SVM-Enhanced Filtering Model for Limit Order Book Dynamics

Abstract

We propose a stochastic filtering model for Limit Order Book (LOB) where the signal process obeys the stochastic differential equation (SDE) whose trend function depend on the history of the LOB in such a way as to reflect the predicted direction of the mid price movement via the Support Vector Machine (SVM). Stochastic dynamics of the LOB is modeled with a queuing system where incoming orders and cancellations of existing orders arrive according to conditionally independent Poisson processes. Finally, numerical example with Nikkei 225 futures are presented.