Market impact is a notion referring to the change in an asset’s price caused by the act of trading the asset itself. In the context of high-frequency trading (HFT), market impact is of particular importance since HFT algorithms trade large volumes of assets on relatively short time scales, resulting in potentially significant changes in the asset price due to the trading itself.
To understand the effect of market impact, consider the following example. Suppose a high-frequency trader wants to sell a large amount of shares in a particular stock. As they sell the shares, the supply of the stock available in the market increases, thereby putting downward pressure on the price. This could cause the price of the stock to decrease, impacting the overall profit the trader gains from the transaction.
Stochastic processes are widely used to model and manage market impact in HFT due to their ability to capture the random and uncertain nature of financial markets. Two primary approaches are common when applying stochastic processes to model market impact: the permanent and temporary market impact models. Both approaches are presented below.
1. **Permanent market impact models:** These models assume that the trading has a lasting effect on the asset price. A common model used for permanent market impact is the linear model proposed by Almgren & Chriss. The model’s specification is as follows:
$$p_t = p_0 + \eta\sum_{i=1}^{t}X_i,$$
where p0 is the initial price, pt is the price at time t, Xi is the trading size (positive for buy orders and negative for sell orders) at each time step, and η is a constant representing the market impact per share traded. In this model, the aggregate market impact is linearly proportional to the cumulative volume traded.
2. **Temporary market impact models:** These models assume that the market impact of a trade is transient, meaning that the price returns to its original level after the trade is executed. A widely used temporary market impact model is the square root model proposed by Bouchaud, Farmer, & Lillo. The model’s specification is:
$$p_t = p_0 + \xi\sum_{i=1}^{t}\operatorname{sgn}(X_i)\sqrt{|X_i|},$$
where sgn (x) is the signum function that indicates the sign of the trade (positive for buy and negative for sell orders) and ξ represents the coefficient for the temporary market impact. In this model, the aggregate market impact is proportional to the sum of the square roots of the trading sizes, emphasizing the diminishing impact for larger trades.
To manage market impact in HFT, traders often use stochastic optimal control techniques such as dynamic programming and reinforcement learning to optimize their trading strategies, considering the influence of the estimated market impact on the overall transaction cost. These stochastic optimal trading strategies, also known as optimal execution strategies, aim to minimize the trading cost resulting from slippage or adverse price impact while taking into account trading constraints such as liquidity and time limits.
In conclusion, market impact is an essential aspect of high-frequency trading as traders need to take into account the potential impact of their trades on asset prices. Stochastic processes are employed as models for market impact, providing a framework to systematically manage the influence of trading on asset prices, thus aiding in the optimization of trading strategies to minimize the transaction costs associated with market impact.