Anchored VWAP: The Three Lines Our Engine Removes, and What That Costs
An anchored VWAP is an ordinary volume-weighted average price that starts where you tell it to instead of at midnight. That is the entire difference, and in code it is three lines shorter.
Our engine does not ship one. crates/vike-indicators has exactly one VWAP, batch_vwap, and its parameter list is literally empty — there is no anchor argument to pass. Searching the whole workspace for "anchored" turns up a walk-forward validation mode and nothing else. What follows is the honest version of that: what the engine does compute, the exact sense in which it already computes an anchored line, and where that stops being true.
What anchoring changes
Session VWAP keeps two running numbers: the sum of typical price times volume, and the sum of volume. Divide one by the other and you have the line.
Anchoring changes nothing about that division. It deletes the reset — the three lines that zero both accumulators when a bar starts a new day. Everything else, including the (high + low + close) / 3 typical price, is untouched.
So the two series are not different indicators — they are one indicator with two kinds of amnesia. One forgets everything at 00:00 UTC. The other forgets nothing after the bar you chose, and every behavioural difference below follows from that single deletion.
The function our engine actually ships
batch_vwap lives at base.rs:102 and is seventeen lines long. Per bar it computes the typical price, adds price times volume to one accumulator and volume to the other, and emits the quotient — or nothing at all when volume is zero, which is the right answer rather than a fake one.
The reset comes from a separate place, and it is one line. Columns::from_bars marks a bar as starting a new session when ts / 86_400_000 differs from the previous bar's, which is UTC-day integer division on epoch milliseconds. No exchange calendar, no timezone parameter, no weekly variant. The session VWAP page measures what that boundary does to the line: a median jump of 1.49% of price, and a flipped answer to "is price above the average" on 41.9% of session opens.
The identity hiding in the seed value
Now the part that made this page worth writing, and it comes down to one initialiser.
Columns::from_bars seeds last_day to i64::MIN. No real timestamp equals that, so the very first bar of any series you hand the function is always flagged as a session start. Feed batch_vwap a slice that begins at your chosen bar and it computes an anchored VWAP — exactly, not approximately.
I checked it rather than asserting it. Taking 2,082 anchors from real swing lows across 355 symbols, and comparing batch_vwap(bars[anchor:]) against a plain accumulator started at the anchor: 25,435 of 25,435 values matched bit for bit. Then the first UTC midnight arrives, the engine's copy resets, and over the following week the two drift a median 3.31% apart, 12.01% at the ninetieth percentile.
That is the whole story of the missing feature. The engine can anchor anywhere; it just cannot stay anchored past midnight, because the reset is on a clock rather than on a parameter.
How far apart the two lines get
Drawn side by side on the same bars, the session line and the anchored line separate in a way that depends entirely on the anchor's age.
At one hour old they sit a median 1.58% of price apart, which is mostly noise about a shared starting point. At six hours, 0.77% — they have converged, because both are averaging much the same recent tape. Then the gap widens for good: 3.18% at three days, 5.51% at a week, 13.72% at thirty.
None of that is aesthetic. Across 1.49 million bars the two averages disagreed about whether price was above or below "the" average on 41.0% of them. Two lines, both correctly computed, both called VWAP, giving opposite readings two times in five. When somebody says price reclaimed VWAP, that sentence is underdetermined until they say which one, and most charts do not.
An old anchor goes quiet
Arithmetic, not psychology, is what makes a long-held anchor stop reacting.
Every bar adds its volume to the denominator, so the denominator only grows. A new bar's influence on the average is its share of a total that gets larger every hour, which means the line's own bar-to-bar movement decays as the anchor ages. Measured: the anchored line moved a median 0.3036% of price per hour when one hour old, 0.0611% at a day, 0.0200% at a week and 0.0095% at thirty days — a thirty-two-fold slowdown.
An anchor from last quarter is therefore not a level that reacts to anything you are watching. It is close to a constant, and treating it as dynamic support is treating a horizontal line as though it were listening. I skip anchors older than a few weeks for exactly that reason; a 200-day moving average at least keeps rolling its window forward.
Choosing the anchor honestly
Every anchored VWAP contains one free parameter, and it is the one nobody writes down.
For the figures on this page the anchors are batch_zigzag swing lows at the crate's registered 5% deviation, which at least means a rule picked them. That inherits a caveat worth stating: a zigzag pivot is only written once price has moved 5% away from it, so nobody could have anchored there live on the bar itself. The lag is usually one bar, and it is never negative.
Choosing by eye is the alternative, and its failure mode is the same one that afflicts a fibonacci retracement: the anchor gets picked after the move, from the handful of candidates that make the resulting line look prophetic. When I've gone back through my own charts, the anchors I remembered as obvious were obvious mainly in hindsight. Fix the rule first, then look.
Frequently asked questions
What is anchored VWAP in simple terms? The average price paid per unit of volume since a bar you chose, rather than since midnight. Same arithmetic, different starting point.
Does your engine have an anchored VWAP indicator? No. There is one VWAP function with no parameters at all. Hand it a series beginning at your anchor and it produces the anchored line until the next UTC midnight, which is an identity we verified on 25,435 values, not a workaround we recommend.
Where should I anchor?
Somewhere a rule can find without you. A swing extreme from batch_zigzag, a listing date, an earnings bar — anything defined before the outcome. The volume profile point of control is another level the engine computes without asking your opinion.
Is anchored VWAP better than session VWAP? It is better defined, which is not the same thing. The session boundary is arbitrary on a market that never closes; moving it somewhere deliberate concedes that, and hands you the responsibility instead.
Why does my anchored VWAP not match my platform's?
Check the anchor bar, then the typical price. Ours is (high + low + close) / 3; platforms that use the close alone, or that include the anchor bar differently, compute a different series from identical candles.
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.