We apply modern statistical methods to the estimation and calibration of models of order books in financial equity markets. We develop particle filtering methods suitable for cases where partial observations have no noise. We test calibration methods including Maximum Likelihood and Expectation Maximization. We refine a basic order book model to achieve better agreement with market order book data. We show that an estimator based on Monte Carlo simulations of an order book is a better predictor of future price ticks. |