Large Scale Memory Storage and Retrieval (LAMSTAR) Network

Information technology enables trading of financial products to be much more efficient and effective. In nowadays, 90% of the trading volumes of stocks come from algorithmic trading in market microstructure. In this project, we explore to apply LAMSTAR network to predict the price movement in market microstructure. More specifically, the historical price movement, the current order book statistics, as well as the previous order book statistics will be regarded as three subwords in the LAMSTAR model. The decision layer will provide the price movement predictions, which include (a) price higher than the current offer price, (b) price below the current bid price, and (c) price between the bid and offer. Experiment shows that LAMSTAR is very effective and efficient. It outperforms the three algorithms used for comparison, namely, SVM (Support Vector Machines — see: Cortes and Vapnik, 1995), BP (back propagation network — see Chap. 6), RBF (Radial Basis Function — see: Broomhead and Lowe, 1988) network in both success rate and efficiency (running time). Furthermore, we apply the LAMSTAR to analyze the most significant factors contributing to the price movement.