What ADX actually is
ADX is the third line of a three-line system. The other two carry the direction; ADX carries only the strength.
+DI (plus directional indicator): upward movement, averaged and expressed per unit of true range
-DI (minus directional indicator): downward movement, same treatment
ADX: a smoothed measure of how one-sided those two are, from 0 to 100
The reference is blunt about the division of labour: ADX "is used to measure the strength or weakness of a trend, not the actual direction", and "does not indicate trend direction or momentum, only trend strength." A market crashing hard prints a high ADX just as readily as one rallying hard.
Directional movement by hand
The building block is directional movement (DM), and each bar casts at most one vote. Compare today's bar to yesterday's: the up-move is today's high minus yesterday's high, the down-move is yesterday's low minus today's low. Whichever is larger, and positive, wins the bar. Three bars cover every case:
The two edge cases are where the design shows its logic. An outside bar that stretched both directions votes only for the side that moved more. An inside bar, which broke nothing, votes for nobody. Every bar is reduced to a single question: which boundary did the market push hardest, if any?
From votes to the three lines
The chain from votes to finished indicator is four steps of Wilder arithmetic, using the smoothing you met in Parts 15 and 19.
Smooth the +DMs, the -DMs and the True Ranges over 14 periods.
+DI = smoothed +DM divided by smoothed TR, times 100, and likewise for -DI. Dividing by true range is the clever part: it expresses directional push per unit of volatility, so a wild market does not look trendy merely for being wild.
DX = the absolute difference between the DIs, divided by their sum, times 100. This is the one-sidedness score, and a quick example shows what it rewards. If +DI is 30 and -DI is 10, DX = 20/40 x 100 = 50. If +DI is 22 and -DI is 18, DX = 4/40 x 100 = 10. Similar amounts of total movement; completely different scores. DX is high only when the votes pile up on one side.
ADX = a 14-period smoothed average of DX. First value is a simple average, then (prior ADX x 13 + current DX) / 14, exactly the ATR pattern.
One property falls out of the construction: a bar's directional movement can never exceed its true range, so +DI and -DI sum to 100 or less, running high in strong trends and sagging in ranges.
So the finished number reads: how one-sided has directional movement been, per unit of volatility, averaged twice. The double averaging is why the line is smooth, and why it is slow.
Reading the dial
ADX is bounded 0 to 100, but in practice it lives low. The references do not fully agree on the bands, which is worth knowing before you memorise anything: the ChartSchool reference reports that "Wilder suggests that a strong trend is present when ADX is above 25 and no trend is present when ADX is below 20," with a gray zone between and many analysts simply using 20, while the other reference bands it higher, calling readings above 40 strong and above 50 exceptional, and a third stretches the top further still, reserving 75 to 100 for rare, possibly overextended trends. The disagreement is the lesson: these are conventions describing a continuous dial, not tested constants.
THE DIAL IN PRACTICE
below 20 no trend: crossover and breakout tools are
in their whipsaw zone
20 - 25 gray zone: conventions disagree here
above 25 trending: trend-following tools have the
conditions they assume
rising ADX the trend, whichever way it points, is
getting more one-sided
falling ADX the trend is losing one-sidedness.
NOT a reversal signal: direction lives
in the DI lines, never in ADXThe reading that burns most newcomers is a falling ADX during a rally. It does not mean price will fall; it means the one-sidedness is fading, and a lower value on a strength dial is less strength, nothing more. Direction was never in this line.
The lag, honestly
ADX inherits every delay this series has catalogued, twice over. The votes are smoothed, the ratio of smoothed values is taken, and that ratio is smoothed again. The reference states the consequence plainly: ADX "is a lagging indicator; that is, a trend must have established itself before the ADX will generate a signal that a trend is under way."
There is also a warm-up cost: because of Wilder's smoothing, the reference notes "it can take around 150 periods of data to get true ADX values." On a fresh chart with a short history, the ADX you see is not yet the ADX the formula defines.
Neither is a flaw to engineer away: a strength reading that reacted to every bar would just be the noise it exists to filter. But it does mean ADX confirms trends rather than catching their starts; expect to arrive after the turn, on purpose.
