The Volume-Timing Profile Inside One Bar, and Why a Time Share Is Misleading

A forming bar tells you how much time is left before it closes. It does not tell you, on its own, whether the volume that has already arrived is normal for this point in the bar or unusual. A common shortcut is to take the volume of recent closed bars, multiply by the share of time elapsed, and compare that against the running total. The number looks precise. It is wrong for a structural reason worth understanding before building anything on top of it. A time share treats every minute of a bar as equal Dividing elapsed time by bar duration and multiplying by average volume assumes volume arrives at a constant rate through the bar. On almost any instrument that trades in a session, it does not. The first minutes after open, the window where London and New York trade at the same time, and the minutes bracketing a scheduled release all pull in far more than their even share of a bar's volume; the middle of a quiet bar barely moves at all. A flat time-share figure blends these two regimes into one number that matches neither. Ten minutes into a fifteen-minute bar that opened during the overlap, a time share already expects two-thirds of the volume to be in. The actual shape of that bar's history says otherwise. The fix is a per-minute profile, not a flat rate The unevenness lives inside the smallest bars already available. Each one-minute bar sits inside exactly one bar of the higher timeframe, and where it sits inside that bar, the second minute in, the fifth, the very last, comes straight from its own timestamp. Instead of one number for the whole bar, build one reference per minute offset: for minute five of a fifteen-minute bar, look at what had accumulated by minute five in every finished bar of a history window, and use that as the comparison point for minute five specifically. Minute one gets its own reference, minute fourteen gets its own. The result is a profile shaped like the session actually behaves rather than a straight line from zero to the bar's average. Median over average keeps one busy day from lying for weeks Building that profile from an average introduces a second, separate mistake. A single release day that produces several times the normal volume in one bar pulls that minute's average well above where an ordinary session sits, and keeps pulling it for as long as the history window still covers that day. A median does not move that way: the middle of a sorted sample stays close to where the ordinary bars are, and one outlier day changes the answer only once outliers make up a large share of the sample. This is the mechanism behind the reference figure in a free panel called Candle Clock Desk, and it applies to any per-minute volume profile, not only that one. Small samples need a floor before they are trusted A profile built from too few finished bars is not merely conservative, it is fragile. A median of two samples is just whichever of the two happens to be larger; a median of five is barely more stable. Any implementation of this idea needs a minimum bar count per minute offset before it reports a number at all, and it needs to say plainly when that minimum has not been reached rather than quietly showing a figure built on a thin sample. Treating "not enough history yet" as its own state, separate from "quiet", keeps noise from being read as a signal. Reading the ratio without turning it into a signal Once the profile exists, comparing the current bar's running volume against the reference for the current minute gives a ratio. A ratio near 1.0 means the bar is filling the way it normally does at this point; well above 1.0 means unusually busy; well below means unusually quiet. The ratio describes participation, not direction. Nothing in the construction says whether the volume is buying or selling, and treating an elevated reading as a trade signal on its own reintroduces the same assumption the time-share method made: that one number carries more meaning than it does. Where it is useful is as context. An elevated ratio next to a scheduled release on the calendar, or next to a visible jump on the tape, corroborates what the two already show separately. An elevated ratio with nothing on the calendar and no visible move is worth noting and not much more. One practical limit follows directly from where the data comes from. The profile needs sub-bars to build its minute-by-minute reference points, so it has nothing to compute against on the timeframe that is itself the smallest unit; a per-minute profile of a one-minute bar has exactly one minute in it. Every timeframe above that has M1 bars to break itself into, and the more of them a broker keeps in history, the further back the reference window can reach.