Real-Time Market Abuse Detection with a Stochastic Parameter Model

This paper develops a new model of market abuse detection in real time. Market abuse is detected, as Minenna (2003) proposed, on the basis of prediction intervals. The model structure is based on the discrete-time, extended market model introduced by Monteiro, Zaman, Leitterstorf (2007) to analyze the market cleanliness. Parameters of the expected return equation are assumed, however, to be time-varying and estimated under the state-space framework using the extended Kalman filter postulated by Chou, Engle, Kane (1992) to capture the GARCH effect in returns. QML estimation is performed on intraday data; its utilization is proposed as an alternative to the continuous time modeling by Minenna (2003). This framework is generalized to the bivariate case which enables the analysis of daily open/close data. The paper also extends procedures of the statistical verification of the estimated state-space model to include the uncertainty arising from time-invariant parameters.