Indicators · Volume
Negative Volume Index (NVI)
The one measure here that moves only when volume falls. Its premise is that quiet sessions and busy ones are traded by different people, an interesting hypothesis, calibrated on a market that no longer exists.
What it is calculated from
Two inputs, and one comparison. Each session, ask whether volume was lower than the session before:
- Volume fell. Add the session’s percentage price change to the index: NVI = previous NVI × (1 + (close − previous close) ÷ previous close)
- Volume rose or was unchanged. The index does not move at all.
The Positive Volume Index is the same construction with the condition reversed. Both are conventionally started at 1000, and between them they partition every session in the series into exactly one of the two indices, which is the neatest thing about the pair and the source of its one durable insight.
The hypothesis, stated fairly
The measure exists because of a specific claim about who trades when. On heavy-volume days the market fills with participants reacting to news; on quiet days it is left to those who are trading for reasons that do not depend on a headline. If that is true, the price drift on quiet days carries better information than the drift on busy ones.
It is a real hypothesis with a plausible mechanism, and it deserves more respect than most indicator folklore. What it does not have is the evidence that would let anyone quote a threshold from it. The published readings come from studies of one exchange in the decades before off-exchange trading, electronic market making and scheduled index rebalancing, all of which changed what makes a session quiet.
Where the construction is weak
The filter compares one bar with one bar
"Volume fell" means lower than the immediately preceding session, and nothing more. A session with three times normal volume following one with four times normal volume qualifies as quiet, and its price change is credited to the patient, informed participants the hypothesis is about.
Comparing against a volume baseline instead would express the intended idea far better, a session below its twenty-day average is genuinely quiet in a way that "below yesterday" is not. That is not how the measure is defined, and redefining it stops it being comparable with any published series, so the defect has to be lived with and known about.
Half the sessions are discarded
By construction the index ignores every busy session, which means long stretches where it records nothing while price moves substantially. That is intentional, and it has the consequence visible in the figure: the line is a series of plateaus, and its shape at any moment depends on how the volume happened to alternate rather than on the size of anything.
The level is arbitrary
Like every cumulative series in this library, on-balance volume included, the number it reaches depends entirely on the start date. Only direction and the relationship to its own long average can be read, and any commentary quoting the level is quoting a decision someone made about where to begin.
How it is used, and what that use can support
| Reading | The claim | What it actually supports |
|---|---|---|
| Index above its one-year average | A bull market: quiet-day drift has been upward. | A slow regime label that changes a few times a decade. Descriptive, and far too coarse to time anything. |
| Index below its one-year average | A bear market by the same logic. | The same, and it lags badly at turns because the average is a year long. |
| NVI rising while PVI falls | Quiet-day buyers accumulating against crowd selling. | A genuine split in the series worth noticing. It is not evidence about who was on which side, the labels are an interpretation. |
| Both indices rising | Broad agreement. | Only that the price rose on both kinds of session, which is what a trend is. The measure adds nothing here. |
Why the pair is still worth knowing
One insight survives all of the above, and it is not about forecasting. Splitting a price series by a volume condition and accumulating each half separately is a general technique, and this pair is the oldest well-known example of it. It makes visible something no single series shows: whether the price change in a period was concentrated in the busy sessions or the quiet ones.
That question is worth asking of any instrument, and the answer occasionally is striking, an advance made almost entirely on quiet days is a different animal from one made on heavy ones. The measure’s failing is not the question; it is that "quiet" was defined against a single previous bar and that a fifty-year-old threshold was attached to the result. Both are fixable by anyone computing it from their own data, which is the most useful thing this page can say about it.
Frequently asked questions
What is the Negative Volume Index?
A cumulative index that changes only on sessions when volume was lower than the session before. On those days it adds the percentage price change; on every other day it stays exactly where it was. Its companion, the Positive Volume Index, does the opposite; it moves only when volume rose. The pair was popularised by Norman Fosback in the 1970s, building on earlier work by Paul Dysart.
What is the reasoning behind it?
That quiet sessions and busy ones are driven by different participants. The argument is that heavy volume brings in the crowd reacting to news, while on quiet days the market is left to those who trade without needing a headline, so the price drift on low-volume days is claimed to reveal the better-informed side. It is a genuinely interesting hypothesis and it is a hypothesis, not a finding: the mechanism is plausible and the evidence for it is thin.
How is it different from on-balance volume?
On-balance volume adds or subtracts the whole session’s volume according to the direction of the close, so every session changes it and the size of the change is the volume. This index ignores volume as a quantity entirely. It only asks whether volume fell, and then accumulates the price change. One is a volume total classified by price; the other is a price total filtered by volume.
What does the level mean?
Nothing, in the same way that no cumulative line has a meaningful level. The index is conventionally started at 1000 and the number it reaches depends on when the accumulation began and how many quiet sessions the period happened to contain. Only its direction, and its relationship to price and to its own moving average, carry any reading.
How is it read?
Usually against a long moving average of itself — a one-year average is the traditional choice — with the index above its average taken as a bull market and below it as a bear market. That is a slow, coarse regime classification rather than a trading signal, and it is the honest use of the measure: it changes state a few times a decade, and it is neither precise nor timely.
Does it still work?
The original claims were made about an exchange and an era with far less off-exchange trading, no high-frequency market making and no index funds rebalancing on a schedule. Since the measure depends entirely on the meaning of a "quiet" day, and what produces a quiet day has changed completely, transferring the published thresholds to today is unsupported. The construction is still computable; the interpretation was calibrated on a different market.
What is its most obvious weakness?
That "volume fell" is a comparison with one previous session, which makes the filter extremely noisy. A session with enormous volume followed by one with slightly less enormous volume counts as a quiet day, and its price change is credited to the informed traders. The measure would be more defensible comparing volume against a baseline average rather than against a single bar, and that is not how it is defined.
Is it worth computing?
As a piece of the field’s history and as a check on your own data, yes; as a decision input, only with the caveats above stated. It appears in most charting packages and in the older literature, so knowing exactly what it does prevents it being read as something it is not, and its construction makes a useful teaching point about what a filtered cumulative series can and cannot show.