Indicator terms

Average directional index

Also called: ADX.

ADX measures the strength of a trend on a 0–100 scale without indicating its direction, commonly read as trending above 25 and directionless below 20.

ADX answers only one question — is there a trend — and deliberately says nothing about which way. That narrowness is what makes it useful as a gate rather than a signal: it tells you which family of strategy is appropriate, not what to do.

It is also slow. ADX is derived from smoothed directional movement, so it confirms a trend well after the trend began and stays elevated after it has ended. Treating a rising ADX as a reason to enter is usually late; treating a low ADX as a reason not to take trend trades is usually right.

Our engine uses it exactly that way. Trend strategies are gated behind ADX of at least 18, EMA-TREND required at least 25, and the mean-reversion and sweep strategies run regardless of the reading. That asymmetry is the point — a single number decides which half of the strategy set is even eligible to compete.

Average true range

Also called: ATR.

ATR is the average of the true range over a lookback period, measuring how far an instrument typically moves per bar regardless of direction.

True range is the largest of three values — the current bar's high-low, and the distance from each of those to the previous close — which makes it robust to gaps. Averaged, it gives a volatility figure in the instrument's own price units rather than a normalised index.

That unit choice makes ATR the most directly useful indicator for risk. A stop of one and a half ATR is a stop scaled to current conditions, so the same rule produces a wide stop in a volatile week and a tight one in a quiet week without any intervention.

On gold this is not optional. XAUUSD volatility varies by a factor of several between quiet Asian sessions and news-driven US ones, so a fixed-dollar stop is far too wide half the time and far too tight the other half. Sizing the stop to ATR and then sizing the position to the stop is the correct order — the lot size calculator handles the second half.

We compared ATR-scaled stops with fixed ones on gold in should a gold stop be fixed, or scaled to volatility?

Bollinger Bands

Bollinger Bands plot a moving average with bands set a number of standard deviations above and below it, expanding and contracting with recent volatility.

Because the bands are volatility-scaled, they adapt automatically to conditions rather than using a fixed distance. A touch of the upper band means price is statistically extended relative to its own recent behaviour, not relative to any absolute level.

The two common uses point in opposite directions. Mean-reversion traders fade band touches; breakout traders treat a band expansion after a squeeze as the start of a move. Both work in the right regime and both fail in the other, which is precisely the regime problem that makes range versus trend classification worth solving.

Our BB-REV strategy fired zero times in twenty days of live trading, having looked reasonable in replay. It was cut for redundancy as much as performance — RANGE-REV already covered the same edge, and duplicated coverage in a tournament means two strategies competing to take the same trade.

Donchian Channel

A Donchian Channel plots the highest high and lowest low over a lookback period, making the boundaries of the recent range explicit as lines on the chart.

It is the simplest possible channel — no averaging, no volatility scaling, just the extremes of the last N bars. A break of the upper line is by definition an N-period high, which makes it the canonical breakout trigger and the basis of the original Turtle trading rules.

Its simplicity is both the strength and the limit. There are no parameters to overfit beyond the lookback length, which makes results relatively trustworthy, but it also has no mechanism for distinguishing a genuine break from a liquidity sweep — every sweep registers as a break.

This is why our surviving breakout strategy uses session structure rather than a raw channel break. SESSION-ORB survived the cull because its edge comes from a time-of-day effect that persists, not from a lookback parameter fitted to one period — which is exactly the distinction walk-forward testing is designed to expose.

Exponential moving average

Also called: EMA.

An exponential moving average weights recent prices more heavily than older ones, so it responds to changes in direction faster than a simple moving average.

The weighting decays geometrically, which means every historical price still contributes something but recent bars dominate. In practice the EMA turns earlier than a simple average of the same length, at the cost of reacting to moves that turn out to be noise.

Popular periods — 20, 50, 200 — are popular because they are popular. There is no property of markets that privileges those numbers, and their apparent significance is partly self-fulfilling, since enough participants watch them to generate reactions.

That self-fulfilment is weaker than it used to be, and our results reflect it. Every EMA-based strategy we tested lost money on XAUUSD, including two that never fired at all because their conditions were too strict for live gold. We kept the full record public rather than quietly deleting them, because the negative result is more useful to a buyer than the positive ones.

Keltner Channel

A Keltner Channel plots bands a multiple of ATR above and below an exponential moving average, giving a volatility envelope based on true range rather than standard deviation.

The construction is close to Bollinger Bands with one substantive difference: the width comes from ATR rather than standard deviation. ATR incorporates gaps, so a Keltner Channel reacts to overnight moves that a standard-deviation band based on closes will miss.

The classic combined use is the squeeze — when Bollinger Bands contract inside the Keltner Channel, volatility is unusually compressed and an expansion is often argued to follow. It is a genuine observation about volatility clustering, though the direction of the eventual expansion is not predicted by it.

We tested this family as SQUEEZE-FADE and TIGHT-BOX-FADE. Both fired zero times in live trading because the conditions they require — genuinely dead tape on gold — effectively never arrive during our session window. That is a useful negative result about XAUUSD specifically rather than about the indicator.

MACD

Also called: Moving average convergence divergence.

MACD plots the difference between two exponential moving averages against a signal line, producing crossovers and a histogram intended to show momentum shifts.

