Education · Reference · Indicators
Momentum: RSI, stochastic, RVI and CCI
These four look very different on a screen and are asking almost the same question: compared with its own recent past, is this move unusual? Where they differ is in what «unusual» is measured against — and in one word that has probably cost traders more money than any other in this field.
Mak · 8 min read · Reference · Indicators, 3 of 8
An oscillator takes a quantity with no natural limits — price — and expresses it on a bounded scale, so that today can be compared with last week. That rescaling is the whole idea, and it carries one consequence you should fix in place now: the scale is relative to a window you chose. Change the window and the same market becomes a different reading.
RSI
Welles Wilder, 1978. Over the last fourteen bars, separate the moves into ups and downs. Average each group. The ratio of average gain to average loss is smoothed, then squeezed onto a 0–100 scale.
So RSI is not measuring how far price moved. It is measuring how one-sided the recent moves were. Fourteen small up bars and no down bars produce a higher reading than a violent rally that had two red bars in it. That distinction explains most of the occasions when RSI seems not to match what you are looking at.
Wilder chose fourteen for daily bars, using a calculator. It has been copied ever since.
Stochastic
Popularised by George Lane — worth phrasing carefully, since the idea is generally traced back to C. Ralph Dystant in 1948, and Lane's contribution was a lifetime of teaching it rather than inventing it.
The question is narrower and more concrete than RSI's: where in the recent range did this bar close? Take the highest high and the lowest low of the last fourteen bars, and ask what percentage of the way up that range today's close sits. That is %K. Smooth it over three bars and you have %D, the slower line people actually watch.
Because it uses the extremes of the range rather than the size of individual moves, the stochastic reaches its boundaries readily — it is designed to. It is at its most informative in a market that is going sideways, where the range is stable and «near the top of it» genuinely means something. It is at its worst in a strong trend, for the reason in the next section.
The word that costs money
«Overbought» does not mean «about to fall». It means «this has been going up persistently». Those are not the same statement, and a strong trend is precisely the situation that produces the first one continuously.
An oscillator pinned near its upper boundary for two hours is not warning you. It is reporting, accurately, that buyers have been in control for two hours. Selling because a line is high is selling because a market is strong — which, written out like that, is clearly not a reason.
The rule of thumb that follows is worth more than any threshold: extremes are informative in a range and actively misleading in a trend, so the first thing to establish is which of the two you are in. That is what the trend page before this one is for, and ADX in particular.
The same applies to divergence — price making a higher high while the oscillator does not. It is a real observation, and what it observes is narrow: this leg was less one-sided than the last one. Sometimes that precedes a turn. Sometimes a trend simply settles into a steadier rhythm and the divergence persists for the rest of the session while price keeps going. Treated as «the character of the move has changed, look closer», it is useful. Treated as a countdown, it is expensive.
Relative Vigor Index
John Ehlers, Stocks & Commodities, January 2002. The premise is a single observation, and an old one: in a rising market, price tends to close above where it opened; in a falling one, below. So rather than measuring the size of the move, measure where within the bar the close landed.
The calculation divides close minus open by high minus low — the body as a fraction of the whole bar — and then, because a single bar of that is unusably noisy, runs it through a short symmetrical weighted average before comparing it with a signal line built the same way.
That smoothing is the Ehlers part, and it is the reason this one links forward: the same author spent a career arguing that most indicators are ruined not by their idea but by their filtering, which is a page of its own in this section.
Where it is weak is inherited from its premise. On instruments and sessions where the open of each bar is effectively the previous close — which is most of a continuously traded future — «close versus open» is a narrower measurement than it sounds. And any bar-based body measure is distorted by the choice of bar: the same flow sliced into volume bars instead of time bars produces different bodies, as the article on candle types sets out.
CCI
Donald Lambert, in Commodities magazine in 1980. The name is misleading and has been for forty years: nothing in the formula is specific to commodities. It runs on anything with a high, a low and a close.
It works on the typical price — the average of high, low and close — and asks how far that has strayed from its own moving average, measured in units of its average deviation. In plain terms: is this unusually far from normal, by this market's own recent standard of normal?
The one detail worth knowing, because it demystifies the ±100 lines everybody draws:
Lambert divided by a constant of 0.015 for one reason — to make roughly 70–80% of readings fall between −100 and +100. The famous levels are a consequence of a scaling choice, not a discovery about markets.
That is worth knowing in both directions. It means ±100 is arbitrary in origin. It also means it is not meaningless: a reading beyond it really is in the minority of that market's recent behaviour, which is exactly what it was engineered to say. Unlike RSI and stochastic, CCI is unbounded, so it can keep going — and in a genuine trend, it does.
What none of the four tell you
- Whether you are in a range or a trend — which is the one thing that decides how to read all four. They cannot tell you, because they are computed the same way in both.
- Who was doing the buying. A one-sided series of bars produced by steady retail buying and one produced by a single participant filling an order look identical here. The tape and absorption are where that distinction lives.
- Anything about size or risk. None of these has any idea how far price can travel against you while the reading stays «extreme». That is volatility, and it is the next page.