Algorithmic & Systematic Execution
Objectives
- Decompose total execution cost into a quoted-spread component and a market-impact component, and write impact as a function of participation rate.
- Choose between schedule-driven (TWAP/VWAP) and cost-driven (Implementation Shortfall) algorithms by reasoning about the urgency–impact trade-off.
- Map the four SVQ TEP execution modes — PAPER, CONFIRM, AUTO, NOTIFY — onto the signal-to-fill control loop and state what each one assumes about your edge.
In the last chapter you learned to read the order book. Now you have to act on it — and acting is where most of your edge quietly leaks away.
Here is the uncomfortable truth that separates a backtest from a track record. A backtest assumes you transact at a price. The real world makes you buy that price. When you send a large order, you are not a passive observer of the market anymore; you are an event in it. You consume the resting liquidity at the best offer, then the next level, then the next, walking up the book and paying more at each rung. You also tip your hand — other participants infer that someone is buying and adjust their quotes against you. The gap between the price you saw when you decided to trade and the average price you actually achieved is called slippage, and it is the single largest controllable cost in systematic trading. Execution is the discipline of giving as little of it back as you can.
Let us build the intuition first, then make it precise.
Think of every execution as a tug-of-war between two forces. The first force pulls you to trade fast. If your signal says the asset is about to move, every minute you wait is a minute the move can happen without you — that is timing risk, sometimes called opportunity cost. The second force pulls you to trade slow. The faster you demand liquidity, the more of the book you eat and the more you move the price against yourself — that is market impact. Trade too quickly and impact destroys you; trade too slowly and the alpha decays before you are filled. Optimal execution is simply finding the trading speed where the marginal cost of impact equals the marginal cost of waiting. Everything that follows is a way of locating that point.
Now the math, kept honest. Decompose the cost of a single child order into two pieces. The first is the spread cost: by crossing to the other side you pay roughly half the bid–ask spread, so this piece is about one-half times the spread, and it is fixed per trade regardless of size. The second is market impact, which grows with how aggressively you participate. The workhorse model the industry leans on says temporary impact scales with your participation rate, the fraction of market volume you represent. Almgren and Chriss formalized the optimization around this — the framework that trades impact against timing risk — while the empirical microstructure literature (Almgren, Thum, Hauptmann & Li 2005; the Grinold–Kahn tradition; the Tóth–Bouchaud strand) is what pinned down the shape of impact itself. That shape is concave: temporary impact grows roughly with the square root of the rate you trade at. So a common shorthand is that cost in basis points is approximately some constant times the square root of your order size divided by the day's volume, times the asset's volatility. The square root matters enormously. It says doubling your size does not double your impact — it multiplies it by about one-point-four. Slicing a large order into many small children is not bookkeeping; it is the lever that bends a concave cost curve in your favor.
That single fact is why execution algorithms exist. They are rules for slicing a parent order into children over time. Three families dominate.
The first is TWAP — Time-Weighted Average Price — which slices the order into equal pieces across equal time intervals. Simple, predictable, and it makes no assumptions about volume. The second is VWAP — Volume-Weighted Average Price — which slices in proportion to expected volume, trading more when the market is naturally busy so your participation rate stays flat and your footprint stays small. VWAP is the standard benchmark a desk is measured against precisely because matching the volume curve hides you in the crowd. Both of these are schedule-driven: they decide when to trade in advance and ignore price.
The third family is the grown-up one. Implementation Shortfall, the framework Andre Perold named, does not target a schedule at all — it targets the decision price, the price at the instant your strategy said "go." It then explicitly trades off impact against timing risk, front-loading when urgency or alpha is high and stretching out when it is low. This is cost-driven execution, and it is the right mental model for systematic trading, because your signal already tells you how urgent the trade is. A fast-decaying signal wants an aggressive shortfall schedule; a slow, patient edge wants something closer to VWAP.
This is exactly the seam where Saqrid lives, and it is worth seeing how cleanly the abstractions line up.
In SVQ, the TEP Protocol panel is the control surface for the signal-to-fill loop, and it hands the actual order off to SVE, the execution engine, through four modes. PAPER routes the child orders to a simulator — no capital, full slippage accounting — so realized shortfall is measured against the model before a cent is at risk. CONFIRM holds each parent order at the gate until it is approved; the human is the final check on a live fill. AUTO closes the loop entirely — TEP's signal fires, the handoff to SVE happens with no human in the path, and the algorithm slices and works the order on its own. NOTIFY sits beside live trading as the listener: it does not place orders, it raises an alert when conditions or fills cross a threshold. The four modes form a ladder of trust in an edge. The step from PAPER to AUTO is earned only when realized execution cost — measured as implementation shortfall against the decision price SVQ stamped at signal time — stays inside the budget the impact model predicted. When SVE's fills come back inside that envelope, the signal was genuine evidence the asset would move; when shortfall swallows the whole edge, the number was noise dressed as alpha.
The throughline of this entire institute is learning when a number is real evidence that an asset will move. Execution is where that question gets its most ruthless test, because the market charges you to find out. A signal that looks profitable on paper but cannot survive its own market impact was never evidence of anything — it was a measurement of liquidity you did not have. The discipline of measuring shortfall, mode by mode, is how you tell the difference.