Purpose
Money Flow splits every bar's volume into the part that came from identified retail order flow and the part that did not, and shows both as one mirrored column per candle: buying stacks above a hairline baseline, selling below it. You see who was pushing, which way, and how hard.
This is the one indicator in the suite where the buy and sell split is measured rather than inferred. Most volume tools, On-Balance Volume included, decide a bar was "buying" because it closed up, which credits a whole session's volume to one side on a one-cent close. Money Flow reads the aggressor side off the consolidated tape print by print, so a bar that closed green on heavy selling reads as heavy selling.
Best used for:
- Seeing who is on the other side of a move: price rising on retail buying while the other cohort sells is a very different setup from both sides buying.
- Catching absorption: one cohort pushing hard while price refuses to follow means somebody sizeable is taking the other side.
- Timing around exhaustion: a cohort that has been dominant for days thinning out is often the first sign a move is running out of fuel.
- Judging conviction behind a breakout, by whether the volume behind it was signed at all or came in unreadable.
- Sanity-checking any other volume read, because this is the only one here that does not guess direction from the close.
What you see on the chart
A strip along the bottom of the price pane, one column per candle, read against the bars above it rather than in a pane of its own:
- Buying above the baseline, selling below it. Direction is position, never colour, so the reading cannot flip on you.
- Blue is institutional-consistent, amber is retail-identified. Hue carries the cohort and nothing else. Blue against amber is the most robust colour-blind-safe pairing available, so both groups stay legible for everyone.
- Column height is measured volume, so net flow reads as the asymmetry between the two sides while total activity stays visible as mass.
- A slate band centred on the baseline is volume the tape could not sign. A noisy bar visibly muddies rather than quietly vanishing.
- Green and red appear nowhere in this strip, deliberately. Those hues already mean candle direction elsewhere on the chart, and reusing them here invites exactly the "green volume means buyers" conflation the indicator exists to prevent.
- Unmeasured windows draw nothing at all. They do not fall back to showing total volume: that is a different quantity, and putting it on this axis would mislead.
How to read it
- Divergence between the cohorts is the signal. Both sides buying is agreement. One side buying into the other's selling is a transfer, and who wins it usually matters more than the candle.
- A thick slate band is a warning, not noise. It means most of that bar's volume could not be signed, so read the cohort split on it loosely or not at all.
- Sustained is more interesting than large. One heavy column is an event; the same cohort dominating session after session is a campaign, and campaigns tend to resolve in their own direction.
- Retail-identified heavy does not mean wrong. It is a participant label, not a quality judgement. What it tells you is who would be forced to react if price turned.
- Below a small imbalance a cohort reads as balanced, not directional. Do not read intent into a near-symmetric column.
- Confluence rule of thumb: a cohort imbalance that lines up with a dark pool level or an option wall is a zone worth planning around; a lone imbalance in a vacuum is something to watch.
How the engine works
The desk reduces the Alpaca SIP consolidated trade tape, the whole market's prints rather than one venue's slice, into per-minute facts: volume cross-tabbed by venue, by sub-penny price precision, and by aggressor side, with unsignable prints kept as their own outcome rather than folded into a guess. Those facts are stored; the cohort split is applied when you read them, so the method can be revised without re-ingesting a single print.
Retail is identified by the Boehmer, Jones, Zhang and Zhang rule published in the Journal of Finance in 2021: a print is retail-identified when it is off-exchange and sub-penny. Reg NMS Rule 612 bans sub-penny quoting but permits sub-penny execution, so on an off-exchange print that precision is the signature of wholesaler price improvement on retail order flow.
Two limits belong in front of you, not buried in our code. The rule under-identifies: audited against 85,000 real retail trades it catches roughly 35 percent of them, so the retail side is a floor rather than a measurement. Both filters are required: measured on our own tape, sub-penny alone accounts for 4.7 to 11.7 percent of lit volume as well, so the venue filter can never be relaxed.
Trade size is deliberately not used to assign a cohort, because it does not identify institutions. Execution algorithms shred large parent orders, which is why AAPL's median print is around 20 shares, and off-exchange retail prints run larger on average than lit ones, so the obvious inference actually inverts. The legend reports the share carried by block-size prints as a size fact and nothing more. For genuinely institutional block flow, use Dark Pool Levels.
That is also why the non-retail side is called institutional-consistent and never simply institutional: it is the residual left after removing identified retail, so it includes market makers and everything the rule missed.
Key settings
Both repaint instantly; neither recomputes anything:
- Opacity: 10 to 100 percent.
- Legend:
full reading,cohort split only, oroff.
Availability
US equities only, and prints are observed with a 15-minute delay. Not available on crypto charts, which have no consolidated tape, nor on pairs charts, which have no volume to count. When the basic Volume indicator is on as well, it sits in a slice below this strip rather than sharing it. Included from the Plus plan up.
Money Flow is an AI Trading Desk indicator. See it on a live chart in the AI Trading Desk, or browseevery guide.
Educational and analysis purposes only. Not financial advice. See theterms.