RSI, MACD and Volume Together: Which Signals Actually Matter?

Reading RSI, MACD and volume together instead of trusting any one of them is the most common advice in technical analysis, and our own scanner is built on it: eight signals, added up into one score. But "use several indicators" only helps if each one adds something the others don't. So we took the score apart. We removed each signal in turn, rebuilt the score without it, and measured what changed. Only one removal clearly hurt, and it wasn't RSI or MACD.

Why combine signals at all

Run any single indicator across a whole exchange and it returns noise. "RSI below 30" gives you a list of coins falling for good reasons. "Volume spike" gives you wash trading and delisting panics. The idea behind combining signals is that a real setup should show up in several independent places at once: momentum turning, volume arriving, structure holding.

The key word is independent. When we broke down our scoring, we wrote that three momentum indicators agreeing is "one signal wearing three hats." That is easy to say. It is also a claim you can test, so we tested it on ourselves.

How we tested it

Rally Radar's score has eight components: volume expansion, price breakout, EMA alignment, MACD, volatility compression, StochRSI, RSI and higher lows. The method is called leave-one-out:

• Remove one component and recompute every hourly score without its points. Everything else stays the same.
• Let each version pick the same number of coins. The full score put 1,540 coin-hours at 85 or above over the period, so each stripped-down version picks its own top 1,540. Removing points lowers every score, so a fixed cut-off of 85 would measure the cut-off rather than the signal.
• Measure against a matched baseline. For each flagged coin we asked whether it reached +10% within 24 hours, then compared that with control coins from the same hour, with similar recent volume and momentum, that the scanner had not rated highly. The gap between the two is the lift.

The data is every hour across 315 Binance USDC pairs from April to late August 2026. With all eight components in place, the score's lift was +4.3 percentage points, with a 95% interval of [+2.0, +6.6], over 1,322 matched episodes. That is our reference point. This comparison is stricter, and runs over a longer window, than the one on our track record page, so the two figures are not meant to match.

What happened when we removed each one

The table shows how the lift changed when each component was taken out. A negative number means the score got worse without it, so the component was doing something. A positive number means the score did slightly better without it. The interval is the range the true effect plausibly sits in; if it spans zero, the change is indistinguishable from noise.

Seven of the eight intervals cross zero. One doesn't.

Volume was the one that mattered

Without the volume component, the score's lift fell from +4.3 to +1.8 points. (That is a drop of 2.5, not 3.0, because the table's change is worked out day by day and then averaged, so every day counts equally however many coins were flagged on it.) It was also the only removal that changed which coins got picked in a big way: the stripped-down score kept just 40% of the full score's selections. Everything else kept between 57% and 85%.

That is notable because volume still feeds the score elsewhere. Both the compression and StochRSI components require volume confirmation before they award points. Even with that back door left open, removing volume as a component of its own did the most damage.

The reason shows up when you check how the components move together. Correlation runs from -1 to 1. Near 0 means two signals tell you unrelated things; near 1 means they mostly say the same thing. Volume's correlation with every other component was low, the highest being 0.27 with price breakout. It is the one signal in the score that consistently tells you something the others don't.

RSI looked better removed, and why

The score did slightly better without RSI: +1.0 points, with an interval of [-1.4, +3.6]. That interval comfortably includes zero, so this does not show RSI is harmful. What it shows is that RSI isn't adding anything the score is missing.

The correlations explain it. RSI correlated 0.68 with EMA alignment and 0.59 with higher lows. When price has been rising cleanly, all three light up together: the moving averages stack, the swing lows step up, and RSI sits in its bullish zone. They are three views of the same uptrend. Count them as three agreeing signals and you are counting one signal three times. That is exactly the trap we warned about, sitting in our own score.

MACD came out slightly useful, at -0.8 [-2.0, +0.5], though not clearly so. Its most interesting number is its correlation of -0.49 with StochRSI. The two tend to fire at different moments. MACD rewards momentum that is already building, and StochRSI rewards a turn up out of oversold. They are not confirming each other; they cover different situations.

Why we haven't changed the score

The obvious next step would be to drop RSI and give volume more weight. We haven't, and these are the reasons.

• This is an in-sample, exploratory study. Every figure was measured on the same five months of data, with no held-out period to check against. It produces a hypothesis, not a finding. Changing what every visitor sees needs a test run on data the idea wasn't drawn from.
• We looked at a lot of numbers. Nine versions of the score across four time horizons gave 36 figures. If every component were pure noise, roughly two of those would still land outside zero by chance. Volume's interval stops at -0.2, barely clear of zero. It is the strongest candidate, not a proven one.
• Leave-one-out can't see redundancy. If RSI and EMA carry the same information, removing either one alone costs nothing, because the other covers for it. Removing both could cost a lot. "RSI is redundant" is a different claim from "RSI is useless".
• The score has a ceiling. The eight components can add up to more than 100, but the score is capped at 100. That cap was active on 17.5% of the full score's top picks, so on those coins a removed component often cost little or nothing.
• None of this is about profit. "Reached +10%" means price touched that level at some point within 24 hours. It may have dropped further first, and nothing here includes stops, exits or fees. When we tested trading the score after costs, on data it had not been tuned on, it showed no edge. That result is on the track record page.

What the study does give us is a ranked shortlist for the next proper test: volume first, then the RSI–EMA–structure overlap. That test will be pre-registered, run once on new data, and published here whichever way it comes out.

What this means for reading RSI, MACD and volume on your own charts

You don't need a scanner for any of this. The lessons carry straight to a chart with three indicators on it:

• Count families, not indicators. RSI, EMA stacking and higher lows all agreeing is one piece of evidence: the trend is up. It is not three. Before you call a setup "confirmed", ask how many different kinds of evidence you have.
• Make volume a requirement, not a bonus. Of the eight things our score looks at, it was the only one whose absence clearly made the ranking worse. A breakout or a momentum turn on flat volume deserves the most suspicion. The accumulation guide covers what good volume looks like on a chart.
• Pair indicators that disagree on timing. MACD and StochRSI rarely fire together because they watch different phases of a move. That is a reason to keep both: between them they cover more situations. It is not a reason to wait for both to agree.
• Test your own combination before you trust it. If your system adds up five indicators, take one out and check whether your past signals actually change. If they don't, that indicator was decoration.

For the full points table behind the score, read what makes a good crypto setup. For how we test claims like these, and why a win rate proves nothing, see do crypto scanners actually work. And for how far ahead the score tends to flag a rally, how early altcoin rallies show up on a scanner has the numbers.

← Back to all posts