What the stochastic actually is

George Lane developed the indicator in the late 1950s (Stochastic oscillator). The premise is that prices tend to close near the extremes of the recent range just before turning points, and the calculation is a single ratio:

%K = (Current Close - Lowest Low) / (Highest High - Lowest Low) x 100

The lowest low and highest high are taken over the look-back period, typically 5, 9 or 14 bars, with 3-period smoothing standard. %D is a 3-period simple moving average of %K, and the signal is the faster %K line crossing %D.

That produces three versions (StockCharts ChartSchool):

• Fast, defaults (14,3): Fast %K is the raw calculation, Fast %D is a 3-period SMA of it

• Slow, defaults (14,3): Slow %K is Fast %K smoothed with a 3-period SMA, Slow %D a 3-period SMA of Slow %K

• Full, defaults (14,3,3): Full %K is Fast %K smoothed by a chosen number of periods, Full %D a moving average of that

The conventional thresholds are above 80 for overbought and below 20 for oversold.

If the ratio feels familiar from Part 17, it should: Bollinger's %b "is derived from the formula for stochastics" (Bollinger Bands), measuring position between two bands instead of a high and a low.

Notice what the "slow" version is: Slow %D is a 3-period average of a 3-period average of the raw ratio, a smoothing of a smoothing, so everything Part 14 established about moving averages and lag applies twice over.

Every input is a price

Lane was specific. The oscillator "follows the speed or the momentum of price," he said, and "as a rule, the momentum changes direction before price." And equally specific about what it did not do: it "doesn't follow price, it doesn't follow volume or anything like that."

 EVERY INPUT IS A PRICE
   %K = (close - lowest low of N) / (highest high of N - lowest low of N) x 100
          -----   --------------      ----------------   --------------
          price       price                 price             price
   then %D and the slow lines are averages of averages of this ratio.
   Three distinct price inputs; nothing else enters at any step.

The fair reading of Lane's claim is that the oscillator does not track the price level, and there it is true and useful: it can sit at 90 whether the asset trades at $12 or $64,000, bounded on a fixed 0-to-100 scale like RSI (by a different formula), which MACD lacks, as Part 16 showed.

But "does not track the price level" is a far smaller claim than "does not follow price." This is Part 13's thesis in its purest form: an indicator computed only from past prices cannot contain information that was not already in them. The stochastic re-expresses the recent range and observes nothing outside it.

Two charts, one reading of 95

Here is the consequence that matters when you place a trade: %K divides by the range, so the range's size cancels out completely.

Chart A's whole fourteen-bar range is four tenths of one percent, and calling that close extended is arithmetic noise. Chart B has run twenty-five percent and is holding near its high, which is what a strong trend looks like. Both fire the same crossover, because the denominator that makes them different is the thing the formula divides away. That is not a flaw in a setting or a market. It is what a percentage of a range is.

Why it sits at 80 for weeks

The second structural property is saturation: %K is capped at 100 the moment the close equals the highest high of the look-back. In a market where every bar closes at a new fourteen-bar high, the numerator equals the denominator and %K reads exactly 100, on the first such bar and the twentieth. A merely strong trend and the strongest trend on record print the identical number; there is no room above 100 to express the difference.

The reference material is blunt: "Overbought readings aren't necessarily bearish. Securities can become overbought and remain overbought during a strong uptrend," with the mirror-image caveat for oversold readings in a downtrend.

It is the Part 15 lesson by another mechanism: RSI 70 means average gains ran 2.33 times average losses, the signature of a healthy uptrend; here the reading simply pins at the top and stops carrying information exactly when the market is moving most. Either way, "overbought" is a description of strength wearing the name of a warning.

What the testing shows, and what it does not

The 2023 crypto study this series has drawn on tested four strategy families on BTC/USDT and ETH/USDT, using White's Reality Check and Hansen's SPA test with slippage of 0.001 and commission of 0.0003 (Chen, Wang & Yang, 2023): the exponential moving average crossover, RSI, Bollinger Bands and MACD. The Stochastic Oscillator is not one of them. Parts 14 through 17 each had a tested result to report; this one does not.

Nor does the broader data-snooping literature hand one over. Hsu and Kuan examined "a more complete universe of trading techniques, including not only simple rules but also complex trading strategies" (Hsu & Kuan, 2005), and their result cuts both ways: significantly profitable rules "do exist in the data from relatively 'young' markets (NASDAQ Composite and Russell 2000) but not in the data from relatively 'mature' markets", the Dow and the S&P 500. A genuine positive finding, net of costs, in the younger indices; not a finding about the stochastic.

