VWAP: The Session Reset, and What Crossing It Actually Returned
VWAP is the volume-weighted average price of everything traded so far in the current session. Two words in that sentence do the work — volume-weighted, which most explanations cover, and so far, which almost none of them do.
Where the counter goes back to zero is what gives the indicator its character. Below I name the exact code in our trading indicators engine, state what happens at a session boundary down to the last bit, and measure 1,816,661 real crossings on 8.9 million hourly bars.
What VWAP actually is
Take each bar's typical price, weight it by that bar's volume, and divide the accumulated total by the accumulated volume. Nothing else happens.
Our detector batch_vwap keeps exactly two numbers between bars: a running sum of price times volume, and a running sum of volume. That pair is the entire state. Each bar it adds (high + low + close) / 3 multiplied by the bar's volume to the first, adds the volume to the second, and divides. When a bar trades nothing at all, the denominator is zero and the engine emits no value rather than a fake one.
That structure makes VWAP unlike every average on the golden cross page. A 50-bar mean forgets the fifty-first bar. This forgets nothing — until it forgets everything at once.
The session reset, exactly as the engine does it
Here is the line that defines the indicator, and it is three lines of code rather than a philosophy.
The engine marks a bar as starting a new session when its timestamp, divided by the number of milliseconds in a day, differs from the previous bar's. On such a bar both accumulators are set to zero before the bar is added. There is no blending, no partial decay and no configurable anchor: the boundary is UTC midnight, and the old session's volume is gone the instant the clock rolls over.
I measured that consequence rather than describing it, and it is not small. Across 371,113 sessions the line jumped a median 1.49% of price at the boundary, and 5.01% at the ninetieth percentile. On 41.9% of session opens, the reset alone flipped the answer to "is price above the average?" — same price, same tape, a different verdict, purely because a new day began.
That is also why a VWAP drawn on a daily chart is meaningless. One bar per session makes the accumulator a single term, so the line lands on that bar's own typical price and stays there forever. Every screenshot of a daily VWAP is a plot of (high + low + close) / 3 wearing somebody else's name.
What VWAP is on the first bar of a session
On the opening bar the arithmetic collapses.
Divide tp × volume by volume and the volume cancels: VWAP equals that bar's typical price, exactly. I checked this on all 371,038 session opens in the sample — 90.12% matched to the bit, and every one of the remaining 36,664 was within a single unit in the last place, maximum 1.0 ulp. That residue is floating-point rounding failing to invert its own multiplication, not a different number.
So on the first bar of any session, "price is above VWAP" reduces to "the close sits in the upper part of this one bar". It knows nothing about the day, because nothing about the day has happened. The scan bears that out: bars in that position resolved 47.8% long against 49.5% five hours later, the weakest slot in the session.
What crossing the VWAP actually returned
Crossing the line is the most-traded rule here, so that is what I scored, on a bracket built to be readable.
Every close that moved from one side of the line to the other counted as a signal — 1,816,661 of them across 355 pairs, which is 4.9 crossings per session. Each was resolved first-touch over the next 24 hours against a symmetric 2% bracket, so break-even sits at exactly 50% before fees. A bar tagging both levels was scored a loss.
Crosses up resolved 49.0%. Crosses down, scored short, resolved 49.5%. Then the number that settles it: applying that identical bracket to every eligible bar returns 48.9% long and 49.8% short, so the upward cross beat doing nothing by 0.1 points and the downward one trailed it by 0.3. Both sit inside the noise of a sample this size, and both sit under break-even before a single fee.
Is price above VWAP bullish?
This is the claim the indicator is mostly sold on, and it splits in two once you measure it two ways.
On the bracket above, bars closing over the line resolved 49.1% long across 3,807,267 observations, against 48.7% for bars closing under it — a real but tiny separation, and both under the 50% break-even. On forward return the ranking inverts: the median 24-hour return from above the line was −0.188%, while from below it was 0.000%. A rule cannot be both the better side to buy and the worse side to hold; what it can be is uninformative, measured twice.
There is a reason this feels more predictive than it is. Price above the session average usually means the session has already gone up, so the state and the thing it is claimed to forecast are the same fact separated by a few hours.
Where VWAP fails
Two real crossings, both scored by the rule above, both losing money in the ways this indicator loses money.
One is the reclaim that settles nothing. Price crosses back over the line, the chart looks decisive, and the average is dragged up behind it fast enough that the level offers no support when the next push down arrives.
Another is the boundary artefact. A short taken on a lost VWAP late in a session inherits a line built from an entire day of volume; the same trade an hour later faces a line rebuilt from almost nothing. I've watched the identical setup grade differently on either side of midnight, and no amount of confirmation from a candlestick reversal fixes it, because the disagreement is with the clock rather than with the tape.
Anchoring is the usual answer, and it is a genuine improvement over the default — an anchored VWAP starts from an event you chose rather than from a timezone. Note what that concedes: the boundary was always arbitrary, and moving it somewhere defensible is admitting so.
When I look at one of these trades now, the clock is the first thing I check, before the level.
Frequently asked questions
What is VWAP in simple terms? Roughly, the average price paid per unit of volume since the session began. Big trades pull it more than small ones, and it starts over at UTC midnight.
Is VWAP a good indicator? As a fair-value reference for execution it is excellent, which is what institutions use it for. As a directional trigger it did not clear its own base rate here: crossings resolved 49.0% and 49.5% against 48.9% and 49.8%.
What is the difference between VWAP and a moving average? Two things. VWAP weights by volume, and it resets — a 20-period average carries a fixed window forward forever, while this carries the whole session and then nothing.
Why does my VWAP look different from my broker's? Almost always the anchor. Ours resets on the UTC day; platforms that reset on an exchange session, or that anchor to a swing low, are computing a different series from the same candles.
Does VWAP work on crypto? The arithmetic does. The session does not: a market that never closes has no natural boundary, so the UTC reset is a convention imposed on a tape that has no opinion about it. The MACD page covers a related indicator whose lag is at least honest about being arbitrary, and the RSI page covers one whose value depends on where your history starts.
This is educational material, not financial advice. Every figure here was measured on past bars, past behavior generalizes poorly to future bars, and trading carries real risk of loss — size any position so that being wrong stays survivable.