What MACD actually computes

MACD was created by Gerald Appel in the late 1970s (MACD). The standard build, written MACD(12,26,9), is three pieces:

• MACD line = the 12-period EMA minus the 26-period EMA

• Signal line = the 9-period EMA of the MACD line

• Histogram = the MACD line minus the signal line

That is the whole indicator. Every component is an exponential moving average of past prices, or an average of those averages. Nothing here observes volume, order flow, or anything outside the price series the chart already shows you.

Traders use it three ways: the signal-line crossover (MACD crossing its own signal line), the zero-line crossover (MACD crossing zero, which happens precisely when the 12 EMA crosses the 26 EMA), and divergence between MACD and price.

Worth pausing on the second one. A zero-line cross is not an independent event. MACD is above zero exactly when the fast EMA is above the slow one, so "MACD crossed zero" and "the 12 crossed the 26" are the same sentence in different clothes. The moving-average crossover from Part 14 is sitting inside this indicator, wearing a different panel.

The number itself means nothing

Here is the property that almost every MACD tutorial skips. MACD is what is formally called an absolute price oscillator: it deals in the actual prices of moving averages rather than in percentage changes. The consequence, stated plainly on the reference page, is that it "will show greater levels for higher priced securities and smaller levels for lower priced securities," so MACD values are not directly comparable across different securities.

This is the exact opposite of RSI. Part 15 showed that RSI's 70 is a real, checkable statement, that average gains have run about 2.33 times average losses. MACD has no equivalent. There is no threshold to memorise because the scale is different for every asset, and different for the same asset at different price levels. Anyone telling you a particular MACD reading is high is telling you nothing portable.

Lag on top of lag

Part 14 established that a moving average cannot turn until the move it is meant to catch is already underway, because it averages prices that have already printed. MACD inherits that problem and adds to it.

The MACD line is the difference of two lagging averages, so it already trails price. The signal line is then a nine-period EMA of that, which trails the MACD line. A signal-line crossover therefore fires when a lagging series crosses a smoothed version of itself, which is a second-order lag: late, and then late again. The histogram is simply the gap between those two, so its turning points necessarily come before the crossover prints, which is why traders watch it. That earlier turn is not free information: it is the same series with one smoothing step removed, so it changes direction more often, which is the ordinary cost of less smoothing rather than an edge.

None of this makes MACD useless as a description. It makes it structurally incapable of being early.

What the testing actually shows

The most directly relevant evidence comes from the same crypto study Part 15 drew on, which tested four rule families on BTC/USDT and ETH/USDT: exponential moving-average crossovers, RSI, Bollinger Bands and MACD (Chen, Wang & Yang, 2023). The researchers did not test one MACD setting; they swept the fast period from 10 to 25, the slow period from 25 to 40 and the signal period from 6 to 12, then applied White's Reality Check and Hansen's test with a slippage assumption of 0.001 and a commission rate of 0.0003.

The best MACD configurations out of that sweep looked respectable in sample: an annualised 34.1% on Bitcoin with a Sharpe ratio of 0.574, and 33.5% on Ethereum with a Sharpe of 0.462. Those are best-of-sweep figures rather than the result of a single setting chosen in advance. Out of sample, from 20 December 2021 to 31 October 2023, both failed to generate significant profits.

But the sharper finding is this. After correcting for data-snooping with White's Reality Check and Hansen's test, MACD profits were not statistically significant in the in-sample period either, not just out of sample. They did not clear the bar even on the data the parameters were chosen from. The authors' overall conclusion across the strategies they tested was that profitable in-sample results generally failed to survive out of sample, a pattern they read as consistent with market efficiency in crypto.

That correction is the benchmark for this kind of work, set by Hsu and Kuan when they applied White's Reality Check and Hansen's SPA test across a universe of nearly 40,000 trading rules, the same study Part 14 drew on (Hsu & Kuan, 2005). A rule that cannot clear it has not been shown to work; it has been shown to look good among the many rules that were tried, which is exactly what sweeping three parameters across dozens of combinations invites.

