Stochastic RSI: An Oscillator on an Oscillator
The stochastic RSI applies the stochastic formula to relative strength index values rather than to price, producing a faster oscillator scaled from zero to one hundred. Because it is a transformation of a transformation, it reaches its extremes far more often than either input does.
How it works
The stochastic formula asks where the current value sits within its own recent range. Applied to price, it asks where today’s close sits between the highest high and lowest low of the last N bars.
The stochastic relative strength index applies that same question to RSI values instead. Where does today’s relative strength index sit between its own highest and lowest readings of the last fourteen periods?
So the input to the calculation is not price at all — it is an indicator that was itself computed from price. That stacking is the entire design, and everything else on this page follows from it.
What stacking actually buys
It is more responsive, and that is the only thing it gains. Rescaling the relative strength index against its own recent range stretches small movements into large ones, so turns that were a few points on the original become swings from 20 to 80 on this.
No new information enters at any point. The inputs are the same closes, run through two transformations rather than one. A tool built from a tool cannot know more than the tool it was built from.
Which is why it spends so much time at its extremes. When the relative strength index makes a new 14-period high, this reads 100 by definition — and it stays at 100 for as long as that keeps happening. Pinning is the normal behaviour of the calculation in a trend, not evidence of anything.
Treating a pinned reading as “extremely overbought” is the standard misuse. It means the underlying index has been at the top of its own range, which in an advance is what you would expect.
The overlap nobody measures
On this site’s shared 576-bar history, the bar-to-bar changes in a 14-period relative strength index and a 14-period stochastic correlate at 0.87. They are close to the same instrument already, before anyone combines them.
Every other oscillator pair measured on the same data sits between 0.53 and 0.71 — momentum against the relative strength index at 0.71, the moving average convergence divergence histogram against it at 0.71, the commodity channel index against the stochastic at 0.63.
Those figures are the honest context for adding a fourth oscillator to a chart. The tools on the menu are not independent measurements of different things; they are variations on the same reading, and the confluence page covers why stacking correlated inputs manufactures confidence rather than evidence.
In practice: the costs of the extra layer
It has four settings where a plain relative strength index has one: the index period, the stochastic lookback, and the two smoothings applied to the result. Four dials tuned on past data will always produce something that looks excellent on that data.
A common confusion worth clearing up: this is not the two indicators plotted together. It is one applied to the other. A chart with a stochastic pane and a relative strength index pane shows two readings; this shows one reading that has been through two calculations.
Neither layer looks at volume. Whatever independence you are hoping to add to a chart, it will not come from another price transformation — it has to come from a genuinely different input.
On longer timeframes the difference from a plain index shrinks, because both are summarising more data and the rescaling has less relative effect. The extra layer earns its keep, if anywhere, on fast charts.
A gap moves both layers simultaneously. The index jumps, and because it jumped to the edge of its recent range, this pins immediately. Two indicators reacting is one event entering two calculations.
It produces no price. Every stop has to come from structure — a swing low, a level, a volatility multiple. An oscillator reading of 12 is not somewhere you can place an order.
And the faster tool produces more signals, each costing 2% of a typical bar’s range in round-trip costs on this history. That is the concrete price of responsiveness.
What the stochastic RSI is not
It is not two indicators agreeing. It is one indicator inside another.
It is not more sensitive to price. It is more sensitive to the relative strength index, which is itself a smoothed view of price.
It is not a better overbought signal. Extremes are more frequent here by construction, which makes each one mean less rather than more.
And it is not independent of anything already on your chart. At 0.87 correlation between its two ingredients, it is a third view of a thing you have twice.
When it fails
The dominant failure is trading the extremes in a trend. Pinned at 100 through an advance, it reads as maximally overbought on every bar, and each bar is an invitation to sell into strength. The reading is correct and the inference is not.
The second failure is parameter tuning. With four settings, a configuration that fits the last year perfectly is always available, and it will not be the configuration that fits next year.
A third is the double-lag problem in reverse. The extra rescaling makes the tool turn earlier, which in a range means turning earlier on noise. Faster is only better where the moves are real.
A fourth is confusing it with the stochastic. They have similar names, similar scales and different inputs, and a strategy specified for one behaves differently on the other.
And a fifth is stacking it with the plain index for confirmation. One is computed from the other. They cannot disagree in any way that carries information, which is the clearest possible case of the correlated-inputs problem.
The original data
Bar-to-bar change correlations across five oscillators on this site’s shared 576-bar history: relative
strength index against stochastic 0.87, against momentum 0.71, against the moving average convergence
divergence histogram 0.71, against the commodity channel index 0.71; the lowest pair in the whole set is
0.53. The matrix is in research/series-measurements.json, produced by site/measure_series.py.
Correlations of changes, not levels, is the measurement that matters and almost nobody publishes it. Levels correlate trivially, because every oscillator tracks price. Changes correlating at 0.87 means the second tool is telling you what the first already said, on the same bar, most of the time — which is the number to compute before adding anything to a chart. Two lines of code on your own data, and it settles arguments about indicator selection that otherwise run forever.
Related
RSI is the input this is computed from. Stochastic is the formula being applied to it. And choosing indicators carries the wider measured question of which tools actually add information.
I added the stochastic relative strength index because my ordinary one felt slow, and what I actually got was the same information arriving earlier and being wrong more often. The honest version of what I wanted was a shorter lookback on the tool I already had.
— Michael Whitman
This page is educational, not financial advice. Test every idea on your own charts before risking money.