MACD, ADX and Trend Strength

MACD is a derivative with a lag, and the histogram is its slope

MACD is the 12-period EMA minus the 26-period EMA, the signal line is a 9-period EMA of MACD, and the histogram is the difference between the two. Work it once and the whole thing is transparent: $82.40 − $80.10 = $2.30, the signal is $1.85, so the histogram is +$0.45, having been −$0.04 the day before. What crossed was not price but one average over another — a faster lag crossing a slower lag — and the signal line then smoothed that crossing by another nine periods. So three lines of the same closes exist on one chart, each one a further restatement of the last, and none of them contains a fact that the price series does not already contain. That is not an argument against using it; it is an argument against mistaking it for evidence. Indicators that are functions of price are *compressions* of price: they delete the short-term noise to make the medium-term direction legible, and the price of that legibility is lag. MACD is useful exactly to the extent that the compression helps you see something the raw candles make hard to see — the tendency of a trend to maintain separation between a fast and a slow window. The histogram is the best part of the construction for that purpose, because it is the rate of change of the separation: a histogram expanding above zero says the trend is accelerating, one shrinking toward zero says it is decelerating, and that turn usually comes before the crossover itself. Reading the histogram rather than the crossover is reading the derivative rather than the level, and it is earlier without being predictive. The magnitude matters as much as the sign, and it is what separates reading from pattern-matching. $2.30 on an $84.20 stock is 2.73% of price — a modest separation of two averages. The identical MACD value on a $12 stock would be an enormous separation relative to its scale, and the same value on a $400 stock would be almost nothing. So the number has to be normalised, either as a share of price or against the instrument’s own history of readings, before it can be compared with anything. The same logic applies to the signal line: a 9-period average of a quantity that is itself an average of averages is a heavily smoothed object, and the crossover of a smoothed quantity with its own smoothed average is a lagged event twice over. The three lines, worked — MACD = 82.40 − 80.10: $2.30 ← · Histogram = MACD − signal = 2.30 − 1.85: +$0.45 (was −$0.04 yesterday: a $0.49 swing) · Scale check = 2.30 ÷ 84.20: 2.73% of price — modest, not extreme · What actually crossed: A 12-period average over a 26-period average — not price Reading the histogram rather than the crossover is the practical improvement: the histogram is the rate of change of the separation, so it decelerates before the two lines cross. It is earlier, and it is still a statement about the past.

ADX: the one measurement that is not the same data again

ADX answers a different question. It measures how directional the market has been, without saying in which direction: a strong downtrend and a strong uptrend both produce a high reading. It is built from directional movement — the amount by which the current range extends above the prior high, against the amount it extends below the prior low — averaged over a window and then smoothed, which is why the customary interpretation is bands rather than a line: above about 25 the market is trending, below about 20 it is ranging, and between them it is undecided. Today’s reading of 27.4 sits 2.4 points above the trending line, which is a mild trend rather than a powerful one. The value of ADX is precisely that it is not another function of the same closing prices. A crossover, an oscillator and a moving average all describe the recent path of price; ADX describes the *shape* of that path — whether the movement is concentrated in one direction or scattered around a level. That makes it the conditioning variable for everything else in this rung. A MACD crossover with ADX at 27 is a continuation signal in a market that has been going somewhere; the identical crossover with ADX at 14 is mechanically guaranteed to happen as the two lines wander around each other in a range, and it will be followed by another one in the opposite direction within weeks. It is the same lesson the moving average taught, arriving through a different measurement: the regime decides whether the signal has any content. Two honest caveats. First, ADX is a lagging average of a lagging average, so it tells you a trend is in force after it has been in force, which means it will read high near the end of extended moves and low at the start of new ones — an unavoidable consequence of measuring a property that only exists over a window. Second, the thresholds are conventions rather than laws, and they are not symmetric across instruments: a calm large-cap may never exceed 20 in a year in which it trends 30%, while a volatile small-cap can exceed 40 in a two-week range. As with every indicator in this subject, the reading that means something is the instrument’s own reading compared with its own history, and the absolute threshold is a starting convention rather than a rule of nature. • ADX measures how directional the market has been, not which way it went — high readings accompany both trends. • Customary bands: above about 25 trending, below about 20 ranging, in between undecided. • It is the conditioning variable for every other signal: a crossover in a range is generated mechanically. • It lags, so it confirms a trend late and reads low at the start of new ones. • Thresholds are conventions — compare the instrument with its own history rather than with a universal number. The classic misread is treating a rising ADX as a buy signal. It is a measure of strength, and a rising ADX with price falling is a strengthening downtrend — the single most dangerous combination to be long into. The direction comes from structure; ADX only says how much the market is committing to whatever direction that is.