One more note, on a study you will see cited. A 2008 paper in Applied Economics Letters set out to test MACD and RSI rules across sixty years of the London FT30, and its headline result circulates widely as evidence that both beat buy-and-hold. That specific finding did not survive the verification pass behind this series, so it is not repeated here. Treat confident citations of it with the same caution you would apply to any statistic you have not traced yourself.

Reputation versus evidence

Using MACD honestly

• Read shape, never level. The cross, the slope and the side of zero carry information. The number on the axis does not.

• Do not treat the zero cross as confirmation of the EMA cross. They are the same event. Counting it twice is how a single piece of evidence starts to feel like two.

• Expect it to be late, and size for that. A second-order lag means the entry is well into the move, so a stop placed at the signal is often placed at the worst possible spot.

• Be suspicious of tuned settings. Sweeping fast, slow and signal periods is precisely the search that data-snooping corrections exist to punish, and in the crypto test the tuned winners did not clear that bar.

• Use it as a momentum description alongside independent evidence. Structure, levels and volume are not derived from the same two EMAs. MACD is.

Common mistakes

• Comparing MACD values across assets - the scale is the asset's own price units, so the comparison is meaningless.

• Trading the crossover mechanically - it is the latest of the lagging signals, being an average of an average.

• Counting the zero cross as a second signal - it is the 12/26 EMA cross by another name.

• Optimising 12, 26 and 9 until the backtest shines - that is the search the statistics are designed to catch.

• Reading the histogram as a leading indicator - it turns slightly sooner and produces correspondingly more false turns.

MACD on ApeX Omni

ApeX Omni's charts carry MACD alongside the price action this series has taught you to read. The honest use on a perpetuals venue is narrow: let it describe whether momentum is building or fading, and take the actual decision from structure, levels and volume, which are not computed from the same two averages. The lag matters more here than on a spot chart, because entering late on leverage means entering with less room between your fill and your liquidation price, exactly the arithmetic from Part 9. If you trade a crossover, put the stop where the structure says rather than where the indicator turned, and trigger it on the mark price as Part 12 argued. Trade the chart on ApeX Omni, and treat the crossover as a description of what already happened.

The bottom line

MACD is two lagging averages subtracted from each other, smoothed again, and plotted on a scale that means something different for every asset you apply it to. That construction explains both of its real limits: the signal cannot be early, and the level cannot be compared. What it can do is show you, at a glance, whether short-term momentum is running ahead of or behind the longer trend, which is a genuine if modest thing to know. What the evidence does not support is the way it is sold. Swept across parameters and corrected for the number of combinations tried, its crypto profits were not statistically significant in sample or out of it. Read the shape, ignore the number, and never count the zero cross twice.

Next in this series: Bollinger Bands, the envelope that supposedly marks when price has gone too far, and what actually happens at the edges.

Frequently asked questions

What does MACD stand for and what does it measure? Moving Average Convergence Divergence. It measures the gap between a fast and a slow exponential moving average of price, showing whether short-term momentum is running ahead of or behind the longer trend.

What are the standard MACD settings? The conventional build is MACD(12,26,9): a 12-period EMA minus a 26-period EMA for the MACD line, and a 9-period EMA of that line as the signal. The histogram is the difference between the two.

Is a MACD crossover a reliable buy signal? The evidence is weak. In a 2023 crypto study that swept MACD parameters and applied data-snooping corrections with transaction costs, MACD profits were not statistically significant in the in-sample period or out of sample.

Why is MACD different on different assets? Because it is an absolute price oscillator, expressed in the price units of the asset itself. Higher-priced securities produce larger MACD values, so readings are not comparable across assets or across very different price levels.

Is there an overbought MACD level like RSI's 70? No. MACD is unbounded and asset-specific, so no fixed threshold exists. Only the shape of the lines, the crossover, the slope and the side of zero, carries portable information.

What is the MACD histogram for? It plots the distance between the MACD line and its signal line, so it turns slightly before a crossover completes. That earlier turn also means more false signals, so it is a sensitivity dial rather than a better signal.

Explore more from this series: Part 9: Leverage and Liquidation | Part 14: Moving Averages | Part 15: RSI Explained | Part 17: Bollinger Bands (Coming Soon)

This article is for educational and informational purposes only and is not financial, investment, or legal advice. Do your own research before making any trading decision.