The accurate position is narrow: in the evidence base this series has used, the stochastic was not among the rules tested. Not a failed test, and not a pass either.

What can be said is structural rather than statistical: the stochastic is computed from nothing but past prices, its slow and full variants smooth twice, its reading discards the size of the range, and it saturates at both ends. Those facts hold whether or not a test is ever run: limits on what the number can tell you, not a prediction of how it would perform.

Reputation versus evidence

Using the stochastic honestly

• Read it as range position, because that is what it is. It answers one question, where in the recent high-low range did this bar close, and not how far the market travelled.

• Check the range before you read the reading. A 95 on a fourteen-bar range of half a percent is a different world from a 95 after a twenty-five percent run.

• Do not treat 80 as a ceiling. The reading is capped at 100 by construction, so persistence at the top is what a strong trend produces.

• Know which version you are running. Fast, Slow and Full are three amounts of smoothing on the same ratio; the crossover you act on depends on the choice.

• Do not claim it is tested. No tested result for it appears in the studies that vetted the other four families.

Common mistakes

• Shorting a reading above 80 by itself - the reading saturates, and a genuinely strong market will hold there.

• Ignoring the width of the range - the calculation divides it away, so the signal looks the same whether the market moved a quarter of a percent or a quarter of its value.

• Counting a %K/%D cross as independent confirmation - %D is an average of %K, so the two lines are not two opinions.

• Assuming "momentum indicator" means it sees something price does not - a close, a period high and a period low go in, and nothing else does.

• Borrowing the other indicators' evidence - the tested results in Parts 14 to 17 belong to those indicators, not to this one.

The Stochastic Oscillator on ApeX Omni

ApeX Omni's charts carry the stochastic in all three variants. The defensible use on a perpetuals venue is as a fast readout of where the current bar closed inside its recent range, paired with something that knows how wide that range is: volatility is the missing variable, so size the position off the actual size of the move, as Part 17 argued with bandwidth. A high reading is not a reason to fade a trend, and on leverage the cost of fading one that keeps going is a liquidation price that does not care how overbought the oscillator looked, as Part 9 covered. If you take a signal from it, put the stop where structure says and trigger it on the mark price, per Part 12. Trade the chart on ApeX Omni, and let the stochastic describe the bar rather than predict the next one.

The bottom line

The stochastic is the clearest of the five indicators in this series about what it is doing, and the least clearly described. The calculation is one plain ratio: where did this close land between the recent low and the recent high? A real and useful question, but the answer is not momentum, not extension, and not evidence that a move is finished, because the range that would tell you any of that is the exact quantity the formula divides out. Add a ceiling at 100, a slow version that averages an average, and an absence from the tests that judged the other four families, and the position is clear. Use it to see where the bar closed. Take the decision from the range it closed in.

That closes the indicator run. Across five parts the pattern held: each tool is a transformation of prices you already have, and each carries lag as the price of being smooth. Where one has a famous threshold, as RSI does at 70 and the stochastic at 80, it describes strength while sounding like a warning. That is not a reason to clear them off your chart. It is a reason to stop asking them for a verdict they were never built to give.

Frequently asked questions

What does the Stochastic Oscillator measure? Where the current close sits within the high-low range of the look-back period, on a 0-to-100 scale: close minus lowest low, divided by highest high minus lowest low, times 100.

Who created it? George Lane developed it in the late 1950s. He described it as following the speed or momentum of price, with momentum as a rule changing direction before price.

What is the difference between fast, slow and full stochastic? All three use the same ratio and differ only in smoothing: Fast is the raw %K with a 3-period SMA as %D; Slow smooths Fast %K with a 3-period SMA and averages that again for %D; Full lets you set the smoothing period, defaulting to 14, 3 and 3.

Does a reading above 80 mean sell? Not on its own. A security can become overbought and remain overbought during a strong uptrend, and the reading is capped at 100, so it cannot show a trend strengthening once pinned.

Why can two very different charts give the same reading? Because the size of the range cancels out of the formula: a fourteen-bar range of 0.4% and one of 25% both read 95 if the close sits nine tenths of the way up.

Has the Stochastic Oscillator been tested academically? No tested result for it appears in the studies this series has drawn on: the 2023 crypto study tested exponential moving average crossovers, RSI, Bollinger Bands and MACD, not the stochastic. That is an absence of evidence rather than evidence against it.

Explore more from this series: Part 14: Moving Averages | Part 15: RSI Explained | Part 16: MACD Explained | Part 17: Bollinger Bands

This article is for educational and informational purposes only and is not financial, investment, or legal advice. Do your own research before making any trading decision.