Stochastic Oscillator: %K, %D and Smoothing

Start with raw %K

Raw %K is 100 × (close − lowest low) / (highest high − lowest low) for the chosen window. It measures where the latest close sits within that range. Fidelity describes the fast stochastic construction. The name does not mean the indicator estimates a full probability distribution for the next price.

A signal line, often called %D, averages %K. Slow and full variants introduce additional smoothing or expose more parameters. Consequently, a label such as stochastic 14 is incomplete: it may omit the %K smoothing, %D length, averaging type and price-field convention used by a platform.

A worked range and smoothing example

Take a hypothetical highest high of 80, lowest low of 60 and close of 75. Raw %K is 100 × 15 / 20 = 75. If the next two completed observations produce raw readings of 60 and 45, a three-observation simple %D is (75 + 60 + 45) / 3 = 60.

If those raw readings are first smoothed into a slow %K series, its own %D requires further observations. Averaging the same three values twice without maintaining the rolling sequence is not an equivalent shortcut. Name each intermediate series in a worksheet so the two smoothing stages remain visible.

Now consider an unchanged close of 75 after an old high of 80 leaves the lookback. With a remaining high of 76 and low of 60, raw %K becomes 93.75. The increase came from a narrower historical range. It did not require a higher closing price or a newly executed trade.

Handle limits and live bars

If the window high equals its low, the denominator is zero and a defined software convention is needed. An unfinished bar can also update the close and either extreme repeatedly. Comparing a live reading with a final historical reading without recording the timestamp can make the same implementation appear inconsistent.

Traditional upper and lower threshold regions indicate closeness to a range edge. They are not independent forecasts of a turn. Price can remain near repeated highs in an advance, while a range-bound series may generate many crossings that would incur significant spread and commission if every one were traded.

Use one reproducible interpretation

A proposed crossover rule should state which series crosses which, whether equality counts and whether the bar must close. It also needs a policy for a signal appearing immediately after a missing-data interval. Those details determine the sequence of eligible trades more than the visual color of the oscillator.

Compare Williams %R, which is a rescaled version of the same raw range location when inputs match. Contrast RSI, which uses changes rather than high-low placement. Use the backtesting guide to evaluate any rule separately from the definition of the indicator itself.