The standard construction subtracts a 26-period EMA from a 12-period one and then plots a 9-period EMA of that difference as the signal line. The histogram is simply the gap between the two, which makes MACD an indicator built entirely from other indicators.

That construction is also its main weakness. Every layer adds lag, so a MACD crossover confirms a move that a plain moving average already confirmed, later. Divergence between MACD and price is the more interesting reading, though it is notoriously prone to appearing repeatedly before a trend actually ends.

Our engine does not use MACD. That is not a criticism of the indicator so much as a consequence of the timeframes we trade — on M1 to M15 gold, an indicator built from 26-period averages is describing conditions that have already changed by the time it updates.

Moving average

Also called: MA.

A moving average is the mean price over a fixed number of recent periods, recalculated each bar, used to smooth price into a single directional line.

Every moving average is a lagging measure by construction — it is an average of what has already happened, so it cannot lead price. The trade-off is fixed: a shorter period tracks price closely and produces many false turns, a longer one is smoother and slower.

The two common variants differ in weighting. A simple moving average weights every period equally, while an exponential one weights recent periods more heavily and therefore turns sooner. Neither is superior; they lag differently.

Moving-average crossovers are the most heavily tested idea in retail trading and the results in our own set are unambiguous. EMA-PULLBACK lost $391 across seven months of 2026 and was negative in five of them, EMA-TREND lost $61 at a 33% win rate, and EMA-BOUNCE lost $40 at 20%. All three are cut, and the pattern is consistent enough to be a warning about the whole family.

The best-known moving-average signal is tested in does the golden cross work?

Pivot points

Pivot points are support and resistance levels calculated arithmetically from the previous period's high, low and close, plotted in advance for the current session.

The central pivot is the average of the previous high, low and close, with support and resistance levels derived by simple formulae above and below it. The whole set is fixed before the session begins, which is unusual — most levels only become visible after price has already reacted to them.

Their usefulness comes mainly from convention. Enough participants plot the same standard pivots that reactions at those prices are partly self-fulfilling, which is a real effect even though the arithmetic behind it is arbitrary.

Our PIVOT-BOUNCE strategy fired zero times live and was cut as redundant with RANGE-REV. That redundancy is worth noting as a general point: several distinct-sounding strategies in our original set turned out to be identifying the same setups, and a tournament that dedupes by strategy type is the mechanism that exposed it.

Whether fixed price levels actually hold on gold is tested for round numbers in do round numbers work as gold support and resistance?

Relative strength index

Also called: RSI.

RSI is a bounded oscillator measuring the ratio of average gains to average losses over a lookback period, conventionally read as overbought above 70 and oversold below 30.

The calculation compares the size of recent up moves against recent down moves and normalises the result between 0 and 100. It is a momentum measure, not a valuation one — a high reading means price has been rising strongly, which is a different statement from "price is too high".

The conventional interpretation inverts in a trending market and this is where most RSI money is lost. A strong uptrend holds RSI above 70 for extended periods, so selling every overbought reading means selling into strength repeatedly. The indicator is not failing; it is being asked a question it does not answer.

Used as a filter rather than a signal it is more defensible: requiring RSI to be out of the extreme before taking a mean-reversion trade removes the worst entries. Notably, our best-performing mean-reversion strategy, VWAP-MEAN-REV, deliberately does not require an oscillator extreme — and a large part of why it works is that it actually fires when the RSI-gated versions do not.

How RSI signals actually performed on XAUUSD is measured in our study does RSI work on gold?

Stochastic oscillator

Also called: Stochastics, Stochastic indicator.

The stochastic oscillator measures where the current close sits within the high-low range of a lookback period, expressed as a percentage between 0 and 100.

A reading of 80 means price closed near the top of its recent range, 20 near the bottom. Unlike RSI it is a position measure rather than a momentum one, though the two are correlated enough that they usually agree.

Its distinguishing quirk is a tendency to pin at extremes during strong moves and stay there. Traders new to it interpret a pinned reading as an imminent reversal, when in practice it is a symptom of a trend doing exactly what a trend does.

We tested it combined with an EMA filter as STOCH-EMA and cut it in the 2026 review — no edge survived out-of-sample testing. It joins the broader pattern in our results: oscillator-based entries on XAUUSD did not produce an edge at any of the timeframes we scan, which is why none of the four strategies we ship is oscillator-driven.

We put the 80/20 rule to the test on gold in the stochastic oscillator on gold, tested.

VWAP

Also called: Volume weighted average price.

VWAP is the average price of an instrument weighted by volume over a defined session, widely used as a benchmark for whether current price is expensive or cheap.

Because it is volume-weighted, VWAP represents where most of the day's business was actually done rather than where price merely passed through. Institutional desks measure execution quality against it, which gives the level genuine mechanical significance rather than purely technical significance.

Session anchoring matters and is frequently done wrong. A VWAP anchored to a rolling window is a different measure from one anchored to the session open, and on a 24-hour instrument the choice of session start materially changes the line.

VWAP-MEAN-REV is our best backtested strategy at +$132 and the engine in mixed conditions. It fades price back toward VWAP without demanding an oscillator extreme, which is a large part of why it fires when the more heavily filtered strategies do not. It was also the worst performer in choppy conditions at −$72 — the same logic, in the wrong regime.