What Is the MACD Indicator, and What Has Testing Shown?
The MACD subtracts a slow exponential moving average of price from a fast one, plots a second average of that difference as a signal line, and leaves the gap between them as a third series.

The MACD takes two exponential moving averages of price, one fast and one slow, and plots the distance between them as a single line. A second exponential average of that line is drawn on top of it, and the distance separating the two becomes a third series, so one calculation puts three things on your screen.
Two platform references document the arithmetic, and the formulas below come from them. What a reader is told to do with the result is interpretation, so every reading in this post is attributed to whoever published it.
What does the MACD indicator show?
The MACD shows how far apart two moving averages of price have drifted. Sierra Chart's Technical Studies Reference classifies it as a trend-following momentum study and describes the measured quantity as the relation between two price averages, and its MACD study prints three series from that one calculation.
When the fast average sits above the slow one, the MACD line is positive. When the slow average is higher, the line is negative. Zero is the level at which the two averages agree, which Coakley, Marzano and Nankervis describe in their 2016 study as the point where the fast and slow averages show no difference.
The second series is an average of the MACD line itself, which Sierra Chart names the Moving Average of the MACD; Quantower's documentation calls it the signal line. The third series is the gap between the first two, and Sierra Chart calls it the MACD Difference. If you go looking for the word histogram, neither the Sierra Chart MACD study page nor Quantower's MACD page uses it.
In a platform's study list, "the MACD" is a family. The contents page of Sierra Chart's Technical Studies Reference carries seven studies with MACD in the name, including volume-weighted and Bollinger-band variants and a signal-generating system, so the study you pick decides what gets computed.
Who created the MACD?
Two of the sources read for this post credit Gerald Appel, and Quantower's page names nobody. Sierra Chart's MACD page states that he developed it and gives no date. Coakley, Marzano and Nankervis say the indicator was introduced by Appel, and their reference list gives the work as Technical Analysis Power Tools for Active Investors, dated 1999.
Appel's own text was not read here, so nothing in this post describes what he wrote or what he recommended. That gap matters for the 12, 26 and 9 period lengths in particular. Quantower's page calls them common periods, the 2016 study fixes the same three numbers in its description of the indicator, and Sierra Chart prints no numbers at all. None of the three presents them as a tested optimum.
Sierra Chart instead exposes the three lengths and the averaging method as inputs and leaves the choice to the user, which is a fair reflection of the situation: the periods are a convention, and the software treats them as a setting.
How is the MACD calculated?
Three formulas, in the order the platform computes them. Subtract the slow exponential moving average of the input series from the fast one to get the MACD line. Take an exponential moving average of that line, over the third length, to get the signal line. Subtract the second from the first to get the difference series.
| Series | What it is computed from | Vendor names |
|---|---|---|
| MACD line | Fast EMA of price minus slow EMA of price | "MACD" on both platforms |
| Signal line | An EMA of the MACD line, over the third length | "Moving Average of the MACD" (Sierra Chart), "signal line" (Quantower) |
| Difference | MACD line minus signal line | "MACD Difference" (Sierra Chart) |
Quantower's page lists the common periods as 12 for the fast average, 26 for the slow and 9 for the signal line. Sierra Chart documents the same three inputs without numbers and adds an averaging choice: the formula is written with exponential averages, and six alternatives can be substituted, among them Wilders, simple, weighted and linear-regression averages. Two traders with the same three lengths and different averaging types are reading two different lines.
Sierra Chart also states when each series becomes defined. The MACD line exists from the first bar; the signal line and the difference exist only from the bar given by the longer of the fast and slow lengths plus the signal length. Neither vendor's formula refers to the instrument, so the same calculation runs on any series the platform will chart.
What is a MACD crossover, and what does a zero-line cross mean?
A crossover is the moment the MACD line passes through its own signal line; a zero-line cross is the moment the MACD line passes through zero, when the two underlying averages are equal. The readings attached to those events belong to the people describing them, so here is who says what.
Quantower's documentation lists three uses. It presents a crossover as a trigger a trader may act on, long or short. It describes a cross of the zero line as a trend-reversing signal, useful in its account for identifying a correction or the start of a new trend. It describes a divergence — a new price extreme that the MACD does not confirm — as something that can be read as weakness in the existing trend.
