The Intraday Volume Profile

Plot the average fraction of a stock's daily volume that trades in each 5-minute bucket, averaged over many days, and you get a shape that's remarkably consistent across names and markets: a spike right at the open as overnight orders and price discovery flood in, a long gentle sag through the middle of the day as activity quiets down, and a second spike into the close as index funds, closing-price benchmarked orders, and end-of-day rebalancing concentrate. This is the U-shaped intraday volume profile, and it's one of the most reliable regularities in market microstructure.

The shape matters directly for execution: an algorithm trying to minimize its market impact wants to trade more when the market is naturally trading more (because a given order size is a smaller fraction of a busy market's volume, so it moves the price less) and less when the market is quiet. VWAP (volume-weighted average price) execution algorithms are built explicitly around this profile — they schedule an order's child slices to match the expected fraction of volume in each period of the day, so that if the historical profile says 8% of the day's volume typically trades in the first 15 minutes, the algo aims to have traded roughly 8% of the parent order by then too.

Worked example. A stock's historical volume profile says the first 30 minutes of trading typically carry 12% of the day's volume, the middle 5 hours carry 55%, and the last 30 minutes carry 10% (the rest spread across the remaining periods). A portfolio manager needs to buy 100,000 shares over the full day using a VWAP algorithm. The algo's schedule targets roughly $100{,}000 \times 0.12 = 12{,}000$ shares in the first 30 minutes, and continues to allocate the remaining shares proportionally to the expected volume in each subsequent bucket, re-forecasting as the day's actual volume comes in and deviates from the historical average.

Second example, an earnings day. The same stock reports earnings before the open. The historical U-shaped profile is no longer a good guide — volume in the first 30 minutes might be 35% of the day's total instead of the usual 12%, because of the concentrated reaction to news. A VWAP algorithm using a static historical profile on an earnings day systematically under-trades early and is forced to catch up later, often trading more aggressively (and at worse prices) into the close than intended. Good execution systems detect this and either switch to a different benchmark or use a dynamically re-estimated profile blending the historical shape with the day's realized volume so far.

What this means in practice

Every desk that reports execution quality against a VWAP benchmark is implicitly relying on this profile — deviating from it (trading too much too early, or falling behind and dumping shares near the close) is exactly what shows up as underperformance against VWAP. It's also relevant to signal research: a feature computed from volume needs to be normalized against the expected volume-at-time-of-day, or it will just be re-discovering the U-shape rather than anything about the specific stock.

See TWAP, VWAP & POV for how the profile gets used directly in scheduling an order, and Optimal Execution for how it interacts with market-impact minimization more generally.