What the evidence says, briefly
ADX itself is like ATR: a measurement rather than a forecast, so there is no win rate to test on the strength reading alone. The claims built on top of it are a different matter. Wilder did attach a signal: buy when +DI crosses above -DI with ADX confirming strength, sell on the reverse, the entry bar's extreme keeping the signal in force. That is a directional, testable rule, and no tested result for it appears in the evidence base this series has used; the 2023 crypto study's four families did not include it. The same goes for the appealing idea that filtering a crossover system by ADX would have rescued Part 14's whipsaw losses: it is a hypothesis a backtest could check, not a result this series can cite. Backtest write-ups claiming exactly that do circulate, but they trace to trading-platform blogs rather than peer-reviewed tests, the same provenance problem Part 14 found under the golden-cross statistics. What stands on its own is narrower: the dial measures one-sidedness per unit of volatility, and checking it before deploying a trend-following tool is a question of fit, not forecasting.
Reputation versus evidence
Using ADX honestly
Ask it one question: is this market trending? That is the question Part 14's whipsawed traders never asked, and the only one this line answers.
Get direction elsewhere. From the DI lines if you must, but better from the structure, levels and volume this series built in Parts 4 to 8. ADX itself points nowhere.
Treat the bands as furniture, not physics. Below 20 is quiet, above 25 is trending, and the zone between is exactly as ambiguous as the disagreeing references suggest.
Expect it to be late, by design. Use it to characterise the market you are in, not to catch the turn into a new one.
Give it history. On a freshly loaded chart, the first stretch of ADX values has not converged yet; the line needs on the order of 150 periods before it means what the formula says.
Common mistakes
Reading ADX as bullish or bearish - it is unsigned; the same 35 can be a strong rally or a strong collapse.
Selling because ADX turned down - that is a strength reading fading, not a direction changing.
Trading the DI crossover mechanically - it is a directional signal with no tested result behind it in this series' evidence base.
Trusting ADX on a short chart - the smoothing needs far more history than 14 bars to settle.
Using it as a standalone system - it was built as a filter for other tools, and it measures exactly one thing.
ADX on ApeX Omni
On a perpetuals venue, ADX slots in one step before every tactic this series has covered. The workflow on ApeX Omni: read the ADX on your trading timeframe before choosing the tactic, not after. Below 20, the market is in the regime where Part 14 showed crossover systems bleeding six small losses at a time, so trend-following entries are fighting their own assumptions; range behaviour around the levels of Part 4 fits better. Above 25 and rising, trend-following tools have the conditions they assume, and pullback entries toward structure make more sense than fading strength, which Part 15 already warned against. In both regimes the sizing discipline is unchanged: measure the stop with Part 19's ATR, size off the stop, and let leverage fall out last, because ADX describes the market's character while ATR prices its movement: what game is this market playing, and what does a seat cost?
The bottom line
ADX is a one-question indicator, and the question is a good one. Most trend-following losses do not come from bad entries; they come from running a trending-market tactic in a market that is not trending; until you measure that, you are guessing. ADX measures it: one vote per bar, normalised by volatility, scored for one-sidedness, smoothed until only the character remains. The price of that smoothness is lag, the number carries no direction, and the entry signal built on it remains untested in the evidence this series relies on. Use it the way its own arithmetic suggests: not as a signal, but as the answer to whether your actual signals are standing in the right market.
Next in this series: VWAP, and a genuine first: after seven indicator articles computed from price alone, the first one that brings a second source of information to the chart.
Frequently asked questions
What does ADX measure? The strength of a trend on a 0-to-100 scale, regardless of its direction. It is built by scoring each bar's directional movement, normalising by true range, measuring how one-sided the result is, and smoothing over 14 periods.
Who created ADX? J. Welles Wilder, in his 1978 book New Concepts in Technical Trading Systems, alongside RSI and ATR. The directional movement system includes +DI and -DI, which carry direction, and ADX, which carries only strength.
What is a good ADX value for a trend? The conventions disagree: Wilder is reported as suggesting a strong trend above 25 and no trend below 20, many analysts use 20 as the key level, and some references call readings above 40 strong. Treat the bands as rough furniture on a continuous dial.
Does a falling ADX mean the trend is over? It means the trend is losing one-sidedness. It says nothing about direction, and price can keep rising or falling while ADX declines. Direction is read from the DI lines or from price structure, not from ADX.
Is the DI crossover a reliable buy signal? It is Wilder's designed entry, buying when +DI crosses above -DI with ADX confirming strength, but no tested result for it appears in the studies this series has drawn on. Treat it as untested rather than proven.
Why does ADX need so much data? Because of the layered Wilder smoothing, the reference notes it can take around 150 periods for the calculation to converge to true values. On a short chart history, the plotted line has not settled yet.
Explore more from this series: Part 13: Do Indicators Actually Work? | Part 14: Moving Averages | Part 19: ATR Explained | Part 21: VWAP Explained (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.