Sierra Chart's MACD Crossover System turns the first of those into drawn output. It places an up arrow on the bar where the MACD line crosses the signal line from below and a down arrow on the cross from above, and the arrow's distance from the bar is itself an input, set as a percentage of the bar's range. The page labels the arrows buy and sell signals and sets out the conditions and the drawing rule; it says nothing about what price does afterwards.
Treated as arithmetic, a crossover is the same event as any moving average crossover, one level up: the thing doing the crossing is already a difference of two averages, and the thing being crossed is an average of that difference.
Does the MACD work?
One formal test of MACD rules was located for this post, and it is a currency study. Coakley, Marzano and Nankervis, in the International Review of Financial Analysis in 2016, tested a universe of 113,148 trading rules across 22 dollar exchange rates over the stretch from October 1997 through June 2015. Of that universe, 294 rules traded the signal-line cross and a further 294 traded the zero-line cross.
Before any correction for data snooping, the two MACD classes had the highest proportion of significant rules of any class in the study. Averaged across currencies, 62% of the signal-line rules and 55% of the zero-line rules were significant on the authors' excess-return measure, against 49% for RSI rules and 42% for Bollinger Band rules.
Then the authors applied a stepwise test for data snooping, which asks how many rules survive once you account for having tried thousands. The survivors thin out sharply, and the MACD's share of them is modest.
| Rule class (that study) | Significant before correction | Best rule's annualised excess return, interest differentials included | Rules robust after correction, per currency |
|---|---|---|---|
| MACD, signal-line cross | 62% | 6.71% | 27 of 294 on average |
| MACD, zero-line cross | 55% | 6.03% | 52 of 294, and none at all for several currencies |
| RSI | 49% | 10.79% | 9.1% of rules, as reported by the authors |
| Bollinger Bands | 42% | 20.64% | 0.8% of rules, as reported by the authors |
The averages hide how concentrated the result is. For the signal-line MACD, exactly seven rules out of 294 survived in 15 of the 22 currencies — about 2% — and the average of 27 is carried by the Indian rupee at 219 and the South African rand at 181. For the zero-line variant the authors report it as the exception among the newer indicators: its snooping-adjusted p-value stays above 0.05 for currencies including the Australian and Canadian dollars, the Norwegian krone and sterling.
What do the 2016 study's MACD figures leave out?
Three things, all of them stated in the paper. Coakley, Marzano and Nankervis exclude transaction costs by design: they say in their excess-return section that they do not use them, on the grounds that quoted spreads in their data overstate the cost of large FX trades. A futures trader pays commissions and crosses a real spread on every entry and exit.
The returns in the table are also the best rule in each currency, picked after the results were known. A 6.71% average means the single best of 294 MACD rules, selected with hindsight, averaged that much. It is not the return of a MACD rule you could have chosen in 1997, and the authors' whole data-snooping exercise exists because that distinction changes the answer.
The third limit is scope, and it is the one that matters most to anyone reading this on a futures site. The study is spot and forward currencies, with no futures contract in its sample. Two Park and Irwin reports were read alongside it: their 2004 review, which surveys 92 studies published between 1988 and 2004 across stock, currency and futures markets and tabulates the trading systems each one used, and their 2005 data-snooping-free test of trading rules in 12 US futures markets. The term MACD appears in neither report, so neither extends the currency result to an index or a commodity contract. A rule tested on currencies is evidence about currencies.
Do different MACD settings perform better?
The 2016 currency study does not answer that question, and the reason sits in its parameter grid. Its MACD grid varies three things: a delay filter of up to five days, a holding period of up to 20 months, and a fixed band multiplier. Six delays, seven holding periods and seven band values make the 294 rules in each MACD class.
The three averaging lengths are not in that grid. The authors fix the fast, slow and signal averages at 12, 26 and 9 days in their description of the indicator and vary everything else, so their results say nothing about whether a different set of lengths would have done better or worse.
That leaves the period question open to optimisation, with nothing in the studies read here comparing one set of MACD lengths against another. Searching a grid of lengths on your own data and keeping the best one is exactly the procedure the stepwise correction in that paper was built to discipline, and the gap between the pre-correction 62% and the handful of surviving rules is a measure of how much a grid search can flatter itself. Our post on curve fitting covers what that does to a backtest.
How much data does the MACD need, and what are its units?
At 12, 26 and 9 periods the signal line and the difference series need 35 bars before they exist, by the condition Sierra Chart states, which is the longer of the two average lengths plus the signal length: 26 plus 9. On one-minute bars that is 35 minutes; on five-minute bars it is 2 hours 55 minutes; on 15-minute bars, 8 hours 45 minutes.
