Almgren-Chriss Model

The model formalizes the execution problem as a trade-off between two conflicting components:

• Expected Cost (Mean): The deterministic cost arising from permanent and temporary market impact. Selling faster increases this cost.
• Uncertainty (Variance): The risk of adverse price movements due to volatility. Holding the asset longer increases this risk. The objective function minimizes a linear combination: Expected Cost + λ * Variance of Cost, where λ (lambda) is the coefficient of risk aversion. A higher λ forces the strategy to trade faster to reduce exposure to price risk.

Almgren-Chriss distinguishes between two types of market impact that affect the price process:

• Permanent Impact: A linear function of the cumulative volume traded. It represents the information leakage that permanently shifts the equilibrium price as the market infers the presence of a large seller.
• Temporary Impact: A function of the instantaneous rate of trading. It represents the liquidity premium paid to attract counterparties and decays quickly after the trade completes. This decomposition allows the model to penalize both total size and trading speed separately.

For a risk-averse trader (λ > 0), the closed-form solution yields a hyperbolic trajectory rather than a linear one.

• The optimal trading speed is highest at the beginning of the liquidation period and decays over time.
• This front-loading minimizes the variance penalty by reducing exposure early, even though it incurs higher temporary impact costs initially.
• The trajectory is 'static' because it is computed once at time zero and does not update based on intraday price realizations, distinguishing it from dynamic programming approaches.

By varying the risk aversion parameter λ, the model traces out an efficient frontier in the space of expected cost versus variance.

• Each point on the frontier represents a Pareto-optimal strategy where cost cannot be reduced without increasing risk, and vice versa.
• The extremes of the frontier correspond to a risk-neutral strategy (λ=0, minimum cost, maximum variance) and a minimum-variance strategy (λ→∞, instantaneous liquidation).
• This framework allows a trading desk to select a strategy aligned with its specific risk mandate.

The continuous-time model is often implemented in a discrete-time framework for practical computation:

• The liquidation horizon T is divided into N equal intervals.
• The model solves for the optimal sequence of share quantities x_0, x_1, ..., x_N to sell in each interval.
• The resulting linear-quadratic structure allows the optimal holding path to be computed efficiently using tridiagonal matrix solvers, making it suitable for real-time pre-trade analysis.

While foundational, the base model makes simplifying assumptions that extensions address:

• Arithmetic Brownian Motion: Assumes prices follow a random walk with zero drift, ignoring momentum or mean-reversion signals.
• Linear Impact: Assumes permanent impact is linear, whereas empirical evidence often supports concave functions (e.g., square-root models).
• No Volume Dynamics: Ignores time-varying market volume, which is addressed by incorporating Volume-Weighted Average Price (VWAP) constraints or volume clock time.
• No Limit Orders: The model assumes all execution is via market orders, excluding the possibility of capturing the spread with passive orders.

The primary cost function minimized by the Almgren-Chriss framework. It quantifies the difference between the decision price (the price when the trading intention was formed) and the final execution price, capturing both explicit commissions and implicit market impact costs. The model treats this as a trade-off between certain loss from impact and uncertain loss from price risk.