MACD: Sweet anticipation?
Moving average convergence/divergence (MACD) is one of today’s most popular trend-following indicators in analytical software packages and online services, but research shows the indicator isn’t always so reliable, particularly in short-swing markets.
Developed by Gerald Appel, MACD, like many approaches, uses averages to smooth out fluctuations and reveal the underlying trend. MACD uses two moving averages. One, typically a 26-unit moving average (26 MA), defines the trend; another, typically a 12-unit moving average (12 MA), indicates a change in 26 MA by crossing over it.
The problem with regular moving averages is that they are slow. Notice on the 40-day cycle chart that the crossover comes after the price peak.
MACD attempts to compensate for this delay by anticipating crossovers. Rather than waiting for crossovers, MACD reacts when the averages begin to converge (or diverge).
The MACD accomplishes this by computing delta, the difference between the 26 MA and the 12 MA. When crossovers occur, delta crosses zero. MACD tries to anticipate delta’s crossing of zero by measuring when delta begins to trend toward zero.
MACD measures the trend of delta by yet another average, usually a nine-period moving average of delta. Delta signals its own trend by crossing its moving average.
Thus, MACD is a double exponential average crossover system that tries to anticipate crossovers by signaling when the distance between averages begins to converge.
In theory, MACD should outperform other moving average systems. However, the reverse appears to be true, according to several studies.
For instance, The Encyclopedia of Technical Market Indicators reports that MACD substantially underperforms a 40-week simple moving average crossover rule.
The Dow Jones Irwin Guide to Trading Systems notes that a five-year performance study shows MACD results inferior to Richard Donchian’s 5-and-20 moving average method and the moving average combination of 4, 9 and 18 days.
Flaws of anticipation
Other testing corroborates these findings. The “anticipation” seems to be counterproductive.
The chart of Standard & Poor’s 500 prices illustrates the problem with anticipating. Prices rarely move smoothly. The whole advantage of using moving averages is that they are slow and give signals when a market’s trend is well under way. Trend traders believe the best time to trade is well into the trend, with momentum. While anticipation occasionally picks a top, it more likely picks a false jiggle and gets whipsawed.
The 20-cycle and 40-cycle charts show a subtle aspect of backtesting systems that use exponential averages: Results may vary widely depending on how the averages are “initialized” — that is, what time spans are used.
Ironically, the MACD is a “low pass” filter. The technique filters out high-frequency bumps and allows low-frequency trends to pass through. The standard 12-26-9 MACD combination tends to grow ineffective at approximately 40 time units, although the shorter cycles do not pass through.
The MACD provides a good example of the counterintuitive nature of system design. The delta with a nine-period moving average, intended to overcome the lag between price peaks and signals, actually increases the phase lag for intermediate-term and short-term cycles.
This supports the findings that MACD’s long-term trading results underperform simple systems while MACD’s short-term trading grows progressively worse.
Proponents of MACD generally acknowledge these findings, then elaborate that, well, it’s an indicator, not a signal, and must be used with other indicators, secret sauce and good judgment. So far, there have not been enough of these trading systems incorporating MACD that are precise enough to test historically.
Facing such evidence, you must question the indicator’s popularity. It likely was built on these items:
* Its claim to anticipate is alluring. * It has been touted as an indicator rather than a signal. * MACD provides a parascientific license to generate plenty of action, excitement, anxiety and commissions.
MACD plays an important role in the market ecology, but it just doesn’t happen to be a tool that makes money when used by itself.
The industry has plenty of good systems. The catch is that they require considerable emotional aptitude. Many traders avoid dealing with their emotions by trying to anticipate, optimize and otherwise abandon their systems.
You cannot escape the fact that systems are, by nature, low-pass price filters. They do not predict — they either just dict (say what’s happening) or postdict (say what’s happened).
Working to anticipate the future can be a distraction from the important task of dealing with the present.