Quant Academy Math · Probability · Quant Finance
Phase 16 - Lesson 16.4

Optimal Execution (the Almgren–Chriss Framework)

Choosing a trade schedule that balances the cost of trading fast against the risk of trading slow.

⏱ 60 min● Advanced🔗 Prereqs: 16.3
↖ Phase 16 hub
Builds on: 16.3 supplied the temporary/permanent impact costs this framework trades off against risk.
Leads to: 16.5 turns the dealer’s side into a market-making problem with inventory risk.

Learning Objectives

Click a status chip to cycle: Not started → In progress → Studied → Practiced → Needs review → Mastered.

Key Vocabulary

Optimal execution
Choosing how to split a parent order over time to minimize expected cost plus a penalty on its risk.
Almgren–Chriss
The canonical mean–variance execution model with linear permanent/temporary impact and price risk.
Trading trajectory
The planned remaining holdings \(x_k\) (or trade sizes \(n_k\)) over the execution horizon.
Timing risk
Variance of cost from price moves while the position is still open; grows the slower you trade.
Risk aversion \(\lambda\)
The weight on cost variance; larger \(\lambda\) means more urgency (faster liquidation).
Efficient frontier
The set of trajectories with minimal expected cost for each level of cost variance.
TWAP
Time-weighted average price: trade equal amounts each interval; the \(\lambda\to0\) Almgren–Chriss solution.

Intuition & Motivation

Intuition
You must liquidate \(X\) shares by time \(T\). Trade it all at once and you pay huge temporary impact; trade it slowly and you sit exposed to price moves that could go against you. Almgren–Chriss makes this precise as a mean–variance problem: minimize expected execution cost plus \(\lambda\) times the variance of that cost. The expected-cost term rewards trading slowly (less impact); the variance term rewards trading quickly (less exposure). The solution is a smooth exponential liquidation whose speed is set by your risk aversion \(\lambda\): patient traders approach a straight-line TWAP, urgent traders front-load. There is no single best schedule - only an efficient frontier, and you pick your point on it by how much timing risk you can stomach.

Setup

Liquidate \(X\) shares over \(N\) intervals of length \(\tau=T/N\). Let \(x_k\) be shares remaining after step \(k\) (so \(x_0=X,\ x_N=0\)) and \(n_k=x_{k-1}-x_k\) the shares sold in step \(k\). With permanent impact \(\gamma\), temporary impact \(\eta\), and price volatility \(\sigma\), the expected shortfall and its variance are

\[\E[\mathcal{C}]=\tfrac12\gamma X^2+\sum_{k=1}^{N}\frac{\eta}{\tau}\,n_k^2,\qquad \Var[\mathcal{C}]=\sigma^2\sum_{k=1}^{N}\tau\,x_k^2.\] (16.7)

The permanent term \(\tfrac12\gamma X^2\) is schedule-independent (it depends only on total size). The temporary term \(\sum \eta n_k^2/\tau\) punishes trading fast; the variance term \(\sigma^2\sum\tau x_k^2\) punishes holding inventory long. Everything interesting is the tension between the last two.

The mean–variance objective and the frontier

Choose the trajectory to minimize

\[\min_{x_1,\dots,x_{N-1}}\ \E[\mathcal{C}]+\lambda\,\Var[\mathcal{C}],\] (16.8)

As \(\lambda\) sweeps from \(0\) to \(\infty\), the optimal trajectories trace the efficient frontier: each is the minimum-variance strategy for its expected cost (or vice versa). \(\lambda=0\) is the risk-neutral trader who only minimizes impact; large \(\lambda\) is the risk-averse trader who liquidates fast to cut exposure.

Key Idea
The Almgren–Chriss trade-off in one line: trade fast to cut timing risk, trade slow to cut market impact. Risk aversion \(\lambda\) picks the balance; the answer is a schedule, not a price.

