Do Trading Indicators Actually Work?
Do indicators work is a question that can only be answered against a control. Tested here, entries taken on a MACD cross performed the same as entries taken on a fixed schedule with the same stop and target, which means the indicator described the move rather than anticipating it.
The question is usually argued. It is answerable, but only if the indicator is tested against something — and the something almost nobody uses is entering at random.
How it works
Every MACD bullish cross across 576 bars was taken as a long, with the stop and the target both at 1.5 times the median bar range.
26 signals. 14 reached the target, 11 were stopped, one was still open at the end.
On its own that looks like a working indicator. A little better than even money, which is what a tutorial would show you and stop.
Against a control
The same 26 trades, entered on a fixed schedule instead — every 21st bar, with no reference to any indicator at all.
14 reached the target. 12 were stopped.
The indicator and the calendar performed the same.
That is the measurement the question needs, and it is the one that is almost never run. A signal is only worth its complexity if it beats entering without one, and here it did not.
Neither arm’s absolute record means anything. Both are longs on a history built to drift upward, so both are flattered by the same amount — which is exactly why the comparison is readable even though the individual numbers are not.
That is what a control is for. It absorbs everything the two arms share — the drift, the volatility, the cost, the stop distance — and leaves only the thing being tested.
The result that disagrees
The identical test on the slower view of the identical history: nine crosses, 3 wins and 6 losses. Nine scheduled entries, 7 and 2.
On that sample the indicator looks actively harmful.
Both tests are honest and they contradict each other. The difference is that one has 26 trades and the other has nine, and nine trades cannot distinguish anything — as the trading journal page puts it, 25 trades leaves a win rate uncertain by about 20 points either way.
So the honest conclusion is the weaker one: on this data the MACD cross did not beat a schedule, and anyone claiming a stronger result in either direction from a sample this size is reading noise.
Why lag is structural
Across 14 swing lows, the confirming cross arrived a median of 4 bars later.
That is not a flaw in the MACD. Every indicator is computed from prices that have already printed, so a turn can only be reported after enough bars exist to establish it. The moving average page shows the same arithmetic in its simplest form.
An indicator is a description of what has happened. Treating a description as a prediction is the mistake, and it is built into the way they are usually taught rather than into the tools.
What they are actually good for
Three uses survive the test above, and all three are smaller claims than “signal”.
Measurement. ATR tells you how far this instrument moves, which sets a stop distance you would otherwise guess. Nothing on this page argues against that, and it is the most valuable thing an indicator does.
Filtering. A condition that stops you taking a trade — the VWAP strategies page measures one — changes what you skip rather than what you enter.
Consistency. A rule you apply the same way every time is testable; a judgment is not. That is worth something even when the rule itself is no better than a schedule.
A worked example
Take your indicator’s signal and count every firing across a long history. Every one, not the good ones.
Then take the same number of entries on a schedule — every twentieth bar, ignoring the chart completely.
Trade both with the same stop and the same target.
If the indicator does not beat the schedule, it is not an entry signal. It may still be a useful measurement, and that is a different job with a different test.
The original data
Across our study of 24,971 trading videos, 147 ask whether indicators work. The median one gets 1,055 views — the second-lowest of any subject measured here, behind only forex stocks.
86% never pass 50,000 views, and the median length is 8.7 minutes.
The corpus carries description text for 27 of those 147, and across those 27, three mention invalidation, failure, or what a bad read looks like.
A field of 147 videos asking whether the tools work, of which about one in nine mentions a case where a read was wrong. Meanwhile the MACD itself has an audience many times this one — the question is much less popular than the tool.
When it fails
The test proves too much
One rule, one instrument, one synthetic history is not a verdict on indicators. It is a demonstration of how the question has to be asked, and the answer will differ by market, by rule and by period.
You test only the good conditions
Running the test on a trending stretch and skipping the range above is how every indicator passes. The control has to see the same bars.
You conclude they are useless
Not what the measurement says. It says the cross was not an edge over a schedule on this data. Measurement and filtering are untouched by that, and they are most of what indicators are for.
You judged from the right-hand edge
At the moment it fires, the three that won and the six that lost are the same picture. Which is why the count matters more than any example either of us could pick.
Related
Choosing indicators is the next question: if they are measurements, which ones measure different things?
MACD is the tool tested on this page, explained properly.
And what nobody tells beginners has the scarcity numbers that make small samples unavoidable.
I use indicators every day and I would not describe any of them as a signal. What they do for me is measure something I would otherwise be estimating by eye - how far price has stretched, how wide the bars are, where the average sits. The moment I started treating them as answers rather than measurements, they stopped helping.
— Michael Whitman, from this video
This page is educational, not financial advice. Test every idea on your own charts before risking money.