What Is the KST Oscillator?
KST oscillator sums four rate-of-change measurements, each smoothed and weighted differently, into a single line intended to capture several market cycles at once. Its four lookback periods and four weights give it eight settings, which is eight places a result can be fitted to history.
The KST is the most heavily constructed indicator on this site: four measurements, four smoothings, four weights, then a signal line on top. Understanding what that construction buys and what it costs is the whole subject.
How it works
Rate of change is the base measurement. Price now against price N bars ago, as a percentage. Four of those are computed over four different lookbacks — short, medium, long and very long.
Each is smoothed with its own moving average, then multiplied by its own weight — conventionally 1, 2, 3 and 4, with the longest window carrying the most.
The reasoning is that markets move on several cycles simultaneously. A short rhythm inside a medium one inside a long one — and a single lookback can only see one of them.
The cost is eight parameters. Four periods, four weights, plus the smoothing lengths and a signal line. Every one is a dial, and every dial is somewhere a historical result can be improved without improving anything real.
Why it signals late
Every input is already smoothed before it is combined. So the sum inherits the lag of all four, and the longest-weighted component contributes the most delay while carrying the heaviest weight.
Then a signal line is applied on top, adding one more layer. A crossing of that line is therefore a smoothed reading of a smoothed reading of four smoothed readings.
That is not a flaw to be tuned out. Smoothing and delay are the same operation seen from two sides, and no arrangement of averages produces calm without it.
A worked example
Take this site’s shared series. Direction runs average 2.01 bars with a longest of 11. Median bar range is 0.493, ninetieth percentile 1.101.
The default periods are 10, 15, 20 and 30. Against two-bar average runs, the 30-period component is spanning roughly fifteen direction changes.
A measurement covering fifteen alternations is close to a constant. It averages the noise flat and contributes very little that varies — while carrying the heaviest weight in the sum.
So the indicator’s most influential component is its least informative one. That is the structural criticism, and it is visible in the arithmetic rather than a matter of opinion.
What the construction is really doing
Rate of change is unbounded. Unlike RSI, the KST has no ceiling or floor, so “high” and “low” are relative to the instrument’s own history rather than to fixed levels.
Which means the zero line carries most of the meaning. Above zero is net positive momentum across the blended windows; below is net negative. The absolute value is comparable only to itself.
And the signal-line crossing is the conventional trigger — which is the same construction as a MACD, applied to a more elaborate input.
The original data
On this site’s shared series: direction runs average 2.01 bars with a longest of 11. Breakouts continued in 85% of 39 twenty-bar events. A round trip costs 0.0098, about 2% of the median bar range of 0.493.
Those two figures frame the problem. Runs are short, so a slow indicator will confirm most of them after they have ended; and breakouts continue 85% of the time, so the late confirmations that do arrive are pointing at moves with real persistence behind them. The lag is expensive on the noise and affordable on the few that matter.
Where the idea came from
Martin Pring published it in the early 1990s, building on a long tradition of rate-of-change work and on the observation that different market participants operate on different horizons.
That observation is sound. A pension fund rebalancing quarterly, a swing trader holding for weeks and a market maker holding for seconds are all acting on the same instrument on completely different clocks, and any single lookback captures at most one of them.
Whether summing four windows recovers that structure is the open question. Adding measurements together produces a blend, not a decomposition — the four cycles go in and one number comes out, so whatever separation existed in the inputs is discarded at exactly the point it might have been useful.
A genuine multi-horizon approach would keep them apart and let each speak. Displaying four lines rather than one sum would say more, and would also look far less decisive, which is probably why the summed version is the one that became popular.
When it fails
The characteristic failure is optimising the eight settings on historical data. With four periods and four weights there is always a combination that would have caught last year’s turns cleanly, and finding it takes minutes. The resulting chart looks authoritative because it was fitted to look that way, and the settings describe one sample rather than any property of markets. An indicator with eight dials is not eight times more informative than one with one — it is eight times easier to fool yourself with.
A second failure is expecting it to be early. It is the most heavily smoothed indicator most people will use, and it confirms rather than anticipates.
A third is reading its level as overbought. It is unbounded; there is no level at which it is extended by definition.
A fourth is using it on short timeframes, where the long-weighted component averages to a constant and the whole construction collapses toward a slow moving average.
And a fifth is treating the default settings as measured. They are Martin Pring’s choices, published and widely copied, and no published work establishes them as optimal for anything you trade.
Related
Moving average covers the smoothing applied to all four components. RSI covers the bounded alternative and why the level means something there. And technical analysis covers the wider tradition.
The name is a joke — Martin Pring called it Know Sure Thing — and the joke is the most honest part of the indicator. Everything else about it is a bet that combining four lagging measurements produces something that is not lagging, which is not how lag works.
— Michael Whitman
This page is educational, not financial advice. Test every idea on your own charts before risking money.