The optimal trajectory is exponential

Minimizing the quadratic (16.8) gives a linear recursion whose solution is a hyperbolic/exponential decay of the holdings:

\[x_k=X\,\frac{\sinh\!\big(\kappa(T-t_k)\big)}{\sinh(\kappa T)},\qquad \kappa=\sqrt{\frac{\lambda\sigma^2}{\eta}}\ \ (\text{continuous-time limit}),\] (16.9)

The urgency \(\kappa\) grows with risk aversion \(\lambda\) and volatility \(\sigma\), and shrinks with temporary impact \(\eta\): more risk or vol → liquidate faster; more impact → stretch out. Two limits are illuminating:

Worked Example - Reading the exponential schedule
1
With \(\kappa T=1\) you liquidate faster than linear early on: after half the horizon \((t=T/2)\) you hold \(x=X\sinh(0.5)/\sinh(1)=X\cdot0.521/1.175\approx0.44X\).
2
A pure TWAP would hold exactly \(0.5X\) at the half-way point - so positive risk aversion has you ahead of schedule (less inventory outstanding).
3
Raise \(\lambda\) (or \(\sigma\)): \(\kappa\) rises, the curve bows further toward the origin - more front-loading, more impact paid to buy down risk.
4
Raise \(\eta\) (impact): \(\kappa\) falls, the curve straightens toward TWAP - expensive impact makes patience worth the risk.

Interactive: execution simulator

Compare front-loaded, TWAP, and back-loaded schedules for the same parent order and see the realized cost distribution.

Interactive: Almgren–Chriss trajectory and expected cost

Common Mistakes to Avoid
  • Believing there is one optimal schedule: the output is an efficient frontier; you must choose \(\lambda\).
  • Setting \(\lambda=0\) by default: that ignores timing risk and gives TWAP, which can be far too slow for a risky or alpha-bearing order.
  • Forgetting permanent impact is schedule-independent here: it cannot be optimized away by re-timing (in the linear model).
  • Treating the frontier as static: volatility, impact, and alpha change intraday, so the optimal urgency changes too.
Quant Practitioner Tips
  • Map \(\lambda\) to an intuitive urgency/half-life; clients understand ‘finish 63% by time \(1/\kappa\)’ better than a risk-aversion number.
  • Add an alpha/drift term when the order carries a signal: expected price drift tilts the schedule (front-load to capture decaying alpha).
  • Re-solve intraday as volume and volatility update; static schedules leak P&L in changing conditions.
  • Benchmark realized execution against the model’s predicted cost and variance to calibrate \(\eta,\gamma\).

Knowledge Check

Q1 Medium
The central trade-off in the Almgren–Chriss framework is between:
Commissions and taxes
Market impact (from trading fast) and timing risk (from trading slow)
Bid and ask prices
Permanent and temporary impact only
Q2 Medium
As the risk-aversion parameter \(\lambda\to0\), the Almgren–Chriss optimal trajectory becomes:
Immediate full liquidation
A straight line - equal slices per interval (TWAP)
Back-loaded to the final interval
Undefined
Q3 Hard
In the exponential solution, the urgency \(\kappa=\sqrt{\lambda\sigma^2/\eta}\) increases when:
Temporary impact \(\eta\) increases
Volatility \(\sigma\) or risk aversion \(\lambda\) increases
Total size \(X\) increases
The horizon \(T\) increases

Practical Exercise

You must liquidate \(X=1{,}000{,}000\) shares over one day. (a) Write the mean–variance objective you are minimizing and identify which term pushes you to trade fast and which to trade slow. (b) Explain qualitatively how the optimal schedule shifts if intraday volatility doubles. (c) Your desk wants ‘the optimal schedule.’ Explain why that request is under-specified.

▶ Show full solution

(a) Minimize \(\E[\mathcal{C}]+\lambda\Var[\mathcal{C}]=\big(\tfrac12\gamma X^2+\sum_k\tfrac{\eta}{\tau}n_k^2\big)+\lambda\sigma^2\sum_k\tau x_k^2\). The temporary-impact term \(\sum \eta n_k^2/\tau\) grows when trades \(n_k\) are large, so it pushes you to trade slowly (spread the \(n_k\)); the variance term \(\lambda\sigma^2\sum\tau x_k^2\) grows when holdings \(x_k\) stay large, so it pushes you to trade fast (shed inventory).

