Function vwap

Volume-Weighted Average Price over a trailing window of period bars:

VWAP[i] = Σ typicalPrice · volume / Σ volume      over bars i−period+1 .. i
typicalPrice = (high + low + close) / 3
Copy

The average price actually traded over the window, as opposed to the average of the bars' prices: a bar on ten times the volume moves it ten times as far. Appends one column; undefined for the first period − 1 rows, like every other count-window study.

Composed from two kernels: typicalPriceValues (shared with CCI, MFI and Keltner when they land) and rollingWeightedMeanValues (shared with a volume-weighted moving average, which is the same kernel over close).

Which VWAP this is — a design decision, not a lookup

There is no single VWAP and no TA-Lib reference to defer to. Two definitions are in real use:

1. Rolling — the ratio of sums over the last period bars. This is what ships. It has the shape every other study in this package has (bar-count window, length-preserving warm-up, no state), it is what a charting package means by "VWAP with a period", and it composes with the rest of the vocabulary without a new mechanism.
2. Anchored — the same ratio accumulated from a fixed bar (the session open, most often) with no window. That is the VWAP of the intraday execution desk, and it is not a special case of (1): a count window is emitted only once it spans period rows, so period = length yields one value at the last bar, not the running line. An anchored VWAP is a ratio of two running sums (cumulativeValues, already a kernel) and is only useful with a reset — which makes it a session-anchored study, and it is deferred to that phase rather than shipped here without one.

period is required for the same reason: a rolling VWAP has no conventional length the way RSI has 14, and a silent default would be an invented one.

Inputs

Four, each named by an option defaulting to its DEFAULT_OHLCV name — the rule atr established, applied once more. The price weighted is the typical price, the usual convention. There is deliberately no price option to choose another: it would be a knob with one conventional value. A close-weighted VWAP is available without it — point high and low at the close column too and the typical price is the close — and if that shape earns a name it should be a vwma study on the same kernel, not a mode here.

Definition, verified

Checked bar-for-bar against a pandas replication ((tp·v).rolling(n).sum() / v.rolling(n).sum()) in the oracle fixture, on a volume series with several bars at five to nine times the typical volume — so that a VWAP which dropped the weighting is distinguishable from the mean of typical prices (the generator asserts the two differ by at least a quarter of a price unit somewhere in the fixture). Agreement to 1e-9.

Edges

• A window with no volume at all (Σ volume = 0) → undefined. There is nothing to weight by, so there is no answer; not 0, and not the plain mean. (It is 0 / 0 in the kernel, which needs no guard — see rollingWeightedMeanValues for why one would be dead code.)
• A gap in any input drops that bar from both sums, numerator and denominator alike, so the window's answer is the VWAP of the bars it does have — rather than a number quietly biased toward zero by a volume whose price went missing. As with sma, a window is emitted once it spans period rows, computed from whichever of them are present.
• Scales with price, not with volume: doubling every price doubles it; doubling every volume leaves it unchanged. Both are pinned as property tests.