MACD Indicator Guide: Crossovers, Histograms and Divergences

The Moving Average Convergence Divergence – MACD – was developed by Gerald Appel in the late 1970s and published in full in The Moving Average Convergence-Divergence Trading Method (1979, Signalert Corporation). Appel was a money manager and systems researcher who wanted a single indicator that combined trend-following (via moving averages) with momentum (via the difference between two averages of different speeds).

The core insight: if a fast moving average is rising faster than a slow moving average, the gap between them is widening. That widening gap is a measure of momentum. When the gap stops widening and begins to narrow, momentum is fading – even if price is still rising. MACD plots that gap as a line so you can see its direction at a glance.

A second derivative – the histogram – shows the difference between the MACD line and its 9-period EMA (the "signal line"). When the histogram grows taller above zero, momentum is accelerating upward. When it shrinks, momentum is decelerating. This makes MACD a three-level read on trend: direction (MACD line above/below zero), strength (MACD line slope), and change-of-strength (histogram size).

MACD line = EMA(12, close) − EMA(26, close).

Signal line = EMA(9, MACD line).

Histogram = MACD line − Signal line.

The three defaults – 12, 26, 9 – come from Appel's original work and have stuck as the near-universal standard. Technically, these correspond to roughly two weeks (12 days), one month (26 days), and two weeks (9 days of the MACD's own smoothing) on daily bars. Appel experimented with 19/39 and other pairs in different market conditions and noted that fast pairs produce more whipsaws while slow pairs lag more.

EMAs are used rather than SMAs because exponential smoothing weights recent data more heavily, making the indicator react faster to new trends. The formula EMA_t = α × price_t + (1−α) × EMA_{t−1} with α = 2/(N+1) gives the standard definitions.

A bullish signal-line crossover occurs when the MACD line crosses above its 9-EMA signal line. A bearish crossover is the opposite. These are the most commonly cited MACD signals in retail literature and also the most problematic: they fire constantly in sideways markets, producing whipsaws that bleed accounts through spread and commission costs even when individual losses are small.

Academic evidence is mixed. Wong, Manzur & Chew (2003, Applied Financial Economics) tested MACD crossovers on the Singapore STI and found modestly positive returns during trending periods and flat-to-negative returns during ranges. Chong & Ng (2008) tested MACD on the London FT30 from 1935–1994 and found raw crossover returns of around 1–2% above buy-and-hold before costs – a margin that transaction costs often consume entirely.

The practical fix is a trend filter: only take bullish crossovers above the zero line, only take bearish crossovers below it. This restriction roughly halves the number of trades but dramatically improves the win rate because you only trade with the larger trend. On FX majors with modern spreads, filtered crossovers typically show Sharpe 0.3–0.6 versus unfiltered Sharpe near zero.

The MACD line crosses zero when the 12-EMA crosses the 26-EMA. This is a slower but more reliable signal than a signal-line cross – it represents an actual moving-average crossover in the price domain, not just a change in the MACD line relative to its own smoothing.

Zero-line crosses typically lag by 5–15 bars depending on volatility, so they are used for confirmation rather than entry timing. A swing trader might take a signal-line cross as a tentative entry signal and add to the position once MACD clears zero, confirming the trend. A more conservative trader waits for the zero-line cross before taking any exposure.

The zero-line itself is a trend filter. Some traders require MACD to be above zero for long-only strategies and below zero for short-only strategies, ignoring the raw 12/26 EMA crossover price trigger and letting MACD's own state machine gate the trades.

The histogram = MACD − signal. When the histogram rises above zero, MACD is above its 9-EMA, meaning momentum is still positive. When the histogram is above zero and growing, momentum is accelerating. When the histogram is above zero and shrinking, momentum is decelerating – an early warning that a signal-line cross is coming.

Alexander Elder in Trading for a Living (1993) popularised the concept of trading the slope of the histogram rather than its absolute value. A "bullish" histogram was one where the most recent bar was higher than the previous bar, regardless of whether the histogram was above or below zero. This is an earlier, leading version of the signal-line cross signal at the cost of more false positives.

Histogram reversal combined with divergence is one of the stronger MACD patterns. If price makes a new high, MACD makes a new high, but the histogram fails to exceed its prior peak, momentum acceleration has stalled – a subtle early warning that the move is topping. Elder called this a "first swing" warning.

