Stochastic Buy and Sell Signals Tested: 4,263 %K/%D Crosses Made 0.00 Pips Before Costs, and Half Reversed Within 15 Minutes

The %K/%D cross is how most people are taught to read Stochastic: the fast line crosses above the slow line, buy; it crosses below, sell. Better still if it happens under 20 or over 80. It is also the signal every free “Stochastic arrows” indicator draws. I have already tested Stochastic as an overbought/oversold fade. This test is only about the crosses.

The short answer: on EURUSD M5 the cross was a coin with a fee attached. The zone version was a slightly better coin with the same fee.

How were the crosses tested?

• Data: 16,925 EURUSD M5 bars, June 3 to August 26, 2026, split at July 15 into two halves.
• Settings: 14/3/3, the MT4 default, and 5/3/3, the popular fast version.
• Signal: %K crosses above %D = buy, below = sell, confirmed on the closed bar. Entry at the open of the next bar.
• Filters: every cross; only crosses below 20 (buys) or above 80 (sells); the same with 30/70; only crosses in the middle between 20 and 80.
• Exits: after 3, 6 or 12 bars (15 minutes, 30 minutes, 1 hour), or at the next opposite signal.
• Cost: 1 pip per trade, taken from my spread and swap test. One trade at a time.
• Baseline: the same number of random entries, 20 runs, to see what “no signal” costs.

What did 4,263 Stochastic crosses pay?

Pips per trade, EURUSD M5, 1 pip of cost included in “net”.

Every row is the cost of the trade and nothing else. Before costs, the cross earned between -0.17 and +0.04 pips per trade, which is zero with rounding noise on both sides. The random entries landed in the same place. The fast 5/3/3 did not change the picture: -0.89 net at 30 minutes, -1.29 at one hour.

Flipping the signal did not rescue it either. Selling the buy cross and buying the sell cross also lost exactly 1.00 pips per trade over 30 minutes. When a signal and its opposite lose the same amount, the signal carries no information about direction.

Why do so many crosses fail at once?

Because %K and %D sit very close together most of the time, and a 5-minute bar can push one across the other and back. With 14/3/3:

• The median gap between one cross and the next was 3 bars.
• 53.9% of crosses were followed by an opposite cross within 15 minutes.
• 14.3% were reversed on the very next bar.

That is why “hold until the opposite cross” was the worst exit, winning only 28.5% of trades. Most trades ended one to three bars after they started, paying the full pip each time. The same pattern showed up in my MACD crossover test, where 63% of 1,482 signals reversed within a few bars. A cross of two lines that follow each other closely is mostly noise.

Do crosses in the overbought and oversold zones work better?

A little. That is the textbook advice, and it was the best of the filters I tried:

The ordering is exactly what the textbooks say: zone crosses beat all crosses, and middle crosses were the worst. The size of the difference is the problem. The best filter found about 0.16 pips per trade. A normal EURUSD account pays 1 pip.

Letting the zone trades run until the next opposite zone signal raised the win rate to 60.6% and made things worse, -1.08 per trade. The average winner was 5.67 pips and the average loser 11.47. It is the same shape as the Bollinger Bands fade on M5: many small wins paid for by a few large losses.

Why is the zone cross so late?

I measured how far price moved in the 30 minutes before each signal and in the 30 minutes after it, both in the direction of the trade:

Pips in the direction of the signal, before costs. Negative “before” means price had been moving against the new trade.

A buy cross below 20 comes after a drop, on average 4.17 pips in half an hour. That is simply how the indicator is built: %K can only be under 20 after price has fallen toward the bottom of its range. What the cross does not deliver is the bounce. After entry, price moved 0.11 pips the right way. The indicator described a move that had already happened. The RSI divergence test found the same thing, with 75% of the move gone before the signal.

Did the time of day matter?

Less than the cost. Zone crosses held for 30 minutes, by session (UTC):

The quiet Asian hours were the least bad for a mean-reversion signal, which fits what I found in session vs all-day trading. No session came within 0.7 pips of paying for itself.

So how would I use Stochastic buy and sell signals?

• Not as a stand-alone entry on EURUSD M5. Every cross, every filter and every exit I tried lost money after a 1-pip cost, and the plain cross made nothing even before it.
• If you use crosses, only use the zone crosses. Below 20 for buys, above 80 for sells. They were the only version with something positive before costs.
• Ignore crosses in the middle of the range. They were worse than random.
• Do not wait for “one more cross” to confirm. Half of all crosses are reversed within 15 minutes, and waiting only moves you further from the move that already happened.
• Change the chart before changing the numbers. On H1 the picture was different. The settings, timeframes and exit that worked are in my Stochastic crossover settings for MT4 guide. The comparison with MACD, RSI and moving averages is in Stochastic cross vs MACD.

Frequently asked questions

Is the Stochastic %K/%D cross a good buy and sell signal?

Not on EURUSD 5-minute charts. On 16,925 bars the default 14/3/3 produced 4,263 crosses. Held for 30 minutes they made 0.00 pips per trade before costs and lost 1.00 pip after a 1-pip cost, about the same as random entries.

When should I buy with the Stochastic?

The least bad buy signal in this test was %K crossing above %D while %K was still below 20. Those crosses made +0.16 pips per trade before costs over 30 minutes, which is less than a normal spread.

How often does a Stochastic crossover fail?

Often and quickly. 53.9% of %K/%D crosses were followed by an opposite cross within 3 bars, and 14.3% were reversed on the next bar.

Are Stochastic crosses in the middle of the range useful?

No. Crosses between 20 and 80 lost 1.21 pips per trade over 30 minutes, worse than random entries at 0.97.

Is 5,3,3 better than 14,3,3 for crossover signals?

Not on M5. The fast 5/3/3 gave 4,623 crosses and lost 0.89 pips per trade over 30 minutes, against 1.00 for 14/3/3. Both are the cost of the trade with no edge on top.