What the stochastic oscillator is

The question the calculation asks 

Most of the indicators treated so far in this module read closes and do something to the level or the change. This one reads three numbers and does something to their geometry. It takes the highest high reached over a stated number of completed bars, the lowest low reached over the same bars, and the latest close, and it reports the position of that close between the two extremes as a percentage of the distance between them.

The distance from the low to the close is divided by the distance from the low to the high, and the result is multiplied by one hundred. Nothing else enters the calculation. Not the opens, not the order in which the bars arrived, not how far the range travelled in price terms, not how long the instrument spent anywhere inside it. The output is a statement about one bar's close relative to a band, and the band is redrawn from scratch on every new bar.

Two consequences follow immediately from that division. The reading cannot leave the range from zero to one hundred, because the close of a bar inside the window cannot be higher than the highest high of the window or lower than its lowest low. And the reading is dimensionless: a band ten points tall and a band a thousand points tall both produce a number on the same scale, so the height of the band is discarded by the arithmetic that produces the percentage.

The window is built from highs and lows 

The band is drawn from the extremes of the bars, not from their closes, and that choice separates this calculation from the averages and the change measures already covered. A wick that reached far beyond every close in the window still sets the edge of the band, and it goes on setting it for as many bars as the lookback holds it. One unusually wide bar therefore changes every reading that follows it, until it falls out of the window and the band snaps back to whatever the remaining bars describe.

That is worth holding onto, because it is a source of movement in the plotted line that has nothing to do with the latest bar. The reading can drop sharply on a bar whose own close barely moved, purely because an old extreme aged out of the window and the denominator changed underneath it. Traditions that describe the line's slope as momentum are describing something the arithmetic does not separate: the numerator and the denominator both move, and the plotted value does not record which of them did.

Why there are two lines 

The raw range position, computed fresh on every bar, is conventionally called the fast line. Plotted on its own it is jagged, because a single bar's close landing near either edge of the band moves it the full width of the scale. The second line is that first line smoothed: an average of the last three of its values, computed exactly as the moving average lesson described, and conventionally called the slow line or the signal line.

The fast and slow variants are named for which pair of lines is plotted, and the difference is one averaging step. In the fast variant the raw range position is the main line and its three period average is the second. In the slow variant the arithmetic moves along by one: the three period average becomes the main line, and an average of that becomes the second. The slow variant is therefore not a different measurement. It is the same measurement with one more smoothing applied to both lines, which is why its swings are shallower and why each of its turns arrives later than the equivalent turn in the fast variant.

Charting packages differ in what they label. Some plot the fast variant and call it stochastic, some plot the slow variant under the same name, and some expose the lookback and both averaging lengths as settings without saying which combination corresponds to which traditional name. Two charts of the same instrument over the same bars can therefore show visibly different stochastic lines while both are correctly implemented, and the settings panel is the only place that resolves which calculation is on screen.

Why the reading pins to an edge in a strong move 

The bounded scale has a consequence that governs how the indicator behaves in exactly the conditions it is most often consulted about. When a series moves persistently in one direction, each new bar sets a new extreme of the window and closes near that extreme. The numerator and the denominator then grow together, the ratio stays close to one, and the reading sits near the top of the scale bar after bar without an upper limit on how long it can stay there.

A reading pinned at an edge is not a malfunction, and it is not a measurement of intensity either. It reports, correctly, that recent closes have been landing at the recent extreme. It cannot report how far the move has travelled, because the height of the band was divided out, and it cannot report how long the condition has persisted, because each bar's reading is computed independently of the last. A series that has held the top of its range for three bars and one that has held it for thirty produce the same number.

The conventional readings, and what they assert 

Two conventions are attached to the scale in most trading literature. The first draws horizontal bands at eighty and twenty, and describes readings above and below them as overbought and oversold. Those words are labels for a range position and nothing more: a reading above the upper band states that recent closes sat in the top fifth of the recent band, which the previous section showed can persist for as long as a directional move persists. The same literature that names the bands also records that they hold for extended stretches, which is why practitioners who use them at all conventionally describe them as a description of the window rather than as a statement about what follows.

The second convention concerns the two lines crossing. Because one line is an average of the other, a crossing is an arithmetic event with a fixed meaning: the latest range position has moved to the other side of the average of the last few range positions. Traditions that treat crossings as meaningful disagree about whether a crossing inside the middle of the scale carries the same content as one near an edge, and about whether the frequency of crossings on a short lookback leaves any of them informative. There is no agreed test that settles either question on a given chart.

The conditions under which the reading is useless 

Four conditions make the number technically correct and descriptively empty, and all four follow from the division rather than from any implementation detail.

• A persistent directional move saturates the scale, as the worked example above shows. The reading holds at an edge, stops distinguishing one bar from the next, and reports the same value whether the move has just begun or has run for weeks.
• A very narrow band makes the denominator small, so a trivial move in the numerator swings the reading across the whole scale. On a quiet session the line can travel from one edge to the other on price movement too small to be visible on the chart above it.
• A window in which the highest high equals the lowest low makes the denominator zero and the ratio undefined. Charting packages resolve this differently, some carrying the previous value forward and some plotting the midpoint, so two implementations can print different numbers for the same flat window.
• A gap redraws the band in one step. When a session opens away from the previous close, the new extreme enters the window immediately, and the reading changes by an amount that reflects the gap rather than anything the current bar did.

The common thread is that the indicator discards the two things a reader of the chart above it can see directly: how large the band is, and how the instrument moved through it. A percentage of a range that is never stated is a compressed description, and the compression is the reason the calculation is quick to read and the reason it cannot answer questions about size or duration.

Where practitioners disagree 

The first disagreement is about the lookback and the averaging lengths. A short lookback produces a line that reaches both edges frequently, a long one produces a line that reaches them rarely and later, and no setting removes both of those properties because they are one quantity read from opposite ends. Named default combinations circulate widely and are conventionally repeated, but they originate as choices made on particular instruments and particular bar intervals decades ago, and no procedure in common use establishes that a default transfers to a different instrument or a different timeframe.

The second is whether this indicator adds anything beside the bounded measures already covered in this module. One position holds that a range position and a normalised measure of recent change read the same bars through different arithmetic, so their agreement is one history counted twice, which is the redundancy the leading and lagging lesson set out. Another holds that a measure built from highs and lows behaves differently around the edges of a range from a measure built from closes, and that the difference is real even where the two lines usually agree. Both descriptions of the arithmetic are accurate, which is why the argument recurs rather than resolving, and neither side has produced a test of the question that the other accepts.