed most of the trades. A filter earns its keep by how much it refuses, not by how often it agrees.

What was the best combination overall?

The best line in the table is the morning and evening star, the pattern with the least marketing behind it, held four times longer than a five-minute trader would hold anything, confirmed by an indicator that hurt every other rule I applied it to.

I want to be careful about how much weight that carries. The star’s win rate barely moved when MACD was added — 52.4% to 52.1% — so the improvement in average pips came from which trades were dropped, not from a better hit rate. That is exactly the shape of a result that does not repeat. The honest reading is that the star is mildly positive here and MACD did not ruin it, which is different from MACD helping.

So is any of this worth putting on a chart?

Three things came out of the comparison that I would keep:

• Candles and oscillators are not two opinions. They agree on the same bar 6.7% of the time. If you are stacking them for confidence, you are stacking two things that rarely look at the same event.
• Widen the window and confirmation becomes meaningless. An hour either side and MACD said yes to 46.4% and no to 45.7%. You will always find the answer you wanted.
• Slow filters beat fast ones, again. The EMA crossover kept 7.6% of the patterns and was the only combination under two hours that cleared the spread.

What none of it supports is the idea that a shape plus an indicator is a system. The full settings grid — wick ratios, body sizes, holds, sessions and the seven variants out of fifty-nine that survived costs — is in my candlestick pattern settings guide for MT4.

Frequently asked questions

Should you confirm candlestick patterns with an indicator?

Not with a fast one. Requiring MACD to agree within 30 minutes made engulfing bars worse — -1.25 pips a trade against -0.64 unfiltered at a two-hour hold, and -1.85 against -0.89 at four hours. A slower EMA 12/26 crossover as the filter turned the same rule positive at +1.31 pips over 125 trades.

How often does MACD agree with a candlestick pattern?

On the same bar, 6.7% of the time. Widen the window to an hour either side and MACD fired near 92.1% of engulfing bars, agreeing 46.4% of the time and contradicting 45.7% — statistically a coin flip.

Are candlestick patterns better than MACD or RSI?

They are equivalent on this data. Engulfing bars won 44.8% of trades, an EMA 12/26 crossover 44.8%, RSI 14 42.8% and MACD 41.9%. All four lost money after a 1-pip spread.

What is the best candlestick and indicator combination?

Morning and evening stars with MACD agreement, held four hours: 52.1% wins and +1.29 pips net across 119 trades. The unfiltered star was almost as good at +0.92, so most of the credit belongs to the pattern and the holding period rather than to MACD.

Why does adding a filter reduce performance?

Because a filter that is true half the time is not selecting anything — it just splits your trades in two. MACD’s signal line fires roughly once an hour on 5-minute EURUSD, so it does not restrict the pattern set meaningfully. The EMA crossover kept only 7.6% of patterns, which is what a real filter looks like.

How many signals do candlestick patterns produce?

Over 16,926 EURUSD 5-minute bars: 1,644 engulfing bars, 718 stars and 124 pin bars, against 1,467 MACD signal crosses, 1,003 RSI zone exits and 672 EMA 12/26 crossovers. Full frequency breakdowns are in the candlestick pattern test.

What data was this run on?

16,926 EURUSD 5-minute bars, June 3 to August 26, 2026, with costs charged at the 1-pip spread measured in my spread and swap costs test. The same set is behind my RSI divergence against overbought comparison.