Almgren-Chriss Optimal Execution

Almgren-Chriss matters because execution is never just a cost problem. It is always a cost versus risk problem.

The Intuition

If you want to buy a large position, two bad things pull in opposite directions:

• trade too fast and you push the market against yourself
• trade too slowly and you risk adverse price movement before finishing

Almgren-Chriss turns that tension into an explicit optimization problem. The model is not valuable because desks follow its textbook trajectory literally. It is valuable because it gives a clean language for urgency.

That is why the framework survives. It organizes the execution problem even when the exact cost specification is only approximate.

Inventory and Trading Schedule

Let $X_0$ be the initial shares you need to buy or sell. In a discrete trading schedule, choose trades $n_1,\dots,n_N$ over N periods, with remaining inventory

$$ x_k = X_0 - \sum_{j=1}^k n_j, \qquad x_0 = X_0,\quad x_N = 0. $$

The choice variable is the trajectory $x_k$: how quickly inventory is reduced over time.

The Objective

In the classic formulation, choose the schedule to minimize

$$ \mathbb{E}[C] + \lambda \operatorname{Var}(C), $$

where:

• C is total execution cost relative to the decision benchmark
• $\mathbb{E}[C]$ is expected impact cost
• $\operatorname{Var}(C)$ is timing risk from price uncertainty during execution
• $\lambda$ is risk aversion or urgency

The variance term depends on volatility and remaining inventory over the schedule, so higher volatility makes slow execution more expensive in risk terms.

This is the whole economic idea.

• low $\lambda$: patient execution, less temporary impact, more timing exposure
• high $\lambda$: front-loaded execution, more immediate impact, less timing exposure

The model is therefore not merely "optimal execution." It is a knob that makes the speed-risk trade-off explicit.

Temporary and Permanent Impact

The textbook model separates two effects.

Temporary impact

This is the price concession paid because you trade quickly right now. It vanishes after the trade is completed.

Permanent impact

This is the part of the price move that remains after your trade because the market re-anchors to a new level.

In a simple linear-cost version, the schedule-dependent part of expected cost has terms like

$$ \sum_{k=1}^N \eta n_k^2, $$

where $\eta$ controls temporary impact. Permanent impact is often modeled separately as a schedule-invariant shift in expected cost for a fixed order size. The exact algebra differs by normalization, but the interpretation is stable:

• larger $\eta$ penalizes fast concentrated trading
• larger $\sigma$ raises timing risk
• larger $\lambda$ makes the schedule more urgent

What the Optimal Schedule Means

You do not need the closed-form solution memorized to use the model well.

What matters is the direction of the comparative statics:

• higher volatility -> execute faster
• higher risk aversion -> execute faster
• higher temporary impact -> execute more slowly
• longer available horizon -> execute more slowly

That is the practical content.

If your chosen schedule does not move the way these inputs suggest, either the parameters are wrong or the model is being applied outside its useful range.

A Worked Scenario

Suppose two desks each need to buy the same position.

Scenario A: calm market

• low volatility
• stable spread
• weak urgency

The model favors a smoother schedule. Inventory declines gradually because timing risk is modest and there is value in avoiding concentrated impact.

Scenario B: stressed market

• high volatility
• greater risk aversion

Now the optimal trajectory front-loads. The desk accepts more immediate impact to reduce the risk of still holding large inventory while the market moves violently.

That is the right way to read Almgren-Chriss. Not as "there is one optimal curve," but as "the inventory path changes systematically when the cost-risk environment changes."

Why the Model Is Still Useful When Literally Wrong

Real execution is messier than the textbook setup:

• impact can be transient rather than purely permanent
• liquidity varies intraday
• alpha decays during the order
• fills depend on queue position and venue choice

So why keep the model?

Because it still answers the first-order question:

should this order behave more like a patient schedule or more like an urgent schedule?

That is often the right level of abstraction for research and pre-trade analysis.

Where It Breaks

Almgren-Chriss becomes fragile when:

• volatility inputs are stale
• impact coefficients are estimated from the wrong regime
• the order interacts with strong intraday seasonality or event risk
• correlated order flow or crowding violates the single-agent logic
• the execution problem is dominated by queue dynamics rather than smooth participation

The model should therefore be used as a scenario tool, not as an oracle.

In Practice

Use these rules:

• treat $\lambda$, volatility, and impact coefficients as regime-dependent inputs
• study how the trajectory changes under parameter stress, not only at one point estimate
• use the model to frame urgency and feasibility, even if the live desk re-plans continuously
• compare its implications against simpler benchmarks such as TWAP or participation caps
• remember that queue effects and alpha decay can dominate the textbook optimum

Common Mistakes

• Treating the closed-form schedule as universally optimal rather than model-conditional.
• Using stale volatility or impact estimates.
• Ignoring the difference between temporary and permanent impact.
• Reading $\lambda$ as a free tuning trick instead of an economic urgency parameter.
• Forgetting that execution models are judged by realized implementation outcomes, not elegance.

Connections

This primer supports Chapter 18's execution framework and schedule design. It connects directly to implementation shortfall, square-root impact, capacity analysis, and the broader question of how signal decay and market risk should change execution urgency.