A study of price manipulation behaviours surveillance based on quantized features

Price manipulation usually does not contain explicitillegitimate activities (i.e. financial rumour spreading, equity supply or demand squeezing), instead, includes submission and cancellation of limit orders, which appear to be normal trading behaviours. This paper proposes a Hidden Markov Model based system for detecting price manipulation behaviours in the capital markets. This paper starts from a thorough study of three primary types of price manipulation strategies, from which the intrinsic patterns of the manipulation is extracted through features extraction module, composed of wavelet transformation and gradient method. The extracted features are modelled by Hidden Markov Model, where the intentions of the trading are distinguished, quantitated and designated through the hidden states, which generate the variables that can be directly observed from the market. To overcome the non-stationary nature of the financial data, an adaptive mechanism is proposed for adaptively updating the model. Experimental evaluations for the new proposed system are conducted based on real financial data from NASDAQ and London Stock Exchanges as well as the simulated stock prices. Evaluations show that the proposed system stably outperforms the selected bench market models.