The MACD crossover is one of the most widely taught entry signals in trading. It appears in almost every beginner course, on every charting platform by default, and in an enormous number of strategies that are otherwise unrelated to each other.
The logic is sound. MACD measures the gap between a fast moving average and a slow one. When that gap starts widening upward — when the MACD line crosses above its own nine-period average, the signal line — momentum has just turned. It is a real observation about price, not superstition.
So we tested it. On 12 September 2026 we ran a bullish MACD crossover across every stock in the S&P 500 on Trade Manager. It produced 1,761 trades — the largest sample we have published — and won 569 of them.
It still lost almost the entire account. This guide is why, and the reason is one percentage point.
The rules we tested
One condition, on the daily timeframe, across all 503 index members:
- Trigger — a bullish MACD crossover, 12 / 26 / 9. The MACD line (the 12-period EMA minus the 26-period EMA) crossing above its 9-period signal line. These are the default settings, unchanged.
- Exits — a 6% target and a 3% stop, fixed, no discretion. A 2-to-1 payoff, which is the risk-reward ratio most commonly recommended to new traders.
There is deliberately no confirmation filter. We wanted the signal itself on trial, not a combination that would make it ambiguous which component was responsible for the result.
The result
| Metric | Value |
|---|---|
| Closed trades | 1,761 |
| Won | 569 |
| Lost | 1,192 |
| Win rate | 32.3% |
| Break-even win rate required | 33.3% |
| Maximum drawdown | −99.4% |
| Longest losing streak | 26 trades |
| Average result per trade | −0.15% |
| Average hold | 5 bars |
| Universe and window | 503 of 503 tickers, 18 March – 10 September 2026 |
Sixty-six of the 1,761 trades had not reached either exit when the window closed and were valued at the last price. Excluding those, the win rate is 31.5% rather than 32.3%. Both numbers sit below the threshold, so the conclusion does not depend on which you use.
Why it lost, in one line of arithmetic
A strategy that wins 569 times looks like a working strategy. That intuition is what this result is useful for dismantling, because the number that decides the outcome is fixed before a single trade is taken:
Break-even win rate = stop ÷ (target + stop)
With a 6% target and a 3% stop, that is 3 ÷ (6 + 3) = 33.3%. The strategy scored 32.3%. It was short by a single percentage point.
One point sounds like a rounding error, and on any individual trade it is invisible. The average trade in this test lost 0.15% — an amount no trader would notice, react to, or remember. There is no bad trade in this ledger to point at. There is no moment where something obviously went wrong.
What there is instead is repetition. A 0.15% average loss compounded across 1,761 trades is not a small loss; it is the account. The equity curve fell 99.4% from its peak, and included a stretch of 26 consecutive losing trades — a sequence that no real person would sit through without concluding something had broken. Nothing had. That is simply what a slightly-negative edge looks like when it is repeated enough times.
The same signal with different exits
Nothing about MACD changes if you move the exits. The requirement does:
| Target / stop | Break-even win rate | 32.3% clears it? |
|---|---|---|
| +3% / −3% | 50.0% | No |
| +6% / −3% (this test) | 33.3% | No — short by 1.0 |
| +7% / −3% | 30.0% | Yes |
| +9% / −3% | 25.0% | Yes |
This is not a recommendation to widen the target until the maths works. A wider target is reached less often, so the win rate falls as you move down the rows — you cannot hold one column still and move the other. The point is narrower and more useful: this strategy sat one point from viability, and the lever that decides which side of the line it lands on is your exits, not the indicator.
What a large sample buys you
With 106 signals you can reasonably argue a result was luck. With 1,761 it is much harder. That is the one genuinely good thing about this test: MACD crosses its signal line often, so the sample is large enough that the 32.3% is unlikely to be noise.
It cuts the other way too. A strategy with a small negative edge and very few trades loses slowly enough to be survivable. The same edge traded 1,761 times does not give you that room. Frequency amplifies whatever edge you have, including a negative one.
What the MACD crossover is still good for
None of this makes MACD useless, and the result should not be read as "MACD does not work." The crossover is a momentum observation, and it was often correct — 569 times it was followed by a 6% gain before a 3% loss.
What it is not is a complete strategy. It says momentum has turned. It does not say by how much, for how long, or whether the move ahead is large enough to clear your target before your stop. Those are the questions your exits answer, and this test is what happens when the exits are chosen from a rule of thumb rather than from the signal's own behaviour.
We have now published the same shape of failure three times. The hammer candle missed its bar by 4.2 points. The breakout strategy missed by 7.7. MACD missed by 1.0. Three different setups, three different reasons to believe in them, and one identical cause of death.
How to run this test yourself
The result above is one strategy on one window. Yours will differ, and the point of a backtest is that you do not have to take anyone's word for it:
- Open the strategy builder and pick a universe — the S&P 500, the NASDAQ 100, or a handful of tickers you already follow.
- Add a MACD Crossover block as the trigger. The defaults are 12 / 26 / 9, crossing above the signal line.
- Set your target and stop, then work out the break-even win rate before you run anything: stop ÷ (target + stop). Write the number down.
- Run the backtest and compare the win rate it reports against the number you wrote down. That comparison is the result — not the profit figure.
- If it lands close to the line, treat "close" as a failure rather than a near miss. One point short is still short, and this page is what that costs.
Backtests run bar by bar against real historical data, using the same evaluation code as the live scanner, so a strategy that tests well behaves the same way when you switch alerts on.
Three caveats that apply to every backtest here
Survivorship bias. The test uses index membership as it stands today. Companies removed from the S&P 500 during the period are not included, and removals skew toward poor performers, so real-world results would likely be worse than shown.
Frictionless fills. Trades are assumed to fill at exactly the target or stop price, with no slippage and no commissions. At 1,761 trades this matters more than usual: commissions alone would deepen the loss considerably.
Position sizing. The equity model compounds a single position sequentially, which is what produces a drawdown as extreme as 99.4%. No sensible trader would risk the whole account on every trade, and with realistic sizing the loss would be far smaller. The win rate, the trade count, and the one-point gap are the durable findings here — the drawdown figure describes the model's sizing, not a forecast for your account.
The first two caveats push in the same direction, which is worth sitting with: this strategy failed to clear its break-even line under assumptions that flattered it.