The units matter too. The MACD line is a difference between two price averages, so it is denominated in the market's own price units and has no fixed range. A reading of 3.0 means three of whatever unit the market happens to quote in, so there is no level at which the line is objectively high.
| MACD | RSI | |
|---|---|---|
| Computed from | Difference of two price averages | Ratio of up closes to down closes |
| Range | Unbounded, in price units | 0 to 100 |
| Fixed reference levels | Zero only | 60 and 40, in the 2016 study's description of it |
Because the RSI sits on a fixed scale, the same RSI reading means the same thing on any instrument. A MACD reading does not, because the unit it is denominated in changes with the market.
What does the MACD mean on a funded account?
Nothing by itself, and that is the thing to be precise about on an evaluation account or a funded simulated one. An indicator produces a reading. On our accounts the drawdown line is the one performance rule that can end an account, and no study on your chart has any awareness of where that line sits.
The arithmetic that governs a funded account runs the other way round from a signal. Your maximum loss is set by the gap between your balance and the drawdown line, which on our accounts sits below the highest daily close your account has recorded, by a fixed amount, and is checked continuously while the session runs. The drawdown rules give the amounts and the lock point. Divide that gap by the dollar risk of the trade you are considering and you have the number of attempts you can afford, which is the calculation our post on position sizing works through.
A crossover tells you none of that, and because Sierra Chart's formulas are indexed by bar rather than by clock, the same three settings on a one-minute chart and on a 15-minute chart are two different calculations. The honest summary of the evidence read here: MACD rules were significant far more often than they survived a correction for having been chosen out of thousands, in a market that is not futures, before costs.
FAQ
What are the standard MACD settings? Quantower's documentation lists 12 for the fast exponential moving average, 26 for the slow one and 9 for the signal line, calling those the common periods. Sierra Chart publishes no default numbers on its MACD page: it exposes the three lengths as inputs along with an averaging type that can be switched from exponential to one of six alternatives. Quantower's page labels those lengths as common ones, and neither source read here presents them as a tested optimum.
What does it mean when the MACD crosses the zero line? Mechanically it means the fast and slow moving averages are equal at that bar, which is how Coakley, Marzano and Nankervis describe the event in their 2016 paper. Quantower's documentation calls a zero-line cross a trend-reversing signal and lists identifying a correction or the beginning of a new trend among its uses. In that paper's tests, the zero-line variant was the one newer indicator class that failed to show robust rules for several currencies once data snooping was accounted for.
Has the MACD been tested on futures? Not in the studies read for this post, which is not the same as saying no such test exists. The one formal test located, Coakley, Marzano and Nankervis (2016), covers 22 currencies against the US dollar, traded spot and one-day forward, and holds no futures contract. Park and Irwin's 2004 review of the technical-analysis literature and their 2005 data-snooping-free test of US futures markets do not mention the MACD anywhere in their text. Treat results from currency markets as evidence about currency markets.
Is the MACD better than the RSI? In the 2016 currency study, MACD signal-line rules were significant more often before correction — 62% against 49% — while the best RSI rule in each currency returned more, averaging 10.79% a year against 6.71%, and the authors report 9.1% of RSI rules surviving their data-snooping correction. That is one study, one asset class, and no transaction costs, which is not enough to rank two indicators for any other market.
Sources
- Sierra Chart — Technical Studies Reference: MACD
- Sierra Chart — Technical Studies Reference: MACD Crossover System
- Sierra Chart — Technical Studies Reference (contents)
- Quantower Help — MACD (Moving Average Convergence/Divergence)
- Coakley, Marzano & Nankervis — How profitable are FX technical trading rules? (accepted manuscript, International Review of Financial Analysis, 2016; University of Essex Research Repository)
- Park & Irwin — The Profitability of Technical Analysis: A Review (AgMAS Project Research Report 2004-04, University of Illinois)
- Park & Irwin — The Profitability of Technical Trading Rules in US Futures Markets: A Data Snooping Free Test (AgMAS Project Research Report 2005-04, University of Illinois)
- Lumen Futures Help Center — Drawdown explained
Educational content about futures markets and simulated trading. Not investment advice, and not a solicitation to trade. Trading futures involves substantial risk of loss. Read the full risk disclosure.