"Slippage and Market Impact: The Almgren-Chriss Framework for Practitioners"

Ask a trader what slippage is and you will hear something like "the difference between the price I wanted and the price I got." True, but incomplete in a way that costs money. Slippage is not one thing; it is the sum of several distinct costs with different causes, different scaling laws, and different remedies. Practitioners who lump them together end up optimizing the wrong variable. This article separates the components and then introduces the framework — Almgren and Chriss (2001) — that made execution cost a solvable engineering problem rather than a lament.

Decomposing "the price I got"

Take a decision to buy at the moment the mid-price is 5000.00, and a final average fill of 5000.85. Where did the 0.85 go?

The half-spread. Crossing from mid to the ask is the fee for immediacy — unavoidable for aggressive orders and analyzed in depth in our article on spread economics.

Temporary impact. Your order consumed depth; the book repriced against you while you traded, then partially recovered after you stopped. This component is mechanical — it exists because liquidity at any instant is finite — and it scales with how fast you demand liquidity relative to what the market replenishes.

Permanent impact. Some of the price move does not revert, because the market read information in your flow. Other participants cannot distinguish your rebalancing from someone else's insight, so persistent one-sided pressure shifts the consensus price itself.

Drift. While you worked the order, the market moved for its own reasons. Over a short execution this is noise; over a long one it is risk — and this is precisely the tension the framework below formalizes.

Empirically, impact is strongly nonlinear. A robust finding across decades of institutional data is the approximate square-root law: expected impact grows roughly with the square root of order size relative to typical volume. Doubling your order does not double your impact cost — it multiplies it by about 1.4 — but the total dollars paid grow faster than linearly. The practical reading: size, not frequency, is what makes execution expensive, and impact costs are the binding constraint on how much capital any given strategy can run. This is the microstructural root of the truism that strategies "hit capacity."

The Almgren-Chriss insight: execution is a risk-return problem

Before 2001, execution advice was folklore. Almgren and Chriss reframed it as optimization. Trade fast and you pay impact with certainty; trade slowly and you pay less impact but accept variance — the market may run away while you work. Impact cost is deterministic-ish and front-loaded; timing risk is stochastic and grows with duration.

Their contribution was to write both as functions of the trading trajectory — how much remains unexecuted at each moment — and solve for the family of trajectories minimizing cost for each level of risk tolerance. The result is an efficient frontier of execution, exactly analogous to Markowitz's frontier for portfolios:

• A risk-neutral trader (or one whose signal is slow) trades on a near-linear schedule, minimizing expected impact.
• A risk-averse trader (or one holding fast-decaying information) front-loads: pay more impact early to shrink exposure to drift.
• The curvature of the optimal schedule is set by a single interpretable parameter — the ratio of risk aversion times variance to impact cost.

The model's assumptions are stylized (linear impact, arithmetic price dynamics), and successors have refined every piece. But its enduring value is conceptual: there is no "best execution" in the abstract — only best execution for a stated risk preference and a stated alpha horizon. A desk that cannot articulate those two inputs cannot evaluate its own fills.

From theory to a measurement discipline

The framework earns its keep only when fed with data, which makes measurement the real practitioner's problem. Three requirements:

Timestamp the decision, not the order. Impact and shortfall are defined against the price when the signal fired. If your infrastructure only logs orders, the most important benchmark in transaction-cost analysis is unrecoverable. (This is why GIDEON's audit trail roots every order chain at the originating signal event — the design makes arrival-price analysis a query rather than a reconstruction.)

Separate reverting from permanent moves. Sample the price at fixed horizons after completion (one minute, thirty minutes). The reverting portion was temporary impact — a cost your schedule can reduce. The persistent portion is information leakage — a cost your order-placement style must address.

Aggregate before concluding. Any single execution is dominated by noise. Impact models are statements about averages across hundreds of orders; judging a schedule change on a week of fills is astrology.

What futures traders should take away

In CME futures the framework applies cleanly — single central book, deep visible liquidity, well-behaved volume curves — with one sharpening: because the most liquid contracts trade at a one-tick spread with deep queues, the passive alternative (joining the queue instead of crossing) is unusually attractive, and the real optimization is often between paying impact now and accepting queue risk instead of drift risk.

The summary discipline: know your components, know your scaling law, choose your point on the frontier deliberately — and instrument your pipeline well enough to know whether reality agrees.

References

• Almgren, R. & Chriss, N. (2001). "Optimal Execution of Portfolio Transactions." Journal of Risk, 3(2).
• Almgren, R., Thum, C., Hauptmann, E. & Li, H. (2005). "Direct Estimation of Equity Market Impact." Risk, 18(7).
• Bouchaud, J.-P., Farmer, J.D. & Lillo, F. (2009). "How Markets Slowly Digest Changes in Supply and Demand." In Handbook of Financial Markets. Elsevier.
• Perold, A. (1988). "The Implementation Shortfall: Paper Versus Reality." Journal of Portfolio Management, 14(3).