ADX measures the strength of a trend. It does not say whether that trend is up or down, and it does not turn a threshold into a trade.
That modest definition is useful. It makes ADX a possible filter for a rule that is already defined elsewhere. It also explains the usual mistake: treating a smoothed description of recent price movement as a self-contained forecast.
What the ADX indicator measures
The Average Directional Index, or ADX, is a smoothed measure of directional movement on a scale from zero to one hundred. Its companion lines, +DI and -DI, describe the positive and negative directional components; the ADX line is designed to describe their separation, not the direction of price (StockCharts: Average Directional Index, Charles Schwab: ADX and RSI).
Price bars -> +DM and -DM -> +DI and -DI -> DX -> smoothed ADX
+DI above -DI = positive directional movement dominates the input
-DI above +DI = negative directional movement dominates the input
ADX rising = the directional imbalance is strengthening
ADX falling = the directional imbalance is weakeningThe last two lines are easy to overread. A rising ADX can accompany either an advance or a decline; a falling ADX says that the measured directional imbalance is fading, not that price must reverse. Both the calculation guide and Schwab's explanation make the same division between strength and direction (StockCharts: ADX interpretation, Charles Schwab: what ADX measures).
Why ADX lags by design
ADX is built from price ranges and then smoothed more than once. That smoothing is intentional: it reduces the sensitivity of the line to each individual bar. It also means the reading is necessarily late relative to the raw move that produced it (StockCharts: ADX calculation and smoothing, Charles Schwab: ADX and trend context).
This is not an indictment. A filter is allowed to be late if its job is to reject conditions, rather than to catch the first tick of a move. The relevant question is empirical: after allowing for the delay, does the filter improve a precisely defined rule after costs?
The lookback and smoothing choices are therefore part of the method. Changing them changes when the indicator reacts. A chart that looks persuasive with one setting can be an accidental result of choosing that setting after seeing the period. Freeze it before looking at the test result.
Familiar ADX levels are conventions
Platform guides commonly describe low readings as weak or absent trend conditions and higher readings as stronger trend conditions. They do not publish a universal threshold that works across instruments, timeframes, or trading costs: StockCharts discusses a range rather than a single dividing line, while Schwab uses its own practical ranges (StockCharts: measuring trend strength, Charles Schwab: ADX ranges).
That disagreement is the useful lesson. A threshold is a parameter, not a property of markets. If a rule only works with one narrowly selected threshold, one symbol, and one timeframe, the result may be fitting the sample rather than describing an edge.
The DI crossover has the same limitation. It is a reproducible condition, but a reproducible condition is not automatically a profitable one. A trader still needs to specify the instrument, timeframe, entry timing, exit, protective stop, position-sizing method, and realistic execution assumptions.
What ADX cannot settle
ADX has no information that is independent of the price history used to calculate it. It cannot identify the reason for a move, prove that a trend will continue, or rescue a rule with poor execution.
It can also be visually persuasive in hindsight. Once a large move has occurred, the later rise in ADX looks like confirmation that the move was real. The testable question is different: was the ADX value available before the decision, and did a pre-written rule outperform a relevant alternative on data it never saw during design?
This is where ATR is a helpful comparison. ATR describes the size of recent movement; ADX describes directional strength derived from related range calculations. Neither selects direction on its own. A combined chart is still a hypothesis until its rules are frozen.
How you'd actually test it
Start with a claim small enough to fail. For example: “A pre-defined breakout rule has a different outcome when ADX is above a pre-defined threshold than when it is not.” Do not make the threshold, lookback, or breakout definition adjustable after the test begins.
| Part of the test | Specify before running it |
|---|---|
| Market and data | Instrument, feed, timeframe, timezone, and date range |
| Signal | The exact breakout or directional rule, including when the bar is considered complete |
| ADX filter | Calculation setting, threshold, and whether +DI / -DI are used |
| Risk and exits | Stop logic, exit logic, sizing rule, and treatment of overlapping signals |
| Execution | Spread, commission, slippage, and any unavailable trading hours |
| Evaluation | In-sample design period, untouched out-of-sample period, and comparison rule without the filter |
Compare the base rule with the identical rule plus the frozen ADX filter. Report the trade count and the full distribution of results, not only the better curve. Then test whether the result remains plausible across different market regimes and reasonable nearby settings. Out-of-sample testing explains why unseen data matters; parameter sensitivity explains why a single neat setting is not enough; real trading costs explains why friction belongs in the result.
realbacktesting is a trading-software studio for cTrader built around verifiable work: its published cBot figures are cTrader-native backtests that readers can reproduce in their own cTrader. The standard for an ADX rule is the same: define it before the outcome, include costs, and see whether it survives data it did not help select.
Frequently asked
Does ADX show trend direction?
No. ADX is intended to measure trend strength. +DI and -DI provide the directional components, while price and the surrounding rule still determine how a trader interprets direction.
Is there a best ADX setting?
No universal setting is established by the indicator itself. The setting and threshold are parameters that need to be fixed and tested for the instrument, timeframe, and rule being studied.
Can ADX predict a reversal?
No. A falling ADX describes weakening measured directional movement. It does not establish that price will reverse, and any reversal rule needs its own test.
The stubborn takeaway
ADX is a filter, not a verdict. If its setting or threshold changes after the chart is known, the indicator is describing your hindsight more than the market.