Inferring Trade Direction using State Space Approaches

To study liquidity in financial market, it is often of importance to identify which of the two parties in a transaction is supplying liquidity. Using a basic state space form method as the building block, this study introduces extensions to infer the direction of a trade. In addition, it assesses the classification performance when applied to a market where explicit trade and quote data are available. The proposed and implemented models include conditional heteroskedasticity and intraday volatility patterns. Results from a simulation study using the proposed models as data generating processes reveals accurate parameter estimations. In terms of the signing performance, these methods turn out to have higher percentage of sign agreement with the well-know Lee and Ready (1991) algorithm than the State Space Form Approximation method.