Does the crossover beat doing nothing?

A crossover is an instruction to be in or out, and once it is stated that way it becomes a testable claim rather than a chart pattern. The claim is not “the crossover catches the big move” — every trend signal catches some of a trend. It is the comparative one: being exposed only when the signal says so produced a better result than simply holding the same instrument, over the same period, at the same cost. Testing it means writing four numbers down beside the buy-and-hold benchmark over an identical window: total return, maximum drawdown, the fraction of days actually in the market, and the number of round trips. The useful comparison is not raw return, because a system that sits out two-thirds of the time is not competing on return — it is competing on return *per unit of exposure*, and on the drawdown it avoided while it sat out. A crossover strategy that earns less than holding but spends a third of the time invested and halves the drawdown may be the better trade for a specific account, and may be worse for another. The numbers make that a decision instead of an argument. What the honest numbers usually show is a recognisable trade-off. The lag that is built into the construction means the signal enters after a move has started and exits after it has turned, so the largest single advance is partly missed. In exchange, the rule is on the sidelines for much of the grind that follows, and it is that absence — not clever entries — that the drawdown improvement comes from. It is worth being precise about this, because the popular story credits the indicator with catching trends when its measurable contribution is more often avoiding stretches of the market it describes as directionless. Then the costs, which are where many versions of the test end. Every crossover is a round trip, and a round trip pays the spread and any commission. A signal that fires forty times a year on a name with a ten-basis-point spread gives away several percent annually before anything else — which is frequently the entire measured edge. A test that omits costs is not a slightly optimistic test; it is a test of a strategy that does not exist. Parameter sensitivity has to be reported alongside the result. The twelve-twenty-six-nine convention is a habit inherited from the era of weekly chart paper, not an optimum, and moving the three lengths shifts every number in the comparison. If the conclusion survives across a band of settings, the indicator is telling you something about momentum; if it holds only at one set of values, it is telling you about that price history. The use that survives testing most reliably is not the standalone entry but the filter. Requiring a trade to agree with the relationship between the two averages, or requiring the histogram to be on the right side of zero, adds a condition rather than generating a signal — and a condition is cheaper to justify because it is refining something you were going to do anyway. It still has to be tested, and tested as a filter rather than admired in hindsight. The question is never whether the lines look convincing on the chart you remember. It is whether acting on them, with costs included, beat not acting at all. • Benchmark against holding the same instrument over the same window. • Report four numbers: return, drawdown, time in market, and round trips. • Most of the measured benefit comes from being absent, not from being early. • Include costs, and report the parameter band rather than one setting. Run the same test with the ADX reading split into trending and flat periods. If the crossover only adds value when the trend measure is high, the indicator is not a signal at all — it is a way of describing an environment, and the environment is what should be forecast.

Sources

• MACD: construction, signal line and histogramAppel, “Technical Analysis: Power Tools for Active Investors”
• Average Directional Index and directional movementWilder, “New Concepts in Technical Trading Systems”
• Why derived indicators restate rather than add informationStandard treatments of technical-indicator redundancy