Stochastic Oscillator: Formula, Settings, and What Tests Reveal

The stochastic oscillator starts with the latest close’s position within a recent high-low range. Its familiar %K and %D lines can smooth that raw reading, helping traders read momentum, identify relatively high or low readings, and define trading signals.

Across 101,952 configuration evaluations, we separated two choices that often get bundled into “slower stochastic settings”: smoothing the oscillator itself and smoothing its signal line. They change different parts of a trading rule. A quieter line also need not deliver a better result.

What does the stochastic oscillator measure?

A raw reading of 80 places the close four-fifths of the way from the window’s lowest low to its highest high. It does not mean price has risen 80%, or that there is an 80% chance of a reversal.

For a nonzero price range:

Raw %K = 100 × (close − lowest low) / (highest high − lowest low)
Smoothed %K = simple moving average of raw %K
%D = simple moving average of the chosen %K line

The high and low come from the same chosen lookback. With a range from 90 to 110 and a close of 106, raw %K is 80. A close of 94 gives 20. The current candle’s high and low belong to the range.

TradingView’s stochastic documentation describes the construction and the common 14-period range with three-period smoothing. Periods mean candles; fourteen hourly candles and fourteen daily candles describe different amounts of time.

Fast, slow, and full stochastic settings

The terminology can make two identical-looking settings mean different things:

• Fast stochastic uses raw %K and usually a three-period average for %D.
• Slow stochastic first averages raw %K over three periods, then averages that line again for %D.
• Full stochastic lets you choose the range lookback and both smoothing lengths.

The familiar 14/3/3 full setting means a fourteen-candle range, a three-candle average of raw %K, and a three-candle average of that smoothed %K. StockCharts explains the fast, slow, and full variants.

Two consecutive averages add memory. They do not convert the oscillator into a forecast, and changing %D does not change %K itself.

Common stochastic signals

20 and 80 are common oversold and overbought reference levels. For raw %K, below 20 places the current close in the bottom fifth of its recent range; above 80 places it in the top fifth. A smoothed %K averages recent range-position readings, so it can remain below 20 after the current close has left that lower fifth. A persistent trend can keep producing high or low readings.

A trader might wait for %K to recover above 20, or for %K to cross above %D while the lines are low. These are different triggers. The first compares a line with a fixed level; the second compares two moving lines and may add a level condition.

Divergence asks a different question again: does a new price extreme have a matching oscillator extreme? Our tests below do not automate divergence or reproduce George Lane’s complete method. They isolate two simple, explicit rules so changes in the settings remain interpretable.

Why stochastic can rise when the close does not

The latest close is only one part of the formula. The range can change when an older high or low leaves it.

Select Old high remains, then Old high expires. Every close in this constructed example is 100. While the five-candle window still contains a high of 120 and a low of 99, raw %K is about 4.76. When that old high leaves, the highest high becomes 101 and raw %K becomes 50.

Price has not rallied. The reference range has contracted around it. A three-period average spreads the change over several readings, so the smoothed oscillator catches up later.

This is arithmetic implied by a rolling range, rather than a newly discovered trading edge. It matters when a rise in stochastic is casually described as new buying pressure: sometimes part of the rise comes from the price leaving the calculation.

What changing the smoothing actually changed

Our first rule bought when %K crossed back above a lower threshold and exited when %K reached the corresponding upper threshold. With the familiar 20/80 levels, it waited for recovery above 20 and exited at 80 or higher. It skipped an entry if the same close had already reached the upper exit threshold.

The second rule bought a %K/%D upward crossover only when %K was at or below the lower threshold. It exited on a downward crossover. Here, changing %D changes both the entry comparison and the exit comparison; it does not merely make the displayed %K line calmer.

In the displayed controlled example, increasing %K smoothing from one to three reduced recovery-rule executions from 588 to 371. In the separate crossover rule, keeping %K smoothing at three and extending %D from three to five reduced executions from 522 to 322. Both changes reduced activity in all eight historical samples. The complete curves preserve every intermediate setting.

Each buy or sell counts as one execution. These are executed trading rules, so their counts differ from the number of threshold crossings visible on an indicator chart. A signal arriving while the rule already holds a position cannot create an additional entry.

The practical distinction is to name the parameter and the decision it changes. “Use more smoothing” is incomplete. Increasing %K smoothing changes the oscillator fed into every subsequent condition. Increasing %D smoothing changes the moving reference against which %K is compared.

Can more D smoothing create more trades?

Yes, in some configurations of this rule. A separate fixed quote check used a range length of 57, K smoothing of 8, and an entry ceiling of 10. Increasing D smoothing from three to five changed recorded executions from 16 to 18.

We selected this case after inspecting the broad study, then froze the same eight D lengths and repeated them on the other seven historical samples. The increase appeared in one of eight samples; in the other seven, D5 produced fewer executions than D3. Every outcome remains in the download; this selected case does not estimate how often the effect occurs.

The entry requires an upward crossover while K is still at or below the ceiling. Changing D moves that comparison and also moves the downward-cross exit. Making a line smoother therefore does not guarantee fewer executions of the complete rule.

Do fewer trades mean better stochastic settings?

The broader percentage-sized sweeps produced both higher and lower returns when %D smoothing changed from three to five, with all other inputs fixed within each comparison. Reduced activity was therefore not a universal improvement. Those recorded portfolio results also include the sizing, execution, and cost assumptions described below.

Reducing activity can reduce fee exposure, but it also changes entry and exit timing, how long the strategy holds, and which moves it misses. A high winning percentage can coexist with a poor total return if the losing trades or time spent on the wrong side of a move are large enough.

The tested history does not supply one best setting for all markets. A useful next comparison holds the lookback and thresholds fixed, changes one smoothing length, and checks executions, drawdown, costs, and return together. Then repeat that exact comparison on another sample.

Reproduce the low-zone crossover test

This script implements the second rule. The downloadable companion implements the separate threshold-recovery rule. They are research adaptations with explicit entries and exits; neither is a complete stochastic trading system.

//@version=6
strategy("Stochastic low-zone crossover research", overlay=false, initial_capital=10000, default_qty_type=strategy.percent_of_equity, default_qty_value=99, process_orders_on_close=false)
length = input.int(14, "Range length", minval=2)
smooth_k = input.int(3, "K smoothing", minval=1)
smooth_d = input.int(3, "D smoothing", minval=2)
lower = input.int(20, "Entry ceiling", minval=1, maxval=49)
raw_k = ta.stoch(close, high, low, length)
k = ta.sma(raw_k, smooth_k)
d = ta.sma(k, smooth_d)
ready = bar_index >= 500
enter = ta.crossover(k, d) and k <= lower
leave = ta.crossunder(k, d)
if ready and strategy.position_size == 0 and enter
    strategy.entry("Long", strategy.long)
if ready and strategy.position_size > 0 and leave
    strategy.close("Long")
plot(k, title="K", color=color.aqua)
plot(d, title="D", color=color.orange)
hline(20, title="Common lower level")
hline(80, title="Common upper level")

Compare the same idea with the RSI guide and the MACD guide: the line, the trigger, and the trade each answer a different question.