Indicator library · Volatility
Relative Volatility Index
RSI’s arithmetic with the standard deviation of price substituted for the price change. It reports which side volatility has been arriving on, and Dorsey built it to confirm another indicator rather than to signal on its own.
The calculation
- Compute the standard deviation of the closing price over a lookback, conventionally 10 sessions.
- Assign that deviation to the up series when the close rose against the previous one, and to the down series when it fell.
- Smooth both with Wilder’s method over 14 periods, then convert exactly as RSI does: 100 − (100 ÷ (1 + average up ÷ average down)).
One substitution, and it changes what the output means completely. RSI aggregates the sizes of the moves themselves. This aggregates the dispersion of the closes, and sorts it by the direction of the day it happened on. The question it answers is therefore not "how big were the moves" but "was the volatility arriving on up days or on down days".
Why the level says nothing about how volatile the market is
The final step is a ratio, so the amount of volatility divides out. This is worth stating in numbers because the name of the indicator invites the opposite reading.
| Market | What the RVI reports |
|---|---|
| Drifting up half a per cent a day, almost no down days | A high reading, nearly all the (small) deviation landed on up days. |
| Advancing five per cent a day with violent pullbacks | A similar high reading, if the up days still carry most of the deviation. |
| Very quiet, up and down days alike | Near 50, balanced, and saying nothing about the calm itself. |
If the quantity you need is how much the instrument moves, the answer is average true range or the standard deviation itself. Reaching for this because it has volatility in the name is the commonest way it gets misused.
Dorsey built it as a filter
The original guidance is unusually explicit and almost always omitted from descriptions: take signals from another method, and act only when the RVI agrees, long signals above 50, short signals below it. The reasoning is that a price move whose volatility has been arriving on the same side is better supported than one where the dispersion is coming from the other direction.
That is a modest claim and it is testable, which is more than most indicator guidance offers. What it is not is a trigger: on its own the line crosses 50 frequently, parks at an extreme in a sustained trend exactly as RSI does, and produces the same stream of premature counter-trend signals. Using it as Dorsey intended costs nothing and keeps the claim as small as the evidence.
Two ways to test a filter
Because Dorsey's claim is narrow, it is one of the few pieces of indicator guidance that can be checked without much machinery. Take whatever signals your own method produces, split them by whether the index was above or below 50 at the time, and compare the two groups.
Two outcomes are informative and both are useful. If the filtered group performs no differently, the filter is costing you signals for nothing, which is a finding rather than a failure. If it does separate them, the next question is whether the difference survives on a second instrument and a different period, because a filter that helps on one sample and not on another has found the sample. Either way the test takes an afternoon and replaces an opinion with a number.
Where it misleads
| Situation | What goes wrong |
|---|---|
| Read as a volatility level | The final step is a ratio; the amount of volatility divides out. A calm market and a violent one can print the same number. |
| Traded on its own | It was published as a confirming filter. Alone it parks at extremes in a trend and crosses 50 constantly in a range. |
| RSI thresholds reused | 70 and 30 are borrowed conventions here; the useful line in Dorsey's own method is 50. |
| Confused with the relative vigor index | Both are abbreviated RVI and they are unrelated measures. |
| Two windows tuned together | The deviation window and the smoothing window interact, so searching both overfits faster than tuning one period. |
| Short history | A deviation window inside a Wilder smoothing means early values depend heavily on where the data begins. |
What volume adds
Every input here is a closing price, so the index can report that volatility has been arriving on up days without any knowledge of how many participants produced it. A high reading on thinning volume describes dispersion in a book with few orders in it; the same reading on expanding volume describes a market being actively contested on the up side.
Used as Dorsey intended, as a filter on someone else’s signal, that distinction matters directly, because it is the difference between a confirmation worth having and a confirmation produced by an empty market. The volume beside the chart answers it at no cost.
Frequently asked questions
How is the relative volatility index calculated?
Exactly as RSI, with one substitution: where RSI uses the size of each price change, the RVI uses the standard deviation of the closing price over a lookback, conventionally 10 sessions. That deviation is assigned to the up side when the close rose and to the down side when it fell, the two series are smoothed with Wilder’s method over 14 periods, and the result is converted with 100 − (100 ÷ (1 + average up ÷ average down)). Donald Dorsey published it in 1993.
What is it actually measuring?
Whether volatility has been arriving on up days or on down days. It is not a measure of how much volatility there is, a quiet market and a violent one can both read 70, because the reading is a ratio of one kind of volatility to the other. That distinction is the single most common misunderstanding of this indicator, and it follows directly from the division at the end of the formula.
Why did Dorsey build it as a filter rather than a signal?
Because it answers a question that is useful alongside a directional method and nearly useless alone. His own guidance was to take a signal from another indicator only when the RVI agreed — buy signals only above 50, sell signals only below — on the argument that a price move whose volatility has been arriving on the same side is better supported than one where it has not. Read on its own it produces the same stream of premature crossings as any oscillator.
Does a high reading mean high volatility?
No. The reading is bounded 0 to 100 because it is a ratio, so the absolute level of volatility divides out entirely. An instrument moving half a per cent a day and one moving five per cent a day can both read 80, provided the movement in each is concentrated on up days. If what you want is the amount of volatility, that is average true range or the standard deviation itself, not this.
What lookbacks are standard?
A 10-period standard deviation smoothed over 14 periods, which is Dorsey’s original pairing, RSI’s 14 with a 10-day deviation feeding it. Two parameters instead of one means they interact: lengthening the deviation window makes the input steadier, while lengthening the smoothing makes the output steadier, and the two are not interchangeable. There is no reason to treat either number as optimal.
What is the difference between this and the volatility index versions on price highs and lows?
Dorsey later published a refinement computing the measure separately on the highs and on the lows and averaging the two, to reduce the dependence on the closing print. It is a reasonable improvement and a different series. Thresholds from one do not transfer to the other, and two platforms plotting "RVI" may not be plotting the same thing. Note also that RVI is used elsewhere as an abbreviation for the relative vigor index, which is unrelated.
How much history does it need?
More than the two periods suggest, because there are two stages of dependence. The standard deviation needs its own window filled, and then Wilder smoothing carries its seed forward with a decaying weight. Feed several hundred bars and read none of the first hundred if the values are to be compared with another implementation.
Does it work on volume instead of price?
The arithmetic runs, and the result is hard to interpret. The standard deviation of volume is dominated by the same right-skew that makes a volume average awkward, so a single expiry session can carry the reading for the length of both windows. If the question is whether participation is arriving on up days or down days, up/down volume answers it directly and without two layers of smoothing in the way.