WhitmanTrading

How to Use On Balance Volume

To use on balance volume, read its direction rather than its level. It adds a bar's volume to a running total when the close is higher and subtracts it when lower, so the number itself is arbitrary and only the slope and its agreement with price carry information.

On balance volume keeps a running total: add the bar’s volume when the close is higher than the previous one, subtract it when lower. That is the whole calculation, and every property of the indicator follows from its bluntness.

Before you start

A volume feed you trust, because the whole calculation is built on it. In fragmented markets the reported figure covers only part of the trading, which makes the total partial too.

An acceptance that the absolute level is meaningless and only the direction matters. The starting value is arbitrary, so the number has no interpretation on its own.

A decision about what a divergence would have to look like before you act on one. How many bars, how large a disagreement. Deciding afterwards means the chart decides.

The steps

1. Read the slope, never the level

A range-bound stretch of price with a volume histogram beneath it.
The line's direction is the entire output. Illustrative chart - not real market data.

Rising means more volume has occurred on up closes than down closes recently. The number itself depends on where the calculation started and means nothing.

2. Check whether it agrees with price

A slice of price data with participation shown alongside.
Agreement is the ordinary case, not a signal. Illustrative chart - not real market data.

Both rising, or both falling, is the normal state and confirms nothing beyond itself. It is the disagreement that people are looking for.

3. Define a divergence before looking for one

A long-horizon price series with two diverging measures.
A defined divergence is testable; an eyeballed one is not. Illustrative chart - not real market data.

Price making a higher high while this makes a lower high, sustained over a stated number of bars. Without the number, every wiggle qualifies in hindsight.

4. Remember it ignores the size of the move

A slow-moving stretch of price with uniform treatment.
A tiny up close counts the same as a large one. Illustrative chart - not real market data.

A bar closing a fraction higher adds its entire volume to the positive side. That bluntness is the source of most of the noise in the line.

5. Treat a gap with suspicion

A price series where one bar opens well away from the previous close.
A gap assigns a whole day's volume by its close. Illustrative chart - not real market data.

A bar that gapped and then drifted the other way still contributes all of its volume to whichever side the close landed on, which can be the opposite of what happened.

6. Use it as context, not as a trigger

A section of a price series with supporting context.
It is a second opinion, never a first one. Illustrative chart - not real market data.

The setup comes from structure. This can support or question it, and it does not generate entries on its own without producing a great many of them.

7. Take the stop from the chart

The first half of a price series with an invalidation level.
A running total contains no price levels. Illustrative chart - not real market data.

Where the idea is wrong is structural. On this site’s shared series the ninetieth percentile bar range is 1.101, so a stop inside that band is noise rather than a level.

How to tell it worked

Divergence was defined in bars and magnitude before the chart was examined.

0 trades came from the indicator alone, without a structural setup.

The level was never quoted as a number, only its direction.

And the volume feed was checked once, so the input is known to cover the real market.

What the calculation throws away

A candlestick chart annotated with the round-trip cost of a switch.
Every signal acted on costs a round trip. Illustrative chart - not real market data.

The size of every move. Only the sign of the close matters, so a market that grinds up a fraction each day on heavy volume looks identical to one advancing strongly.

A section of a price series drawn without volume context.
And a thin market produces a line built from very few trades. Illustrative chart - not real market data.

Where inside the bar the volume traded. All of it is assigned by the close, which is one data point from a whole session.

Why divergence is weaker than it looks

A divergence can persist for a long time. Price and the line can disagree for thirty bars before anything resolves, and there is no reading that says how long is too long.

It resolves in both directions. Sometimes price falls to meet the line; sometimes the line rises to meet price, and nothing about the divergence itself indicates which.

And it is defined in hindsight unless you write the rule first. Any two series will disagree somewhere over any window, which is why the number of bars has to be fixed in advance for the observation to mean anything.

The one use that holds up

Confirming that a breakout had participation behind it. Price exceeds a level and the line moves decisively with it, which means the volume on the days that carried price were up-close days rather than down-close ones.

That is a weak confirmation and it is honest about being weak. It does not say the breakout will continue; it says the move was not made on a handful of quiet sessions.

The comparison worth making is against the line’s own recent behaviour, not against a threshold. A move in the line that is large relative to the last twenty bars means something; the same move on a different instrument means nothing without that context.

Everything beyond that is asking a signed tally to answer questions it has no inputs for. It knows which side of the previous close each bar finished on and how many shares traded. That is two facts, and most of the interpretations built on this indicator require a third.

The original data

Of the 24,971 unique videos in research/search-study-corpus.jsonl, 26 mention this indicator in the title, at a median of 15,517 views across 24 channels, and 38% of those titles are instruction-shaped. Divergence generally appears in 99 at 9,671 and the money flow index in 44 at 3,160. The counts come from site/corpus_count.py.

A candlestick series with several gaps, the largest of them marked.
A gap is where the calculation is bluntest. Illustrative chart - not real market data.

26 videos at 15,517, one of the lower instruction proportions in this group at 38%. Most of the coverage explains what the line is rather than how to use it, which fits an indicator whose main described application is a pattern that is hard to define.

A stretch of price bars cut short at a decision point.
Price made a new high and the line did not. Short it? Illustrative chart - not real market data.

The answer to the question on that chart is that a divergence is a question rather than a signal. It can persist for dozens of bars and resolve either way — and on this site’s series 54% of 566 ten-bar windows finished higher, so the default outcome does not favour the short.

When it fails

The failure is a divergence found by looking, and it is available on any chart at any time. Two series that both move will disagree somewhere over some window, so scanning for disagreement always succeeds. The pattern gets noticed after price has already turned, which makes it look reliable in review and produces nothing prospectively, because the rule was never written down in a form that could have been wrong.

The second failure is quoting the level. It depends on the start date.

A third is treating it as a measure of buying pressure. It reads the sign of the close.

A fourth is trusting it on a fragmented volume feed. The input is partial.

A fifth is using it as a trigger. It produces far too many.

And a sixth is ignoring how gaps are handled. A whole session gets assigned by one price.

On balance volume covers the calculation. Volume analysis is what the indicator is a small part of. And volume is the input everything here depends on.

What I actually do

The limitation that matters is that it only reads the sign of the close, never the size. A day that closes a hundredth of a percent higher adds the entire volume of that day to the up side. Once I understood that, the indicator stopped looking like a measure of buying pressure and started looking like what it is — a signed volume tally.

— Michael Whitman

This page is educational, not financial advice. Test every idea on your own charts before risking money.