(b) Doubling \(\sigma\) multiplies the urgency \(\kappa=\sqrt{\lambda\sigma^2/\eta}\) by \(2\), bowing the trajectory more toward the origin: you front-load, liquidating faster to reduce exposure to the now-riskier price, accepting more temporary impact in exchange.

(c) ‘The optimal schedule’ is under-specified because optimality is only defined given a risk aversion \(\lambda\) (a point on the efficient frontier). Without stating how much timing-risk variance the desk will trade for lower expected cost - or an equivalent urgency/deadline - every trajectory from TWAP to immediate is ‘optimal’ for some \(\lambda\).

After the reveal, answer for yourself: If the order also carried a decaying alpha, which direction would that push \(\lambda\)-equivalent urgency?

Lesson Summary

Almgren–Chriss casts optimal execution as minimizing expected implementation cost plus \(\lambda\) times its variance: temporary impact penalizes fast trading while timing risk penalizes slow trading. Sweeping the risk aversion \(\lambda\) traces an efficient frontier of strategies; the optimal trajectory is an exponential liquidation with urgency \(\kappa=\sqrt{\lambda\sigma^2/\eta}\), reducing to TWAP as \(\lambda\to0\) and to immediate liquidation as \(\lambda\to\infty\). There is no single best schedule - only a choice of where to sit on the frontier.

Formula Sheet Additions

Mean–variance execution cost
\[\E[\mathcal{C}]+\lambda\Var[\mathcal{C}]=\big(\tfrac12\gamma X^2+\sum_k\tfrac{\eta}{\tau}n_k^2\big)+\lambda\sigma^2\sum_k\tau x_k^2\]
Impact (trade fast is costly) versus timing risk (trade slow is risky).
Optimal exponential trajectory
\[x_k=X\,\frac{\sinh(\kappa(T-t_k))}{\sinh(\kappa T)},\ \ \kappa=\sqrt{\lambda\sigma^2/\eta}\]
Urgency rises with \(\lambda,\sigma\), falls with \(\eta\); \(\lambda\to0\) gives TWAP.
Error Log Checklist
  • Did I state a risk aversion \(\lambda\) (or urgency) rather than claim one optimal schedule?
  • Did I separate schedule-independent permanent impact from the optimizable terms?
  • Did I let \(\kappa\) respond to volatility/impact correctly?
  • Did I consider alpha/drift if the order carries a signal?

Retrieval Practice

Close the lesson and answer from memory before checking. This is deliberate, effortful recall - the single highest-yield study action.

▶ Show retrieval prompts & answers
Q: State the Almgren–Chriss trade-off and the objective it minimizes.
A: Trade fast to reduce timing risk vs trade slow to reduce market impact; minimize \(\E[\mathcal{C}]+\lambda\Var[\mathcal{C}]\) over the trading trajectory.
Q: What is the shape of the optimal trajectory and its two limiting cases?
A: An exponential/hyperbolic decay \(x_k\propto\sinh(\kappa(T-t_k))\); \(\lambda\to0\) gives a straight-line TWAP, \(\lambda\to\infty\) gives near-immediate liquidation.
Q: How does urgency \(\kappa=\sqrt{\lambda\sigma^2/\eta}\) respond to volatility and impact?
A: It increases with risk aversion and volatility (trade faster) and decreases with temporary impact \(\eta\) (trade slower).

Flashcards

Click to flip. These feed the site-wide spaced-repetition queue.

Almgren–Chriss trade-off
Impact (fast) vs timing risk (slow); minimize \(\E[\mathcal{C}]+\lambda\Var[\mathcal{C}]\).
Optimal trajectory
\(x_k=X\sinh(\kappa(T-t_k))/\sinh(\kappa T)\), \(\kappa=\sqrt{\lambda\sigma^2/\eta}\).
Limits
\(\lambda\to0\Rightarrow\) TWAP; \(\lambda\to\infty\Rightarrow\) immediate liquidation.

Completion Checklist

Confidence / mastery rating
Personal notes

Source References

This lesson synthesizes and paraphrases concepts from the sources below. No copyrighted text, problem sets, or solutions are reproduced. Return to the originals for full depth.

← 16.3Next: 16.5 →