ADX / Trend Strength

The Average Directional Index (ADX) was published by J. Welles Wilder Jr. in 1978 as part of the Directional Movement System, alongside the two directional indicators +DI and −DI. Its purpose is narrow and worth stating plainly: ADX measures how strongly a market is trending without saying which way. A reading of 40 describes decisive movement, not whether that movement is up or down. Direction, when a rule set needs it, must come from the DI pair or from somewhere else entirely.

That property explains why the 15 decoded versions linked from this page use the same number in opposite ways: some require a high reading before taking a trend signal, others require a low one because their edge lives in ranges, and a few fade high readings as stretched moves. All three are defensible, which is why "does ADX work" cannot be answered without the rest of the strategy.

## How the calculation works

Each bar produces directional movement: the portion of the current range extending beyond the previous bar's, upward (+DM) or downward (−DM), counting only the larger of the two. Both series are smoothed with Wilder's averaging over a lookback — 14 by default — and divided by smoothed true range, giving +DI and −DI as percentages comparable across instruments and timeframes. The directional index is then DX = 100 × |+DI − −DI| / (+DI + −DI), a measure of how lopsided they are, and ADX is a second Wilder average of DX.

Two structural facts follow. The absolute value in DX discards the sign, so direction is destroyed at that step and cannot be recovered afterwards. And the series is smoothed twice, so it is slow: a 14-period ADX needs roughly 27 bars before its first valid value, and confirms a trend well after that trend began. Wilder's bands — under 20–25 as no meaningful trend, above 25 as trending, above 40 as strong — came from daily commodity charts and are conventions, not constants.

## Main variants

- **Strength gate.** ADX above a threshold as a permission filter on an entry generated elsewhere — a moving-average cross, breakout, pullback or trendline break. The largest family. - **Range permission.** The inverse condition: a mean-reversion or fade rule trades only while ADX is low. - **Fade the strong reading.** A high or sharply rising ADX read as exhaustion rather than opportunity, usually paired with a short-term overextension condition. - **DI-based systems.** +DI crossing −DI supplies direction while ADX supplies strength, so the indicator generates the trade rather than filtering it. - **Slope rather than level.** Rising versus falling ADX, or ADX against its own moving average, instead of a fixed threshold. - **Substitute strength measures.** The Choppiness Index, Kaufman's Efficiency Ratio and similar ratios of net movement to total movement fill the same slot; several versions swap or combine them. - **Regime and exposure selection.** The reading used to choose which strategy runs, what size it takes, or which options structure fits — including multi-timeframe versions that read strength high and trigger low.

## What typically differentiates implementations

Threshold and lookback are the visible parameters, but the choices around them matter more: level or slope; entries only or exits too; where direction comes from and whether both sides are traded; confirmed close or intrabar reading. Smoothing also differs by platform: replacing Wilder's recursive average with an exponential one draws a different line from identical settings. And above all, what the version is compared against: a filter modifies an existing strategy, so its value is a difference, not a standalone result.

## Common mistakes

Reading ADX as direction: it rises in strong downtrends exactly as in strong uptrends. Treating 25 as a universal boundary rather than a daily-commodity convention to re-check per instrument and timeframe. Conflating level and slope: a high but falling ADX describes a strong trend losing force, which is not a low reading. Ignoring the double lag, then being surprised that confirmation arrives late — that lag is what the fade family is built on. Stacking ADX with Choppiness, Efficiency Ratio and similar measures and calling their agreement confirmation, when they largely measure the same quantity. Trading DI crossovers alone in quiet markets, where the two lines braid around each other. And adding a filter, reading an improved average, and not noticing the trade count collapsed — fewer trades means noisier statistics, and the improvement may rest on a handful of removed outcomes.

## How to evaluate and backtest a version

Start with the baseline: run the underlying strategy without the ADX condition on the same data, costs and execution rules, then add the condition and report both. What matters is the trade count before and after, the removed trades as a distribution rather than an average, and whether any improvement is spread across years or concentrated in one.

Then test the condition, not a setting. Sweep the threshold with the lookback alongside it, and prefer a broad plateau to an isolated peak. Check the filter in both directions: if requiring ADX above the threshold and requiring it below both improve results, the improvement is coming from the sample. Confirm the smoothing convention, and align the warm-up so no trade is taken on an unformed value. Then segment by year and by volatility regime, examine long and short separately, and validate out of sample or with walk-forward on more than one instrument.

Strategies in this concept (58)

Frequently asked questions

What does ADX actually measure?

Trend strength, and nothing else. The calculation takes the absolute difference between +DI and −DI, which throws away the sign, so a rising ADX means the market is moving decisively in some direction — up or down, the number cannot tell you. Direction has to come from the DI lines, from price, or from whatever rule the version uses to trigger. Most confusion around the indicator comes from treating a high reading as bullish; in a sustained decline the reading is high for the same reason.

What ADX level counts as a trend?

Wilder's conventional bands — roughly 20–25 as the boundary, 40 and above as strong — came from daily commodity charts and travel poorly. Because the DI values are normalised by true range, the scale is comparable across instruments, but the typical distribution of readings is not: intraday series and quiet FX pairs spend their time in a different part of the range than trending indices. Treat the threshold as a parameter to be examined across a range of values on the data you actually trade, not as a constant.

Is ADX a lagging indicator?

Yes, and more so than most. It is smoothed twice — once when the directional movement series are averaged, again when DX is averaged into ADX — so a 14-period setting needs roughly 27 bars before it produces a valid value and confirms a trend well after that trend is underway. This is not a defect to be corrected by shortening the lookback, since a short lookback simply produces a noisier reading. It is the reason some catalogued versions require a high reading to join a move while others read the same value as a move already extended.

Should I use the ADX level or its slope?

They describe different conditions and should not be substituted for one another. A level condition asks whether a trend exists; a slope condition asks whether it is gaining or losing force. A reading of 40 that is falling and a reading of 20 that is rising can point to opposite situations, and a rule written only in terms of levels treats them as the same. If a version uses one, it is worth testing the other on the same data — they will not select the same trades.

Do +DI / −DI crossovers work as a standalone entry?

They give the system a direction, which ADX alone cannot, but crossovers are frequent and unstable when the market is quiet: the two lines braid around each other and generate signals with no follow-through. That is why the classic formulation pairs the crossover with an ADX condition, and why most catalogued versions add a trigger or a filter of some other kind. Tested on its own, a DI crossover rule is best treated as the baseline a version has to beat, not as the finished strategy.

Is ADX better than the Choppiness Index or the Efficiency Ratio?

They occupy the same slot in a rule set — all measure how much net movement a market produced relative to the movement it travelled — so they tend to agree, and running two or three together mainly adds parameters rather than information. The practical differences are responsiveness and how the reading is bounded, which affects where a threshold sits and how often the filter switches on. The way to choose is to swap one for another inside an otherwise identical strategy and compare, rather than to stack them and call the agreement confirmation.

All strategy concepts · Explore all strategies