The Indicator Library, by What It Measures

Because it is the part that is usually missing, and it is the part that costs money. An indicator is a compression of price or volume history into one number, and every compression discards something; knowing what a given indicator cannot see is more useful than another paragraph on how to read a crossing. It also keeps the pages honest, an indicator with no stated failure modes is being sold rather than described.

Fewer than you would guess, and ideally ones that read different inputs. Three momentum oscillators tell you almost the same thing three times, because they are all built from the closing price; one momentum measure, one volume measure and one breadth measure disagree with each other usefully. The value of a second indicator lies entirely in it being able to contradict the first.

Sometimes, but not by searching. Defaults like RSI 14 or MACD 12/26/9 have no statistical justification — they are conventions — and their persistence is itself a reason to keep them, since a level many participants watch behaves differently from one nobody does. Changing a setting because you can state why is analysis; changing it until a signal appears is fitting the tool to the past.

The arithmetic does. The interpretation needs care with anything volume-based, because raw intraday volume is distorted by the open and the close, when a disproportionate share of the session trades in a few minutes. Comparing a mid-morning bar with a closing bar without adjusting for that shape produces signals that are artefacts of the clock. That distortion is the problem this section keeps coming back to.

RSI, RMI, Volume RSI, the stochastic oscillator and the high-low index are bounded. Their output is confined to a fixed scale, usually 0 to 100. MACD, the rate of change, momentum, PMO and the advance/decline line are unbounded. The distinction decides whether a level can be quoted at all: a bounded indicator has the same scale on every instrument, so 70 means something transferable, while an unbounded one carries the instrument’s own price level and volatility in its values, and any threshold has to be drawn from that instrument’s own history.

Because the page is not written yet, and pointing at a page that does not exist is worse than saying so. The section is still described in full (the entry, what the indicator measures, and where it sits in the grouping), and the link appears when the page does. An unlinked entry is therefore a description without a destination rather than a broken promise.

More than most people expect, and almost always in the smoothing. Wilder’s average weights each new value by 1/n while a conventional exponential average uses 2/(n+1), and indicators that specify one are routinely implemented with the other. Seeds differ too: whether the first value is a simple mean of the first block or a recursion started from a single change. Both choices produce lines that look correct and disagree in the third digit for dozens of bars, which is why the pages here state the convention they use rather than leaving it implied.

Beneath it, in its own panel, whenever the indicator is not denominated in price. Overlaying a 0–100 oscillator on a price axis requires a second vertical scale, and a chart with two vertical scales can be made to show almost any relationship by choosing where each scale starts. Two panels sharing one horizontal axis says the same thing without that freedom, which is the convention every figure on this site follows.

No, and taking the distinction seriously is what separates a reference from a sales page. Every measure in this library is a summary of what has already happened, expressed to make one property easier to see: how fast, how broad, how heavily traded. Statements about the future are laid on top of those summaries by analysts; they are additional claims, and each has to be tested on its own rather than inherited from the arithmetic underneath it.