Reducing transaction costs with low-latency trading algorithms

Abstract

We formulate a trade execution problem at the market microstructure level and solve it using dynamic programming. The objective is to sell a single lot of an asset in a short time horizon T, using the imbalance of the top of book bid and ask sizes as a price predictor. The optimization problem takes into account the latency L of the trading algorithm, which affects the prices at which the asset is traded. The solution divides the state space into a ‘trade’ and a ‘no-trade’ region. We calculate the cost of latency per lot traded and demonstrate that the advantage of observing the limit order book can dissipate quickly as execution latency increases. In the empirical section, we show that our optimal policy significantly outperforms a TWAP algorithm in liquidating on-the-run US treasury bonds, saving on average approximately 1/3 of the spread per share if trades are executed with low latency (1 ms).