U.S. equities traded an average of 20.5 billion shares a day in March 2026, up 27.8% year over year, against average daily notional of roughly $1.1 trillion across 2025. Every one of those shares travelled a path that starts inside a research repository and ends inside a matching engine in New Jersey. Most quants can describe the first step and the last one. The eight or so handoffs in between are where the paper alpha gets spent, and they are the part of the job that almost nobody teaches.
This is the full route: signal, target portfolio, order, algorithm, router, protocol, matching engine, tape, clearing house.
For daily Quant articles and market breakdowns:
An alpha model outputs expected returns or scores. It does not output trades. The optimizer sits between the two, and it takes the alpha vector, a risk model, a transaction cost model, and a constraint set, then solves for a target holdings vector. The trade list is the difference between yesterday’s positions and today’s target.
That distinction is where most of the practical decisions get made. Turnover penalties, sector neutrality, beta neutrality, borrow availability, and position limits bind long before signal strength does. A researcher who improves an information coefficient from 0.03 to 0.035 and watches the live P&L stay flat has usually discovered that the constraint set was already binding on those names.
Inside a multi-strategy platform the trade list then hits a second layer. Pods at Millennium, Citadel and Balyasny route through a central risk book that nets offsetting interest across teams before anything reaches the street. Two pods trading the same name in opposite directions pay zero external cost when the internal cross fires, which is a structural edge that a single-strategy fund cannot replicate.
The order management system applies the controls that keep the broker-dealer out of enforcement. SEC Rule 15c3-5 requires pre-trade credit and capital thresholds, duplicate order checks, fat-finger price and size bands, and hard blocks on restricted names, all applied before the order leaves. Short sales need a locate under Reg SHO. Nothing passes on a post-trade basis.
These checks cost time. A software risk layer adds microseconds, an FPGA-based one adds nanoseconds, and firms competing in latency-sensitive strategies pay for the hardware because the alternative is losing the race to the queue.
A portfolio manager sends a parent order for 400,000 shares. The execution algorithm decides how that becomes several thousand child orders across the day.
The scheduling choice is a trade-off between market impact and timing risk. Trade fast and pay impact. Trade slow and carry exposure to price moves that have nothing to do with the signal. The standard square-root law puts temporary impact at roughly the daily volatility multiplied by the square root of participation, so an order equal to 10% of ADV in a name with 2% daily volatility costs on the order of 60 basis points before spread and fees.
The common schedules:
VWAP and TWAP, which target a benchmark and accept whatever the day delivers
Percentage of volume, which caps participation at a fixed rate and stretches the horizon
Implementation shortfall, which front-loads execution when the alpha decays fast
Liquidity seeking, which sits passively and takes size when it appears
The choice of schedule is a statement about signal half-life, and most execution desks never receive that number from research. A signal with a two-day half-life executed evenly across three days captures roughly 62% of the alpha on paper. The same signal executed in half a day captures over 90% and pays several times the impact. That arithmetic decides the schedule, and it can only be done when the researcher hands over the decay curve.
The child order now faces a fragmented venue map. More than a dozen registered exchanges, roughly thirty active alternative trading systems, and a handful of wholesalers, all quoting the same securities.
Comments
Nothing yet. Say the first thing.
Sign in to join the conversation.