Price makes a higher high; MACD makes a lower high. This is regular bearish divergence and is the MACD signal with the strongest empirical support across markets. The intuition: price is reaching for new highs, but the momentum engine (the EMA gap) can no longer sustain that pace. The trend is running on reduced thrust.

Four divergence types exist, identical to those for RSI divergence: regular bullish/bearish (reversal) and hidden bullish/bearish (continuation). Regular divergences are more famous; hidden divergences are arguably more useful for trend-continuation traders.

Rigorous backtests of MACD divergence on liquid instruments (Lento 2007; Chong, Ng & Liew 2014 on Hong Kong equities) show win rates in the 52–58% range with decent reward skew (winners 1.5–2× average losers). These numbers assume strict pivot rules and filter for divergences occurring in overbought/oversold MACD zones. Eyeballed divergences – drawn after the fact on a chart – show dramatically higher win rates because survivorship bias selects the ones that worked.

The chart below summarises win-rate and Sharpe estimates across MACD signal types, averaged across published studies on FX majors and US large-cap equities from 2000–2020. Numbers are directional – treat them as a guide to which signals deserve further backtesting on your own data, not as tradable truth.

These Sharpe numbers are survivorship-laden and regime-dependent; they should not be used as tradable estimates. The takeaway is the ordering: the more filtering you add, the better the risk-adjusted outcome – at the cost of fewer trades.

A single-timeframe MACD is noisy. Combining MACD states across two or three timeframes is how professional trend-followers filter signals:

Trade with the higher-timeframe MACD and enter on the lower-timeframe MACD. If the daily MACD is above zero and rising, only take bullish 1-hour MACD signals (crosses, divergences). Opposite-direction signals on the lower timeframe are ignored.

Histogram alignment on entry timeframe. Enter only when the trading-timeframe histogram has reversed in your favour for at least two bars – avoids chasing signals into noise.

Exit when higher-timeframe momentum fades. Even if the lower timeframe is still bullish, a flattening daily MACD histogram is a reason to tighten stops. The larger trend that justifies holding is weakening.

This nested approach is what swing traders mean when they say "trade the pullback in the trend" – it operationalises the pullback via MACD states.

Trap 1 – Sideways whipsaws. In a range, MACD oscillates around zero and generates 10+ crossovers per month, most of which fail. If your market is ranging (ADX < 20 / narrow daily range), either stand down or switch to range-based tools.

Trap 2 – Overfit parameters. Changing 12/26/9 to 11/27/8 to "optimise" on historical data is classic curve-fitting. Any change must survive out-of-sample testing before you trade it live.

Trap 3 – Divergence without confirmation. MACD divergences can persist for dozens of bars before price reacts, or never play out ("failed divergence"). Require a break of a recent swing or a bearish candle pattern at the second peak before entering.

Trap 4 – Using MACD alone. MACD is a single-factor indicator (momentum). Volume, volatility, and structure are independent factors. Combining MACD with volume analysis catches signals that pure-momentum systems miss.

Trap 5 – News shocks. NFP and FOMC releases can flip MACD instantly; crossovers during news windows are usually driven by volatility, not trend. Either exit before the release or filter crossovers that occur within 60 minutes of a tier-1 event.

Trap 6 – Ignoring regime change. A MACD that worked beautifully for 18 months in a trending market can produce a year of losses in a range. Monitor rolling performance and accept that all indicators have regime-dependence.

Prop firms do not care which indicator you used – they care about your R-distribution and drawdown profile. But MACD fits naturally into a prop-firm evaluation workflow because it gives clean, binary state changes that are easy to log and review:

• Trade filter: "MACD above zero on daily" → long setups only.
• Setup flag: "1-hour MACD cross in direction of filter with histogram > 0.0005" → takeable setup.
• Journal entry: At trade close, log MACD state at entry and exit. Review weekly for patterns – maybe all your losers happen when the histogram peaked 3+ bars before entry.
• Drawdown guardrail: If three consecutive trades on the same side have lost while MACD was in the "valid" state, stand down for the day. The regime may have shifted and your filter is lying.

For evaluation phases where consistency matters more than peak returns, filtered MACD systems tend to produce the kind of steady, slow R-accumulation that firms reward.