Is the ADX gate actually worth it, or is it curve-fitting?

The gate is one fixed threshold applied identically across all markets, which is about as far from curve-fitting as a filter gets. What it demonstrably does is cut trade count sharply in ranges, where crossover systems lose most of their money. What it cannot do is improve the crosses it approves — in the V-reversal case it approves bad ones. It is a coarse filter with a real but bounded benefit, not a guarantee.

Why not just use the 200-day average as the trend filter instead?

A long moving average answers 'is price above its past?' — a directional question. ADX answers 'is price moving one-sidedly?' — a strength question, indifferent to direction. A market can be above its 200-day average and dead flat, or below it and trending furiously. The two filters disagree often enough that they are genuinely different systems, not substitutes.

Does ADX still work now that everyone can compute it?

The reading itself measures something real and unarbitrageable — how one-sided recent movement has been. What decades of popularity have thinned is the edge of the naive entry around round thresholds. The durable value has migrated toward the unglamorous uses: vetoing trades in chop and ranking markets by trendiness, neither of which crowds out.

ADX + DI Directional MovementTrend following7 min read

The gate opens late, and it does not close fast enough for a V-reversal

Every weakness of the system is a price paid for one of its strengths. The lag buys reliability, the gate buys selectivity, and both bills arrive at the same two moments: the start of a trend and the end of one.

Key takeaway

ADX is smoothed twice, so by the time it certifies a trend, a good part of the move has already happened — that is construction, not malfunction
In a fast reversal the DI lines cross repeatedly while ADX is still high from the old trend; the gate approves exactly the whipsaw it exists to prevent
A falling ADX from a high level means the trend is dying, and crosses taken there have the gate's approval and none of its meaning

The lag is structural: you pay the first leg of every trend for certainty about it

ADX is an average of an average — DX values, themselves built from 14-bar smoothed components, smoothed again over 14 bars. That double filtering is why an ADX reading is trustworthy, and it is also arithmetic that cannot be argued with: a market must trend for weeks before ADX admits it is trending.

By the time ADX clears 25, a good part of the move is already done. The system does not claim otherwise — it trades the middle of trends, and the entry toll is the whole first leg.

Shortening the period to make the gate open sooner buys earliness with the only currency available: reliability. A 7-period ADX certifies trends that evaporate, which reintroduces exactly the chop trades the gate existed to refuse. There is no setting that is both early and sure; there is only choosing which failure you prefer.

The V-reversal: where the filter approves the whipsaw

The gate's blind spot is a trend that ends violently and reverses at once. ADX, still elevated from the move that just died, keeps the gate open; the DI lines, reacting fast to the turn, cross — and the system buys a rally that is actually the first bounce of a new downtrend. In a churning top the lines can cross several times in a month, every cross gate-approved, every trade a loss.

This is the exact scenario the detail rules warn about: in a fast reversal both DI lines cross repeatedly and the filter does not save you. The filter was built against one enemy — the trendless range — and the V-turn is a different enemy wearing the range's opposite face: too much movement rather than too little.

The trailing stop is the real defence here. Three ATR from the high fires early in a violent break, usually before the first false cross has a chance to re-enter you.
Losses cluster. The whipsaw sequence produces two or three consecutive gate-approved losers in the same market — normal for the system, and survivable only at the sizing the previous module insists on.
Elevated ADX after a long trend deserves suspicion, not confidence. The highest readings often print near exhaustion, because ADX summarises the recent past at the exact moment the past stops predicting.

The falling-ADX trap: technically above 25, practically over

The level test has a second soft spot: it cannot distinguish an ADX at 28 on its way up from an ADX at 28 on its way down from 45. The first is a trend establishing itself; the second is a trend unwinding into a range. A DI cross taken in the second case has the gate's formal approval and almost none of its intended meaning.

Practitioners patch this by requiring a rising ADX, which genuinely filters the trap and just as genuinely delays every legitimate entry further. This implementation keeps the level test and accepts the trap as one of its costs — the defensible choice for a mechanical system, since 'rising' invites a second parameter and a second argument. Know the cost exists; do not discover it live.

The failure that is actually the system working: long stretches of nothing

In a broad low-volatility regime, ADX sits under 25 across most of a watchlist for months. The system's output over that whole stretch is: no trades. Nothing is broken — the gate is doing the one thing it was built to do — but a rule that mostly says no is psychologically expensive to obey, and this is where users quietly lower the threshold to 20, then to 'about 18 looks fine'.

Every downgrade of the threshold converts the system back toward the plain crossover it was designed not to be. The drought is not a malfunction to fix; it is the fee for being absent from chop.

The system's edge lives in the trades it declines. A quarter of enforced idleness produces no evidence of that — only the absence of losses you cannot see.

Common questions

Is the ADX gate actually worth it, or is it curve-fitting?: The gate is one fixed threshold applied identically across all markets, which is about as far from curve-fitting as a filter gets. What it demonstrably does is cut trade count sharply in ranges, where crossover systems lose most of their money. What it cannot do is improve the crosses it approves — in the V-reversal case it approves bad ones. It is a coarse filter with a real but bounded benefit, not a guarantee.
Why not just use the 200-day average as the trend filter instead?: A long moving average answers 'is price above its past?' — a directional question. ADX answers 'is price moving one-sidedly?' — a strength question, indifferent to direction. A market can be above its 200-day average and dead flat, or below it and trending furiously. The two filters disagree often enough that they are genuinely different systems, not substitutes.
Does ADX still work now that everyone can compute it?: The reading itself measures something real and unarbitrageable — how one-sided recent movement has been. What decades of popularity have thinned is the edge of the naive entry around round thresholds. The durable value has migrated toward the unglamorous uses: vetoing trades in chop and ranking markets by trendiness, neither of which crowds out.

These are documented methods described for study. Nothing here is investment advice, a recommendation, or a claim about future returns — every system on this page has losing periods, and the pages say where.

Reading about a system is not having one.

Plutux is where you write your own rules down, test them against real data, and keep the record your memory would otherwise rewrite. Join the waitlist for early access.